Split (1)

train · 521k rows

text stringlengths 254 1.16M
--- title: Microbiota alteration of Chinese young male adults with high-status negative cognitive processing bias authors: - Hui-Min Xu - Shen-Wei Xie - Tian-Yao Liu - Xia Zhou - Zheng-Zhi Feng - Xie He journal: Frontiers in Microbiology year: 2023 pmcid: PMC10015002 doi: 10.3389/fmicb.2023.989162 license: CC BY 4.0 --- # Microbiota alteration of Chinese young male adults with high-status negative cognitive processing bias ## Abstract ### Introduction Evidence suggests that negative cognitive processing bias (NCPB) is a significant risk factor for depression. The microbiota–gut–brain axis has been proven to be a contributing factor to cognitive health and disease. However, the connection between microbiota and NCPB remains unknown. This study mainly sought to explore the key microbiota involved in NCPB and the possible pathways through which NCPB affects depressive symptoms. ### Methods Data in our studies were collected from 735 Chinese young adults through a cross-sectional survey. Fecal samples were collected from 35 young adults with different levels of NCPB (18 individuals were recruited as the high-status NCPB group, and another 17 individuals were matched as the low-status NCPB group) and 60 with different degrees of depressive symptoms (27 individuals were recruited into the depressive symptom group, as D group, and 33 individuals were matched into the control group, as C group) and analyzed by the 16S ribosomal RNA sequencing technique. ### Results As a result, the level of NCPB correlated with the degree of depressive symptoms as well as anxiety symptoms and sleep quality ($p \leq 0.01$). The β-diversity of microbiota in young adults was proven to be significantly different between the high-status NCPB and the low-status NCPB groups. There were several significantly increased bacteria taxa, including Dorea, Christensenellaceae, Christe -senellaceae_R_7_group, Ruminococcaceae_NK4A214_group, Eggerthellaceae, Family-XIII, Family_XIII_AD3011_group, Faecalibaculum, and Oscillibacter. They were mainly involved in pathways including short-chain fatty acid (SCFA) metabolism. Among these variable bacteria taxa, Faecalibaculum was found associated with both NCPB and depressive symptoms. Furthermore, five pathways turned out to be significantly altered in both the high-status NCPB group and the depressive symptom group, including butanoate metabolism, glyoxylate and dicarboxylate metabolism, propanoate metabolism, phenylalanine, tyrosine, and tryptophan biosynthesis, valine, leucine, and isoleucine degradation. These pathways were related to SCFA metabolism. ### Discussion Fecal microbiota is altered in Chinese young male adults with high status NCPB and may be involved in the biochemical progress that influences depressive symptoms. ## Introduction From the cognitive perspective, negative cognitive processing bias (NCPB), which was characterized as negative bias in attention, explanation, memory, and rumination being accompanied by poor sleep quality, has been reported to be the core feature of depression (Disner et al., 2011; Rozenman et al., 2014; Gobin et al., 2015). The information processing involving sensory, perception, attention, memory, learning, and so on constitutes cognition, which has been proven to be essential for maintaining physical and mental health (Kwak et al., 2016; Bayne et al., 2019; Liu et al., 2020; Ling et al., 2021). For the last decade, the cognitive(Aatsinki et al., 2022) function has been linked with microbiota composition. Initially, germ-free (GF) mice were found to exhibit anxiolytic basal behavior utilizing the elevated plus maze (EPM) compared to the specific pathogen-free (SPF) mice (Neufeld et al., 2011). Furthermore, GF mice were proven to be exhibiting changes in learning, memory recognition, and emotional behavior resulting from the absence of microbiota (Connell et al., 2022). The studies on the fecal microbiota of 8-month-old infants described the association between early microbiota and later fear bias. It observed a lower abundance of Bifidobacterium and a higher abundance of Clostridium with an increased “fear bias” of infants. Chronic antibiotic depletion of microbiota populations alters cognition-related metabolism and the expression of key cognitive signaling molecules, leading to long-lasting effects on cognition (Fröhlich et al., 2016; Connell et al., 2022). The administering probiotics, such as Bifidobacterium longum 1714, modulate the behavior or cognition in both rodents and humans (Savignac et al., 2015; Akbari et al., 2016; Connell et al., 2022). Novel perspectives suggest that the dynamic bidirectional communication systems of the microbiota–gut–brain axis may be a contributing factor to cognitive health and diseases. Still, the exact mechanisms, especially on NCPB, remain unknown (Connell et al., 2022). Traditionally, neurobiological mechanism holds the view that a loss of neural plasticity explains the occurrence of depression (Duman et al., 1999; Li and Wang, 2021). At present, abnormality of the microbiota–gut–brain axis is verified as an important risk factor for depressive symptoms (Aizawa et al., 2016; Caspani et al., 2019; Chung et al., 2019; Berg et al., 2020). The germ-free mice were found to decrease the immobility time in the forced swimming test more than healthy control mice. Furthermore, fecal microbiota transplantation of GF mice derived from patients with major depressive disorder (MDD) leads to aggravating depressive-like behaviors compared with the colonization of the “health” fecal microbiota from control individuals (Zheng et al., 2016). According to Beck’s cognitive theory of depression, individuals suffering from stressful life events might automatically activate negative cognitive schemas, which lead to a negative tendency of cognition (Beck, 1967). The unpredictable chronic mild stress (UCMS) model mouse displays depressive-like behaviors. Fecal microbiota transplantation could have transferred the depression phenotype from UCMS donor mice to naive recipient mice (Chevalier et al., 2020). The studies have revealed that the adverse effects of UCMS-transferred microbiota were alleviated by complementation with a strain of the Lactobacilli genus in the mice (Chevalier et al., 2020). Yet the microbiota–brain–gut axis of NCPB has been largely neglected. The purpose of our study was to identify the adverse effect of NCPB on an individual’s depressive symptoms, explore the role of microbiota in NCPB, and discuss the possible underlying mechanisms by which NCPB affects depression. We evaluated the changes in fecal microbial community structure and composition in subjects with high-status NCPB or depressive symptoms, analyzed the correlation of the variation tendency involved in NCPB with the degree of depressive symptoms, and hypothesized that (a) NCPB positively correlated with depressive symptoms, anxiety symptoms, and poor sleep quality in young adults; (b) the changes in microbiota could be observed in individuals with high-status NCPB, as well as depressive symptoms compared to that in controls; and (c) specific taxa and functional pathways may be found to potentially mediate the affection of NCPB on depressive symptoms. ## Self-report measures The self-designed socio-demographic information questionnaire was used to collect personal information including participants’ age, gender, ethnicity, only-child status, family types, and contact ways. Moreover, based on Beck’s NCPB theory of depression, the negative cognitive processing bias questionnaire (NCPBQ) is designed to assess the degree of an individual’s NCPB (especially for the Chinese population). In our previous studies, NCPBQ has good reliability (Cronbach’s alpha coefficient = 0.89) and validity (all factor loads are over 0.30) in the Chinese population (Yan et al., 2017). It showed good reliability in this study with Cronbach’s alpha coefficient being 0.92 (Jiang et al., 2017; Huang et al., 2021; Li et al., 2021). It is a 4-point scale on four factors: negative attention bias (i.e., sustained attention on negative information), negative memory bias (i.e., tendency to remember or recall negative life events), negative rumination bias (i.e., rumination on personal negative emotions), and negative explanation bias (i.e., tendency to make a negative explanation on events). The total scores of NCPBQ range from 16 to 64, with the higher scores indicating the higher status of NCPB. In the present study, to recruit participants with a high or low level of NCPB, we transformed the linear slope from a continuous variable into three categories on a common way of grouping (bottom $27\%$, top $27\%$, and others) (Notebaert et al., 2020; Ge et al., 2022). Self-Rating Depression Scale (SDS) is a 4-point Likert-type and self-reported scale used to assess the severity of depressive symptoms. A total raw SDS score was obtained by summing the ratings of the 20 items, which were divided by 80 to create a depression severity index. A depression severity index greater than 0.5 was considered to indicate depressive symptoms. With high internal consistency, high validity in differentiating between depressed and non-depressed subjects, and international propagation, SDS has been a worldwide inventory among the most used self-rating scales for measuring depressive symptoms (Zung, 1965). Anxiety symptoms were assessed by self-report with the Chinese version of the Self-rating Anxiety Scale (SAS), which has been validated in the Chinese population in several studies (Zung, 1971; Gong and Chan, 2018; Wu et al., 2019). Cronbach’s alpha coefficient measured in the current study was 0.79. Pittsburgh Sleep Quality Index (PSQI) is a measurement tool used for assessing sleep quality, which has been widely used in clinical and healthy populations all over the world. The higher scores indicate poorer sleep quality. A Chinese version of the PSQI has been validated with adequate reliability (Buysse et al., 1989; Guo et al., 2016). ## Study design and participants The study design was approved by the Human Research Ethics Committee of the Army Medical University (the number was: IEC-C-[B013]-02-J.02). A cross-sectional survey was conducted in Mianyang city, Sichuan province, between November and December 2019, which covered 766 individuals. According to the results of the survey, 18 young adults with high-status NCPB and 17 with low-status NCPB (HS represented the high-status NCPB and LS represented the low-status NCPB) were recruited in the following experimental study. Meanwhile, 27 people with severe depressive symptoms and 33 people as the control were also recruited. In the trial phase, fecal samples were collected to analyze the structure of fecal microbiota after subjects were given 1 month of the same balanced diet. The detailed design is shown in Figure 1. Written informed consent was obtained from all participants. FIGURE 1: Flow diagram of the study. HS group, the high-status NCPB group; LS group, the low-status NCPB group; D group, the depression symptom group; C group, the control group. In the cross-sectional survey, eligible subjects are defined as follows: [1] young adults aged 18–25 years old and [2] young adults who were willing to take part in the survey. All subjects consumed the same balanced diet according to Chinese Recommended Dietary Allowance (Ge, 2011; Yang et al., 2018) for 1 month (Sanz, 2010; Jing et al., 2021) to avoid the influence of diet and forbidden snacks. A total of 766 individuals were approached for participation, and 31 of them were excluded from our analysis. Finally, as many as $95.95\%$ ($\frac{735}{766}$) of young adults were included in the statistical analysis. In the experimental study section, 18 individuals were recruited as the HS group if their scores of NCPBQ were in the top $27\%$. Similarly, another 17 individuals were matched as the LS group with scores of NCPB in the bottom $27\%$ as in previous studies (Notebaert et al., 2020; Ge et al., 2022). Moreover, 27 individuals with an SDS index (see later) higher than 0.6 were recruited into the depressive symptom group (D group), and 33 individuals lower than 0.5 were matched as the control group (C group) (Zung, 1965). Considering that a variety of factors might influence microbiota composition, we excluded subjects in the previous 1 month if any of the following criteria were met, including [1] a history of severe cardiac, pulmonary, hepatic, renal, intestinal diseases, or any kind of tumor; [2] antibiotic, probiotic, prebiotic and synbiotic application, as well as active bacterial, fungal or viral infections, gastrointestinal surgery; [3] a history of other psychiatric diseases except depression (e.g., schizophrenia); and [4] they were not willing to participate in the experiment. All the subjects needed to meet the following conditions: [1] young adults aged 18–25; [2] Han nationality; [3] male; and [4] they were willing to take part in the experiment. ## Cross-sectional survey The cross-sectional survey was carried out after a clear illumination. The participants were asked to complete the questionnaires designed for collecting personal information and estimating the degree of NCPB, depressive symptoms, anxiety symptoms, and sleep quality. ## Fecal sample collection and DNA isolation According to the results of the cross-sectional survey, 18 young adults with high-status NCPB and 17 age- and gender-matched with low status were recruited. Moreover, 27 young adults with high depressive symptoms were selected into the depressive symptom group, and 33 age- and gender-matched young adults with low depressive symptoms were also recruited. The fecal samples collected from the recruited participants in the morning were numbered and stored at −80°C before analyses. According to the instruction, the standard Power Soil DNA Isolation Kit was used to extract DNA (QIAGEN, Germany). ## 16S rRNA gene sequencing and analysis The V3–V4 variable region of the bacteria’s 16S ribosomal RNA (rRNA) gene was amplified by PCR amplification technology with barcode-indexed primers, including 338F (5′-ACTC-CTACGGGAGGCAGCA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′). The PCR amplification system and reactions were consulted in the previous studies (Zhuang et al., 2018). In the experimental section, raw reads were filtered and quality-controlled to remove chimeric sequence reads. All remaining sequence reads were assigned to operational taxonomic units (OTUs) with a $97\%$ threshold of pairwise identity using the UCLUST comparison tool of the quantitative insights into microbial ecology (QIIME) pipeline (version 1.8.01) and then taxonomically classified with the RDP reference database. α-Diversity including the ACE index, Chao index, Simpson index, and Shannon diversity index was calculated for each sample or group. β-Diversity was also evaluated by principle coordinate analysis (PCoA) of weight and unweight UniFrac distances and Bray–Curtis dissimilarity as previously described (Chung et al., 2019). Furthermore, the taxonomic distributions of OTUs were performed and graphics were constructed based on the relative abundance of microbiota in each taxon in the samples and groups. The linear discriminant analysis (LDA) effect size (LEfSe) method was used to identify significant OTUs differentially. Moreover, Mann–Whitney U-test was conducted to assess differences in socio-demographic characteristics, species diversity indexes, and the relative abundance of microbiota among the various taxonomic levels using SPSS 22.0 (IBM, Armonk, NY, USA). Finally, the KEGG database2 was used in the signal pathway analysis to annotate pathways. We analyzed functional pathways by STAMP (version 2.1.3) (Parks et al., 2014) to explore the potential functional properties of the identified microbiota. All tests mentioned earlier were two-tailed, with a p-value of < 0.05 considered statistical significance. ## Statistical analysis In the analysis of cross-sectional survey results, reliability and internal consistency were estimated using Cronbach’s alpha coefficient. Descriptive statistics, t-test, one-way ANOVA or Chi-square test, Pearson’s correlation analysis, and logistic regression analysis were performed with SPSS 22.0 (IBM, Armonk, NY, USA). ## Results of cross-sectional survey A total of 735 young adults were finally enrolled in the statistical analysis. Those subjects were all male adults, ranging from 18 to 26 years old (23.76 ± 3.67). A majority of them were Han ethnicity, who showed less depressive or anxiety symptoms and better sleep quality ($p \leq 0.05$). Detailed socio-demographic variables are shown in Supplementary Table 1. Negative cognitive processing bias is positively correlated with depressive symptoms and anxiety symptoms, as shown in Table 1. Correlations among study variables showed that NCPB positively correlated with depressive symptoms, anxiety symptoms, and sleep quality in young adults. It was also found in the relationship between depressive symptoms, anxiety symptoms, and sleep quality. The results of regression analyses showed that NCPB. could positively predict the degree of depressive symptoms and anxiety symptoms and negatively predict the quality of sleep, covering the proportion of total variance of 16.7, 31.8, and $20.5\%$, respectively. Detailed information is shown in Table 2. ## Comparison of the microbiota in the high-status and low-status NCPB groups As shown in Supplementary Table 2, no significant difference was found in gender, ethnicity, age, only-child status, and the body mass index (BMI) between the high-status and low-status NCPB. We have already verified the sample size of flora analysis and the data collection volume is sufficient before the comparision of the microbiota in different groups, as shown in Supplementary Figure 1. Overall, Adonis unweighted UniFrac dissimilarity metrics showed that the fecal microbial communities significantly differed with different statuses of NCPB (Adonis $$p \leq 0.007$$) (Figures 2A, B), suggesting dissimilar microbiota composition between the high-status and low-status groups. The level of α-diversity was slightly higher in the high-status group, but not statistically significant (Supplementary Table 3 and Figure 2C). FIGURE 2:* β-Diversity measures (A,B) and α-diversity measures (C) including (C-1) ACE, (C-2) Chao, (C-3) Shannon, and (C-4) Simpson of the fecal microbiota in HS and LS groups. HS, the high-status NCPB; LS, the low-status NCPB. In addition, the composition of the fecal microbiota was different in the high-status NCPB and the low-status groups. A total of 18 pivotal discriminatory OTUs were identified by using the Random Forest algorithm, including five OTUs (assigned to the genus of Bacteroides, Faecalibaculum, Family_XIII_unclassified, Christensenellaceae_R-7_group, and Coriobacteriales_unclassified) were overrepresented in the HS group, while six OTUs (assigned to the genus of Bacteroides, Bacteria_unclassified, and Butyricimonas) were overrepresented in the low-status NCPB group (Figure 3A). The taxonomic compositions of fecal microbiota in the two groups are shown in Figure 3B. FIGURE 3: Heat map of relative abundance at the level of OTU (A) and composition (B) of the fecal microbiota at the level of (B-1) phylum, (B-2) class, (B-3) order, and (B-4) family in HS and LS groups. HS, the high-status NCPB; LS, the low-status NCPB. Further analysis revealed that several targets were significantly higher in the high-status group, including four families (Family_XIII, Christensenellaceae, Peptococcaceae, and Eggerthellaceae) and 12 genera (Faecalibaculum, Family_XIII_unclassified, Ruminococcaceae_UCG-010, Ruminoco- ccaceae_unclassified, Eggerthellaceae_unclassified, Dorea, Chris- tensenellaceae_R7_group, Ruminococcaceae_NK4A214_group, Eub- acterium, Peptococcus, Family_XIII_AD3011_group, and Oscillibacter) (Supplementary Table 4). The heat map of 50 top genera in HS and LS group was showed in Figure 4A. FIGURE 4: Heat map of 50 top genera in HS and LS groups (A) and difference of meaningful bacterial taxa (B) in HS and LS groups. (B-1) Taxonomic represents a statistical difference in groups. (B-2) Histogram of the LDA scores for differential abundant genera. HS, the high-status NCPB; LS, the low-status NCPB. Linear discriminant analysis effect size analysis [LDA scores (log10) >2] identified nine meaningful pregnancy taxa in the high-status NCPB group, inclusive of Dorea, Christensenellaceae, Christensenellaceae_R_7_group, Ruminococcaceae_NK4A214_group, Eggerthellaceae, Family -_XIII, Family_XIII_AD3011_group, Faecalibaculum, and Oscillibacter (Figure 4B). ## Shared bacteria associated with NCPB and depressive symptoms As shown in Supplementary Table 5, young adults in the depressive symptom group also had more severe anxiety symptoms. No significant difference was found in gender, ethnicity, age and only-child status, and BMI between the depressive symptom group and normal controls. α-Diversity and β-diversity of the fecal microbiota were not significantly different, as shown in Supplementary Table 6 and Supplementary Figure 2A. Still, analyses revealed several taxon targets for the depressive symptom group, as shown in Figure 5 and Supplementary Figure 2B and Supplementary Table 7. LEfSe was also performed [LDA scores (log10) >1.5] and identified the meaningful taxa. Agathobacter, Faecalibaculum (LDA score was 1.9992248338), and Ruminococcus_2 were significantly increased in the depressive symptom group, while Bilophila, Eggerthella, Deltaproteobacteria, Erysipelotrichaceae_UCG_003, and Ruminococcus_gnavus_group were decreased (Figure 5). FIGURE 5: Boxplots of relative abundance of several genera with a significant difference in depression symptom and the control groups. (A) Difference of meaningful bacterial taxa in depression symptom and the control groups and (B) in depressive symptom and control groups. (B-1) Taxonomic represents a statistical difference in groups. (B-2) Histogram of the LDA scores for differentially abundant genera. The results of the fecal microbial structure indicated a significant fecal microbial imbalance in young male adults with a high level of NCPB or depressive symptoms. NCPB and depressive symptom groups shared similar bacteria (Figure 6A). Faecalibaculum was increased in both NCPB and the depressive symptom group, and a graphical representation is shown in Figure 6B. FIGURE 6: Analysis of commonly associated genera with NCPB and depressive symptom groups. (A) The common associated genera with NCPB and depressive symptom groups (B) as well as the results of Faecalibaculum. HS, the high-status NCPB; LS, the low-status NCPB; D, the depression symptom group; C, the control group. ## Predictive microbiota functional profiling and shared pathways associated with NCPB and depressive symptoms Functional profiling of microbial communities was predicted based on OTUs. A total of 11 pathways were found significantly different in the predictive microbiota functional profiling between the high-status and low-status NCPB groups. The predicted pathways mainly included compound metabolism (carbohydrate, amino acid metabolism, and lipid), membrane transport, cell growth, and death pathways. Detailed results are displayed in Figure 7A ($p \leq 0.05$). FIGURE 7: Analysis of functional pathways in NCPB groups and depressive symptom groups. (A) The high-status and low-status NCPB groups, (B) the depressive symptom and control groups, (C) and shared pathways associated with NCPB and depressive symptoms. HS, the high-status NCPB; LS, the low-status NCPB. Nevertheless, the predictive microbiota functional profiling was changed relative to depressive symptoms. In the functional pathway analyses, a total of 25 pathways were significantly enriched and 47 were depleted in the depressive symptom group. These pathways mainly included compound metabolism (e.g., carbohydrate, amino acid metabolism, nucleotide, and lipid), biosynthesis processes (e.g., lysine, arginine, fatty acid, phenylalanine, tyrosine, and tryptophan), degradation (e.g., lysine, fatty acid, and RNA), and signal transduction (e.g., two-component system), with detailed results partly presented in Figure 7B ($p \leq 0.01$) and whole in Supplementary Figure 3 ($p \leq 0.05$). In addition, the results also showed that NCPB and depressive symptom groups shared similar pathways. Several common pathways related to both NCPB and depressive symptoms were also found including butanoate metabolism, glyoxylate and dicarboxylate metabolism, propanoate metabolism, phenylalanine, tyrosine and tryptophan biosynthesis, valine, leucine, and isoleucine degradation (Figure 7C). ## NCPB is positively correlated with depressive symptoms and anxiety symptoms Depression is considered to be one of the most common mental disorders, and its pathogenesis is still unknown. In many cases, it is a common outcome of exposure to traumatic events, which influence information processing, modify attention or memory functions, and promote dysfunctional interpretations of current experiences (LeMoult and Gotlib, 2019; Qu et al., 2019; Vidańa et al., 2020). Referring to Beck’s model, cognitive distortions and negative shelf schema, often based on childhood experiences, bring about depression and co-occur with mood disorders (Sanchez et al., 2017; Beevers et al., 2019; Li et al., 2021; Noworyta et al., 2021). Apart from substantial research examining depressive cognitive content as a vulnerability factor (e.g., negative thoughts), a promising line of research highlights the role of NCPB in the development, maintenance, and relapse/recurrence of depressive symptoms or clinical depression (De Raedt and Koster, 2010; Dagmara et al., 2019; LeMoult et al., 2020). In our studies, the relationship between NCPB, depressive symptoms, anxiety symptoms, and sleep quality among 735 Chinese young adults were discussed. In line with former studies, we certified that the scores on NCPBQ positively correlated with scores of depressive symptoms, anxiety symptoms, and sleep quality in Chinese young male adults. Further results in regression analyses showed that NCPB could positively predict the degree of depressive symptoms and anxiety symptoms and negatively predict sleep quality, covering the proportion of total variance of 16.37, 31.8, and $20.5\%$. There is substantial evidence for general cognitive deficits in depression which is revealed in the literature (Joormann and Gotlib, 2010; Koster et al., 2011; Joormann and Vanderlind, 2014; Sanchez et al., 2017; Van den Bergh et al., 2018). Patients with cognitive deficits are typically characterized by impairments in attention, motor skills, working memory, and executive functions, which are prominent features of psychosis (LeMoult and Gotlib, 2019). Pathology-congruent interpretative biases are found in the prodromal phase; then, this presents an exciting new treatment possibility (Aronoriaga-Rodriguez and Fernandez-Real, 2019). Cognitive models have been proposed originally as etiological theories of depression (Mathews and MacLeod, 2005; Bomyea et al., 2017; Poole et al., 2017; Buckley et al., 2020). ## The composition of the microbiota is significantly associated with the level of NCPB Nowadays, the microbiota–gut–brain axis has been proposed as a key regulator of stress responses, providing possibilities for the prevention and treatment of stress-related disorders (Gubert et al., 2020). The fecal microbiome seems to exert psychological effects affecting cognitive and emotional reactivity (Proctor et al., 2017; Vuong et al., 2017; Foster et al., 2021). Studies have found that disturbances in the homeostasis of the microbiota, such as a consequence of antibiotics, result in alterations at neural, hormonal, and immunological levels and are involved in the physiological stress response and behavior in both animals and humans, including patients with depression (Campos et al., 2016; Zheng et al., 2016; Sarkar et al., 2018). However, it is still unknown on the connection between the NCPB and the consequences of fecal microbial composition structure transform. In our studies, we assessed the changes in the composition of the bacteria at different levels of NCPB. The significantly diverse fecal bacteria β-diversity and increase in the relative abundance of several microbiota, such as Family_XIII, Christensenellaceae, Peptococcaceae, and Eggerthellaceae, confirmed that the fecal bacteria comparison is variations of the alteration of the NCPB levels (Figure 2). We speculated that the degrees of our volunteers’ depressive symptoms might weaken the shift of microbial composition structure which has been verified in patients with MDD in previous studies (Ritchie et al., 2022). There were neither α-diversity nor β-diversity significant differences at the phylum level caused by aggravated depressive symptoms (Supplementary Figure 2A). Nevertheless, we found that some of the relative abundances of fecal microbiota, such as Deltaproteobacteria, had changed significantly, as shown in Figure 5. The cognition disorders involved in emotional disturbance, such as anxiety, have been associated with the imbalance of intestinal flora via the stress response in the clinical data (Ritchie et al., 2022). Several specific microbial families and genera have been associated with cognitive decline, anxiety behaviors, and affective disorders, such as Bifidobacterium lactis CNCM I-2494, Lactobacillus bulgaricus, Streptococcus thermophilus, and Lactobacillus lactis. The probiotic mixture containing these microbiota have been reported substantially to alter brain activity during the emotional reactivity test in healthy volunteers (Cryan and Dinan, 2012). To the best of our knowledge, how the gut microbiome might vary specifically from patients with depression with and without the NCPB specifier had not been explored previously. ## Faecalibaculum was the common bacterium associated with both NCPB and depressive symptoms More interestingly, in our studies, among these flora community compositions converts, Faecalibaculum, which belongs to Erysipelotrichaceae, was the common bacterium associated with both NCPB and depressive symptoms at the genus level. It was not only higher in the high-state NCPB group but also in the depressive symptom group (Figure 6). It has been proven that Faecalibaculum rodentium transplanted from chronic social defeat stress (CSDS)-susceptible mice might induce the anhedonia-like behavior in the antibiotic cocktail (ABX)-treated WT and Ephx2KO mice, which did not show depressive behaviors in the exposure of CSDS. Ingestion of Faecalibaculum for 14 days induced anhedonia-like and depression behaviors of ABX-treated Ephx2KO mice, accompanied by the increased expression of proinflammatory factors, such as interleukin-6 in the blood and reduction of synaptic proteins expression in the prefrontal cortex (Wang et al., 2021). The combination of probiotics and prebiotics contains Lactobacillus with Faecalibaculum, Blautia, or Bifidobacterium spp. was proven to normalize the gut microbiome diversity and improve depressive-like behavior of the chronic stress-induced depression and anxiety in mice model (Westfall et al., 2021). Furthermore, growing evidence indicates that Faecalibaculum enriched “Western diet” or high-fat diet (HFD) fed mice, especially positively correlated with serum proinflammatory cytokines such as TNF-α, IL-6, and LBP in HFD-fed mice, and closely related to the metabolism disorders involved in energy production or adiposity (Skonieczna-Żydecka et al., 2018; Ma et al., 2019; Wei et al., 2020). Faecalibaculum was considered a gut antigen causing the abnormal function of the microbiota–gut–brain axis (Oleskin and Shenderov, 2016; Dalile et al., 2019). A high sugar diet and innate lymphoid cell 3 (ILC3) promoted that the outgrowth of F. rodentium could displace Th17-inducing microbiota and posing risk for metabolic syndrome in mice (Kawano et al., 2022). Activity ileal ILC3 might regulate the ileal Treg/T helper 17 cells ratio and impact the production of hippocampal and prefrontal cortex chemotactic in the stress-induced behavioral deficits mice model, which could be relieved by the combination of probiotics and prebiotics and promoted behavioral resilience to the chronic and recurrent stress by normalizing gut microbiota populations (Westfall et al., 2021). These studies have revealed a connection between stress-induced depression and anxiety-like behavioral impairments depending on the microbiota–gut–brain axis and immune regulation system, and most of them were based on animal research rather than clinical studies. ## Shared metabolic pathways associated with NCPB and depressive symptoms were mainly involved with SCFAs We continuously analyzed the functional profiling of microbiota communities predicted based on OTUs. It is revealed that 11 functional pathways were significantly different and mainly focused on compound metabolisms, such as carbohydrate, amino acid metabolism and lipid, membrane transport, cell growth, and death pathways (Figure 7A). It has been proven that the production of several metabolic responses impacts a variety of life functions, including neurological functions (Kennedy et al., 2017; Martin et al., 2020). The pathways, including the phenylalanine, tyrosine and tryptophan biosynthesis, butanoate metabolism, propanoate metabolism, glyoxylate and dicarboxylate metabolism, were significantly altered between the high-status and low-status NCPB in our studies. And they were reported to be involved into the cognitive dysfunction in rodents and patients with depression. In which five pathways, such as butanoate metabolism, glyoxylate, and dicarboxylate metabolism, propanoate metabolism, phenylalanine, tyrosine, and tryptophan biosynthesis, valine, leucine, and isoleucine degradation, were invovled with environment responses such as two-component system, and the neurotrophy relevantly pathways such as folate biosynthesis and biosynthesis of amino acids. The production of these pathways, especially the SCFAs, such as butanoate, phenylalanine, tyrosine, and tryptophan, has been proven to modulate multiple biological processes including neurological function. Produced by the most endogenous microbiota, SCFAs readily pass through the mucosa layer and cell membranes and exert toxic effects on mammal cells at high concentrations, particularly affecting the functions of the nervous system by serving as nutrients, metabolites, or regulators involved in the operation of various kinds of nervous cells (Valles-Colomer et al., 2017). A clear association has been found between lower levels of SCFAs and decreased representation of obligate anaerobes such as the Faecalibaculum, Lachnospiraceae, and Ruminococcaceae in human fecal microbiota (D’Amato et al., 2020). The fecal microbiota transplant contained those microbiotas including Faecalibaculum and Lachnospiraceae from aged donor mice into young adult recipients altered the abundance of bacteria associated with SCFAs production and impacted the cognitive function (D’Amato et al., 2020). Several studies have revealed the potential ability of Erysipelotrichaceae, Bifidobacterium, Faecalibaculum, Bacteroides, and Romboutsia to produce SCFAs in mice (Smith et al., 2013; Dalile et al., 2019). A correlation between SCFA levels and the abundance of Faecalibaculum, Romboutsia, Bacteroides, and Turicibacter were found in rats. Faecalibaculum was the most strongly positively correlated with the levels of SCFA levels, which are well known as the endogenous ligands of PPARγ, and crucially together with increasing PPARγ expression, promoting the PPARγ/MAPK/NF-κB signaling pathway connected with the metabolism and immune system (Zhang et al., 2022). In addition, SCFAs, especially butyrate production, provide energy substrates for colonocytes, mitigate inflammation, and regulate satiety for their host. Their deficiency or redundancy not only leads to metabolic diseases but is also involved in depression and other mood disorders (Oleskin and Shenderov, 2016). Michels et al. [ 2017] and Skonieczna-Żydecka et al. [ 2018] demonstrated that emotional problems in Belgian children and Polish depressive women. were associated with a significantly higher concentration of butyrate, isobutyrate, valerate, and isovalerate. Hence, the results of fecal microbial structure and function alteration indicated a significant fecal microbial imbalance in Chinese young male adults with high levels of NCPB with depressive symptoms. It may be a hint that explains the mechanism of higher levels of NCPB involved with severe depressive symptoms. ## Conclusion This is the first study on the connection between microbiota and NCPB, exploring the co-exist microbiota of NCPB and depressive symptoms by sequencing and analyzing the fecal microbiota of 735 Chinese young adults who have to take a balanced diet for 1 month and restricted of snacks. The NCPB status of these young adults was positively correlated with the degree of depressive symptoms as well as anxiety symptoms and sleep quality ($p \leq 0.01$). We, fortunately, found several target bacteria taxa, including Dorea, Christensenellaceae, Christensenellaceae_R_7_group, Ruminococcaceae_NK4A214_group, Eggerthellaceae, Family-XIII, Family_XIII_AD3011_group, Faecalibaculum, and Oscillibacter that were significantly enriched in the high-status NCPB group than in the low-status group. These bacteria taxa were mainly involved in pathways including SCFA metabolism. Furthermore, Faecalibaculum, which was proven involved with the depressive-like phenotypes in laboratory animals, was also found significantly increased in the depressive symptom group compared with the control group. Five pathways, such as butanoate metabolism, glyoxylate, and dicarboxylate metabolism, turned out to be significant altered in both the high-status NCPB group and the depressive symptoms group. The results hint at the common bacteria taxa and pathways involved with either NCPB or depressive symptoms and provided some clues for the crosstalk of NCPB and depressive symptoms via the microbiota–brain–gut axis. ## Limitations There was an exploratory study that needs to be replicated across larger samples and compared with a healthy control group. ## Data availability statement The data presented in this study are deposited in the BIG submission, the accession numbers are CRA009867 and CRA009825. Available from https://ngdc.cncb.ac.cn/gsa/s/m0k5fYEp and https://ngdc.cncb.ac.cn/gsa/s/OD562MZb. ## Ethics statement The studies involving human participants were reviewed and approved by the study of the mechanism of overgeneralized autobiographical memory in three stages of suicide in depression. The studies described in our manuscript is part of the Ethics Committee. The patients/participants provided their written informed consent to participate in this study. ## Author contributions Z-ZF guided the studies and critically revised the manuscript. XH conceived and designed the experiments. S-WX, H-MX, T-YL, and XH collected and analyzed the data. XH, H-MX, and XZ wrote the manuscript. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2023.989162/full#supplementary-material ## References 1. Aatsinki A., Kataja E., Munukka E., Lahti L., Keskitalo A., Korja R.. Infant fecal microbiota composition and attention to emotional faces.. (2022) 22 1159-1170. DOI: 10.1037/emo0000924 2. Aizawa E., Tsuji H., Asahara T.. Possible association of Bifidobacterium and Lactobacillus in the gut microbiota of patients with major depressive disorder.. (2016) 202 254-257. DOI: 10.1016/j.jad.2016.05.038 3. Akbari E., Asemi Z., Daneshvar R. K., Bachmani F., Kouchaki E., Tamtaji O. R.. Effect of probiotic supplementation on cognitive function and metabolic status in Alzheimer’s disease: a randomized, double-blind and controlled trial.. (2016) 8. DOI: 10.3389/fnagi.2016.00256 4. Aronoriaga-Rodriguez M., Fernandez-Real J. M.. Microbiota impacts on chronic inflammation and metabolic syndrome related cognitive dysfunction.. (2019) 20 473-480. DOI: 10.1007/s11154-019-09537-5 5. Bayne T., Brainard D., Byrne R. W., Chittka N., Heyes C., Matter J.. What is cognition?. (2019) 29 R608-R615. DOI: 10.1016/j.cub.2019.05.044 6. Beck A. T.. (1967) 7. Beevers C. G., Mullarkey M. C., Dainer-Best J., Stewart R. A., Shumake J.. Association between negative cognitive bias and depression: a symptom-level approach.. (2019) 128 212-227. DOI: 10.1037/abn0000405 8. Berg G., Rybakova D., Fischer D., Cernava T., Vergès M., Charles T.. Microbiome definition re-visited: old concepts and new challenges.. (2020) 8. DOI: 10.1186/s40168-020-00875-0 9. Bomyea J., Johnson A., Lang A. J.. Information processing in PTSD: evidence for biased attentional, interpretation, and memory processes.. (2017) a4 218-243. DOI: 10.5127/pr.037214 10. Buckley T. C., Blanchard E. B., Neill W. T.. Information processing and PTSD: a review of the empirical literature.. (2020) 4 218-243. DOI: 10.1016/s0272-7358(99)00030-6 11. Buysse D. J., Reynolds C. F., Monk T. H., Berman S. R., Kupfer D. J.. The Pittsburgh sleep quality index-a new instrument for psychiatric practice and research.. (1989) 28 193-213. DOI: 10.1016/0165-1781(89)90047-4 12. Campos A. C., Rocha N. P., Nicoli J. R., Vieira L. Q., Teixeira M. M., Teixeira A. L.. Absence of gut microbiota influences lipopolysaccharide-induced behavioral changes in mice.. (2016) 312 186-194. DOI: 10.1016/j.bbr.2016.06.027 13. Caspani G., Kennedy S., Foster J. A.. Gut microbial metabolites in depression: understanding the biochemical mechanisms.. (2019) 6 454-481. DOI: 10.15698/mic2019.10.693 14. Chevalier G., Siopi E., Guenin-Macé L., Pascal M., Laval T., Rifflet A.. Effect of gut microbiota on depressive-like behaviors in mice is mediated by the endocannabinoid system.. (2020) 11. DOI: 10.1038/s41467-020-19931-2 15. Chung Y. C. E., Chen H. C., Chou H. L., Chen I. M., Lee M. S., Chuang L. C.. Exploration of microbiota targets for major depressive disorder and mood related traits.. (2019) 111 74-82. DOI: 10.1016/j.jpsychires.2019.01.016 16. Connell E., Le Gall G., Pontifex M. G., Sami S., Cryan J. F., Clarke G.. Microbial-derived metabolites as a risk factor of age-related cognitive decline and dementia.. (2022) 17. DOI: 10.1186/s13024-022-00548-6 17. Cryan J. F., Dinan T. G.. Mind-altering microorganisms: the impact of the gut microbiota on brain and behaviour.. (2012) 13 701-712. DOI: 10.1038/nrn3346 18. D’Amato A., Di Cesare Mannelli L., Lucarini E., Man A. L., Le Gall G., Branca J. J. V.. Faecal microbiota transplant from aged donor mice affects spatial learning and memory via modulating hippocampal synaptic plasticity- and neurotransmission-related proteins in young recipients.. (2020) 8. DOI: 10.1186/s40168-020-00914-w 19. Dagmara M., Aleksandra A., Artur D., Dorota F., Andrzej C., Łukasz G.. Resilience and cognitive biases mediate the relationship between early exposure to traumatic life events and depressive symptoms in young adults.. (2019) 254 26-33. DOI: 10.1016/j.jad.2019.05.008 20. Dalile B., Van Oudenhove L., Vervliet B., Verbeke K.. The role of short-chain fatty acids in microbiota-gut-brain communication.. (2019) 16 461-478. DOI: 10.1038/s41575-019-0157-3 21. De Raedt R., Koster E. H. W.. Understanding vulnerability for depression from a cognitive neuroscience perspective: a reappraisal of attentional factors and a new conceptual framework.. (2010) 10 50-70. DOI: 10.3758/CABN.10.1.50 22. Disner S. G., Beevers C. G., Haigh E. A. P., Beck A. T.. Neural mechanisms of the cognitive model of depression.. (2011) 12 467-477. DOI: 10.1038/nrn3027 23. Duman R. S., Malberq J., Thome J.. Neural plasticity to stress and antidepressant treatment.. (1999) 46 1181-1191. DOI: 10.1016/S0006-3223(99)00177-8 24. Foster J. A., Baker G. B., Dursun S. M.. The relationship between the gut microbiome-immune system-brain axis and major depressive disorder.. (2021) 12. DOI: 10.3389/fneur.2021.721126 25. Fröhlich E. E., Farzi A., Mayerhofer R., Reichmann F., Jačan A., Wagner B.. Cognitive impairment by antibiotic-induced gut dysbiosis: analysis of gut microbiota-brain communication.. (2016) 56 140-155. DOI: 10.1016/j.bbi.2016.02.020 26. Ge K.. The transition of Chinese dietary guidelines and food guide pagoda.. (2011) 20 439-446. PMID: 21859664 27. Ge M., Sun X., Huang Z.. Correlation between parenting style by personality traits and mental health of college students.. (2022) 2022. DOI: 10.1155/2022/6990151 28. Gobin C. M., Banks J. B., Fins A. I., Tartar J. L.. Poor sleep quality is associated with a negative cognitive bias and decreased sustained attention.. (2015) 24 535-542. DOI: 10.1111/jsr.12302 29. Gong J., Chan R. C. K.. Early maladaptive schemas as mediators between childhood maltreatment and later psychological distress among Chinese college students.. (2018) 259 493-500. DOI: 10.1016/j.psychres.2017.11.019 30. Gubert C., Kong G., Renoir T., Hannan A. J.. Exercise, diet and stress asmodulators of gut microbiota: implications for neurodegenerative diseases.. (2020) 134. DOI: 10.1016/j.nbd.2019.104621 31. Guo S. R., Sun W. M., Liu C.. Structural validity of the pittsburgh sleep quality index in chinese undergraduate students.. (2016) 7. DOI: 10.3389/fpsyg.2016.01126 32. Huang C. W., Kang Y. W., Zhang B. R., Feng Q. Y., Liu Z. H., Zhang F.. Mediating role of soldiers’ fear of evaluation in negative cognitive bias and society anxiety.. (2021) 42 563-567. DOI: 10.16781/j.0258-879x.2021.05.0563 33. Jiang J., Xie S. R., Li L., Li J., Liu Y. B., Xu W. J.. Characterstics of negative cognitive processing bias and relationship with depression in a troop stationed on plateau.. (2017) 39 1891-1895. DOI: 10.16016/j.1000-5404.201704041 34. Jing Y., Han S., Chen J., Lai Y., Cheng J., Li F.. Gut microbiota and urine metabonomics alterations in constitution after Chinese medicine and lifestyle intervention.. (2021) 49 165-1193. DOI: 10.1142/S0192415X21500567 35. Joormann J., Gotlib I. H.. Emotion regulation in depression: relation to cognitive inhibition.. (2010) 24 281-298. DOI: 10.1080/02699930903407948 36. Joormann J., Vanderlind W. M.. Emotion regulation in depression: the role of biased cognition and reduced cognitive control.. (2014) 2 402-421. DOI: 10.1177/2167702614536163 37. Kawano Y., Edwards M., Huang Y., Bilate A. M., Araujo L. P., Tanoue T.. Microbiota imbalance induced by dietary sugar disrupts immune-mediated protection from metabolic syndrome.. (2022) 185 3501-3519.e20. DOI: 10.1016/j.cell.2022.08.005 38. Kennedy P. J., Cryan J. F., Dinan T. G., Clarke G.. Kynurenine pathway metabolism and the microbiota-gut-brain axis.. (2017) 112(Pt. B) 399-412. DOI: 10.1016/j.neuropharm.2016.07.002 39. Koster E. H. W., De Lissnyder E., Derakshan N., De Raedt R.. Understanding depressive rumination from a cognitive science perspective: the impaired disengagement hypothesis.. (2011) 31 138-145. DOI: 10.1016/j.cpr.2010.08.005 40. Kwak Y. T., Yang Y., Koo M. S.. Depression and cognition.. (2016) 15 103-109. DOI: 10.12779/dnd.2016.15.4.103 41. LeMoult J., Gotlib I. H.. Depression: a cognitive perspective.. (2019) 69 51-66. DOI: 10.1016/j.cpr.2018.06.008 42. LeMoult J., Humphreys K. L., Tracy A., Hoffmeister J. A., Ip E., Gotlib I. H.. Meta-analysis: exposure to early life stress and risk for depression in childhood and adolescence.. (2020) 59 842-855. DOI: 10.1016/j.jaac.2019.10.011 43. Li L., Han L., Gao F. Q., Chen Y. M.. The role of negative cognitive processing bias in the relation between shyness and self-disclosure among college students.. (2021) 29 1260-1265. DOI: 10.16128/j.cnki.1005-3611.2021.06.028 44. Li X., Wang J.. Abnormal neural activities in adults and youths with majordepressive disorder during emotional processing: a meta-analysis.. (2021) 15 1134-1154. DOI: 10.1007/s11682-020-00299-2 45. Ling Z., Zhu M., Yan X., Cheng Y., Shao L., Liu X.. Structural and functional dysbiosis of fecal microbiota in Chinese patients with Alzheimer’s disease.. (2021) 8. DOI: 10.3389/fcell.2020.634069 46. Liu S., Gao J., Zhu M., Liu K., Zhang H. L.. Gut microbiota and dysbiosis in Alzheimer’s disease: implications for pathogenesis and treatment.. (2020) 57 5026-5043. DOI: 10.1007/s12035-020-02073-3 47. Ma H., Zhang B., Hu Y., Wang J., Liu J., Qin R.. Correlation analysis of intestinal redox state with the gut microbiota reveals the positive intervention of tea polyphenols on hyperlipidemia in high fat diet fed mice.. (2019) 67 7325-7335. DOI: 10.1021/acs.jafc.9b02211 48. Martin K. S., Azzolini M., Ruas J. L.. The kynurenine connection: how exercise shifts muscle tryptophan metabolism and affects energy homeostasis, the immune system, and the brain.. (2020) 318 C818-C830. DOI: 10.1152/ajpcell.00580.2019 49. Mathews A., MacLeod C.. Cognitive vulnerability to emotional disorders.. (2005) 1 167-195. DOI: 10.1146/annurev.clinpsy.1.102803.143916 50. Michels N., Van de Wiele T., De Henauw S.. Chronic psychosocial stress and gut health in children: associations with calprotectin and fecal short-chain fatty acids.. (2017) 79 927-935. DOI: 10.1097/psy.0000000000000413 51. Neufeld K. M., Kang N., Bienenstock J., Foster J. A.. Reduced anxiety-like behavior and central neurochemical change in germ-free mice.. (2011) 23 255-264,e119. DOI: 10.1111/j.1365-2982.2010.01620.x 52. Notebaert A., Ferriby A., Sinning A.. Differences in item statistics between first order and second order laboratory practical examination questions.. (2020) 34. DOI: 10.1096/fasebj.2020.34.s1.06757 53. Noworyta K., Cieslik A., Rygula R.. Neuromolecular underpinnings of negative cognitive bias in depression.. (2021) 10. DOI: 10.3390/cells10113157 54. Oleskin A. V., Shenderov B. A.. Neuromodulatory effects and targets of the SCFAs and gasotransmitters produced by the human symbiotic microbiota.. (2016) 27 30971-30984. DOI: 10.3402/mehd.v27.30971 55. Parks D. H., Tyson G. W., Hugenholtz P., Beiko R. G.. STAMP: statistical analysis of taxonomic and functional profiles.. (2014) 30 3123-3124. DOI: 10.1093/bioinformatics/btu494 56. Poole L., Cao H. W., Yu Y., Li M.. Resilience and cognitive bias in Chinese male medical freshman.. (2017) 8. DOI: 10.3389/fpsyt.2017.00158 57. Proctor C., Thiennimitr P., Chattipakorn N., Chattipakorn S. C.. Diet, gut microbiota and cognition.. (2017) 32 1-17. DOI: 10.1007/s11011-016-9917-8 58. Qu W., Liu S., Zhang W., Zhu H., Tao Q., Wang H.. Impact of traditional Chinese medicine treatment on chronic unpredictable mild stress-induced depression-like behaviors: intestinal microbiota and gut microbiome function.. (2019) 10 5886-5897. DOI: 10.1039/c9fo00399a 59. Ritchie G., Strodl E., Parham S., Bambling M., Cramb S., Vitetta L.. An exploratory study of the gut microbiota in major depression with anxious distress.. (2022) 1 595-604. DOI: 10.1016/j.jad.2022.10.001 60. Rozenman M., Amir N., Weersing V. R.. Performance-based interpretation bias in clinically anxious youths: relationships with attention, anxiety, and negative cognition.. (2014) 45 594-605. DOI: 10.1016/j.beth.2014.03.009 61. Sanchez A., Duque A., Romero N., Vazquez C.. Disentangling the interplay among cognitive biases: evidence of combined effects of attention, interpretation and autobiographical memory in depression.. (2017) 41 829-841. DOI: 10.1007/s10608-017-9858-5 62. Sanz Y.. Effects of a gluten-free diet on gut microbiota and immune function in healthy adult humans.. (2010) 1 135-137. DOI: 10.4161/gmic.1.3.11868 63. Sarkar A., Harty S., Lehto S. M., Moeller A. H., Dinan T. G., Robin I. M. D.. The microbiome in psychology and cognitive neuroscience.. (2018) 22 611-636. DOI: 10.1016/j.tics.2018.04.006 64. Savignac H. M., Tramullas M., Kiely B., Dinan T. G., Cryan J. F.. Bifidobacteria modulate cognitive processes in an anxious mouse strain.. (2015) 287 59-72. DOI: 10.1016/j.bbr.2015.02.044 65. Skonieczna-Żydecka K., Grochans E., Maciejewska D., Szkup M., Schneider-Matyka D., Jurczak A.. Faecal short chain fatty acids profile is changed in polish depressive women.. (2018) 10 1939-1952. DOI: 10.3390/nu10121939 66. Smith P. M., Howitt M. R., Panikov N., Michaud M., Gallini C. A., Bohlooly-Y M.. The microbial metabolites, short-chain fatty acids, regulate colonic Treg cell homeostasis.. (2013) 341 569-573. DOI: 10.1126/science.1241165 67. Valles-Colomer M., Falony G., Darzi Y., Tigchelaar E. F., Wang J., Tito R. Y.. The neuroactive potential of the human gut microbiota in quality of life and depression.. (2017) 4 623-632. DOI: 10.1038/s41564-018-0337-x 68. Van den Bergh N., Hoorelbeke K., De Raedt R., Koster E. H. W.. Remediation of depression-related cognitive impairment: cognitive control training as treatment augmentation.. (2018) 18 907-913. DOI: 10.1080/14737175-2018.1537783 69. Vidaña A. G., Forbes C. N., Gratz K. L., Tull M. T.. The influence of posttraumatic stress disorder and recurrent major depression on risk-taking propensity following trauma script exposure among patients with substance use disorders.. (2020) 102. DOI: 10.1016/j.addbeh.2019.106181 70. Vuong H. E., Yano J. M., Fung T. C., Hsiao E. Y.. The microbiome and host behavior.. (2017) 40 21-49. DOI: 10.1146/annurev-neuro-072116-031347 71. Wang S., Ishima T., Qu Y., Shan J., Chang L., Wei Y.. Ingestion of Faecalibaculum rodentium causes depression-like phenotypes in resilient Ephx2 knock-out mice: a role of brain-gut-microbiota axis via the subdiaphragmatic vagus nerve.. (2021) 1 565-573. DOI: 10.1016/j.jad.2021.06.006 72. Wei C., Xu J. X., Gang L., Liu T., Guo X. L., Wang H. J.. Ethanol extract of propolis prevents high-fat diet-induced insulin resistance and obesity in association with modulation of gut microbiota in mice.. (2020) 130. DOI: 10.1016/j.foodres.2019.108939 73. Westfall S., Caracci F., Estill M., Frolinger T., Shen L., Pasinetti G. M.. Chronic stress-induced depression and anxiety priming modulated by gut-brain-axis immunity.. (2021) 12. DOI: 10.3389/fimmu.2021.670500 74. Wu S. J., Xu Z. D., Zhang Y. T., Liu X. F.. Relationship among psychological capital, coping style and anxiety of Chinese college students.. (2019) 54 264-268. DOI: 10.1708/3281.32545 75. Yan X. F., Zhang R., Feng Z. Z.. Development of negative cognitive processing bias questionnaire.. (2017) 39 2329-2334. DOI: 10.16016/j.1000-5404.201707160 76. Yang Y. X., Wang X. L., Leong P. M., Zhang H. M., Yang X. G., Kong L. Z.. New Chinese dietary guidelines: healthy eating patterns and food-based dietary recommendations.. (2018) 27 908-913. DOI: 10.6133/apjcn.072018.03 77. Zhang X., Zhang B., Peng B., Wang J., Hu Y., Wang R.. Different dose of sucrose consumption divergently influences gut microbiota and PPAR-γ/MAPK/NF-κB pathway in DSS-induced colitis mice.. (2022) 14. DOI: 10.3390/nu14132765 78. Zheng P., Zeng B., Zhou C., Liu M., Fang Z., Xu X.. Gut microbiome remodeling induces depressive-like behaviors through a pathway mediated by the host’s metabolism.. (2016) 21 786-796. DOI: 10.1038/mp.2016.44 79. Zhuang Z. Q., Shen L. L., Li W. W., Fu X., Zeng F., Gui L.. Gut microbiome is altered in patients with Alzheimer’s disease.. (2018) 68 1337-1346. DOI: 10.3233/JAD-180176 80. Zung W. W.. A self-rating depression scale.. (1965) 12 63-70. PMID: 14221692 81. Zung W. W.. A rating instrument for anxiety disorders.. (1971) 12 371-379. PMID: 5172928
--- title: Antimicrobial protein REG3A regulates glucose homeostasis and insulin resistance in obese diabetic mice authors: - Patrick Gonzalez - Alexandre Dos Santos - Marion Darnaud - Nicolas Moniaux - Delphine Rapoud - Claire Lacoste - Tung-Son Nguyen - Valentine S. Moullé - Alice Deshayes - Gilles Amouyal - Paul Amouyal - Christian Bréchot - Céline Cruciani-Guglielmacci - Fabrizio Andréelli - Christophe Magnan - Jamila Faivre journal: Communications Biology year: 2023 pmcid: PMC10015038 doi: 10.1038/s42003-023-04616-5 license: CC BY 4.0 --- # Antimicrobial protein REG3A regulates glucose homeostasis and insulin resistance in obese diabetic mice ## Abstract Innate immune mediators of pathogen clearance, including the secreted C-type lectins REG3 of the antimicrobial peptide (AMP) family, are known to be involved in the regulation of tissue repair and homeostasis. Their role in metabolic homeostasis remains unknown. Here we show that an increase in human REG3A improves glucose and lipid homeostasis in nutritional and genetic mouse models of obesity and type 2 diabetes. Mice overexpressing REG3A in the liver show improved glucose homeostasis, which is reflected in better insulin sensitivity in normal weight and obese states. Delivery of recombinant REG3A protein to leptin-deficient ob/ob mice or wild-type mice on a high-fat diet also improves glucose homeostasis. This is accompanied by reduced oxidative protein damage, increased AMPK phosphorylation and insulin-stimulated glucose uptake in skeletal muscle tissue. Oxidative damage in differentiated C2C12 myotubes is greatly attenuated by REG3A, as is the increase in gp130-mediated AMPK activation. In contrast, Akt-mediated insulin action, which is impaired by oxidative stress, is not restored by REG3A. These data highlight the importance of REG3A in controlling oxidative protein damage involved in energy and metabolic pathways during obesity and diabetes, and provide additional insight into the dual function of host-immune defense and metabolic regulation for AMP. Antimicrobial protein REG3A improved insulin sensitivity by combating oxidative stress and activating AMPK in skeletal muscle in HFD-induced obese mice overexpressing human REG3A in the liver. ## Introduction Metabolic syndrome is a serious health condition that affects 25-$35\%$ of American and European adults1,2, and is characterized by abdominal obesity, insulin resistance, hypertension, and hyperlipidemia resulting from the increasing prevalence of obesity. This contributes to the pathogenesis of multiple chronic diseases, such as type 2 diabetes mellitus (T2D), non-alcoholic fatty liver disease (NAFLD), cardiovascular diseases, stroke, and cancers. Patients with metabolic disorders may also face a higher risk for various failures in the body’s defenses, including an increased risk of infection, changes in immune monitoring and modulation, lack of activation of T cells that have become resistant to insulin in a context of low-grade inflammation and oxidative stress. Immune-endocrine interrelationships represent a major circuitry for maintaining proper homeostasis and energy metabolism, this cross-organ communication being highly conserved throughout evolution. A large body of evidence indicates functional links between metabolic control with the innate immune system, the primary and most ancient defense against infection3. For example, the activation of NF-κB by a pathogen sensing system (Toll-like receptors) or the tumor necrosis factor (TNF) signaling blocks systemic insulin action and glucose metabolism in response to nutrient deprivation or sepsis as an adaptive response to reallocate energy use to the affected organs whose acute requirements need to be met4,5. Antimicrobial peptides and proteins (AMPs), which are fundamental effectors of the innate immune response by exerting microbicidal activities against multiple types of microorganisms, are components of the immune-metabolic crosstalk. They include defensins, C-type lectins of the REG3 family, and cathelicidins. The expression of a number of them is finely modulated by nutrients and metabolic hormones, such as the adipokines visfatin and leptin6–8. Some AMPs such as cathelicidin LL37/CRAMP and alpha-melanocyte-stimulating hormone α-MSH have been implicated in energy and glucose regulation by stimulating insulin production or inhibiting hepatic gluconeogenesis, respectively9–12. This suggests a paradigm shift towards a dual function of innate host defense and metabolism for AMP, which, unlike pattern recognition receptors, are secreted molecules that can act at a distance from their production site to ensure metabolic adaptation of the organism to the environment and nutrition. In the intestine, human regenerating islet-derived protein 3-alpha (REG3A), also known as hepatocarcinoma-intestine-pancreas/pancreatic associated protein or HIP/PAP, is a secreted 16 kDa carbohydrate-binding protein that functions as an AMP against Gram-positive bacteria13. Antibacterial activity requires N-terminal proteolytic processing by trypsin or elastase of intestinal REG3 proteins generating a 15 kDa protein. This cleavage does not occur for the extra-intestinal and circulating forms of REG3A14–16. REG3A also helps shape the symbiotic interactions of the human host and the gut microbiota through an antioxidant capacity to scavenge toxic reactive oxygen species (ROS), thus preserving the bacterial symbionts and the mucosal barrier17. In tissues, the action of REG3 goes beyond its activity towards microbes. REG3 genes are expressed in a limited number of tissues and cell types under physiological conditions, and in greater numbers during tissue damage and inflammation particularly in response to interleukin-6 (IL6) family cytokines, where they promote tissue repair and healing. These undoubtedly positive actions of REG3 proteins have been reported in vivo in several internal tissues and in the skin17–26. The mechanisms by which REG3A initiates intracellular signaling remain poorly defined, and involve, depending on the context, tissue and cell type, carbohydrate ligand recognition, carbohydrate-protein interaction, or protein-protein interaction. Several potential surface receptors and membrane-associated proteins (such as glycoprotein 130 (gp130), fibronectin, epidermal growth factor receptor (EGFR), exostosin-like glycosyltransferase 3 (EXTL3), syndecan2, annexin3) have been proposed to interact with REG3A and transmit the REG3A signal into cells27–33. This report deals with the action of REG3A on type 2 diabetes mellitus and obesity which remains largely unknown, being mainly studied in type 1 diabetes in humans and experimental animals, as a potential inducer of protection and regeneration of pancreatic beta cells34–46. Our findings reveal a previously unknown mechanism by which REG3A regulate glucose homeostasis by combating oxidative stress and activating AMPK in skeletal muscle, and suggest an additional level of regulation of metabolism by lectin-mediated innate immunity. ## Hepatic expression of human REG3A decreases adiposity in aged mice We have begun to explore the metabolic effects of REG3A by characterizing the metabolic phenotype of healthy and diabetic C57BL/6 mice overexpressing human REG3A. We used previously generated homozygous transgenic C57BL/6 mice expressing human REG3A in hepatocytes (TG-REG3A) under the control of the mouse albumin gene promoter47 and fed them a regular diet ($51.7\%$ carbohydrates, $21.4\%$ protein, $5.1\%$ fat). Transgenic hepatocytes secreted REG3A into blood vessels through the basolateral membranes, resulting in basal serum REG3A levels of 242 ± 14 ng/mL in females ($$n = 50$$) and 316 ± 17 ng/mL in males ($$n = 50$$). This allowed us to study the action of highly concentrated circulating REG3A on metabolic homeostasis beyond the impact on liver as this extrahepatic action is currently largely unknown. Although weight differences existed between TG-REG3A mice and wild-type (WT) mice, they were modest and were observed in females rather than males. In female TG-REG3A mice, weight gain was lower than in WT mice at 4, 6, and 18 months of age (21.6 ± 3.4 g vs. 22.8 g ± 1.6 g, $$P \leq 5.10$$–4; 24.7 ± 2.6 g vs. 25.2 ± 6.6 g, $$P \leq 0.032$$; 28.2 ± 3.8 g vs. 30.1 ± 4.4 g; $$P \leq 7.10$$–3, respectively). No significant difference was observed at 12 and 24 months (Supplementary Figure 1). TG-REG3A male mice showed no significant difference in weight compared with WT male mice until 18 months of age, after which a difference was observed at 24 months of age (32.2 g ± 5.4 g vs. 33.8 ± 3.9 g, respectively; $$P \leq 0.04$$) (Fig. 1a). We selected male REG3A-transgenic mice ($$n = 8$$) and control mice ($$n = 8$$) at 4 months of age and studied their phenotypes over time. At this time, body weight was 29.0 ± 0.6 g and 28.0 ± 0.6, respectively ($$P \leq 0.34$$). We monitored body composition by Magnetic Resonance Imaging (MRI) and found that TG-REG3A mice showed a reduction in fat mass compared with WT mice at all times in the 2-year analysis (−$21\%$, −$59\%$, −$40\%$, −$42\%$, and −$43\%$ at 4, 6, 12, 18, and 24 months, respectively) with a concomitant increase in lean mass at 6, 12, 18 months (Fig. 1b, c). These differences in body composition were not explained by reduced food intake (Fig. 1d and Supplementary Fig. 2a). In fact, cumulative food intake was even higher (+$20\%$) in TG-REG3A than WT mice at 6 and 24 months (Fig. 1d and Supplementary Fig. 2a). In addition, bomb calorimetry measurements showed no defect in intestinal energy and macronutrient absorption in TG-REG3A mice compared with WT mice (Fig. 1e-h). Lastly, the spontaneous locomotor activity and energy expenditure were also similar between the two groups (Fig. 1i, j). Indirect calorimetry measurements of oxygen consumed (VO2) and carbon dioxide produced (VCO2) yielded a higher respiratory exchange ratio (RER; VCO2/VO2) in TG-REG3A mice than in WT mice at several time points, indicating that carbohydrates rather than fat are oxidized during the dark (active) phase for energy production. In particular, we observed an accelerated transition from beta-oxidation to glycolysis. During the normal light phase, the RER of TG-REG3A mice increased more rapidly than that of WT mice, suggesting that TG-REG3A mice preferentially use carbohydrate rather than fat during this period (Fig. 1k, l and Supplementary Fig. 2b, c). This difference persisted at 24 months and indicates the ability of transgenic mice to use glucose rather than fat during the fasting period throughout their lives. Fig. 1Suppression of body fat accumulation in transgenic mice expressing human REG3A in the liver.a Changes in body weight in male REG3A-transgenic mice (REG3A, $$n = 73$$) and littermate wild-type (WT) controls ($$n = 53$$) fed with ad libitum standard chow (2830 kcal/kg). The difference between WT and REG3A groups is not significant (P = NS) except 24 months (two-sided Wilcoxon rank sum test). Red dots: 16 randomly selected 4-month old mice (8 WT, 8 REG3A) for further metabolic characterization. b–l Metabolic phenotype of the 16 selected mice mentioned above. b Fat and (c) lean mass in WT ($$n = 8$$) and REG3A mice ($$n = 8$$) over time. d–j Feeding behavior, activity and metabolic performance of the same 8 WT and 8 REG3A mice at the age of 6 months. d Cumulative food intake over a 3.5-day period. e Food efficiency as calculated by the ratio of body weight gain over food ingested in kJ. f Feces produced per day. g Intestinal absorption as the ratio of amount of food intake over fecal output within a 24 h period. h Energy content in feces measured with a bomb calorimeter. i Locomotor activity averaged over two 12 h period. j Energy expenditure measured in a calorimetric chamber. k Respiratory quotient (VCO2/VO2) during a day-night cycle. l Fatty acid (FA) oxidation measured by respirometry over a day-night cycle. Data are averages ± SEM. ∗$p \leq 0.05$; ∗∗$p \leq 0.01$; ∗∗∗$p \leq 0.005$ by Anova test (a–c) or two-way repeated measures Anova (d, j–l) both followed by Tukey post hoc multiple comparison test. NS or no statistical indication, no significance. ## Increased sensitivity to insulin in REG3A transgenic mice fed a regular diet To determine whether the above reduction in body fat and increase in glucose oxidation in TG-REG3A mice are accompanied by specific metabolic traits, we examined glucose levels under basal conditions and in insulin and glucose tolerance tests. Levels of fasting blood glucose was lower in 6-month old TG-REG3A chow-fed mice compared to age-matched WT (WT, 100 ± 6 mg/dL; REG3A, 77 ± 3 mg/dL) and equivalent in the two groups of mice at 24 months (WT, 115 ± 5 mg/dL; REG3A, 120 ± 7 mg/dL, Fig. 2a). In response to a glucose challenge (2 g/kg), 6-month-old TG-REG3 mice exhibit mild glucose intolerance, but 24-month-old mice do not (Fig. 2b); their peak glucose values are higher and their clearance times longer compared with control mice (248 ± 10 vs. 190 ± 5 mg/dL at 30 min, and 230 ± 12 vs. 177 ± 8 mg/dL at 45 min, $p \leq 0.01$). At 60 min, glucose levels were no longer different from those of control mice. This was associated with a decrease in insulin output at the 30-min time point after glucose gavage. It should be noted that 24-month-old TG-REG3A mice with normal glucose tolerance showed a reduction in insulin output at times 0, 15, and 30 min after gavage, in contrast to the increase observed in aged WT mice (Fig. 2c). Plasma C-peptide was lower at baseline in TG-REG3A mice than in WT mice at 6 and 24 months. In both mice and at both ages, it was elevated in response to glucose, indicating similar responsiveness of their endocrine pancreas (Supplementary Fig. 3a). According to the homeostasis model assessment of insulin resistance (HOMA-IR), TG-REG3 mice showed lower HOMA scores than WT mice, indicating that TG-REG3 mice improved, and maintained their better insulin sensitivity with advanced age (Fig. 2d). Of note, 24-month-old WT mice show a trend toward basal hyperinsulinemia compared with 6-month-old WT mice ($$p \leq 0.17$$), which is reflected by the increase in the HOMA-IR score (Fig. 2c, d). This pattern of basal hyperinsulinemia in aged WT mice combined with relatively preserved insulin sensitivity (Fig. 2e) is consistent with previously reports in elderly human subjects48 and 18-month-old C57bl/6 mice49. In an insulin tolerance test (0.75 units/kg), TG-REG3 mice became more hypoglycemic after insulin injection than WT mice at 6 and 24 months (Fig. 2e). We analyzed the contribution of liver, adipose tissue, and skeletal muscle in improving insulin sensitivity in TG-REG3A mice by assessing the level of Akt phosphorylation at Ser473 in these tissues. Compared with WT mice, increased phosphorylation of Akt on Ser473 was observed in the liver of TG-REG3A mice at 6 months (but not at 24 months) of age, whereas no differences were found in subcutaneous white adipose tissue (WAT), tibialis anterior, and soleus muscles (Supplementary Fig. 3b, c).Fig. 2REG3A transgenic mice fed a standard diet are hypersensitive to insulin. Glucose homeostasis was assessed in 6- and 24-month-old REG3A transgenic mice (REG3A) and wild-type (WT) mice ($$n = 8$$ for each group, same as Fig. 1) fed ad libitum with standard chow. a Fasting blood glucose. b Blood glucose curves under oral glucose tolerance tests OGTT (baseline at time $t = 0$ right before glucose gavage). Right: area under the curve of OGTT. c Insulin levels during same OGTT as Fig. 2b. d Insulin sensitivity estimated by HOMA-IR (homeostatic model assessment for insulin resistance). e Blood glucose curves under insulin tolerance tests ITT (baseline at time $t = 0$ right before insulin injection). Right: area under the curve of ITT. Asterisks indicate significant differences between WT and REG3A at 6 months of age. Data are averages ± SEM. ∗$p \leq 0.05$; ∗∗$p \leq 0.01$ by Anova (a, right b, c, d, right e) or repeated measures Anova (left b, left e) both followed by a post hoc test. NS or no statistical indication, no significance. ## Improved insulin sensitivity and reduced hyperinsulinemia in REG3A transgenic mice fed a high fat diet To determine whether obese TG-REG3A mice had improved whole-body glucose homeostasis, they were fed a high-fat diet (HF260; 5505 kcal/kg) for 8 weeks, in which $60\%$ of energy came from fat and $27\%$ from carbohydrate. We found that females were more resistant to HFD-induced obesity and metabolic alteration than males, in agreement with previous report50. Glucose and insulin tolerance were better in females than in males ($$P \leq 0.004$$ for OGTT and $$P \leq 3.10$$–4 for ITT; two-way Anova). This better insulin sensitivity of females is comparable regardless the genotype is WT or transgenic (Supplementary Fig. 4). We also found that systemic insulin sensitivity did not correlate with circulating REG3A levels in either the male or female groups, suggesting that changes in circulating levels do not affect the extent of insulin sensitivity control. We then proceeded with male mice with clearly established metabolic disease. The amount of food ingested was similar in WT and REG3A transgenic mice throughout the diet (Fig. 3a). Weight gain was also similar in WT and REG3 transgenic mice with weight increasing by 55-$60\%$ during this period (44.9 g ± 2.7 g in WT vs. 42.7 g ± 9.7 g in REG3A-TG) (Fig. 3b). Basal fasting blood glucose after two months of high-fat feeding did not differ between the two groups of mice (220 ± 7 mg/dL in WT; 224 ± 8 mg/dL in REG3A-TG). In response to oral glucose, the blood glucose escalation curves were largely similar between the two groups of mice, indicating similar glucose tolerance under HF260 (Fig. 3c). The obese REG3A-TG mice showed a decrease in basal fasting C-peptide compared with the obese WT mice (Fig. 3c–e). In response to glucose, plasma insulin and C-peptide were elevated in obese WT mice compared with basal levels before gavage. In contrast, obese REG3A-TG mice showed lower stimulated insulin and C-peptide (Fig. 3d, e). Thus, REG3A-TG obese mice require $50\%$ less insulin than WT obese mice to achieve similar glucose homeostasis, suggesting that REG3A may protect against the deleterious effects of a high-fat diet on insulin sensitivity and insulin secretion. When the HOMA-IR was calculated, there was no significant difference between the two mouse lines, although a small trend of decrease was noted in the REG3A group (Fig. 3f). In an insulin tolerance test, REG3A-TG obese mice exhibited a greater hypoglycemic response to exogenous insulin than WT obese mice, consistent with the increased insulin sensitivity they exhibit when fed a standard diet (Fig. 3g). Preserved glucose homeostasis in REG3A-TG obese mice occurs without any change in body weight (Fig. 3b). Furthermore, comparable levels of Akt phosphorylation (Ser473) in the liver, tibialis anterior, soleus, and WAT of obese REG3A-TG and control mice seem to exclude that the mechanism underlying the enhanced metabolic phenotype of obese REG3A-TG mice is an increase in Akt-dependent insulin signaling (Supplementary Figure 5a-d). The morphology of the pancreatic beta islets appeared normal in the transgenic mice. We detected no difference between islets from TG-REG3A and WT obese mice after immunostaining for insulin, suggesting that the lower insulin secretion under basal conditions and during glucose tolerance test in high-fat-fed REG3A-TG mice would not be related to altered pancreatic beta cell function (Fig. 3h). These results establish that REG3A overexpression is associated with reduced insulin secretion in response to hyperglycemic challenge and improved insulin sensitivity under high-fat feeding. Fig. 3Improved insulin sensitivity and reduced insulin secretion in obese REG3A transgenic mice fed a high fat diet. Four-month-old REG3A transgenic mice (REG3A) and wild-type (WT) mice ($$n = 8$$ REG3A, $$n = 9$$ WT, new cohorts different from Figs. 1 and 2) were fed a high-fat, butter-enriched diet (5505 kcal/kg) for 8 weeks and then analyzed. a cumulative food intake. b Body weight gain. c Blood glucose curves under oral glucose tolerance tests OGTT. Right: area under the curve of OGTT. d–f Blood tests (d) Blood insulin, e C-peptide concentrations and f Insulin resistance indices (HOMA-IR). g Blood glucose curves under insulin tolerance tests ITT. Right: area under the curve of ITT assays. ( $$n = 7$$ per group). h Representative images of insulin staining on pancreas sections from REG3A transgenic and WT mice fed standard (CD) or a high-fat (HFD) diet, and quantification of insulin positive areas. Scale bar: 200 µm. Data are averages ± SEM. ∗$p \leq 0.05$ by Anova test (d, e, h), one-way repeated measures Anova (a, b, left c, left g) both followed by post-hoc test, Wilcoxon rank sum test (right c, right g). NS or no statistical indication, no significance. ## REG3A reduces hyperglycemia and insulin resistance in ob/ob mice We next hypothesized that the administration of a full-length recombinant REG3A protein (rcREG3A, ALF5755) would improve metabolic homeostasis and insulin resistance and tested that hypothesis first in the ob/ob mice, a congenital leptin-deficient model of severe obesity, hyperinsulinemia and insulin resistance. We have previously shown that ALF5755 is a biologically active protein that binds to oligo- and polysaccharides, but not to monosaccharides17, and exhibits cytoprotective and antioxidant activity promoting tissue regeneration in vivo20,23,51,52. Twelve-week-old ob/ob mice were infused with a daily dose of 9 µg or 43 µg of rcREG3A (or an equivalent volume of buffer) using an osmotic pump-based delivery system for 28 days (Fig. 4a). This resulted in serum REG3A levels of 100 ± 28 and 250 ± 44 ng/mL for 9 µg or 43 µg of administered rcREG3A per day, respectively (Fig. 4b). Both quantities of infused rcREG3A were chosen to fit within the range of circulating REG3A levels as found in vivo in TG-REG3A mice. Body weight and fat mass gains were similar between ob/ob mice infused with vehicle or rcREG3A (Fig. 4c, d). However, significant differences were observed in several blood metabolic parameters, including a reduction in fasting concentrations of nonesterified (free) cholesterol, triglycerides, neutral lipids, and glucose by $29\%$, $74\%$, $23\%$, and $30\%$, respectively (Fig. 4e, f). There was no difference in hepatic lipid concentration and liver steatosis score (Supplementary Fig. 6a, b). For mice injected with 9 µg of rcREG3A, glucose concentration decreased from 283 ± 24 mg/dL on day 1 to 193 ± 6 mg/dL 11 days later and, for mice injected with 43 µg of rcREG3A, from 316 ± 32 mg/dL to 234 ± 19 mg/dL. Glucose levels remained stable afterwards for mice injected with both rcREG3A doses (Fig. 4f). In an insulin tolerance test, the insulin response was significantly improved in ob/ob mice given either dose of rcREG3A compared with the vehicle group, demonstrating that administration of rcREG3A over a 4-week period greatly improves systemic insulin sensitivity (Fig. 4g, h). With the same levels of glucose clearance whether or not the mice were treated with rcREG3A, there was a significant decrease in circulating insulin levels at 15 min and a trend at 30 min after a glucose tolerance test, suggesting better control of insulin resistance in those treated with rcREG3A (Fig. 4i–k). This was illustrated by a stabilization of HOMA-IR values in rcREG3A-treated mice, and a further increase in the vehicle group, indicating attenuation of insulin resistance by rcREG3A (Fig. 4l). There was no significant difference in insulin sensitivity (HOMA-IR) of ob/ob mice receiving a dose of 9 µg or 43 µg per day at D11 or D25 after rcREG3A infusion (D11: $$P \leq 0.39$$; D25: $$P \leq 0.063$$), so we continued with the larger dose of rcREG3A. Overall, ob/ob mice treated with rcREG3A showed an increase in the hypoglycemic effect of exogenous insulin as well as a decrease in circulating insulin levels under glucose-stimulated conditions. These results conclusively demonstrate that insulin resistance was significantly attenuated by chronic REG3A treatment in both diet-induced obesity and that resulting from the ob/ob model. Fig. 4Ob/ob mice given a recombinant REG3A protein are less hyperglycemic and insulin resistant. Three-month-old Ob/Ob mice were given a daily subcutaneous dose of 9 µg or 43 µg of a recombinant human REG3A protein (rcREG3A) for 28 days ($$n = 10$$ per group). Control ob/ob mice received an equivalent volume of buffer (vehicle). Glucose homeostasis was assessed on days 11 and 25 after starting rcREG3A or buffer treatment. a Diagram of the experimental protocol. OGTT: oral glucose tolerance test. ITT: insulin tolerance test. b–e Measurement of indicated markers on day 29 (end of study). b Serum concentrations of REG3A. c Body weight. d Adipose tissue content. e Serum levels of lipids. f Fasting blood glucose on days 11 and 25 after onset of treatment. g, h Blood glucose curves under insulin tolerance tests on (g) day 14 and (h) day 28. I, j Blood glucose curves under glucose tolerance tests on (i) day 11 and (j) day 25. k Blood insulin concentrations and l insulin resistance indices (HOMA-IR) after oral glucose intake on day 25. Data are averages ± SEM. ∗$p \leq 0.05$; ∗∗$p \leq 0.01$; ∗∗∗$p \leq 0.005$ by Anova test (b–f, k, l, right panels of g, h, i and j) or one-way repeated measures Anova (left panels of g, h, i and j) both followed by post-hoc test. NS or no statistical indication, no significance. ## REG3A reduces glucose tolerance and dyslipidemia in prediabetic mice To examine whether the insulin-sensitizing action of rcREG3A is effective in non-genetically obese mouse models with insulin resistance, we tested the effect of rcREG3A on glucose metabolism in C57BL/6 WT mice with HFD-induced prediabetes (Fig. 5a). After 10 weeks of HFD feeding (HF235; 4655 kcal/kg; $37.5\%$ carbohydrate, $27.5\%$ fat, $17\%$ protein), WT mice given 43 µg per day of rcREG3A for 28 days showed improvement in metabolic syndrome and associated traits (changes in lipid profile, leptin, glucose tolerance) not seen in mice given the vehicle (Fig. 5). Weight gain and adiposity were equivalent in mice treated with rcREG3A or vehicle (Fig. 5b, c). In contrast, significant reductions in blood lipid and leptin levels of about $30\%$ were observed in mice given rcREG3A which is consistent with the results obtained in ob/ob mice, as well as a trend toward reduced hepatic steatosis (Fig. 5d–f). While the HF235 diet did not alter fasting blood glucose compared with the chow diet, rcREG3A-treated mice had lower blood glucose levels than control mice (Fig. 5g). In response to glucose challenge, rcREG3A significantly improved glucose tolerance and reduced glucose-stimulated insulin secretion by −$56\%$ at 15 min and −$45\%$ at 30 min (Fig. 5h, i). This differs from the impaired glucose tolerance and threefold increase in insulin secretion required to adapt to the glucose challenge in HF235-fed mice receiving a vehicle (Fig. 5i). We detected no difference between islets from mice fed HF235 and treated, or not, with rcREG3A after immunostaining for insulin (Supplementary Fig. 7). Such improvements in glucose homeostasis in rcREG3A-treated obese mice was not accompanied by Akt phosphorylation on Ser473 in the peripheral tissues analyzed (tibialis anterior, liver) (Supplementary Fig. 8). These observations indicate that rcREG3A reduces the metabolic consequences of an unbalanced fatty diet, including the correction of dyslipidemia, by decreasing insulin secretion in response to glucose at a prediabetic stage. This suggests that REG3A may enhance insulin action through an Akt-independent mechanism and thus contribute to curb the development of insulin resistance. Fig. 5Obese prediabetic mice given a recombinant REG3A protein showed improved glucose homeostasis and increased glucose uptake by skeletal muscle. After 10 weeks of HFD (4655 kcal/kg), WT mice received 43 µg per day of a recombinant REG3A protein (rcREG3A) subcutaneously for 28 days ($$n = 9$$). Mice on a standard diet (chow, CTD) also received rcREG3A for the same period ($$n = 5$$). HFD-fed ($$n = 10$$) and CTD-fed control mice ($$n = 5$$) received an equivalent volume of buffer (vehicle). a Diagram of the experimental protocol. OGTT: oral glucose tolerance test. ITT: insulin tolerance test. b–k Measurement of indicated markers at the end of rcREG3A or buffer treatment. b Body weight over time. c Body fat and lean mass. d Histological score of liver steatosis. e Serum levels of lipids. f Serum levels of leptin. g Fasting blood glucose. h Blood glucose curves under oral glucose tolerance tests OGTT. Right: area under the curve of OGTT. i blood insulin concentrations during glucose tolerance test. j Blood glucose curves during insulin tolerance test. Right: area under the curve of ITT. k Hyperinsulinemic-euglycemic clamp in mice fed a chow (CTD, $$n = 10$$) or a high-fat (HFD, $$n = 14$$) diet and treated subcutaneously with 43 µg per day of a recombinant REG3A protein (rcREG3A) or an equivalent volume of buffer for 28 days. GUR: Whole-body glucose utilization rate. GIR: glucose infusion rate; EGP: endogenous glucose production. l Tissue sensitivity to insulin estimated by the quantification of glucose uptake in the indicated tissues. Data are averages ± SEM. ∗$p \leq 0.05$; ∗∗$p \leq 0.01$; ∗∗∗$p \leq 0.005$ by Anova test (c, e, f, g, right h, i, k, l) or one-way repeated measures Anova (left h, left j) both followed by post-hoc test, Wilcoxon rank sum test (right j). NS or no statistical indication, no significance. ## REG3A increases glucose uptake in glycolytic skeletal muscles, but not in oxidative skeletal muscles and WAT To provide insights into the mechanisms that contribute to the control of glucose metabolism by rcREG3A, and to identify the tissue(s) where insulin action is enhanced, we performed hyperinsulinemic-euglycemic clamp (3mU/minKg) studies in WT mice with HF235-induced prediabetes or fed a chow diet and given rcREG3A or vehicle for 28 days. We showed that the rcREG3A-treated group had a glucose turnover rate comparable to that of the vehicle group in both the standard and high-fat diets. The rate of glucose infusion to maintain euglycemia was similar in mice given rcREG3A or a vehicle in response to HF235, indicating equivalent whole-body insulin sensitivity in both setups. Endogenous glucose production was also the same in all groups indicating no particular effect of rcREG3A on this parameter (Fig. 5k). Labeled 2 deoxy-glucose was injected at the end of the clamp in order to measure glucose uptake in skeletal muscles either oxidative (soleus) or glycolytic (tibialis anterior) as well as fat pad (Fig. 5l). As shown in Fig. 5, the tibialis anterior of HF235 fed mice that received rcREG3A showed a significant increase in insulin-stimulated glucose uptake compared with vehicle-treated mice, whereas the latter developed skeletal muscle insulin resistance during HF235. In contrast, adipose tissue and soleus showed glucose uptake comparable to that of control mice (Fig. 5l). This suggests that REG3A regulates glucose homeostasis and insulin sensitivity in a tissue-specific manner, this phenomenon is presumably related to the fast muscle glucose clearance in the targeted skeletal fiber muscles. ## REG3A attenuates oxidative protein damage and activates AMPK in skeletal muscle We, and others, have previously demonstrated that REG3A is a coupling agent between innate immunity and organ homeostasis through multiple functionalities, such as antimicrobial, antioxidant, survival factor, and regulator of inflammatory response in various damaged tissues, where REG3A promotes tissue repair and regeneration1,2,17,20,21,23,53. Serum levels of tumor necrosis factor α (TNF-α), IL-6, and interferon-γ were measured in mice overexpressing REG3A or infused with rcREG3A to define their relationship with REG3A-triggered metabolic changes. Elevations in cytokine levels were not substantial in response to HFD feeding or leptin deficiency, and did not differ significantly in the presence of the transgene or recombinant REG3A (except for IL6 in mice fed HF235 and given rcREG3A) (Supplementary Fig. 9). Because REG3A is a ROS scavenger, we wondered whether REG3A’s ability to counteract oxidative damage, particularly in muscle, could help improve glucose homeostasis, which is chronically impaired by high-fat diets and obesity. Protein carbonylated levels were measured in tissues and muscle cells by derivatization of carbonyl groups with 2,4-dinitrophenylhydrazine (DNPH) or biotin-hydrazine, followed by immunodetection with a DNPH-specific antibody (Fig. 6b) or by affinity capture (Fig. 6j). We found higher levels of protein carbonylation in the tibialis muscle of ob/ob mice than in that of HF235-fed mice (the latter showing low levels of oxidative damage comparable to that seen during a standard diet), an effect significantly attenuated by rcREG3A in ob/ob mice (Fig. 6a). Similarly, quantification of protein carbonylation in differentiated C2C12 myotubes yielded 4-fold higher values under glucose oxidase (GO)-induced oxidative stress than in the absence of GO. The accumulation of oxidized proteins in C2C12 cells was significantly reduced by rcREG3A, indicating antioxidant activity of REG3A in skeletal muscle cells and tissue (Fig. 6b). To determine whether reduction of muscle oxidative damage by REG3A improves glucose control, we tested basal and insulin-induced phosphorylation of Akt in oxidized C2C12 cells. GO completely inhibited insulin-induced phosphorylation of Akt on Ser473 in C2C12 cells and rcREG3A did not reverse the oxidative stress-induced impairment of insulin action (Fig. 6c). We turned to AMP-activated protein kinase (AMPK), another key regulator of metabolism, and showed that the level of phosphorylation of AMPK (i.e., the active form of AMPK) was higher in rcREG3A-treated C2C12 cells than in untreated cells under baseline and oxidative stress conditions (Fig. 6 d). The effect of rcREG3A on AMPK activation was dose-dependent in C2C12 myotubes and a trend was observed in primary myocytes (Fig. 6 e, f). Importantly, we found a positive association between rcREG3A and AMPK phosphorylation, and a neutral effect on Akt activation in the tibialis of HF235-fed prediabetic and ob/ob mice (Fig. 6 g, h). Next, we investigated how the increase in AMPK activity was enhanced by REG3A and the specific contributions of glycoprotein 130 (gp130) signaling. Reg3 family members are linked to components of several signaling pathways, depending on the tissue, cell, and disease type. These include the gp130 receptor complex, which is one of the triggers of muscle AMPK activation through IL6 and ciliary neurotrophic factor (CNTF) signaling3,54. On the other hand, IL-6 family cytokines (CNTF, LIF, IL6) that use the gp130 pathway are positive signals triggering changes in gene expression of Reg3b, the murine homolog of REG3A, in many cell types4,5,55–57. Finally, the gp130-JAK2-STAT3 cascade is reported to mediate the effects of Reg3b6–8,58. We found that REG3A-mediated AMPK activation is gp130 dependent, as gp130 silencing significantly suppressed AMPK activation in C2C12 cells stimulated with rcREG3A (Fig. 6i). We then hypothesized that the carbonyl levels of gp130 and the insulin receptor might be affected by REG3A given the significant reduction in muscle protein carbonylation associated with REG3A in vivo and in vitro. We used affinity capture and streptavidin-linked beads, followed by immunodetection with antibodies specific for gp130 or the insulin receptor, to assess the amount of these oxidized membrane proteins. Addition of rcREG3A to GO-exposed C2C12 cell culture significantly reduces carbonylation of gp130 but not of the insulin receptor (Fig. 6j), suggesting that REG3A, by protecting gp130 from oxidative damage, releases the gp130-dependent AMPK activation pathway. Fig. 6The antioxidant rcREG3A increases AMPK activation in the skeletal muscle.a Quantification of normalized protein carbonylation levels in tibialis muscle of wild-type (WT) and REG3A transgenic (REG3A) mice fed a standard diet (CD, $$n = 8$$), mice fed a high-fat diet (HFD, $$n = 4$$), and ob/ob mice receiving 43 µg per day of recombinant REG3A protein (rcREG3A, $$n = 4$$–6) for 28 days. Control mice fed the HFD diet and ob/ob received an equivalent volume of buffer (vehicle). b Left: Coomassie blue staining: protein loading control. Middle: Anti-DNP immunoblots displaying total carbonylated proteins from C2C12 myoblasts in the absence (−) or presence (+) of recombinant REG3A protein (rcREG3A) and exposed or not to glucose oxidase (GO)-induced oxidative stress. Eluate: carbonylated protein enrichment using biotin tagging and avidin affinity chromatography. Right: densitometry quantification of immunoblots performed on affinity-enriched carbonylated proteins (Eluate lane). c, d Representative immunoblots using the indicated metabolic markers in extracts of C2C12 myoblasts treated or not with insulin and recombinant REG3A protein (rcREG3A) without or with oxidative stress (GO). AKT and AMPK immunoblots are shown as protein loading controls. Below: densitometry quantification of the immunoblots. Three independent experiments. e–f Anti-phospho-AMPK (pAMPK) immunoblotting in mouse C2C12 cells (e) and primary myocytes (f) for different doses of rcREG3A. Bar chart: densitometric analysis. g, h Representative immunoblots for the indicated markers in muscle extracts (g) in the high fat diet (HFD)-fed WT mice of Fig. 5 and (h) in the ob/ob mice of Fig. 4 treated with rcREG3A or buffer (vehicle). Each lane represents 1 mouse. Bar chart: Densitometric analysis. i Representative immunoblots for the indicated markers in C2C12 myoblasts with or without rcREG3A treatment and with downregulated gp130 expression (gp130 siRNA) or not (Control siRNA). Bar chart: densitometric analysis. Two independent experiments. j Total and carbonylated levels (oxidized) of GP130 and insulin receptor (InsR alpha) in C2C12 myoblasts under oxidative stress conditions treated with a recombinant REG3A protein (rcREG3A) or buffer (vehicle). Two independent experiments. Means of three independent experiments ± SEM. ∗$p \leq 0.05$; ∗∗$p \leq 0.01$ by Anova test followed by post-hoc test (b–f, i) or Student’s t test (a, g, h, j). NS or no statistical indication, no significance. ## Changes in REG3A expression in metabolic organs of obese and diabetic patients We explored the human relevance of our findings by investigating whether altered glucose metabolism in obese/diabetic patients could be correlated with aberrant REG3A expression. We analyzed endogenous REG3A expression using publicly available large-scale microarray datasets from human tissues of individuals with different metabolic phenotypes. As expected, REG3A mRNA expression is mainly restricted to the intestine and pancreas, and low or nearly absent in liver, adipose tissue, and skeletal muscle under physiological conditions (Supplementary Fig. 10a). We selected metabolic disorder datasets deposited in the Gene Expression Omnibus (GEO)59–63 and examined REG3A gene expression in major metabolic tissues (i.e. liver, pancreas, WAT, vastus lateralis muscle). Significant changes were observed in WAT and mostly muscle between insulin-sensitive and diabetic states, but not in liver and pancreas. A significant decrease in REG3A expression in muscle was observed in the insulin-resistant and diabetic conditions compared with the insulin-sensitive condition, suggesting that REG3A in skeletal muscle may be involved in glucose-insulin homeostasis (Fig. 7a). The public availability of skeletal muscle transcriptomic data obtained over time after metabolic surgery (2, 12, 24, 52 weeks; GSE 135066) allowed us to study any changes in REG3A gene expression. While REG3A expression was low in the muscle before surgery, it increased significantly upon resolution of the disease before returning to basal levels at 52 weeks (Fig. 7b). The temporal variation in REG3A coincided with transient changes in the expression of many genes, including those related to insulin signaling, AMPK signaling and fatty oxidation, and mitochondrial function in skeletal muscle (Fig. 7c–e). The level and timing of REG3A induction were not shared with the vast majority of molecules in the antimicrobial peptide and protein class to which REG3A belongs (Fig. S10b–d). Only REG3 gamma (REG3G) and Defensin Alpha 5 (DEFA5) showed similar expression dynamics to REG3A during the disease resolution period, whereas the expression of almost the entire set of AMPs analyzed was negatively modulated ($$n = 25$$) or not modulated at all ($$n = 56$$) after bariatric surgery. This in silico analysis suggests a strong link between improved muscle insulin sensitivity and REG3 proteins. Fig. 7Endogenous upregulation of REG3A in skeletal muscle is linked to improved metabolism after bariatric surgery in obese individuals.a, b REG3A mRNA expression level assessed from transcriptomes of indicated tissues in patients with metabolic disease. a NGT: normal glucose tolerance (Hb1Ac less than $6\%$); IGT: impaired glucose tolerance (HbA1c between 6 and $6.5\%$); T2D: type 2 diabetes (HbA1c greater than $6.5\%$); b IS: insulin-sensitive; IR insulin-resistant. IS et IR were defined by measuring the glucose disposal rate during an hyperinsulinemic euglycemic clamp in Wu et al. Endocrinology, et al. 2007. c–f Molecular changes over time in the vastus lateralis skeletal muscle of 16 patients who underwent bariatric surgery. c REG3A mRNA expression. d Gene enrichment of “Insulin signaling pathway” in KEGG. e Gene enrichment of “Activated AMPK stimulates fatty acid oxidation in muscle” in Reactome. f Spearman correlation between REG3A expression and enrichment score of the “mitochondrial uncoupling pathway” in Reactome. ∗$p \leq 0.05$; ∗∗$p \leq 0.01$; ∗∗∗$p \leq 0.005$ by Anova test followed by post-hoc test. NS or no statistical indication, no significance. ## Discussion Mediators of the immune response help control energy metabolism, and disruption of one or more of these inter-organ signals can have dramatic effects on the pathogenesis of insulin homeostasis in obesity and type 2 diabetes mellitus64. Chronic systemic and local low-grade inflammation and oxidative stress mediators are important pathogenic factors in this setting, suggesting that control of inflammatory/oxidative defects may help resolve insulin resistance and diabetes. Indeed, evidence in human and animal models indicate that metabolic oxidative stress is a trigger in the pathogenesis of insulin resistance, diabetes and its complications. In diabetes and obesity, glucolipotoxicity among other processes prompts an overproduction of ROS overwhelming the scavenging capabilities of metabolic tissues. Increase in non-scavenged intermediates leads to oxidative damage. Protein carbonylation that irreversibly impairs protein structure and function is an example of such damage. Carbonylation of insulin receptor substrates 1 and 2 by ROS contributes to a loss of insulin responsiveness, resulting in compensatory insulin hypersecretion. High ROS concentrations and subsequent overactivation of ROS-responsive pathways enhance oxidative damage and lower insulin efficacy65–67. In addition, elevated ROS production caused by mono-oleoyl-glycerol stimulates basal insulin secretion in pancreatic ß cells68. In turn, the accumulation of oxidized, misfolded and aggregated proteins exacerbates oxidative stress levels, insulin resistance and diabetes. We and others have shown that acute-phase REG3 protein expression increases after tissue injury. The resulting elevated blood concentration of REG3A has been proposed as a possible prognostic biomarker for many human acute and chronic diseases, including pancreatitis, inflammatory bowel disease, graft versus host disease, and ischemic stroke69–72. A growing body of research has shown that overexpression of REG3A promotes the resolution of inflammation and tissue repair in different experimental settings by controlling key repair mechanisms such as wound repair, immune cell recruitment, and proliferation. In the liver, REG3A pairs up with extracellular matrix components of inflamed tissue, and the resulting high local concentration of REG3A in the damaged area makes it a particular effective antioxidant that contributes to better-controlled regeneration after tissue injury51. In this study, we establish links between oxidative stress regulation and metabolic regulation though REG3A. We report here that increase of REG3A, both as a transgene and as a recombinant protein, controls elevated basal insulin levels during aging, high-fat feeding, and in obese ob/ob mice and prevents glucose-induced hyperinsulinemia. REG3A reduces hyperglycemia and dyslipidemia in prediabetes and type 2 diabetes in obese rodents. A basis for the increased whole-body insulin sensitivity was found to be the effectiveness of REG3A in reducing oxidative stress and activating AMPK signaling in skeletal muscle. This phenomenon is not associated with changes in body fat levels for the recombinant REG3A protein. In contrast, in REG3A transgenic mice, improvement of metabolic phenotype was associated with changes in primary adiposity. Because ectopic lipid accumulation is clearly associated with insulin-resistant states, proper maintenance of insulin homeostasis in aging REG3A mice may be a by-product of the marked reduction in body fat in these mice. It is conceivable that the observed effect of REG3A protein on glucose uptake in glycolytic skeletal muscle and not in fat helps counteract excessive triglyceride storage in adipocytes and explains the decrease in fat mass and increase in lean mass in REG3A transgenic mice over time. In addition, it should be noted that glycolytic muscles (tibialis anterior) are more vulnerable to sarcopenia than oxidative muscles (soleus) possibly related to a higher generation of ROS in glycolytic muscles during aging73–75. The attenuation of muscle oxidative damage by REG3A stems from its scavenging activity against two deleterious ROS, namely, the superoxide anion and hydroxyl radical, both of which are known to be abundantly produced during chronic metabolic disease76,77. Similar ROS scavenging activity involved in the protection of murine hematopoietic stem cells has been reported for other lectins78. Because insulin resistance in skeletal muscle is the major disruption of insulin action in type 2 diabetes79,80, increasing muscle action of REG3A may contribute to the decrease of insulin resistance in diabetes and obesity. The decrease in membrane carbonylated protein content by REG3A in skeletal muscle strongly suggests that the significant changes in insulin sensitivity in obese/diabetic mice administered REG3A are related to the antioxidant activity of REG3A in skeletal muscle. Such a REG3A-mediated mechanism of action most likely promotes induction of gp130-AMPK signaling and thus bypasses the typical impairment of insulin signaling, leading to minimization of the amount of insulin needed to control blood glucose and reduce hyperinsulinemia. Similar effects of reduced insulin hypersecretion and dyslipidemia are observed with REG3A transgenic mice during aging and obesity after feeding with a high-fat diet. While the recombinant REG3A protein bypasses the AKT-dependent insulin pathway by playing on the AMPK pathway in muscle, we found indications that the REG3A transgene and consequently the resulting overexpression of circulating and intrahepatocyte REG3A may play on the enhancement of alternative Akt and AMPK signaling pathways in the liver (Supplementary Fig. 11). These results suggest that REG3A contribute to a strong control of glucose homeostasis while significantly reducing hyperinsulinemia and improving insulin sensitivity. Together, this study shows that REG3A significantly reduces oxidative protein damage in skeletal muscle and improves glucose homeostasis and insulin sensitivity by activating AMPK in obese/diabetic mice. Through this mechanism, REG3A could potentially preserve the fitness of metabolic organs such as adipose tissue, muscle, liver and pancreas. Our results suggest that a therapeutic approach based on REG3A could break the vicious circle between hyperinsulinemia and altered signaling by oxidative stress, which, if not controlled, contributes to disease progression. In this regard, we demonstrate that muscle expression of REG3 proteins is altered after bariatric surgery and is associated with improved muscle insulin sensitivity. The direct systemic activity on insulin resistance, including its impact on muscle glucose uptake and fat/lean mass reduction, as well as the activity in modulating the gut microbiome17, suggest that REG3A may be a therapeutic approach complementary to existing treatments for type 2 diabetes, such as metformin and GLP1 agonists. This combined, with the good clinical safety profile of REG3A (ALF-5755) observed in a phase I study involving healthy volunteers and a phase II study in patients with acute liver failure23,51, are major arguments for translational exploration of REG3A in patients with type 2 diabetes. The question of whether REG3A significantly alters metabolism toward weight increase is critical and could undermine the potential clinical uses of REG3A. This was rigorously examined in this study, particularly because previous work reported a spontaneous increase in body weight in mice expressing REG3A in hepatocytes at ages 6-27 weeks81. We cast serious doubt on the validity of the Secq et al. study regarding the obesogenic capacity of REG3A by pointing out an internal inconsistency in the data82 that the authors’ response unfortunately did not resolve83. To clarify this issue, we followed the weight curve for 24 months of a large number of male and female REG3A transgenic mice and wild-type controls fed a regular standard diet. 125 REG3A transgenic mice (50 males, 75 females) and 102 WT mice (50 males, 52 females) were studied. We show that weight changes are similar between TG-REG3A and WT males over time, and that modest weight differences are observed between transgenic and WT females. Feeding REG3A transgenic and WT mice fatty diets resulted in comparable levels of induced obesity among them. These findings indicate that overexpression of REG3A does not result in an aberrant increase in body weight and are in agreement with those obtained with Reg3b and Reg3g, the murine homologs of REG3A84. Overall, this study suggests that REG3A, by mitigating oxidative stress, which is enhanced by metabolic syndromes triggered by Western diets, preserves the function of selective metabolic pathways involved in glucose homeostasis, namely, in this study, gp130-dependent AMPK activation. The antioxidant action of REG3A on the gp130-AMPK pathway does not preclude such a mechanism from triggering the action of other putative REG3A receptors, even though we did not find endogenous expression of EXTL3 in C2C12 muscle cells. Enhanced muscle glucose uptake through REG3A signaling would help reduce insulin requirements in obese, diet-induced insulin-resistant mice. This attenuated hyperinsulinism would ultimately control the progression of insulin resistance and diabetes. Based on the results obtained in mice and in silico, and from a clinical perspective, this study suggests that the administration of a recombinant human REG3A protein may be a valuable approach to reduce peripheral insulin resistance and its associated comorbidities in overweight and type 2 diabetic patients. ## Mouse models Animal studies were performed in compliance with the institutional and European Union guidelines for laboratory animal care and approved by the Ethics Committee of CE2A-03-CNRS-Orléans (Accreditation N°01417.01). The number of Mus musculus mice used complied with institutional ethical rules and was consistent with common practice in the fields of metabolism studies. Conventional mice were produced and housed in the CNRS SEAT and Institut André Lwoff animal care facilities (Paris-Saclay University, Villejuif). All the REG3A-transgenic mice had circulating human REG3A levels within the 100–500 ng/mL range. Wild-type and REG3A-transgenic mice had the same C57BL/6 N genetic background. The diets used in this study were as follows: standard chow A03 (2830 kcal/kg; $51.7\%$ carbohydrates, $5.1\%$ fat, $21.4\%$ protein), HF235 (4655 kcal/kg; $37.5\%$ carbohydrates, $27.5\%$ fat, $17\%$ protein) or HF260 (5505 kcal/kg; $60\%$ fat, $27\%$ carbohydrates) (SAFE, Augy, France). After 8 weeks on the diet, the mice were first tested for insulin or glucose tolerance and then briefly anesthetized with isoflurane to implant a filled minipump on the right back. Twelve-week-old leptin-deficient male Ob/Ob mice (B6.Cg-Lepob/J) were obtained from Charles River. For rcREG3A administration, an Alzet miniosmotic pump (model 2004) was implanted subcutaneously in 13-week-old mice delivering 9 or 43 µg/day of human recombinant REG3A for 4 weeks. Control groups were infused with an equivalent volume of buffer. Changes in body composition (i.e. fat and lean mass) were determined by quantitative magnetic resonance (EchoMRI 900; Echo Medical Systems, Houston, Texas, USA) in aging mice, in chow and HFD-fed mice and in ob/ob mice on the last day of rcREG3A or buffer treatment (day 29). Body composition was expressed as a percentage of body weight. ## Recombinant human REG3A protein (rcREG3A) The rcREG3A protein (ALF5755) is a biological drug that corresponds to the addition of one amino‐terminal methionine to the sequence of the secreted (i.e., lacking the 26–amino acid signal sequence) form of the human endogenous REG3A (HIP/PAP) (NP_620355). It was produced in Escherichia coli, purified to ≥$99\%$ and released in batches in compliance with the clinical grade manufacturing process by PX’Therapeutics (Grenoble, France). The batches of ALF5755 used were supplied by Alfact Innovation. ## Glucose and insulin tolerance tests Insulin tolerance tests were performed in mice fasted for 5 h and injected subcutaneously with a rapid-acting insulin (0.5 units/kg or 0.75 units/kg body weight of C57Bl/6N or ob/ob mice respectively; Novorapid flexpen 100 U/mL, NovoNordisk A/S). A glucose tolerance test was performed in mice that were fasted overnight (18 h) and then received a glucose solution at 2 g/kg body weight via oral gavage (glucose $30\%$ CMD Lavoisier). Tail blood was sampled at the indicated time points for glucose and insulin measurements. Blood was immediately centrifuged, and the plasma was separated and stored at −20 °C until assayed. Glucose levels were measured with a glucometer (Glucofix mio, Menarini Diagnostics). Insulin sensitivity was assessed by calculating the slope of the decrease in glucose concentration as a function of time after insulin injection. Surrogate index of insulin resistance (HOMA-IR) was calculated from fasting blood glucose and plasma insulin concentrations as follows: HOMA = (GxI)/405(with glucose (G) expressed as mg/dL and insulin (I) expressed as µU/ml; ref. 85. ## Indirect calorimetry The metabolic screening was done using indirect calorimetry. Animals were analyzed for whole energy expenditure (kcal/h), oxygen consumption and carbon dioxide production (VO2 and VCO2, where V is the volume), respiratory exchange ratio (RER = VCO2/VO2), food intake (g) and locomotor activity (counts/hour) using calorimetric cages with bedding, food and water (Labmaster, TSE Systems GmbH, Germany). Gas ratio was determined using an indirect open-circuit calorimeter, which monitored O2 and CO2 concentrations by volume at the inlet ports of a tide cage with an airflow of 0.4 L/min, with regular comparisons to an empty reference cage. Whole energy expenditure was calculated according to the Weir equation, using respiratory gas exchange measurements. The flow was previously calibrated with O2 and CO2 mixture of known concentrations (Air Liquide, S.A. France). Animals were individually housed in a cage with lights on from 7 a.m. to 7 p.m. and an ambient temperature of 22 ± 1 °C. All animals were acclimated to their cages for 48 h before experimental measurements. Data regarding food and water consumption were collected every 40 min, and all ambulatory movements recorded during the entire experiment, with the aid of an automated online measurement system combining highly sensitive feeding and drinking sensors and an infrared beam-based activity monitoring system. Gas and movement detection sensors were operational during both light and dark phases, allowing for continuous recording. Animals were monitored for body weight and composition at the beginning and end of the experiment. Data analysis was carried out with Excel XP using extracted raw values of VO2 consumption, VCO2 production (ml/h), and energy expenditure (kcal/h). Subsequently, each value was normalized by whole lean tissue mass extracted from the EchoMRI analysis. ## Calorimetric bomb A calorimetric bomb and a fecalogram were performed in the Functional Coprology Laboratory of the Pitié-Salpétrière hospital. Feces were collected over a period of 5 days and 4 nights; feed intake (FI) and fecal output (FO) were recorded. The net absorption balance was calculated as FI-FO, and the percentage of intestinal absorption as 100(FI-FO)/FI. Dry matter content (dry weight) in g per 100 g of feces (%) was determined by weighing the feces before and after dehydration overnight in an incubator at +70 °C. Total combustion of feces under oxygen allowed calculation of energy leakage via measurement of the amount of heat released. The determination of fecal lipids was performed by a titration technique after extraction of lipids by organic solvents (van de Kamer et al., 1949), and that of proteins by the determination of atomic nitrogen by HPLC. These assays were used to calculate fecal nitrogen and lipid flows and, by subtraction, the carbohydrate flow. ## Quantification of islet insulin content The pancreas was fixed in $4\%$ paraformaldehyde overnight, followed by dehydration in graded ethanol solution and embedding in paraffin wax. For each embedded pancreas, five nonconsecutive 5-μm-thick sections (50-μm apart) were mounted on slides and subjected to hematoxylin/eosin staining and anti-insulin primary antibody labeling (RRID:AB_631835; 1 µg/mL). The labeled sections were scanned with Scan 3D Histech P250 Flash III. Quantification of the area of insulin-producing islets was related to the total area of pancreatic sections using QuPath software (version 0.3). ## Hyperinsulinemic-euglycemic clamp 5-week-old C57BL/6 N mice were fed chow ($$n = 10$$) or high-fat ($$n = 14$$) diet for 10 weeks, and an Alzet miniosmotic pump (model 2004) was implanted subcutaneously in 11-week-old mice delivering 43 µg/day of human rcREG3A ($$n = 12$$) or buffer treatment ($$n = 12$$) for the last 4 weeks. Then, the hyperinsulinemic-euglycemic clamp experiments were carried out for conscious mice after a 5 h fast. Insulin perfusions in the jugular vein were at 3mU/minKg body weight and 15 g/dL glucose at variable rate to maintain euglycemia (chow diet: rcREG3A 139.0 ± 14.0 mg/dl [$$n = 5$$], buffer 130.2 ± 8.9 mg/dl [$$n = 5$$]; high-fat diet: rcREG3A 129.7 ± 12.3 [$$n = 7$$], buffer 125.2 ± 8.9 mg/dl [$$n = 7$$]). Glucose infusion rates (GIRs) were calculated at the end of a 80-min clamp. To determine [3H]-glucose and 2-[14C]-DG concentrations, plasma samples were deproteinized with ZnSO4 and Ba(OH)2, dried, resuspended in water, and counted in scintillation fluid for detection of 3H and 14C. Tissue 2-[14C]-DG-6-phosphate (2-DG-6-P) content was determined in homogenized samples that were subjected to an ion-exchange column to separate 2-DG-6-P from 2-[14C]-DG. ## ELISA assays The plasma concentrations of REG3A, leptin, insulin, and C-peptide were measured in duplicates using the PancrePAP (Dynabio), mouse leptin ELISA (Crystal Chem; RRID:AB_2722664), mouse ultrasentitive insulin ELISA (ALPCO; RRID:AB_2792981) and mouse C-peptide ELISA (Crystal Chem Inc; Catalog # 90050) kits, respectively, according to the manufacturer’s instructions. Plasma concentrations of IL-6, IFN-γ, and TNF-α were measured in triplicate using the ELLA microfluidic platform (Bio-Techne, Minneapolis, MN, USA) according to the manufacturer’s instructions. ## Neutral lipid quantification Lipids corresponding to 1 mg of liver tissue or 10 µL of plasma of HFD-fed and ob/ob mice were extracted according to Bligh and Dyer86 in dichloromethane/methanol/water (2.5:2.5:2.1, v/v/v), in the presence of internal standards: 4 µg stigmasterol, 4 µg cholesteryl heptadecanoate, 8 µg glyceryl trinonadecanoate. The dichloromethane phase was evaporated to dryness and dissolved in 30 µl of ethyl acetate. 1 µl of lipid extract was analyzed by gas-liquid chromatography on a FOCUS Thermo Electron system using a Zebron-1 Phenomenex fused silica capillary columns (5 m × 0.32 mm i.d, 0.50 µm film thickness). The oven temperature was programmed from 200 °C to 350 °C at a rate of 5 °C per min and the carrier gas was hydrogen (0.5 bar). The injector and the detector were at 315 °C and 345 °C, respectively. ## Cell culture Mouse C2C12 myoblasts were grown in a DMEM medium (Gibco) containing inactivated $10\%$ fetal calf serum (FBS), 2 mM Glutamine and $1\%$ penicillin/streptomycin. To induce differentiation, cells were allowed to reach 80–$90\%$ confluence in complete medium and then incubated in DMEM with $2\%$ of horse serum. Cells were usually used for experiments 5 days after differentiation. In RNAi experiments, siRNA transfection was performed at day 4 after induction of differentiation using the Viromer Blue transfection kit (Viromer) and a murine GP130- targeting siRNA (ThermoFisher Scientific, RNA ID: MSS236904). The cells were used 24 h after transfection. In experiments performed under oxidative stress conditions, differentiated C2C12 cells were incubated with 10mU of glucose oxidase in fresh DMEM medium with $2\%$ horse serum for 3 h to generate H2O2 as a by-product of metabolism of medium glucose, and then the cells were lysed. Primary myocytes were maintained in proliferation medium consisting of DMEM-F12 Glutamax (Gibco) supplemented with $20\%$ of inactivated FBS and $2\%$ UltroserG (Pall Life Sciences). The medium was replaced every 2 days. Primary myocytes were seeded at a density of 30,000 cells/cm2 on matrigel-coated plates (BD biosciences) in the proliferation medium for 6 h before removing the medium for differentiation medium (DMEM-F12 Glutamax supplemented with $2\%$ inactivated horse serum). Myocytes were kept for 4 days to achieve differentiation. For glucose oxidase treatment, myotubes were serum-starved and incubated for 3 h with glucose oxidase, then switched to serum-free medium containing 100 nmol/l insulin (30 min). ## Immunoblotting Whole-cell lysates were made in ice-cold lysis buffer supplemented with phosphatase and protease inhibitors (50 mM Tris-Cl pH7.4, 150 mM NaCl, $1\%$ NP40, $0.25\%$ sodium deoxycholate, 1 mmoL/L Na3VO4, 20 mmoL/L NaF, 1 μg/mL aprotinin, 10 μg/mL pepstatin, 10 μg/mL leupeptin, 1 μM phenylmethylsulfonyl fluoride). Tissue lysates were obtained from flash-frozen tissue, which was lysed by bead beating in lysis buffer. Protein quantification was performed using the Bio-Rad protein assay kit and bovine serum albumin. 30 µg of protein was resolved in a $12\%$ polyacrylamide gel in Tris-Glycine SDS buffer (Invitrogen), before being electrotransferred onto nitrocellulose membranes (Whatman, Dominique Dutscher). Membranes were blocked in $5\%$ non-fat milk in $0.1\%$ Tween 20 Tris-buffered saline for 1 h and probed with the primary antibodies. AMPK (RRID:AB_915794; dilution 1:1000), Thr172P-AMPK RRID:AB_2169396; dilution 1:1000), AKT (RRID:AB_329827; dilution 1:1000), Ser473P-AKT RRID:AB_2315049; dilution 1:1000) and tubulin (RRID:AB_2288042; dilution 1:1000) antibodies were purchased from Cell Signaling Technologies. The gp130 antibody was obtained from Santa Cruz Biotechnology (RRID:AB_647629; 0.2 µg/mL). ## Protein lysate for analysis of oxidative changes Differentiated C2C12 cells were lysed in deoxycholate lysis buffer (DOC) containing $2\%$ sodium deoxycholate, benzonase and protease inhibitors (cOmplete Protease Inhibitor Cocktail, Roche), 20 mM Tris-HCl pH 8.8, 2 mM EDTA, 2 mM iodoacetamide. Homogenates were centrifuged at 10,000 g for 10 min at 4 °C. DOC-insoluble fractions were solubilized in SDS lysis buffer containing $1\%$ sodium dodecyl sulfate SDS, 20 mM Tris-HCl pH 8.8, 2 mM EDTA, protease inhibitors, 2 mM iodoacetamide. ## Protein carbonylation The carbonyl groups of a DOC-insoluble protein fraction were derivatized to 2,4-dinitrophenylhydrozone (DNPH) by reaction with 2,4-dinitrophenylhydrazine in $0.1\%$ trifluoroacetic acid solution, incubated 15 min in the dark at room temperature. The coupling reactions were stopped by a 1 M Tris solution containing $30\%$ glycerol. Samples were resolved on $10\%$ SDS-PAGE and transferred to nitrocellulose membranes (Whatman, Dominique Dutscher) by electroblotting. Membranes were blocked in $5\%$ non-fat milk in $0.1\%$ Tween 20 Tris-buffered saline for 1 h and probed with a primary antibody against the dinitrophenyl group (DNP) (Millipore, RRID:AB_10850321; dilution 1:50) and then with HRP-conjugated goat anti-rabbit antibody (RRID:AB_390191; dilution 1:2000) to reveal carbonylated proteins using a chemiluminescent reagent (Pierce). To determine the amount of total protein, the same fraction of insoluble DOC protein was resolved on a $10\%$ acrylamide gel for Coomassie blue staining. The amounts of carbonylated and total proteins were determined by densitometry using Image J software; the results were presented as the ratio of carbonylated to total proteins. ## Oxidation of the GP130 and insulin receptor DOC-insoluble protein fractions were precipitated with trichloroacetic acid (TCA), protein pellets were washed with ethanol/ethyl acetate solution ($\frac{50}{50}$) and resuspended 100 mM sodium acetate pH 6.0, 20 mM NaCl, $1\%$ SDS and the carbonyl groups were derivatized by reaction with biotin hydrazide for 2 h before reduction with sodium cyanoborohydride for 30 min at 4 °C. After derivatization, the proteins were precipitated with TCA, the protein pellets were washed with ethanol/ethyl acetate solution ($\frac{50}{50}$) and resuspended in SDS lysis buffer containing $1\%$ SDS, 20 mM Tris-HCl pH 8.8, 2 mM EDTA, protease inhibitors, and 2 mM iodoacetamide. An aliquot of each sample was frozen for later used as input for Western blotting while the remainder was diluted in equilibration buffer containing 100 mM potassium phosphate, 150 mM NaCl, 400 mM ammonium sulfate pH 7.2. Biotin-derived proteins were captured on a 50 µL streptavidin mutein matrix (Roche), non-specific binding was removed by washing in 100 mM potassium phosphate, 150 mM NaCl pH 7.2, and biotinylated proteins were eluted in 1X Laemmli loading sample buffer and processed for anti-gp130 and insulin receptor immunoblotting. Input aliquots and eluted protein fractions were resolved on $12\%$ SDS-PAGE and transferred onto nitrocellulose membranes by electroblotting. Membranes were blocked as described previously and probed with streptavidin-biotinylated HRP complex (GE healthcare), rabbit polyclonal gp130 (RRID:AB_647629; 0.2 µg/mL) and insulin receptor β (RRID:AB_631835; 1 µg/mL) antibodies (Santa Cruz Biotechnology), followed by HRP-conjugated goat anti-rabbit antibody (RRID:AB_390191; dilution 1:2000) to reveal proteins using chemiluminescent reagent (Pierce). The amount of proteins was determined by densitometry using Image J software; results were presented as the ratio of carbonylated proteins (biotinylated captured protein) on total proteins (input). ## In silico studies Bulk tissue gene expression for REG3A was assessed from the GTEx portal (gtexportal.org; version V8). The Gene Expression Omnibus (GEO) data repository was used to study the expression profile of REG3A in liver, pancreas, white adipose tissue and skeletal muscle of patients with type 2 diabetes. We analyzed 21 publicly available datasets for normalized differential gene expression using limma and edgeR R packages. Enrichment scores were calculated using the ssGSEA algorithm based on the KEGG and Reactome collections of the Molecular Signatures database (MSigDB). Pearson correlation coefficients were calculated between REG3A expression and enrichment scores to determine the relevant pathways associated with REG3A. Local false discovery rates were measured using the Benjamini-Hochberg method. All calculations were performed with the R programming language (v.4.1.3). P values < 0.05 were deemed statistically significant. ## Statistics and reproducibility Differences between two sample groups were tested using Student’s one-tailed t test or Wilcoxon test. Changes over time (cumulative food intake, energy expenditure, respiratory exchange ratio, fat oxidation, weight gain, OGTT, ITT) were assessed by two-way repeated measures ANOVA followed by Tukey’s post-comparison test. Differences between more than two sample groups were tested by ANOVA followed by Tukey’s post comparison test when conditions for normality were met. Otherwise, we evaluated by the Kruskal–Wallis test followed by a pairwise Wilcoxon test with Benjamini–Hochberg correction. Results are presented as means ± SEM or box plots. P-values < 0.05 were deemed statistically significant. Raw data are available in Supplementary Data and figures. ## Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. ## Supplementary information Peer Review File Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Supplementary Data 6 Supplementary Data 7 Reporting Summary The online version contains supplementary material available at 10.1038/s42003-023-04616-5. ## Peer review information Communications Biology thanks Yuping Lai and the other, anonymous, reviewer for their contribution to the peer review of this work. Primary Handling Editors: Gabriela da* Silva Xavier and Joao Valente. Peer reviewer reports are available. ## References 1. Moore JX, Chaudhary N, Akinyemiju T. Metabolic syndrome prevalence by race/ethnicity and sex in the United States, National Health and Nutrition Examination Survey, 1988-2012. Prev. Chronic Dis. (2017) 14 E24. DOI: 10.5888/pcd14.160287 2. Scuteri A. Metabolic syndrome across Europe: different clusters of risk factors. Eur. J. Prev. Cardiol. (2015) 22 486-491. DOI: 10.1177/2047487314525529 3. Lackey DE, Olefsky JM. Regulation of metabolism by the innate immune system. Nat. Rev. Endocrinol. (2016) 12 15-28. DOI: 10.1038/nrendo.2015.189 4. Agrawal N. The Drosophila TNF eiger is an adipokine that acts on insulin-producing cells to mediate nutrient response. Cell Metab. (2016) 23 675-684. DOI: 10.1016/j.cmet.2016.03.003 5. DiAngelo JR, Bland ML, Bambina S, Cherry S, Birnbaum MJ. The immune response attenuates growth and nutrient storage in Drosophila by reducing insulin signaling. Proc. Natl Acad. Sci. USA (2009) 106 20853-20858. DOI: 10.1073/pnas.0906749106 6. Hau CS. Visfatin enhances the production of cathelicidin antimicrobial peptide, human β-defensin-2, human β-defensin-3, and S100A7 in human keratinocytes and their orthologs in murine imiquimod-induced psoriatic skin. Am. J. Pathol. (2013) 182 1705-1717. DOI: 10.1016/j.ajpath.2013.01.044 7. Kanda N, Watanabe S. Leptin enhances human beta-defensin-2 production in human keratinocytes. Endocrinology (2008) 149 5189-5198. DOI: 10.1210/en.2008-0343 8. Lago F, Dieguez C, Gómez-Reino J, Gualillo O. Adipokines as emerging mediators of immune response and inflammation. Nat. Clin. Pract. Rheumatol. (2007) 3 716-724. DOI: 10.1038/ncprheum0674 9. Brennan MB. Alpha-melanocyte-stimulating hormone is a peripheral, integrative regulator of glucose and fat metabolism. Ann. N. Y. Acad. Sci. (2003) 994 282-287. DOI: 10.1111/j.1749-6632.2003.tb03191.x 10. Schneeberger M. Reduced α-MSH underlies hypothalamic ER-stress-induced hepatic gluconeogenesis. Cell Rep. (2015) 12 361-370. DOI: 10.1016/j.celrep.2015.06.041 11. Sun J. Cathelicidins positively regulate pancreatic β-cell functions. FASEB J. (2016) 30 884-894. DOI: 10.1096/fj.15-275826 12. Solana R, Pawelec G, Tarazona R. Aging and innate immunity. Immunity (2006) 24 491-494. DOI: 10.1016/j.immuni.2006.05.003 13. Cash HL, Whitham CV, Hooper LV. Refolding, purification, and characterization of human and murine RegIII proteins expressed in Escherichia coli. Protein Expression Purification (2006) 48 151-159. DOI: 10.1016/j.pep.2006.01.014 14. Graf R. A family of 16-kDa pancreatic secretory stress proteins form highly organized fibrillar structures upon tryptic activation. J. Biol. Chem. (2001) 276 21028-21038. DOI: 10.1074/jbc.M010717200 15. Mukherjee S. Regulation of C-type lectin antimicrobial activity by a flexible N-terminal prosegment. J. Biol. Chem. (2009) 284 4881-4888. DOI: 10.1074/jbc.M808077200 16. Medveczky P, Szmola R, Sahin-Tóth M. Proteolytic activation of human pancreatitis-associated protein is required for peptidoglycan binding and bacterial aggregation. Biochem. J. (2009) 420 335-343. DOI: 10.1042/BJ20090005 17. Darnaud M. Enteric delivery of regenerating family member 3 alpha alters the intestinal microbiota and controls inflammation in mice with colitis. Gastroenterology (2018) 154 1009-1023.e14. DOI: 10.1053/j.gastro.2017.11.003 18. Gallo RL, Hooper LV. Epithelial antimicrobial defence of the skin and intestine. Nat. Rev. Immunol. (2012) 12 503-516. DOI: 10.1038/nri3228 19. Gironella M. Experimental acute pancreatitis in PAP/HIP knock-out mice. Gut (2007) 56 1091-1097. DOI: 10.1136/gut.2006.116087 20. Haldipur P. HIP/PAP prevents excitotoxic neuronal death and promotes plasticity. Ann. Clin. Transl. Neurol. (2014) 1 739-754. DOI: 10.1002/acn3.127 21. Livesey FJ. A Schwann cell mitogen accompanying regeneration of motor neurons. Nature (1997) 390 614-618. DOI: 10.1038/37615 22. Lörchner H. Myocardial healing requires Reg3β-dependent accumulation of macrophages in the ischemic heart. Nat. Med. (2015) 21 353-362. DOI: 10.1038/nm.3816 23. Moniaux N. Human hepatocarcinoma-intestine-pancreas/pancreatitis-associated protein cures fas-induced acute liver failure in mice by attenuating free-radical damage in injured livers. Hepatology (Baltimore, Md.) (2011) 53 618-627. DOI: 10.1002/hep.24087 24. Wang L. Intestinal REG3 lectins protect against alcoholic steatohepatitis by reducing mucosa-associated microbiota and preventing bacterial translocation. Cell Host Microbe (2016) 19 227-239. DOI: 10.1016/j.chom.2016.01.003 25. Zheng X. HIP/PAP protects against bleomycin-induced lung injury and inflammation and subsequent fibrosis in mice. J. Cell Mol. Med. (2020) 24 6804-6821. DOI: 10.1111/jcmm.15334 26. Wu Y. Hyperglycaemia inhibits REG3A expression to exacerbate TLR3-mediated skin inflammation in diabetes. Nat. Commun. (2016) 7 13393. DOI: 10.1038/ncomms13393 27. Christa L. HIP/PAP is an adhesive protein expressed in hepatocarcinoma, normal Paneth, and pancreatic cells. Am. J. Physiol. (1996) 271 G993-G1002. PMID: 8997243 28. Levetan CS. Discovery of a human peptide sequence signaling islet neogenesis. Endocr Pract (2008) 14 1075-1083. DOI: 10.4158/EP.14.9.1075 29. Mueller CM, Zhang H, Zenilman ME. Pancreatic Reg I binds MKP-1 and regulates Cyclin D in pancreatic-derived cells. J. Surg. Res. (2008) 150 137-143. DOI: 10.1016/j.jss.2008.03.047 30. Ba IA-TV. Regenerating Islet-derived 1α (Reg-1α) protein is new neuronal secreted factor that stimulates neurite outgrowth via exostosin tumor-like 3 (EXTL3) receptor . J. Biol. Chem.* (2012) 287 4726-4739. DOI: 10.1074/jbc.M111.260349 31. Lai Y. The antimicrobial protein REG3A regulates keratinocyte proliferation and differentiation after skin injury. Immunity (2012) 37 74-84. DOI: 10.1016/j.immuni.2012.04.010 32. Loncle C. IL17 functions through the novel REG3β-JAK2-STAT3 inflammatory pathway to promote the transition from chronic pancreatitis to pancreatic cancer. Cancer Res. (2015) 75 4852-4862. DOI: 10.1158/0008-5472.CAN-15-0896 33. Liu X. REG3A accelerates pancreatic cancer cell growth under IL-6-associated inflammatory condition: Involvement of a REG3A–JAK2/STAT3 positive feedback loop. Cancer Lett. (2015) 362 45-60. DOI: 10.1016/j.canlet.2015.03.014 34. Dungan KM, Buse JB, Ratner RE. Effects of therapy in type 1 and type 2 diabetes mellitus with a peptide derived from islet neogenesis associated protein (INGAP). Diabetes Metab. Res. Rev. (2009) 25 558-565. DOI: 10.1002/dmrr.999 35. Fleming A, Rosenberg L. Prospects and challenges for islet regeneration as a treatment for diabetes: a review of islet neogenesis associated protein. J. Diabetes Sci. Technol. (2007) 1 231-244. DOI: 10.1177/193229680700100214 36. Gross DJ. Amelioration of diabetes in nonobese diabetic mice with advanced disease by linomide-induced immunoregulation combined with Reg protein treatment. Endocrinology (1998) 139 2369-2374. DOI: 10.1210/endo.139.5.5997 37. Mittermayer F. Addressing unmet medical needs in type 1 diabetes: a review of drugs under development. Curr. Diabetes Rev. (2017) 13 300-314. DOI: 10.2174/1573399812666160413115655 38. Okamoto H. The Reg gene family and Reg proteins: with special attention to the regeneration of pancreatic beta-cells. J. Hepatobiliary Pancreat. Surg. (1999) 6 254-262. DOI: 10.1007/s005340050115 39. Parikh A, Stephan A-F, Tzanakakis ES. Regenerating proteins and their expression, regulation and signaling. Biomol. Concepts (2012) 3 57-70. DOI: 10.1515/bmc.2011.055 40. Román CL, Maiztegui B, Del Zotto H, Gagliardino JJ, Flores LE. INGAP-PP effects on β-cell mass and function are related to its positive effect on islet angiogenesis and VEGFA production. Mol. Cell Endocrinol. (2018) 470 269-280. DOI: 10.1016/j.mce.2017.11.009 41. Rosenberg L. A pentadecapeptide fragment of islet neogenesis-associated protein increases beta-cell mass and reverses diabetes in C57BL/6J mice. Ann. Surg. (2004) 240 875-884. DOI: 10.1097/01.sla.0000143270.99191.10 42. Chen Z, Downing S, Tzanakakis ES. Four decades after the discovery of regenerating islet-derived (Reg) proteins: current understanding and challenges. Front Cell Dev Biol (2019) 7 235. DOI: 10.3389/fcell.2019.00235 43. Unno M. Production and characterization of Reg knockout mice: reduced proliferation of pancreatic beta-cells in Reg knockout mice. Diabetes (2002) 51 S478-S483. DOI: 10.2337/diabetes.51.2007.S478 44. Watanabe T. Pancreatic beta-cell replication and amelioration of surgical diabetes by Reg protein. Proc. Natl Acad. Sci. USA (1994) 91 3589-3592. DOI: 10.1073/pnas.91.9.3589 45. Xiong X. Pancreatic islet-specific overexpression of Reg3β protein induced the expression of pro-islet genes and protected the mice against streptozotocin-induced diabetes mellitus. Am. J. Physiol. Endocrinol. Metab. (2011) 300 E669-E680. DOI: 10.1152/ajpendo.00600.2010 46. Shin JH, Seeley RJ. Reg3 Proteins as Gut Hormones?. Endocrinology (2019) 160 1506-1514. DOI: 10.1210/en.2019-00073 47. Simon M-T. HIP/PAP stimulates liver regeneration after partial hepatectomy and combines mitogenic and anti-apoptotic functions through the PKA signaling pathway. FASEB J. (2003) 17 1441-1450. DOI: 10.1096/fj.02-1013com 48. Fink RI, Revers RR, Kolterman OG, Olefsky JM. The metabolic clearance of insulin and the feedback inhibition of insulin secretion are altered with aging. Diabetes (1985) 34 275-280. DOI: 10.2337/diab.34.3.275 49. Marmentini C. Aging reduces insulin clearance in mice. Front. Endocrinol. (Lausanne) (2021) 12 679492. DOI: 10.3389/fendo.2021.679492 50. Pettersson US, Waldén TB, Carlsson P-O, Jansson L, Phillipson M. Female mice are protected against high-fat diet induced metabolic syndrome and increase the regulatory T cell population in adipose tissue. PLoS ONE (2012) 7 e46057. DOI: 10.1371/journal.pone.0046057 51. Moniaux N. The Reg3α (HIP/PAP) lectin suppresses extracellular oxidative stress in a murine model of acute liver failure. PLoS ONE (2015) 10 e0125584. DOI: 10.1371/journal.pone.0125584 52. Lieu H-TT. HIP/PAP accelerates liver regeneration and protects against acetaminophen injury in mice. Hepatology (2005) 42 618-626. DOI: 10.1002/hep.20845 53. März-Weiss P. Expression of pancreatitis-associated protein after traumatic brain injury: a mechanism potentially contributing to neuroprotection in human brain. Cell Mol. Neurobiol. (2011) 31 1141-1149. DOI: 10.1007/s10571-011-9715-0 54. Watt MJ. CNTF reverses obesity-induced insulin resistance by activating skeletal muscle AMPK. Nat. Med. (2006) 12 541-548. DOI: 10.1038/nm1383 55. Broekaert D. Comparison of leptin- and interleukin-6-regulated expression of the rPAP gene family: evidence for differential co-regulatory signals. Eur. Cytokine Netw. (2002) 13 78-85. PMID: 11956024 56. Nishimune H. Reg-2 is a motoneuron neurotrophic factor and a signalling intermediate in the CNTF survival pathway. Nat. Cell Biol. (2000) 2 906-914. DOI: 10.1038/35046558 57. Tohma Y. Reg gene expression in periosteum after fracture and its in vitro induction triggered by IL-6. Int J Mol Sci (2017) 18 E2257. DOI: 10.3390/ijms18112257 58. Loncle C. REG3β plays a key role in IL17RA protumoral effect-response. Cancer Res. (2016) 76 2051. DOI: 10.1158/0008-5472.CAN-15-3355 59. Misu H. A liver-derived secretory protein, selenoprotein P, causes insulin resistance. Cell Metab. (2010) 12 483-495. DOI: 10.1016/j.cmet.2010.09.015 60. Fadista J. Global genomic and transcriptomic analysis of human pancreatic islets reveals novel genes influencing glucose metabolism. Proc Natl Acad Sci USA (2014) 111 13924-13929. DOI: 10.1073/pnas.1402665111 61. Civelek M. Genetic regulation of adipose gene expression and cardio-metabolic traits. Am. J. Hum. Genet. (2017) 100 428-443. DOI: 10.1016/j.ajhg.2017.01.027 62. Wu X. The effect of insulin on expression of genes and biochemical pathways in human skeletal muscle. Endocrine (2007) 31 5-17. DOI: 10.1007/s12020-007-0007-x 63. Gancheva S. Dynamic changes of muscle insulin sensitivity after metabolic surgery. Nat. Commun. (2019) 10 4179. DOI: 10.1038/s41467-019-12081-0 64. Priest C, Tontonoz P. Inter-organ cross-talk in metabolic syndrome. Nat Metab. (2019) 1 1177-1188. DOI: 10.1038/s42255-019-0145-5 65. Evans JL, Goldfine ID, Maddux BA, Grodsky GM. Oxidative stress and stress-activated signaling pathways: a unifying hypothesis of type 2 diabetes. Endocr. Rev. (2002) 23 599-622. DOI: 10.1210/er.2001-0039 66. Gao D. The effects of palmitate on hepatic insulin resistance are mediated by NADPH Oxidase 3-derived reactive oxygen species through JNK and p38MAPK pathways. J. Biol. Chem. (2010) 285 29965-29973. DOI: 10.1074/jbc.M110.128694 67. Hoehn KL. Insulin resistance is a cellular antioxidant defense mechanism. Proc. Natl Acad. Sci. USA (2009) 106 17787-17792. DOI: 10.1073/pnas.0902380106 68. Saadeh M. Reactive oxygen species stimulate insulin secretion in rat pancreatic islets: studies using mono-oleoyl-glycerol. PLoS ONE (2012) 7 e30200. DOI: 10.1371/journal.pone.0030200 69. Iovanna JL. Serum levels of pancreatitis-associated protein as indicators of the course of acute pancreatitis. Multicentric Study Group on Acute Pancreatitis. Gastroenterology (1994) 106 728-734. DOI: 10.1016/0016-5085(94)90708-0 70. Ferrara JLM. Regenerating islet-derived 3-alpha is a biomarker of gastrointestinal graft-versus-host disease. Blood (2011) 118 6702-6708. DOI: 10.1182/blood-2011-08-375006 71. Gironella M. Anti-inflammatory effects of pancreatitis associated protein in inflammatory bowel disease. Gut (2005) 54 1244-1253. DOI: 10.1136/gut.2004.056309 72. Sands M. Antimicrobial protein REG3A and signaling networks are predictive of stroke outcomes. J. Neurochem. (2022) 160 100-112. DOI: 10.1111/jnc.15520 73. Balagopal P, Schimke JC, Ades P, Adey D, Nair KS. Age effect on transcript levels and synthesis rate of muscle MHC and response to resistance exercise. Am. J. Physiol. Endocrinol. Metab. (2001) 280 E203-E208. DOI: 10.1152/ajpendo.2001.280.2.E203 74. Cartee GD. Aging skeletal muscle: response to exercise. Exerc Sport Sci Rev (1994) 22 91-120. DOI: 10.1249/00003677-199401000-00006 75. Capel F, Buffière C, Patureau Mirand P, Mosoni L. Differential variation of mitochondrial H2O2 release during aging in oxidative and glycolytic muscles in rats. Mech. Ageing Dev. (2004) 125 367-373. DOI: 10.1016/j.mad.2004.02.005 76. Houstis N, Rosen ED, Lander ES. Reactive oxygen species have a causal role in multiple forms of insulin resistance. Nature (2006) 440 944-948. DOI: 10.1038/nature04634 77. McMurray F, Patten DA, Harper M-E. Reactive oxygen species and oxidative stress in obesity-recent findings and empirical approaches. Obesity (Silver Spring) (2016) 24 2301-2310. DOI: 10.1002/oby.21654 78. Hinge A, Bajaj M, Limaye L, Surolia A, Kale V. Oral administration of insulin receptor-interacting lectins leads to an enhancement in the hematopoietic stem and progenitor cell pool of mice. Stem Cells Dev. (2010) 19 163-174. DOI: 10.1089/scd.2009.0128 79. da Silva Rosa SC, Nayak N, Caymo AM, Gordon JW. Mechanisms of muscle insulin resistance and the cross-talk with liver and adipose tissue. Physiol. Rep. (2020) 8 e14607. DOI: 10.14814/phy2.14607 80. Defronzo RA. Banting Lecture. From the triumvirate to the ominous octet: a new paradigm for the treatment of type 2 diabetes mellitus. Diabetes (2009) 58 773-795. DOI: 10.2337/db09-9028 81. Secq V. PAP/HIP protein is an obesogenic factor. J. Cell Physiol. (2014) 229 225-231. DOI: 10.1002/jcp.24438 82. Gonzalez P, Moniaux N, Bréchot C, Faivre J. Is the Reg3α (HIP/PAP) protein really an obesogenic factor?. J. Cell Physiol. (2016) 231 1. DOI: 10.1002/jcp.25046 83. Secq V. Response to ‘is the Reg3α (HIP/PAP) protein really an obesogenic factor?’. J. Cell Physiol. (2016) 231 2. DOI: 10.1002/jcp.25130 84. Bluemel S. The role of intestinal C-type regenerating islet derived-3 lectins for nonalcoholic steatohepatitis. Hepatol Commun (2018) 2 393-406. DOI: 10.1002/hep4.1165 85. Matthews DR. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia (1985) 28 412-419. DOI: 10.1007/BF00280883 86. Bligh EG, Dyer WJ. A rapid method of total lipid extraction and purification. Can J. Biochem. Physiol. (1959) 37 911-917. DOI: 10.1139/y59-099
--- title: Secular trends in physical fitness of rural Chinese children and adolescents aged 7–18 years from 1985 to 2019 authors: - Chengyue Li - Alimujiang Yimiti Taerken - Qian Li - Adilijiang Selimu - Hao Wang journal: Scientific Reports year: 2023 pmcid: PMC10015040 doi: 10.1038/s41598-023-31190-x license: CC BY 4.0 --- # Secular trends in physical fitness of rural Chinese children and adolescents aged 7–18 years from 1985 to 2019 ## Abstract The main purpose of the study was to evaluate the secular trends in physical fitness of Chinese rural children and adolescents aged 7–18 from 1985 to 2019. The speed, muscular strength, explosive power fitness, cardiorespiratory fitness, and flexibility were investigated by National Survey on Students’ Constitution and Health in 1985, 2000, 2010 and 2019. During the period 1985–2000, the physical fitness of Chinese rural children and adolescents improved except for flexibility, and almost all of them reached the maximum increase rate. During the period 2000–2010, in addition to the improvement of flexibility, other fitness showed downward trends, and the decline ranges were large. During the period 2010–2019, the speed of boys rebounded, flexibility, explosive power and muscular strength continued to decrease. Meanwhile, speed, flexibility and muscular strength in girls rebounded, and the explosive power continued to decline. From 2000 to 2019, the body mass index increase accelerated. This study shows that some components of physical fitness of Chinese rural children and adolescents have shown positive trends in recent years, especially for girls and adolescents aged 13–15 years. However, it may also imply inequality between sexes and ages, which provides a reference for the focus of the country's physical fitness and health monitoring and intervention measures. ## Introduction Physical fitness is a multicomponent construct that is closely related to the ability to perform physical activity1. Although its measurement varies from country to country, core items usually include endurance running (reflecting cardiorespiratory fitness), standing long jump (reflecting explosive power), 1-min pull-ups and sit-ups (reflecting muscular strength), sit-and-reach (reflecting flexibility), and 50-m (or 60-m) dash (reflecting speed)2. These tests cover different functions and structures of the body`s movement, including musculoskeletal, cardiorespiratory, circulatory, endocrine metabolic and psycho-neurological functions2. Given the current concerns about the declining fitness of the world's children and adolescents and its potential association with nutritional and health status in adulthood3,4, cardiovascular fitness is often the focus of attention, with its association with greater BMI and fat mass5, while speed, flexibility and strength performances are core fitness for children and adolescents to participate in multiple forms of physical activity6. In China, the "Research on Physical Shape, Function and Fitness of Chinese Children and Adolescents" project started in 19797, which was the first national growth and development survey performed by the former state Physical Culture and Sports Commission, covering only 16 provinces/municipalities directly under the central government8. Starting in 1985, five central ministries and commissions, including the Ministry of Education, established the "National Students fitness and Health Survey" system, in which almost all provinces, autonomous regions, and municipalities directly under the central government participated in the study once every five years and continued to increase the scope of the study afterward. Since 1985, eight surveys have been performed, providing comprehensive scientific data on the physical development of children and adolescents in China, and many achievements have been made during this period (the latest survey was conducted in 2019). Since the “reform and opening of China” in 1978, the national economy has increased rapidly, and the primary manifestation of this was the improvement of people's material life. The level of diet and nutrition of children and adolescents has improved significantly, the growth potential has been stimulated, and physical fitness has begun to improve. However, since around the twenty-first century, the height development of Chinese children and adolescents has slowed down while their weight has increased dramatically9, and the situation of overweight and obesity has become serious10, and physical fitness has begun to decline or fluctuate2,11. Previous studies have shown that only speed and flexibility rebounded among Chinese children and adolescents in 2014, while other fitness components continued to decline to vary degrees, and although the urban–rural gap decreased, the advantage of rural children and adolescents in endurance fitness and explosive strength also narrowed11,12. Meanwhile, recent regional studies have shown that the physical fitness of children and adolescents in rural areas, although better than in urban areas, has been on a decreasing trend13. However, these findings do not reflect the Chinese overall situation. Current national studies of secular trends in children and adolescents’ physical fitness are mainly time-scale changes, rely on cross-sectional designs, usually compare only a few time points for children and adolescent age groups or joint age groups and have small sample sizes12,14–16. Recently, several systematic reviews on the secular trend of cardiorespiratory fitness and muscular strength of children and adolescents have indicated that in recent years, muscular strength, measured by grip strength, has improved in some countries17, the changes in cardiorespiratory fitness, measured by 20-m shuttle run, have tended to be stable18, and muscular endurance, measured by sit-ups19 and explosive power, measured by standing long jump, have worsened20. Although these analyses provide the secular trends of children and adolescents in China, the report does not distinguish the differences between urban and rural areas, lacks some test items, and the data are relatively old. In recent years, more attention has been given to research on muscular strength and cardiorespiratory fitness16–21, although due to its significant role in health and recent negative trends worldwide6,15,18–21, there is still a lack of comprehensive research on some health-related physical fitness, especially speed and flexibility fitness. An inverted U-shaped relationship between body mass index (BMI) and physical fitness has been shown, suggesting that malnutrition and overweight obesity could have a negative impact on physical fitness2,22,23. In 1985, the prevalence of malnutrition among rural children and adolescents was as high as $24.2\%$, approximately 2.7 times higher than that in urban areas, but the prevalence of overweight and obesity was less than $1.0\%$24. Subsequently, China actively carried out programs for "rural revitalization"25, physical education reform, and the “nutrition improvement program for rural students in compulsory education”26 and so on. Malnutrition in rural areas has been greatly improved, and physical fitness have improved, but recent studies have shown that obesity among rural children and adolescents has increased dramatically, the increasing rate of obesity in rural areas is greater than that in urban areas27, and too many obese individuals in rural areas can also lead to a decline in physical fitness. In addition, the increase in sedentary time and the decrease in physical activity of children and adolescents in recent years has also led to a negative impact on physical fitness28. In summary, it is necessary to carry out secular trends research on rural children and adolescents in China, which can not only provide a reference for Chinese physical health, physical education and public health policies but also make efforts to supplement the research on physical fitness in recent years and guide future global physical health research and health monitoring29. Therefore, this paper conducts an analysis of the secular trends of Chinese children and adolescents’ physical fitness in rural areas over a 34-year period by the Chinese National Surveillance on Students' Constitution and Health (CNSSCH). Specifically, we aim to [1] investigate the secular trends of physical fitness of five fitness components of Chinese children and adolescents aged 7–18 years in rural during the entire period, [2] understand the changes between subgroups (age, sex) and different periods, so as to find inequalities in the health of Chinese children and adolescents. ## Study design and subjects Data were obtained from test scores of Han Chinese children and adolescents aged 7–18 in rural areas by CNSSCH30–33 in 1985, 2000, 2010, and 2019. CNSSCH was organized by the Ministries of Education, Health, Science and Technology, the State Ethnic Affairs Commission, and the State Sports General Administration of the People’s Republic of China30–33. Multistage stratified cluster sampling was used to maintain consistent sampling and assessment methods across survey years with the class as the sampling unit. The sampling procedure was performed as previously described in detail2,24. 29 provinces/autonomous regions/municipalities directly under the central government (34 overall), excluding Hong Kong, Macau, Taiwan, Hainan, and Chongqing, were included in 1985, and Hainan and Chongqing were included in the latter three surveys. This study only included participants of the Han ethnicity, who account for $92\%$ of the total Chinese population, from 26 mainland provinces and 4 municipalities of mainland China, excluding Tibet (where the Han ethnicity is a minority). Since 1985, children and adolescents in each province, except Tibet, were stratified into three levels according to their socioeconomic status (upper, moderate, and low) and then, in turn, stratified by urban and rural areas according to their place of residence, with at least 50 Han Chinese students in each age group included in the survey. The classifications of urban and rural were based on the revised criteria for designated towns issued in 198434. It has not changed since the initial classification in 1985, which means that if an area initially classified as rural experienced urbanization, it remained classified as rural. The exclusion criteria for participants were: [1] suffering from important organ diseases such as heart, lung, liver, and kidney; [2] abnormal physical development (e.g., pygmyism, gigantism); [3] those with physical disabilities or deformities; [4] those with acute illnesses, or those who suffered from acute illnesses in the last month of the testing period and had not recovered their physical strength; [5] girls who were menstruating (The girls were asked about their menstrual status in each age group by the female internist, and only they were asked "with or without" being menstruating). All participants were grouped by sex and age, with 1 year being an age group and 24 age groups in total. Participants with missing data or illogical test results were excluded. From 1985 to 2019, 160,588, 263,421, 262,765, 262,661, 259,757 and 260,448 boys and 160,888, 262,667, 262,847, 262,687, 262,727 and 260,839 girls aged 7–18 were tested for BMI (1985 data missing), speed, explosive power, flexibility, muscular strength and cardiorespiratory fitness, respectively. The number of boys and girls tested in 2000, 2010, and 2019 ranged from 51,000 to 54,000 with a rate of approximately 1:1 for each age group, and the number of those tested in 1985 was approximately twice the other survey years. There were similar numbers in each age group. See Tables 1 and 2 for details. Table 1Comparison of physical fitness scores of five fitness components of Chinese boys of different age categories from 1985 to 2019.Age categories(year)1985 (a)2000 (b)2010 (c)2019 (d)FSignificant post hoc comparisons #R2B &AgeNMSDAgeNMSDAgeNMSDAgeNMSD7–12Body mass index (kg/m2)9.49 (1.71)27,12015.912.339.50 (1.71)26,93816.972.959.48 (1.71)26,83318.033.593370.270**b vs c, b vs d, c vs d0.9990.11150-m dash (s)9.50 (1.71)51,34210.191.099.49 (1.71)27,1179.971.149.50 (1.71)26,89610.031.209.48 (1.71)26,49610.031.28266.836*a vs b, a vs c, a vs d, b vs c, b vs d0.590− 0.005Standing long jump (cm)9.50 (1.71)51,342140.8421.709.49 (1.71)27,156152.2224.239.50 (1.71)26,916149.7024.759.48 (1.71)26,522143.5826.771694.614a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.2030.158Stand/sit-and-reach (cm)9.50 (1.71)51,3425.524.509.48 (1.71)27,1304.854.809.50 (1.71)26,9216.445.349.48 (1.71)26,7305.946.21474.044a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.2380.019oblique body pull-ups (n)9.50 (1.71)51,34218.0810.809.47 (1.71)25,29329.0014.789.50 (1.71)26,86528.5719.319.40 (1.67)25,59724.6120.174028.403a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.5260.25950-m × 8 shuttle run (s)9.50 (1.71)51,342114.6711.589.48 (1.71)27,101119.5615.399.49 (1.70)26,795123.8417.409.41 (1.67)25,414126.4619.2541,119.917a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.9970.35313–15Body mass index (kg/m2)14.01 (0.82)13,29018.202.6414.00 (0.82)13,47719.083.0114.01 (0.82)13,19720.413.911572.592*b vs c, b vs d, c vs d0.9780.11650-m dash (s)14.00 (0.82)25,6758.650.7114.01 (0.82)13,2778.330.7914.00 (0.82)13,4648.330.9214.01 (0.82)13,0338.090.961458.596a vs b, a vs c, a vs d, b vs d, c vs d0.943− 0.015Standing long jump (cm)14.00 (0.82)25,675185.0722.9214.01 (0.82)13,305201.5324.1514.00 (0.82)13,466199.7326.2814.01 (0.82)13,062198.3028.571846.750*a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.6750.446Stand/sit-and-reach (cm)14.00 (0.82)25,6758.805.6214.01 (0.82)13,1417.706.0514.00 (0.82)13,4678.896.7114.01 (0.82)13,1258.177.57116.352a vs b, a vs d, b vs c, b vs d, c vs d0.126− 0.012Pull-ups (n)14.00 (0.82)25,6753.233.1714.01 (0.82)13,2944.374.5014.00 (0.82)13,4413.454.8514.01 (0.82)12,9772.944.48313.363a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.009− 0.0031000-m run (s)14.00 (0.82)25,675255.3524.5714.01 (0.82)12,915268.2833.1114.01 (0.82)13,216284.2140.8114.01 (0.82)12,962277.2647.472320.016a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.8440.78916–18Body mass index (kg/m2)17.01 (0.82)13,41419.882.4417.00 (0.82)13,44220.252.7716.98 (0.82)12,87721.583.771135.429b vs c, b vs d, c vs d0.8820.08850-m dash (s)17.00 (0.82)25,6407.880.5617.01 (0.82)13,3987.580.5817.00 (0.82)13,4017.680.7416.98 (0.82)12,7827.670.87702.666a vs b, a vs c, a vs d, b vs c, b vs d0.556− 0.007Standing long jump (cm)17.00 (0.82)25,640212.6920.1217.01 (0.82)13,423227.5919.9617.00 (0.82)13,435226.2622.1916.98 (0.82)12,823220.0025.601890.707a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.4260.311Stand/sit-and-reach (cm)17.00 (0.82)25,64013.205.8417.01 (0.82)13,23911.356.9017.00 (0.82)13,43112.556.9616.98 (0.82)12,82010.917.64434.280a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.623− 0.056Pull-ups (n)17.00 (0.82)25,6406.903.8817.01 (0.82)13,4177.344.5217.00 (0.82)13,3925.495.3116.98 (0.82)12,8144.334.741338.465a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.702− 0.071000-m run (s)17.00 (0.82)25,640233.0420.4217.01 (0.82)13,251243.8024.4717.00 (0.82)13,417255.8431.5616.98 (0.82)12,720262.8339.263894.106a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.9910.886AllBody mass index (kg/m2)12.48 (3.46)53,82417.462.9612.50 (3.45)53,85718.313.2412.44 (3.45)52,90719.494.024703.305b vs c, b vs d, c vs d0.9850.10650-m dash (s)12.50 (3.45)102,6579.231.3412.48 (3.46)53,7928.971.4112.50 (3.45)53,7619.021.4612.44 (3.45)53,2118.971.56622.830a vs b, a vs c, a vs d, b vs c, c vs d0.774− 0.008Standing long jump (cm)12.50 (3.45)102,657169.8537.4912.48 (3.46)53,884183.1739.9612.50 (3.45)53,817181.3341.1212.45 (3.45)52,407175.9243.081710.962a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.3880.259Stand/sit-and-reach (cm)12.50 (3.45)102,6578.266.0412.46 (3.46)53,5107.166.2912.50 (3.45)53,8198.586.6212.44 (3.45)52,6757.717.22536.586a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.045− 0.008N, is the sample size; M, is the mean, and SD is the standard deviation.#One-way analysis of variance (ANOVA) with the Bonferroni post hoc test.&Sample-weighted linear regression.Represents $p \leq 0.05$; represents $p \leq 0.01$; represents $p \leq 0.001$, same below. Table 2Comparison of physical fitness scores of five fitness components of Chinese girls of different age categories from 1985 to 2019.Age categories(year)1985 (a)2000 (b)2010 (c)2019 (d)FSignificant post hoc comparisons R2B AgeNMSDAgeNMSDAgeNMSDAgeNMSD7–12Body mass index (kg/m2)9.50 (1.71)26,99015.632.359.50 (1.71)26,97116.352.569.51 (1.71)26,96917.303.132596.387b vs c, b vs d, c vs d0.9880.08850-m dash (s)9.50 (1.71)51,34110.711.209.50 (1.71)26,96410.531.239.50 (1.71)26,95210.611.209.50 (1.71)26,64210.461.21291.723a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.767− 0.006Standing long jump (cm)9.50 (1.71)51,341133.0820.29.50 (1.71)26,995140.4722.639.50 (1.71)26,953137.2122.779.50 (1.71)26,669132.7723.55860.430a vs b, a vs c, b vs c, b vs d, c vs d0.0300.039Stand/sit-and-reach (cm)9.50 (1.71)51,3417.664.639.50 (1.71)26,9996.274.869.50 (1.71)26,9549.465.339.50 (1.71)26,71010.576.323771.308a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.5240.0821-min sit-ups (n)9.50 (1.71)51,34116.6010.759.50 (1.71)26,97924.0511.309.50 (1.71)26,93320.6710.329.49 (1.71)26,59525.3510.864973.544a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.7280.23150-m × 8 shuttle run (s)9.50 (1.71)51,341121.2412.299.50 (1.71)26,967125.7316.179.49 (1.70)26,770128.4516.169.43 (1.68)25,538130.2018.102495.489a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.9950.27113–15Body mass index (kg/m2)14.00 (0.82)13,42518.612.4714.00 (0.82)13,45819.242.6814.00 (0.82)13,21420.483.331484.940b vs c, b vs d, c vs d0.9530.09850-m dash (s)14.00 (0.82)25,6719.580.8114.00 (0.82)13,4199.500.8414.00 (0.82)13,4229.741.0014.00 (0.82)13,0849.511.03201.148a vs b, a vs c, a vs d, b vs c, c vs d0.0030.000Standing long jump (cm)14.00 (0.82)25,671158.3717.9314.00 (0.82)13,431166.4018.6214.00 (0.82)13,426161.1319.1314.00 (0.82)13,100156.8321.93691.314a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.000− 0.001Stand/sit-and-reach (cm)14.00 (0.82)25,67110.345.5314.00 (0.82)13,2768.705.8914.00 (0.82)13,41611.446.3914.00 (0.82)13,16612.717.251027.921a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.3950.0621-min sit-ups (n)14.00 (0.82)25,67120.979.9914.00 (0.82)13,43129.7310.6014.00 (0.82)13,41325.7510.1214.00 (0.82)13,10830.6010.843467.113a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.7080.257800-m run (s)14.00 (0.82)25,671231.2724.1214.00 (0.82)13,269246.9630.1014.01 (0.82)13,418262.1133.8114.00 (0.82)12,951256.5338.703704.561a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.8780.88416–18Body mass index (kg/m2)17.00 (0.82)13,48520.282.3117.00 (0.82)13,45920.292.3816.98 (0.81)12,91721.093.12411.715b vs d, c vs d0.7270.04250-m dash (s)17.00 (0.82)25,5509.440.8117.00 (0.82)13,4719.330.8617.00 (0.82)13,4199.691.0316.98 (0.81)12,7329.661.18474.788a vs b, a vs c, a vs d, b vs c, b vs d0.5080.007Standing long jump (cm)17.00 (0.82)25,550162.5317.8817.00 (0.82)13,497171.6917.8917.00 (0.82)13,425166.8818.7216.98 (0.81)12,789162.7021.16822.869a vs b, a vs c, b vs c, b vs d, c vs d0.0410.055Stand/sit-and-reach (cm)17.00 (0.82)25,55012.775.5717.00 (0.82)13,29810.826.2617.00 (0.82)13,42413.586.4916.98 (0.81)12,88214.147.08717.771a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.2010.0381-min sit-ups (n)17.00 (0.82)25,55021.1710.1817.00 (0.82)13,49332.739.6517.00 (0.82)13,42027.679.9116.98 (0.81)12,79331.6810.735190.252a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.6410.296800-m run (s)17.00 (0.82)25,550229.0123.0017.00 (0.82)13,333241.8926.0117.00 (0.82)13,437252.3228.4716.99 (0.81)12,594257.2634.123903.492a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.9940.861AllBody mass index (kg/m2)12.50 (3.45)53,90017.533.10212.50 (3.45)53,88818.063.088312.44 (3.44)53,10019.013.632790.353b vs c, b vs d, c vs d0.9630.07750-m dash (s)12.49 (3.45)102,56210.111.1912.50 (3.45)53,8549.971.2012.49 (3.45)53,79310.161.2012.44 (3.45)52,45810.031.24279.091a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.049− 0.001Standing long jump (cm)12.49 (3.45)102,562146.7423.5312.50 (3.45)53,923154.7425.1112.50 (3.45)53,804150.5824.9412.44 (3.44)52,558146.0526.381575.503a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.0140.029Stand/sit-and-reach (cm)12.49 (3.45)102,5629.615.5312.50 (3.45)53,5738.005.8212.50 (3.45)53,79410.986.1512.45 (3.43)52,75811.986.994468.133a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.4180.0661-min sit-ups (n)12.49 (3.45)102,56218.8310.6612.48 (3.45)53,90327.6411.3712.49 (3.45)53,76623.6810.6312.45 (3.44)52,49628.2011.2112,094.560*a vs b, a vs c, a vs d, b vs c, b vs d, c vs d0.6990.252 ## Measurements The height was measured to the nearest 0.1 cm by mechanical height and sitting height meter. The subjects did not wear shoes, and their heel, sacrum, and two shoulder blades were in contact with the column in a "three points and one line" standing posture. The weight was measured to the nearest 0.1 kg by an electronic weight meter or lever scale. The subjects stood barefoot in the center of the weight meter for 3 to 5 s, and the value was recorded. Boys wore shorts and girls wore shorts and short-sleeved shirts. BMI was calculated as weight in kilograms divided by height in meters squared [weight(kg)/height(m)2]. Survey participants were given complete physical fitness tests at all survey sites following the same protocol. All physical fitness tests were administered in physical education classes by specially trained physical education teachers who had passed a measurement test. A school physician was present to prevent injuries to children and adolescents during the physical fitness tests, and a program director was present to monitor that the physical fitness tests were conducted as required and to provide the necessary guidance. A group of trained field investigators measured five physical fitness: explosive power (standing long jump, SLJ), speed (50-m dash, D50), flexibility (sit/stand-and-reach, SR), muscular strength, and cardiorespiratory fitness following standardized procedures. The specific test procedures and details are described in previous studies2,35. According to the differences in physical fitness by age and sex, muscular strength was assessed by oblique body pull-ups (OPU) for boys aged 7–12, pull-ups (PU) for boys aged 13–18, and 1-min sit-ups (SU) for girls aged 7–18 years. Cardiorespiratory fitness was assessed by 50-m × 8 shuttle run (50SR) for boys and girls aged 7–12, a 1000-m run (1000R) for boys aged 13–18, and an 800-m run (800R) for girls aged 13–18. From 1985 to 2000, the flexibility test was stand-and-reach. For safety reasons, since 2005, sit-and-reach has been used to measure flexibility. All the measuring instruments were consistent in each survey year and calibrated before use. All the students in the final analysis took each test simultaneously. Nearly $100\%$ of participants performed all tests on the same day. Note that smaller values for speed and endurance fitness tests represent better performance, while larger values for other fitness tests represent better performance. ## Statistical analysis All results for physical fitness were summarized as the mean (M) and standard deviation (SD). Akima splines were used to establish the change curve of each sex-age group, with the x-axis as the years and the y-axis as the results. The sample-weighted linear regression was used to assess the secular trends in means of five fitness components and BMI, with the independent variable being the year and the dependent variable being the test score. The fitting degree was expressed as R-squared, and the regression coefficient (B) represented the value of annual change. Mean differences among all subgroups were tested by one-way analysis of variance (ANOVA) with the Bonferroni post hoc test to verify significance between every two survey years. The level of statistical significance was set at 0.05. All physical fitness indicators from 1985 to 2019 were divided into 3 stages: 1985–2000 as the 1st stage, 2000–2010 as the 2nd stage, and 2010–2019 as the 3rd stage, and the increased range per decade in each sex-age category were calculated for the 3 stages. The calculation formula is:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} & {\text{Average change value of an indicator per decade }}\left({\text{/10a}} \right) \\ & \;\;\; = \left({{\text{average value of the subsequent survey year}} - {\text{average value of the previous survey year}}} \right) \\ & \;\;\;\;\;\;\;/\left({{\text{subsequent survey year}} - {\text{previous survey year}}} \right) \times 10. \\ \end{aligned}$$\end{document}Average change value of an indicator per decade/10a=average value of the subsequent survey year-average value of the previous survey year/subsequent survey year-previous survey year×10. To understand the differences between subgroups (age, sex), data from each survey were divided into 3 age categories, 7–12 years old (primary school), 13–15 years old (junior middle school) and 16–18 years old (junior high school) according to the Chinese educational phases in each sex. Data are processed as above. All analyses were conducted using IBM SPSS version 27.0 (IBM Corp, Armonk, NY, USA) and GraphPad Prism 9.3.1 (GraphPad Software, Inc, CA, USA). ## Ethics statement The studies involving human participants were reviewed and approved by the Medical Research Ethics Committee of the Peking University Health Science Center (IRB00001052-19095). All participants and guardians participated voluntarily and written informed consent by the participant’ legal guardian/next of kin were obtained before the survey. All methods were performed in accordance with relevant guidelines and regulations (detailed rules and regulations of Chinese National Surveillance on Students' Constitution and Health). ## BMI During the period 2000–2019, the BMI of boys and girls in all age groups increased (Fig. 1). There were significant differences in the BMI of boys and girls between the three survey years [boys: $F = 3370.270$ (7–12 years), 1572.592 (13–15 years), 1135.429 (16–18 years), 4703.305 (7–18 years); girls: 2596.387 (7–12 years), 1484.940 (13–15 years), 411.715 (16–18 years), 2790.353 (7–18 years), all $p \leq 0.001$] (Tables 1, 2). In 2019, compared with 2000, the BMI of boys and girls in all age categories increased (all $p \leq 0.05$). During the period 2000–2010 and 2010–2019, the BMI of boys and girls in general increased (all $p \leq 0.05$) but there was an exception in that the BMI of girls aged 16–18 years had no significant change ($p \leq 0.05$) (Tables 1, 2). In terms of the rate of change, the increase accelerated in the 3rd stage for boys and girls in all age categories compared with the 2nd stage, especially in the age of 16–18 (Fig. 7).Figure 1Secular trends (means) in body mass index for Chinese boys and girls aged 7–18 in rural from 1985 to 2019. ## Speed During the entire period, the speed of boys and girls in all age groups improved first, then worsened, and finally rebounded slightly, with slight differences among different age groups (Fig. 2). There were significant differences in the results of D50 of boys and girls between the four survey years [boys: $F = 266.836$ (7–12 years), 1458.596 (13–15 years), 702.666 (16–18 years), 622.830 (7–18 years); girls: 291.723 (7–12 years), 201.148 (13–15 years), 474.788 (16–18 years), 279.091 (7–18 years), all $p \leq 0.001$] (Tables 1, 2). Boys aged 13–15 showed a linear decline in the results of D50 over 34 years ($p \leq 0.05$), which represented the improvement in speed. During the entire 34-year period, in other words, 2019 compared with 1985, boys' and girls' speed improved (decreased in the results of D50) (all $p \leq 0.05$) but there was an exception in that the speed of girls aged 16–18 years worsened ($p \leq 0.05$). From each period, during the period 1985–2000, the speed improved in all age categories and overall for both sexes (all $p \leq 0.05$); during the period 2000–2010, only speed for boys aged 13–15 years had no significant change ($p \leq 0.05$), while that others of age categories improved (all $p \leq 0.05$); during the period 2010–2019, speed for boys aged 13–15 and 7–18 years, girls aged 7–15 and 7–18 years improved, and there were some exceptions that speed of boys aged 7–12 and 16–18 years, girls aged 16–18 years had no significant changes (all $p \leq 0.05$) (Tables 1, 2). In terms of the rate of change, the rate of improvement in the 3rd stage was higher for girls aged 7–18 years than in the 1st stage, while the opposite was observed for boys. The improvement in the 3rd stage for both boys and girls was mainly concentrated in the age categories of 13–15 years, and girls aged 7–12 years also improved to some extent (Fig. 7).Figure 2Secular trends (means) in 50-m dash tests for Chinese boys and girls aged 7–18 in rural from 1985 to 2019. Dots and solid lines are the means and Akima splines, respectively. Upward sloping lines represent poorer performance over time and downward sloping lines represent better performance. ## Explosive power During the entire period, the explosive power of boys and girls in all age groups improved first then worsened (Fig. 3). There were significant differences in the results for SLJ of boys and girls between the four survey years [boys: $F = 1694.614$ (7–12 years), 1846.750 (13–15 years), 1890.707 (16–18 years), 1710.962 (7–18 years); girls: 860.430 (7–12 years), 691.314 (13–15 years), 822.869 (16–18 years), 1575.503 (7–18 years), all $p \leq 0.001$] (Tables 1, 2). In 2019, compared with 1985, the results for SLJ of boys in all age categories improved (all $p \leq 0.05$), while girls in general worsened (13–15 and 7–18 years, all $p \leq 0.05$) but the results for SLJ girls aged 7–12 and 16–18 years had no significant changes (all $p \leq 0.05$). From each period, during the period 1985–2000, the results for SLJ improved in all age categories and overall for both sexes (all $p \leq 0.05$); during the period 2000–2010, the results for SLJ for boys and girls in all age categories worsened (all $p \leq 0.05$) and continued to decline during the period 2010–2019 (Tables 1, 2). In terms of the rate of change, the rate of decline in the 3rd stage was higher for boys and girls aged 7–18 years than in the 2nd stage, especially in boys, and the rate of decline increased to more than 3 times in the 2nd stage. However, the rate of decline in the 3rd stage was significantly faster for boys aged 7–12 and 16–18 years than for boys aged 13–15 years, and the range of decline was very large compared to the 2nd stage. The rate of decline in the 3rd stage for boys aged 13–15 years increased only slightly compared to the 2nd stage. There was little difference in the rate of decline for girls in the three age categories in the 3rd stage (Fig. 7).Figure 3Secular trends (means) in standing long jump tests for Chinese boys and girls aged 7–18 in rural from 1985 to 2019. Dots and solid lines are the means and Akima splines, respectively. Upward sloping lines represent better performance over time and downward sloping lines represent poorer performance. ## Flexibility During the entire period, the flexibility of boys in all age groups worsened first, then improved, and finally worsened. The flexibility of girls in all age groups worsened during the period 1985–2000 and then constantly improved after 2000. There were some differences among the different age groups (Fig. 4). There were significant differences in the results of SR of boys and girls between the four survey years [boys: $F = 474.044$ (7–12 years), 116.352 (13–15 years), 434.280 (16–18 years), 536.586 (7–18 years); girls: 3771.308 (7–12 years), 1027.921 (13–15 years), 717.771 (16–18 years), 4468.133 (7–18 years), all $p \leq 0.001$] (Tables 1, 2). In 2019, compared with 1985, the results for SR of boys aged 13–18 and 7–18 years worsened (all $p \leq 0.05$) but boys aged 7–12 years improved ($p \leq 0.05$). The results for SR of girls in all age categories improved (all $p \leq 0.05$). From each period, during the period 1985–2000, the results for SR worsened but improved during the period 2000–2010 in all age categories and overall for both sexes (all $p \leq 0.05$); during the period 2010–2019, the results for SR of boys in all age categories worsened but the opposite for girls (all $p \leq 0.05$) (Tables 1, 2). In terms of the rate of change, the rates of decline in the 3rd stage were higher for boys in all age categories and overall than in the 1st stage, especially in boys aged 16–18 years. Although the results for SR of girls in all age categories and overall continued to increase during the period 2000–2019, the rates of increase in the 3rd stage were not as fast as those in the 2nd stage (Fig. 7).Figure 4Secular trends (means) in stand/sit-and-reach tests for Chinese boys and girls aged 7–18 in rural from 1985 to 2019. Dots and solid lines are the means and Akima splines, respectively. Upward sloping lines represent better performance over time and downward sloping lines represent poorer performance. ## Muscular strength During the entire period, the muscular strength of girls in all age groups improved first, then worsened, and finally improved. The muscular strength of boys in all age groups improved during the period 1985–2000 and then constantly worsened after 2000 (Fig. 5). There were significant differences in the results of OPU for boys aged 7–12 years, PU for boys aged 13–18 years and SU for girls between the four survey years [boys: $F = 4028.403$ (7–12 years), 313.363 (13–15 years), 1338.465 (16–18 years); girls: 4973.544 (7–12 years), 3467.113 (13–15 years), 5190.252 (16–18 years), 4468.133 (7–18 years), all $p \leq 0.001$] (Tables 1, 2). In 2019, compared with 1985, the results for OPU of boys aged 7–12 years improved but the results for PU of boys aged 13–18 years worsened (all $p \leq 0.05$). The results for SU of girls in all age categories and overall improved (all $p \leq 0.05$). From each period, during the period 1985–2000, the results for muscular strength tests for boys and girls improved but worsened during the period 2000–2010 in all age categories (all $p \leq 0.05$); during the period 2010–2019, the results for tests of boys in all age categories continued to worsen but the opposite for girls (all $p \leq 0.05$) (Tables 1, 2). In terms of the rate of change, the rate of increase in the 3rd stage was lower for girls in general than in the 1st stage but there was an exception in that of girls aged 7–12 years was faster. Although the results for tests of boys in all age categories continued to decline in the 3rd stage, the decline accelerated for boys aged 7–12 years while the decline slowed for boys aged 13–18 years compared to the 2nd stage (Fig. 7).Figure 5Secular trends (means) in muscular strength tests for Chinese boys (7–12-years-old: oblique body pull-ups; 13–18-years-olds: pull-ups) and girls (7–18-years-old: 1-min sit-ups) in rural from 1985 to 2019. Dots and solid lines are the means and Akima splines, respectively. Upwards sloping lines represent better performance over time and downwards sloping lines represent poorer performance. ## Cardiorespiratory fitness During the entire period, the cardiorespiratory fitness of boys and girls in most age groups worsened (the means of tests increased) (Fig. 6). There were significant differences in the results of 50SR for boys and girls aged 7–12 years, 1000R for boys aged 13–18 years and 800R for girls between the four survey years [boys: $F = 41$,119.917 (7–12 years), 2320.016 (13–15 years), 3894.106 (16–18 years); girls: 2495.489 (7–12 years), 3704.561 (13–15 years), 3903.492 (16–18 years), 4468.133 (7–18 years), all $p \leq 0.001$] (Tables 1, 2). Boys and girls aged 7–12 and 16–18 years showed linear upward trends in the results of tests over 34 years (all $p \leq 0.05$), which represented the decline in cardiorespiratory fitness. In 2019, compared with 1985, the cardiorespiratory fitness of boys and girls in all age categories worsened (the means of 50SR, 1000R and 800R increased) (all $p \leq 0.05$). From each period, during the period 1985–2000 and 2000–2010, the cardiorespiratory fitness for boys and girls in all age categories worsened (all $p \leq 0.05$); during the period 2010–2019, the cardiorespiratory fitness of boys and girls 7–12 and 16–18 years continued to worsen but improved for boys and girls 13–15 years (all $p \leq 0.05$) (Tables 1, 2). In terms of the rate of change, the rate of decline in the 3rd stage was lower for boys and girls aged 7–12 and 16–18 years than in the 2nd stage. The rate of decline in the 2nd stage was highest for boys and girls in general (Fig. 7).Figure 6Secular trends (means) in cardiorespiratory fitness tests for Chinese boys (7–12-years-olds: 50-m × 8 shuttle run; 13–18-years-olds: 1000-m run) and girls (7–12-years-olds: 50-m × 8 shuttle run; 13–18-years-olds: 800-m run) in rural from 1985 to 2019. Dots and solid lines are the means and Akima splines, respectively. Upward sloping lines represent poorer performance over time and downward sloping lines represent better performance. Figure 7Change in the mean ($95\%$ confidence interval) differences in the physical fitness test results of Chinese rural children and adolescents in different age categories and genders from 1985 to 2019 per decade. Muscular strength is assessed by oblique body pull-ups for boys aged 7–12, pull-ups for boys aged 13–18, and 1-min sit-ups for girls aged 7–18 years; cardiorespiratory fitness is assessed by 50 m × 8 shuttle run for boys and girls aged 7–12, a 1000-m running for boys aged 13–18, and a 800-m running for girls aged 13–18. From 1985 to 2000, the flexibility test was stand-and-reach. Since 2005, sit-and-reach has been used to measure flexibility. ## Discussion The results showed that during the period 1985–2000, the physical fitness of Chinese rural children and adolescents improved in most aspects except for flexibility, and most of them reached the largest increases; during the period 2000–2010, except for flexibility, all other fitness decreased significantly; during the period 2010–2019, the speed of boys rebounded and flexibility, explosive power and muscular strength continued to decline. The largest rates of decline in fitness were reached in general except for muscular strength in boys aged 13–18 years. Speed, flexibility and muscular strength of girls rebounded and explosive power continued to decline. The cardiorespiratory fitness of boys and girls had significant downward trends in general during the entire period but improved significantly for boys and girls in junior middle school, and the decline slowed down in primary school and junior high school. The largest increases in BMI in boys and girls occurred in the period between 2010 and 2019. In* general, some components of fitness of Chinese children and adolescents in rural has shown positive trends in recent years, especially for girls. The speed for both sexes improved after 2010, which was similar to other Chinese studies2,11,12,35 and a Japanese study (2013–2019)36, which both demonstrated positive trends in recent years. Some studies found that Chinese urban children and adolescents also show positive trends in speed after 2010 as well, but positive trends were more pronounced in rural areas11,12. Speed in children and adolescents aged 8–15 years in Mozambique, Africa, declined continuously from 1992 to 201237. The trends in Slovenia are almost consistent with our study15, with speed quality first decreasing and then rebounding. However, studies from some developed countries6,38–42 showed different results that speed remained stable in children after 2000, having previously improved or worsened or stabilized. There are some exceptions, such as the decline in speed of Portuguese girls from $\frac{2003}{2008}$ to $\frac{2008}{201338}$ and in the Netherlands from 2006 to $\frac{2015}{201743.}$ Our study also found that during the period 2010–2019, the speed of adolescents aged 16–18 years had no significant changes, and some studies also provided some evidence35,44. A systematic review by Fühner et al.45 showed that speed in children and adolescents had been increasing since 2002 and declined to a minimum in the 1980s, while rural Chinese boys bottomed out in 1985 and girls in 2010. Other systematic reviews, which were not quantified, showed inconsistent trends in speed across countries1,44. Moreover, we found significant differences in speed items between countries, with straight-line dashes (e.g., $\frac{30}{50}$/60-m dash) being related to the ability to move quickly; speed-agility (e.g., 10 × 5 m shuttle run) also included the ability to change body position/orientation quickly and accurately in response to stimuli46. Three types of speed were selected in Italy, and the results of dashes remained almost constant over the last 30 years, but the shuttle run declined40. Therefore, we estimate that the differences in secular trends in speed may be related to the test items, although it is known that speed is highly genetically determined1. In short, the current worldwide change trends of speed vary across the world. The explosive power of boys and girls began to improve in 1985 but worsened after 2000, and the rate of decline increased. The trend was similar to other Chinese studies2,11,12,35 in Chinese urban areas11,12. However, we found that the explosive power of rural boys improved from 2015 to 2018 in Shanghai, China13, and improved in Japan in the last decade36. Some developed countries, such as Slovenia (until 2014)21, Lithuania39, Italy41, Poland47, and Brazil48, reported negative trends at the beginning of the twenty-first century or throughout the observed period, with many of these studies showing no significant changes15,39,41,47 or improvement for girls in recent years21. The explosive power of Croatian and German boys had no significant changes and improved for girls slightly since the beginning of the twenty-first century14,49. Slovenia had stable or positive trends (especially for girls) from 2014 to 2019 after the previous downward trend15. Positive trends were observed among boys and girls in the United States in the twentieth century and in Greece in the early twenty-first century50,51. Tomkinson et al. converged on the standing long jump performance of 10,940,801 children and adolescents from 29 countries during the period 1960–2017 and found that the rate of improvement was steady from the 1960s to the 1980s, slowed in the 1990s, and then declined thereafter20, which is generally consistent with the results of this paper. In addition to the above results using the standing long jump as an explosive power test, we also found that the performance of continuous leaping (Finland)52, horizontal jump (Portugal)38, high jump (Netherlands)43, and vertical jump (rural Poland)6 worsened or remained stable or in recent years stabilized52 and, in general, did not show positive trends. Several reviews also pointed to the negative trends of explosive power in most countries1,44, but we found that explosive power remained stable in children and adolescents (especially girls) in some countries around the 2110s. Meanwhile, we found downward trends for girls and boys, and the decline for boys accelerated, suggesting the need to focus on explosive power in boys in China and globally. We observed that flexibility worsened during the period 1985–2000 and then improved during the period 2000–2010 for both sexes, but the trends were different for boys (worsened) and girls (improved) during the period 2010–2019. Overall, the flexibility worsened in boys and improved in girls. The trends were similar to nationwide research and Chinese urban areas11, but there were different trends in some regions of China (first stable then worsened or improved)13,35. The flexibility of boys worsened and improved in girls in Hong Kong from 1998 to 201553 which was similar to our findings A 30-year negative trend was observed in Africa37, but a positive trend was observed in Japan in the previous decade36. For some developed countries, such as Croatia14, Portugal38, Lithuania39, rural Germany42, Dutch43, Poland (Krakow)47, and Brazil48, a large number of studies have shown that flexibility has worsened. Other studies (e.g., rural Poland6, Slovenia15, Italy41, Germany49, Canada54, and Brazil55) found negative or steady trends in boys and steady or positive trends in girls in recent years. Meanwhile, an upward trend was found in Greece for both sexes50. In conclusion, most of the studies found that flexibility worsened, and there were positive trends for girls in recent years which was consistent with this study. In addition, we also observed a significant decline in higher-age boys, the exact reasons for which need to be explored in depth. The trends of muscular strength showed differences among boys and girls with boys having mostly negative trends and girls having mostly positive trends during the period 1985–2019. Muscular strength in boys worsened and in girls improved in recent years. The trends were similar to nationwide research and Chinese urban areas2,11,12. However, Xinjiang, China, found positive trends in recent years only in boys aged 13–18 and worsened in both boys and girls in Hong Kong (sit-ups for both sexes)35,53, which differs from this study and suggests differences between different regions of China. The performance of sit-ups improved in Japan but declined after 201936. In some developed countries, increasing or stable sit-up performance was observed in Croatia14, Slovenia15, Greece50 Portugal38 and Germany42,49,56, and negative trends were found only in Brazil48,55 and the United Kingdom57. On the other hand, for the bent-arm hangs test, we found that most studies showed downward trends, such as in Slovenia15, Dutch43, Lithuania39, and the UK57. We also found a decline in backward overhead medicine ball throws in Poland (Krakow)47. Kaster et al.19 estimated secular trends of sit-up performance for 9,939,289 children and adolescents aged 9–17 years from 31 countries/regions from 1964 to 2017 and found that most countries showed positive trends. Although there was a negative international trend after 2010, the lack of data for a large number of developing countries made the interpretation of the results incomplete19. These different secular trends might be related to the specifics of the different tests used for muscle strength, such as bent-arm hang/flexed-arm hang, pull-up, sit-up, handgrip, etc., which emphasize arm and shoulder belt strength, abdomen strength, etc. Meanwhile, the selection of tests varies between countries and is not always consistent between boys and girls. Cardiorespiratory fitness significantly worsened in Chinese rural children and adolescents. The negative trends were similar to nationwide researchs and Chinese urban areas2,11,35,53, but improvements were also observed for children aged 7–12 years from 2005 to 201411,12. Cardiorespiratory fitness improved in Japan but worsened after 201936. Previous studies have shown negative trends in cardiorespiratory fitness in children and adolescents in many countries around the world2,11,13,14,37,39,48,54,55 and some studies have found that cardiorespiratory fitness has been stable (boys or girls or both sexes) in recent years35,41,42,49,56. However, there are also studies reporting positive trends, such as Greece51, rural Poland (girls)6, and Slovenia15,58. Fühner et al.45 and Tomkinson et al.19 identified stabilization and possible improvement after 2010 or 2000. During the period 2000–2019, the BMI of boys and girls in all age groups increased and the increase accelerated, especially in those aged 16–18 years. It has been shown that both higher and lower BMI can have a detrimental effect on physical fitness22,59. Overweight and obesity individuals tend to perform less physical activity, have longer screen time, consume more calories60 and have a substantial decrease in physical activity. The study from CNSSCH showed that the prevalence and increased rate of overweight and obesity among rural Chinese boys were higher than those among girls24, which may be one of the reasons for the more obvious improvement in girls' physical fitness in recent years. However, a slow or stagnant or negative increase in the prevalence of overweight and obesity was observed in recent years61,62, which partly explains the improvements in some physical fitness in children and adolescents during the period 2010–2019. In addition, some studies have noted that nutritional status is an important factor affecting physical fitness regardless of the levels of physical activity63, and Dong et al.64 also found that children and adolescents with high level of physical activity and high socioeconomic status were associated with better physical fitness, and children and adolescents with obesity and longer TV viewing time were associated with worse physical fitness. Most of these factors were independently and significantly associated with physical fitness. Physical fitness is influenced by several factors and there are also some interactions among these factors, but nutritional status might be the key factor of physical fitness in children and adolescents. In addition, children and adolescents in lower age groups were less resistant to food temptations than those in higher age groups60, and their physical inactivity was more likely to lead to overweight or obesity65. It was worth noting that the secular trends in some fitness (cardiorespiratory fitness, speed, muscular strength) were found to be more “positive” for both sexes during junior middle school than for other educational phases. Other studies did not demonstrate such distinctly different trends across age categories as this paper (possibly due to the age range limitations of the participants or the lack of focus on age differences), and some worldwide studies found more positive trends in sit-ups (with a smaller rate of decline) in recent years in children than in adolescents19, while cardiorespiratory fitness showed almost no difference18. Since participants of our study were divided based on the Chinese educational stage and observed significantly different trends for this age category, we believe that it might be related to some Chinese-specific factors. One study found that the increased rate of overweight and obesity among Chinese adolescents aged 14–17 years was smaller than that of children aged 7–13 years in recent years61, while a study from CNSSCH (Henan, China) showed a rapid increase in the prevalence of overweight and obesity in students aged 10–12 and 16–18 years from 2010 to 201966. From our study, we found that the increase in BMI was faster in adolescents at higher ages from 2010 to 2019, but these findings cannot fully explain the positive trend of adolescents aged 13–15 years in recent years. Therefore, we estimate that another occurrence of this phenomenon is related to the junior high school entrance examination for physical education. Piloting from the twenty-first century, 31 administrative districts in China included the exam of physical education in the total score (physical education and culture scores) of the junior high school examination, signifying the full implementation that the exam of physical education was included among the junior high school entrance examination; initially, the exam of physical education accounted for $5\%$ of the total score of the junior high school entrance examination, and then, in response to the decline in the physical fitness of students nationwide and the need for sports power strategy, the score of the physical education examination was constantly increased (other cultural scores remained almost unchanged) and even doubled in some districts67. In addition, the test programs have become more diverse, but the 1000-m running for boys and 800-m running for girls belong to the mandatory test programs, which may explain the significant improvements in cardiorespiratory fitness in adolescents aged 13–15 years. Although we are unclear how this initiative has impacted them (e.g., voluntary or pressured participation in physical activity), it is clear from the CNSSCH that from 2005 to 2014, the rates of the good and excellent health status of physical fitness rose significantly for 13–15 years than for 16–18 years68,69. On the other hand, they proves some evidence. In contrast, the National College Entrance Examination (NEMT or Gaokao) and junior middle school entrance examination do not include physical education subjects or do not count toward the total score, and students are not highly motivated to exercise. Meanwhile, students at junior high school are under more academic pressure and spend more time being sedentary,70, which may worsen their health status. In the early twenty-first century, the nation and society have paid great attention to the decrease in physical fitness of children and adolescents, and a series of initiatives have been taken. For example, the Healthy Physical Education Curriculum Model of China71, developed by Professor Ji, was introduced to improving students' physical fitness by helping them enjoy physical education and engage in at least one sport as a hobby. The curriculum must focus on the three key elements of "sports loading, physical fitness training and motor skills". Students should have approximately 10 min of physical readiness training in each session, and the exercise intensity should be at least $75\%$ in each session, with an average heart rate of 140–160 beats per minute per session. Currently, the model has been promoted nationwide and a large number of Chinese physical education teachers have been trained to use the model35,72. The government implemented the Opinions of the Central Committee of the Communist Party of China on Strengthening Youth Sports to Enhance the Physical Fitness of Young People in 200745 while emphasizing the importance of school physical education and ensuring that students exercise for one hour every day at school. For the first time, Children and adolescents’ sports health promotion has been elevated to the level of national strategy. In the years that followed, the government issued more than 88 policies including the promotion of sports, reductions in academic burden, and promotion of physical fitness2,73. To improve the health of poor rural students, the state launched the "Nutrition improvement program for rural compulsory education students" in 2011, with the financial department providing nutritional meal subsidies for rural compulsory education students (approximately 7–15 years) in poor areas of central and western areas. In the beginning, each student was provided with a subsidy of 3 yuan per study day, which was increased to 4 yuan in 2014 and 5 yuan in 2021, and the subsidy amounted to 34.8 billion yuan in 202174,75. Through this, the average height and weight of students increased, and the gap between urban and rural areas narrowed; micronutrient deficiencies such as anemia decreased; the intake of foods rich in high-quality protein and micronutrients such as fish, poultry, meat, eggs and milk increased, and rural students’ nutrition levels improved74,75. A study from CNSSCH found an increase in the percentage of children and adolescents meeting one hour of in-school physical activity76. These findings may be related to improvements in physical fitness in recent years. In addition, the Central Committee of the Communist Party of China and the State Council issued the "Outline of the Healthy China 2030 Plan" to improve the physical fitness of the whole population as one of the strategic goals in 201677. In 2019, the Health China Action Promotion Committee issued the "Health China Action (2019–2030)"78, which clearly states that by 2022 and 2030, the proportion of students meeting the national physical fitness standards (National student physical health standard, revised in 2014) will reach at least $50\%$ and $60\%$, respectively. The Physical Education Law of the People's Republic of China, which is amended by the Standing Committee of the 13th National People's Congress on June 24, 2022, stipulates that physical education subjects will be included in NEMT from January 202379. We predict that in the future the physical fitness of Chinese children and adolescents will show more positive trends. Our study spans more than three decades and a long time interval, providing not only a report of secular trends in physical fitness but also an exploration of changes in different year phases. Since there are differences between urban and rural areas, such as economic and political, separate analyses of the physical fitness of rural children and adolescents are beneficial for the development of future rural promotion strategies. In contrast to including only a few age groups, this study includes 12 age groups for each sex from childhood to adulthood and is divided again according to educational phases, which facilitates the characteristics of physical fitness in boys and girls at different stages of growth and development, and our study did find different secular trends in cardiorespiratory fitness. This study has several limitations. First, the muscular strength and cardiorespiratory fitness tests differ across age categories for boys and girls, which does not facilitate comparisons between them. Then, the assessment of rural area being done only in 1985 and urban areas are also possibly included in later study waves. This may increase the differences in economics, policies, etc. between the rural areas selected, and secular trends may not be applicable to all rural areas. Finally, it has been demonstrated that physical activity, nutritional status, and dietary habits can have an effect on physical fitness, and this study did not include these variables or conduct a correlation study. ## Conclusion Our results are nearly consistent with previous studies in China, and it complements the most recent data and evidence from rural areas. Our study showed that from 1985 to 2019, although the physical fitness of children and adolescents in rural China previously experienced negative trends, some components of physical fitness have begun to improve in recent years. At the same time, we also found that certain physical fitness have shown negative trends in recent years, with varying trends for gender and age subgroups. This implies that despite favorable trends over the past decade, there are inequalities in the physical fitness development of Chinese children and adolescents, which may also contribute to future health inequalities, pointing to the need for China to focus on physical fitness and health equity among children and adolescents in the future. Preferential policies for rural areas, promoting physical activity, reducing academic pressures, reducing sedentary time and preventing obesity could all be effective countermeasures. ## References 1. Masanovic B. Trends in physical fitness among school-aged children and adolescents: A systematic review. Front Pediatr. (2020.0) 8 627529. DOI: 10.3389/fped.2020.627529 2. Dong Y. Trends in physical fitness, growth, and nutritional status of Chinese children and adolescents: A retrospective analysis of 1·5 million students from six successive national surveys between 1985 and 2014. Lancet. Child. Adolese. (2019.0) 3 871-880. DOI: 10.1016/S2352-4642(19)30302-5 3. Zhao MM, Zhou ZT, Sun YX, Li J. Correlation between physical fitness and blood pressure in school aged children. Chin. J. Prev. Contr. Chron. Dis. (2022.0) 30 205-208. DOI: 10.16386/j.cjpccd.issn.1004-6194.2022.03.010 4. Ortega FB, Ruiz JR, Labayen I, Lavie CJ, Blair SN. The Fat but Fit paradox: What we know and don’t know about it. Br. J. Sports. Med. (2018.0) 52 151-153. DOI: 10.1136/bjsports-2016-097400 5. Henriksson P. Body composition, physical fitness and cardiovascular risk factors in 9-year-old children. Sci. Rep. (2022.0) 12 2665. DOI: 10.1038/s41598-022-06578-w 6. Bartkowiak S. Physical fitness of rural polish school youth: Trends between 1986 and 2016. J. Phys. Act. Health. (2021.0) 18 789-800. DOI: 10.1123/jpah.2020-0712 7. Zhang YH, Sun JZ, Li N. Analysis of height and weight growth changes of children and adolescents in China from 1943 to 2014. Chin. J. Sch. Health. (2016.0) 37 1578-1581. DOI: 10.16835/j.cnki.1000-9817.2016.10.044 8. Ji CY, Hu PJ, He ZH. Secular growth trends in the Chinese urban youth and its implications on public health. J. Peking. Univ. Health Sci. (2007.0) 2 126-131. DOI: 10.19723/j.issn.1671-167x.2007.02.030 9. Liu ZM, Yang SR, Fang JQ, Li X. Long-term trend of growth and development for primary and middle school students in China from 1985 to 2014. Mod. Prev. Med. (2017.0) 44 3321-3325 10. Wang S, Dong YH, Wang ZH, Zhou ZY, Ma J. Trends in overweight and obesity among Chinese children of 7–18 years old during 1985–2014. Chin. J. Prev. Med. (2017.0) 51 300-305. DOI: 10.3760/cma.j.issn.0253-9624.2017.04.005 11. Wu J, Yuan SM. Dynamic analysis of physical function and fitness of Chinese students from 1985 to 2014. J. Beijing. Sport. Univ. (2019.0) 42 23-32. DOI: 10.19582/j.cnki.11-3785/g8.2019.06.003 12. Ao D, Wu F, Yun CF, Zheng XY. Trends in physical fitness among 12-year-old children in urban and rural areas during the social transformation period in China. J. Adolesc. Health. (2019.0) 64 250-257. DOI: 10.1016/j.jadohealth.2018.08.021 13. Yan YJ, Yan YL. Changes of health and physical fitness of children and adolescents in Shanghai, 2015–2018. Mod. Prev. Med. (2021.0) 48 67-73 & 109 14. Kasović M, Štefan L, Petrić V. Secular trends in health-related physical fitness among 11–14-year-old Croatian children and adolescents from 1999 to 2014. Sci. Rep. (2021.0) 11 11039. DOI: 10.1038/s41598-021-90745-y 15. Radulović A, Jurak G, Leskošek B, Starc G, Blagus R. Secular trends in physical fitness of Slovenian boys and girls aged 7 to 15 years from 1989 to 2019: A population-based study. Sci. Rep. (2022.0) 12 10495. DOI: 10.1038/s41598-022-14813-7 16. Huotari P, Gråstén A, Huhtiniemi M, Jaakkola T. Secular trends in 20 m shuttle run test performance of 14- to 15-year-old adolescents from 1995 to 2020. Scand. J. Med. Sci. Sports. (2022.0). DOI: 10.1111/sms.14290 17. Dooley FL. A systematic analysis of temporal trends in the handgrip strength of 2,216,320 children and adolescents between 1967 and 2017. Sports. Med. (2020.0) 50 1129-1144. DOI: 10.1007/s40279-020-01265-0 18. Tomkinson GR, Lang JJ, Tremblay MS. Temporal trends in the cardiorespiratory fitness of children and adolescents representing 19 high-income and upper middle-income countries between 1981 and 2014. Br. J. Sports. Med. (2019.0) 53 478-486. DOI: 10.1136/bjsports-2017-097982 19. Kaster T. Temporal trends in the sit-ups performance of 9,939,289 children and adolescents between 1964 and 2017. J. Sports. Sci. (2020.0) 38 1913-1923. DOI: 10.1080/02640414 20. Tomkinson GR. Temporal trends in the standing broad jump performance of 10,940,801 children and adolescents between 1960 and 2017. Sports. Med. (2021.0) 51 531-548. DOI: 10.1007/s40279-020-01394-6 21. Đurić S. Secular trends in muscular fitness from 1983 to 2014 among Slovenian children and adolescents. Scand. J. Med. Sci. Sports. (2021.0) 31 1853-1861. DOI: 10.1111/sms.13981 22. Chen G, Chen J, Liu J, Hu Y, Liu Y. Relationship between body mass index and physical fitness of children and adolescents in Xinjiang, China: A cross-sectional study. BMC Public Health (2022.0) 22 1680. DOI: 10.1186/s12889-022-14089-6 23. Hsu CY. Can anthropometry and body composition explain physical fitness levels in school-aged children?. Children (Basel). (2021.0) 8 460. DOI: 10.3390/children8060460 24. Song Y. National trends in stunting, thinness and overweight among Chinese school-aged children, 1985–2014. Int. J. Obes (Lond) (2019.0) 43 402-411. DOI: 10.1038/s41366-018-0129-7 25. 25.Central People's Government of the People's Republic of China. Opinions on implementing the strategy of rural revitalization (accessed 18 September 2022); http://www.gov.cn/zhengce/2018-02/04/content_5263807.htm (2018–02–04). 26. 26.Central People's Government of the People's Republic of China. Decided to launch and implement the nutrition improvement plan for rural compulsory education students (accessed 18 September 2022); http://www.gov.cn/ldhd/2011-10/26/content_1979016.htm. (2011–10–26). 27. Cheng WL. Height, weight and prevalence of overweight and obesity among 10–15 years old children in China, 2010–2016. Chin. J. Public. Health. (2021.0) 37 520-524. DOI: 10.11847/zgggws1126803 28. Lu YH. The association of different sedentary patterns and health-related physical fitness in Pre-schoolers. Front. Pediatr. (2022.0) 9 796417. DOI: 10.3389/FPED.2021.796417 29. Lang JJ. Top 10 international priorities for physical fitness research and surveillance among children and adolescents: A twin-panel delphi study. Sports. Med. (2022.0). DOI: 10.1007/s40279-022-01752-6 30. 30.CNSSCH Association. Report on the 1985th National Survey on Students' Constitution and Health. (People's Educational Publication, 1987). 31. 31.CNSSCH Association. Report on the 2000th National Survey on Students' Constitution and Health. (Higher Educational Press, 2000). 32. 32.CNSSCH Association. Report on the 2010th National Survey on Students' Constitution and Health. (Higher Educational Press, 2012). 33. 33.CNSSCH Association. Report on the 2019th National Survey on Students' Constitution and Health. (Higher Educational Press, 2022). 34. Morgan SL, Dutt AK, Costa FJ, Aggarwal S, Noble AG. The impact of the growth of township enterprises on Rural-Urban transformation in China. The Asian City: Processes of Development, Characteristics and Planning. The GeoJournal Library, 30 (1994.0) 1978-1990 35. Bi C, Zhang F, Gu Y, Song Y, Cai X. Secular trend in the physical fitness of Xinjiang children and Adolescents between 1985 and 2014. Int. J. Environ. Res. Public. Health. (2020.0) 17 2195. DOI: 10.3390/ijerph17072195 36. Kidokoro T, Tomkinson GR, Lang JJ, Suzuki K. Physical fitness before and during the COVID-19 pandemic: Results of annual national physical fitness surveillance among 16,647,699 Japanese children and adolescents between 2013 and 2021. J. Sport. Health. Sci. (2022.0). DOI: 10.1016/j.jshs.2022.11.002 37. Dos Santos FK. Secular trends in physical fitness of Mozambican school-aged children and adolescents. Am. J. Hum. Biol. (2012.0) 27 201-206. DOI: 10.1002/ajhb.22638 38. Costa AM, Costa MJ, Reis AA, Ferreira S, Martins J, Pereira A. Secular trends in anthropometrics and physical fitness of young Portuguese school-aged children. Acta. Med. Port. (2017.0) 30 108-114. DOI: 10.20344/amp.7712 39. Venckunas T, Emeljanovas A, Mieziene B, Volbekiene V. Secular trends in physical fitness and body size in Lithuanian children and adolescents between 1992 and 2012. J. Epidemiol. Community. Health. (2017.0) 71 181-187. DOI: 10.1136/jech-2016-207307 40. Vandoni M. The temporal association between body characteristics and speed performance over twenty-five years in Italian adolescents. Children (Basel). (2022.0) 9 521. DOI: 10.3390/children9040521 41. Lovecchio N, Giuriato M, Carnevale Pellino V, Valarani F, Codella R, Vandoni M. Italian physical fitness decline: A true fact or a mindset? A 10-year observational perspective study. Int. J. Environ. Res. Public. Health. (2020.0) 17 8008. DOI: 10.3390/ijerph17218008 42. Eberhardt T, Bös K, Niessner C. Changes in physical fitness during the COVID-19 pandemic in German children. Int. J. Environ. Res. Public. Health. (2022.0) 19 9504. DOI: 10.3390/ijerph19159504 43. Anselma M, Collard DCM, van Berkum A, Twisk JWR, Chinapaw MJM, Altenburg TM. Trends in Neuromotor fitness in 10-to-12-year-old Dutch children: A comparison between 2006 and 2015/2017. Front. Public. Health. (2020.0) 8 559485. DOI: 10.3389/fpubh.2020.559485 44. Eberhardt T. Secular trends in physical fitness of children and adolescents: A review of large-scale epidemiological studies published after 2006. Int. J. Environ. Res. Public. Health. (2020.0) 17 5671. DOI: 10.3390/ijerph17165671 45. Fühner T, Kliegl R, Arntz F, Kriemler S, Granacher U. An update on secular trends in physical fitness of children and adolescents from 1972 to 2015: A systematic review. Sport. Med. (2021.0) 51 303-320. DOI: 10.1007/s40279-020-01373-x 46. Giuriato M, Codella R, Lovecchio N, Carnevale Pellino V, Vandoni M, Nevill AM. Speed agility trends in children according to growth. Ann. Hum. Biol. (2021.0) 48 271-279. DOI: 10.1080/03014460.2021.1928285 47. Kryst Ł, Żegleń M, Artymiak P, Kowal M, Woronkowicz A. Analysis of secular trends in physical fitness of children and adolescents (8–18 years) from Kraków (Poland) between 2010 and 2020. Am. J. Hum. Biol. (2022.0). DOI: 10.1002/ajhb.23829 48. Gaya AR. Temporal trends in physical fitness and obesity among Brazilian children and adolescents between 2008 and 2014. J. Hum. Sport. Exerc. (2020.0) 15 549-558. DOI: 10.14198/jhse.2020.153.07 49. Hanssen-Doose A. Population-based trends in physical fitness of children and adolescents in Germany, 2003–2017. Eur. J. Sport. Sci. (2021.0) 21 1204-1214. DOI: 10.1080/17461391.2020.1793003 50. Smpokos EA, Linardakis M, Papadaki A, Lionis C, Kafatos A. Secular trends in fitness, moderate-to-vigorous physical activity, and TV-viewing among first grade school children of Crete, Greece between 1992/93 and 2006/07. J. Sci. Med. Sport. (2012.0) 15 129-135. DOI: 10.1016/j.jsams.2011.08.006 51. Pinoniemi BK, Tomkinson GR, Walch TJ, Roemmich JN, Fitzgerald JS. temporal trends in the standing broad jump performance of United States children and Adolescents. Res. Q. Exerc. Sport. (2021.0) 92 71-81. DOI: 10.1080/02701367 52. Jaakkola T, Gråsten A, Huhtiniemi M, Huotari P. Changes in the continuous leaping performance of Finnish adolescents between 1979 and 2020. J. Sports. Sci. (2021.0) 40 1532-1541. DOI: 10.1080/02640414.2022.2091344 53. Poon ET, Tomkinson G, Huang WY, Wong SHS. Temporal trends in the physical fitness of Hong Kong adolescents between 1998 and 2015. Int. J. Sports. Med. (2022.0). DOI: 10.1055/a-1738-2072 54. Colley RC. Trends in physical fitness among Canadian children and youth. Health. Rep. (2019.0) 30 3-13. DOI: 10.25318/82-003-x201901000001-eng 55. Blasquez Shigaki G, Batista MB, Paludo AC, Vignadeli LFZ, Serassuelo Junior H, Ronque ERV. Secular trend of physical fitness indicators related to health in children. J. Hum. Growth. Dev. (2019.0) 29 381-389. DOI: 10.7322/jhgd.v29.9537 56. Spengler S, Rabel M, Kuritz AM, Mess F. Trends in motor performance of first graders: A comparison of cohorts from 2006 to 2015. Front. Pediatr. (2017.0) 5 206. DOI: 10.3389/fped.2017.00206 57. Sandercock GRH, Cohen DD. Temporal trends in muscular fitness of English 10-year-olds 1998–2014: An allometric approach. J. Sci. Med. Sport. (2019.0) 22 201-205. DOI: 10.1016/j.jsams 58. Morrison SA, Sember V, Leskošek B, Kovač M, Jurak G, Starc G. Assessment of secular trends and health risk in pediatric cardiorespiratory fitness from the Republic of Slovenia. Front. Physiol. (2021.0) 12 644781. DOI: 10.3389/fphys.2021.644781 59. Qin G, Qin Y, Liu B. Association between BMI and health-related physical fitness: A cross-sectional study in Chinese high school students. Front. Public. Health. (2022.0) 10 1047501. DOI: 10.3389/fpubh.2022.1047501 60. Zhu SQ, Zhang YJ. Analysis of behavioral risk factors for overweight and obesity among children and adolescents (7–17 years old) in China. Chin. J. Prev. Contr. Chron. Dis. (2022.0) 30 491-496. DOI: 10.16386/j.cjpccd.issn.1004-6194.2022.07.003 61. Guo Y, Yin X, Wu H, Chai X, Yang X. Trends in overweight and obesity among Children and adolescents in China from 1991 to 2015: A meta-analysis. Int. J. Environ. Res. Public. Health. (2019.0) 16 4656. DOI: 10.3390/ijerph16234656 62. Hu X. Trends of overweight and obesity among children and adolescents aged 7–17 in 16 provinces of China from 2000 to 2018. J. Hyg. Res. (2022.0) 51 568-573. DOI: 10.19813/j.cnki.weishengyanjiu.2022.04.012 63. Malicevic S. Is the physical fitness of schoolchildren dependent on their physical activity levels and nutritional status? The experience from Serbia. Nutr. Hosp. (2022.0) 39 506-512. DOI: 10.20960/nh.03861 64. Dong X. Physical activity, screen-based sedentary behavior and physical fitness in Chinese adolescents: A cross-sectional study. Front. Pediatr. (2021.0) 9 722079. DOI: 10.3389/fped.2021.722079 65. Ren SS. Correlation between physical activity and nutritional status among Chinese Children and adolescents. Chin. J. Sch. Health (2022.0) 43 14-18. DOI: 10.16835/j.cnki.1000-9817.2022.01.004 66. Zhang Y. Trends of overweight and obesity prevalence in school-aged children among Henan Province from 2000 to 2019. Front. Public. Health. (2022.0) 10 1046026. DOI: 10.3389/fpubh.2022.1046026 67. Bai YL, Nie RX, Li XD. Historical changes, evolutionary characteristics and future directions of the reform of physical education for junior high school entrance examination. Chin. Exam. (2022.0) 3 15-23. DOI: 10.19360/j.cnki.11-3303/g4.2022.03.003 68. Song Y. Trends of prevalence of excellent health status and physical fitness among Chinese Han students aged 13 to 18 years from 1985 to 2014. Beijing Da Xue Xue Bao Yi Xue Ban (2022.0) 52 317-322. DOI: 10.3760/cma.j.cn112150-20191121-00877 69. Zhang JS. Analysis on the trend of prevalence of excellent and good physical fitness and health status among Chinese Han students aged 13 to 18 years and related influencing factors from 1985 to 2014. Zhonghua Yu Fang Yi Xue Za Zhi (2020.0) 54 981-987. DOI: 10.3760/cma.j.cn112150-20191121-00877 70. Wang D. Improving school physical education to increase physical activity and promote healthy growth of Chinese school-aged children-time for action. J. Sport. Health. Sci. (2017.0) 6 384-385. DOI: 10.1016/j.jshs.2017.10.001 71. Dong CX, Lv HM. Theoretical basis and practical basis for establishing key points of the Healthy Physical Education Curriculum Model of China. Chin. Sport. Sci. (2020.0) 40 24-31. DOI: 10.16469/j.css.202006004 72. Ji L. A Re-study on the theoretical and practical problems of healthy physical education curriculum model of China. J. Beijing. Sport. Univ. (2019.0) 42 12-22. DOI: 10.19582/j.cnki.11-3785/g8.2019.06.002 73. Dong Y. Individual-, family-, and school-level ecological correlates with physical fitness among Chinese school-aged children and adolescents: A national cross-sectional survey in 2014. Front. Nutr. (2021.0) 8 684286. DOI: 10.3389/fnut.2021.684286 74. Ma L. National childhood obesity-related intervention systems and intervention programs in China in 1949 to 2020: A narrative review. Obesity (Silver Spring) (2022.0) 30 320-337. DOI: 10.1002/oby.23316 75. Zhang Q. A decade of review and prospects for improving the nutrition and health of Chinese primary and secondary school students. J. Hyg. Res. (2022.0) 51 696-699. DOI: 10.19813/j.cnki.weishengyanjiu.2022.05.003 76. Yan X. Comparison of status of physical activity time at school and influencing factors in students in China, 2010 and 2014. Zhonghua Liu Xing Bing Xue Za Zhi (2020.0) 41 373-378. DOI: 10.3760/cma.j.issn.0254-6450.2020.03.018 77. 77.Central Committee of the Communist Party of China and the State Council. Outline of the Healthy China 2030 Plan (accessed 18 September 2022); http://www.gov.cn/xinwen/2016‑10/25/content_5124174.htm (2016‑10‑25). 78. 78.Health China Action Promotion Committee. Health China Action (2019–2030) (accessed 18 September 2022); http://www.gov.cn/xinwen/2019-07/15/content_5409694.htm (2019‑07‑15). 79. 79.General Administration of Sport. The Physical Education Law of the People's Republic of China is amended (accessed 18 September 2022); https://www.sport.gov.cn/n31/n20067006/c24405447/content.html (2022‑06‑25).
--- title: A machine learning approach for early prediction of gestational diabetes mellitus using elemental contents in fingernails authors: - Yun-Nam Chan - Pengpeng Wang - Ka-Him Chun - Judy Tsz-Shan Lum - Hang Wang - Yunhui Zhang - Kelvin Sze-Yin Leung journal: Scientific Reports year: 2023 pmcid: PMC10015050 doi: 10.1038/s41598-023-31270-y license: CC BY 4.0 --- # A machine learning approach for early prediction of gestational diabetes mellitus using elemental contents in fingernails ## Abstract The aim of this pilot study was to predict the risk of gestational diabetes mellitus (GDM) by the elemental content in fingernails and urine with machine learning analysis. Sixty seven pregnant women (34 control and 33 GDM patient) were included. Fingernails and urine were collected in the first and second trimesters, respectively. The concentrations of elements were determined by inductively coupled plasma-mass spectrometry. Logistic regression model was applied to estimate the adjusted odd ratios and $95\%$ confidence intervals. The predictive performances of multiple machine learning algorithms were evaluated, and an ensemble model was built to predict the risk for GDM based on the elemental contents in the fingernails. Beryllium, selenium, tin and copper were positively associated with the risk of GDM while nickel and mercury showed opposite result. The trained ensemble model showed larger area under curve (AUC) of receiver operating characteristic curve (0.81) using fingernail Ni, Cu and Se concentrations. The model was validated by external data set with AUC = 0.71. In summary, the results of the present study highlight the potential of fingernails, as an alternative sample, together with machine learning in human biomonitoring studies. ## Introduction Gestational diabetes mellitus (GDM) is one of the most common pregnancy complications threatening both maternal and fetal health1. The prevalence of GDM in China is $14.8\%$, which is the largest worldwide2. GDM is known as impaired glucose tolerance during pregnancy. In healthy pregnant women, the demand of insulin increases to store glucose for later stages of pregnancy. However, the dysfunction of pancreatic β-cell occurred in GDM pregnant women resulted in insufficient of insulin and causes hyperglycemia3. In literature, there is increasing evidence that certain heavy metals in pregnant women is associated with the risk of GDM. For example, several meta-analyses have reported that increased levels of arsenic (As)4, iron (Fe)5 and cadmium (Cd)6 were associated with the risk of GDM. However, contradictory results have been reported. For instance, two recently published studies evaluated the association between multi-elements and GDM. One of them reported that urinary Ni was positively associated with GDM7 while another reported no significant association between Ni and GDM8. More studies are required for confirming an association between heavy metals and GDM. Studies involving nail samples (i.e., fingernails, toenails) in human biomonitoring (HBM) have proliferated in recent years due to the ease of collection and their biological properties compared with blood and urine9. The major advantages of nails over blood and urine are that samples represent long-term accumulation; collection is simple, easy and non-invasive; and storage and transport are also simple, easy, and convenient. Due to the slow growth rate of nails, nails record exposure over periods ranging from a few weeks to more than a year10. The correlation between elemental contents in nail samples and diabetes have been evaluated in various studies. Copper (Cu)11 and selenium (Se)12 were inversely associated with the risks of diabetes and obesity. In contrast, exposure to mercury (Hg) and nickel (Ni)13 increased the risk of diabetes. There has been only one study, however, assessing the relationship between elements in nails and GDM. That study evaluated the correlation between As in toenails and the risk of GDM14. Understanding the correlation between nail elements and GDM may promote the use of nails as a simple, non-invasive way of monitoring the risk for GDM in clinical applications. Machine learning analysis is an emerging trend in the field of HBM. Conventional statistical models describe the features of data based on various assumptions rather than predicting the risk of disease development. In contrast, machine learning aims at developing models through general learning algorithms from data to predict outcomes15. Machine learning has been widely applied for predicting and/or classifying different diseases based on elemental contents in the body. The concentration of six elements in cerebrospinal fluid was used to predict the risk of Parkinson’s disease using Support Vector Machine Model16. In another study, healthy individual and nasopharyngeal carcinoma (NPC) patients were accurately classified according to the elemental contents in their blood serum. This study highlights the potential of early diagnosis of NPC. In terms of GDM prediction by machine learning, several biomarkers in blood have provided acceptable prediction for the risk of GDM17–19. The many known risk factors of GDM, such as age, pre-pregnancy BMI and family history of diabetes20, have also been proved to be valuable in the prediction of GDM using machine learning20–22. However, these risk factors failed to fully elucidate the etiology of GDM. As mentioned previously, elements were associated with the risk of GDM. Machine learning is a good way to explore and establish the predictive value of elements together with conventional risk factors for the incidence of GDM. Moreover, fingernail samples used in this study reflected elemental contents in pregnant women well before the onset of GDM, demonstrating a great potential for early prediction on the risk of GDM. To the best of our knowledge, this is the first study applying machine learning for GDM prediction based on the elemental contents determined in fingernail samples. The present nested case–control study aimed to demonstrate the ability of machine learning to predict GDM based on analysis of elements in fingernails. Twenty-seven elements were monitored in fingernails by ICP-MS after acid digestion. The risk of GDM was predicted by ensemble subspace model using the elemental contents in fingernails as well as the clinical information. The performance of optimized prediction model trained by fingernails elemental contents was also evaluated. ## Basic characteristics The basic characteristics of control and GDM pregnant women are listed in Table 1. No significant associations were observed for the characteristics assessed between control and GDM group. One outlier was observed from the GDM patient fingernail elemental contents while four patient urinary elemental contents were found missing. Hence, only 33 and 30 GDM patients were included in the fingernails and urine statistical analyses, respectively. Table 1Basic characteristics of control and GDM pregnant women. CharacteristicTotal ($$n = 67$$)Control ($$n = 34$$)GDM ($$n = 33$$)p valueEducation level0.33Junior high school10 ($14.9\%$)4 ($12.1\%$)6 ($17.6\%$)High school12 ($17.9\%$)7 ($21.2\%$)5 ($14.7\%$)Junior college20 ($29.9\%$)7 ($21.2\%$)13 ($38.2\%$)University/above25 ($37.3\%$)15 ($45.5\%$)10 ($29.4\%$)Income (CNY per year)0.85 < 100,00019 ($28.4\%$)8 ($24.2\%$)11 ($32.4\%$)100,000–200,00032 (47.8)18 ($54.5\%$)14 ($41.2\%$)200,000–300,0008 ($11.9\%$)3 ($9.1\%$)5 ($14.7\%$)300,000–400,0004 ($6.0\%$)2 ($6.1\%$)2 ($5.9\%$)400,000–500,0004 ($6.0\%$)2 ($6.1\%$)2 ($5.9\%$)Passive smoking0.39Yes16 ($23.9\%$)6 ($18.2\%$)10 ($29.4\%$)No51 ($76.1\%$)27 ($81.8\%$)24 ($70.6\%$)Physical activity pattern0.55Low strength28 ($41.8\%$)13 ($39.4\%$)15 ($44.1\%$)Middle strength37 ($55.2\%$)18 ($54.5\%$)19 ($55.9\%$)High strength2 ($3.0\%$)2 ($6.1\%$)0 ($0.0\%$)Family history of diabetes1.00Yes65 ($97.0\%$)32 ($97.0\%$)33 ($97.1\%$)No2 ($3.0\%$)1 ($3.0\%$)1 ($2.9\%$)Parity0.631 time34 ($50.7\%$)18 ($54.5\%$)16 ($47.1\%$) > 1 time33 ($49.3\%$)15 ($45.5\%$)18 ($52.9\%$)AgeMean (SD)30.7 (± 4.3)30.9 (± 4.1)30.5 (± 4.5)0.65Pre-pregnancy BMIMean (SD)21.9 (± 3.5)22.2 (± 3.7)21.6 (± 3.3)0.29p value of age and pre-pregnancy BMI were calculated by Mann–Whitney U test;p value of education level, income, passive smoking, physical activity pattern, family history of diabetes and parity were calculated by Pearson Chi-square test. ## Elemental contents in fingernails The detection rates, median concentrations, and the interquartile ranges (IQR) of elements in fingernails are summarized in Table 2. The detection rates of beryllium (Be), arsenic (As), molybdenum (Mo), cerium (Ce) and mercury (Hg) were below $90\%$. The concentrations of Be (p value < 0.001), selenium (Se) (p value = 0.003), tin (Sn) (p value = 0.009) and antimony (Sb) (p value = 0.032) in fingernails of GDM patient were significantly higher than those of the control group while the concentration of nickel (Ni) (p value = 0.029) and mercury (Hg) (p value = 0.005) in fingernails of GDM patients showed the opposite trend. Table 2Elemental concentrations in fingernails of control and GDM group. ElementsBelow LOD (%) ($$n = 67$$)Control ($$n = 34$$)GDM ($$n = 33$$)p valueLi0 ($0.0\%$)16.13 (12.70–27.11)15.97 (12.67–20.46)0.866Be27 ($40.3\%$) < LOD (< LOD-0.18)0.16 (0.09–0.74) < 0.001Mg0 ($0.0\%$)90.02 (77.40–96.87)83.17 (72.88–96.29)0.807Al0 ($0.0\%$)19.16 (13.74–30.19)21.09 (15.35–27.85)0.603V0 ($0.0\%$)40.26 (28.83–58.33)41.19 (26.46–65.88)1Cr0 ($0.0\%$)292.10 (184.39–511.32)330.30 (219.68–489.79)0.730Mn0 ($0.0\%$)340.44 (224.68–571.81)321.89 (265.13–500.33)0.885Fe0 ($0.0\%$)26.54 (21.53–33.47)26.21 (20.03–33.51)0.577Co0 ($0.0\%$)16.14 (11.89–23.99)17.53 (13.35–30.27)0.377Ni0 ($0.0\%$)793.81 (505.64–1603.92)409.46 (313.36–890.30)0.029Cu0 ($0.0\%$)3972.89 (3186.18–5258.24)4398.95 (3798.94–5369.89)0.278Zn0 ($0.0\%$)81.42 (76.63–87.00)87.23 (76.33–96.86)0.121As37 ($55.2\%$)3.30 (< LOD-43.76) < LOD (< LOD-21.65)0.306Se0 ($0.0\%$)461.54 (400.94–543.12)532.95 (481.76–665.16)0.003Sr0 ($0.0\%$)770.16 (443.18–1065.63)756.93 (492.83–1093.87)1Mo13 ($19.4\%$)5.95 (1.00–17.79)4.61 (0.45–18.67)0.826Cd0 ($0.0\%$)23.91 (16.75–57.42)21.26 (11.09–32.90)0.158Sn0 ($0.0\%$)315.44 (199.85–576.30)561.75 (295.81–733.55)0.009Sb0 ($0.0\%$)39.27 (27.95–57.71)49.02 (35.15–88.80)0.032Ba0 ($0.0\%$)867.26 (573.90–1375.13)752.44 (552.27–1283.23)0.551La0 ($0.0\%$)10.98 (8.28–16.54)11.47 (7.67–17.02)0.945Ce17 ($25.4\%$)7.62 (< LOD-21.87)7.84 (0.99–16.38)0.884Hg11 ($16.4\%$)92.09 (44.09–141.39)32.43 (0.02–76.56)0.005Tl0 ($0.0\%$)0.37 (0.20–0.47)0.38 (0.26–0.59)0.307Pb0 ($0.0\%$)617.40 (277.10–1073.17)560.91 (364.66–865.72)0.846Bi4 ($6.0\%$)5.34 (2.04–12.38)4.67 (2.83–10.90)0.812U0 ($0.0\%$)3.79 (2.78–6.30)4.99 (2.63–9.35)0.246Concentration presented in median (IQR) (ng/g).Concentration presented in median (IQR) (µg/g).Bolded: p value calculated by Mann–Whitney U test. ## Associations between elemental concentration in fingernails and GDM The association between the elemental concentrations in fingernails and GDM was calculated by logistic regression model. Table 3 lists the crude odd ratio with the $95\%$ confidence interval of each element. Among the 27 elements, Se (OR: 43.49, $95\%$ CI 3.54–847.67), Sn (OR: 2.32, $95\%$ CI 1.13–5.55) and Be (OR: 1.52, $95\%$ CI 1.23–1.97) were positively association with GDM. Ni (OR: 0.50, $95\%$ CI 0.25–0.92) and Hg (OR: 0.73, $95\%$ CI 0.58–0.89) were negatively associated with GDM. We further analyzed the association between elements and the risk of GDM based on the tertiles of elemental concentrations (Table 4). The risk of GDM increased with the concentrations of Be (OR: 8.64, $95\%$ CI 2.04–45.44 in the highest tertile), Cu (OR: 8.08. $95\%$ CI 1.93–41.78 in the second tertile), Se (OR: 4.67, $95\%$ CI 1.23–19.73 in the highest tertile) and Sn (OR: 6.78, $95\%$ CI 1.68–32.34 in the highest tertile). Significant positive dose–response relationships were observed for Be (adjusted p for trend: 0.090) and Sn (adjusted p for trend: 0.090). The risk of GDM decreased with increased concentrations of Ni (OR: 0.020, $95\%$ CI 0.05–0.77 in the highest tertile) and Hg (OR: 0.21, $95\%$ CI 0.05–0.84 in the second tertile and OR: 0.10, $95\%$ CI 0.02–0.41 in the highest tertile). Significant negative dose–response relationship was observed for Hg (adjusted p for trend: 0.081).Table 3Adjusted odd ratio (OR) of elements in fingernails for the risk of GDM.ElementOR ($95\%$CI)p valueAdjusted p valueElementOR ($95\%$CI)p valueAdjusted p valueLi0.71 (0.29, 1.52)0.4050.810Sr1.03 (0.43, 2.51)0.9430.943Be1.52 (1.23, 1.97)0.0000.000Mo0.99 (0.88, 1.11)0.8580.918Mg1.67 (0.16, 20.87)0.6760.810Cd0.56 (0.25, 1.15)0.4210.810Al1.32 (0.51, 3.46)0.5690.810Sn2.32 (1.13, 5.55)0.0370.077V1.22 (0.44, 3.50)0.7030.810Sb2.11 (0.99, 5.31)0.0760.200Cr0.84 (0.34, 1.99)0.6830.810Ba0.87 (0.38, 1.91)0.7200.810Mn1.25 (0.48, 3.37)0.6500.810La0.95 (0.47, 1.84)0.8840.918Fe0.71 (0.16, 2.99)0.6470.810Ce1.03 (0.94, 1.13)0.5300.810Co1.28 (0.51, 3.34)0.5920.810Hg0.73 (0.58, 0.89)0.0030.041Ni0.50 (0.25, 0.92)0.0350.077Tl1.64 (0.62, 4.56)0.3220.810Cu4.26 (0.80, 26.80)0.1010.342Pb1.15 (0.59, 2.30)0.6790.810Zn5.43 (0.40, 141.57)0.2620.390Bi1.08 (0.93, 1.28)0.3320.810As0.95 (0.84, 1.07)0.4210.810U1.42 (0.78, 2.75)0.2720.810Se43.49 (3.54, 847.67)0.0060.041Odd ratio adjusted for education level, income, passive smoking, physical activity pattern, family history of diabetes, parity, age and pre-pregnancy BMI.Bold: p value < 0.05 by logistic regression analysis. FDR correction was indicated in adjusted p value (significant level: p value < 0.1).Table 4Adjusted odd ratio (OR) for the risk of GDM according to the tertiles of fingernail elemental concentration. ElementTertileOR ($95\%$CI)p valueElementTertileOR ($95\%$CI)p valueLiQ1refrefCuQ1refrefQ20.48 (0.13, 1.75)0.274Q28.08 (1.93, 41.78)0.007Q30.90 (0.26, 3.11)0.863Q32.44 (0.65, 9.86)0.193P for trend0.945P for trend0.624BeQ1refrefZnQ1refrefQ22.78 (0.72, 11.85)0.147Q20.30 (0.07, 1.23)0.106Q38.64 (2.04, 45.44)0.006Q32.61 (0.68, 10.74)0.169P for trend0.090P for trend0.493MgQ1refrefAsQ1refrefQ20.68 (0.17, 2.63)0.581Q22.18 (0.61, 8.24)0.235Q30.91 (0.24, 3.48)0.884Q31.03 (0.27, 4.02)0.963P for trend0.945P for trend0.945AlQ1refrefSeQ1refrefQ21.27 (0.34, 4.81)0.720Q23.00 (0.82, 11.93)0.103Q31.61 (0.46, 5.82)0.457Q34.67 (1.23, 19.73)0.028P for trend0.945P for trend0.140VQ1refrefSrQ1refrefQ20.87 (0.25, 3.01)0.822Q21.13 (0.31, 4.22)0.849Q31.38 (0.36, 5.51)0.644Q31.07 (0.27, 4.30)0.925P for trend0.945P for trend0.945CrQ1refrefMoQ1refrefQ21.52 (0.42, 5.74)0.524Q21.05 (0.29, 3.77)0.944Q31.35 (0.32, 5.88)0.679Q31.17 (0.31, 4.46)0.816P for trend0.945P for trend0.945MnQ1refrefCdQ1refrefQ21.55 (0.45, 5.58)0.492Q20.27 (0.06, 1.00)0.056Q31.34 (0.37, 5.02)0.657Q30.32 (0.07, 1.39)0.135P for trend0.945P for trend0.493FeQ1refrefSnQ1refrefQ21.01 (0.27, 3.80)0.985Q22.38 (0.62, 9.88)0.215Q31.09 (0.29, 4.20)0.896Q36.78 (1.68, 32.34)0.010P for trend0.945P for trend0.090CoQ1refrefSbQ1refrefQ22.62 (0.65, 11.71)0.186Q20.94 (0.25, 3.53)0.930Q31.24 (0.34, 4.53)0.746Q32.46 (0.68, 9.67)0.179P for trend0.945P for trend0.493NiQ1refrefBaQ1refrefQ20.38 (0.10, 1.39)0.152Q20.76 (0.21, 2.68)0.669Q30.20 (0.05, 0.77)0.024Q30.74 (0.19, 2.90)0.660P for trend0.140P for trend0.945LaQ1refrefPbQ1refrefQ21.06 (0.29, 3.93)0.928Q21.57 (0.44, 5.78)0.489Q31.06 (0.30, 3.77)0.930Q31.15 (0.28, 4.81)0.845P for trend0.945P for trend0.945CeQ1refrefBiQ1refrefQ22.46 (0.65, 10.08)0.194Q22.88 (0.76, 11.89)0.127Q31.01 (0.27, 3.82)0.989Q30.98 (0.28, 3.44)0.972P for trend0.945P for trend0.945HgQ1refrefUQ1refrefQ20.21 (0.05, 0.84)0.034Q20.76 (0.20, 2.83)0.685Q30.10 (0.02, 0.41)0.003Q32.28 (0.66, 8.28)0.199P for trend0.081P for trend0.606TlQ1refrefQ20.80 (0.21, 2.93)0.735Q31.37 (0.36, 5.27)0.639P for trend0.945Odd ratio adjusted for education level, income, passive smoking, physical activity pattern, family history of diabetes, parity, age and pre-pregnancy BMI.ref: reference group.p for trend was adjusted by FDR correction. Bold indicates p value < 0.05 by logistic regression analysis. ## Prediction performance of ensemble model Machine learning analysis can help predict diseases including GDM. In the present study, we utilized machine learning to evaluate the correlation between multi-element contents and the risk of GDM based on the results from the traditional statistical analysis. According to the logistic regression analysis shown above, Be, Ni, Cu, Se, Sn and Hg were found significantly associated with the risk of GDM. Due to the low detection rate of Be and Hg, they were not included in the machine learning algorithm to minimize the bias generated by the accumulation of larger portion of data points. The training data set was firstly trained by SVM, KNN, DA, ensemble and NB to select the most accurate algorithm for further analysis. The performances of trained models were compared using the AUC of ROC. The average AUC of the ensemble subspace algorithm resulted in the highest AUC (0.78) with the smallest standard deviation among the tested algorithms. Because this indicate a higher reproducibility, it was used in further analysis (Supplementary Fig. S1). In the first attempt, a single element was used to train models; the results are shown in Supplementary Table S5. In order to further enhance the accuracy of the trained models, multi-element models were employed to improve the prediction performance. All possible combinations were evaluated by the ensemble model with the best results summarized in Table 5. The highest AUC obtained was 0.81 using Ni, Cu and Se as predictors. When Sn was added to the trained model, the AUC did not change significantly while the accuracy and sensitivity of models decreased significantly to 0.65 and 0.71 respectively. Hence, the combination of Ni, Cu and Se was evaluated together with the basic characteristics of participants to maximize the prediction accuracy. Finally, six different basic characteristics were added on top of the combination of Ni, Cu and Se to evaluate the prediction performance. However, the prediction performance of trained models deteriorated due to the decrease of sensitivity (Supplementary Table S6). Hence, models trained by Ni, Cu and Se without any basic characteristics were further validated by testing data set in the subsequent study. Table 5Prediction performance of multi-element model trained by fingernail elemental contents. No. of element in groupElementAUCSensitivityAccuracyBalanced accuracyF-valueMatthews correlation coefficientp value of permutation test4Ni, Cu, Se, Sn0.800.650.710.660.600.330.023Ni, Cu, Se0.810.760.780.780.780.57 < 0.012Ni, Cu0.780.530.680.670.620.360.021Cu0.730.720.730.730.720.450.02AUC, area under the receiver operating characteristic curve. No. of permutations: 100. ## Comparison of prediction performance of models constructed from fingernails and urine data The association of urinary elements with GDM and the prediction performance with urinary elements are detailed in Supplementary Table S3. Among the 10 elements (Li, Mg, Ni, Cu, Zn, As, Se, Sr, Mo, Sn) with detection rates at $90\%$ or above, none showed significant association between control and GDM groups. There was also lack of significant dose–response relationships between the target elements and the risk of GDM (Supplementary Table S8). The machine learning analysis of urinary elements was carried out like fingernails. As shown in Supplementary Fig. S2, kNN and ensemble models resulted in similar AUCs. The training process of kNN was much faster than that of the ensemble model; hence the kNN model was used for training models with urinary elemental contents. The best prediction combination was given by urinary Cu, Se and Sn concentrations in addition to pre-pregnancy BMI, physical activity pattern and parity (Supplementary Table S11). After selecting the optimized models for fingernails and urine respectively, the testing data set was used to validate the prediction performance of both models. Figure 1 shows that the trained fingernail model (AUC: 0.71) performed better than the urine model (AUC: 0.49) in predicting GDM.Figure 1ROC curve of best predictions given by nails and urine with ensemble model and KNN model respectively. The prediction performance of trained models were validated by testing data set. The AUC of fingernail and urine prediction models were 0.71 and 0.49, respectively. ## Discussion This is the first study predicting the risk of GDM based on the elemental content of fingernails using a machine learning algorithm. A similar approach has been used to evaluate the risk of GDM based on the metabolites of urine23. Conventional statistical models have been widely applied to evaluate the association between elements and GDM in many studies14,24,25. However, no studies have applied machine learning for this purpose. In the present study, we first used conventional statistical models, and found significant associations of Be, Ni, Se, Sn, Sb, Cu and Hg with GDM (Table 2 to Table 4). We here present the first report of a significant association between Be concentration and GDM. Nevertheless, statistical models cannot conclusively determine the risk of GDM solely by the association with elements. According to Senat, et al., many other basic characteristics including age, pre-pregnancy BMI, and family history of diabetes are general risk factors for GDM26. Passive smoking24; parity25 has also been reported as a potential risk factor. Machine learning can take into account these many factors. Hence, machine learning analysis was implemented to find the hidden pattern in multi-factorial data collected from pregnant women with and without GDM, and then predict the risk of GDM with trained models. Numerous machine learning models for the prediction of GDM have been reported27,28; however, there is no consensus as to which one is best. As shown in Supplementary Fig. S1, the prediction performances of 15 machine learning algorithms were compared using the training data set. Ensemble models and SVM models resulted in similar AUC in the trained models, but the ensemble subspace model was more reproducible, suggesting it would be more reliable for the data in the present study. The major advantage of ensemble models over SVM is that ensemble models use multiple single models to form a new model. As a result, the prediction performance of an ensemble algorithm is usually better than a single algorithm29. After selecting the algorithms, different combinations of elements as well as basic characteristics were used to train models to obtain the highest accuracy. The model was firstly trained by single element content in fingernails. The results of single element models (Supplementary Table S5) show that only the ensemble model trained by Cu level in fingernails provided acceptable prediction performance. Multiple studies have reported that multi-elements exposure is significantly associated with GDM7,8. Hence, we also evaluated the performance of models trained by multiple elements, ranging from two element combinations to four element combinations. Table 5 shows that when the number of elements increased, the prediction performance of the trained model also increased. The trained model was validated by an external testing data set. Figure 1 shows that acceptable accuracy was obtained by the trained model, which suggested that the concentrations of Cu, Ni and Se were important predictors for GDM. In the present study, addition of the basic characteristics of pregnant women did not improve the prediction performance of the machine learning models. It indicated that the models used in the present study worked better for numerical variables but not categorical variables30. The elements used to train the predictive model were similar to most of the other studies. The correlation between circulating Cu level and GDM was summarized using the data from 14 published studies. It was concluded that high serum Cu was positively associated with the risk of GDM, especially among Asians during the third trimester31. Multiple systematic reviews and meta-analyses have focused on the association between Se and GDM. Those studies were consistent in concluding that Se concentrations were low in women with GDM compared with normal women, while the present study shows an opposite trend. The studies involved in those reviews determined serum Se level in either second or third trimesters32–34; while in this study Se levels were measured in the first trimester. Studies reporting the correlation of blood or urinary Ni with GDM are limited, and the conclusions are inconsistent. The present study found significant negative association between fingernail Ni level and GDM while the above mentioned studies reported no significant association8 and positive association7, respectively. Our results show that the correlation between fingernail elements and GDM is different from that of blood and urine. Although the trained model in the present study did not include basic characteristic as predictors, our models highlighted fingernail Cu, Ni and Se concentrations as potential predictors for GDM. To the best of our knowledge, this is the first study demonstrating the prediction of GDM by elemental contents using machine learning. Our model outperformed the models trained by serum triglyceride and fasting plasma glucose level (AUC: 0.68)17. Our trained model also performed comparably to another model trained by cytosine-phosphate-guanine levels in blood (AUC: 0.82)19. Although excellent prediction models constructed by putrescine and microRNA with AUC 0.95 and 0.91, respectively, have been reported, studies using those models did not include external validation by a testing data set18,27. Our prediction model was validated by a testing data set and resulted in AUC 0.71, which indicated acceptable performance. Another major highlight of the present study is that fingernail samples were collected in the first trimester. To date, many studies involving nail samples utilized nail clippings collected either during a later stage of pregnancy or postpartum35. Information obtained from nail samples represents exposure from a few weeks to a few months before collection36. As a result, the association observed using those samples is closely related to the middle to later stage of pregnancy. In contrast, the fingernail samples used in this study represent exposure during the first few weeks of gestation, if not before, which is much earlier than the identification of GDM. But this is what prediction means: Anticipating a* problem before it develops. The model used in this pilot study highlights the ability of fingernail Cu, Ni and Se levels to predict GDM because it was predicting the risk of GDM before the development of GDM. In the present work, we collected both urine and fingernail samples from the same individual and predicted the risk of GDM with their elemental contents through machine learning analysis. One of the major advantages of using fingernails rather than urine is that the elemental detection rate in fingernails is higher than that in urine. The elemental analysis revealed more than $90\%$ of 24 elements in fingernail samples, while the same analysis could detect only 9 elements in urine samples. For fingernails, it should be pointed out that although the detection rates of Be and Hg were relatively low, our model found that they had a significant association with the risk of GDM. In terms of the prediction performance of the trained model, prediction by fingernail elemental contents provided acceptable predictive accuracy for the testing data set while the prediction by urinary elemental contents was similar to random guessing, as the AUC was 0.49 for the external validation result of a urine prediction model (Fig. 1)37. It was mainly due to the low elemental detection rate and no significant difference in elemental concentrations between control and GDM patients for urine sample (Supplementary Table S3). Although it is expected that the use of urine sample will remain dominant in HBM studies, this pilot study highlights that fingernails are a potential alternative sample for predicting the risk of GDM. However, there are several important limitations that should be considered in interpreting the results of the present study. Firstly, the sample size was relatively small. A larger sample (more than 1000 pregnant women in total) will be utilized in the future study to compare the prediction performance of models with other studies38. Secondly, the reason why the results of this study with regard to the correlation between some of the elements with GDM were not consistent with past studies is not known. For example, As content in urine or blood is well-known for its correlation with GDM but no significant association was observed in the present study39,40. To date, there is only one study reported As content in toenails in association with GDM, and it found that As content in toenails collected 2 weeks postpartum was significantly associated with GDM14. Our study utilized fingernails, collected in the first trimester. The influence of type of nails and the specific stage of pregnancy needs to be thoroughly examined in future studies, and other reasons for these inconsistencies need to be explored. Thirdly, the urinary elemental detection rates in the present study were low, which affected the results of machine learning. ## Conclusion To the best of our knowledge, this is the first study demonstrating the application of machine learning analysis to the prediction of GDM using the elemental contents in fingernails. Our study provides additional evidence for the positive association between elemental contents in fingernails and GDM. The results indicate that Ni, Cu and Se concentrations, in particular, in fingernails are important factors for the prediction of GDM by ensemble subspace models. In contrast with fingernails, the elemental contents in urine failed to predict the risk of GDM due to the low detection rate for most of the elements. The present study highlights the potential of GDM prediction in early pregnancy using the elemental contents in fingernails. Further large scale studies are required to verify the correlation of elemental contents in fingernails and GDM. It should be pointed out that long-term exposure information provided by fingernails may also help in elucidating the mechanistic relationship between elements and GDM development. ## Study population This pilot, nested case–control study was based on the Shanghai Maternal-Child Pairs cohort study conducted at the School of Public Health, Fudan University, Shanghai. A cohort of Shanghai pregnant women were recruited from September 2016 to December 2017. Eligible women were those who: [1] were over 20 years old; [2] were free of serious chronic disorders (e.g., diabetes, high blood pressure, heart disease, etc.); [ 3] did not smoke or drink alcohol. Of these, 34 with GDM were recruited for the evaluation of the association between elemental exposure and the risk of GDM. Diagnosis of GDM was performed by oral glucose tolerance test (OGTT) during gestational weeks 24–28. 34 pregnant women without GDM were selected as control group by propensity score matching. The control group was matched with the experimental group in terms of age, living district, pre-pregnancy BMI, family yearly income, education level, infant sex, parity, passive smoking, and physical activity pattern. ## Ethic statement All methods were carried out in accordance with relevant guidelines and regulations. The study was approved by the Institutional Review Boards (IRB) of the School of Public Health, Fudan University (IRB#2016-04-0587). Informed consent was obtained for each participant at the time of enrollment. ## Data collection A face-to-face interview was conducted by a trained nurse with each participant using a standard questionnaire to collect information on age, pre-pregnancy BMI, education level, family yearly income, passive smoking, physical activity pattern, parity and family history of diabetes. Information on the oral glucose tolerance test (OGTT) of pregnant women, and infant sex were retrieved from medical records. ## GDM diagnosis The diagnosis of GDM was based on the guideline published by the Ministry of Health (MOH) of China41. A 75 g OGTT was performed on pregnant women during gestational weeks 24–28. Pregnant women were identified as having GDM if any one of the following criteria was met: fasting plasma glucose ≥ 5.1 mmol/L, 1 h plasma glucose ≥ 10.0 mmol/L or 2 h plasma glucose ≥ 5.1 mmol/L. ## Sample collection Fingernail samples (width larger than 0.2 mm) were provided by pregnant women during 12–16 weeks of gestation. They were asked to collect nails from all fingers with a stainless-steel nail clipper. Fingernail samples were stored in a zip-locked plastic bag at room temperature and transported to Hong Kong Baptist University (HKBU) for elemental analysis. For each individual, the mass of fingernail samples used for ICP-MS analysis ranged from 5 to 10 mg. Urine samples (10 mL) were collected at around 16 weeks of gestation. Samples were stored in polypropylene centrifuge tubes at -80 °C until analysis. ## Elemental analysis The concentration of 27 elements in fingernails was determined by inductively coupled plasma-mass spectrometry (ICP-MS) at HKBU while the concentration of the same 27 elements in urine was determined by ICP-MS in Shanghai. Fingernail samples were washed using the method recommended by the International Atomic Energy Agency (IAEA)42. The subsequent acid-digestion was performed using a modified method reported in our previous study43. In brief, fingernail samples were transferred to 15 mL polypropylene tubes and then rinsed by acetone followed by ultrapure water three times. The washed samples were oven-dried at 60 °C overnight. Samples were then mineralized with 3 mL of concentrated nitric acid and 1 mL of $30\%$ hydrogen peroxide in a microwave digestion system. The digested solution was transferred to an acid-washed beaker and evaporated on a hotplate until nearly dry. 5 µL of 1 µg/mL germanium standard solution was added as an internal standard and ultrapure water was used to make up the sample solution to 5 mL. The solution was then analyzed by ICP-MS. Multi-element standard solutions (Li, Be, Mg, V, Cr, Mn, Ni, As, Se, Mo, Cd, Sn, Sb, La, Ce, Hg, U: 0.2–10 ng/mL; Co, Tl, Bi: 0.04–2 ng/mL; Sr, Pb: 0.8–40 ng/mL; Cu, Ba: 2–100 ng/mL; Fe: 4–200 ng/mL; Al-Zn: 8–400 ng/mL) were prepared by appropriate dilution of 1000 µg/mL of stock standard solution for the quantification of elements by external calibration. Since certified reference material (CRM) for nails was not available, hair CRM was employed for method validation. 5 mg of hair CRM or fingernail sample was digested and analyzed by ICP-MS. The standard solution used in the calibration was analyzed every 20 samples to ensure no significant instrumental drift occurred during the analysis. Urine samples were thawed at room temperature and vortexed (IKA, Germany) for 5 s. Then each sample was centrifuged (Microfuge 16, Beckman Coulter, USA) at 4000 rpm for 1 min to remove debris. 1 mL of supernatant was diluted with $1\%$ (v/v) concentrated nitric acid (HNO3) to 10 mL and vortexed for 30 s to ensure complete mixing. Other relevant details can be found in the Supplementary Materials. ## Statistical analysis Limit of Detections (LODs) were calculated by multiplying the standard deviation of 7 consecutive measurements of blank standard solution by three, and then dividing that number by the slope of the calibration curve. Calculated concentrations below LOD were assigned the value as LOD/2 for further analysis. Elements with detection rates higher than $90\%$ were included in the machine learning algorithm. The basic characteristics of the participants are summarized in Table 1. The education level, income, passive smoking, physical activity pattern, family history of diabetes and parity were compared by Pearson chi-square test. Mann–Whitney U test was used to examine the age and pre-pregnancy BMI. The association between elements and the risk of GDM were evaluated by Mann–Whitney U test and logistic regression analysis. Odd ratio (OR) and $95\%$ confidence intervals (CIs) were calculated with elemental concentration as continuous variables and categorical variables according to the tertile distribution (The first tertile was used as the reference group) for the risk of GDM. The regression model was adjusted for age, pre-pregnancy BMI, education level, income, passive smoking, physical activity pattern, family history of diabetes, and parity. The Pearson correlation between fingernail and urinary elemental concentration was determined by MATLAB R2021b software. Unless otherwise specified, a two-tailed p value < 0.05 was defined as statistically significant. False discovery rate (FDR) was employed for multiple testing with significant level defined at p value < 0.1. Machine learning analysis was performed by MATLAB R2021b software. Maternal age and pre-pregnancy BMI were input as continuous variables while passive smoking, physical activity pattern, parity and family history of diabetes were input as categorical variables. The prediction accuracy of multiple models, including ensemble models, discriminant analysis (DA), support vector machine (SVM), k-nearest neighbor (kNN) and Naive Bayes (NB), were evaluated. The data was split into a training data set consisting of 51 individuals ($75\%$ of total individual) and a testing data set consisting of 16 individuals ($25\%$ of total individual, control to patient ratio = 1:1). Ensemble modeling using a random subspace algorithm resulted in the highest area under curve (AUC) in the receiver operating characteristic (ROC) curve for the trained model; these models were used in all analysis in this pilot study. The trained ensemble models were optimized by the training data set with tenfold cross validation repeated five times. The performance of optimized models was evaluated by the test data set. Based on the result of the logistic regression analysis mentioned above, nickel (Ni), copper (Cu), selenium (Se), and tin (Sn) concentrations in fingernails were input as predictors in the ensemble model. Age, pre-pregnancy BMI, passive smoking, physical activity pattern, parity and family history of diabetes were incorporated in machine learning analysis to further optimize the prediction models. Permutation test was employed as the negative control. The p value of the permutation test was determined as the fraction obtained by the 100 permutations which were higher than the real accuracy44. Balanced accuracy, F-value, Matthews correlation coefficient were also calculated with MATLAB. ## Supplementary Information Supplementary Information. The online version contains supplementary material available at 10.1038/s41598-023-31270-y. ## References 1. Diagnosis and classification of diabetes mellitus. Diabetes Care (2014.0) 37 81-90. DOI: 10.2337/dc14-S081 2. Gao C, Sun X, Lu L, Liu F, Yuan J. Prevalence of gestational diabetes mellitus in mainland China: A systematic review and meta-analysis. J. Diabetes Investig. (2019.0) 10 154-162. DOI: 10.1111/jdi.12854 3. Plows JF, Stanley JL, Baker PN, Reynolds CM, Vickers MH. The pathophysiology of gestational diabetes mellitus. Int. J. Mol. Sci. (2018.0) 19 3342. DOI: 10.3390/ijms19113342 4. Salmeri N. Maternal arsenic exposure and gestational diabetes: A systematic review and meta-analysis. Nutrients (2020.0) 12 1-19. DOI: 10.3390/nu12103094 5. Kataria Y, Wu Y, Horskjær PH, de Mandrup-Poulsen T, Ellervik C. Iron status and gestational diabetes—a meta-analysis. Nutrients (2018.0) 10 1-15. DOI: 10.3390/nu10050621 6. Filippini T, Wise LA, Vinceti M. Cadmium exposure and risk of diabetes and prediabetes: A systematic review and dose-response meta-analysis. Environ. Int. (2022.0) 158 106920. DOI: 10.1016/j.envint.2021.106920 7. Wang X. Exposure to multiple metals in early pregnancy and gestational diabetes mellitus: A prospective cohort study. Environ. Int. (2020.0) 135 105370. DOI: 10.1016/j.envint.2019.105370 8. Wang Y. Multiple metal concentrations and gestational diabetes mellitus in Taiyuan China. Chemosphere (2019.0) 237 124412. DOI: 10.1016/j.chemosphere.2019.124412 9. Gil F, Hernández AF. Toxicological importance of human biomonitoring of metallic and metalloid elements in different biological samples. Food Chem. Toxicol. (2015.0) 80 287-297. DOI: 10.1016/j.fct.2015.03.025 10. Alves A. Human biomonitoring of emerging pollutants through non-invasive matrices: State of the art and future potential. Anal. Bioanal. Chem. (2014.0) 406 4063-4088. DOI: 10.1007/s00216-014-7748-1 11. Sukumar A, Subramanian R. Relative element levels in the paired samples of scalp hair and fingernails of patients from New Delhi. Sci. Total Environ. (2007.0) 372 474-479. DOI: 10.1016/j.scitotenv.2006.10.020 12. Xu R, Chen C, Zhou Y, Zhang X, Wan Y. Fingernail selenium levels in relation to the risk of obesity in Chinese children: A cross-sectional study. Med. (United States) (2018.0) 97 1-5 13. Mehra R, Juneja M. Fingernails as biological indices of metal exposure. J. Biosci. (2005.0) 30 253-257. DOI: 10.1007/BF02703706 14. Farzan SF. Maternal arsenic exposure and gestational diabetes and glucose intolerance in the New Hampshire birth cohort study. Environ. Heal. (2016.0) 15 1-8. DOI: 10.1186/s12940-016-0194-0 15. Bzdok D, Altman N, Krzywinski M. Statistics versus machine learning. Nat. Methods (2018.0) 15 233-234. DOI: 10.1038/nmeth.4642 16. Maass F. Elemental fingerprint: Reassessment of a cerebrospinal fluid biomarker for Parkinson’ s disease. Neurobiol. Dis. (2020.0) 134 104677. DOI: 10.1016/j.nbd.2019.104677 17. Hu M. Elevated serum triglyceride levels at first prenatal visit is associated with the development of gestational diabetes mellitus. Diabetes. Metab. Res. Rev. (2022.0) 38 1-7. DOI: 10.1002/dmrr.3491 18. Liu C. Putrescine as a novel biomarker of maternal serum in first trimester for the prediction of gestational diabetes mellitus: A nested case-control study. Front. Endocrinol. (Lausanne) (2021.0) 12 1-8. DOI: 10.3389/fendo.2021.759893 19. Liu Y, Wang Z, Zhao L. Identification of diagnostic cytosine-phosphate-guanine biomarkers in patients with gestational diabetes mellitus via epigenome-wide association study and machine learning. Gynecol. Endocrinol. (2021.0) 37 857-862. DOI: 10.1080/09513590.2021.1937101 20. Artzi NS. Prediction of gestational diabetes based on nationwide electronic health records. Nat. Med. (2020.0) 26 71-76. DOI: 10.1038/s41591-019-0724-8 21. Gibbone E, Wright A, Campos RV, Anzoategui S, Nicolaides KH. Maternal cardiac function at 19–23 weeks’ gestation in prediction of gestational diabetes mellitus. Ultrasound Obs. Gynecol (2021.0) 58 77-82. DOI: 10.1002/uog.23589 22. Liu H. Machine learning risk score for prediction of gestational diabetes in early pregnancy in Tianjin, China. Diabetes. Metab. Res. Rev. (2021.0) 37 e3397. DOI: 10.1002/dmrr.3397 23. Scott HD. Metabolic dysfunction in pregnancy: Fingerprinting the maternal metabolome using proton nuclear magnetic resonance spectroscopy. Endocrinol. Diabetes Metab. (2021.0) 4 e00201. DOI: 10.1002/edm2.201 24. Leng J. Passive smoking increased risk of gestational diabetes mellitus independently and synergistically with prepregnancy obesity in Tianjin. China. Diabetes. Metab. Res. Rev. (2017.0) 33 1-9 25. Zhou Z. Prospective association of metal levels with gestational diabetes mellitus and glucose: A retrospective cohort study from South China. Ecotoxicol. Environ. Saf. (2021.0) 210 111854. DOI: 10.1016/j.ecoenv.2020.111854 26. Senat MV, Deruelle P. Gestational diabetes mellitus. Gynecol. Obstet. Fertil. (2016.0) 44 244-247. DOI: 10.1016/j.gyobfe.2016.01.009 27. Yoffe L. Early diagnosis of gestational diabetes mellitus using circulating microRNAs. Eur. J. Endocrinol. (2019.0) 181 565-577. DOI: 10.1530/EJE-19-0206 28. Wu YT. Early prediction of gestational diabetes mellitus in the chinese population via advanced machine learning. J. Clin. Endocrinol. Metab. (2021.0) 106 E1191-E1205. DOI: 10.1210/clinem/dgaa899 29. Ren Y, Zhang L, Suganthan PN. Ensemble classification and regression-recent developments, applications and future directions. IEEE Comput. Intell. Mag. (2016.0) 11 41-53. DOI: 10.1109/MCI.2015.2471235 30. Jia H, Cheung YM. Subspace clustering of categorical and numerical data with an unknown number of clusters. IEEE Trans. Neural Networks Learn. Syst. (2018.0) 29 3308-3325. DOI: 10.1109/TNNLS.2017.2728138 31. Lian S, Zhang T, Yu Y, Zhang B. Relationship of circulating copper level with gestational diabetes mellitus: A meta-analysis and systemic review. Biol. Trace Elem. Res. (2021.0) 199 4396-4409. DOI: 10.1007/s12011-020-02566-1 32. Kong FJ, Ma LL, Chen SP, Li G, Zhou JQ. Serum selenium level and gestational diabetes mellitus: A systematic review and meta-analysis. Nutr. J. (2016.0) 15 1-10. DOI: 10.1186/s12937-016-0211-8 33. Askari G, Iraj B, Salehi-Abargouei A, Fallah AA, Jafari T. The association between serum selenium and gestational diabetes mellitus: A systematic review and meta-analysis. J. Trace Elem. Med. Biol. (2015.0) 29 195-201. DOI: 10.1016/j.jtemb.2014.09.006 34. Xu W. The association between serum selenium level and gestational diabetes mellitus: A systematic review and meta-analysis. Diabetes. Metab. Res. Rev. (2022.0). DOI: 10.1002/dmrr.3522 35. White AJ, O’Brien KM, Jackson BP, Karagas MR. Urine and toenail cadmium levels in pregnant women: A reliability study. Environ. Int. (2018.0) 118 86-91. DOI: 10.1016/j.envint.2018.05.030 36. Lum JTS, Chan YN, Leung KSY. Current applications and future perspectives on elemental analysis of non-invasive samples for human biomonitoring. Talanta (2021.0) 234 122683. DOI: 10.1016/j.talanta.2021.122683 37. Fawcett T. An introduction to ROC analysis. Pattern Recognit. Lett. (2006.0) 27 861-874. DOI: 10.1016/j.patrec.2005.10.010 38. Wang J. An early model to predict the risk of gestational diabetes mellitus in the absence of blood examination indexes: Application in primary health care centres. BMC Pregnancy Childbirth (2021.0) 21 1-8. DOI: 10.1186/s12884-021-04295-2 39. Ashley-Martin J. Association between maternal urinary speciated arsenic concentrations and gestational diabetes in a cohort of Canadian women. Environ. Int. (2018.0) 121 714-720. DOI: 10.1016/j.envint.2018.10.008 40. Xia X. Association between serum arsenic levels and gestational diabetes mellitus: A population-based birth cohort study. Environ. Pollut. (2018.0) 235 850-856. DOI: 10.1016/j.envpol.2018.01.016 41. Yang HX. Diagnostic criteria for gestational diabetes mellitus (WS 331–2011). Chin. Med. J. (Engl) (2012.0) 125 1212-1213. PMID: 22613589 42. 42.Ryabukhin, Y. S. Activation analysis of hair as an indicator of contamination. IAEA Rep.IAEA/RL/50, 1–135 (1978). 43. Chen B, Lum JTS, Huang Y, Hu B, Leung KSY. Integration of sub-organ quantitative imaging LA-ICP-MS and fractionation reveals differences in translocation and transformation of CeO2 and Ce3+ in mice. Anal. Chim. Acta (2019.0) 1082 18-29. DOI: 10.1016/j.aca.2019.07.044 44. Pan J. Detecting awareness in patients with disorders of consciousness using a hybrid brain-computer interface. J. Neural Eng. (2014.0) 11 56007. DOI: 10.1088/1741-2560/11/5/056007
--- title: Deficiency of gluconeogenic enzyme PCK1 promotes metabolic-associated fatty liver disease through PI3K/AKT/PDGF axis activation in male mice authors: - Qian Ye - Yi Liu - Guiji Zhang - Haijun Deng - Xiaojun Wang - Lin Tuo - Chang Chen - Xuanming Pan - Kang Wu - Jiangao Fan - Qin Pan - Kai Wang - Ailong Huang - Ni Tang journal: Nature Communications year: 2023 pmcid: PMC10015095 doi: 10.1038/s41467-023-37142-3 license: CC BY 4.0 --- # Deficiency of gluconeogenic enzyme PCK1 promotes metabolic-associated fatty liver disease through PI3K/AKT/PDGF axis activation in male mice ## Abstract Metabolic associated fatty liver disease (MAFLD) encompasses a broad spectrum of hepatic disorders, including steatosis, nonalcoholic steatohepatitis (NASH) and fibrosis. We demonstrated that phosphoenolpyruvate carboxykinase 1 (PCK1) plays a central role in MAFLD progression. Male mice with liver Pck1 deficiency fed a normal diet displayed hepatic lipid disorder and liver injury, whereas fibrosis and inflammation were aggravated in mice fed a high-fat diet with drinking water containing fructose and glucose (HFCD-HF/G). Forced expression of hepatic PCK1 by adeno-associated virus ameliorated MAFLD in male mice. PCK1 deficiency stimulated lipogenic gene expression and lipid synthesis. Moreover, loss of hepatic PCK1 activated the RhoA/PI3K/AKT pathway by increasing intracellular GTP levels, increasing secretion of platelet-derived growth factor-AA (PDGF-AA), and promoting hepatic stellate cell activation. Treatment with RhoA and AKT inhibitors or gene silencing of RhoA or AKT1 alleviated MAFLD progression in vivo. Hepatic PCK1 deficiency may be important in hepatic steatosis and fibrosis development through paracrine secretion of PDGF-AA in male mice, highlighting a potential therapeutic strategy for MAFLD. Phosphoenolpyruvate carboxykinase 1 (Pck1) is an enzyme involved in glucose production that also regulates lipogenesis and has been linked to liver steatosis. Here the authors report that deficiency of Pck1 in the liver leads to nonalcoholic fatty liver disease via activation of the RhoA/PI3K/AKT pathway in a study with male mice. ## Introduction Metabolic dysfunction-associated fatty liver disease or ‘MAFLD’ (formerly known as ‘NAFLD’) is the most common chronic liver disease worldwide, affecting nearly $25\%$ of US and European adults1. MAFLD is defined by the presence of steatosis in >$5\%$ of hepatocytes, regardless of alcohol consumption or other concomitant liver diseases, especially the presence of obesity and T2DM2. It entails a wide spectrum of hepatic clinical conditions, spanning from uncomplicated steatosis to nonalcoholic steatohepatitis (NASH), a more serious form of liver damage hallmarked by irreversible pathological changes such as inflammation, varying degrees of fibrosis, and hepatocellular damage, which is more likely to develop into cirrhosis and hepatocellular carcinoma3. Although multiple parallel insults, including oxidative damage, endoplasmic reticulum stress, and hepatic stellate cell (HSC) activation, have been proposed to explain the pathogenesis of MAFLD, the underlying mechanisms remain unclear4. In gluconeogenesis, glucose is generated from non-carbohydrate substrates, such as glycerol, lactate, pyruvate, and glucogenic amino acids, mainly in the liver, to maintain glucose levels and energy homeostasis. Phosphoenolpyruvate carboxykinase 1 (PCK1) is the first rate-limiting enzyme in gluconeogenesis and converts oxaloacetate to phosphoenolpyruvate in the cytoplasm5. Our previous studies showed that PCK1 deficiency promotes hepatocellular carcinoma progression by enhancing the hexosamine-biosynthesis pathway6. However, PCK1 regulates not only glucose homeostasis but also lipogenesis by activating sterol-regulatory element-binding proteins7. Patients lacking PCK1 function present diffuse hepatic macrosteatosis concomitant with hypoglycemia and hyperlactacidemia8. Similarly, mice with reduced Pck1 expression develop insulin resistance, hypoglycemia, and hepatic steatosis, indicating the important role of PCK1 in regulating both glucose homeostasis and lipid metabolism9,10. However, the precise role of PCK1 in MAFLD progression is not well-understood. The phosphoinositide 3-kinase/protein kinase B (PI3K/ATK) pathway plays a critical role in regulating cell growth and metabolism. This pathway is activated in response to insulin, growth factors, energy, and cytokines and, in turn, regulates key metabolic processes such as glucose and lipid metabolism and protein synthesis11. AKT promotes de novo lipogenesis primarily by activating sterol regulatory element-binding protein12. PI3K/AKT dysregulation leads to numerous pathological metabolic conditions, including obesity and type 2 diabetes13. MAFLD is characterized by dysregulated glucose and lipid metabolism in the liver. Although the PI3K/AKT pathway is a key regulator for sensing metabolic stress, its exact role in MAFLD progression is unclear14,15. In this study, we explored the role of Pck1 in a mouse model. We determined the molecular mechanisms underlying disordered lipid metabolism, inflammation, and fibrosis induced by Pck1 depletion. We also delineated the functional importance of the PI3K/AKT pathway and paracrine secretion of PDGF-AA as its effectors in steatohepatitis, providing a potential therapeutic strategy for treating MAFLD. ## PCK1 is downregulated in patients with NASH and mouse models of MAFLD To determine whether PCK1 is involved in MAFLD, we first examined hepatic gene expression in a published transcriptome dataset [Gene Expression Omnibus (GEO): GSE126848] containing samples from 14 healthy participants, 12 patients with obesity, 15 patients with NAFLD, and 16 patients with NASH16. Bioinformatics analysis showed that 32 genes were markedly changed in obesity, NAFLD, and NASH; 12 genes were considerably downregulated and 20 genes were upregulated (Supplementary Fig. 1a–c). Notably, PCK1 was gradually reduced in patients with obesity, NAFLD, and NASH (Fig. 1a, b). Downregulation of PCK1 mRNA was also observed in a similar dataset (GSE89632) (Fig. 1b). Moreover, immunohistochemistry (IHC) assays showed that hepatic PCK1 protein levels were significantly lower in patients with NASH than in healthy participants (Fig. 1c). Similarly, PCK1 mRNA and protein levels decreased in the liver of mice fed a high-fat diet with drinking water containing fructose and glucose (HFCD-HF/G) for 24 weeks (Fig. 1d, e).Fig. 1PCK1 is downregulated in patients with NASH and mouse models of MAFLD.a Genes downregulated in patients with health ($$n = 14$$), obesity ($$n = 12$$), NAFLD ($$n = 15$$), and NASH ($$n = 16$$) from GSE126848 dataset. b Relative PCK1 mRNA levels of health ($$n = 14$$), obesity ($$n = 12$$), NAFLD ($$n = 15$$), and NASH ($$n = 16$$) in GSE126848 and of health ($$n = 19$$), and NASH ($$n = 24$$) in GSE89632 datasets. The box plots show the medians (middle line) and the first and third quartiles (boxes), whereas the whiskers show 1.5× the IQR above and below the box. Unpaired, two-sided Mann–Whitney U test P values are depicted in the plots, and the significant P value cutoff was set at 0.05. c PCK1 expression in normal individuals and patients with NASH and semi-quantitative analyses of immunohistochemistry (IHC) data (health, $$n = 10$$; NASH, $$n = 36$$). Scale bars: 50 µm. d, e mRNA (d) and protein (e) levels of PCK1 in the livers of WT mice fed with chow diet or HFCD-HF/G ($$n = 6$$). n was the number of biologically independent mice. The samples were derived from the same experiment and the blots were processed in parallel. f, g PCK1 mRNA (f) and protein (g) levels in MIHA cells treated with BSA or PA-BSA. h Relative levels of indicated genes in MIHA cells treated with 0.2 mM PA. i, j Representative ATF3 expression in normal individuals and patients with NASH (i) and semiquantitative analyses of IHC data (j) (health, $$n = 10$$; NASH, $$n = 36$$). Scale bars: 50 µm. k Chromatin immunoprecipitation assays were performed in MIHA cells with or without PA treatment using an antibody against ATF3, IgG, or H3. l Protein levels of PCK1 in MIHA cells infected with either shControl or shATF3 treated with 0.2 mM PA. The samples were derived from the same experiment and the blots were processed in parallel. m Correlation analysis of ATF3 mRNA level with PCK1 in human NAFLD/NASH liver samples (GSE135251, $$n = 206$$). For f and h, $$n = 3$.$ Data are expressed as the mean ± SEM; n.s., not significant. p values obtained via two-tailed unpaired Student’s t tests, one-way analysis of variance with Tukey’s post hoc test, or non-parametric Spearman’s test. Source data are provided as a Source Data file. Next, palmitic acid (PA) was used to mimic the liver steatosis of patients with MAFLD in vitro17. Cell growth was assessed in a CCK8 assay after treatment with different concentrations of PA (Supplementary Fig. 1d). Interestingly, the PCK1 mRNA and protein levels were downregulated in a dose-dependent manner during 24 h PA stimulation (Fig. 1f, g), suggesting that the transcription of PCK1 was inhibited in response to lipid overload. We screened several known regulators of PCK1 (Supplementary Fig. 1e, f) and found that activating transcription factor 3 (ATF3), a transcriptional repressor of PCK118, was upregulated upon PA stimulation (Fig. 1h). Similarly, ATF3 expression was remarkably upregulated in liver samples derived from patients with NASH and MAFLD model mice (Fig. 1i, j, Supplementary Fig. 1g, h). Chromatin immunoprecipitation assays revealed that the binding of ATF3 to the PCK1 promoter increased following PA administration (Fig. 1k). ATF3 knockdown considerably enhanced PCK1 promoter activity and restored PCK1 expression in human hepatocytes treated with PA (Fig. 1l), while overexpression of ATF3 played a opposite effects on PCK1 promoter activity (Supplementary Fig. 1i, j). Furthermore, correlation analysis revealed that ATF3 was negatively correlated with the PCK1 mRNA level based on the GEO database (GEO: GSE135251) (Fig. 1m). These results indicate that increased lipid intake led to the upregulation of the repressor ATF3, impairing PCK1 transcription in patients with NASH and mouse models. ## L-KO mice exhibit a distinct hepatic steatosis phenotype To explore the role of Pck1 in fatty liver disease, wild-type (WT) and liver-specific Pck1-knockout mice (L-KO) mice were fed a chow diet for 24 weeks (Fig. 2a). Hepatic-specific depletion of PCK1 was confirmed by performing immunoblotting (Fig. 2b). Starting at 16 weeks, L-KO mice showed an increased body weight compared with WT mice, however, the results of the glucose tolerance test (GTT) and insulin tolerance test (ITT) did not significantly differ (Fig. 2c). Moreover, increased liver weight was observed in L-KO mice (Fig. 2d). Alanine transaminase (ALT) and aspartate transaminase (AST) levels were higher in L-KO mice, indicating liver injury (Fig. 2e). In addition, total triglyceride (TG), total cholesterol (TC), and free fatty acids (FFAs) in the liver tissues and serum were elevated in L-KO mice compared to those in WT mice (Fig. 2f, g). Histochemistry and enzyme-linked immunosorbent assay (ELISA) showed that L-KO mice had prominent hepatic steatosis, increased inflammatory infiltration, and high levels of TNF-α, whereas hepatic fibrosis was not observed (Fig. 2h–j). Additionally, PCK1 deficiency significantly increased the mRNA levels of genes related to fatty acid transport and inflammation, whereas there were no significant changes in fibrosis-related genes in L-KO mice fed the chow diet (Fig. 2k). These data suggest that L-KO mice exhibited a distinct hepatic steatosis phenotype and liver injury even when fed normal chow. Fig. 2L-KO mice fed chow diet exhibited a distinct hepatic steatosis phenotype.a Schematic diagram of the mouse model fed the chow diet, $$n = 10$$/group. b PCK1 protein expression in WT and L-KO mouse intestine, liver, spleen, kidney, white adipose, and brown adipose confirmed by immunoblotting. The samples were derived from the same experiment and the blots were processed in parallel. This experiment was repeated for three times with similar results. c Body weight, glucose tolerance test (GTT), and insulin tolerance test (ITT) were measured in WT and L-KO mice ($$n = 10$$). d Liver weight of WT and L-KO mice ($$n = 10$$). e–g Determination of alanine aminotransferase (ALT), aspartate aminotransferase (AST), total triglycerides (TG), total cholesterol (TC), and free fatty acid (FFA) levels in the serum or liver tissues ($$n = 10$$). h Paraffin-embedded liver sections were stained with hematoxylin and eosin (H&E), Sirius red, and F$\frac{4}{80.}$ Frozen sections stained with Oil Red O. Scale bars: 50 µm. i Quantification of liver sections of WT and L-KO mice fed the chow diet ($$n = 10$$). j Levels of TNF-α and IL-6 in the liver tissues ($$n = 10$$). k Quantitative PCR analysis of liver mRNA expression ($$n = 10$$). For immunoblotting, the samples were derived from the same experiment and the blots were processed in parallel. n was the number of biologically independent mice. Data are expressed as the mean ± SEM; n.s., not significant. p values obtained via two-tailed unpaired Student’s t tests. Source data are provided as a Source Data file. ## Hepatic loss of Pck1 promotes inflammation and fibrogenesis in MAFLD mice To explore whether an unhealthy diet could exacerbate pathologic changes in L-KO mice, WT and L-KO mice were fed HFCD-HF/G (Fig. 3a)19,20. Starting at 4 weeks, L-KO mice showed significant weight gain (Fig. 3b). The GTT and ITT showed that L-KO mice developed a more severe form of glucose intolerance and insulin resistance compared with those in WT mice (Supplementary Fig. 2a, b). L-KO mice had heavier livers compared with WT mice (Fig. 3c), although there was no significant difference in the liver weight ratio (Supplementary Fig. 2c). Insulin, AST, ALT, TC, TG, and FFAs increased in the serum and liver homogenates of L-KO mice, suggesting more serious liver injury and lipid metabolism disorder (Fig. 3d, e, Supplementary Fig. 2d, e). Analyses of L-KO liver sections revealed increased fat droplets, more severe fibrosis, and greater macrophage infiltration (Fig. 3f, Supplementary Fig. 2f). Furthermore, L-KO mice exhibited higher NAFLD activity scores (NAS score) and higher TNF-α and IL-6 levels (Fig. 3g, h). In addition, the expression of inflammatory factors, lipogenic enzymes, and fibrogenesis-associated genes was upregulated in L-KO mice (Supplementary Fig. 2g). In summary, mice lacking hepatic Pck1 showed substantial liver inflammation and fibrosis after being fed the HFCD-HF/G.Fig. 3PCK1 ablation accelerates inflammation and fibrogenesis in MAFLD model.a Schematic diagram of mouse model fed the HFCD-HF/G. b, c Body weight ($$n = 11$$ per group) and liver weight ($$n = 8$$ per group) were measured in WT and L-KO mice. d, e Serum levels of insulin, alanine aminotransferase (ALT), and aspartate aminotransferase (AST) were measured ($$n = 8$$). f Paraffin-embedded liver sections were stained with hematoxylin and eosin (H&E), Sirius red, α-SMA immunostaining, and F$\frac{4}{80}$ immunostaining. Frozen sections stained with Oil Red O. Scale bars: 50 µm. g NAFLD activity scores (NAS) for each group ($$n = 8$$). h Levels of TNF-α and IL-6 in the liver tissues detected using enzyme-linked immunosorbent assay (ELISA) ($$n = 8$$). i Schematic showing the administration protocol for AAV8-TBG-control or AAV8-TBG-Pck1 in WT and L-KO mice for experiments shown in j–l, $$n = 6$$/group. j Liver weight and body weight. k Paraffin-embedded liver sections were stained with H&E, Sirius red staining and Oil Red O staining. Scale bars: 50 µm. l Quantifications of Sirius red staining, Oil Red O staining, and NAS. Data are expressed as the mean ± SEM; n.s., not significant. p values obtained via two-tailed unpaired Student’s t tests. Source data are provided as a Source Data file. Besides, we used a genetic model with hepatic deficiency in phosphatase and tensin homolog (PTEN) to induce MAFLD. We generated a mouse model with biallelic deletion of Pck1 and Pten in the liver (cPtenf/fPck1f/f) (Supplementary Fig. 3a, b). At 6 months, cPtenf/fPck1f/f mice had higher liver/body weight ratios compared with cPtenf/f mice (Supplementary Fig. 3c). ALT, AST, IL-6, TNF-a, TC, TG, and FFAs levels significantly increased in the serum and liver tissues of cPtenf/fPck1f/f mice, suggesting more serious liver injury and lipid metabolism disorder in cPtenf/fPck1f/f mice (Supplementary Fig. 3d, e). Histological analysis of cPtenf/fPck1f/f liver sections revealed increased fat droplets, more severe fibrosis, and greater macrophage infiltration (Supplementary Fig. 3f, g). In agreement with increased inflammation and fibrosis, elevated expression levels of genes related to inflammation and fibrosis were observed in cPtenf/fPck1f/f mice (Supplementary Fig. 3h). In summary, hepatic Pck1 depletion showed substantial liver steatosis, inflammation and fibrosis in PTEN-null livers. ## AAV-mediated restoration of hepatic PCK1 alleviates the MAFLD phenotype in Pck1-deficient mice We then investigated whether adeno-associated virus (AAV)-based Pck1 replacement therapy could reverse ongoing liver derangement, which is typically observed in patients with MAFLD. After 10 weeks of chow diet or HFCD-HF/G feeding, WT and L-KO mice were injected through the tail vein with AAV serotype 8 (AAV8) vector expressing Pck1 under the control of a liver-specific promoter (thyroxine-binding globulin, TBG), AAV8-TBG-Pck1 or AAV8-TBG-control (Fig. 3i). The expression of Pck1 mRNA and PCK1 protein levels was identified (Supplementary Fig. 4a, b). Interestingly, mice with PCK1 re-expression showed a lower liver weight, body weight, serum liver enzymes, and serum lipid contents (Fig. 3j, Supplementary Fig. 4c, d). Moreover, lipid deposition, inflammation, and fibrosis significantly improved in L-KO mice injected with AAV8-TBG-Pck1 (Fig. 3k-l). Hepatic gene expression analyses indicated that the expression of genes involved in inflammation and liver fibrosis was greatly attenuated by PCK1 restoration in L-KO mice (Supplementary Fig. 4e). Overall, these data support that forced PCK1 expression in the liver protects against MAFLD in mice. ## Transcriptomic and metabolomics analyses confirmed that the loss of Pck1 promotes hepatic lipid accumulation To comprehensively investigate the role of Pck1 deficiency in MAFLD development, we performed RNA-seq analysis of liver samples from L-KO and WT mice fed the normal chow diet or HFCD-HF/G for 24 weeks. Gene Ontology analysis indicated that lipid metabolic processes were remarkably upregulated in L-KO mice fed the HFCD-HF/G (Fig. 4a). The volcano plot showed that genes involved in fatty acid uptake, such as solute carrier family 27 member 1 (Slc27a1) and fatty acid translocase (Cd36), and lipid droplet synthesis, such as cell death-inducing DFFA like effector C (Cidec) and cell death-inducing DFFA like effector A (Cidea), were upregulated in response to the HFCD-HF/G (Fig. 4b). Gene Set Enrichment Analysis (GSEA) revealed that the PPAR signaling pathway was prominently upregulated in L-KO mice fed either diet (Fig. 4c, Supplementary Fig. 5a, b). Several genes selected from the dataset were independently validated by quantitative polymerase chain reaction (qPCR) and immunoblotting and found to be significantly overexpressed in L-KO mice (Fig. 4d, e). Furthermore, genes involved in the glycerol 3-phosphate (G3P) pathway were upregulated in L-KO mice (Fig. 4f). Metabolomics analysis showed that compared with WT mice fed the HFCD-HF/G, L-KO mice had significantly higher G3P and PA levels (Fig. 4g, h). As G3P is a substrate for TG synthesis and PA is a key intermediate metabolite in de novo lipogenesis, Pck1 ablation may promote substrate accumulation for lipid synthesis. Fig. 4Loss of PCK1 promotes lipid accumulation according to transcriptome and metabolome analyses. RNA sequencing was performed on the livers of WT ($$n = 4$$) and L-KO ($$n = 5$$) mice fed the HFCD-HF/G. a Gene Ontology analysis of all significantly changed genes in top 10 biological processes. b Volcano plot representation of significantly up- and downregulated genes. c Gene Set Enrichment Analysis plot (left) of enrichment in “PPAR signaling pathway” signature; heatmap (right) of significantly upregulated PPAR target genes. d, e Quantitative PCR (d) and immunoblot (e) analysis of indicated genes or protein expression in mouse liver tissues. f Relative mRNA expression of key genes in G3P pathway ($$n = 8$$). g Upregulated metabolites detected by untargeted metabolomics ($$n = 6$$). h Relative level of G3P and PA in mouse liver tissues ($$n = 6$$). Data are expressed as the mean ± SEM. p values obtained via two-tailed unpaired Student’s t tests. Source data are provided as a Source Data file. To further examine the function of PCK1 in steatosis in vitro, we overexpressed (PCK1-OE) using the AdEasy adenoviral vector system and knocked out PCK1 (PCK1-KO) using the CRISPR-Cas9 system in human hepatocytes (Supplementary Fig. 5c, d). We found that PCK1-OE attenuated the accumulation of lipid droplets, whereas PCK1-KO facilitated lipid accumulation (Supplementary Fig. 5e, f). Collectively, these results suggest that hepatic Pck1 deficiency leads to lipid accumulation by promoting the expression of lipogenic genes and accumulation of substrates related to lipid synthesis (Supplementary Fig. 5g). ## Hepatic Pck1 deficiency leads to HSC activation via PI3K/AKT pathway RNA-seq analysis indicated that the PI3K/AKT pathway was specifically activated in L-KO mice fed the HFCD-HF/G (Fig. 5a, b). Immunoblotting revealed p-AKT (S473) and p-AKT (T308), which are two activated forms of AKT, and downstream c-MYC were significantly upregulated in both the livers and primary hepatocytes of L-KO mice fed the HFCD-HF/G (Fig. 5c, d). qPCR confirmed that genes related to the PI3K/AKT pathway were highly expressed in L-KO mice (Supplementary Fig. 6a). Similarly, p-AKT (S473 and T308) significantly decreased in human PCK1-OE cells but increased in PCK1-KO cells after 0.2 mM PA treatment (Fig. 5e, f).Fig. 5Hepatic PCK1 deficiency leads to HSC activation via PI3K/AKT pathway.a Pathway enrichment analysis of significantly upregulated genes in L-KO mice. b Gene Set Enrichment Analysis (GSEA) plot of enrichment in PI3K/AKT pathway. c–f Immunoblot analysis of AKT and p-AKT (S473 or T308) in mouse liver tissues (c), primary hepatocytes from HFCD-HF/G feeding mice (d), PCK1-OE (e), and PCK1-KO (f) MIHA cells with or without 0.2 mM palmitic acid (PA) treatment. The samples were derived from the same experiment and the blots were processed in parallel. g Schematic flow chart of co-culture models. h, i Quantitative PCR analysis of fibrosis-related genes in LX-2 cells co-cultured with PCK1-KO (h) or PCK1-OE (i) MIHA cells. j, k Western blotting of fibrosis-related protein in liver tissues (j) or primary HSCs from HFCD-HF/G feeding mice (k) ($$n = 3$$) l, m Relative mRNA expression (l) and immunofluorescence images (m) of ACTA2/α-SMA, COL1A1, and COL3A1 in LX-2 cells co-cultured with PCK1-KO MIHA cells treated with AKT inhibitor MK2206 (10 μM). Scale bars: 25 µm. Data are expressed as the mean ± SEM. p values obtained via two-tailed unpaired Student’s t tests or one-way ANOVA with Tukey’s post hoc test. Source data are provided as a Source Data file. To clarify the role of PI3K/AKT pathway activation, transcriptome data were further analyzed. Interestingly, Col1a1, Col3a1, and Lama2, which are primary components of the extracellular matrix (ECM), were upregulated, as shown in the heat map of the PI3K/AKT pathway (Supplementary Fig. 6b). Moreover, GSEA analysis revealed that ECM-receptor interaction was upregulated in L-KO mice (Supplementary Fig. 6c). Because ECM deposition is typically considered as the key event underlying liver fibrosis, we predicted that the activation of the PI3K/AKT pathway promotes fibrosis in L-KO mice. HSCs are major ECM secretors; thus, we performed co-culture assays with human hepatocyte (MIHA) and human hepatic stellate cell lines (LX-2) (Fig. 5g). Interestingly, the mRNA levels of ACTA2 (α-SMA, an HSC activation marker), COL1A1, and COL3A1 increased in LX-2 cells co-cultured with PCK1-KO cells but decreased in LX-2 cells co-cultured with PCK1-OE cells (Fig. 5h, i). Similarly, COL1A1, COL3A1, and α-SMA expression increased in the liver tissues and primary HSCs of L-KO mice (Fig. 5j, k), which was confirmed in IHC analysis of COL3A1 (Supplementary Fig. 6d). However, these increases were partially reversed by MK2206, an AKT inhibitor (Fig. 5l, m). Collectively, these data suggest that the loss of PCK1 in hepatocytes induces HSC activation and ECM formation by activating the PI3K/AKT pathway. ## Paracrine PDGF-AA from hepatocytes promotes HSC activation Hepatocytes elicit several fibrogenic actions in a paracrine manner to promote HSC activation21. Thus, PCK1-mediated hepatic fibrosis may be involved in paracrine disorders. To test this hypothesis, several pro-fibrotic factors were screened; Pdgfa was found to be significantly elevated in the liver tissues of L-KO mice (Fig. 6a). Bioinformatics analysis confirmed that PDGFA was significantly increased in patients with NAFLD and NASH (Fig. 6b). Pdgfa encodes a dimer disulfide-linked polypeptide (PDGF-AA), and the chronic elevation of PDGF-AA in the mouse liver induces fibrosis22. Immunoblotting and ELISA revealed increased PDGF-AA expression in the liver tissues, primary hepatocytes, and plasma of L-KO mice (Fig. 6c–f). Moreover, the PDGF-AA concentration markedly increased in the culture medium of PCK1-KO cells but decreased in that of PCK1-OE cells treated with 0.2 mM PA (Fig. 6g, h). Correspondingly, platelet-derived growth factor receptor alpha (PDGFRA), which encodes the PDGF-AA receptor, increased in LX-2 cells co-cultured with PCK1-KO cells but decreased in LX-2 cells co-cultured with PCK1-OE cells (Fig. 6i, j). To determine whether the pro-fibrogenic effect was mediated by PDGF-AA secretion, we treated the cells with a neutralizing antibody against PDGF-AA. As expected, the increases in α-SMA, COL1A1, and COL3A1 in LX-2 cells co-cultured with PCK1-KO cells were reversed by anti-PDGF-AA treatment (Fig. 6k).Fig. 6Paracrine PDGF-AA from hepatocytes promotes hepatic stellate cell (HSC) activation.a Expression levels of genes related to fibrogenesis ($$n = 8$$ for each group). b Relative PDGFA mRNA levels of health ($$n = 14$$), obesity ($$n = 12$$), NAFLD ($$n = 15$$), and NASH ($$n = 16$$) in GSE126848 and of health ($$n = 19$$), and NASH ($$n = 24$$) in GSE89632 datasets. The box plots show the medians (middle line) and the first and third quartiles (boxes), whereas the whiskers show 1.5× the IQR above and below the box. Unpaired, two-sided Mann–Whitney U test P values are depicted in the plots, and the significant P value cutoff was set at 0.05. c, d PDGF-AA protein levels in primary hepatocytes (c) or liver tissues (d) detected by western blotting. The samples were derived from the same experiment and the blots were processed in parallel. e, f PDGF-AA levels in liver tissues (e) or serum (f) were detected using enzyme-linked immunosorbent assay (ELISA) ($$n = 8$$). g, h Secreted PDGF-AA levels in conditioned medium with PCK1-KO (g) or PCK1-OE (h) MIHA cells treated with 0.2 mM palmitic acid (PA). i, j mRNA levels of PDGFRA in cell lysate of LX-2 cells co-cultured with PCK1-KO (i) or PCK1-OE (j) MIHA cells treated with PA. k Protein level in LX-2 cells co-cultured with PCK1-KO MIHA cells containing nonspecific rabbit IgG or a PDGF-AA blocking antibody. I Immunohistochemistry (IHC) analysis of PCK1, p-AKT (S473), and PDGF-AA in mouse liver sections (from serial sections). Scale bars: 50 µm. m, n Levels of PDGF-AA (m) or PDGFA (n) in conditioned medium or cell lysate of PCK1-KO MIHA cells treated with AKT inhibitor MK2206 (10 μM). o Protein levels in PCK1-KO MIHA cells treated with AKT inhibitor MK2206 (10 μM). The samples were derived from the same experiment and the blots were processed in parallel. For g, h, i, j, m and n, $$n = 3$.$ Data are expressed as the mean ± SEM; n.s., not significant. p values obtained via two-tailed unpaired Student’s t tests or one-way ANOVA with Tukey’s post hoc test. Source data are provided as a Source Data file. A review of the transcriptome data showed that Pdgfa appeared in the heat map of the PI3K/AKT pathway (Supplementary Fig. 6b). The IHC results showed that p-AKT (S473) was positively correlated with PDGF-AA (Fig. 6l). The AKT inhibitor MK2206 significantly blocked the increase in PDGFA levels in the supernatants or cells lysates of PCK1-KO cells (Fig. 6m–o). Taken together, these data confirm that PCK1 deficiency promoted PDGF-AA expression through the PI3K/AKT pathway and activated HSCs through hepatocyte-HSC crosstalk. ## PCK1 deficiency promotes the activation of the PI3K/AKT/PDGF-AA axis by activating RhoA signaling in hepatocytes Rho GTPases, which cycle between active GTP-bound and inactive GDP-bound conformations, activate the PI3K/AKT pathway23–25. Considering that PCK1 catalyzes the conversion of oxaloacetate to phosphoenolpyruvate, consuming GTP to generate GDP, we predicted that PCK1 deficiency alters intracellular GTP homeostasis. To test this hypothesis, high-performance liquid chromatography (HPLC) analysis was conducted to detect the intracellular levels of GTP. Interestingly, intracellular GTP levels decreased in PCK1-OE cells (Supplementary Fig. 7a) but increased in PCK1-KO cells (Supplementary Fig. 7b). Considering that Rho GTPases are activated when combined with GTP26, we examined the proteins levels of several Rho GTPases in the mouse liver tissues and found that GTP-bound RhoA significantly increased and inactivated RhoA, p-RhoA (S188), decreased in L-KO mice (Fig. 7a–c). Similar results were observed in the primary hepatocytes of mice fed the HFCD-HF/G (Supplementary Fig. 7c). Consistently, after PA treatment, the levels of GTP-bound RhoA decreased and p-RhoA (S188) expression increased in PCK1-OE cells, whereas the opposite results were observed in PCK1-KO cells (Fig. 7d–g). Moreover, we found that the addition of 5’-GTP, 2Na+ (salt of guanosine triphosphate) activated the RhoA/PI3K/AKT pathway in primary hepatocytes (Supplementary Fig. 7d). However, inhibition of de novo guanine nucleotide synthesis by mycophenolic acid (MPA) robustly inhibited the RhoA/PI3K/AKT pathway by decreasing the extent of phosphorylated AKT (Supplementary Fig. 7e). Next, RhoA inhibitor (Rhosin) or shRhoA was used to determine whether PI3K/AKT activation depends on RhoA. Immunoblotting, ELISA, and qPCR assays showed that both Rhosin and shRhoA blocked the increase in the activated forms of AKT and PDGF-AA in the PCK1-KO cell lysate and supernatant, as well as ACTA2, COL1A1, and COL3A1 expression in LX-2 cells co-cultured with PCK1-KO hepatocytes (Fig. 7h–j, Supplementary Fig. 7f–h). To further evaluate the involvement of RhoA in HSC activation, we isolated primary hepatocytes from HFCD-HF/G-fed WT or L-KO mice; treated the cells with MK2206, Rhosin, or dimethyl sulfoxide vehicle; and then co-cultured the cells with primary HSCs from chow-fed WT mice. The activation of co-cultured HSCs was partially eliminated by the inhibition of AKT or RhoA in primary hepatocytes (Supplementary Fig. 7i). Moreover, PCK1 and p-RhoA (S188) were downregulated in samples from patients with NASH, whereas p-AKT (S473) and PDGF-AA levels were upregulated (Fig. 7k, l). These data indicate that PCK1 ablation stimulated the PI3K/AKT/PDGF-AA axis by activating RhoA.Fig. 7PCK1 deficiency promotes activation of PI3K/AKT/PDGF-AA axis by activating RhoA in hepatocytes.a Immunoblotting analysis of indicated proteins in mouse liver tissues. The samples were derived from the same experiment and the blots were processed in parallel. b Immunohistochemistry (IHC) analysis of p-RhoA (S188) in mouse liver tissues. Scale bars: 50 µm. c–e Relative levels of active RhoA were measured using G-LISA colorimetric RhoA activation assay in mouse liver tissues (c) ($$n = 8$$), PCK1-KO (d) and PCK1-OE (e) MIHA cells treated with 0.2 mM palmitic acid (PA). f, g Immunoblots of p-RhoA (S188) and RhoA in PCK1-KO (f) and PCK1-OE (g) MIHA cells with or without 0.2 mM PA treatment. The samples were derived from the same experiment and the blots were processed in parallel. h Expression of indicated proteins in PCK1-KO MIHA cells after addition of Rhosin (30 µM). i Levels of PDGF-AA in the supernatant of PCK1-KO MIHA cells treated with Rhosin (30 µM). j Relative mRNA expression of ACTA2, COL1A1, and COL3A1 in LX-2 cells co-cultured with PCK1-KO MIHA cells treated with Rhosin (30 µM). k IHC analysis of PCK1, p-RhoA (S188), p-AKT (S473), and PDGF-AA in normal individuals and patients with NASH (from serial sections). Scale bars: 50 µm. l Semi-quantitative analyses of immunohistochemistry data of health and NASH human tissues for indicated proteins (health, $$n = 10$$, NASH, $$n = 10$$). The cell culture experiments were repeated for three times independently with similar results. For d, e, i and j, $$n = 3$.$ Data are expressed as the mean ± SEM. p values obtained via one-way ANOVA with Tukey’s post hoc test. Source data are provided as a Source Data file. ## Genetic or pharmacological disruption of RhoA and AKT1 reduced progressive liver fibrosis in vivo To explore whether blocking RhoA/PI3K/AKT could rescue the MAFLD phenotype in L-KO mice, the genetic and pharmacological disruption of RhoA and AKT1 was performed in vivo (Fig. 8a). Pharmacological inhibition of AKT1 or RhoA led to improved glucose intolerance (Supplementary Fig. 8a) and insulin resistance (Supplementary Fig. 8b). The increase in liver weight was also prevented (Fig. 8b), whereas the body weight decreased only in the MK2206 treatment group (Supplementary Fig. 8c). Additionally, Rhosin or MK2206 administration attenuated the levels of inflammatory factors (e.g. AST, ALT, IL-6, and TNF-α) as well as TG and FFA levels, in the serum and liver tissues (Fig. 8c, Supplementary Fig. 8d, e). Similarly, histochemistry revealed reduced liver steatosis, inflammation, and fibrosis in Rhosin- or MK2206-treated mice (Fig. 8d, Supplementary Fig. 8f). Additionally, α-SMA, COL1A1, COL3A1, PDGF-AA, and p-AKT (S473, T308) expression and GTP-bound RhoA levels decreased, whereas p-RhoA (S188) expression increased in the treatment group (Supplementary Fig. 8g, h). MK2206 or Rhosin treatment also reduced the expression of genes related to inflammation and fibrosis (Supplementary Fig. 8i). These results indicate that the pharmacological inhibition of AKT or RhoA can alleviate the clinical phenotypes of MAFLD.Fig. 8Pharmacological inhibition or genetic silencing of AKT1 or RhoA alleviates MAFLD development in vivo. L-KO mice were fed the HFCD-HF/G for 24 weeks, and therapeutic treatments were initiated at different times. a Schematic diagram of in vivo pharmacological inhibition (b–d) and pSECC lentivirus-mediated silencing (e–g) of AKT1 or RhoA. b, c Liver weight (b), and serum alanine aminotransferase (ALT), aspartate aminotransferase (AST), TNF-α, and IL-6 (c) (DMSO, $$n = 6$$; MK2206, $$n = 6$$; Rhosin, $$n = 5$$). d Paraffin-embedded liver sections were stained with hematoxylin and eosin, Sirius Red, or immunostained for F$\frac{4}{80}$, COL3A1, and α-SMA. Frozen sections were stained with Oil Red O. Scale bars: 50 µm. e Quantification of body weight and liver weight in pSECC-sgAkt1 and pSECC-sgRhoA L-KO mice ($$n = 6$$). f Plasma levels of total triglycerides (TG), total cholesterol (TC), and free fatty acids (FFA) ($$n = 6$$). g NAFLD activity scores of liver sections ($$n = 6$$). h Model depicting the critical role of PCK1 in controlling MAFLD progression. n was the number of biologically independent mice. Data are expressed as the mean ± SEM; n.s., not significant. p values obtained via one-way ANOVA with Tukey’s post hoc test. Source data are provided as a Source Data file. Next, we used pSECC, a lentiviral-based system that combines the CRISPR system and Cre recombinase, to silence RhoA or AKT1 in L-KO mice27. In agreement with the results of the experiments described above, AKT or RhoA depletion in vivo partially prevented and liver weight and body weight gain (Fig. 8e). Moreover, serum AST, ALT, TC, TG, and FFA levels significantly decreased in sgAkt1- or sgRhoA-treated mice compared with those in sgCtrl-treated L-KO mice (Fig. 8f, Supplementary Fig. 9a). Histological analysis revealed that the NAS score, hepatic fibrosis, and F$\frac{4}{80}$+ macrophage counts decreased in sgAkt1 and sgRhoA mice, indicating the attenuation of the MAFLD phenotypes (Fig. 8g, Supplementary Fig. 9b, c). Consistently, genes or proteins involved in hepatic inflammation and fibrosis were significantly downregulated in the livers of sgAkt1 and sgRhoA mice, accompanied by the reduced expression of PDGF-AA (Supplementary Fig. 9d, e). These results indicate that the genetic inhibition of AKT or RhoA can alleviate the clinical phenotypes of MAFLD. ## Discussion We found that the hepatic gluconeogenic enzyme PCK1 plays an important role in MAFLD progression. The expression of PCK1 was diminished in the livers of patients or mice with MAFLD. Moreover, the deletion of PCK1 significantly exacerbated hepatic steatosis, fibrosis, and inflammation in mouse models fed the HFCD-HF/G. Mechanistically, loss of PCK1 not only promotes steatosis by enhancing lipid deposition, but also induces fibrosis through HSC activation via paracrine secretion of PDGF-AA, thus promoting MAFLD progression (Fig. 8h). Abnormal lipid metabolism is characteristic of MAFLD. Previous studies predicted that altered lipid homeostasis was caused by abnormal expression of genes related to lipid metabolism28. However, recent studies demonstrated that disruption of gluconeogenesis also leads to abnormal lipid metabolism. A deficiency of fructose-1,6-bisphosphatase 1 and glucose-6-phosphatase catalytic subunit, which are key enzymes in gluconeogenesis, results in severe hepatic steatosis and hypoglycemia, indicating that the suppression of gluconeogenesis also disrupts lipid homeostasis17,29. As the first rate-limiting enzyme in gluconeogenesis, it is currently unclear whether PCK1 plays a critical role in MAFLD development. We identified a robust decrease in PCK1 expression in the livers of MAFLD mice and patients with NAFLD/NASH, causing severe hepatic steatosis and confirming that disordered hepatic gluconeogenesis affects lipid homeostasis. Previous reports showed that PCK1 expression is increased in several obesity/diabetes mouse models, such as ZDF rats and ob/ob and db/db mice, and the disease progression of MAFLD is positively correlated with obesity and type 2 diabetes mellitus30–32. Interestingly, we found that PCK1 expression was downregulated in a diet-induced murine model. This discrepancy may be related to differences in the animal models used in different studies. Widely used rodent models for genetic forms of obesity and diabetes, such as ob/ob and db/db mice, exhibit increased plasma glucocorticoids, which may drive PCK1 expression32,33. Another explanation is that the high-fat diet supplemented with high fructose/glucose in drinking water might suppress PCK1 expression34. Under a high-fat diet, PA decreased PCK1 expression via SIRT3 inhibition35. Additionally, acetylation, ubiquitination, and phosphorylation modulate PCK1 expression36. A high glucose level was reported to destabilize PCK1 by stimulating its acetylation, thus promoting its ubiquitination and subsequent degradation37. In this study, we found that ATF3, a member of the basic leucine zipper family of transcription factors38, transcriptionally repressed Pck1 upon PA overload in vitro and in the mouse model. This result agrees with those of previous studies suggesting that ATF3 is upregulated in patients with NAFLD and murine models and inhibits the expression of PCK1 in alcoholic fatty liver disease17,39,40. Therefore, in this study, we found that PCK1 markedly decreased in MAFLD, and PA inhibited PCK1 transcription by upregulating ATF3. Numerous studies using PCK1 agonists or whole-body Pck1 knockdown mice have verified that PCK1 can affect lipid metabolism41,42. In the present study, liver-specific Pck1 knockout induced significant hepatic steatosis even under normal feeding conditions. This is very important because it is uncommon for single-gene ablation to cause spontaneous steatosis unless a high-fat diet is used. Moreover, mice with liver Pck1 deficiency present aggravated inflammation when fed a high-fat high-fructose diet, contrasting with the results of a previous study showing that whole-body Pck1 knockdown prevents hepatic inflammation43. This discrepancy may be related to differences between diets and animal models, as whole-body Pck1 knockdown may have unexpected effects on glucolipid metabolism. The macronutrient composition of the diet, the duration of feeding, and the genetic background of the animals are all important variables in determining disease severity in preclinical MAFLD models44. To establish an ideal MAFLD model, several attempts have been made to use HFD in combination with fructose and sucrose-enriched drinking water45,46. In this study, we have constructed the MAFLD model using the mouse strain derived from the C57BL/6 J and 129S6/SvEvTac, which may serve as a suboptimal model to investigate advanced stages of MAFLD. In light of the fact that MAFLD prevalence is higher in men than in women at all ages47, we initially used male mice, but not female mice, in the present studies; therefore, the generalizability of the findings to female mice is not guaranteed. Lipid accumulation is characteristic of steatosis. Emerging evidence indicates that increased fatty acid uptake is associated with lipid accumulation48,49. Previous studies demonstrated that the loss of PCK1 in the liver disturbed hepatic cataplerosis and led to the accumulation of TCA cycle intermediates. The slowed TCA cycle impaired fatty acid oxidation, resulting in fat accumulation in the liver50. Additionally, elevated plasma TGs and FFAs could contribute to body weight gain in L-KO mice. Notably, HFCD-HF/G-induced glucose intolerance and insulin resistance in L-KO mice play essential roles in obesity and whole-body metabolism. In this study, genes involved in fatty acid uptake such as Cd36 and Slc27a1 were highly expressed in L-KO mice. In addition, a lipid droplet-associated protein Cidec was increased by both the chow and HFCD-HF/G diet, and was shown to be upregulated in patients with NAFLD and L-KO mice, suggesting that PCK1 ablation promotes lipid droplet formation51,52. Abnormal levels of metabolites also contribute to TG accumulation in the liver, with the G3P pathway contributing to over $90\%$ of TG synthesis53. As our metabolomics data showed that G3P and PA were significantly upregulated in L-KO mice, PCK1 deficiency may promote hepatic lipid accumulation by enhancing the expression of Cd36, Slc27a1, and Cidec and the levels of metabolic substrates such as G3P and PA. However, the precise mechanism by which PCK1 regulates the G3P pathway and expression levels of Cd36 and Slc27a1 must be further analyzed. Fibrosis is another characteristic of MAFLD and drives the transition from simple steatosis to NASH. Activation of HSCs through the secretion of profibrotic cytokines, such as TGF-β and PDGF, is a key event in liver fibrosis54. A recent study identified high mobility group protein B1, secreted by fructose-1,6-bisphosphatase 1-deficient hepatocytes, as the main mediator activating HSCs, revealing important crosstalk between hepatocytes and HSCs via paracrine signaling. Herein, PDGF-AA was secreted by PCK1-deficient hepatocytes and acted in a paracrine manner to activate HSCs. Increased deposition of ECM and activation of HSCs were observed in PDGFA-transgenic mice; however, the mechanism mediating PDGF-AA upregulation in fibrosis remains unclear22. Here, we demonstrated that PCK1 deficiency promoted PDGF-AA secretion by activating the RhoA/PI3K/AKT pathway. Mechanistically, PCK1 deletion may increase intracellular GTP levels, thus promoting the activation of RhoA and further activating the PI3K/AKT pathway. Most Rho GTPases cycle between an active GTP-bound and an inactive GDP-bound form, a process that is regulated by guanine nucleotide exchange factors, GTPase-activating proteins, and guanine nucleotide dissociation inhibitors. Guanine nucleotide exchange factors can activate Rho GTPases by catalyzing the exchange of GDP for GTP when the intracellular concentration of GTP is high55. Several members of the Rho-GTPase family, such as Rac1, RhoA, and RhoC, can be activated by increased concentrations of intracellular GTP56–59. We found that PCK1 deficiency activated RhoA by increasing the levels of intracellular GTP, therefore activating the downstream PI3K/AKT pathway. Moreover, the genetic and pharmacological disruption of RhoA and AKT1 can effectively mitigate MAFLD phenotypes in L-KO mice. In addition, RhoA and AKT inhibitors can reportedly inhibit the progression of MAFLD through other pathways, such as the NF-κB60, Hippo61,62, and Notch63 signaling pathways. However, off-target effects of these inhibitors cannot be completely ruled out. RhoA and AKT inhibitors are currently only in Phase 3 trials or preclinical studies for the treatment of liver fibrosis or clinical tumors, with therapeutic potential for MAFLD64–66. In conclusion, hepatic PCK1 deficiency promoted lipid deposition and fibrosis in a murine MAFLD model. Moreover, hepatic PCK1 loss activated the RhoA/PI3K/AKT pathway, which increased PDGF-AA secretion and promoted HSC activation in male mice. AKT/RhoA inhibitors reduced progressive liver fibrosis, providing a potential therapeutic strategy for MAFLD treatment. ## Animal models Animal experiments were approved by the Animal Experimentation Ethics Committees of Chongqing Medical University and performed in accordance with the Guide for the Care and Use of Laboratory Animals. Pck1f/f mice on a 129S6/SvEv background were purchased from the Mutant Mouse Resource & Research Center (MMRRC: 011950-UNC; Bar Harbor, ME, USA) and Alb-Cre mice on a C57BL/6 background were purchased from Model Animal Research Center of Nanjing University (Nanjing, China). To generate liver-specific Pck1-knockout mice (L-KO), Alb-Cre mice were crossed with Pck1f/f mice. Pck1f/f mice from the same breeding step were used as controls (wild-type, WT). Male WT and L-KO mice at 7–9 weeks old were fed the HFCD-HF/G (D12492: $60\%$ Kcal fat, with drinking water containing 23.1 g/L fructose and 18.9 g/L glucose; Research Diets, New Brunswick, NJ, USA) ($$n = 11$$ per group) or control chow diet (D12450J: $10\%$ Kcal fat, with tap water; Research Diets) ($$n = 10$$ per group) for 24 weeks. Food and drinking water were provided ad libitum. Alb-Cre; Ptenf/f (cPtenf/f) mice (kindly provided by Prof. Yujun Shi, Sichuan University, Chengdu, China) were on a C57BL/6 background and genotyping of the mice was performed as previously described67. L-KO mice were crossed with cPtenf/f mice to generate Alb-Cre; Ptenf/+Pck1f/+ (cPtenf/+Pck1f/+). cPtenf/+Pck1f/+ mice were crossed to breed Alb-Cre; Ptenf/fPck1f/f (cPtenf/fPck1f/f) mice. Littermates that were negative for the Cre transgene were used as WT controls. Primers used for the Pck1, Pten and Cre have been provided in Supplementary Table 2. Male mice were used in all experiments. All mice were housed in temperature-controlled (23 °C) pathogen-free facilities with a 12 h light-dark cycle and humidity ($50\%$ ± $10\%$) conditions. For AAV8 transduction, AAV8-TBG-control and AAV8-TBG-Pck1 were purchased from the Shanghai Genechem Co., Ltd. (Shanghai, China) and injected via the tail vein following 10 weeks of HFCD-HF/G feeding (2 × 1011 genome copies/mouse). The efficiency of virus infection and expression of PCK1 in mouse hepatocytes were confirmed by western blot analysis. After a total of 24 weeks of HFCD-HF/G feeding, the mice were sacrificed for analysis. Genetic and pharmacological inhibition of AKT1 or RhoA were performed in vivo. Genetic depletion of mouse RhoA or AKT1 in vivo was conducted using pSECC (#60820; Addgene, Watertown, MA, USA), a lentiviral-based system that combined both the CRISPR system and Cre recombinase. After HFCD-HF/G feeding for 16 weeks, male L-KO mice were injected with pSECC-sgCtrl, pSECC-sgAkt1, or pSECC-sgRhoA through the tail vein at 1 × 109 genome copies per mouse, with two booster injections at 7 day intervals. The pharmacological inhibition of AKT1 or RhoA was conducted using MK2206 or Rhosin. After HFCD-HF/G feeding for 16 weeks, the mice were divided into 3 groups and intraperitoneally injected with vehicle solution ($$n = 6$$), MK2206 (AKT inhibitor, 50 mg/kg, every 3 days) ($$n = 5$$), or Rhosin (RhoA inhibitor, 20 mg/kg, every 3 days) ($$n = 6$$) for 8 weeks. All mice were sacrificed for further study after HFCD-HF/G feeding for 24 weeks. ## Liver tissues from patients with NASH Liver tissue collection was approved by the Institutional Ethics Committees of Chongqing Medical University and Xin Hua Hospital (project license number: XHEC-C-2012-023). Informed consent was obtained from all participants. Paraffin-embedded normal ($$n = 10$$) and NASH human liver samples ($$n = 36$$) were kindly provided by Dr. Jiangao Fan, Dr. Xiaojun Wang, and Dr. Yalan Wang. Human liver samples with NASH were obtained either by liver biopsy for diagnostic purpose or from surgical liver resections. All liver specimens were evaluated independently by three experienced pathologists, who are blinded to clinical data, according to the NAFLD activity score (NAS), defined as the sum of steatosis, inflammation and hepatocyte ballooning. Patients with a NAS score ≥5 were considered likely to have NASH68. Normal control samples were recruited from samples obtained for the exclusion of liver malignancy during major oncological surgery. Exclusion criteria were the presence of other causes of liver disease, including alcoholic fatty liver disease (>30 g/day for men, >20 g/day for women), chronic infection with hepatitis B and/or C virus, primary biliary cirrhosis, haemochromatosis, autoimmune hepatitis, and Wilson’s disease, as well as the use of anti-obesity, glucose-lowering, and/or lipid-lowering pharmacological treatments. The general characteristics of the NASH human liver samples are listed in Supplementary Table 1. ## Cell culture and treatment All cell lines were grown in Dulbecco’s modified eagle medium (DMEM) supplemented with $10\%$ fetal bovine serum, 100 µg/mL of streptomycin, and 100 U/mL of penicillin at 37 °C in $5\%$ CO2. All cells were negative for mycoplasma. Short tandem repeat tests were performed to ensure the authenticity of the cells. Bovine serum albumin (BSA, $10\%$), different concentrations of palmitic acid (PA), 10 µM MK2206 (AKT inhibitor), 40 µM Rhosin (RhoA inhibitor), or blocking antibody against PDGF-AA (2 µg/mL) was added to the medium. For in vitro co-culture assays, the human hepatic stellate cell (HSC) line LX-2 and human hepatocyte line MIHA cells (provided by Dr Ben C.B. Ko, The Hong Kong Polytechnic University, Hongkong, China) were used. LX-2 was pre-cultured in the lower chamber for 12 hours, and PCK1-OE or PCK1-KO cells (with or without indicated treatment) were seeded in Transwell inserts (#3401, Corning, NY, USA) that were subsequently loaded into the LX2-containing wells. The cells and supernatants were harvested for further analysis after 48 h of co-culture. ## Isolation of primary mouse hepatocytes Primary hepatocytes were isolated and cultured from the livers of WT and L-KO mice fed the HFCD-HF/G as described previously69. Briefly, following anesthesia, the inferior vena cava was cannulated and the liver was perfused in situ with 40 mL pre-warmed EGTA solution and 40 mL solution containing 0.35 mg/mL pronase (no. P5147, Sigma-Aldrich, St. Louis, MO, USA), followed by 40 mL solution containing 0.55 mg/mL collagenase (no. V900893, Sigma-Aldrich). After perfusion, the liver was crushed, and hepatocytes were released into the DMEM. The cell suspension was filtered through a 100 μm cell strainer and centrifuged at 50 ×g at 25 °C for 3 min. After washing three times, the cells were suspended in DMEM supplemented with 10 mM glucose, $10\%$ fetal bovine serum, 100 nM insulin (P3376, Beyotime Biotechnology, Shanghai, China), and 100 nM dexamethasone (D8040, Solarbio Life Sciences, Beijing, China), and then plated on 60 mm diameter plastic plates69. After cell attachment, the medium was replaced with serum-free media, and the cells were used for experiments on the following day. ## Isolation of primary mouse hepatocytes and primary HSCs HSCs were isolated from the mice as described previously70. Briefly, after perfusion with solutions containing protease and collagenase, the crushed liver was digested with a solution containing $1\%$ DNase (10104159001, Roche Diagnostics GmbH, Mannheim, Germany), 0.5 mg/mL protease, and 0.55 mg/mL collagenase for 25 min. The cell suspension was filtered through a 70 µm cell strainer, centrifuged at 580 × g for 10 min at 4 °C, and washed twice with Gey’s balanced salt solution (GBSS). The cells were subjected to gradient centrifugation on a $9.7\%$ Nycodenz (1002424, Axis-Shield, Oslo, Norway) to isolate HSCs, which were then plated onto collagen-coated plates. The cells were cultured in DMEM containing $10\%$ (vol/vol) fetal bovine serum and $1\%$ penicillin-streptomycin71. ## Construction of adenovirus, lentivirus, and stable cell lines AdGFP and AdPCK1 adenoviruses were generated using the AdEasy system as described previously5. The PCK1 knockout (PCK1-KO) MIHA cell line was constructed using the CRISPR-Cas9 system (from Prof. Ding Xue, the School of Life Sciences, Tsinghua University, Beijing, China), as described previously5. To knock down ATF3 expression in MIHA cells, three pairs of oligonucleotides encoding short hairpin RNAs (shRNAs) targeting ATF3 or negative control shRNA were cloned into the pLL3.7 vector (from Prof. Bing Sun, Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences, China). The lentiviruses were obtained by transient transfection of the psPAX2 packaging plasmid and pMD2.G envelope plasmid in HEK293T cells by using Lipofectamine 3000 Transfection Reagent (Thermo Fisher Scientific, Waltham, MA, USA). Transfection efficiency was validated by western blotting. Information on the reagents is listed in Supplementary Table 2. The pSECC lentiviral vector cloning and packaging strategy have been described previously27. ## RNA extraction and real-time PCR Total RNA was extracted from the liver tissues or cell lines using Trizol reagent (15596018, Invitrogen, Carlsbad CA, USA) according to the manufacturer’s instructions. RNA was reverse-transcribed using a PrimeScript RT reagent Kit with gDNA Eraser (Cat: RR047A, TAKARA, Shiga, Japan), and qPCR was performed on a Bio-Rad CFX96 machine (Hercules, CA, USA). The mRNA levels of selected genes were calculated after normalization to β-actin by using the 2(-∆∆C(T)) method. Primer sequences are provided in Supplementary Table 3. ## Immunoblotting Mouse liver tissues and cells were homogenized and lysed in Protein Extraction Reagent (P0013, Beyotime Biotechnology) supplemented with protease inhibitor cocktail (04693159001, Roche, Basel, Switzerland). Equal amounts of protein lysates were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred onto polyvinylidene fluoride membranes (Millipore, Billerica, MA, USA). The antibodies are listed in Supplementary Table 4. β-ACTIN protein was used as a loading control. ## BODIPY staining The cells were fixed in $4\%$ formaldehyde for 30 min at room temperature, permeabilized with $0.1\%$ Triton X-100 for 5 min at room temperature, and stained with BODYPY (D3823, Invitrogen) for 60 min. Stained sections were analyzed using a Leica confocal microscope (Leica TCS SP8, Wetzlar, Germany). ## CCK8 assay The cells were seeded at 5 × 103 cells /well in 96-well plates and treated with different concentrations of PA for 24 h. 10 μL CCK-8 solution (#C0005, Topscience, Shanghai, China) was added to each well and incubated at 37 °C for 1 h. The absorbance was measured at 450 nm. ## ELISA and G-LISA Serum insulin (P3376, Beyotime Biotechnology), TNFα (PT512, Beyotime Biotechnology), IL-6 (PI326, Beyotime Biotechnology), and PDGF-AA (SEA523Mu, Cloud-Clone Corp., Wuhan, China) concentrations in the liver tissue and plasma were quantified using mouse ELISA kits according to the manufacturer’s instructions. The levels of secreted PDGF-AA in the cell culture supernatants were determined using a human ELISA kit (SEA523Hu, Cloud-Clone Corp). Active RhoA was detected in a colorimetric RhoA activation assay (G-LISA) (BK124, Cytoskeleton, Denver, CO, USA). ## Immunofluorescence LX-2 cells were fixed in $4\%$ formaldehyde for 25 min and then incubated in $10\%$ normal goat serum for 1 h. The cells were incubated with primary α-SMA antibodies. Specific signals were visualized using secondary antibodies (ZF-0316, Zsbio, Beijing, China). For nuclear staining, the cells were treated with 1 μg/mL DAPI (10236276001, Roche Diagnostics). The samples were detected with a laser-scanning confocal microscope (Leica TCS SP8). ## Chromatin immunoprecipitation (ChIP) assay Chromatin immunoprecipitation (ChIP) and ChIP-quantitative real-time PCR (ChIP-qPCR) assays for MIHA cells were performed as described previously72. Briefly, sonicated chromatin was used for the immunoprecipitation assay. The pre-cleared supernatants were incubated with a monoclonal antibody against ATF3 overnight at 4 °C, followed by a 4 h of incubation with protein A/G agarose beads (LOT: 3460992, Millipore). The purified DNA fragments bound by ATF3 were analyzed using qPCR. IgG and histone H3 were used as negative and positive controls, respectively. The primers used for real-time PCR analysis of PCK1 are listed in Supplementary Table 3. ## Glucose and insulin tolerance tests The glucose tolerance test (GTT) and insulin tolerance test (ITT) were performed at 2 or 1 weeks prior to sacrifice. The animals were fasted for 16 or 6 h, and then glucose solution (2 g/kg body weight) or insulin (0.75 U/kg body weight) was administered via an intraperitoneal injection, respectively. Venous tail blood samples were collected at 0, 30, 60, 90, and 120 min post-administration to assess blood glucose levels using a glucose meter (HGM-114, OMRON, Kyoto, Japan). ## Luciferase reporter assay MIHA cells were treated with 0.2 mM palmitic acid (PA) in 12-well plates for 24 h and then co-transfected with 0.5 μg of pReceiver-M02-ATF3 or shControl or shATF3 plasmid, 0.5 μg of luciferase reporter plasmids pGL3-basic or pGL3-PCK1, and 25 ng of pRL-TK-Renilla (as transfection control) for 48 h. Then, cells were assayed for luciferase activity via the Dual Luciferase Assay Kit (Promega, Madison, WI, USA). All experiments were performed thrice and presented as mean ± standard deviation (SD). ## Biochemical analysis The serum levels of aspartate transaminase (AST), alanine transaminase (ALT), total triglyceride (TG), total cholesterol (TC), and free fatty acids (FFA) were determined with an automated biochemical analyzer (Hitachi 7600, Tokyo, Japan). TG, TC, and FFA levels in the mouse liver tissues were measured with commercial kits according to the manufacturer’s protocol (TG: cat. no. BC0625; TC: cat. no. BC1985; FFA: cat. no. BC0595, Solarbio Life Sciences). ## Histological analysis Paraffin blocks were sectioned into 4 μm slices and used for hematoxylin and eosin (HE) staining, Sirius red staining, and immunohistochemistry (IHC) assay according to standard protocols. Frozen liver tissue sections were stained with Oil Red O (G1260, Solarbio). For pathological grading, all liver specimens were scored by two experienced pathologists according to the NAFLD activity score (NAS), defined as the sum of steatosis (0–3), inflammation (0–3), and hepatocyte ballooning (0–2). An NAS score ≥5 was considered to indicate NASH. Samples were scanned using a slide scanner (Pannoramic DESK, 3D Histech kft, Hungary). For quantitative analysis, the areas of lipid droplets and Sirius red staining were quantified using ImageJ software (version 1.6.0; NIH, Bethesda, MD, USA). Immunohistochemical staining was semi-quantitatively analyzed using the immunoreactive scoring system73. The percentage of positive cells was graded on a scale of 0−4: (0: negative, 1: 0–$25\%$, 2: 26–$50\%$, 3: 51–$75\%$, 4: 76–$100\%$). The signal intensity was scored on a scale of 0–3: 0 = negative; 1 = weak; 2 = moderate; and 3 = strong. Thus, the final immunoreactive score = (score of staining intensity) × (score of percentage of positive cells). ## Transcriptomic analyses Using Trizol reagent, total RNA was isolated from the liver tissues of WT and L-KO mice fed a chow diet or HFCD-HF/G. The RNA quality was checked with a Bioanalyzer 2200 (Agilent Technologies, Santa Clara, CA, USA). cDNA libraries were prepared using an NEBNext® Ultra™ Directional RNA Library Prep Kit, NEBNext® Poly (A) mRNA Magnetic Isolation Module, NEBNext® Multiplex Oligos according to the manufacturer’s instructions (New England Biolabs, Ipswich, MA, USA). Genes with fold-change >2.0 or <0.5 and false discovery rate <0.05 were considered to be significantly differentially expressed. The volcano, heat, and bubble maps were generated using the ‘ggplot2’ or ‘ggpubr’ packages in R (version 3.6.3; The R Project for Statistical Computing, Vienna, Austria). Gene set enrichment analysis was performed using ‘enrichplot’ packages74. The RNA-seq data files have been deposited to the Gene Expression Omnibus database (www.ncbi.nlm.nih.gov/geo/) under accession number GSE162211. ## Untargeted metabolomics Following sacrifice, the livers were snap-frozen in liquid nitrogen and stored at −80 °C until analysis. Untargeted metabolomics was performed using an ultra-high performance liquid chromatography apparatus (Agilent 1290 Infinity LC, Agilent Technologies) coupled to a quadrupole time-of-flight (TripleTOF 6600, AB SCIEX, Framingham, MA, USA) at Shanghai Applied Protein Technology Co., Ltd. (Shanghai, China). Metabolites with a variable importance in projection value >1 were evaluated using Student’s t-test. $p \leq 0.05$ was considered to indicate statistically significant results. ## HPLC anaIysis of celluIar nucIeotides Cellular nucleotides were extracted according to published procedures75. Briefly, 1 × 106 cells were washed with phosphate-buffered saline and quenched with liquid nitrogen, and then vigorously mixed with methanol and acetonitrile (1:1, v:v). After incubation on ice for 15 min, the samples were centrifuged at 12,000× g at 4 °C for 10 min. Cellular nucleotides were separated and quantified using a C18 column (Agilent Eclipse XDB-C18, 4.6 × 250 mm, average particle size 5 μm) assembled on the Waters Alliance e2695 Separations Module (Milford, MA, USA). Acetonitrile ($5\%$) and 50 mM KH2PO4 (pH 6.5) containing 10 mM tetrabutylammonium bromide were used as mobile phase A, and acetonitrile was used as mobile phase B. All samples were separated in the mobile phase at a flow rate of 1 mL/min for 30 min at 22 °C. No degradation of individual nucleotides or changes in the ratios of nucleotide mixtures was detected during the experiments. The GTP concentration in the samples was calculated based on the slope of the calibration curves generated using pooled authentic samples (to mimic the matrix), and guanosine-5′-triphosphoric acid disodium salt was used as a standard. ## GEO database mining Raw data in the GSE126848, GSE89632, and GSE135251 datasets were downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/geo/). R package DESeq2 was used to analyze gene expression levels between different samples, such as NASH vs. health, NAFLD vs. health, and obesity vs. health. Genes showing \|log2 fold-change \| >1 and false discovery rate <0.05 were considered to present differential expression. ## Statistics Statistical analyses were performed using GraphPad Prism 6.0 software (GraphPad, Inc., La Jolla, CA, USA). Data were represented as the mean ± standard error of the mean unless otherwise stated. Significant differences between the means of two groups were determined using Student’s t-test. One-way analysis of variance was used to determine statistical significance for experiments with more than two groups followed by Tukey’s post hoc test. For correlation analysis, Spearman’s correlation coefficient was used. $p \leq 0.05$ was considered to indicate statistically significant results. ## Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. ## Supplementary information Supplementary Information Peer Review File Reporting Summary The online version contains supplementary material available at 10.1038/s41467-023-37142-3. ## Source data Source Data ## Peer review information Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available. ## References 1. Eslam M. A new definition for metabolic dysfunction-associated fatty liver disease: An international expert consensus statement. J. Hepatol. (2020) 73 202-209. DOI: 10.1016/j.jhep.2020.03.039 2. Eslam M, Sanyal AJ, George J. MAFLD: A Consensus-Driven Proposed Nomenclature for Metabolic Associated Fatty Liver Disease. Gastroenterology (2020) 158 1999-2014.e1991. DOI: 10.1053/j.gastro.2019.11.312 3. Schuster S, Cabrera D, Arrese M, Feldstein AE. Triggering and resolution of inflammation in NASH. Nat. Rev. Gastroenterol. Hepatol. (2018) 15 349-364. DOI: 10.1038/s41575-018-0009-6 4. Arab JP, Arrese M, Trauner M. Recent Insights into the Pathogenesis of Nonalcoholic Fatty Liver Disease. Annu. Rev. Pathol. (2018) 13 321-350. DOI: 10.1146/annurev-pathol-020117-043617 5. Tuo L. PCK1 negatively regulates cell cycle progression and hepatoma cell proliferation via the AMPK/p27(Kip1) axis. J Exp Clin. Cancer Res. (2019) 38 50. DOI: 10.1186/s13046-019-1029-y 6. Xiang J. Gluconeogenic enzyme PCK1 deficiency promotes CHK2 O-GlcNAcylation and hepatocellular carcinoma growth upon glucose deprivation. J. Clin. Invest. (2021) 131 e144703. DOI: 10.1172/JCI144703 7. Xu D. The gluconeogenic enzyme PCK1 phosphorylates INSIG1/2 for lipogenesis. Nature (2020) 580 530-535. DOI: 10.1038/s41586-020-2183-2 8. Santra S. Cytosolic phosphoenolpyruvate carboxykinase deficiency presenting with acute liver failure following gastroenteritis. Mol. Genet. Metab. (2016) 118 21-27. DOI: 10.1016/j.ymgme.2016.03.001 9. She P. Phosphoenolpyruvate carboxykinase is necessary for the integration of hepatic energy metabolism. Mol. Cell Biol. (2000) 20 6508-6517. DOI: 10.1128/MCB.20.17.6508-6517.2000 10. Millward CA. Phosphoenolpyruvate carboxykinase (Pck1) helps regulate the triglyceride/fatty acid cycle and development of insulin resistance in mice. J. Lipid Res. (2010) 51 1452-1463. DOI: 10.1194/jlr.M005363 11. Hoxhaj G, Manning BD. The PI3K-AKT network at the interface of oncogenic signalling and cancer metabolism. Nat. Rev. Cancer (2020) 20 74-88. DOI: 10.1038/s41568-019-0216-7 12. Porstmann T. PKB/Akt induces transcription of enzymes involved in cholesterol and fatty acid biosynthesis via activation of SREBP. Oncogene (2005) 24 6465-6481. DOI: 10.1038/sj.onc.1208802 13. Huang X, Liu G, Guo J, Su Z. The PI3K/AKT pathway in obesity and type 2 diabetes. Int. J. Biol. Sci. (2018) 14 1483-1496. DOI: 10.7150/ijbs.27173 14. Chen J. HIF-2α upregulation mediated by hypoxia promotes NAFLD-HCC progression by activating lipid synthesis via the PI3K-AKT-mTOR pathway. Aging (Albany NY) (2019) 11 10839-10860. DOI: 10.18632/aging.102488 15. Chi Y, Gong Z, Xin H, Wang Z, Liu Z. Long noncoding RNA lncARSR promotes nonalcoholic fatty liver disease and hepatocellular carcinoma by promoting YAP1 and activating the IRS2/AKT pathway. J. Transl. Med. (2020) 18 126. DOI: 10.1186/s12967-020-02225-y 16. Suppli MP. Hepatic transcriptome signatures in patients with varying degrees of nonalcoholic fatty liver disease compared with healthy normal-weight individuals. Am. J. Physiol. Gastrointest. Liver Physiol. (2019) 316 G462-g472. DOI: 10.1152/ajpgi.00358.2018 17. Fang J. Hepatic IRF2BP2 Mitigates Nonalcoholic Fatty Liver Disease by Directly Repressing the Transcription of ATF3. Hepatology (2020) 71 1592-1608. DOI: 10.1002/hep.30950 18. Allen-Jennings AE, Hartman MG, Kociba GJ, Hai T. The roles of ATF3 in liver dysfunction and the regulation of phosphoenolpyruvate carboxykinase gene expression. J. Biol. Chem. (2002) 277 20020-20025. DOI: 10.1074/jbc.M200727200 19. Asgharpour A. A diet-induced animal model of non-alcoholic fatty liver disease and hepatocellular cancer. J. Hepatol. (2016) 65 579-588. DOI: 10.1016/j.jhep.2016.05.005 20. Liu XJ. Characterization of a murine nonalcoholic steatohepatitis model induced by high fat high calorie diet plus fructose and glucose in drinking water. Lab. Invest. (2018) 98 1184-1199. DOI: 10.1038/s41374-018-0074-z 21. Kucukoglu O, Sowa JP, Mazzolini GD, Syn WK, Canbay A. Hepatokines and adipokines in NASH-related hepatocellular carcinoma. J. Hepatol. (2021) 74 442-457. DOI: 10.1016/j.jhep.2020.10.030 22. Thieringer F. Spontaneous hepatic fibrosis in transgenic mice overexpressing PDGF-A. Gene (2008) 423 23-28. DOI: 10.1016/j.gene.2008.05.022 23. Higuchi M, Masuyama N, Fukui Y, Suzuki A, Gotoh Y. Akt mediates Rac/Cdc42-regulated cell motility in growth factor-stimulated cells and in invasive PTEN knockout cells. Curr. Biol. (2001) 11 1958-1962. DOI: 10.1016/S0960-9822(01)00599-1 24. Dou C. P300 Acetyltransferase Mediates Stiffness-Induced Activation of Hepatic Stellate Cells Into Tumor-Promoting Myofibroblasts. Gastroenterology (2018) 154 2209-2221.e2214. DOI: 10.1053/j.gastro.2018.02.015 25. Calvayrac O. The RAS-related GTPase RHOB confers resistance to EGFR-tyrosine kinase inhibitors in non-small-cell lung cancer via an AKT-dependent mechanism. EMBO Mol. Med. (2017) 9 238-250. DOI: 10.15252/emmm.201606646 26. Hodge RG, Ridley AJ. Regulating Rho GTPases and their regulators. Nat. Rev. Mol Cell Biol. (2016) 17 496-510. DOI: 10.1038/nrm.2016.67 27. Sánchez-Rivera FJ. Rapid modelling of cooperating genetic events in cancer through somatic genome editing. Nature (2014) 516 428-431. DOI: 10.1038/nature13906 28. Snaebjornsson MT, Janaki-Raman S, Schulze A. Greasing the Wheels of the Cancer Machine: The Role of Lipid Metabolism in Cancer. Cell Metab. (2020) 31 62-76. DOI: 10.1016/j.cmet.2019.11.010 29. Vily-Petit J. Intestinal gluconeogenesis prevents obesity-linked liver steatosis and non-alcoholic fatty liver disease. Gut (2020) 69 2193-2202. DOI: 10.1136/gutjnl-2019-319745 30. Yoon JC. Control of hepatic gluconeogenesis through the transcriptional coactivator PGC-1. Nature (2001) 413 131-138. DOI: 10.1038/35093050 31. Friedman SL, Neuschwander-Tetri BA, Rinella M, Sanyal AJ. Mechanisms of NAFLD development and therapeutic strategies. Nat. Med. (2018) 24 908-922. DOI: 10.1038/s41591-018-0104-9 32. Samuel VT. Fasting hyperglycemia is not associated with increased expression of PEPCK or G6Pc in patients with Type 2 Diabetes. Proc. Natl. Acad. Sci. USA (2009) 106 12121-12126. DOI: 10.1073/pnas.0812547106 33. Imai E. Characterization of a complex glucocorticoid response unit in the phosphoenolpyruvate carboxykinase gene. Mol Cell Biol. (1990) 10 4712-4719. PMID: 2388623 34. Lundsgaard AM. Mechanisms Preserving Insulin Action during High Dietary Fat Intake. Cell Metab. (2019) 29 50-63.e54. DOI: 10.1016/j.cmet.2018.08.022 35. Guo X. The Role of Palmitoleic Acid in Regulating Hepatic Gluconeogenesis through SIRT3 in Obese Mice. Nutrients (2022) 14 1482. DOI: 10.3390/nu14071482 36. Wang Z, Dong C. Gluconeogenesis in Cancer: Function and Regulation of PEPCK, FBPase, and G6Pase. Trends Cancer (2019) 5 30-45. DOI: 10.1016/j.trecan.2018.11.003 37. Jiang W. Acetylation regulates gluconeogenesis by promoting PEPCK1 degradation via recruiting the UBR5 ubiquitin ligase. Mol. Cell (2011) 43 33-44. DOI: 10.1016/j.molcel.2011.04.028 38. Hai T, Wolfgang CD, Marsee DK, Allen AE, Sivaprasad U. ATF3 and stress responses. Gene Expr. (1999) 7 321-335. PMID: 10440233 39. Tsai WW. ATF3 mediates inhibitory effects of ethanol on hepatic gluconeogenesis. Proc. Natl. Acad. Sci. USA (2015) 112 2699-2704. DOI: 10.1073/pnas.1424641112 40. Tu C. Cardiolipin Synthase 1 Ameliorates NASH Through Activating Transcription Factor 3 Transcriptional Inactivation. Hepatology (2020) 72 1949-1967. DOI: 10.1002/hep.31202 41. Gut P. Whole-organism screening for gluconeogenesis identifies activators of fasting metabolism. Nat Chem Biol (2013) 9 97-104. DOI: 10.1038/nchembio.1136 42. Hakimi P. Phosphoenolpyruvate carboxykinase and the critical role of cataplerosis in the control of hepatic metabolism. Nutr Metab (Lond) (2005) 2 33. DOI: 10.1186/1743-7075-2-33 43. Satapati S. Mitochondrial metabolism mediates oxidative stress and inflammation in fatty liver. J. Clin. Invest. (2015) 125 4447-4462. DOI: 10.1172/JCI82204 44. Jahn D, Kircher S, Hermanns HM, Geier A. Animal models of NAFLD from a hepatologist’s point of view. Biochim. Biophys. Acta. Mol. Basis Dis. (2019) 1865 943-953. DOI: 10.1016/j.bbadis.2018.06.023 45. Zhou R. Intestinal α1-2-Fucosylation Contributes to Obesity and Steatohepatitis in Mice. Cell Mol. Gastroenterol. Hepatol. (2021) 12 293-320. DOI: 10.1016/j.jcmgh.2021.02.009 46. Tsuchida T. A simple diet- and chemical-induced murine NASH model with rapid progression of steatohepatitis, fibrosis and liver cancer. J Hepatol. (2018) 69 385-395. DOI: 10.1016/j.jhep.2018.03.011 47. Lonardo A. Sex Differences in Nonalcoholic Fatty Liver Disease: State of the Art and Identification of Research Gaps. Hepatology (2019) 70 1457-1469. DOI: 10.1002/hep.30626 48. Miquilena-Colina ME. Hepatic fatty acid translocase CD36 upregulation is associated with insulin resistance, hyperinsulinaemia and increased steatosis in non-alcoholic steatohepatitis and chronic hepatitis C. Gut (2011) 60 1394-1402. DOI: 10.1136/gut.2010.222844 49. Doege H. Silencing of hepatic fatty acid transporter protein 5 in vivo reverses diet-induced non-alcoholic fatty liver disease and improves hyperglycemia. J. Biol. Chem. (2008) 283 22186-22192. DOI: 10.1074/jbc.M803510200 50. Méndez-Lucas A. PEPCK-M expression in mouse liver potentiates, not replaces, PEPCK-C mediated gluconeogenesis. J. Hepatol. (2013) 59 105-113. DOI: 10.1016/j.jhep.2013.02.020 51. Langhi C, Baldán Á. CIDEC/FSP27 is regulated by peroxisome proliferator-activated receptor alpha and plays a critical role in fasting- and diet-induced hepatosteatosis. Hepatology (2015) 61 1227-1238. DOI: 10.1002/hep.27607 52. Xu MJ. Fat-Specific Protein 27/CIDEC Promotes Development of Alcoholic Steatohepatitis in Mice and Humans. Gastroenterology (2015) 149 1030-1041.e1036. DOI: 10.1053/j.gastro.2015.06.009 53. Alves-Bezerra M, Cohen DE. Triglyceride Metabolism in the Liver. Compr. Physiol. (2017) 8 1-8. PMID: 29357123 54. Tsuchida T, Friedman SL. Mechanisms of hepatic stellate cell activation. Nat. Rev. Gastroenterol. Hepatol. (2017) 14 397-411. DOI: 10.1038/nrgastro.2017.38 55. Zhang B, Zhang Y, Shacter E, Zheng Y. Mechanism of the guanine nucleotide exchange reaction of Ras GTPase-evidence for a GTP/GDP displacement model. Biochemistry (2005) 44 2566-2576. DOI: 10.1021/bi048755w 56. Wawrzyniak JA. A purine nucleotide biosynthesis enzyme guanosine monophosphate reductase is a suppressor of melanoma invasion. Cell Rep. (2013) 5 493-507. DOI: 10.1016/j.celrep.2013.09.015 57. Hallett MA, Dagher PC, Atkinson SJ. Rho GTPases show differential sensitivity to nucleotide triphosphate depletion in a model of ischemic cell injury. Am. J. Physiol. Cell Physiol. (2003) 285 C129-C138. DOI: 10.1152/ajpcell.00007.2003 58. Mondin M. Alterations in cytoskeletal protein expression by mycophenolic acid in human mesangial cells requires Rac inactivation. Biochem. Pharmacol. (2007) 73 1491-1498. DOI: 10.1016/j.bcp.2006.12.025 59. Bianchi-Smiraglia A. Microphthalmia-associated transcription factor suppresses invasion by reducing intracellular GTP pools. Oncogene (2017) 36 84-96. DOI: 10.1038/onc.2016.178 60. Han MH. Flavonoids Isolated from Flowers of Lonicera japonica Thunb. Inhibit Inflammatory Responses in BV2 Microglial Cells by Suppressing TNF-α and IL-β Through PI3K/Akt/NF-kb Signaling Pathways. Phytother. Res. (2016) 30 1824-1832. DOI: 10.1002/ptr.5688 61. Zhao Y. PI3K Positively Regulates YAP and TAZ in Mammary Tumorigenesis Through Multiple Signaling Pathways. Mol. Cancer Res. (2018) 16 1046-1058. DOI: 10.1158/1541-7786.MCR-17-0593 62. Jeong SH, Lim DS. Insulin receptor substrate 2: a bridge between Hippo and AKT pathways. BMB Rep. (2018) 51 209-210. DOI: 10.5483/BMBRep.2018.51.5.095 63. Gutierrez A, Look AT. NOTCH and PI3K-AKT pathways intertwined. Cancer Cell (2007) 12 411-413. DOI: 10.1016/j.ccr.2007.10.027 64. Chien AJ. MK-2206 and Standard Neoadjuvant Chemotherapy Improves Response in Patients With Human Epidermal Growth Factor Receptor 2-Positive and/or Hormone Receptor-Negative Breast Cancers in the I-SPY 2 Trial. J. Clin. Oncol. (2020) 38 1059-1069. DOI: 10.1200/JCO.19.01027 65. Schmid P. Capivasertib Plus Paclitaxel Versus Placebo Plus Paclitaxel As First-Line Therapy for Metastatic Triple-Negative Breast Cancer: The PAKT Trial. J. Clin. Oncol. (2020) 38 423-433. DOI: 10.1200/JCO.19.00368 66. Yoon C. Chemotherapy Resistance in Diffuse-Type Gastric Adenocarcinoma Is Mediated by RhoA Activation in Cancer Stem-Like Cells. Clin. Cancer Res. (2016) 22 971-983. DOI: 10.1158/1078-0432.CCR-15-1356 67. Tong Z. Pancreas-specific Pten deficiency causes partial resistance to diabetes and elevated hepatic AKT signaling. Cell Res. (2009) 19 710-719. DOI: 10.1038/cr.2009.42 68. Brunt EM, Kleiner DE, Wilson LA, Belt P, Neuschwander-Tetri BA. Nonalcoholic fatty liver disease (NAFLD) activity score and the histopathologic diagnosis in NAFLD: distinct clinicopathologic meanings. Hepatology (2011) 53 810-820. DOI: 10.1002/hep.24127 69. Hernández-Alvarez MI. Deficient Endoplasmic Reticulum-Mitochondrial Phosphatidylserine Transfer Causes Liver Disease. Cell (2019) 177 881-895.e817. DOI: 10.1016/j.cell.2019.04.010 70. Mederacke I, Dapito DH, Affò S, Uchinami H, Schwabe RF. High-yield and high-purity isolation of hepatic stellate cells from normal and fibrotic mouse livers. Nat. Protoc. (2015) 10 305-315. DOI: 10.1038/nprot.2015.017 71. Choi WM. Glutamate Signaling in Hepatic Stellate Cells Drives Alcoholic Steatosis. Cell Metab. (2019) 30 877-889.e877. DOI: 10.1016/j.cmet.2019.08.001 72. Yang F. GSTZ1-1 Deficiency Activates NRF2/IGF1R Axis in HCC via Accumulation of Oncometabolite Succinylacetone. Embo j (2019) 38 e101964. DOI: 10.15252/embj.2019101964 73. Yao F. A targetable LIFR-NF-κB-LCN2 axis controls liver tumorigenesis and vulnerability to ferroptosis. Nat. Commun. (2021) 12 7333. DOI: 10.1038/s41467-021-27452-9 74. Yu G, Wang LG, Han Y, He Q. Y. clusterProfiler: an R package for comparing biological themes among gene clusters. Omics (2012) 16 284-287. DOI: 10.1089/omi.2011.0118 75. Hannan JP. HPLC method to resolve, identify and quantify guanine nucleotides bound to recombinant ras GTPase. Anal. Biochem. (2021) 631 114338. DOI: 10.1016/j.ab.2021.114338 76. Arendt BM. Altered hepatic gene expression in nonalcoholic fatty liver disease is associated with lower hepatic n-3 and n-6 polyunsaturated fatty acids. Hepatology (2015) 61 1565-1578. DOI: 10.1002/hep.27695 77. Govaere O. Transcriptomic profiling across the nonalcoholic fatty liver disease spectrum reveals gene signatures for steatohepatitis and fibrosis. Sci. Transl. Med. (2020) 12 eaba4448. DOI: 10.1126/scitranslmed.aba4448
--- title: 'Global research on syndemics: a meta-knowledge analysis (2001-2020)' authors: - Md Mahbub Hossain - Nobonita Saha - Tahmina Tasnim Rodela - Samia Tasnim - Tasmiah Nuzhath - Tamal Joyti Roy - James N. Burdine - Helal Uddin Ahmed - E. Lisako J. McKyer - Banga Kamal Basu - Ping Ma journal: F1000Research year: 2023 pmcid: PMC10015119 doi: 10.12688/f1000research.74190.2 license: CC BY 4.0 --- # Global research on syndemics: a meta-knowledge analysis (2001-2020) ## Abstract Background: Syndemics or synergies of cooccurring epidemics are widely studied across health and social sciences in recent years. Methods: We conducted a meta-knowledge analysis of articles published between 2001 to 2020 in this growing field of academic scholarship. Results: We found a total of 830 articles authored by 3025 authors, mostly from high-income countries. Publications on syndemics are gradually increasing since 2003, with rapid development in 2013. Each article was cited more than 15 times on average, and most ($$n = 604$$) articles were original studies. Syndemics research focused on several areas, including HIV/AIDS, substance abuse, mental health, gender minority stressors, racism, violence, chronic physical and mental disorders, food insecurity, social determinants of health, and coronavirus disease 2019. Moreover, biopsychosocial interactions between multiple health problems were studied across medical, anthropological, public health, and other disciplines of science. Conclusions: The limited yet rapidly evolving literature on syndemics informs transdisciplinary interests to understand complex coexisting health challenges in the context of systematic exclusion and structural violence in vulnerable populations. The findings also suggest applications of syndemic theory to evaluate clinical and public health problems, examine the socioecological dynamics of factors influencing health and wellbeing, and use the insights to alleviate health inequities in the intersections of synergistic epidemics and persistent contextual challenges for population health. ## Amendments from Version 1 Based on the reviewers suggestions we have expanded on the data selection criteria to alert readers about possible inclusion of all literatures using syndemic theory despite the theoretical discrepancies in the framing. We have also expanded the discussion section to provide additional critical evaluation of the retrieved articles, highlighted the on-going methodological issues in the field and added additional recommendations for future research based on our findings. ## Introduction Addressing health inequities in marginalized populations requires a complex understanding of the epidemiological burden of multiple health problems and associated factors that determine the health statuses and outcomes in that context. 1 – 3 Theoretical frameworks may help in examining different dimensions of population health, adopting intellectual inspirations from scholarly concepts that emerged in the past across scientific discipline. 4, 5 Amongst many contemporary theories, “syndemic(s) theory” offers critical perspectives on the relationships between diseases and biopsychosocial factors that not only explain the high burden of diseases in populations but also sustain a series of adverse health and social outcomes. 6, 7 Conceptually, syndemics have three fundamental components that characterize them. 6, 8, 9 Firstly, two or more health problems cluster together that can be assessed epidemiologically or described as co-morbidity or multimorbidity. Secondly, syndemic diseases or conditions interact among themselves using biological, psychological, or social pathways. Lastly, syndemics are associated with social, structural, and contextual forces that precipitate disease clustering and progression in the first place, which may include but are not limited to poverty, segregated housing, systematic exclusion from opportunities, enslavement, colonialism, neo-colonialism, and neo-liberal economic measures that result in disproportionate distribution of wealth, lack of access to resources and services that may improve health and wellbeing, and other socioeconomic inequalities. 6, 8 – 10 Since its conceptualization by Merrill Singer in the 1990s, 10 – 12 syndemic(s) theory has become one of the eminent ideas that has influenced scholarly discourses in different disciplines, including social epidemiology and medical anthropology, investigating the dynamics of persistent health inequities in disadvantaged population groups such as homeless, racial and ethnic minorities, and people affected by chronic diseases or social problems. 6, 9 One of the earliest examples used in theorizing syndemics was substance abuse, violence, and HIV/AIDS (SAVA). 8, 10 The case of the SAVA syndemic illustrated the clustering of these three health problems where a high burden of substance abuse and violence were prevalent in people living with HIV/AIDS, predominantly in the systematically oppressed communities in the United States. 10 Most of them were urban poor, people of color, and deprived of opportunities to live healthy lives in the first place. Another syndemic of HIV and Hepatitis C virus (HCV) is recognized in people who use drugs (PWUD). 9, 13 Nearly 2.3 million out of 36.7 million people living with HIV in 2015 were infected with HCV. The co-infection of HIV and HCV significantly increases the risks of advanced liver disease and associated adverse health outcomes compared to those with HCV infection alone. 9, 14, 15 Moreover, HCV facilitates the pathogenesis and disease progression of HIV, thus interacting with each other in multiple pathways that may result in adverse health outcomes among the affected individuals. 14 These infections are highly prevalent in PWUD, who are more likely to share syringes and engage in other high-risk health behaviors. In addition to these pathological and psychological challenges, these people share similar socioeconomic marginalization driven by structural forces that transform their life choices, health behaviors, and biopsychosocial outcomes. 9, 14 Rather than presenting the disease burden and their correlates only, syndemic(s) theory highlights the complicated relationships between these co-existing health problems that can be biological, psychological, or social in nature. More importantly, studying these problems in the context of their shared determinants and interactions between multiple constructs provides a broader understanding to address the problems, which is less probable if these problems are examined individually without exploring their synergistic characteristics. 9, 13 A growing body of literature indicates that the academic and professional interests in syndemics have increased over the past three decades. 6 – 8 A review of syndemics associated with HIV/AIDS identified 60 articles published in 2019 alone, 6 which reported co-conditions such as substance abuse ($$n = 40$$; $67\%$), high-risk sexual behavior ($$n = 36$$; $60\%$), depression ($$n = 36$$; $60\%$), interpersonal violence ($$n = 35$$; $58\%$), stigma ($$n = 19$$; $32\%$), sexually transmitted infections (STIs) ($$n = 16$$; $27\%$), trauma ($$n = 14$$; $23\%$), and noncommunicable diseases (NCDs) ($$n = 6$$; $10\%$). Such inclusive nature of those syndemics literature informs the scope of interdisciplinary research using tools from diverse sources to answer common questions of interest in different contexts and populations. The increasing recognition of research on syndemics can potentially specify, describe, and explain bio-social interactive pathways advancing both science and practice. Another review identified 143 journal articles, 23 book chapters, and 29 other types of publications. 9 In this review, the authors reported five thematic categories of studies that included syndemics ($12\%$) with cooccurring diseases with bib-bio and bio-social interactions, potential syndemics ($18\%$) where those interactions are referred but not fully articulated, socially determined heightened health burden of diseases ($15\%$) that described social conditions associated with poor health without identifying disease clusters or interactions, harmful disease clusters ($17\%$) that identified disease clusters and social factors without mentioning their interactions, and adverse additive co-morbidities ($38\%$) describing diseases and social determinants through an additive approach to adverse health outcomes without examining the evidence on interactions. Furthermore, a systematic literature review assessed a potential syndemic comprising of HIV, HCV, intimate partner violence, and posttraumatic stress disorder (PTSD). 16 In this review, the authors reported childhood physical and sexual abuse and intimate partner violence as social sources of trauma, which was associated with elevated unsafe health behavior leading to a higher burden of HIV infections and subsequent health outcomes. Existing literature provides an overview of diverse health problems and methodological measures adopted by different authors to study syndemics and associated conditions, 6, 9 which highlight both the complexity of the current evidence base and the necessity of improving our understandings of syndemics applied in different frontiers of knowledge. Primary studies and analytical reviews offer syntheses of evidence focusing on specific research questions relevant to a field or a problem of interest, which are useful for evaluating evidence in that scenario. However, such focused approaches may not provide a comprehensive overview of the entire scientific landscape or describe how research on a domain or topic evolved over time. In this regard, meta-knowledge analyses offer quantitative assessments of research identifying the overall status and characteristics of research. Through scientometric and bibliometric measures, meta-knowledge studies identify top contributing scholars, institutions, journals, and countries that may promote further research collaborations. In addition, meta-knowledge studies aim to identify research hotspots where most studies have emphasized previously, thus inform research trends and explore research areas that are not examined extensively. Recognizing areas that require further studies is one of the many goals of knowledge development in a field or topic of interest. To the best of our knowledge, there is no meta-level study on syndemics, which informs an overview of the current status and the evolution of global knowledge on syndemics-related research. Such studies, if conducted systematically, can inform the scholars and practitioners to understand the historical development of knowledge in this field, explain the status of intellectual advancements, and guide future research and scientific advancements in this area of growing interest in health sciences research. ## Objectives In the current study, we primarily aimed to analyze the characteristics and trends of the global research literature on syndemics. Secondarily, we evaluated the most prolific authors, institutions, journals, affiliating countries, and funding institutions contributing to syndemics-related research. Lastly, we mapped the major knowledge domains on syndemics highlighting the intellectual development in this field. ## Data source In this study, we adopted meta-knowledge methods that have been used in previous research. 17, 18 To retrieve scientometric data on syndemics, we accessed the Web of Science (WoS) core collection that included multiple citation sources such as Science Citation Index-Expanded (SCI-Expanded), the Social Sciences Citation Index (SSCI), the Arts & Humanities Citation Index (A&HCI), and the Emerging Sources Citation Index (ESCI). The selection of WoS as the data source was informed by several advantages that it offers. Firstly, WoS provides extensive coverage of more than 20,000 journals, making it one of the most widely used databases for bibliometric studies. 19 Moreover, WoS includes publications not only from biomedical sources but also social sciences and other scholarly disciplines, thus making the bibliographic collective more inclusive in nature. 20, 21 As syndemics are associated with a wide range of biopsychosocial issues relevant to different scientific disciplines, 9, 22 WoS is likely to provide a transdisciplinary overview of the research landscape in this topic. ## Data criteria and extraction We used “syndemic” keyword in the topic field for searching the titles, abstracts, and keywords in the WoS database collection. The search process was structured using several eligibility criteria. First, we included citations published from January 1, 2000, to December 31, 2020. Due to discrepancy in the framing of syndemics in the published literature, the coexistence of multiple health problems without specifying the underlying bio-bio or bio-social relationships is often studied within the syndemic framework. 9 Thus studies titled syndemics can be included in the analysis despite the potential differences in the scope of their work. This issue has been elaborated in the limitations section. Second, we reviewed the titles and abstracts of the retrieved citations and excluded studies on topics such as “parasyndemicolpate” OR “syndemicolpate” that could have provided false positive entries. Third, we limited our search in citations published in English language, thus we excluded citations published in languages other than English. Lastly, we included all publication types such as original articles, reviews, commentaries, editorials, letters, and book chapters considering a low number of citations in the emerging field of knowledge. For all eligible citations, the complete bibliometric data on publication records, authorship, institutional affiliations, funding information, keywords, and citations data were extracted for subsequent analytic steps. No alterations were made for the primarily collected data. ## Data analysis The corpus of the eligible literature was used for descriptive analysis, social network analysis, and conceptual structure analysis using measures that have been used in previous knowledge mapping studies. 17, 19, 23, 24 The descriptive analyses on the key bibliometric characteristics such as total citations, publication trends, top ten cited articles, prolific authors, h-index and g-index of the authors, contributing journals, research institutions, affiliating countries of the authors were analyzed using Microsoft Excel 2021 and R software (version 4.1.2). 25 Knowledge mapping offers visual representations of the social connectedness among affiliating authors, institutions, and countries that reflect the research collaborations in a topic. We used a free software called the VOSviewer (version 1.6.16) that applies a natural language processing algorithm (NLP), which represents the units of analysis as a circular node within the map. 26 The size of the node portrays the volume (e.g., number of publications in a dataset) and the position represents the resemblance with other nodes in the same set. As a result, closer nodes are more alike than nodes far apart from each other in the final map. The lines connecting multiple nodes represent the relationships among those nodes, whereas the thickness of those lines indicate the strength of that relationship. Finally, the color of the node indicates the cluster to which each node has been allocated to. In this way, all nodes in the map are clustered together based on their affiliation. 26 To make the map, VOSviewer uses the SMACOF algorithm, 27 which minimizes the function: VX1………Xn=∑i<jSijXi−Xj2 under the constraints: 2nn−1∑i<jSijXi−Xj=1 where: n–the number of nodes in a given network, X i –the locations of node i within a two-dimensional space, \|\|X i −X j \|\|–the Euclidean distance between nodes i and j. VOSviewer builds clusters of nodes by maximizing the following function: Vc1………cn=∑i<jδci,cjsij−γ where: c i –the cluster to which node i is assigned, δ (c 1, c j)–a function that equals one if c i = c j; and zero otherwise, γ–a resolution parameter that determines the level of detail of the clustering (the higher γ is, the higher the number of clusters). This method uses a distance-based approach to construct the bibliometric maps in three steps. 26, 28 At the first step, it normalizes the differences between multiple nodes. At the next step, it builds a two-dimensional map where the distance between multiple nodes reflects the similarities between those nodes. Further, it combines closely related nodes into clusters sharing similar bibliometric properties allowing visual representation of the research field. Moreover, we conducted a co-occurrence analysis of keywords to evaluate the conceptual relationships between multiple topics. The frequency of co-occurrence of two or more keywords indicated the strength of their association, whereas multiple keywords appearing within a cluster highlighted the topical foci or knowledge base within the research landscape. This approach was used to map multiple clusters with an overview of research domains and a graphical evolution of those domains across years. Lastly, we developed a three-field plot connecting the current literature with cited references using KeyWords Plus in the WoS database. This was used to generate the most frequently appearing words or phrases in cited sources, depicting the intellectual linkages between the existing studies on syndemics and cited research articles. This meta-knowledge study is registered in the Open Science Framework, and the data on eligible studies can be found in the underlying data. ## Overview of the publications We found a total of 830 articles eligible for this bibliometric study, including 604 original articles, 75 reviews, and 151 other types of publications from 314 scholarly sources (Table 1). These documents were authored by 3025 individual authors, on average most articles were authored by more than three authors Most ($$n = 750$$) articles had multiple authors, whereas 80 articles were authored by a single author. The collaboration index was 4.03 suggesting more than four co-authors per article index calculated only using the multi-authored article set. Figure 1 shows the citation retrieval process. Table 2 provides an overview of the top 10 cited articles on syndemics published from 2003 to 2019, with the average citation per article per year ranging from 12.81 to 124.67. Seven of these articles focused on HIV-related syndemics, whereas the remaining articles discussed HIV in syndemic-related discourses emphasizing other health issues such as substance abuse, tuberculosis, violence, nutritional disorders, chronic diseases, and social determinants of health. Table 2.* \| Title \| Authors \| Journals \| Publication year \| Total citations \| Average citations per year \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Syndemics and public health: Reconceptualizing disease in bio-social context \| Singer, M; Clair, S \| Medical Anthropology Quarterly \| 2003 \| 480 \| 25.26 \| \| The Global Syndemic of Obesity, Undernutrition, and Climate Change: The Lancet Commission report \| Swinburn, B A. et al. \| Lancet \| 2019 \| 374 \| 124.67 \| \| Psychosocial health problems increase risk for HIV among urban young men who have sex with men: preliminary evidence of a syndemic in need of attention \| Mustanski, B. et al. \| Annals of Behavioral Medicine \| 2007 \| 328 \| 21.87 \| \| HIV and Tuberculosis: a Deadly Human Syndemic \| Kwan, C.K.; Ernst, J.D. \| Clinical Microbiology Reviews \| 2011 \| 310 \| 28.18 \| \| Antiretroviral Therapy for Prevention of Tuberculosis in Adults with HIV: A Systematic Review and Meta-Analysis \| Suthar, A.B. \| Plos Medicine \| 2012 \| 233 \| 23.3 \| \| Syndemics and the biosocial conception of health \| Singer, M. et al. \| Lancet \| 2017 \| 229 \| 45.8 \| \| Sexual Compulsivity, Co-Occurring Psychosocial Health Problems, and HIV Risk Among Gay and Bisexual Men: Further Evidence of a Syndemic \| Parsons, J.T. et al \| American Journal of Public Health \| 2012 \| 206 \| 20.6 \| \| Syndemics, sex and the city: Understanding sexually transmitted diseases in social and cultural context \| Singer, M. et al. \| Social Science & Medicine \| 2006 \| 205 \| 12.81 \| \| Syndemic Theory and HIV-Related Risk Among Young Transgender Women: The Role of Multiple, Co-Occurring Health Problems and Social Marginalization \| Brennan, J. et al. \| American Journal of Public Health \| 2012 \| 199 \| 19.9 \| \| Substance Abuse, Violence, and HIV in Women: A Literature Review of the Syndemic \| Meyer, J. et al. \| Journal of Women’s Health \| 2011 \| 178 \| 16.18 \| ## Publication trends Figure 2 highlights an increasing trend of scholarly publications on syndemics since 2003. Only 33 articles were published before 2010, whereas the frequency of publication increased significantly since 2013. The annual growth rate of the scholarly publications on syndemics was calculated as $10.47\%$. Figure 2.: Trends of scholarly publications on syndemic. ## Top contributors in syndemics research Table 3 provides an overview of the top ten contributing authors who contributed to the published manuscripts on syndemics. Among the authors, Safren S.A. had the highest number of publications ($$n = 26$$), followed by Singer M ($$n = 24$$) and Halkitis P.N. ($$n = 17$$). Moreover, Singer M. had the highest number of citations ($$n = 949$$), followed by Stall R. ($$n = 679$$) and Mendenhall E. ($$n = 555$$). Singer M. also had the highest h-index [12] and g-index [24] among the top authors. Table 3. \| Author \| Number of Publications \| Year of first publication \| h-index \| g-index \| Total citations \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Safren SA \| 26 \| 2010 \| 9 \| 22 \| 497 \| \| Singer M \| 24 \| 2003 \| 12 \| 24 \| 949 \| \| Halkitis PN \| 17 \| 2010 \| 11 \| 17 \| 484 \| \| Mayer KH \| 14 \| 2015 \| 6 \| 14 \| 229 \| \| Stall R \| 14 \| 2005 \| 11 \| 14 \| 679 \| \| Mendenhall E \| 13 \| 2014 \| 8 \| 13 \| 555 \| \| O'Cleirigh C \| 12 \| 2015 \| 6 \| 12 \| 200 \| \| Gilbert M \| 9 \| 2017 \| 4 \| 6 \| 45 \| \| Pitpitan EV \| 9 \| 2013 \| 6 \| 9 \| 102 \| \| Reisner SL \| 9 \| 2009 \| 8 \| 9 \| 349 \| Most articles on syndemics were published in AIDS and Behavior journal ($7.23\%$, $$n = 60$$), followed by Annals of Behavioral Medicine ($3.73\%$, $$n = 31$$) and AIDS Care ($2.89\%$, $$n = 24$$). Four out of the top ten journals were associated with AIDS and STI-related topics, whereas all journals were related to epidemiological, behavioral, anthropological, social, and public health sciences (Table 4). Table 4. \| Journal's name \| Total publications on syndemics \| Percentage \| Impact factor (Journal citations reports 2019) \| \| --- \| --- \| --- \| --- \| \| AIDS and Behavior \| 60 \| 7.22% \| 3.147 \| \| Annals of Behavioral Medicine \| 31 \| 3.73% \| 4.475 \| \| AIDS Care Psychological and Socio-medical Aspects of AIDS HIV \| 24 \| 2.89% \| 1.894 \| \| Archives of Sexual Behavior \| 18 \| 2.17% \| 3.131 \| \| Social Science Medicine \| 17 \| 2.05% \| 3.616 \| \| American Journal of Public Health \| 16 \| 1.93% \| 6.464 \| \| BMC Public Health \| 15 \| 1.81% \| 2.521 \| \| Global Public Health \| 15 \| 1.81% \| 1.791 \| \| Journal of Acquired Immune Deficinecy Syndromes \| 15 \| 1.81% \| 3.475 \| \| Annals of Anthropological Practice \| 14 \| 1.69% \| \| ## Institutional research collaborations Table 5 shows the top 10 institutions that were affiliated with syndemics-related research. The majority of the studies were authored by scholars from the University of California System ($11.93\%$, $$n = 99$$), followed by Harvard University ($8.19\%$, $$n = 68$$), and the Johns Hopkins University ($6.63\%$, $$n = 55$$). Figure 3 shows extensive collaborations between the key affiliating institutions with a higher publication frequency ($$n = 5$$ or above), which highlights the interconnectedness of top contributing institutions in collaborative research on syndemics. ## Global research collaborations Global contributors at the country level show 76 countries that published at least one document on syndemics-related topics. Among those countries, the top 10 contributors are listed in Table 6. Most articles originated from the United States ($74.46\%$, $$n = 618$$), followed by Canada ($11.93\%$, $$n = 99$$) and UK ($6.14\%$, $$n = 51$$). Figure 4 shows global collaborations among participating countries (with at least one publication), highlighting a high volume of syndemics research from high-income countries compared to low- and middle-income countries. Moreover, North American and other high-income countries had stronger collaborative ties on syndemics research that informs higher cumulative production of scientific research from these regions. ## Research areas on syndemics Several research areas were identified within the broader umbrella of syndemics using clusters of keywords that overlapped with each other. We mapped keywords that had a frequency of 5 and above across the collective bibliography. A total of 104 top keywords were identified and clustered to generate common areas of research in multiple clusters highlighted in the same color. Figure 5 shows the distribution of keyword clusters where the greater size of a circle represents a higher number of publications on that keyword and the thickness of connecting line between circles represents the intensity of coexistence of these terms in the literature. Figure 5.: Research hotspots in syndemics research. Amongst multiple research hotspots, the first cluster highlighted in red consisted of keywords related to syndemic theory and associated concepts such as syndemic, substance use, intimate partner violence, depression, health disparities, transgender, gay, bisexual, food insecurity, and intersectionality (Figure 5). Several studies were identified that used these keywords, indicating their relevance to this research cluster. For example, Couture et al., adopted the syndemic framework to evaluate the effects of comorbid psychosocial problems on the risk of physical and sexual violence on female entertainment and sex workers (FESW) in Cambodia. 29 They reported a high burden of client-perpetuated violence that was associated with housing insecurity, substance use, and psychological distress. FESW with two psychosocial conditions had twice the odds (adjusted odds ratio [AOR] = 2.08; $95\%$ confidence interval [CI] 1.00-4.31), whereas women with 5-6 psychosocial conditions had eightfold higher odds (AOR = 8.10; $95\%$ CI 3.4-19.31) of violence, highlighting a syndemic model of cooccurring psychosocial problems. Another study examined the prevalence and correlates of trauma in South African youth living in syndemic HIV risk. 30 More than $99\%$ of the participating youths experienced at least one potentially traumatic event (PTE), whereas a high PTE score associated with high food insecurity among adolescent men (AOR 2.63, $95\%$ CI = 1.36-5. 09) and women (AOR = 2.57, $95\%$ CI = 1.55-4.26, respectively). This study reported biopsychosocial pathways including depression and inconsistent condom use as pathways of syndemic of trauma and HIV in that context. The second cluster illustrated in green includes keywords on syndemics of infectious diseases such as HIV, AIDS, tuberculosis, STIs, hepatitis c, antiretroviral therapy, adherence, comorbidity, and coinfection. For example, Dyer et al., conducted a cohort study among 301 men who have sex with men (MSM) to assess syndemic relationships. 31 They reported that depression symptoms were associated with sexual compulsiveness (odds ratios [OR]: 1.88, $95\%$ CI = 1.1, 3.3) and stress (OR: 2.67, $95\%$ CI = 1.5, 4.7); sexual compulsiveness was associated with stress (OR: 2.04, $95\%$ CI = 1.2, 3.5); substance misuse was associated with intimate partner violence (IPV) (OR: 2.57, $95\%$ CI = 1.4, 4.8); stress was associated with depression symptoms (OR: 2.67, $95\%$ CI = 1.5, 4.7), sexual compulsiveness (OR: 2.04, $95\%$ CI = 1.2, 3.5) and IPV (OR: 2.84, $95\%$ CI = 1.6, 4.9). Also, men who reported three or more syndemic constituents (three or more conditions) were engaged in high-risk sexual behavior compared to men who had two or fewer health conditions (OR: 3.46, $95\%$ CI = 1.4-8.3). Moreover, a review by Meyer and colleagues identified 45 articles that emphasized SAVA syndemic and associated conditions such as HIV-associated risk-taking behaviors, mental health, utilization of health services and medication adherence, and a bidirectional relationship between violence and HIV/AIDS. 32 This review highlighted the complex relationships and associated outcomes of poor decision making and high-risk behavior in the context of the SAVA syndemic. In the third cluster highlighted in blue, several keywords were identified, including mental health, MSM, sexual compulsivity, stigma, resilience, internet, and aging. Studies in this cluster focused on health behavior and syndemic relationships in gay and bisexual men. For example, a study by Parsons and colleagues examined 1,033 HIV-negative gay and bisexual men living in the U.S. and found that more than $62\%$ of men reported having at least one syndemic condition. 33 Also, HIV-related risk behavior was associated with polydrug use, sexual compulsivity, being single, and being Latino. Moreover, the risk was highest among participants with three or more syndemic conditions. Another study from Mexico conducted by Pitpitan et al. found that MSM with a high number of syndemic conditions showed an increased prevalence of sexual risk-taking. 34 Moreover, MSM who were out to more people showed a weaker association between high-risk sexual behavior and syndemic conditions suggesting outness or disclosure of same-sex preference as a resilience factor from a syndemic perspective. A fourth cluster highlighted in yellow included keywords related to psychosocial and health behavior-related keywords such as violence, substance abuse, HIV testing, homophobia, and suicide. This research cluster emphasized the growing number of studies that examined the psychosocial health of MSM and associated syndemic conditions and relationships. For example, a study by Herrick and colleagues recruited 1551 MSM and found that different life-course predictors such as internalized homophobia and victimization were significantly associated with syndemic condition as well as psychosocial health conditions including stress, depressive symptomology, substance abuse, compulsive sexual behavior, and intimate partner violence. 35 Moreover, the authors used a nested negative binomial analysis and found that the overall life course significantly explained the variability in syndemic outcomes (chi[2] = 247.94; $P \leq .001$; df = 22). Another study by Ferlatte and colleagues examined the data from a survey of 8382 Canadian gay and bisexual men and found that suicidal ideation and attempts were associated with individual marginalization and psychosocial health problems such as mental disorders, substance use, STIs, and HIV risks. 36 In addition, individuals with three or more psychosocial problems had higher odds of experiencing suicidal ideation [6.90 (5.47-8.70) times] and suicide attempts [16.29 (9.82-27.02)] compared to participants with no such problems. These relationships show the complex nature of syndemic relationships among people living under psychosocial stressors that impacts their health and wellbeing. The fifth research cluster highlighted in purple color consisted of keywords including coronavirus disease 2019 (COVID-19), obesity, diabetes, chronic disease, poverty, and social determinants of health. This cluster emphasizes a research domain that includes noncommunicable diseases and contemporary health problems such as the COVID-19 pandemic. For example, Mendenhall and colleagues conducted a study in Kenya and found that adults with diabetes shared a complex social and medical framework associated with their health conditions. People with diabetes also had comorbid anxiety, depression, and infectious diseases such as HIV/AIDS, malaria, and tuberculosis. 7 The authors also reported that social problems were associated with biophysical suffering, whereas women had a higher burden of psychosocial distress and somatic symptoms such as multimorbidity compared to men. People with diabetes reported not only concurrent anxiety and depression but also common infections, including malaria, tuberculosis, and HIV/AIDS. Another study from Puerto Rico found that the subaltern status negatively affected obesity rates that could be attributable to limited federal assistance for health insurance and healthier food items. Moreover, weight mismanagement and a lack of healthcare providers were amongst the psychosocial challenges that were associated with the obesity syndemic in this population. Furthermore, recent studies focused on the relevance of syndemic perspectives on coronavirus disease (COVID-19) pandemic. For example, Gutman and colleagues argued that coinfections such as malaria and other parasitic diseases with SARS-CoV-2 could result in detrimental health outcomes necessitating increased testing and disease surveillance during the COVID-19 pandemic. 22 Moreover, Perez-Escamilla and colleagues discussed the persistent effects of food and nutrition insecurity associated with poor maternal and child health outcomes that can be exacerbated during the COVID-19 pandemic. 37 Furthermore, long-standing health inequities and systemic racism are associated with multimorbidity, which may have syndemic relationships leading to adverse health outcomes in marginalized communities. 38 In this regard, Poteat and colleagues used the syndemic framework to discuss the syndemic conditions in Black Americans who experienced psychosocial stressors such as mortgage redlining, history of enslavement, political gerrymandering, lack of access to healthcare, job discrimination, and health care provider bias. 39 The authors argued that racial disparities in COVID-19 require acknowledging and addressing structural racism and determinants of these chronic disparities among the affected individuals. These articles highlight the biopsychosocial challenges and their relationships that share common determinants and affect population health, predominantly in vulnerable population groups. Three more clusters with fewer keywords appeared in the intersections of major clusters reported above. The sixth cluster identified in pink color consisted of keywords such as men who have sex with men, condom use, transgender women, and India. Moreover, keywords such as AIDS, HCV, and tuberculosis (TB) colored in brown formed the seventh research cluster on syndemics. Lastly, scattered nodes of keywords colored in orange included HIV infection, South Africa, pregnancy, qualitative, sexual minority, network analysis, and mental illness. These clusters highlight diverse topics that overlap with other clusters and highlight the interconnectedness of the keywords as well as research topics that are common across research domains. ## Evolution of knowledge in the field of syndemics research Research domains within the scientific field of syndemics evolved over the years, which is highlighted in Figure 6. Since most articles were published in recent years, keywords used until 2015 highlight the scholarly themes of earlier publications on syndemics. These keywords are marked in purple, which included HIV/AIDS, substance abuse, prevention, syndemic theory, coinfection, social determinants, social inequality, internet, mortality, sexually transmitted diseases, and health policy. Moreover, keywords colored in bluish-purple show a transition of those topics being used in publications around 2016, which include syndemics, tuberculosis, obesity, harm reduction, health disparities, poverty, and aging. Furthermore, keywords used across publications during 2017-18 are presented in green color, where most keywords such as syndemic, men who have sex with men, gay and bisexual men, diabetes, domestic violence, food insecurity, global health, public health, drug use, intimate partner violence, HIV risk, testing, and prevention were identified. Lastly, keywords in yellow represent topics used in articles published around and after 2019. These recent topics included COVID-19, pandemic, social determinants of health, noncommunicable diseases, network analysis, latent class analysis, pre-exposure prophylaxis, and climate change. Figure 6.: Evolution of major research topics in the field of syndemics. An assessment of the cited references provided another perspective on how the current research publications cited and used previous scholarly items. Figure 7 represents the cited sources on the left, the most widely used keywords from the titles of cited publications in the middle, the top contributing authors on the right. Common keywords from cited sources inform the most relevant topics used from previous research, which included psychosocial health problems, gay, united states, substance use, risk, united states, prevalence, depression, health, women, and syndemics. Across these cited sources, Singer et al. authored five articles published from 1994 to 2017. 7, 8, 10, 12, 40 One of earliest articles that was widely cited across syndemics literature was the article describing the development and psychometric assessment of the Center for Epidemiological Studies-Depression (CES-D) scale by Radolf, published in 1977. 38 Furthermore, top published authors such as Singer and Stall have been contributing to syndemics research across most topics that are frequently cited across current publications. Figure 7.: Intellectual contributions of influential articles and authors on commonly cited topics related to syndemics. ## Discussion This study evaluated the global scientific landscape of syndemics research using bibliometric measures. The findings of this study suggest a slow yet gradual increase in syndemics-related publications, which has accelerated since 2013. In recent years, syndemic conditions are increasingly examined in both primary and review articles that draw intellectual inspirations from biopsychosocial literature, highlighting a transdisciplinary trend in syndemics research. Moreover, a higher proportion of original articles rather than reviews and other publication types indicates a rising body of empirical research on syndemic(s) theory and associated concepts. Furthermore, syndemics-related publications were published in general and specialty journals emphasizing the intersection of coexisting diseases, their shared determinants, and complex relationships between biosocial constructs as commonly seen in social medicine and allied fields of knowledge. Most of the syndemics-related publications were affiliated with authors and institutions from high-income countries, and the U.S. had the highest contributions. In addition, heavy collaborations among authors and institutions in high-income countries indicate active research across major research entities. While some of the low- and middle-income countries (LMICs) such as Brazil, Mexico, China, South Africa, and India are engaged in syndemics research, their individual and collective contributions appeared to be limited, which informs a critical research gap in LMICs. This is consistent with previous research that reported a low scholarly output from LMICs in different areas of health sciences. 41 – 44 In the case of syndemics research, the historical and persistent gaps in research capacities in LMICs are likely to be compounded by the fact that syndemics have been primarily conceptualized and extensively studied by scholars and institutions in high-income countries. It is necessary to increase the research capacities in LMICs with a focus on syndemics, as those countries experience a high burden of infectious and noncommunicable diseases and poor social determinants of health. As the key concepts of syndemic(s) theory suggest, 6, 7, 10 studying syndemics in such underprivileged contexts can potentially offer unique and diverse scientific perspectives on the biopsychosocial formation of health and illness in respective populations. It may enrich the current knowledge base on syndemics and inform the future transformations of medical and social care as well as disease prevention globally. Syndemics research focused on diverse topics, among which HIV-related syndemics were extensively studied across the literature. As evident in research hotspots, HIV/AIDS and associated health behaviors and outcomes were common in more than one cluster of keywords. This frequent appearance of HIV/AIDS in the syndemics research landscape can be attributable to several factors. First, HIV/AIDS-related conditions were amongst the earliest syndemics that were conceptualized and examined in the history of the syndemic theory. For this reason, many articles focusing on non-HIV syndemics also referred to the case of previous HIV syndemics such as SAVA. 9, 10 Secondly, syndemics literature from the US and other developed countries present studies and cases of multiple health problems and their synergies within a syndemic framework, 6, 9 where the study populations were urban poor from inner-city neighborhoods. As we learn from social epidemiological and anthropological studies, 9, 33 biosocial challenges associated with HIV/AIDS were highly prevalent in these population groups. Therefore, the academic discourses on the social reality of health and wellbeing in these marginalized people would remain incomplete without studying HIV/AIDS and related issues. Lastly, literature from the global context, particularly from Sub-Saharan countries, highlighted the burden of HIV/AIDS in the context of the continued burden of infectious and chronic diseases in those contexts. 6, 9 For this reason, we can find studies that provide national and regional assessments of HIV/AIDS-related syndemics from those countries, which also compared their findings with what was known from similar studies conducted in developed nations. As many scholars described HIV/AIDS as a pandemic rather than just a disease outbreak, 45, 46 global studies comparing and connecting the concepts and associated evidence revealed the worldwide relevance of HIV/AIDS in syndemics research. In addition to HIV/AIDS, syndemics literature also discussed infectious diseases such as tuberculosis and COVID-19. While tuberculosis research was often studied as a comorbid condition in people living with HIV/AIDS, 6, 9 recent literature examined the relevance of comorbid diseases that are relevant to susceptibility to COVID-19 and subsequent health outcomes. 39, 47 – 49 Given the growing burden of biopsychosocial challenges associated with COVID-19, 38, 50 – 55 it is necessary to investigate bio-bio and bio-social interactions between coexisting health problems in COVID-19 patients and survivors. Such research may reveal the true burden of disease clustering, biopsychosocial relationships, shared determinants, and health outcomes in the context of this pandemic. Despite a growing body of evidence that is mostly epidemiological and predominantly cross-sectional in nature, transdisciplinary and longitudinal investigations would be necessary to understand syndemic aspects of this pandemic. Furthermore, COVID-19 has demonstrated the vulnerability of people with noncommunicable diseases who are more likely to have adverse outcomes. 38, 39, 49 Limited literature exists on syndemics of non-HIV chronic diseases, 9 which necessitates a more comprehensive yet inclusive study of communicable and noncommunicable diseases from a syndemic perspective. Social determinants of health constituted a significant proportion of syndemics research. This is consistent with the fundamental idea of syndemic conditions that have common social factors and bio-social interactions. 9, 10, 12 However, there were distinct patterns in research on social factors that influenced health status, disease development, and subsequent outcomes in study populations. As evident in research hotspots, sexual and gender minorities such as men who have sex with men, bisexuals, and transgender people were frequently studied in the syndemics literature. Although many of those studies examined health problems and associated factors in the context of HIV/AIDS, 6 studies have also presented their psychosocial vulnerability due to minority stressors such as social stigma and other determinants of health in the affected individuals. The added value of those studies may include, but are not limited to, a broader understanding of how gender roles and norms may have biosocial interactions with pre-existing health conditions, contribute to the progression of diseases, and impact their health and wellbeing at the individual and population level. Such research findings must be translated to clinical and social decision-making addressing the stressors that affect health outcomes in gender minorities. It is also crucial to highlight the existing debate within the field of syndemics is the framing and the statistical approaches used to establish relationships between the diseases and outcomes. Considering the complex multilevel involvement of syndemics Tsai et al. and colleagues believe that most empirical studies purporting to validate the theory lack in statistical robustness. Tsai identified that most studies in the field of syndemics demonstrate a statistically significant association between psychosocial problems and health outcomes using simple linear models, very few studies assess the how these psychosocial problems interact to magnify the outcomes as they only use summative approach for measuring the effect. Additionally, the summative approach is only successful to measure the changes when the effect sizes are small. 56 Authors suggested to use the summative model using the count variable approach to conduct a latent factor analysis for confirming that a single parameter model is adequate. They emphasized on sharpening the theory’s prediction approaches by incorporating advanced systems level approaches. 56, 57 Stall and colleagues argued in favor of the summative approach saying that majority of the existing data in this field comes from self-reported data from cross-sectional studies. Applying more rigorous data analysis method would require redesigning studies such as long-term cohorts to collect data to establish the theory. 58 However, considering the current nature of the funding, implementing such resource intensive cohort study is not feasible. Due to the physical and psychological burden of these diseases (especially HIV) on highly vulnerable marginalized communities such as young gay black men or transgender women, the current approach is acceptable as it can demonstrate the dangerous existence of these co-morbidities highlighting the needs for prevention efforts from practitioners and policy makers. Moreover, future syndemics research may need to extend the scope and depth of conceptual and empirical investigations on other psychosocial stressors such as racism, poverty, lack of education, unemployment, inadequate access to health and social care, xenophobia, and other means of marginalization. In recent years, public interests in those issues reflect not only their relevance to the social oppression experienced by minorities but also inform a critical need to examine the biopsychosocial dynamics of such stressors to understand how they may determine health and social outcomes in respective populations. ## Directions for future research The current literature emphasizes diseases and their interactions among individuals in the social context. However, socioecological perspectives may inform the need for research in several under-investigated areas that may provide a more accurate understanding of how syndemics work in individuals and populations. 59, 60 For example, the role of caregivers and family members is critical for the psychosocial wellbeing of the affected individuals. 44, 61, 62 Moreover, caring for someone with one or more chronic diseases is reported to be associated with adverse health and social outcomes among family caregivers. 63, 64 Since they share the same or similar social and contextual factors relevant to the respective syndemics, future research should investigate the roles of and impacts on informal caregivers alongside primarily affected individuals in syndemic scenarios. Furthermore, syndemic conditions, as well as their determinants and consequences, may not have equal impacts on people of different age groups. Disease dynamics and associated social forces are likely to be different in children, adolescents, young adults, and older adults with varying sociodemographic characteristics. 9, 65, 66 Future research should examine syndemics in different population groups, associated biopsychosocial relationships, and multiple outcomes. From a health services perspective, syndemics may impose unique challenges to the affected individuals requiring personalized care that may address multiple health concerns at a time. The recent technological advancements have facilitated the digitalization of health services with a focus on personalized or precision health. 67 – 69 Syndemics research in the digital age should explore avenues of integrating interventions that target intersections of coexisting diseases and associated factors in specific contexts. Such integrative technological measures may inform spatial and temporal challenges through real-time measurements, 70, 71 and enable the practitioners to mitigate the burden of multimorbidity using evidence-based digital and traditionally delivered interventions. Given the complex nature of syndemics, implementation research on coexisting diseases, common determinants, cumulative impacts, and shared sociocultural aspects should be conducted, and the findings should be incorporated in public health and social welfare policymaking. Despite a growing need for translating the current evidence to transform clinical and social care for syndemics, it is necessary to acknowledge that the current literature does not emphasize implementation dimensions of complex health and social problems. Researchers and decision-makers should recognize this gap as the persistence of comorbid disease would be critical for common healthcare operations such as patient-provider interactions and the delivery of health services. Institutional measures, including sensitizing the stakeholders and establishing evidence translation systems for practitioners, should be prioritized for minimizing knowledge-to-practice gaps in this regard. Syndemics research highlights global health disparities by emphasizing the roles of context-specific forces and determinants of health. 9, 13 It is critical to foster collaborative efforts among researchers across the globe to expand the ongoing research in the field of syndemics by including diverse perspectives. The present study has highlighted which institutions, mainly from high-income countries, have expertise in syndemics research. An ideal next step would be to focus on developing long-term collaboration by resource sharing and capacity building with scholars from LMICs to amplify their work. International public health organizations can establish global strategic planning groups that help fund collaborative work to address syndemics across nations. These groups could also concentrate on implementing multilevel strategies involving researchers, policymakers, health workers, and other key stakeholders to achieve a sustainable solution. Therefore, local and global health systems should be examined from a syndemic perspective, which may enable health system strengthening that results in a better understanding of concurrent syndemics, future population health challenges, and how to respond to the same in a systematic way. Such insights on syndemics would also necessitate the active participation of global health organizations and member countries to use their collective resources to combat global health crises such as infectious disease outbreaks, food insecurity, climate migration, and health inequities that continue to affect populations. In these processes, the role of political commitment and collaborations will be of paramount importance. From a health promotion perspective, it is critical to engage communities who are vulnerable to syndemics to increase their awareness on social determinants of cooccurring diseases. 72, 73 The public should be informed through targeted and mass media interventions regarding common challenges and preventive measures. However, empowering community-based organizations to assess their risks and address the same using local resources would be far more impactful. Such efforts may require technical assistance and external support, which should be organized through public health institutions. As syndemics researchers argue that context-specific factors and biopsychosocial interactions make syndemics unique to individuals and populations, 9, 40, 48 participatory research and action plans may create opportunities to exchange knowledge and develop context-appropriate interventions for syndemics. Syndemics research highlights an intersection of academic contributions from multiple scholarly disciplines, including medical anthropology, social epidemiology, health policy, health promotion, clinical sciences, and social sciences. 6, 30, 60, 72 The combined use of multiple quantitative and qualitative methods has complemented the investigations of complex syndemic problems in current literature, which couldn’t have been understood using any single-best method. 6, 9, 16 Arguably, syndemics necessitate transdisciplinary approaches to integrate multiple research methodologies to answer complex research questions. Such integrative measures may help the scholars to apprehend the ontology and phenomenology of syndemics from shared epistemological perspectives. Global research on syndemics informs the deteriorating effects of cooccurring health problems in populations living under chronic social stressors. 8, 10, 31, 59 Understanding these complex scenarios may offer insights on humanitarian challenges that are infrequently studied and discussed in contemporary health and social policy discourses. Notably, the production of health in a population reflects its commitment to improve people’s wellbeing and maximize public welfare through regulating factors that may be harmful to individual and social health. Such initiatives may also include mandates for better access to health services irrespective of social, economic, cultural, geographic, and other differences across individuals and communities. However, many high-income countries such as the United States are far behind in ensuring equitable access to health. 74 Previous research on political determinants of health informs critical challenges such as systematic racism, implicit bias, environmental injustice, and other structural factors that share the policy responses to public health problems. 74 – 77 From an ethical perspective, it is a shared responsibility of healthcare providers, researchers, and organizations to advocate for using science for bringing social justice through meaningful changes in the socio-political determinants of syndemics and health disparities in marginalized populations. ## Limitations There are several limitations of this meta-knowledge study, which are necessary to understand the scope and findings of this study. Also, the limitations of this study would inform future research addressing the theoretical, methodological, and empirical shortcomings of the current findings. One such limitation is the choice of the database; although WoS is one of the most inclusive sources of bibliographic data, it may not contain all articles published on syndemics. Therefore, this study could not include potential studies that may exist elsewhere. Although this limitation is very common across knowledge mapping studies, 17, 19, 23 we encourage methodologists and other scholars to continue intellectual discourses on how the global knowledge community can find opportunities to harmonize citations data across databases. This is extremely challenging as journals are not universally indexed, and databases do not contain bibliographic data on all key variables of interest. This challenge necessitates an integration of technological advancements and cooperation between database authorities to facilitate a uniform distribution of scholarly resources globally. Another challenge was the framing of syndemics in the published literature. As found in the previous synthesis of empirical research, 9 many scholars do not examine all criteria of syndemics in their studies. For example, the coexistence of multiple health problems without specifying the bib-bio or bio-social relationships is often studied within the syndemic framework. Also, it is possible to describe all criteria of syndemic without using the framework, which is perhaps a major challenge leading to a limited observation of literature on complex health problems. We recommend wide scholarly communication of syndemic theory and adoption of relevant keywords and explanations whenever possible. This may eliminate the existing knowledge biases and improve future mapping of the global knowledge landscape. Lastly, despite using contemporary knowledge mapping approaches, this study may not inform population-level estimates or determinants of syndemics. As we discussed earlier, the field of syndemics research is evolving, and this study highlighted key research areas within this growing field of knowledge. We recommend further primary studies and evidence synthesis on disease clusters that have synergistic effects, interactions, and shared sociocultural factors that may inform specific insights on context-specific syndemics. ## Conclusions In this meta-knowledge study on global research on syndemics, we found a limited yet growing body of scientific literature on syndemics and associated population health problems. Most studies on syndemics were published from high-income countries, and they included diverse topics ranging from STIs to various social determinants of health. Further research is needed to understand the dynamics of multiple coexisting infectious and noncommunicable diseases across global populations with a focus on cumulative disease burden, biopsychosocial relationships, and common structural forces that are associated with health and wellbeing. The findings of this study highlight the scope of syndemics to inform advanced research, health policymaking, and practices, not only through focusing on a single disease but also addressing health inequities in the intersections of multiple health problems through inclusive, context-specific, socioculturally appropriate, and evidence-based approaches. ## Underlying data Open Science Framework: A meta-analysis of global research on syndemics [2001-2020]. https://doi.org/10.17605/OSF.IO/8N9R6. 78 This project contains the following file: -DataFile-Authors-DOI-Titles.xlsx-Flowchart of the citations retrieval process [1].pdf Data are available under the terms of the Creative Commons Zero “No rights reserved” data waiver (CC0 1.0 Public domain dedication). ## References 1. Braveman PA, Kumanyika S, Fielding J. Health disparities and health equity: the issue is justice.. (2011) 101 S149-S155. DOI: 10.2105/AJPH.2010.300062 2. Gee GC, Ford CL. Structural racism and health inequities: Old issues, New Directions1.. (2011) 8 115-132. DOI: 10.1017/S1742058X11000130 3. Quinn SC, Kumar S. Health inequalities and infectious disease epidemics: a challenge for global health security.. (2014) 12 263-273 4. Dover DC, Belon AP. The health equity measurement framework: a comprehensive model to measure social inequities in health.. (2019) 18 1-12. PMID: 30606218 5. Friedman DJ, Starfield B. Models of population health: their value for US public health practice, policy, and research.. (2003) 93 366-369. DOI: 10.2105/AJPH.93.3.366 6. Mendenhall E, Singer M. What constitutes a syndemic? Methods, contexts, and framing from 2019.. (2020) Publish Ahead of Print 213-217. DOI: 10.1097/COH.0000000000000628 7. Singer M, Bulled N, Ostrach B. Syndemics and the biosocial conception of health.. (2017) 389 941-950. DOI: 10.1016/S0140-6736(17)30003-X 8. Singer M. A dose of drugs, a touch of violence, a case of AIDS, part 2: Further conceptualizing the SAVA syndemic.. (2006) 34 39-54 9. Singer M, Bulled N, Ostrach B. Whither syndemics?: Trends in syndemics research, a review 2015–2019.. (2020) 15 943-955. DOI: 10.1080/17441692.2020.1724317 10. Singer M. (1996) 11. Singer M, Snipes C. Generations of suffering: Experiences of a treatment program for substance abuse during pregnancy.. (1992) 3 222-234. DOI: 10.1353/hpu.2010.0180 12. Singer M. AIDS and the health crisis of the US urban poor; the perspective of critical medical anthropology.. (1994) 39 931-948. DOI: 10.1016/0277-9536(94)90205-4 13. Perlman DC, Jordan AE. The syndemic of opioid misuse, overdose, HCV, and HIV: structural-level causes and interventions.. (2018) 15 96-112. DOI: 10.1007/s11904-018-0390-3 14. Maier I, Wu GY. Hepatitis C and HIV co-infection: a review.. (2002) 8 577-579. DOI: 10.3748/wjg.v8.i4.577 15. Hernandez MD, Sherman KE. HIV/hepatitis C coinfection natural history and disease progression.. (2011) 6 478-482. DOI: 10.1097/COH.0b013e32834bd365 16. Brezing C, Ferrara M, Freudenreich O. The syndemic illness of HIV and trauma: implications for a trauma-informed model of care.. (2015) 56 107-118. DOI: 10.1016/j.psym.2014.10.006 17. Abdalla SM, Solomon H, Trinquart L. What is considered as global health scholarship? A meta-knowledge analysis of global health journals and definitions.. (2020) 5 e002884. DOI: 10.1136/bmjgh-2020-002884 18. Trinquart L, Galea S. Mapping epidemiology’s past to inform its future: metaknowledge analysis of epidemiologic topics in leading journals, 1974–2013.. (2015) 182 93-104. DOI: 10.1093/aje/kwv034 19. Hernández-Torrano D, Ibrayeva L, Sparks J. Mental health and well-being of university students: A bibliometric mapping of the literature.. (2020) 11 1226. DOI: 10.3389/fpsyg.2020.01226 20. Meho LI, Yang K. Impact of data sources on citation counts and rankings of LIS faculty: Web of Science versus Scopus and Google Scholar.. (2007) 58 2105-2125. DOI: 10.1002/asi.20677 21. Wang Q, Waltman L. Large-scale analysis of the accuracy of the journal classification systems of Web of Science and Scopus.. (2016) 10 347-364. DOI: 10.1016/j.joi.2016.02.003 22. Gutman JR, Lucchi NW, Cantey PT. Malaria and parasitic neglected tropical diseases: potential syndemics with COVID-19?.. (2020) 103 572-577. DOI: 10.4269/ajtmh.20-0516 23. Hossain MM. Alice in Wonderland syndrome (AIWS): a research overview.. (2020) 7 389-400. DOI: 10.3934/Neuroscience.2020024 24. Hossain MM. Umbrella Review as an Emerging Approach of Evidence Synthesis in Health Sciences: A Bibliometric Analysis.. (2020). DOI: 10.2139/ssrn.3551055 25. Aria M, Cuccurullo C. bibliometrix: An R-tool for comprehensive science mapping analysis.. (2017) 11 959-975. DOI: 10.1016/j.joi.2017.08.007 26. Van Eck NJ, Waltman L. Software survey: VOSviewer, a computer program for bibliometric mapping.. (2010) 84 523-538. DOI: 10.1007/s11192-009-0146-3 27. Borg I, Groenen PJF. (2005) 28. Eck NJ, Waltman L. (2014) 285-320. DOI: 10.1007/978-3-319-10377-8_13 29. Couture M-C, Evans JL, Moret JD. Syndemic psychosocial health conditions associated with recent client-perpetrated violence against female entertainment and sex workers in Cambodia.. (2020) 49 3055-3064. DOI: 10.1007/s10508-020-01705-y 30. Closson K, Dietrich JJ, Nkala B. Prevalence, type, and correlates of trauma exposure among adolescent men and women in Soweto, South Africa: implications for HIV prevention.. (2016) 16 1-15. DOI: 10.1186/s12889-016-3832-0 31. Dyer TP, Shoptaw S, Guadamuz TE. Application of syndemic theory to black men who have sex with men in the Multicenter AIDS Cohort Study.. (2012) 89 697-708. DOI: 10.1007/s11524-012-9674-x 32. Meyer JP, Springer SA, Altice FL. Substance abuse, violence, and HIV in women: a literature review of the syndemic.. (2011) 20 991-1006. DOI: 10.1089/jwh.2010.2328 33. Parsons JT, Millar BM, Moody RL. Syndemic conditions and HIV transmission risk behavior among HIV-negative gay and bisexual men in a U.S. National sample.. (2017) 36 695-703. DOI: 10.1037/hea0000509 34. Pitpitan EV, Smith LR, Goodman-Meza D. “Outness” as a Moderator of the Association Between Syndemic Conditions and HIV Risk-Taking Behavior Among Men Who Have Sex with Men in Tijuana, Mexico.. (2016) 20 431-438. DOI: 10.1007/s10461-015-1172-1 35. Herrick AL, Lim SH, Plankey MW. Adversity and syndemic production among men participating in the multicenter AIDS cohort study: a life-course approach.. (2013) 103 79-85. DOI: 10.2105/AJPH.2012.300810 36. Ferlatte O, Dulai J, Hottes TS. Suicide related ideation and behavior among Canadian gay and bisexual men: a syndemic analysis.. (2015) 15 1-9. DOI: 10.1186/s12889-015-1961-5 37. Pérez-Escamilla R, Cunningham K, Moran VH. COVID-19 and maternal and child food and nutrition insecurity: a complex syndemic.. (2020) 16 e13036. DOI: 10.1111/mcn.13036 38. Gravlee CC. Systemic racism, chronic health inequities, and COVID-19: A syndemic in the making?.. (2020) 32 e23482. DOI: 10.1002/ajhb.23482 39. Poteat T, Millett GA, Nelson LE. Understanding COVID-19 risks and vulnerabilities among black communities in America: the lethal force of syndemics.. (2020) 47 1-3. DOI: 10.1016/j.annepidem.2020.05.004 40. Singer M, Clair S. Syndemics and public health: reconceptualizing disease in bio-social context.. (2003) 17 423-441. DOI: 10.1525/maq.2003.17.4.423 41. Ahasan R, Alam MS, Chakraborty T. Applications of GIS and geospatial analyses in COVID-19 research: A systematic review.. (2020) 9. DOI: 10.12688/f1000research.27544.1 42. Allen LN, Fox N, Ambrose A. Quantifying research output on poverty and non-communicable disease behavioural risk factors in low-income and lower middle-income countries: a bibliometric analysis.. (2017) 7 e014715. DOI: 10.1136/bmjopen-2016-014715 43. Hossain MM, Sultana A, Tasnim S. Prevalence of mental disorders among people who are homeless: An umbrella review.. (2020) 66 528-541. DOI: 10.1177/0020764020924689 44. Hong YA, Hossain MM, Chou WS. Digital interventions to facilitate patient-provider communication in cancer care: A systematic review.. (2020) 29 591-603. DOI: 10.1002/pon.5310 45. Eisinger RW, Fauci AS. Ending the HIV/AIDS pandemic.. (2018) 24 413-416. DOI: 10.3201/eid2403.171797 46. Parker R. The global HIV/AIDS pandemic, structural inequalities, and the politics of international health.. (2002) 92 343-347. DOI: 10.2105/AJPH.92.3.343 47. Yadav UN, Rayamajhee B, Mistry SK. A syndemic perspective on the management of non-communicable diseases amid the COVID-19 pandemic in low-and middle-income countries.. (2020) 8 508. DOI: 10.3389/fpubh.2020.00508 48. Islam N, Lacey B, Shabnam S. Social inequality and the syndemic of chronic disease and COVID-19: county-level analysis in the USA.. (2021) 75 496-500. DOI: 10.1136/jech-2020-215626 49. Horton R. Offline: COVID-19 is not a pandemic.. (2020) 396 874. DOI: 10.1016/S0140-6736(20)32000-6 50. Cruz MP, Santos E, Cervantes MAV. COVID-19, a worldwide public health emergency.. (2021) 221 55-61. DOI: 10.1016/j.rceng.2020.03.001 51. Lai C-C, Wang C-Y, Wang Y-H. Global epidemiology of coronavirus disease 2019 (COVID-19): disease incidence, daily cumulative index, mortality, and their association with country healthcare resources and economic status.. (2020) 55 105946. DOI: 10.1016/j.ijantimicag.2020.105946 52. Mazumder H, Hossain MM, Das A. Geriatric care during public health emergencies: lessons learned from novel corona virus disease (COVID-19) pandemic.. (2020) 63 257-258. DOI: 10.1080/01634372.2020.1746723 53. Lin J, Guo T, Becker B. Depression is associated with moderate-intensity physical activity among college students during the covid-19 pandemic: Differs by activity level, gender and gender role.. (2020) 13 1123-1134. DOI: 10.2147/PRBM.S277435 54. Hossain MM, Tasnim S, Sultana A. Epidemiology of mental health problems in COVID-19: A review.. (2020) 9 1-16. DOI: 10.12688/f1000research.24457.1 55. Chi X, Becker B, Yu Q. Prevalence and Psychosocial Correlates of Mental Health Outcomes Among Chinese College Students During the Coronavirus Disease (COVID-19) Pandemic.. (2020) 11. DOI: 10.3389/fpsyt.2020.00803 56. Tsai AC, Burns BF. Syndemics of psychosocial problems and HIV risk: A systematic review of empirical tests of the disease interaction concept.. (2015) 139 26-35. DOI: 10.1016/j.socscimed.2015.06.024 57. Tsai AC. Syndemics: a theory in search of data or data in search of a theory?. (2018) 206 117-122. DOI: 10.1016/j.socscimed.2018.03.040 58. Stall R, Coulter RW, Friedman MR. Commentary on “Syndemics of psychosocial problems and HIV risk: A systematic review of empirical tests of the disease interaction concept” by A. Tsai and B. Burns.. (2015) 145 129-131. DOI: 10.1016/j.socscimed.2015.07.016 59. Elifson KW, Klein H, Sterk CE. The value of using a syndemics theory conceptual model to understand the factors associated with obesity in a southern, urban community sample of disadvantaged African-American adults.. (2016) 27 1 60. Batchelder AW, Gonzalez JS, Palma A. A social ecological model of syndemic risk affecting women with and at-risk for HIV in impoverished urban communities.. (2015) 56 229-240. DOI: 10.1007/s10464-015-9750-y 61. Hiel L, Beenackers MA, Renders CM. Providing personal informal care to older European adults: Should we care about the caregivers’ health?.. (2015) 70 64-68. DOI: 10.1016/j.ypmed.2014.10.028 62. Hossain M. Health and well-being of cancer caregivers in a changed role of breadwinners.. (2018) 55 422. DOI: 10.4103/ijc.IJC_266_18 63. Bremer P, Cabrera E, Leino-Kilpi H. Informal dementia care: Consequences for caregivers’ health and health care use in 8 European countries.. (2015) 119 1459-1471. DOI: 10.1016/j.healthpol.2015.09.014 64. Gupta S, Isherwood G, Jones K. Assessing health status in informal schizophrenia caregivers compared with health status in non-caregivers and caregivers of other conditions.. (2015) 15 1-11. DOI: 10.1186/s12888-015-0547-1 65. Tran TD, Biggs B-A, Holton S. Co-morbid anaemia and stunting among children of pre-school age in low-and middle-income countries: a syndemic.. (2019) 22 35-43. DOI: 10.1017/S136898001800232X 66. Kelly PJ, Cheng A, Spencer-Carver E. A syndemic model of women incarcerated in community jails.. (2014) 31 118-125. DOI: 10.1111/phn.12056 67. Bierman AS, Tinetti ME. Precision medicine to precision care: managing multimorbidity.. (2016) 388 2721-2723. DOI: 10.1016/S0140-6736(16)32232-2 68. Melchiorre MG, Papa R, Quattrini S. Integrated Care Programs for People with Multimorbidity in European Countries: eHealth Adoption in Health Systems.. (2020) 2020 1-23. DOI: 10.1155/2020/9025326 69. Hossain MM, Sultana A, Shaik AF. Psychoanalysis in the era of precision psychiatry.. (2020) 47 101840. DOI: 10.1016/j.ajp.2019.10.020 70. Mangin D, Parascandalo J, Khudoyarova O. Multimorbidity, eHealth and implications for equity: a cross-sectional survey of patient perspectives on eHealth.. (2019) 9 e023731. DOI: 10.1136/bmjopen-2018-023731 71. Ahasan R, Hossain MM. Leveraging GIS and spatial analysis for informed decision-making in COVID-19 pandemic.. (2020) 10 7-9. DOI: 10.1016/j.hlpt.2020.11.009 72. Murti M, Wong J, Whelan M. The need for integrated public health surveillance to address sexually transmitted and blood-borne syndemics.. (2019) 45 63-66. DOI: 10.14745/ccdr.v45i23a03 73. Ford N, Wi T, Easterbrook P. Global public health efforts to address HIV and related communicable disease syndemics.. (2020). DOI: 10.1097/COH.0000000000000636 74. Ford CL, Airhihenbuwa CO. Critical Race Theory, Race Equity, and Public Health: Toward Antiracism Praxis.. (2010) 100 S30-S35. DOI: 10.2105/AJPH.2009.171058 75. Pager D, Shepherd H. The Sociology of Discrimination: Racial Discrimination in Employment, Housing, Credit, and Consumer Markets.. (2008) 34 181-209. DOI: 10.1146/annurev.soc.33.040406.131740 76. Williams DR. Stress and the Mental Health of Populations of Color: Advancing Our Understanding of Race-related Stressors.. (2018) 59 466-485. DOI: 10.1177/0022146518814251 77. Northridge ME, Shepard PM. Environmental racism and public health.. (1997) 87 730-732. DOI: 10.2105/AJPH.87.5.730 78. Hossain M. A meta-knowledge analysis of global research on syndemics (2001-2020).. (2022, August 26). DOI: 10.17605/OSF.IO/8N9R6 79. Syndemics of psychosocial problems and HIV risk: A systematic review of empirical tests of the disease interaction concept.. (2015) 139 26-35. DOI: 10.1016/j.socscimed.2015.06.024 80. Commentary on "Syndemics of psychosocial problems and HIV risk: A systematic review of empirical tests of the disease interaction concept" by A. Tsai and B. Burns.. (2015) 145 129-31. DOI: 10.1016/j.socscimed.2015.07.016 81. Syndemics and Health Disparities: A Methodological Note.. (2016) 20 423-30. DOI: 10.1007/s10461-015-1260-2 82. Syndemics: A theory in search of data or data in search of a theory?. 206 117-122. DOI: 10.1016/j.socscimed.2018.03.040 83. Syndemic theory, methods, and data. (2022) 295. DOI: 10.1016/j.socscimed.2021.114656
--- title: 'Targeting Acanthamoeba proteins interaction with flavonoids of Propolis extract by in vitro and in silico studies for promising therapeutic effects' authors: - Imran Sama-ae - Suthinee Sangkanu - Abolghasem Siyadatpanah - Roghayeh Norouzi - Julalak Chuprom - Watcharapong Mitsuwan - Sirirat Surinkaew - Rachasak Boonhok - Alok K. Paul - Tooba Mahboob - Najme Sadat Abtahi - Tajudeen O. Jimoh - Sónia M.R. Oliveira - Madhu Gupta - Chea Sin - Maria de Lourdes Pereira - Polrat Wilairatana - Christophe Wiart - Mohammed Rahmatullah - Karma G. Dolma - Veeranoot Nissapatorn journal: F1000Research year: 2023 pmcid: PMC10015121 doi: 10.12688/f1000research.126227.3 license: CC BY 4.0 --- # Targeting Acanthamoeba proteins interaction with flavonoids of Propolis extract by in vitro and in silico studies for promising therapeutic effects ## Abstract Background: Propolis is a natural resinous mixture produced by bees. It provides beneficial effects on human health in the treatment/management of many diseases. The present study was performed to demonstrate the anti- Acanthamoeba activity of ethanolic extracts of Propolis samples from Iran. The interactions of the compounds and essential proteins of Acanthamoeba were also visualized through docking simulation. Methods: The minimal inhibitory concentrations (MICs) of Propolis extract against Acanthamoeba trophozoites and cysts was determined in vitro. In addition, two-fold dilutions of each of agents were tested for encystment, excystment and adhesion inhibitions. Three major compounds of Propolis extract such as chrysin, tectochrysin and pinocembrin have been selected in molecular docking approach to predict the compounds that might be responsible for encystment, excystment and adhesion inhibitions of A. castellanii. Furthermore, to confirm the docking results, molecular dynamics (MD) simulations were also carried out for the most promising two ligand-pocket complexes from docking studies. Results: The minimal inhibitory concentrations (MICs) 62.5 and 125 µg/mL of the most active Propolis extract were assessed in trophozoites stage of Acanthamoeba castellanii ATCC30010 and ATCC50739, respectively. At concentrations lower than their MICs values ($\frac{1}{16}$ MIC), Propolis extract revealed inhibition of encystation. However, at $\frac{1}{2}$ MIC, it showed a potential inhibition of excystation and anti-adhesion. The molecular docking and dynamic simulation revealed the potential capability of Pinocembrin to form hydrogen bonds with A. castellanii Sir2 family protein (AcSir2), an encystation protein of high relevance for this process in Acanthamoeba. Conclusions: The results provided a candidate for the development of therapeutic drugs against Acanthamoeba infection. In vivo experiments and clinical trials are necessary to support this claim. ## Amendments from Version 2 In this version, we need this additional grant (WU-SAH $\frac{0005}{2023}$), to be acknowledged in this article, from School of Allied health Sciences-Walailak University for the APC reimbursement that was paid to F1000 Journal. ## Introduction Acanthamoeba, a free-living ameba, is a causative agent of fatal granulomatous amoebic encephalitis (GAE), Acanthamoeba keratitis (AK), Acanthamoeba pneumonia (AP), cutaneous acanthamoebiasis, and disseminated acanthamoebiasis found in humans 1. In healthy individuals with contact lenses, Acanthamoeba keratitis is increasingly being recognized as a serious sight-threatening ocular infection in public health worldwide 2. Acanthamoeba life cycle includes an active trophozoite stage and a dormant cyst stage. The trophozoite stage is the motile form that acquires nutrients, neutral pH, adequate food supply, ambient temperature, and balanced osmolality, while the cyst is triggered by extreme conditions, such as food crisis, hyper- or hypo-osmolarity, temperature, and excessive acid to basic conditions. Regarding the Acanthamoeba keratitis, the cyst form can be found in the acceptor cornea and is difficult to treat due to the resilient nature of the cyst. Current treatment regimens usually include standard anti- Acanthamoeba drugs, biguanide and diamidine, for an effective treatment against cysts 3. However, long-term treatment has also been suggested to induce a resistant Acanthamoeba cyst form due to a non-specific symptom at the early stage of AK, which share other common features, such as eye pain and redness 4. Propolis or bee glue is a mixture of honeybees and natural products of different parts of plants 5 that is used for the construction and repairing beehives. Propolis hardens the cell wall of beehives, contributes to an aseptic internal environment 6, and acts as a protective barrier against predators. In addition, Propolis property contains several biological activities such as anti-inflammation, anti-proliferation, antioxidant, anti-diabetic, and antimicrobial activities 7– 9. Therefore, this study sought to evaluate an amoebicidal activity and anti- Acanthamoeba encystation, excystation and anti-adhesion by Propolis extracts that could offer an alternative treatment strategy for Acanthamoeba infection. Molecular docking simulation was included to predict a predominant binding mode of small molecules derived from Propolis with essential proteins from Acanthamoeba spp., to identify a relevant stable protein-ligand complex for future drug development. ## Preparation extracts Three Propolis samples were collected from Sardasht county, Boroujen city and Kermanshah city from Iran. The raw materials of Propolis were cut into small pieces, homogenizing (20 g) with 50 mL absolute ethanol and incubated at room temperature for seven days without shaking. Then, the extract was filtered through Whatman No. 1 filter paper, and the alcoholic extract was evaporated under vacuum with a rotary evaporator until it was dry. Dried extracts were preserved at 4°C and re-suspended in dimethyl sulfoxide (DMSO) at 100 mg/mL concentration before use. ## Culture of Acanthamoeba castellanii Acanthamoeba castellanii non-pathogenic strain (ATCC 30010) and Acanthamoeba castellanii pathogenic strain (ATCC 50739) were kindly given by Asst. Prof. Dr. Rachasak Boonhok, Walailak University. Trophozoites were grown in 75 cm 2 tissue culture flasks in Peptone Yeast Extract Glucose Broth (PYG) medium containing proteose peptone $0.75\%$ (w/v), yeast extract $0.75\%$ (w/v) and glucose $1.5\%$ (w/v) (purchased from HiMedia Laboratories Pvt. Ltd., Mumbai, India), without shaking at 28°C as described previously 10. For cysts, trophozoites were transferred from the PYG medium to the Neff’s encystment medium (NEM) containing 0.1 M KCl, 8 mM MgSO 4·7H 2O, 0.4 mM CaCl 2·2H 2O, 1 mM NaHCO 3, 20 mM ammediol (purchased from RCI Labscan Limited, Bangkok, Thailand) and were cultured in this medium for seven days to obtain mature cysts. After that, mature cysts were harvested and washed twice using 10 mL sterile phosphate-buffer saline (PBS). ## Determination of minimal inhibitory concentration (MIC) The MIC was determined by the micro-dilution method using serially diluted (two-fold) Propolis extracts. Determination of the MIC of the Propolis extract was examined according to a previous study 10. Stock solution of extracts (4 μL) were transferred into the first well of 96-well microplates, including 196 μL PYG medium to obtain a final concentration of 2,000 µg/mL. A two-fold serial dilution of the extracts were prepared in 96-well assay microplates to obtain concentrations in the range of 7.8–1,000 μg/mL in PYG medium. Then, 100 µL trophozoites or cysts (2×10 5 cells/mL) were added. The final volume in each well was 200 μL. Plates were incubated for 24 hours at 28°C. The percentage of cell viability was determined using $0.2\%$ trypan blue, obtained by manual counting under inverted microscopy (Nikon, Tokyo, Japan). The relative percentage of parasite viability was defined as: (mean of the treated parasite/mean of the control) × 100. The lowest concentration of extract that inhibited $90\%$ of A. castellanii growth was recorded as the MIC. The commercial antibiotic agent, chlorhexidine was used as positive control, while $1\%$ DMSO was used as negative (untreated) control. ## Anti-encystation on Acanthamoeba castellanii Anti-encystation was performed as previously studied 11 with modifications. Briefly, Acanthamoeba trophozoites (5×10 5 cells/mL) were incubated in Neff’s medium in a 96-well plate containing Propolis extracts at different concentrations ($\frac{1}{2}$ MIC, $\frac{1}{4}$ MIC, $\frac{1}{8}$ MIC, $\frac{1}{16}$ MIC). Plates were incubated at 28°C for seven days, and the total amoebae number was counted using a hemocytometer (Boeco, Hamburg, Germany). Subsequently, the sodium dodecyl sulfate (SDS, $0.5\%$ final concentration) was added and incubated for 1 hour to dissolve trophozoites and immature cysts. The remaining cysts were counted using a hemocytometer after the addition of SDS. To quantify encystation, the percentage of Acanthamoeba encystation was determined as follows: (total number of amoebae post-SDS treatment/total number of amoebae pre-SDS treatment) × 100. Phenylmethylsulfonyl fluoride (PMSF) (10 mM final concentration) was used as a positive control, whereas $1\%$ DMSO was used as a negative control. ## Anti-excystation on Acanthamoeba castellanii For excystation, Acanthamoeba cysts (5×10 5 cells/mL) were incubated with various concentrations of Propolis extracts ($\frac{1}{2}$ MIC, $\frac{1}{4}$ MIC, $\frac{1}{8}$ MIC, $\frac{1}{16}$ MIC) in PYG medium in 96-well plate at 28°C for seven days 12. The effects of the extract on excystation were observed under an inverted microscope. The total amoebae were counted using a hemocytometer while SDS ($0.5\%$ final concentration) was added and incubated for 1 hour to dissolve trophozoites and immature cysts. The remaining cysts were counted after the addition of SDS. To quantify excystation, the percentage of Acanthamoeba excystation was determined as follows: (total number of amoebae pre-treatment with SDS − total number of amoebae post-SDS treatment)/total number of amoebae pre-SDS treatment) × 100. PMSF (10 mM final concentration) and $1\%$ DMSO were used as positive and negative control, respectively. ## Anti-adhesion on Acanthamoeba castellanii The anti-adhesion assay was modified as previously reported 13. Trophozoites (4 × 10 5 cells/mL) were added to each well of a 96-well polystyrene microtiter plate supplemented with $\frac{1}{2}$ MIC, $\frac{1}{4}$ MIC, $\frac{1}{8}$ MIC, $\frac{1}{16}$ MIC of Propolis extract. Plates were incubated at 28°C without shaking for 24 hours. After incubation, a removing step to discard unbound trophozoites was performed. Plates were washed once with 0.1 M PBS, then air dried for 30 minutes at room temperature. The wells were stained with $0.1\%$ crystal violet assay for 30 minutes. The crystal violet was eliminated, and the plates were washed with water and air dried. An aliquot of DMSO was added to the well and the absorbance was read at OD 570 nm. Wells containing trophozoites with $1\%$ DMSO were used as control. The percentage of inhibition was calculated by following the formula: percentage of inhibition = (control OD – test OD/control OD) × 100. ## Cytotoxicity assay The cytotoxic effects of the most active Propolis extract were evaluated using the Vero cell line (ECACC 84113001, RRID:CVCL_0059). Cells were cultured in Dulbecco’s Modified Eagle’s medium (DMEM) (Merck KGaA, Darmstadt, Germany) supplemented with $10\%$ FBS (Sigma Aldrich, St. Louis, USA), and $1\%$ antibiotic containing penicillin G (100 units/mL) and streptomycin (100 μg/mL). The culture was incubated at 37°C, humidified with $5\%$ CO 2 in an incubator (non-shaking). After the cells reached $90\%$ confluence, the detachment was performed with trypsin and ethylene diamine tetra-acetic acid (EDTA) and incubated at 37°C in $5\%$ CO 2. Single cells at a density of 1.5 × 10 4 cells/100 μL were seeded into each well of a 96-well polystyrene plate and allowed to attach for 24 hours. Then, 100 μL propolis extract, eye drops, and combined set were gently added. After incubation for 24 hours, the cytotoxic effects were determined using an MTT assay 14, 15. The absorbance was measured using a microplate reader (Biotek, Cork, Ireland) at 570 nm. The survival percentage was calculated using the following equation: ABt and ABu denote the absorbance values of treated and untreated cells, respectively. ## Gas chromatography-mass spectrometry (GC-MS) analysis The Propolis extract (20 mg/mL) was diluted in ethanol (1:10), the solution was centrifuged for 10 minutes at a speed of 10,000 rpm at temperature of 10°C. The solution was used for analysis. GC-MS analysis was performed using Agilent Technology 7890 A (GC) equipped with 5977A Mass Selective Detector (MS) (Agilent, California, USA). A VF-WAXms capillary column of dimensions 30 m × 250 × 0.25 μM was used with helium gas as the carrier at 30 m × 250 × 0.25 μM at a flow rate of 1 mL/minute. The column temperature was initially programmed at 60°C, which was increased to 160°C at 10°C/minute and further increased to 325°C at 2.5°C/minute, hold time for 15 minutes. The mass spectra was collected at 70 eV ionization voltage over the range of m/z 35 to 500 in full scan mode. Chemical constituents were identified by comparing their mass spectral data with those from the Wiley library. ## Data analysis The experiments were repeated in triplicate. All data were recorded and entered into IBM SPSS Statistics (RRID:SCR_016479) version 26.0 (SPSS Inc. Chicago, IL, USA). Data were expressed as mean ± SD. Statistical analysis was conducted using a two-tailed unpaired Student’s t-test. $p \leq 0.05$ was considered statistically significant in all analyses. ## The three-dimensional (3D) structures prediction The effect of Propolis compounds on essential proteins of A. castellanii was investigated using the computational modelling method. This study focuses on three critical proteins: the Sir2 family protein, the mannose-binding protein, and the G protein-coupled receptor. The I-TASSER server was used to predict the 3D structures of these proteins 16, 17. FASTA sequences of A. castellanii Sir2 family protein (AcSir2) (NCBI Reference Sequence: XP 004358245.1) 18, A. castellanii mannose-binding protein (AcMBP) (GenBank: AAT37865.1) 19, and A. castellanii G protein-coupled receptor (AcGPCR) (GenBank: ELR16814.1) 18 were used as inputs, with no constraints or applied templates. The most confidently predicted model was constructed using the most significant templates in the threading alignments. Then, the quality of the predicted 3D model was further improved using ModRefiner 20. Finally, the stereochemical quality of the protein structures was determined using PROCHECK (RRID:SCR_019043) 21. ## Preparation of protein and ligand structures for molecular docking In this study, we used molecular docking to measure the binding energies of major compounds of Propolis such as pinocembrin, chrysin, and tectochrysin to those of A. castellanii essential proteins such as AcSir2, AcMBP, and AcGPCR to identify potential protein targets. Prior to the molecular docking process, the protein structures were dehydrated to expose only amino acid residues. Then, polar hydrogens were assigned to the protein structure, nonpolar hydrogens were merged, and Kollman charges were added to amino acid residues. The partial charges and atom types were assigned to stabilized protein structures and saved the files in the PDBQT formats (Protein Data Bank (PDB), Partial Charge (Q), and Atom Type (T)). For the preparation of the ligand, the PubChem database was queried for the 3D structures of pinocembrin (PubChem CID: 68071) 22, chrysin (PubChem CID: 5281607) 23, and tectochrysin (PubChem CID: 5281954) 24. Next, polar hydrogens and Gasteiger charges were introduced to the ligand structures, and nonpolar hydrogens were merged. Finally, the ligand structures were saved in the PDBQT format for stabilized ligand structures. After the receptor and ligand structures were prepared, the grid maps representing the system in the actual docking process were calculated with AutoGrid4 software version 4.2. The dimension of the grid was set to sufficiently cover the whole receptor structure (126 × 126 × 126 Å), with a spacing of 0.608 Å. All procedures were carried out using the AutoDock Auxiliary Tool (ADT) version 4.2 25, 26. ## Molecular docking of Propolis compounds to Acanthamoeba castellanii Sir2 family protein, mannose-binding protein, and G-protein coupled receptor AutoDock4 version 4.2 25, 26 was chosen for this purpose. Each docking step consisted of 50 GA runs with a maximum population size of 200 units. The total energy evaluation for each docking was 2,500,000 units. The average mutation rate was 0.02, the average cross-over rate was 0.80, and the average elitism value for each docking was 1. The Lamarckian Genetic Algorithm was used to combine local search (using the Solis and Wets algorithm) and global search (using the Genetic Algorithm alone) 27. This parameter was used to perform 10,000 independent docking runs on each ligand. This step was repeated five times to ensure the results were accurate. The protein-ligand lowest binding energy (ΔGbind) and the inhibitor constant were determined using AutoDock Auxiliary Tool (ADT) version 4.2 25, 26. ## Molecular dynamics (MD) simulation MD simulations were performed using the Desmond module (RRID:SCR_014575) from Schrödinger suite (RRID:SCR_014879) 28. In this process, hydrogen bonds were assigned according to standard procedures. The optimized potentials for liquid simulations (OPLS) force field were then applied to the protein and ligand complexes. The energy of the complexes was minimized after submerging them in a transferable intermolecular potential with 3 points (TIP3P) water model at a distance of 10 Å from the center of the box. The system was then neutralized by adding sodium and chloride ions, mimicking the in vivo environment. Molecular dynamic simulations were performed for 100 ns using ensembles of constant numbers of particles, pressure, and temperature (NPT) with a recording interval of 100 ps. The temperature was set to be 310.15 K and a pressure of roughly 1.01325 bar 29, 30. The following formula was used to determine the root mean square deviation (RMSD) trajectories of the protein-ligand interaction: RMSDx=1N∑$i = 1$N(ri′(tx))−ri(tref))2 where N is the number of chosen atoms, r' is the position of the chosen atoms in a frame x after they have overlapped in the reference frame, where frame x is captured at time t x, and t ref is the reference time. Each additional simulation frame required a new repeat of this process 28. The protein residues' root mean square fluctuation (RMSF) trajectories were determined using the following formula: RMSFi=1T∑$t = 1$T<(ri′(t))−ri(tref))2> where T stands for the trajectory time interval used to calculate the RMSF, r ′ stands for the position of the atoms in residue I following superposition in the reference, r i stands for the position of residue I, t ref stands for the reference time, and the angle brackets signify that the square distance is averaged on the atoms in the selected residue 28. The ligand atoms' RMSF trajectories were estimated using the following formula: RMSFi=1T∑$t = 1$T<(ri′(t))−ri(tref))2 where T is the trajectory time interval used to calculate the RMSF, r' is the position of atom I in the reference at time t following superposition on the reference frame, t ref is the reference time, and r is the location of atom I in the reference at time t ref 28. Desmond Schrödinger's module's simulation interaction diagram tool was used to analyze protein-ligand interactions, protein-ligand RMSD, and protein and ligand RMSF 28– 30. ## Molecular Mechanics Generalized Born Surface Area (MM-GBSA) free energy calculation The Prime Molecular Mechanics Generalized Born Surface Area (MM-GBSA) approach 31, which integrates the GBSA continuum solvent model 32, was used to calculate the contributions of enthalpy and entropy-related components toward the binding of the ligand-protein complex. The contributions from molecular mechanics energies, polar solvation, and nonpolar solvation terms were estimated (kcal/mol) using the equation: ΔGbind=Gcomplex−Gprotein−Gligand Where, ΔG bind = Calculated binding free energy of complex G complex = Binding free energy of minimized complex G protein = Binding free energy of receptor G ligand = Binding free energy of unbound ligand ## Protein and ligand visualization The proteins and ligands in this study were visualized using BIOVIA Discovery Studio version 21.1.0.20298 (RRID:SCR_015651) software 33 and the Mol Viewer 34. ## Drug likeliness prediction of the ligands using SwissADME analysis Drug-likeness profiles of ligands were unraveled through SwissADME, a free web tool to evaluate pharmacokinetics, drug-likeness, and medicinal chemistry of small molecules 35. ## Pharmacokinetics and toxicity prediction of the ligands The pharmacokinetic properties of the ligands, such as chemical absorption, distribution, metabolism, excretion, and toxicity (ADMET), were analyzed using the pkCSM ADMET descriptors algorithm methodology, an approach to the prediction of pharmacokinetic properties that relies on graph-based signatures 36. In brief, the canonical SMILES of the ligands (pinocembrin, chrysin, and tectochrysin) acquired from the PubChem database (RRID:SCR_004284) were used for input data, and ADMET profiles were generated. The Caco-2 permeability, intestinal absorption (human), and skin permeability were estimated to predict the absorption level of the ligands. The steady-state volume of distribution (VDss), fraction unbound (human), blood-brain barrier (BBB) permeability, and central nervous system (CNS) permeability were evaluated to predict the distribution of the ligands in various tissues. To predict the metabolism of the ligands in the human body, the ligands were determined whether they are likely to be CYP2D6/CYP3A4 substrates (the two main subtypes of cytochrome P450) or Cytochrome P450 inhibitors or not. To predict the excretion of the ligands, total compound clearance was measured. The compounds also determined whether they are likely going to be renal organic cation transporter 2 (OCT2) substrates or not. Finally, the toxicity of the ligands was predicted by AMES toxicity, hERG I/II inhibitor, oral rat acute toxicity (LD 50), oral rat chronic toxicity (LOAEL), hepatotoxicity, skin sensitization, and Minnow toxicity. The ADMET properties of the pinocembrin, chrysin, and tectochrysin are presented in the Table 5. To predict the absorption level of the compounds, water solubility, Caco-2 permeability, intestinal absorption (human), and skin permeability were estimated. A compound is easy to absorb if Caco-2 permeability is high. The Caco-2 permeability is considered as high if it has an apparent permeability coefficient (Papp) > 8 × 10 -6 cm/s (or log Papp > 0.90). The results showed that all ligands were predicted to have high Caco-2 permeability. About the human intestinal absorption prediction, a compound is poorly absorbed if absorbance is less than $30\%$. The results proved that all compounds were considered to have a good absorption. With regards to skin permeability, if a compound has a logKp > -2.5, the compound is predicted to have a relatively low skin permeability. The results indicated that all compounds were predicted to have good skin permeability. Table 5. \| Property \| Predicted Value \| Predicted Value.1 \| Predicted Value.2 \| Unit \| \| --- \| --- \| --- \| --- \| --- \| \| Property \| Pinocembrin \| Chrysin \| Tectochrysin \| Unit \| \| Property \| 68071 \| 5281607 \| 5281954 \| Unit \| \| Absorption \| Absorption \| Absorption \| Absorption \| Absorption \| \| Water solubility \| -3.538 \| -3.538 \| -3.641 \| log mol/L \| \| Cancer coli-2 (CaCo-2) permeability \| 1.152 \| 0.945 \| 1.248 \| log Papp in 10-6 cm/s \| \| Intestinal absorption (human) \| 92.417 \| 93.761 \| 95.229 \| % Absorbed \| \| Skin Permeability \| -2.808 \| -2.739 \| -2.758 \| log Kp \| \| P-glycoprotein substrate \| Yes \| Yes \| Yes \| Yes/No \| \| P-glycoprotein I inhibitor \| No \| No \| No \| Yes/No \| \| P-glycoprotein II inhibitor \| No \| No \| Yes \| Yes/No \| \| Distribution \| Distribution \| Distribution \| Distribution \| Distribution \| \| VDss (human) \| -0.386 \| 0.403 \| -0.047 \| log L/kg \| \| Fraction unbound (human) \| 0.022 \| 0.136 \| 0.119 \| Fu \| \| BBB permeability \| 0.42 (readily cross the BBB) \| 0.047 \| 0.003 \| log BB \| \| CNS permeability \| -2.047 \| -1.912 (penetrate the CNS) \| -1.992 (penetrate the CNS) \| log PS \| \| Metabolism \| Metabolism \| Metabolism \| Metabolism \| Metabolism \| \| CYP2D6 substrate \| No \| No \| No \| Yes/No \| \| CYP3A4 substrate \| No \| No \| Yes \| Yes/No \| \| CYP1A2 inhibitor \| Yes \| Yes \| Yes \| Yes/No \| \| CYP2C19 inhibitor \| Yes \| Yes \| Yes \| Yes/No \| \| CYP2C9 inhibitor \| Yes \| Yes \| Yes \| Yes/No \| \| CYP2D6 inhibitor \| No \| No \| No \| Yes/No \| \| CYP3A4 inhibitor \| No \| No \| Yes \| Yes/No \| \| Excretion \| Excretion \| Excretion \| Excretion \| Excretion \| \| Total Clearance \| 0.122 \| 0.405 \| 0.457 \| log mL/min/kg \| \| Renal OCT2 substrate \| No \| No \| No \| Yes/No \| \| Toxicity \| Toxicity \| Toxicity \| Toxicity \| Toxicity \| \| AMES toxicity \| No \| No \| No \| Yes/No \| \| hERG I inhibitor \| No \| No \| No \| Yes/No \| \| hERG II inhibitor \| No \| No \| No \| Yes/No \| \| Oral rat acute toxicity (LD 50) \| 1.586 \| 2.289 \| 2.042 \| mol/kg \| \| Oral rat chronic toxicity \| 2.059 \| 0.955 \| 0.744 \| log mg/kg-bw/day \| \| Hepatotoxicity \| No \| No \| No \| Yes/No \| \| Skin sensitization \| No \| No \| No \| Yes/No \| \| Minnow toxicity \| 1.683 \| 1.746 \| -0.04 \| log mM \| To predict the distribution of the compounds in various tissues, the VDss, Fraction unbound (human), BBB permeability, and central CNS permeability were evaluated. The VDss is relatively low if lower than 0.71 L/kg (log VDss < -0.15). Whereas it is high if higher than 2.81 L/kg (log VDss > 0.45). The results demonstrated that the VDss of the pinocembrin was higher than the chrysin and tectochrysin. The higher the VD is, the more of a ligand is distributed in tissue rather than plasma, as demonstrated in the pinocembrin. With regards to BBB and CNS permeability, the results indicated that the pinocembrin readily crossed the BBB, while chrysin and tectochrysin might penetrate the CNS. As cytochrome P450 is responsible for the metabolism of many drugs in liver, to predict metabolism of compounds, the compounds were determined whether they are likely to be CYP2D6/CYP3A4 substrates (the two main subtypes of cytochrome P450) or Cytochrome P450 inhibitors or not. The result predicted that all compounds were not substrates and inhibitors of the CYP2D6. However, all of them were inhibitors of CYP1A2, CYP2C19, and CYP2C9. Moreover, the tectochrysin was a substrate and inhibitor of CYP3A4. To predict the excretion of the compounds, total compounds clearance was measured. It was also determined whether the compounds were likely going to be renal organic cation transporter 2 (OCT2) substrates or not. With regards to total compounds clearance, the total clearance of the tectochrysin is the highest, followed by chrysin, and pinocembrin. Remarkably, all compounds were not predicted to be renal OCT2 substrates. Finally, to predict the toxicity of the compounds, AMES toxicity, hERG I/II inhibitor, oral rat acute toxicity (LD 50), oral rat chronic toxicity (LOAEL), hepatotoxicity, skin sensitization, and Minnow toxicity were predicted. Notably, the results predicted that none of the compounds were mutagenic or hERG I/II inhibitors, and none of them showed hepatotoxicity or skin sensitization. ## Anti- Acanthamoeba activities The effect of Propolis extracts on both strains of Acanthamoeba was examined, and the results exhibited as MIC are presented in Table 1 37. Propolis extract from the Kermanshah city exhibited the most inhibitory activity. The values of the minimum inhibitory concentration (MIC) ranged from 62.5 to 125 µg/mL in trophozoite form. But this extract had no inhibitory activity against cysts at 1,000 µg/mL concentration. In the positive control, chlorhexidine exhibited MIC values of 8 to 16 and 32 to 64 µg/mL for trophozoite and cyst forms, respectively. Therefore, Propolis from Kermanshah city was chosen for further study. Table 1. \| Propolis extract \| MIC value (µg/mL) \| MIC value (µg/mL).1 \| MIC value (µg/mL).2 \| MIC value (µg/mL).3 \| \| --- \| --- \| --- \| --- \| --- \| \| Propolis extract \| ATCC 30010 \| ATCC 30010 \| ATCC 50739 \| ATCC 50739 \| \| Propolis extract \| Trophozoite \| Cyst \| Trophozoite \| Cyst \| \| Sardasht county \| >1,000 \| >1,000 \| >1,000 \| >1,000 \| \| Boroujen city \| >1,000 \| >1,000 \| >1,000 \| >1,000 \| \| Kermanshah city \| 62.5 \| >1,000 \| 125 \| >1,000 \| \| Chlorhexidine \| 8 \| 32 \| 16 \| 64 \| ## Anti-encystation of Acanthamoeba castellanii To assess the effect of Propolis extract on A. castellanii encystation, PMSF was used as a positive control in Neff’s medium. According to the data presented in Figure 1, the results revealed that the Propolis-Kermanshah city exhibited inhibition of A. castellanii encystation at all concentrations. The formation of mature cysts significantly reduced after Propolis extract treatment at $\frac{1}{16}$ MIC on both strains of A. castellanii ATCC50739 ($21\%$) (Figure 1A) and A. castellanii ATCC30010 ($17\%$) (Figure 1B). In the 5 mM PMSF group, encystation was reduced to $6\%$ and $8\%$ for ATCC50739 and ATCC30010, respectively. Figure 1.: The effect of Propolis extract on encystation. The Propolis extract reduced the encystation in a dose-dependent manner on ( A) A. castellanii ATCC50739 and ( B) A. castellanii ATCC30010. The experiments were repeated three times, and the average values are presented with error bars representing standard deviations. ; significantly different at a P value of <0.05 by Student’s t test. PMSF, phenylmethylsulfonyl fluoride; ATCC, American Type Culture Collection.* ## Anti-excystation of Acanthamoeba castellanii The effect of Propolis extract treatment on excystation was assessed in PYG medium. The excystation rate decreased to $44\%$ and $42\%$ after exposure to high concentrations of propolis extract at $\frac{1}{2}$ MIC of trophozoites (Figure 2A and B). PMSF significantly inhibited Acanthamoeba excystation ($13\%$ and $4\%$) at 5 mM concentration. Figure 2.: The effect of Propolis extract on excystation. Propolis extract reduced the excystation on ( A) A. castellanii ATCC50739 and ( B) A. castellanii ATCC30010. The experiments were repeated three times, and the average values are presented with error bars representing standard deviations. ; significantly different at a P value of <0.05 by Student’s t test. PMSF, phenylmethylsulfonyl fluoride; ATCC, American Type Culture Collection.* ## Anti-adhesion assay To evaluate the influence of Propolis extract on the adhesion properties of Acanthamoeba trophozoites, the adhesion of trophozoites to the plastic surface varied and depended on the concentrations of extract. The strongest anti-adhesion was observed in trophozoites treated with $\frac{1}{2}$ MIC concentration of extract (Table 2) 37 in both strains of Acanthamoeba when compared with the untreated control. Table 2. \| Concentration of Propolis extract \| Anti-adhesion (%) \| Anti-adhesion (%).1 \| \| --- \| --- \| --- \| \| Concentration of Propolis extract \| ATCC50739 \| ATCC30010 \| \| 1/2 MIC \| 55.02 ± 4.14 \| 65.79 ± 3.11 \| \| 1/4 MIC \| 41.07 ± 7.53 \| 51.8 ± 0.77 \| \| 1/8 MIC \| 17.82 ± 3.68 \| 39.21 ± 3.23 \| \| 1/16 MIC \| 13.59 ± 3.13 \| 20.46 ± 2.46 \| \| Untreated \| 100 ± 0 \| 100 ± 0 \| ## Toxicity After 24 hours of treatment with Propolis extract, the number of viable cells was constant at low concentrations, ranging from 8–64 µg/mL. However, the survival rate of Vero cells was lower when treated with the extract at concentrations of 128–1,000 µg/mL. ## GC-MS analysis of Propolis extract The GC-MS analysis of the Propolis-Kermanshah city extract allowed the identification of 52 compounds (Supplementary Table 1 38). Chrysin ($18.86\%$) was the main compound present in the Propolis extract, followed by pinocembrin ($15.02\%$), and tectochrysin ($9.88\%$), respectively. ## 3D structure prediction of AcSir2, AcMBP, and AcGPCR The optimal 3D structural models of AcSir2, AcMBP, and AcGPCR were constructed using I-TASSER server and the top 10 threading templates. Then, the best C-score model was selected and refined. The refined 3D structure models of AcSir2, AcMBP, and AcGPCR are illustrated in Figure 3 39, 40. The AcSir2 is a protein located inside the nucleus. The protein consists of 536 amino acids (aa) that contain the SIR2 super-family region (aa residues 36–297) and YEATS family region (aa residues 443–524). The AcMBP is a large protein located at the cell membrane. The protein consists of 833 aa, and some residues such as aa residues 732–760 are transmembrane proteins. The AcGPCR is a protein also located in the cell membrane. The protein consists of 456 aa, and some of them are transmembrane proteins, such as aa residues 182–202, 214–236, 248–274, 286–305, 311–332, 353–375, and 381–401. The protein contains the lung seven-transmembrane receptor region (aa residues 140–413). The stereochemical quality of protein structures was analyzed using PROCHECK. The Ramachandran plot of the AcSir2 model identified $71.0\%$ of the residues in the most favored regions, $25.8\%$ of the residues in the other allowed regions, and only $3.2\%$ of the residues in disallowed regions. The Ramachandran plot of the AcMBP model discovered $64.7\%$ of the residues in the most favored regions, $32.3\%$ of the residues in the other allowed regions, and only $3.0\%$ of the residues in disallowed regions. The AcGPCR model's Ramachandran plot determined $85.5\%$ of the residues in the most favored regions, $13.3\%$ in other allowed regions, and only $1.2\%$ in disallowed regions. Figure 3.: The predicted three-dimensional structures of AcSIR2, AcMBP, and AcGPCR. AcSIR2: blue represents the SIR2 superfamily region, yellow represents the YEATS family region. AcMBP: Orange represents transmembrane proteins; green represents a domain of an unknown function. AcGPCR: purple represents Lung seven-transmembrane receptor. AcSIR2, A. castellanii Sir2 family protein; AcMBP, A. castellanii mannose-binding protein; AcGPCR, A. castellanii G protein-coupled receptor. ## Molecular docking of Propolis compounds to A. castellanii Sir2 family protein, mannose-binding protein, and G-protein coupled receptor The molecular docking of Propolis compounds such as: pinocembrin, chrysin, and tectochrysin against three essential proteins of A. castellanii was performed using AutoDock 4. The results are illustrated in Figure 4– Figure 6 40, 41. The Pinocembrin demonstrated good binding potential to the AcSir2 with binding energy (ΔGbind) of -7.63 kcal/mol and the inhibitory constant (Ki) of 2.57 µM. The compound interacts with the residues Glu147 through the conventional hydrogen bond (H-bond), Thr476 through Pi-lone pair, Phe477 through Pi-Pi T-shaped, Ser478 through conventional H-bond and Pi-Lone Pair, and Val482 through conventional H-bond. Chrysin exhibited a high affinity for AcSir2, with a ΔGbind of -8.05 kcal/mol and a Ki of 1.26 µM. The compound interacts with the residues Glu147 via the conventional H-bond, Phe272 via Pi-Pi T-shaped, Thr476 via Pi-lone pair, and Ser478 via Pi-donor H-bond. Tectochrysin exhibited an excellent affinity for AcSir2, with a ΔGbind of -8.12 kcal/mol and a Ki of 1.12 µM. The compound interacts with residues Arg122 through carbon or Pi-donor H-bond, Leu123 through alkyl or Pi-alkyl, Gly124 through conventional H-bond, Ile269 through alkyl or Pi-alkyl, Phe272 through Pi-Pi T-shaped, Phe477 through Pi-Pi T-shaped, and alkyl or Pi-alkyl, Ser478 through Pi-lone pair, and carbon or Pi-donor H-bone, and Pro479 through alkyl or Pi-alkyl (Figure 4). Figure 4.: The interaction of pinocembrin, chrysin, and tectochrysin toward the AcSIR2 protein predicted by molecular docking. ( A) Binding site of the ligands toward the AcSIR2 protein, purple compound represents pinocembrin, red compound represents chrysin, yellow compound represents tectochrysin. ( B) A schematic representation of the detailed interactions of the ligand atoms with the protein residues. ( C) Binding affinity and inhibitory constant prediction of propolis compounds against Sir2 family protein of Acanthamoeba castellanii. AcSIR2, A. castellanii Sir2 family protein. Figure 5.: The interaction of pinocembrin, chrysin, and tectochrysin toward the AcMBP protein predicted by molecular docking.( A) Binding site of the ligands toward the AcMBP protein, purple compound represents pinocembrin, red compound represents chrysin, yellow compound represents tectochrysin. ( B) A schematic representation of the detailed interactions of the ligand atoms with the protein residues. ( C) Binding affinity and inhibitory constant prediction of propolis compounds against mannose-binding protein of Acanthamoeba castellanii. AcMBP, A. castellanii mannose-binding protein. Figure 6.: The interaction of pinocembrin, chrysin, and tectochrysin toward the AcGPCR protein predicted by molecular docking. ( A) Binding site of the ligands toward the AcSIR2 protein, purple compound represents Pinocembrin, red compound represents Chrysin, yellow compound represents Tectochrysin. ( B) A schematic representation of the detailed interactions of the ligand atoms with the protein residues. ( C) Binding affinity and inhibitory constant prediction of propolis compounds against G protein-coupled receptor of Acanthamoeba castellanii. AcGPCR, A. castellanii G protein-coupled receptor; AcSIR2, A. castellanii Sir2 family protein. Pinocembrin demonstrated a weak binding affinity for AcMBP with ΔGbind of -6.34 kcal/mol and the Ki of 22.62 µM. The compound interacts with residues Pro593 through conventional H-bond and Pi-alkyl, Cys610 through Pi-sulfur, Thr625 through Pi-sigma, and Cys632 through Pi-alkyl. Chrysin had a very low affinity for AcMBP, with ΔGbind of -6.15 kcal/mol and the Ki of 30.92 µM. The compound interacts with the residues Pro263, Val267, and Pro317 through Pi-alkyl; Cys327 through Pi-sulfur and Pi-alkyl; Asp366 and Asn367 through conventional H-bond; and Phe369 through Pi-Pi Stacked. Tectochrysin showed a low affinity for AcMBP, with ΔGbind of -6.32 kcal/mol and the Ki of 23.21 µM. The compound interacts with residues Pro593 and Pro594 through Conventional H-bond; Glu596 through Pi-sigma; Cys610, Cys627, and Cys632 through Pi-sulfur; and Cys612 and Cys632 through Pi-Alkyl (Figure 5). Pinocembrin demonstrated a weak binding affinity for the AcGPCR with ΔGbind of -7.04 kcal/mol and the Ki of 6.96 µM. The compound interacts with residues Ile275 through Pi-alkyl, Phe278 through conventional H-bond and unfavorable donor-donor interaction, Leu279 through Pi-alkyl, Asp283 through Amide-Pi stacked, Lys284 through Pi-alkyl, and Arg346 through conventional H-bond. With a ΔGbind of -6.98 kcal/mol and Ki of 7.7 µM, chrysin exhibited a low affinity for AcGPCR. The compound interacts with AcGPCR in the same way as the pinocembrin-AcGPCR complex, except for the residue Phe278, whose conventional H-bonding did not occur for this compound. Finally, tectochrysin exhibited a low affinity for AcGPCR, with ΔGbind of -7.17 kcal/mol and the Ki of 5.51 µM. The compound interacts with AcGPCR the same as the pinocembrin-AcGPCR complex, except for the residue Phe278 in which there was no interaction for this compound (Figure 6). Based on the molecular docking result, pinocembrin, chrysin, and tectochrysin demonstrated inhibition potential towards the AcSir2 protein. Tectochrysin showed the most robust inhibition, followed by chrysin and pinocembrin. Thus, the molecular dynamics of these complexes were then simulated to understand the dynamic motions and analyze the stabilities of these protein-ligand complexes. ## Molecular dynamic simulations of apo and bound forms of AcSir2 protein The dynamic motions of the apo and docked complexes were further analyzed by molecular dynamic simulations at 100 ns using the Desmond module of Schrödinger's suite. The results of MD simulations of the apo and bound forms of AcSir2 protein are illustrated in Figure 7 and Figure 8 40, 42, respectively. For an MD run of 100 ns, the RMSD and the RMSF were predicted for the apo and bound forms. A ligand’s interaction can ward off unfolding and stabilize the protein 43. Hence, we analyzed the protein’s secondary structures before and after docking to understand the conformational changes due to ligand binding. The RMSD quantifies the average change in displacement of a selection of atoms relative to a reference frame for a particular frame. Figure 7A, Figure 8A, 8D, and 8G demonstrated the protein RMSD from the simulation of the AcSir2 apo form, pinocembrin-AcSir2 complex, chrysin-AcSir2 complex, and tectochrysin-AcSir2 complex, respectively. The Protein RMSD (P-RMSD) shows how the RMSD of a protein has changed over time (left Y-axis). After aligning all of the protein frames with the backbone of the reference frame, the atoms are chosen to figure out the P-RMSD. During the simulation, the calculation of the P-RMSD can give information about how the structure is built. For the ligand RMSD (L-RMSD), the L-RMSD value (right Y-axis) shows how stable the ligand is concerning the protein and its binding pocket. ' Lig fit Prot' illustrated the RMSD of a ligand after the protein-ligand complex was aligned in the reference protein backbone and the RMSD of the ligand heavy atoms was determined. If the observed values exceed the P-RMSD by a significant amount, the ligand almost certainly has diffused away from its initial binding site. The mean values of P-RMSD of the apo-AcSir2 was around the 10 Å (Figure 7A). The P-RMSD values of the Pinocembrin-AcSir2 complex wildly deviated at around the first 12 ns. After that, the fluctuation was regular at the end of the simulation. An average of RMSD values is stable after 12 ns, indicating that the system has equilibrated during this simulation. Furthermore, for the L-RMSD values of the Pinocembrin-AcSir2 complex, the observed values are significantly lower than the P-RMSD, so the ligand has likely fixed in its initial binding site (Figure 8A, Supplementary Figure 1, Supplementary video 1 found as Underlying data 38). The P-RMSD values of the chrysin-AcSir2 complex deviated from 0 to 14 Å in the first duration of 18 ns and slightly decreased during 18 ns to 50 ns. The fluctuation was regular at the end of the simulation after 70 ns, with the highest point about 15 Å and the lowest point about 10 Å. An average of RMSD values is constant after 70 ns, indicating that the system has equilibrated during this simulation. The L-RMSD values of this complex show significantly lower than the RMSD of the protein during the first 90 ns. However, the values dramatically increase around 90 ns and are higher than the P-RMSD. Almost obviously, the ligand may have diffused away from its initial binding site (Figure 8D, Supplementary Figure 2, Supplementary video 2 found as Underlying data 38). The fact that the P-RMSD values of the tectochrysin-AcSir2 complex strongly deviated throughout the simulation shows that a substantial conformational change has occurred in the protein. However, the overall average is relatively stable after 60 ns, indicating that the system has equilibrated during this simulation. The L-RMSD values of this complex are lower than the RMSD of the protein during the first 80 ns. However, the values gradually increase during the simulation time of 80–90 ns, then more extensive than the P-RMSD at approximately 90 ns. Clearly, the ligand may have diffused away from its initial binding site (Figure 8G, Supplementary Figure 3, Supplementary video 3 found as Underlying data 38). The RMSF can characterize local changes in the protein chain and the positions of the ligand atoms. Figure 7B, Figure 8B, 8E, and 8H demonstrated the protein and ligand RMSF from the simulation of the apo-AcSir2, pinocembrin-AcSir2 complex, chrysin-AcSir2 complex, and tectochrysin-AcSir2 complex, respectively. For the protein RMSF (P-RMSF), the peaks in this plot correspond to the protein regions that fluctuate the most during the simulation. The P-RMSF of the apo-AcSir2 strongly fluctuated at amino acid residues Pro19, Pro112, Cys192-Gly212, Pro318-Ala357, Arg364-Met378, Thr409-Glu412, Pro423, His432, Ala434-Pro436, and Pro515-Ala536 (Figure 7B). The P-RMSF of the pinocembrin-AcSir2 complex strongly fluctuated at amino acid residues Pro112, Pro204, Asp317, Pro320, Pro324, Pro334, Pro423, Pro436, Val450, and Pro515. However, these residues were not ligand contacts residues (Figure 8B). The P-RMSF of the chrysin-AcSir2 complex immensely fluctuated at amino acid residues Pro19, Pro112, Pro204, Asp317, Pro324, Pro334, Pro423, His432, Pro436, Lys451, and Pro515. These residues were also not ligand contacts residues (Figure 8E). The P-RMSF of the tectochrysin-AcSir2 complex wildly fluctuated at amino acid residues Pro19, Pro112, Pro204, Val319, Pro324, Pro423-Thr437, Thr443, Pro151, and Gly525. Almost all these residues were not ligand contacts residues, except for the Pro204 position (Figure 8H). The result illustrated that the binding of a ligand can prevent protein unfolding and stabilize it. During the simulation, the interactions of the protein with the ligand can be monitored. The protein-ligand contacts diagrams for the pinocembrin-AcSir2 complex, chrysin-AcSir2 complex and tectochrysin-AcSir2 complex are illustrated in Figures 8C, 8F, and 8I, respectively. The stacked bar charts demonstrated that all complexes exhibited H-bonds, hydrophobic interactions, ionic bonds, and water bridges during the simulation. The PRIME MM-GBSA binding free energy values of the ligand-AcSir2 complexes are given in Table 3. Figure 7.: The P-RMSD and P-RMSF of the apo-AcSIR2 protein.( A) Plot of the P-RMSD of the apo-AcSIR2 protein. ( B) Plot of the P-RMSF of the apo-AcSIR2 protein. P-RMSD, protein root mean square deviation; P-RMSF, protein root mean square fluctuation; AcSIR2, A. castellanii Sir2 family protein. Figure 8.: Simulation interactions diagram of pinocembrin, chrysin and tectochrysin with essential proteins.( A–C) Simulation interactions diagram of pinocembrin-AcSIR2 complex. (A) Plot of protein-ligand RMSD. ( B) Plot of protein RMSF. ( C) Histogram of protein-ligand contacts categorized by type of interactions: hydrogen bonds (green), hydrophobic (purple), ionic (magenta), and water bridges (blue). (D-F) Simulation interactions diagram of chrysin-AcSIR2 complex. ( D) Plot of protein-ligand RMSD. ( E) Plot of protein RMSF. ( F) Histogram of protein-ligand contacts categorized by type of interactions: hydrogen bonds (green), hydrophobic (purple), ionic (magenta), and water bridges (blue). ( G–I) Simulation interactions diagram of tectochrysin-AcSIR2 complex. ( G) Plot of protein-ligand RMSD. (H) Plot of protein RMSF. ( I) Histogram of protein-ligand contacts categorized by type of interactions: hydrogen bonds (green), hydrophobic (purple), ionic (magenta), and water bridges (blue). RMSD, root mean square deviation; RMSF, root mean square fluctuation; AcSIR2, A. castellanii Sir2 family protein. TABLE_PLACEHOLDER:Table 3. ## Drug likeliness prediction of the ligands using SwissADME After careful analysis of the drug-likeness properties of the ligands using SwissADME, the result indicated that the compound properties are within the range of drug-likeness based on various filters such as Lipinski 44, Ghose 45, Veber 46, Egan 47, and Muegge 48 (Table 4). Table 4. \| Drug likeness \| Drug likeness.1 \| Drug likeness.2 \| Drug likeness.3 \| \| --- \| --- \| --- \| --- \| \| Filters \| Pinocembrin \| Chrysin \| Tectochrysin \| \| Lipinski \| Yes; 0 violation \| Yes; 0 violation \| Yes; 0 violation \| \| Ghose \| Yes \| Yes \| Yes \| \| Veber \| Yes \| Yes \| Yes \| \| Egan \| Yes \| Yes \| Yes \| \| Muegge \| Yes \| Yes \| Yes \| \| Bioavailability Score \| 0.55 \| 0.55 \| 0.55 \| ## Discussion The pharmaceutical activities of natural products have historically been screened because they are thought to have key roles in drug discovery, are inexpensive and rarely have undesirable side effects. Propolis has been known for a long time and attracted scientific interest due to its biological activities such as anti-viral, anti-bacterial, anti-fungal, anti-protozoal, anesthetic, antioxidant, anti-tumoral, anti-cancer, anti-hepatotoxic, anti-mutagenic, anti-septic and anti-inflammatory activities, in addition to being utilized for its cytotoxic activity 49, 50. In vitro studies of its anti-parasitic effect were reported against Leishmania spp., Trypanosoma spp., Plasmodium spp., Cryptosporidium spp., Giardia spp., Toxoplasma gondii, Trichomonas vaginalis, and Blastocystis spp. 51. In literature, Propolis extract has reported amoebistatic activity between 2.0 and 6.0 mg/mL and its effects were amoebicidal at 8.0 mg/mL or higher 52. Here, the anti- Acanthamoeba activities of three Propolis extracts from different cities in Iran were screened. The highest activity was obtained from the Propolis ethanolic extract of Kermanshah city, and the MIC against trophozoites was 62.5 µg/mL and 125 µg/mL for A. castellanii ATCC30010 and ATCC50739, respectively. Propolis composition included more than 180 different types of chemicals 53 depending on several factors such as extraction method, source of plant, season, and local flora 54. Flavonoid compounds like chrysin and pinocembrin are commonly identified in Romanian, Turkish and Polish Propolis. The type of Uruguayan propolis mentioned other flavonoid compounds like tectochrysin, galangin and kaempferol 55. This study revealed the main compounds of Propolis from Kermanshah city were chrysin, pinocembrin, and tectochrysin. To determine the cytotoxic effect of the extract at a concentration of at least 0.128 mg/mL was demonstrated against Vero cells. Our data agreed with Vural et al. 56, in which the Propolis concentration at higher than 7.81 mg/mL caused corneal epithelial cell damage. The safe concentration of Propolis at 1.4 mg/kg per day was also recommended 57. A. castellanii ATCC50739 and ATCC30010 were tested for their encystment capability in Neff’s medium. The results seemed to encyst and presented the mature cysts in both media for seven days. The process of Acanthamoeba is an essential for the survival under unfavorable conditions 58. The double wall of the Acanthamoeba cyst is resistant to many drugs and chemicals and leads to clinical drug resistance 59. As only single cyst surviving in the cornea stroma after initial successful treatment, they can regularly excyst and lead to reinfection 60. Thus, the inhibition of encystation process during the treatment of Acanthamoeba infections can lead to more favorable outcomes and enhances the potential of Acanthamoeba keratitis treatment. In this study, PMSF was used as a positive control to block serine proteinase, providing a significant inhibition of encystation. The data were in agreement with the results of Leitsch et al. 61 in which the PMSF inhibited the proteolytic activity at the early stage of encystation. The Propolis extract at low concentration ($\frac{1}{16}$ MIC) was able to inhibit the encystation of A. castellanii ATCC50739 and ATCC30010. The low concentration caused a reduction in the level of encystation of around 80–$90\%$. The high concentrations ($\frac{1}{2}$-$\frac{1}{8}$ MIC) gave a < $20\%$ reduction in the encystment levels, which suggests that low concentration of Propolis extract is suitable for inhibiting the encystment process. The main mechanism underlying inhibition of encystation by Propolis remain largely unknown. Aqeel et al. 62 mentioned phenolic compounds such as resveratrol and demethoxycurcumin are strong antioxidants with Acanthamoeba growth inhibitory effects in vitro. It raises the possibility that antioxidant activity may be required to inhibit Acanthamoeba encystation. Furthermore, Mahboob et al. 63 reported that other phenolic compounds i.e., ester of caffeic acid and quinic acid, demonstrated the inhibitory effect on encystation by scavenging reactive oxygen species within Acanthamoeba cytoplasm. The use of therapeutic agents for Acanthamoeba infection may lead to cyst formation, a drug-resistance stage, and transformation of cysts to trophozoites that lead to recurrence of infection 64. The fluids or some microorganisms in eye infection may provide an appropriate condition to induce excystation of surviving Acanthamoeba cyst 65. This reason remains a challenge for Acanthamoeba keratitis prevention. Although Propolis extract from Kermanshah city did not inhibit the growth of Acanthamoeba cysts at 1,000 µg/mL, it exhibited excystment inhibition at 62.5 and 31.25 µg/mL. This evidenced that it prevented the recurrence of infection because there was no change of morphological transformation from cysts to trophozoites. However, it remains unclear on how Propolis inhibited Acanthamoeba excystation. Maslinic acid, a natural triterpene found in olives and Propolis, has been shown to inhibit parasitic proteases enzymes 66. These proteases enzymes are normally secreted within the first 24 hours, which may indicate an important role of the enzyme in excystation 67. The first step in the pathogenesis of Acanthamoeba infection is the adhesion to the surface of the host tissues. Subsequently, the adhesion to host cells, Acanthamoeba produce proteinase enzymes that work in concert to produce a potent cytopathic effect (CPE) involving killing of the host cells, degradation of epithelial basement, and penetration into the deeper layers of the cornea 68. In this study, we showed that Propolis possess anti-amoebic properties and the capability to reduce amoebae adhesion on plastic plate. The highest adhesion was noticed in the control group, which was an untreated agent. Similar results were obtained in the current study, where anti-adhesion was observed in plastic plate and contact lenes belonging to Curcuma longa extract 69, Annona muricata and Combretum trifoliatum extracts 70, and Garcinia mangostana and their pure compounds 13. Based on our results, we recognize the importance of developing Propolis extract to eliminate or inhibit the pathogenicity of Acanthamoeba. Therefore, the determination the main target of the pathogen in silico has been studied. A molecular docking simulation was carried out to investigate the binding affinities of the major compounds (chrysin, pinocembrin and tectochrysin) from the Propolis extract and essential proteins in Acanthamoeba (AcSir2, AcMBP, and AcGPCR). AcSir2 was classified as a class-IV sirtuin. This protein exhibited functional SIRT deacetylase activity, localized mainly in the nucleus, and its transcription was upregulated during encystation 71. Acanthamoeba mannose-binding protein (AcMBP) is a virulence factor of the free-living amoeba, which is important for adhesion of the pathogen 72. G proteins and GPCRs are well known key regulators of cellular communication and cellular functions including cell cycle, mitosis, and proliferation 73. After blind docking of the three ligands with the AcMBP, the Chrysin bounded to a different binding site on AcMBP. As AcMBP is a virulence factor of the free-living amoeba, which is important for adhesion of the pathogen 72. We hypothesize that this protein might have more than one binding site to help them adhere to the surface. To test this hypothesis, we predicted all this protein's binding sites with PrankWeb 74. The results are consistent with the hypothesis. The results showed that this protein has more than one pocket and some pockets have similar probability scores (Supplementary Figure 4 and Supplementary Tables 2 and 3 38). It might be possible that this protein has more than one binding site.” Our study revealed the potential capability of the pinocembrin, chrysin, and tectochrysin complex to form hydrogen bonds with the AcSir2 protein. The low binding energy indicates strong interactions between the compounds and AcSir2 protein. Sirtuins have been classified into five major classes (I, II, III, IV and V) and conserved from bacteria to humans 71. In some parasites, Sir2 is located mainly in the nucleus and plays a role in cell function, proliferative life span and development under various conditions 75. Notably, the regulation of AcSir2 expression is essential for growth and encystation in A. castellanii. In AcSir2-overexpressing encysting cells, the transcription of cellulose synthase was highly upregulated compared to control cells 71. Moreover, MD simulations indicated that pinocembrin, chrysin, and tectochrysin can interact with AcSir2 protein. Chrysin and tectochrysin may have a probability of diffusing away from their initial binding site. Over the 100 ns of MD simulations, only the pinocembrin remained fixed within its initial binding site. However, pinocembrin, chrysin, and tectochrysin seem to bind and inhibit Cytochrome P450, including CYP1A2, CYP2C19 and CYP2C9. Because cytochrome P450 is primarily found in the liver as a crucial detoxification enzyme in the body. This enzyme oxidises xenobiotics to help their excretion. In addition, the cytochrome P450 system can activate and deactivate many drugs 36. Therefore, these agents should be used carefully in patients taking other drugs to avoid drug-drug interaction. Thus, this compound may be an excellent candidate for future anti- Acanthamoeba drug development. In this study, our successful combination of computational approaches and phenotypic screening led to the identification of compounds with noteworthy activities against Acanthamoeba. ## Conclusions Natural products are one of the essential resources for drug discovery. Considering the pharmacological activities of Propolis extract against Acanthamoeba, its therapeutic potential should be considered. Our study was conducted with extracts of Propolis. Moreover, molecular docking was used as a computational and easily accessible method to propose a binding mode of chrysin, tectochrysin and pinocembrin on a protein target. Molecular docking stimulation indicated that pinocembrin is the strongest binding site on AcSir2 protein. This noteworthy data further allows us to simulate the effects of pinocembrin or its synthetic structural modifications to optimize desirable activities and targets. Nevertheless, our results provide the possibility of finding a new series of anti- Acanthamoeba compounds that can act in combination with conventional drugs as an alternative therapeutic strategy for the treatment of AK. ## Underlying data NCBI Protein: transcriptional regulator, Sir2 family protein [Acanthamoeba castellanii str. Neff]. Accession number XP_004358245.1; https://identifiers.org/ncbiprotein:xp004358245.1 18. NCBI Protein: mannose-binding protein [Acanthamoeba castellanii]. Accession number AAT37865.1; https://identifiers.org/ncbiprotein:AAT37865.1 19. NCBI Protein: G protein coupled receptor, putative [Acanthamoeba castellanii str. Neff]. Accession number ELR16814.1; https://identifiers.org/ncbiprotein:ELR16814.1 18. NCBI PubChem Compound: Pinocembrin. PubChem CID 68071; https://identifiers.org/pubchem.compound:68071 22. NCBI PubChem Compound: Chrysin. PubChem CID 5281607; https://identifiers.org/pubchem.compound:5281607 23. NCBI PubChem Compound: Tectochrysin. PubChem CID 5281954; https://identifiers.org/pubchem.compound:5281954 24. Figshare: In vitro RAW DATA.xlsx. https://doi.org/10.6084/m9.figshare.21213563 37. Figshare: MD simulations Movie.rar. https://doi.org/10.6084/m9.figshare.21213560 38. This project contains the following underlying data: Figshare: Raw data for Molecular docking. https://doi.org/10.6084/m9.figshare.21214079 41. This project contains the following underlying data: Figshare: Raw data for Molecular dynamics (MD) simulation. https://doi.org/10.6084/m9.figshare.21214160 42. This project contains the following underlying data: Figshare: Raw data for the prediction of the three-dimensional structures. https://doi.org/10.6084/m9.figshare.21214184 39. This project contains the following underlying data: Figshare: F1000_raw figures. https://doi.org/10.6084/m9.figshare.21312297 40. This project contains the following underlying data: Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0). ## References 1. Kot K, Łanocha-Arendarczyk N, Kosik-Bogacka D. Immunopathogenicity of. (2021) 22 1261-1280. DOI: 10.3390/ijms22031261 2. Pinto LF, Andriolo BNG, Hofling-Lima AL. The role of. (2021) 120 2717-2729. DOI: 10.1007/s00436-021-07240-6 3. Khan NA. (2006) 30 564-95. DOI: 10.1111/j.1574-6976.2006.00023.x 4. Turner NA, Russell AD, Furr JR. Emergence of resistance to biocides during differentiation of. (2000) 46 27-34. DOI: 10.1093/jac/46.1.27 5. Pasupuleti VR, Sammugam L, Ramesh N. Honey, Propolis, and royal jelly: a comprehensive review of their biological actions and health benefits.. (2017) 2017 1259510-1259531. DOI: 10.1155/2017/1259510 6. Shehu A, Ismail S, Rohin MAK. Antifungal properties of Malaysian Tualang honey and stingless bee Propolis against. (2016) 6 044-050. DOI: 10.7324/JAPS.2016.60206 7. Przybyłek I, Karpiński TM. Antibacterial properties of Propolis.. (2019) 24 2047-2064. DOI: 10.3390/molecules24112047 8. Shi YZ, Liu YC, Zheng YF. Ethanol extract of Chinese Propolis attenuates early diabetic retinopathy by protecting the blood-retinal barrier in streptozotocin-induced diabetic rats.. (2019) 84 358-369. DOI: 10.1111/1750-3841.14435 9. Hwang S, Hwang S, Jo M. Oral administration of Korean Propolis extract ameliorates DSS-induced colitis in BALB/c mice.. (2020) 17 1984-1991. DOI: 10.7150/ijms.44834 10. Mitsuwan W, Bunsuwansakul C, Leonard TE. (2020) 114 194-204. DOI: 10.1080/20477724.2020.1755551 11. Dudley R, Alsam S, Khan NA. The role of proteases in the differentiation of. (2008) 286 9-15. DOI: 10.1111/j.1574-6968.2008.01249.x 12. Anwar A, Numan A, Siddiqui R. Cobalt nanoparticles as novel nanotherapeutics against. (2019) 12 280. DOI: 10.1186/s13071-019-3528-2 13. Sangkanu S, Mitsuwan W, Mahboob T. Phytochemical, anti-Acanthamoeba, and anti-adhesion properties of. (2022) 226 106266. DOI: 10.1016/j.actatropica.2021.106266 14. Mitsuwan W, Wintachai P, Voravuthikunchai SP. (2020) 77 3546-3554. DOI: 10.1007/s00284-020-02164-3 15. Wilson C, Lukowicz R, Merchant S. Quantitative and qualitative assessment methods for biofilm growth: A mini-review.. (2017) 6 1-25. PMID: 30214915 16. Yang J, Yan R, Roy A. The I-TASSER suite: protein structure and function prediction.. (2015) 12 7-8. DOI: 10.1038/nmeth.3213 17. Yang J, Zhang Y. I-TASSER server: new development for protein structure and function predictions.. (2015) 43 W174-W181. DOI: 10.1093/nar/gkv342 18. Clarke M, Lohan AJ, Liu B. ([Dataset]. 2013) 14 R11. DOI: 10.1186/gb-2013-14-2-r11 19. Garate M, Cao Z, Bateman E. (2004) 279 29849-56. DOI: 10.1074/jbc.M402334200 20. Xu D, Zhang Y. Improving the physical realism and structural accuracy of protein models by a two-step atomic-level energy minimization.. (2011) 101 2525-2534. DOI: 10.1016/j.bpj.2011.10.024 21. Laskowski RA, MacArthur MW, Moss DS. (1993) 26 283-291. DOI: 10.1107/S0021889892009944 22. 22 National Center for Biotechnology Information: PubChem Compound Summary for CID 68071, Pinocembrin.[Dataset].2022. http://pubchem.ncbi.nlm.nih.gov/compound/Pinocembrin. (2022) 23. 23 National Center for Biotechnology Information: PubChem Compound Summary for CID 5281607, Chrysin.[Dataset].2022. http://pubchem.ncbi.nlm.nih.gov/compound/Chrysin. (2022) 24. 24 National Center for Biotechnology Information: PubChem Compound Summary for CID 5281954, Tectochrysin.[Dataset].2022. http://pubchem.ncbi.nlm.nih.gov/compound/Tectochrysin. (2022) 25. Morris GM, Huey R, Lindstrom W. AutoDock4 and AutoDockTools4: Automated docking with selective receptor flexibility.. (2009) 30 2785-2791. DOI: 10.1002/jcc.21256 26. Forli S, Huey R, Pique ME. Computational protein-ligand docking and virtual drug screening with the AutoDock suite.. (2016) 11 905-919. DOI: 10.1038/nprot.2016.051 27. Solis FJ, Wets RJB. Minimization by random search techniques.. (1981) 6 19-30. DOI: 10.1287/moor.6.1.19 28. Shaw DE. Desmond Molecular Dynamics System, Tools for Maestro-Desmond interoperability.. (2021) 29. Pant S, Singh M, Ravichandiran V. Peptide-like and small-molecule inhibitors against COVID-19.. (2021) 39 2904-2913. DOI: 10.1080/07391102.2020.1757510 30. Thangavel N, Al Bratty M, Al Hazmi HA. Molecular docking and molecular dynamics aided virtual search of OliveNet™ directory for secoiridoids to combat SARS-CoV-2 infection and associated hyperinflammatory responses.. (2021) 7 627767. DOI: 10.3389/fmolb.2020.627767 31. Shivakumar D, Williams J, Wu Y. Prediction of absolute solvation free energies using molecular dynamics free energy perturbation and the OPLS force field.. (2010) 6 1509-1519. DOI: 10.1021/ct900587b 32. Naresh P, Selvaraj A, Shyam Sundar P. Targeting a conserved pocket (n-octyl-β-D-glucoside) on the dengue virus envelope protein by small bioactive molecule inhibitors.. (2022) 40 4866-4878. DOI: 10.1080/07391102.2020.1862707 33. Discovery studio visualizer. v21.1.0.20298.. (2021) 34. Sehnal D, Bittrich S, Deshpande M. *Mol Viewer: Modern web app for 3D visualization and analysis of large biomolecular structures.. (2021) 49 W431-W437. DOI: 10.1093/nar/gkab314 35. Daina A, Michielin O, Zoete V. SwissADME: a free web tool to evaluate pharmacokinetics, drug-likeness and medicinal chemistry friendliness of small molecules.. (2017) 7 42717. DOI: 10.1038/srep42717 36. Pires DE, Blundell TL, Ascher DB. pkCSM: Predicting small-molecule pharmacokinetic and toxicity properties using graph-based signatures.. (2015) 58** 4066-4072. DOI: 10.1021/acs.jmedchem.5b00104 37. Sama-ae I. figshare (2022). DOI: 10.6084/m9.figshare.21213563 38. Sama-ae I. figshare (2022). DOI: 10.6084/m9.figshare.21213560 39. Sama-ae I. figshare (2022). DOI: 10.6084/m9.figshare.21214184 40. Sama-ae I. figshare (2022). DOI: 10.6084/m9.figshare.21312297 41. Sama-ae I. figshare (2022). DOI: 10.6084/m9.figshare.21214079 42. Sama-ae I. figshare (2022). DOI: 10.6084/m9.figshare.21214160 43. Mazal H, Aviram H, Riven I. Effect of ligand binding on a protein with a complex folding landscape.. (2018) 20 3054-3062. DOI: 10.1039/c7cp03327c 44. Lipinski CA. Lead- and drug-like compounds: The rule-of-five revolution.. (2004) 1 337-341. DOI: 10.1016/j.ddtec.2004.11.007 45. Ghose AK, Viswanadhan VN, Wendoloski JJ. A knowledge-based approach in designing combinatorial or medicinal chemistry libraries for drug discovery. 1. A qualitative and quantitative characterization of known drug databases.. (1999) 1 55-68. DOI: 10.1021/cc9800071 46. Veber DF, Johnson SR, Cheng HY. Molecular properties that influence the oral bioavailability of drug candidates.. (2002) 45 2615-2623. DOI: 10.1021/jm020017n 47. Egan WJ, Merz KM, Baldwin JJ. Prediction of drug absorption using multivariate statistics.. (2000) 43 3867-3877. DOI: 10.1021/jm000292e 48. Muegge I, Heald SL, Brittelli D. Simple selection criteria for drug-like chemical matter.. (2001) 44 1841-1846. DOI: 10.1021/jm015507e 49. Toreti VC, Sato HH, Pastore GM. Recent progress of Propolis for its biological and chemical compositions and its botanical origin.. (2013) 2013 697390. DOI: 10.1155/2013/697390 50. Sforcin JM. Biological properties and therapeutic applications of Propolis.. (2016) 30 894-905. DOI: 10.1002/ptr.5605 51. Asfaram S, Fakhar M, Keighobadi M. Promising anti-protozoan activities of Propolis (bee glue) as natural product: A review.. (2021) 66 1-12. DOI: 10.1007/s11686-020-00254-7 52. Topalkara A, Vural A, Polat Z. (2007) 23 40-5. DOI: 10.1089/jop.2006.0053 53. Kuropatnicki AK, Szliszka E, Krol W. Historical aspects of propolis research in modern times.. (2013) 2013 964149. DOI: 10.1155/2013/964149 54. Dantas Silva RP, Machado BA, Barreto GA. Antioxidant, antimicrobial, antiparasitic, and cytotoxic properties of various Brazilian Propolis extracts.. (2017) 12 e0172585. DOI: 10.1371/journal.pone.0172585 55. Kurek-Górecka A, Keskin Ş, Bobis O. Comparison of the antioxidant activity of Propolis samples from different geographical regions.. (2022) 11 1203. DOI: 10.3390/plants11091203 56. Vural A, Polat ZA, Topalkara A. The effect of Propolis in experimental. (2007) 35 749-54. DOI: 10.1111/j.1442-9071.2007.01620.x 57. Burdock GA. Review of the biological properties and toxicity of bee Propolis (Propolis).. (1998) 36 347-63. DOI: 10.1016/s0278-6915(97)00145-2 58. Blanton WE, Villemez CL. Molecular size and chain length distribution in. (1978) 25 264-267. DOI: 10.1111/j.1550-7408.1978.tb04412.x 59. Huang FC, Shih MH, Chang KF. Characterizing clinical isolates of. (2017) 50 570-577. DOI: 10.1016/j.jmii.2015.10.011 60. Moore MB, McCulley JP, Newton C. (1987) 94 1654-61. DOI: 10.1016/S0161-6420(87)33238-5 61. Leitsch D, Köhsler M, Marchetti-Deschmann M. Major role for cysteine proteases during the early phase of. (2010) 9 611-8. DOI: 10.1128/EC.00300-09 62. Aqeel Y, Iqbal J, Siddiqui R. Anti-acanthamoebic properties of resveratrol and demethoxycurcumin.. (2012) 132 519-23. DOI: 10.1016/j.exppara.2012.09.007 63. Mahboob T, Azlan AM, Tan TC. Anti-encystment and amoebicidal activity of. (2016) 9 866-871. DOI: 10.1016/j.apjtm.2016.07.008 64. Lakhundi S, Khan NA, Siddiqui R. The effect of environmental and physiological conditions on excystation of. (2014) 113 2809-2816. DOI: 10.1007/s00436-014-3941-6 65. Chávez-Munguía B, Omaña-Molina M, González-Lázaro M. Ultrastructural study of encystation and excystation in. (2005) 52 153-158. DOI: 10.1111/j.1550-7408.2005.04-3273.x 66. Zulhendri F, Chandrasekaran K, Kowacz M. Antiviral, antibacterial, antifungal, and antiparasitic properties of Propolis: A review.. (2021) 10 1360. DOI: 10.3390/foods10061360 67. Neal RA. Enzymic proteolysis by. (1960) 50 531-550. DOI: 10.1017/s0031182000025610 68. Panjwani N. Pathogenesis of. (2010) 8 70-79. DOI: 10.1016/s1542-0124(12)70071-x 69. Mitsuwan W, Sangkanu S, Romyasamit C. (2020) 14 218-229. DOI: 10.1016/j.ijpddr.2020.11.001 70. Mitsuwan W, Sin C, Keo S. Potential anti-. (2021) 7 e06976. DOI: 10.1016/j.heliyon.2021.e06976 71. Joo SY, Aung JM, Shin M. The role of the. (2020) 13 368. DOI: 10.1186/s13071-020-04237-5 72. Garate M, Marchant J, Cubillos I. (2006) 47 1056-62. DOI: 10.1167/iovs.05-0477 73. Aqeel Y, Siddiqui R, Manan Z. The role of G protein coupled receptor-mediated signaling in the biological properties of. (2015) 81 22-7. DOI: 10.1016/j.micpath.2015.03.006 74. Jendele L, Krivak R, Skoda P. PrankWeb: a web server for ligand binding site prediction and visualization.. (2019) 47 W345-W349. DOI: 10.1093/nar/gkz424 75. Religa AA, Waters AP. Sirtuins of parasitic protozoa: In search of function(s).. (2012) 185 71-88. DOI: 10.1016/j.molbiopara.2012.08.003 76. The use of dimethyl sulfoxide in contact lens disinfectants is a potential preventative strategy against contracting Acanthamoeba keratitis.. (2016) 39 389-93. DOI: 10.1016/j.clae.2016.04.004 77. In Vitro Evaluation of the Combination of Melaleuca alternifolia (Tea Tree) Oil and Dimethyl Sulfoxide (DMSO) against Trophozoites and Cysts of Acanthamoeba Strains. Oxygen Consumption Rate (OCR) Assay as a Method for Drug Screening.. (2021) 10. DOI: 10.3390/pathogens10040491
--- title: Activation of PPAR-γ prevents TERT-mediated pulmonary vascular remodeling in MCT-induced pulmonary hypertension authors: - Tafseel Hussain - Limin Chai - Yan Wang - Qianqian Zhang - Jian Wang - Wenhua Shi - Qingting Wang - Manxiang Li - Xinming Xie journal: Heliyon year: 2023 pmcid: PMC10015197 doi: 10.1016/j.heliyon.2023.e14173 license: CC BY 4.0 --- # Activation of PPAR-γ prevents TERT-mediated pulmonary vascular remodeling in MCT-induced pulmonary hypertension ## Abstract ### Background It has been demonstrated that elevated telomerase reverse transcriptase (TERT) expression or activity is implicated in pulmonary hypertension (PH). In addition, activation of peroxisome-proliferator-activated receptor γ (PPAR-γ) has been found to prevent PH progression. However, the molecular mechanism responsible for the protective effect of PPAR-γ activation on TERT expression in the pathogenesis of PH remains unknown. This study was performed to address these issues. ### Methods Intraperitoneal injection of monocrotaline (MCT) was used to establish PH. BIBR1532 was applied to inhibit the activity of telomerase. The right ventricular systolic pressure (RVSP) and histological analysis were used to detect the development of PH. The protein levels of p-Akt, t-Akt, c-Myc and TERT were determined by western blotting. Pharmacological inhibition of TERT by BIBR1532 effectively suppressed RVSP, RVHI and the WT% in MCT-induced PH rats. ### Results Pharmacological inhibition of Akt/c-Myc pathway by LY294002 diminished TERT upregulation, RVSP, RVHI and WT% in MCT-PH rats. Activation of PPAR-γ by pioglitazone inhibited p-Akt and c-Myc expressions and further downregulated TERT, thus to reduced RVSP, RVHI and WT% in MCT-treated PH rats. ### Conclusions In conclusion, TERT upregulation contributes to PH development in MCT-treated rats. Activation of PPAR-γ prevents pulmonary arterial remodeling through Akt/c-Myc/TERT axis suppression. ## Graphical abstract Image 1 ## Introduction Pulmonary arterial hypertension is a severe disease characterized by abnormal structure and dysfunction of pulmonary blood vessels caused by multiple factors, leading to the sustained increases in pulmonary vascular resistance and pulmonary arterial pressure, and eventually right heart failure. Currently, the main pathological mechanisms underlying PH compromise persistent pulmonary vasoconstriction, vascular remodeling and thrombosis in situ [1,2]. Pulmonary vascular remodeling has been considered as the major structural alteration of PH, in which aberrant proliferation of pulmonary arterial smooth muscle cells (PASMCs) plays the key role in this process. Therefore, elucidating the molecular mechanisms responsible for PASMCs proliferation in pulmonary vascular remodeling is essential for identifying novel treatments of PH. Telomerase reverse transcriptase (TERT) encodes the catalytic subunit of telomerase to maintain telomerase activation [3,4]. Telomerase reactivation is a prerequisite for cell immortalization, and telomerase over expression exists in more than $85\%$ of human tumours [5,6]. Recent studies have suggested that TERT is upregulated and activated in PH patients and a rat model of PH [7,8], and that TERT regulates cyclin D1 and G1/S phase transitions to promote cell proliferation [9,10]. Knockdown of TERT in an animal model of PH effectively alleviated PASMCs proliferation and pulmonary vascular remodeling [11]. Recently, we have reported that PDGF promotes PASMC proliferation and migration through the Akt/c-MYC/TERT axis in vitro. Furthermore, activation of PPAR-γ with pioglitazone suppressed PDGF-induced TERT expression and telomerase activation, leading to inhibition of PASMC proliferation and migration [12]. However, the underlying mechanisms of how individual targets in the above pathways mediate PASMC proliferation and pulmonary vascular remodeling in vivo are not fully understood. The PI3K/Akt signaling pathway is an intracellular signal transduction pathway that is involved in many cell process, including cell metabolism, proliferation, survival, and angiogenesis [1,2,13,14]. Abnormal activation of the PI3K/AKT pathway has been found to promote PASMCs proliferation in PH [15]. Recent studies have demonstrated that activation of PI3K/Akt mediates TERT upregulation to promote tumour cell proliferation [16]. Besides, inhibition of PI3K/Akt pathway suppresses lung cancer proliferation via blockage of TERT expression [17]. These studies suggest that TERT may be an important target downstream of the PI3K/Akt signaling pathway to promote cell proliferation. However, whether PI3K/Akt also activates TERT and subsequently promotes vascular remodeling in PH remains to be explored. Peroxisome-proliferator-activated receptor γ (PPAR-γ), a family member of nuclear receptors, has beneficial effects on cardiovascular systems that regulating adipogenesis and glucose metabolism [18]. Accumulating evidence have demonstrated that activation of PPAR-γ may be a novel PH therapeutic target. Pioglitazone, one of the most potent and selective synthetic agonists of PPAR-γ receptors, has been found to suppresses PASMCs proliferation and protect against PH development in experimental models [19,20]. Further studies have found that activation of PPAR-γ upregulates tumour suppressor gene PTEN and inhibit PI3K/Akt signaling pathway, thus to prevent pulmonary vascular remodeling in PH model [21,22]. Taken together, these findings lead to our hypothesis that PPAR-γ acts as a pivotal meadiator for PASMCs proliferation/migration and pulmonary vascular remodeling, these effects could be mediated by PI3K/Akt/c-MYC signaling pathway, and subsequently promotes TERT upregulation and pulmonary vascular remodeling. To address the above issues, we used mono-crotaline (MCT)-induced PH rat model to examine the role of PPAR-γ activation by pioglitazone on inhibition of TERT-mediated vascular remodeling in PH, thus to develop new insight and appropriate targets for the management of PH. ## Drugs and reagents MCT (Must Bio-technology, Chengdu, China) was used to establish PH. BIBR1532 (APExBIO Techonologies, Houston, TX, USA) was used to impede the activity of telomerase. LY294002 (HY-MedChemExpress, Monmouth, NJ, USA). Pioglitazone (Cayman Chemical Company, Michigan, USA), was used to active PPAR-γ. Antibody against Akt, phospho-Akt (Ser473), c-Myc and GAPDH were purchased from Cell signaling Technology (Danvers, MA, USA). Polyclonal antibody against TERT were provided by Abcam (Cambridge, MA, USA). ## Animals Male Sprague-Dawley (SD) rats (190–200 g) were purchased from Xi'an Jiaotong University Animal Experiment Centre. All animal experiments were carried out in accordance with the Guide for Care and Use of Laboratory Animals of Xi'an Jiaotong University Animal Experiment Centre and approved by Laboratory Animal Care Committee of Xi'an Jiaotong University. They were kept under standard conditions with same light/dark cycle and housed in wire cages with free access to food and water. ## Experimental design The rats were randomly divided into five groups ($$n = 8$$ each group); control group, MCT group, MCT plus LY294002 treatment group, MCT plus BIBR1532 5 mg/kg treatment group and MCT plus BIBR1532 10 mg/kg treatment group, and MCT plus pioglitazone treatment group. Models of PH rats were established by a single intraperitoneal injection of MCT (60 mg/kg) on day 1 as previously described, pioglitazone (10 mg/kg) was given daily by gavage tube for 28 days. LY294002 (0.03 mg/kg) was given daily by intraperitoneal injection for 28 days. BIBR1532 were injected intraperitoneally with 5 mg/kg or 10 mg/kg twice a week for 28 days, respectively. Rats in control group were received an equal volume of $0.9\%$ NaCl solution. The concentrations of the compounds were chosen based on previous studies [[23], [24], [25], [26]]. ## Measurement of RVSP and RVHI Four weeks after MCT injection, all survived rats were anesthetized for hemodynamic measurements. As described previously, all rats underwent closed-chest right heart catheterization to detect right ventricle systolic pressure (RVSP), which was considered equal to pulmonary arterial pressure (PAP) [27]. After hemodynamic measurements, the harvested hearts were dissected into right ventricle RV, left ventricle LV plus interventricular septum, respectively. Each chamber was weighed separately to assess the right ventricular hypertrophy index (RVHI), a ratio of the weight of RV to LV plus S [RV/(LV + S)]. ## Histological analysis Marginal right upper pulmonary lobes were fixed in $4\%$ paraformaldehyde, and then the fixed lungs tissues were embedded in paraffin and sectioned at a thickness of 5 μm and stained with haematoxylin and eosin (HE) as previously reported [23]. Pulmonary vascular remodeling was evaluated by assessing the percentage of medial wall thickness (%WT). In each rat, ten pulmonary arteries (20–70 μm) were determined for structural integrity using a morphometric image system (Image J) [28]. The %WT was calculated as (external diameter-internal diameter)/external diameter × $100\%$. ## Western blot analysis Lung tissues from rats were lysed in RIPA Lysis Buffer, and then centrifuged at 13,000 rpm at 4 °C for 15 min. The supernatant was collected as sample proteins. Proteins were loaded and separated in SDS-PAGE gel and transferred to Polyvinylidene difluoride (PVDF) membranes as described previously [29]. After being blocked with $5\%$ non-fat milk for 1 h at room temperature, membranes were incubated overnight at 4 °C with primary antibodies against TERT (1:1000 dilution, #ab191523, Abcam), p-Akt (1:1000 dilution, #13008, CST), t-Akt (1:1000 dilution, #4691, CST), c-Myc (1:1000 dilution, #5605, CST) and GAPDH (1:1000 dilution, TA802519, Origene) according to the manufacturer's instructions. Then the membranes were incubated with horseradish peroxidase conjugated with goat anti-rabbit or mouse IgG antibody. Immunoreactive bands were visualized with enhanced chemiluminescence (ECL) kit. Images were digitally captured using a ChemiDoc XPS System. The band densities were quantified using Quality One software (Bio-Rad). ## Statistical analysis All data were presented as mean ± SEM. One-way ANOVA followed by a Tukey post hoc test was conducted to analyse the differences between groups. P value < 0.05 was considered statistically significant. ## Effect of TERT inhibition on MCT-induced PH pulmonary vascular remodeling To investigate whether TERT expression is elevated in lungs of MCT-induced rats, the protein level of TERT was detected by immunoblotting. As shown in Fig. 1A, the TERT expression was highly increased to (1.89 ± 0.28)-fold over control compared with the control group ($P \leq 0.05$). The result indicates that TERT is increased in MCT-treated PH rats. Fig. 1Effect of TERT inhibition on MCT-induced PH pulmonary vascular remodeling. ( A) TERT protein level in lung tissues of rats was determined using immunoblotting, GAPDH served as a loading control ($$n = 4$$ each group). ( B) Representative tracings of RVSP in each group ($$n = 6$$). ( C) Changes of RV/LV + S in each group ($$n = 6$$). ( D) HE-staining photomicrographs of small pulmonary vessels in different groups ($$n = 10$$ per rat). * $P \leq 0.05$ versus control group, #$P \leq 0.05$ versus MCT treated group. Fig. 1 Next, we examined whether suppression of TERT prevent MCT-induced PH development. BIBR1532, a potent and selective TERT inhibitor, was administrated to rat after MCT injection. As shown in Fig. 1A, administration of TERT inhibitor BIBR1532 of low (5 mg/kg) and high (10 mg/kg) dose significantly reduced the protein level of TERT ($P \leq 0.05$ vs. MCT-treated rats), indicating that TERT is effectively inhibited. In MCT-induced PH rats, RVSP significantly increased to 68.04 ± 7.92 mmHg versus 25.62 ± 8.11 mmHg in control rats ($P \leq 0.05$, Fig. 1B), suggesting that PH was successfully established in MCT-treated group. However, inhibition of TERT with BIBR1532 for low and high dose partially reduced RVSP to 43.2 ± 6.36 mmHg or 42.9 ± 6.15 mmHg ($P \leq 0.05$ vs. MCT group, Fig. 1B), respectively. Similar changes were found in RVHI. Fig. 1C showed that RVHI dramatically increased from 0.22 ± 0.06 in control rats to 0.57 ± 0.02 in MCT-treated rats ($P \leq 0.05$). After treatment with BIBR1532 of low and high dose, RVHI decreased to 0.34 ± 0.05 or 0.40 ± 0.07 ($P \leq 0.05$ vs. MCT group), respectively. These results suggest that inhibition of TERT prevents the development of MCT-induced PH. Finally, we assessed the effects of TERT inhibition by BIBR1532 on MCT-induced pulmonary vascular remodeling using histological analysis. As depicted in Fig. 1D, the HE-staining results showed that medial wall thickness of small pulmonary arteries in MCT-treated rats were dramatically elevated compared with control rats. In addition, the quantitative morphometric analysis confirmed that %WT was significantly increased from (12.0 ± 1.0) % in control rats to (28.3 ± 1.5) % in MCT-treated rats ($P \leq 0.05$). However, after administration of BIBR1532 of low and high dose prevented MCT-induced increases in medial wall thickness, and reduced %WT to (17.9 ± 1.5) % or (17.0 ± 0.7) % ($P \leq 0.05$ vs. MCT group), respectively. Taken together, these results indicate that TERT inhibition dramatically prevents the pulmonary arterial remodeling by suppressing the proliferation of PASMCs. ## PI3K/Akt/c-Myc pathway mediates TERT upregulation and pulmonary arterial remodeling in MCT-induced PH rats It has been demonstrated that PI3K/Akt/c-Myc pathway regulates TERT expression in tumour cells. In this study, we investigated whether this above axis was involved in TERT expression in MCT-induced PH rats. As shown in Fig. 2A, administration of LY294002 suppressed MCT induced increases in p-Akt (1.57 ± 0.2 vs. 2.56 ± 0.10) and c-Myc (1.50 ± 0.10 vs. 2.02 ± 0.11). In addition, treatment with LY294002 significantly reduced MCT induced change in TERT expression from (2.18 ± 0.15)-fold to (1.45 ± 0.11)-fold, respectively ($P \leq 0.05$ vs. MCT group, Fig. 2A). Moreover, administration with LY294002 significantly reduced RVSP, RVHI, and wall thickness ($P \leq 0.05$ vs. MCT group, Fig. 2B–D). Together, these results suggest that PI3K/Akt/c-Myc signaling pathway mediates MCT-induced TERT upregulation and pulmonary arterial remodeling. Fig. 2Inhibition of PI3K/Akt/c-Myc prevents MCT-induced PH in rats. ( A) Protein level in lung tissues of rats was determined using immunoblotting, GAPDH served as a loading control ($$n = 4$$ each group. ( B) Representative tracings of RVSP in each group ($$n = 6$$). ( C) Changes of RV/LV + S in each group ($$n = 6$$). ( D) HE-staining photomicrographs of small pulmonary vessels in different groups ($$n = 10$$ per rat). * $P \leq 0.05$ versus control group, #$P \leq 0.05$ versus MCT treated group. Fig. 2 ## PPAR-γ activation by pioglitazone prevents the development of MCT-induced PH rats As shown in Fig. 3A and B, the RVSP and RVHI declined from 68.04 ± 7.92 mmHg and 0.57 ± 0.02 in MCT-treated rats to 45.2 ± 5.4 mmHg and 0.39 ± 0.06 in pioglitazone-treated PH rats, respectively ($P \leq 0.05$). Similar changes were observed in medial thickness of pulmonary arteries. Fig. 3C showed that medial thickness of pulmonary arteries was reduced. Quantitative morphometric analysis confirmed that %WT dropped to (17.7 ± 2.0) % in pioglitazone-treated PH rats ($P \leq 0.05$). These results suggest that activation of PPAR-γ by pioglitazone alleviates the pulmonary arterial remodeling by inhibiting PASMCs proliferation. Fig. 3Pioglitazone prevents the development of MCT-induced PH in rats. ( A) Representative tracings of RVSP in each group ($$n = 6$$). ( B) Changes of RV/LV + S in each group ($$n = 6$$). ( C) HE-staining photomicrographs of small pulmonary vessels in different groups ($$n = 10$$ per rat). * $P \leq 0.05$ versus control group, #$P \leq 0.05$ versus MCT treated group. Fig. 3 ## Mechanisms underlying PPAR-γ activation preventing MCT-induced PH rats We further examined whether this beneficial effect of PPAR-γ activation is associated with the inhibition of Akt/c-Myc/TERT axis. Fig. 4A showed that the level of Akt phosphorylation increased to (2.03 ± 0.18)-fold compared with control group, while this level decreased to (1.5 ± 0.2)-fold over control in the pioglitazone-treated PH rats. Similarly, the level of c-Myc protein reached to (2.0 ± 0.2)-fold increase over control in MCT-treated rats and declined to (1.3 ± 0.1)-fold over control in the MCT plus pioglitazone-treated PH rats (Fig. 4B). In addition, Fig. 4C showed TERT protein level decreased to (1.4 ± 0.1)-fold over control in pioglitazone-treated PH rats ($P \leq 0.05$), suggesting that activation of PPAR-γ suppresses TERT expression by inhibiting Akt/c-Myc signaling pathway. Fig. 4Mechanisms underlying Pioglitazone preventing MCT-induced PH rats. Expression levels of p/t-Akt (A), c-Myc (B), TERT(C) in lung tissues from different groups were determined using western blotting, GAPDH served as loading controls ($$n = 4$$ each group). * $P \leq 0.05$ versus control group, #$P \leq 0.05$ versus MCT treated group. Fig. 4 ## Discussion In this study, we have found that TERT is highly expressed in MCT-induced PH rats and inhibition of TERT expression significantly prevents MCT-induced PH progression by suppressing pulmonary vascular remodeling. In addition, activation of PPAR-γ might be associated with suppression of Akt/c-Myc pathway and subsequently inhibition of TERT expression, thus to alleviate PH development. Our study provides a new molecular basis whereby inhibition of TERT function might benefit for PH management by preventing pulmonary vascular remodeling. Telomerase activity constitutes the addition of noncoding TTAGGG nucleotide repeats, preventing the ends of chromosomes from deteriorating [30]. TERT, an important component of telomerase, maintains telomere homeostasis by lengthening telomeric DNA [31]. It has been confirmed that telomerase activation is an early event in the development of cancers, especially the TERT, which plays an important role in this process [32,33]. TERT is expressed at low levels or not detectable in normal somatic cells and tissues while its expression is upregulated in most carcinoma cells and highly proliferative organs [34]. Recently, several studies suggested TERT expression or activity increase in lungs from patients with idiopathic PH and animal models with PH. These is consistent with the results of our current study, which showed that overexpression of TERT was found in MCT-induced PH rats, which once again verified the reliability of our in vitro experiment [7,8,12,35]. To further investigate the effects of TERT in PH, we used BIBR1532 as a pharmacological inhibitor. The inhibition of TERT resulted in reduced wall thickness, RVSP and right hypertrophy. These results constitute the causal role of TERT in pulmonary vascular remodeling in PH. PI3K/Akt signaling pathway participates in cell proliferation, differentiation, apoptosis, glucose transport and other cellular functions. A variety of growth factors, such as platelet-derived growth factor (PDGF) and hypoxia, activates PI3K/Akt signaling pathway, promoting proliferation and migration of various tumour cells. Recent studies have demonstrated that inhibition of PI3K/Akt pathway inhibits TERT expressions, which further diminishes proliferation of lung adenocarcinoma cells and breast cancer cells [36,37]. These studies suggest that TERT is an important target gene which downstream the PI3K/Akt signaling pathway and promote cell proliferation. In HaCaT cells, chromatin immunoprecipitation experiments have shown that c-MYC regulates the transcriptional activity of TERT by enhanced binding to TERT promoters [38]. In present study, we found that inhibition of PI3K/Akt pathway significantly suppressed c-Myc and TERT upregulation and pulmonary arterial remodeling in MCT induced PH rats model. Peroxisome proliferator-activated receptor γ (PPAR-γ) is a member of the nuclear receptor family. PPAR-γ forms a heterodimer with the retinoic acid X receptor family and binds to DNA containing The repeated 5′-AGGTCA-3′ sequence of PPAR response elements (PPAR response elements, PPREs) regulate downstream target genes and signal pathways, mediate fat metabolism, cell proliferation, differentiation and migration and other physiological processes, which are considered to be one An important endogenous protective molecule [39]. Clinically, PPAR-γ agonists such as pioglitazone have been used to treat type 2 diabetes [40]. In recent years, studies have further suggested that activation of PPAR-γ may have potential clinical application value in a variety of diseases other than diabetes such as cardiovascular diseases, tumours, and inflammation [19]. Another study has indicated that PPAR-γ down-regulates the expression of TERT, depending on Wnt/β-Catenin signal way in GC cell lines [38]. In this study, we found that activation of PPAR-γ by pioglitazone suppressed Akt and c-Myc expressions, which further downregulated TERT, thus to prevent pulmonary vascular remodeling in PH. Our study offers a novel perspective of the mechanisms responsible for the protective effects of PPAR-γ activation on experimental PH. ## Conclusion In summary, we determine that TERT upregulation significantly promotes the development of MCT-induced PH, and activation of PPAR-γ suppresses pulmonary vascular remodeling by inhibition of Akt/c-Myc/TERT axis in MCT-induced PH. Targeting TERT signaling pathways may have potential value in therapeutic interventions for PAH. ## Ethics approval and consent to participate This article does not contain any studies with human participants or animals performed by any of the authors. ## Author contribution statement Tafseel Hussain, Limin Chai, Yan Wang, Qianqian Zhang, Jian Wang, Wenhua Shi, Qingting Wang, Manxiang Li, Xinming Xie: Conceived and designed the experiments; Performed the experiments; Analyzed and interpreted the data; Contributed reagents, materials, analysis tools or data; Wrote the paper. ## Funding statement Xinming Xie was supported by $\frac{10.13039}{501100001809}$National Natural Science Foundation of China [81800052]. ## Consent for publication All authors have seen the manuscript and approved to submit to your journal. ## Data availability statement Data included in article/supp. material/referenced in article. ## Declaration of interest's statement The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. ## References 1. Cool C.D.. The hallmarks of severe pulmonary arterial hypertension: the cancer hypothesis-ten years later. Am. J. Physiol. Lung Cell Mol. Physiol. (2020) 318 L1115-l1130. PMID: 32023082 2. Simonneau G.. Haemodynamic definitions and updated clinical classification of pulmonary hypertension. Eur. Respir. J. (2019) 53 3. Leão R.. Mechanisms of human telomerase reverse transcriptase (hTERT) regulation: clinical impacts in cancer. J. Biomed. Sci. (2018) 25 22. PMID: 29526163 4. Rosen J.. Non-canonical functions of telomerase reverse transcriptase - impact on redox homeostasis. Redox Biol. (2020) 34 5. Chen X.. Therapeutic strategies for targeting telomerase in cancer. Med. Res. Rev. (2020) 40 532-585. PMID: 31361345 6. Roake C.M., Artandi S.E.. Regulation of human telomerase in homeostasis and disease. Nat. Rev. Mol. Cell Biol. (2020) 21 384-397. PMID: 32242127 7. Boucherat O.. The cancer theory of pulmonary arterial hypertension. Pulm. Circ. (2017) 7 285-299. PMID: 28597757 8. Pullamsetti S.S.. Translational advances in the field of pulmonary hypertension. From cancer biology to new pulmonary arterial hypertension therapeutics. Targeting cell growth and proliferation signaling hubs. Am. J. Respir. Crit. Care Med. (2017) 195 425-437. PMID: 27627135 9. Zhang H., Hu N.. Telomerase reverse transcriptase induced thyroid carcinoma cell proliferation through PTEN/AKT signaling pathway. Mol. Med. Rep. (2018) 18 1345-1352. PMID: 29901196 10. Bajaj S.. Targeting telomerase for its advent in cancer therapeutics. Med. Res. Rev. (2020) 40 1871-1919. PMID: 32391613 11. Mouraret N.. Role for telomerase in pulmonary hypertension. Circulation (2015) 131 742-755. PMID: 25550449 12. Zhang Q.. PPARγ activation inhibits PDGF-induced pulmonary artery smooth muscle cell proliferation and migration by modulating TERT. Biomed. Pharmacother. (2022) 152 13. Hoxhaj G., Manning B.D.. The PI3K-AKT network at the interface of oncogenic signalling and cancer metabolism. Nat. Rev. Cancer (2020) 20 74-88. PMID: 31686003 14. Fang L.. RNF43 G659fs is an oncogenic colorectal cancer mutation and sensitizes tumor cells to PI3K/mTOR inhibition. Nat. Commun. (2022) 13 3181. PMID: 35676246 15. Xiao Y.. PDGF promotes the warburg effect in pulmonary arterial smooth muscle cells via activation of the PI3K/AKT/mTOR/HIF-1α signaling pathway. Cell. Physiol. Biochem. (2017) 42 1603-1613. PMID: 28738389 16. Xu Z.. Targeting PI3K/AKT/mTOR-mediated autophagy for tumor therapy. Appl. Microbiol. Biotechnol. (2020) 104 575-587. PMID: 31832711 17. Gu X.. Human Schlafen 5 inhibits proliferation and promotes apoptosis in lung adenocarcinoma via the PTEN/PI3K/AKT/mTOR pathway. BioMed Res. Int. (2021) 2021 18. Legchenko E.. PPARγ agonist pioglitazone reverses pulmonary hypertension and prevents right heart failure via fatty acid oxidation. Sci. Transl. Med. (2018) 10 19. Li F.. Activation of PPARγ inhibits HDAC1-mediated pulmonary arterial smooth muscle cell proliferation and its potential mechanisms. Eur. J. Pharmacol. (2017) 814 324-334. PMID: 28867608 20. Li H.H.. Prostanoid EP(4) agonist L-902,688 activates PPARγ and attenuates pulmonary arterial hypertension. Am. J. Physiol. Lung Cell Mol. Physiol. (2018) 314 L349-l359. PMID: 29146573 21. Zhang Y.. JAGGED-NOTCH3 signaling in vascular remodeling in pulmonary arterial hypertension. Sci. Transl. Med. (2022) 14 eabl5471. PMID: 35507674 22. Tseng V., Sutliff R.L., Hart C.M.. Redox biology of peroxisome proliferator-activated receptor-γ in pulmonary hypertension. Antioxidants Redox Signal (2019) 31 874-897 23. Feng W.. ERK/Drp1-dependent mitochondrial fission contributes to HMGB1-induced autophagy in pulmonary arterial hypertension. Cell Prolif. (2021) 54 24. Afzal S.. Renoprotective and haemodynamic effects of adiponectin and peroxisome proliferator-activated receptor agonist, pioglitazone, in renal vasculature of diabetic Spontaneously hypertensive rats. PLoS One (2020) 15 e0229803. PMID: 33170841 25. Wu X.. LY294002 attenuates inflammatory response in endotoxin-induced uveitis by downregulating JAK3 and inactivating the PI3K/Akt signaling. Immunopharmacol. Immunotoxicol. (2022) 44 510-518. PMID: 35344456 26. Xie X.. Pharmacological preconditioning by TERT inhibitor BIBR1532 confers neuronal ischemic tolerance through TERT-mediated transcriptional reprogramming. J. Neurochem. (2021) 159 690-709. PMID: 34532857 27. Savai R.. Pro-proliferative and inflammatory signaling converge on FoxO1 transcription factor in pulmonary hypertension. Nat. Med. (2014) 20 1289-1300. PMID: 25344740 28. Zhang J.. MicroRNA-483 amelioration of experimental pulmonary hypertension. EMBO Mol. Med. (2020) 12 29. Adibkia K.. Silver nanoparticles induce the cardiomyogenic differentiation of bone marrow derived mesenchymal stem cells via telomere length extension. Beilstein J. Nanotechnol. (2021) 12 786-797. PMID: 34395152 30. Blasco M.A.. Telomeres and human disease: ageing, cancer and beyond. Nat. Rev. Genet. (2005) 6 611-622. PMID: 16136653 31. Greider C.W., Blackburn E.H.. Identification of a specific telomere terminal transferase activity in Tetrahymena extracts. Cell (1985) 43 405-413. PMID: 3907856 32. Yuan X., Xu D.. Telomerase reverse transcriptase (TERT) in action: cross-talking with epigenetics. Int. J. Mol. Sci. (2019) 20 33. Nault J.C.. The role of telomeres and telomerase in cirrhosis and liver cancer. Nat. Rev. Gastroenterol. Hepatol. (2019) 16 544-558. PMID: 31253940 34. Ningarhari M.. Telomere length is key to hepatocellular carcinoma diversity and telomerase addiction is an actionable therapeutic target. J. Hepatol. (2021) 74 1155-1166. PMID: 33338512 35. Türck P.. Blueberry extract decreases oxidative stress and improves functional parameters in lungs from rats with pulmonary arterial hypertension. Nutrition (2020) 70 36. Zhao Y.. Antitumor effects of oncolytic adenovirus-carrying siRNA targeting potential oncogene EphA3. PLoS One (2015) 10 e0126726. PMID: 25978371 37. Wazir U.. Correlation of TERT and stem cell markers in the context of human breast cancer. Cancer Genomics Proteomics (2019) 16 121-127. PMID: 30850363 38. Zheng L.. Regulation of c-MYC transcriptional activity by transforming growth factor-beta 1-stimulated clone 22. Cancer Sci. (2018) 109 395-402. PMID: 29224245 39. Yousefnia S.. The influence of peroxisome proliferator-activated receptor γ (PPARγ) ligands on cancer cell tumorigenicity. Gene (2018) 649 14-22. PMID: 29369787 40. Marion-Letellier R., Savoye G., Ghosh S.. Fatty acids, eicosanoids and PPAR gamma. Eur. J. Pharmacol. (2016) 785 44-49. PMID: 26632493
--- title: Enhancement of bioactive compounds and biological activities of Centella asiatica through ultrasound treatment authors: - Eunjeong Seong - Huijin Heo - Heon Sang Jeong - Hana Lee - Junsoo Lee journal: Ultrasonics Sonochemistry year: 2023 pmcid: PMC10015234 doi: 10.1016/j.ultsonch.2023.106353 license: CC BY 4.0 --- # Enhancement of bioactive compounds and biological activities of Centella asiatica through ultrasound treatment ## Graphical abstract ## Highlights •Centella asiatica leaf is an excellent source of numerous phytochemicals.•Ultrasound as a postharvest elicitor increased accumulation of bioactive compounds.•Ultrasound significantly elevated activities of phenolic triggering enzymes.•Centella asiatica can protect myoblast via modulation ROS, GSH, and MDA levels. ## Abstract Centella asiatica possess various health-promoting activities owing to its bioactive compounds such as triterpenes, flavonoids, and vitamins. Ultrasound treatment during the post-harvest process is a good strategy for eliciting secondary metabolite in plants. The present study investigated the effect of ultrasound treatment for different time durations on the bioactive compounds and biological activities of C. asiatica leaves. The leaves were treated with ultrasound for 5, 10, and 20 min. Ultrasound elicitation (especially for 10 min) markedly elevated the accumulation of stress markers, leading to enhanced phenolic-triggering enzyme activities. The accumulation of secondary metabolites and antioxidant activities were also significantly improved compared with that in untreated leaves. In addition, ultrasound-treated C. asiatica leaves protected myoblasts against H2O2-induced oxidative stress by regulating reactive oxygen species production, glutathione depletion, and lipid peroxidation. These findings indicate that elicitation using ultrasound can be a simple method for increasing functional compound production and enhancing biological activities in C. asiatica leaves. ## Introduction Elicitors are classified as biotic (organic acids, microbes, and hormones) and abiotic (hypoxia, salts, temperature, and light) [1]. Increasing evidence show that the application of elicitors enhances the various secondary metabolites compared with the control levels [2]. These elicitors need to be sustainable, cheap, and generally recognized as safe. Ultrasound technology is widely used in the food industry because it is safe, non-toxic, and eco-friendly [3]. The mechanism underlying the action of ultrasound is the formation of a multiple-bubble system and bubble coalescence process, leading to oxidative burst and the finally inducing a defense system in plants [4]. Previous studies have shown that ultrasound enhances genistein, daidzein, and gamma-aminobutyric acid contents in soybean sprouts and increases resveratrol content in peanut sprouts [5], [6]. In addition, ultrasound treatment triggers membrane ion fluxes and rapidly increases tyrosine ammonia-lyase (TAL) and phenylalanine ammonia-lyase (PAL) activities, followed by increase in phenolic compounds and polyphenol concentrations [7], [8]. Centella asiatica (L) Urban is an herb used to prevent or treat diseases in traditional Chinese medicine in Asian countries [9]. C. asiatica contains various bioactive compounds, including phenolic compounds, triterpenes, minerals, and vitamins [10]. Based on numerous studies, the triterpenes, including asiaticoside, madecasosside, asiatic acid, and madecassic acid, are believed to be the major bioactive compounds of C. asiatica [11]. Previous studies have attributed relieving and therapeutic effects to C. asiatica and its bioactive compounds with regard to cardioprotection, wound healing, and neuroprotection [12]. C. asiatica alleviates neurological diseases by reducing inflammatory factors, repairing abnormal expression of mitochondria-related proteins, and balancing oxidative stress [13]. Pittella et al. [ 2009] investigated the total phenolic and flavonoid constituents and antioxidant activities of C. asiatica leaves and established a positive correlation between antioxidant and antitumor activities [14]. The ameliorative effect of C. asiatica on various diseases may be due to its excellent antioxidant potential. Skeletal muscle is an important component of the body and constitutes approximately $50\%$ of the body mass [15]. Sarcopenia is a generalized loss of muscle mass and function that occurs in the absence of underlying diseases and is characterized by aging-associated progression [16]. The damage caused is irreversible and destructive and is instigated by oxidative stress in skeletal muscle cells [17]. Oxidative stress is induced by the homeostatic impairment of reactive oxygen species (ROS) and leads to several pathological conditions including obesity, inflammation, and aging [18]. A previous study has confirmed that asiatic acid exhibits antioxidative and antiapoptotic effects by reducing ROS production [19]. Anand et al. [ 2012] demonstrated that administration of C. asiatica extract suppressed physical fatigue by increasing glycogen stores and antioxidant enzymes and decreasing lipid peroxidation in a rat model [20]. Therefore, utilizing C. asiatica could be a strategy for the protection of skeletal muscle. However, the accumulation of biological compounds due to post-harvest treatment using ultrasound in C. asiatica leaves and their protective effect on oxidative stress in C2C12 myoblasts remain unknown. The present study aimed to determine the effect of ultrasound treatment on the antioxidant capacity, bioactive compounds content, and biological activity of C. asiatica leaves. ## Ultrasound treatment and sample preparation C. asiatica was purchased from a farm (Hapcheon, Korea) and stored in cold room at 4 ℃. Fifty grams of the sample was transferred into an ultrasonic cleaner (WUC-D10H, Daihan Scientific, Korea, 290×240×150 mm, 230v/40 kHz, 400 W) and treated with ultrasound at 200 W and for 5, 10, and 20 min at 25 ℃. After treatment, the ultrasound treated C. asiatica was stored at −80 ℃ until further use. For analysis, 3 g of lyophilized C. asiatica leaves was subjected to extraction using methanol. ## Determination of triterpenes To confirm the quantities of major four triterpenes including madecassoside, asiaticoside, madecassic acid, and asiatic acid, dried C. asiatica leaves were ground into a fine powder. Each leaf (0.1 g) was subjected to extraction using $80\%$ methanol and sonicated for 30 min. The extracts were filtered through a Toyo No. 2 filter paper (Toyo Ltd., Tokyo, Japan), and diluted to 50 mL. The triterpenes were analyzed using HPLC in a Luna C18 column (250×4.6 mm, 5 µm, Phenomenex, Torrance, CA, USA). A fluorescence detector was used at an excitation of wavelength 205 nm. The mobile phase consisted of $0.05\%$ phosphoric acid in water and acetonitrile. Gradient elution was performed under the following conditions: 0–30 min, $20\%$ B; 30–40 min, $100\%$ B; 40–50 min, $20\%$ B. The flow rate was 1.0 mL/min. A UV detector (UV-2075, Jasco) was used at wavelength of 205 nm. ## Determination of total flavonoid and total polyphenol contents and antioxidant activities The total phenolic and total flavonoid contents in C. asiatica leaves were measured using the Folin–Ciocalteu colorimetric method and another colorimetric method based on one described in a previous study, respectively [21]. ABTS and DPPH radical scavenging activities and reducing power were determined according to a previously described method [22]. ## Determination of chlorophyll content Dried sample (0.05 g) was subjected to extraction using $80\%$ dimethyl sulfoxide and incubated at 65 ℃ for 1 h. The extracts were then centrifuged for 5 min at 10,000× g. The supernatant was assayed using a spectrophotometer (BioTek, Inc., Winooski, VT, USA) at wavelengths of 665 and 648 nm. Chl $a = 14.85$ × A665 – 5.14 × A648. Chl $b = 25.48$ × A648 – 7.36 × A665. Total chlorophyll = 7.49 × A665 + 20.34 × A648 (expressed in mg/g dry weight). ## Determination of vitamins Vitamin E and C levels were determined using HPLC according to a previously described methods [23], [24]. Vitamin E was analyzed using normal phase HPLC in a Lichrospher 100 Diol column (250×4.6 mm, 5 μm, Merck, Berlin, Germany) and vitamin C using reverse phase HPLC in a C18 column (250×4.6 mm, 5 μm, Shisheido, Tokyo, Japan), respectively. ## Determination of flavonoids Lyophilized leaves were extracted with methanol containing $10\%$ phosphoric acid ($0.1\%$ [v/v]) and the mixture was centrifuged at 1000× g for 5 min. Rutin, catechin, and naringin were quantified using HPLC in a Luna C18 column (250 mm × 4.6 mm, 5 μm, Phenomenex, USA). Water (A) and $5\%$ acetic acid in methanol (B) were used as the mobile phase. Gradient elution was carried out under the following conditions: 0–10 min, $0\%$–$20\%$ B; 10–20 min, $20\%$-$40\%$ B; 20–30 min, $40\%$–$50\%$ B; 30–40 min, $50\%$–$70\%$ B; and 40–50 min, $70\%$–$100\%$ B. ## Tyrosine ammonia-lyase (TAL) and phenylalanine ammonia-lyase (PAL) assays TAL and PAL activities of the ultrasound-treated C. asiatica leaves were evaluated as previously described [25]. TAL enzymatic activity was measured by confirming the production of p-coumaric acid from the l-tyrosine in the supernatant at 310 nm. PAL enzymatic activity was determined by confirming the production of trans-cinnamic acid from l-phenylalanine in the supernatant at 290 nm. ## Catalase (CAT) and peroxidase (POD) assays Ultrasound-treated C. asiatica leaves were homogenized in 5 mL phosphate buffer (10 mM, pH 7.4). The extracts were centrifuged (14,240 × g for 15 min) and the supernatant was collected. For the CAT assay, 250 μL of the extract, 2.5 mL of phosphate buffer (10 mM, pH 7.4), and 200 μL of water (100 mM) were mixed in a tube. For the POD assay, 100 μL of the extract, 3 mL of phosphate buffer (0.05 M, pH 6.0), 150 mM of guaiacol, and 200 μL of water (100 mM) were mixed in a tube. Detection was performed using a spectrophotometer at 240 nm and 450 nm, respectively for 30 min. The results were expressed as U/g fresh weight (FW). ## Cell culture and cell viability C2C12 cells were obtained from ATCC (CRL-1772, Manassas, VA, USA). C2C12 myoblasts were seeded in 96-well plates at a density of 5.0×104 cells/mL. After 24 h, the cells were treated with untreated- or ultrasound-treated C. asiatica (50 μg/mL). After 2 h, the culture medium was replaced with hydroperoxide (700 μM) and the samples. After 24 h, 20 μL of MTT (5 mg/mL) was added to each well and incubated for 2 h. The supernatant was then removed, and the blue crystal formazan crystals produced in viable cells were dissolved in dimethyl sulfoxide. ## Cellular reactive oxygen species (ROS), glutathione (GSH), and lipid peroxidation ROS production was determined as previously described [26]. C2C12 cells (5.0×104 cells/mL) were seeded in a 96-well black plate. After 24 h, the cells were pre-incubated with the samples for 5 h. The supernatant was then removed, and 10 μM of DCFH-DA with 700 μM hydroperoxide was added to each well at 37 ℃. ROS levels were determined using a fluorescence spectrophotometer (LS-55; Perkin-Elmer, Norwalk, CT, USA). To determine GSH and MDA levels, C2C12 cells were seeded in 6-well plates at a density of 1.5×105 cells/mL. After 24 h, the culture medium was replaced with an FBS-free medium containing extracts. After 4 h, the cells were treated with 700 μM hydroperoxide for 24 h to induce oxidative stress. The cells were then extracted and centrifuged. To measure the GSH levels, 20 µL of the supernatant was added to 180 µL of a mixture containing glutathione reductase, NADPH, and DTNB. The lipid peroxidation level was determined using the thiobarbituric acid reactive substance (TBARS) assay [26]. ## Statistical analysis Data are representative of two or three independent experiments and were analyzed using GraphPad Prism software version 5 (GraphPad Software Inc., La Jolla, CA, USA) and SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). ## Effects of ultrasound treatment on triterpenes accumulation in C. Asiatica leaves Ultrasound treatment can be used as a post-harvest elicitor to increase the amounts of secondary metabolites [27]. It was reported that elicitors can switch the enzymatic responses to abiotic or biotic stresses, leading to the accumulation of secondary metabolites [28]. The major compounds responsible for bioactivity in C. asiatica are two glycosides, madecassoside and asiaticoside, and corresponding two aglycones, madecassic acid and asiatic acid [29]. The present study investigated the changes in the content of the main triterpenes induced by ultrasonic treatment of C. asiatica leaves. Table 1 shows that C. asiatica leaves contains higher levels of the glycoside form than of the aglycone form. The highest concentration of a secondary metabolite in untreated leaves corresponded to asiaticoside (7.46±0.35 mg/g dry weight), followed by madecassoside (5.59±0.07 mg/g dry weight). Ultrasound treatment increased the amounts of madecassoside, asiaticoside, and madecasic acid in a time-dependent manner. Also, total triterpene content was significantly enhanced at 20 min (19.12±0.08 mg/g dry weight) compared with that in the untreated leaves (15.36±0.21 mg/g dry weight). A previous study showed that ultrasound treatment for 20 min significantly stimulated the secretion of oleanolic acid saponins in marigold hairy root [30]. Puttarak and Panichayupakaranant [2012] reported that the total triterpenes content in C. asiatica leaves was 19.5±0.9 mg/g dry weight [31]. Moreover, our results are in conformation with that of a previous study, which demonstrated that the triterpenes accumulated in C. asiatica leaves were in the glycoside form rather than the aglycone form [31]. Taken together, our findings indicate that ultrasound treatment could enhance triterpenes levels in C. asiatica leaves. Table 1Triterpene contents in Centella asiatica. SamplesMadecassosideAsiaticosideMadecassic acidAsiatic acidTotal(mg/g DW1)Raw (untreated)5.59±0.07b7.46±0.35c0.32±0.02c1.99±0.09a15.36±0.21c5 min6.31±0.48b8.40±0.14b0.39±0.02b1.83±0.05ab16.93±0.69b10 min7.40±0.19a9.30±0.32a0.44±0.02b1.74±0.06bc18.88±0.59a20 min7.50±0.57a9.47±0.07a0.51±0.02a1.64±0.01c19.12±0.08aAll values are means of duplicate, and the mean values in a column followed by different superscript letters are significantly ($p \leq 0.05$) different (Duncan's multiple range test).a1 DW = Dry weight. ## Effects of ultrasound treatment on bioactive compounds in C. Asiatica leaves To investigate the effect of ultrasound treatment, we investigated the changes in the phytonutrient components of C. asiatica. We found that the polyphenol, flavonoid, vitamin, and chlorophyll contents of C. asiatica were greatly altered by ultrasound treatment (Table 2). Total phenolic and total flavonoid constituents were significantly higher in leaves with ultrasound treatment for 5, 10, and 20 min than in untreated leaves. The best time duration for ultrasound treatment was determined to be 20 min, which induced the highest increase of total polyphenol (2255.21±33.35 GAE mg/100 g DW) and total flavonoid contents (2219.80±5.39 CE mg/100 g DW). Additionally, ultrasound treatment impacted the concentration of flavonoid content including that of catechin, naringin, and rutin in C. asiatica. The concentration of catechin, naringin, and rutin in ultrasound untreated C. asiatica was 0.92±0.11, 0.04±0.00, and 0.35±0.03 g/100 g DW, respectively. Post-harvest treatment of C. asiatica for 10 min with ultrasound significantly enhanced the levels of catechin (1.24±0.22 g/100 g DW), naringin (0.07±0.00 g/100 g DW), and rutin (0.45±0.08 g/100 g DW). The increase could be attributed to the triggering of shikimic acid–phenylpropanoid metabolism, resulting in the biosynthesis and accumulation of flavonoids and phenolics [32].Table 2Functional compounds in Centella asiatica. Raw (untreated)5 min10 min20 minTotal polyphenol (GAE2 mg/100 g DW1)1992.70±18.33d2089.83±20.25c2192.12±21.16b2255.21±33.35aTotal flavonoid (CE3 mg/100 g DW)1746.56±32.22d1948.26±31.86c2085.98±54.62b2219.80±5.39a FlavonoidsCatechin (g/100 g DW)0.92±0.11a1.00±0.09a1.24±0.22a1.16±0.01aNaringin (g/100 g DW)0.04±0.00b0.05±0.00ab0.07±0.00a0.05±0.01abRutin (g/100 g DW)0.35±0.03a0.39±0.01a0.45±0.08a0.41±0.02a VitaminsVitamin C (mg/100 g DW)44.34±0.43c57.87±7.08b70.43±2.10a59.00±0.27bα-Tocopherol (mg/100 g DW)13.76±1.18a12.60±0.68a13.57±1.12a13.64±1.09aβ-Tocopherol (mg/100 g DW)0.07±0.03a0.05±0.00a0.06±0.00a0.06±0.01aγ-Tocopherol (mg/100 g DW)0.25±0.13a0.23±0.01a0.23±0.03a0.24±0.03aγ-Tocotrienol (mg/100 g DW)0.04±0.01a0.04±0.00a0.05±0.01a0.04±0.01a ChlorophyllsChlorophyll a (mg/g DW)40.62±6.58a46.01±0.02a47.52±0.31a45.45±2.23aChlorophyll b (mg/g DW)8.97±1.03b10.49±0.08ab10.73±0.11a10.19±0.53abAll values are the means of duplicate trials, and the mean values in a row followed by different superscript letters are significantly different ($p \leq 0.05$) (Duncan's multiple range test).a1 DW = Dry weight.b2 GAE = Gallic acid equivalent.c3 CE = Catechin equivalent. Plant-derived vitamins such as ascorbic acid and tocopherol act as antioxidants and provide various health benefits ranging from providing basic nutrition to reducing the risk of cancer and chronic diseases [33]. Chlorophylls being the most abundant pigments in plants have huge antioxidant potential [34]. Vitamin C content was markedly enhanced by treatment with ultrasound for 10 min (70.43±2.10 mg/100 g DW) compared with that in untreated leaves (44.34±0.43 mg/100 g DW). Meanwhile, no significant changes were noticed in the vitamin E content of C. asiatica after ultrasound treatment. The amounts of chlorophyll a and b ranged between 40.62 and 47.52 mg/g DW and 8.97–10.73, respectively. Chlorophyll a content was not significantly altered, however, Chlorophyll b content was significantly increased by ultrasound treatment for 10 min. Our results thus confirm the recent reports demonstrating the capacity of abiotic elicitors to promote the accumulation of functional compounds, such as flavonoids, and vitamins in various plants [25], [35], [36]. Yu et al. [ 2016] reported that ultrasound treatment enhanced resveratrol contents in peanut sprouts compared with that of the control [6], whereas, Yang et al. [ 2015] showed that it increased the daidzein, genistein, and gamma-aminobutyric acid contents in soybean sprouts [5]. The reason for the increase of these bioactive compounds could be that ultrasound as a stress-elicitor may have evoked ROS generation. As a defense response, secondary metabolite levels are increased to eliminate the detrimental effects of oxidative damage [37]. Based on these findings, post-harvest treatment with ultrasonication for 10 min might be a efficient way to increase phytochemical accumulation in C. asiatica. ## Effect of ultrasound treatment on antioxidant capacities in C. Asiatica leaves The antioxidant activity of ultrasound-treated C. asiatica is shown in Fig. 1. Compared with that in the untreated leaves, treatment with ultrasound for 10 min significantly increased the DPPH radical scavenging activity, ABTS radical scavenging activity, and reducing power 1.75-, 1.35-, and 1.20-fold, respectively. Yu et al. [ 2016] reported that ultrasound-treated romaine lettuce was exhibited a significantly higher DPPH antioxidant activity [38]. Our results are similar to those of Gani et al. [ 2016], who found that the DPPH and ABTS radical scavenging activities of ultrasonic-treated strawberries increased with treatment time [39]. However, antioxidant activities were slightly reduced after 20 min of ultrasound treatment than at 10 min. This reduction in antioxidant activity may be related to the decrease in flavonoid and vitamin C contents. Generally, when plants are exposed to elicitation conditions, they may activate secondary metabolism as a defense strategy for self-protection. However, plants have thresholds for the quantity of secondary metabolites that they can synthesize. Plant with accumulation of high levels of secondary metabolites stimulate an increase in the levels of enzyme that degrade those metabolites via feedback modulation [40]. González and Nazareno [2011] showed that vitamin C and flavonoids (naringin) exhibit high antioxidant activity [41]. Flavonoids and vitamins are known to exhibit antioxidant activities. A previous study reported that triterpene enrichment in C. asiatica extract did not improve its antiradical activity [42]. The triterpenes in C. asiatica have various health-promoting effects; however, they may not be directly responsible for their antiradical capacity. Hence, our findings suggest that ultrasound treatment for 10 min would enhance antioxidant activity by increasing the vitamin C and flavonoid contents in C. asiatica. Fig. 1Effect of ultrasound-treated Centella asiatica on 1,1-diphenyl-2-picrylhydrazyl (DPPH) radical scavenging activity, (2,2-azino-bis-3-ethylbenzothiazoline-6-sulfonic acid (ABTS) radical scavenging activity, and reducing power. The DPPH and ABTS radical scavenging activity and reducing power are expressed as Trolox equivalent mg/100 g dry weight. Different letters (a,b for DPPH, a′,b′,c′,d’ for ABTS, and a′′,b′′,c′′ for reducing power) above the bars indicate significant differences according to Duncan's test ($p \leq 0.05$). ## Effect of ultrasound treatment on enzyme activities in C. Asiatica leaves POD and CAT are stress marker enzymes that play important roles in free radical scavenging [43]. PAL and TAL, which are responsible for the activation of the phenylpropanoid pathway for phenolic biosynthesis, are major markers of plant resistance [44]. TAL and PAL catalyze the conversion of L-tyrosine to p-coumaric acid and L-phenylalanine to trans-cinnamic acid, respectively [35]. To the best of our knowledge, no study has yet demonstrated the accumulation of stress markers in C. asiatica due to ultrasound elicitation during the post-harvest processing. We found that, compared with that in the untreated leaves, ultrasound treatment significantly enhanced CAT and POD activity in a time-dependent manner (Fig. 2A and B). The highest CAT and POD activity was observed at 10 min. Similarly, compared with that in the untreated leaves, PAL and TAL activities in C. asiatica leaves were enhanced 1.30- and 1.55-fold, respectively, after ultrasound treatment for 10 min (Fig. 2C and D). Ampofo and Ngadi [2020] found that ultrasound treatment elicited TAL and PAL activities in bean sprouts [7]. According to a previous study, the disruption of plant tissue leads to H2O2 accumulation in the cell walls, resulting in the induction of defense-related enzymes such as TAL and PAL for the increase of phenolic compound biosynthesis [45]. Therefore, our findings suggest that ultrasound application during the post-harvest processing of C. asiatica enhances the demand for CAT and POD to decompose H2O2, protect plant cells from oxidative damage, and increase the activity of PAL and TAL to synthesize flavonoids and phenolics. Fig. 2Effect of ultrasound treated Centella asiatica on (A) catalase (CAT), (B) peroxidase (POD), and (C) phenylalanine ammonia-lyase (PAL), and (D) tyrosine ammonia-lyase (TAL) activities. Data are presented as the the mean ± standard error ($$n = 3$$). Different letters above the bars indicate significant differences at $p \leq 0.05.$ ## Cytoprotective effect of ultrasound–treated C. Asiatica leaves against H2O2-induced oxidative stress in C2C12 myoblasts ROS accumulation is one of the factors that can cause sarcopenia [46]. Antioxidant defense and lipid peroxidation in the body have been suggested as early biomarkers of sarcopenia [47]. In this study, we investigated whether ultrasound-treated C. asiatica protects myoblasts against oxidative stress, and we confirmed the protective effect, ROS generation, GSH levels, and lipid peroxidation levels. Treatment with the extract (50 μg/mL) did not affect the cytotoxicity of myoblasts (Fig. 3A). Treatment with hydroperoxide (700 μM) reduced cell viability by $32.7\%$ (Fig. 3B). However, treatment with C. asiatica increased cell viability by 24.2 (raw), 42.9 (5 min), 57.9 (10 min), and $47.3\%$ (20 min), compared with that of H2O2-treated cells. As shown in Fig. 3C – F, myoblasts treated with H2O2 showed a significant increase in ROS production, GSH depletion, and lipid peroxidation levels compared to that of control cells, whereas treatment with C. asiatica markedly decreased the oxidative stress-induced ROS production and GSH depletion. The lipid peroxidation level was reduced depending on the time duration of ultrasound treatment; however, it increased slightly with longer sonication time (20 min). Our results showed that ultrasound treatment for 10 min was the most effective in inverting the increase in ROS generation, GSH depletion, and MDA levels. These increased bioactivities might be attributed to the enhanced phytochemical composition of C. asiatica leaves after post-harvest treatment with ultrasound. Triterpenes, flavonoids, and vitamins are well-known to exert antioxidant activity. A previous study reported that supplementation with vitamins C and E alleviated oxidative damage and improved muscle function in aged rodents [48]. Another study showed that C. asiatica extract stimulates muscle protein synthesis resulting in the restoration of normal muscle structure and mass [49]. Therefore, these results suggest that ultrasound-treated C. asiatica leaves play a crucial role in the protection of myoblasts when treated with ultrasound as the elicitor after harvest. Fig. 3Effect of ultrasound treated Centella asiatica extracts (50 μg/mL) on (A) cytotoxicity, (B) protective activity, (C and D) reactive oxygen species production, (E) glutathione depletion, and (F) lipid peroxidation against H2O2-induced C2C12 cells. Data are presented as the the mean ± standard error ($$n = 3$$). ## $p \leq 0.01$ and ###$p \leq 0.001$, versus the control cells; $p \leq 0.05$, $p \leq 0.01$, and $p \leq 0.001$ versus the H2O2-treated cells. Con, Control; GSH, glutathione; MDA, malondialdehyde. ## Conclusion In the present study, the changes in the bioactive compounds and biological activities of ultrasound- treated C. asiatica leaves were investigated. Ultrasound treatment improved the accumulation of secondary metabolites and antioxidant activities in C. asiatica leaves. Ultrasound-treated C. asiatica leaves enhanced the protective effect by modulating ROS, GSH, and MDA levels in H2O2-induced C2C12 cells. Thus, ultrasound treatment can be applied as post-harvest process to stimulate the production of functional compounds in may agricultural products. ## Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. ## References 1. Meena K.K., Sorty A.M., Bitla U.M., Choudhary K., Gupta P., Pareek A., Singh D.P., Prabha R., Sahu P.K., Gupta V.K., Singh H.B., Krishanani K.K., Minhas P.S.. Abiotic stress responses and microbe-mediated mitigation in plants: The omics strategies*. Front. Plant Sci.* (2017) 8 172. DOI: 10.3389/fpls.2017.00172 2. Ramakrishna R., Sarkar D., Manduri A., Iyer S.G., Shetty K.. Improving phenolic bioactive-linked anti-hyperglycemic functions of dark germinated barley sprouts (. J. Food Sci. Technol. (2017) 54 3666-3678. DOI: 10.1007/s13197-017-2828-9 3. Kentish S., Feng H.. Applications of power ultrasound in food processing. Annu. Rev. Food Sci. Technol. (2014) 5 263-284. DOI: 10.1146/annurev-food-030212-182537 4. Bhat R., Kamaruddin N.S.B.C., Min-Tze L., Karim A.A.. Sonication improves kasturi lime (Citrus microcarpa) juice quality. Ultrason. Sonochem. (2011) 18 1295-1300. DOI: 10.1016/j.ultsonch.2011.04.002 5. Yang H., Gao J., Yang A., Chen H.. The ultrasound-treated soybean seeds improve edibility and nutritional quality of soybean sprouts. Food Res. Int. (2015) 77 704-710. DOI: 10.1016/j.foodres.2015.01.011 6. Yu M., Liu H., Shi A., Liu L., Wang Q.. Preparation of resveratrol-enriched and poor allergic protein peanut sprout from ultrasound treated peanut seeds. Ultraso. Sonochem. (2016) 28 334-340. DOI: 10.1016/j.ultsonch.2015.08.008 7. Ampofo J.O., Ngadi M.. Ultrasonic assisted phenolic elicitation and antioxidant potential of common bean (. Ultraso. Sonochem. (2020) 64. DOI: 10.1016/j.ultsonch.2020.104974 8. Wu J., Lin L.. Ultrasound-induced stress responses of Panax ginseng cells: enzymatic browning and phenolics production. Biotechnol. Prog. (2002) 18 862-866. DOI: 10.1021/bp0255210 9. Brinkhaus B., Lindner M., Schuppan D., Hahn E.G.. Chemical, pharmacological and clinical profile of the East Asian medical plant. Phytomedicine (2000) 7 427-448. DOI: 10.1016/s0944-7113(00)80065-3 10. Siddiqui B.S., Aslam H., Ali S.T., Khan S., Begum S.. Chemical constituents of. J. Asian Nat. Prod. Res. (2007) 9 407-414. DOI: 10.1080/10286020600782454 11. Bylka W., Znajdek-Awiżeń P., Studzińska-Sroka E., Dańczak-Pazdrowska A., Brzezińska M.. Phytother. Res. (2014) 28 1117-1124. DOI: 10.1002/ptr.5110 12. Kunjumon R., Johnson A.J., Baby S.. Phytomed. Plus. (2022) 2. DOI: 10.1016/j.phyplu.2021.100176 13. Chanana P., Kumar A.. Possible involvement of nitric oxide modulatory mechanisms in the neuroprotective effect of. Phytother. Res. (2016) 30 671-680. DOI: 10.1002/ptr.5582 14. Pittella F., Dutra R.C., Junior D.D., Lopes M.T.P., Barbosa N.R.. Antioxidant and cytotoxic activities of. Int. J. Mol. Sci. (2009) 10 3713-3721. DOI: 10.3390/ijms10093713 15. Mafakheri S., Flörke R.R., Kanngießer S., Hartwig S., Espelage L., De Wendt C., Schönberger T., Hamker N., Lehr S., Chadt A., Al-Hasani H.. AKT and AMP-activated protein kinase regulate TBC1D1 through phosphorylation and its interaction with the cytosolic tail of insulin-regulated aminopeptidase IRAP. J. Biol. Chem. (2018) 293 17853-17862. DOI: 10.1074/jbc.RA118.005040 16. Gomes M.J., Martinez P.F., Pagan L.U., Damatto R.L., Cezar M.D.D.M., Lima A.R.R., Okoshi K., Okoshi M.P.. Skeletal muscle aging: influence of oxidative stress and physical exercise. Oncotarget (2017) 8 20428-20440. DOI: 10.18632/oncotarget.14670 17. Powers S.K., Morton A.B., Ahn B., Smuder A.J.. Redox control of skeletal muscle atrophy. Free Radic. Biol. Med. (2016) 98 208-217. DOI: 10.1016/j.freeradbiomed.2016.02.021 18. Nakanishi T., Tsujii M., Asano T., Iino T., Sudo A.. Protective effect of edaravone against oxidative stress in C2C12 myoblast and impairment of skeletal muscle regeneration exposed to ischemic injury in Ob/ob mice. Front. Physiol. (2020) 10 19. Wang Z.H.. Anti-glycative effects of asiatic acid in human keratinocyte cells. Biomedicine (Taipei) (2014) 4 19. DOI: 10.7603/s40681-014-0019-9 20. Anand T., Kumar G.P., Ilaiyaraja N., Khanum F., Bawa A.S.. Effect of asiaticoside rich extract from. J. Anim. Vet. Adv. (2012) 7 832-841. DOI: 10.3923/ajava.2012.832.841 21. Sung J., Lee J.. Antioxidant and antiproliferative activities of grape seeds from different cultivars. Food Sci. Biotechnol. (2010) 19 321-326. DOI: 10.1007/s10068-010-0046-6 22. Lee H., Lee J.. Anti-diabetic effect of hydroxybenzoic acid derivatives in free fatty acid-induced HepG2 cells via miR-1271/IRS1/PI3K/AKT/FOXO1 pathway. J. Food Biochem. (2021) 45 e13993. PMID: 34730253 23. Gardner P.T., White T.A.C., McPhail D.B., Duthie G.G.. The relative contributions of vitamin C, carotenoids and phenolics to the antioxidant potential of fruit juices. Food Chem. (2000) 68 471-474 24. Choi Y., Jeong H.S., Lee J.. Antioxidant activity of methanolic extracts from some grains consumed in Korea. Food Chem. (2007) 103 130-138. DOI: 10.1016/j.foodchem.2006.08.004 25. Jeong H., Sung J., Yang J., Kim Y., Jeong H.S., Lee J.. Effect of sucrose on the functional composition and antioxidant capacity of buckwheat (. J. Funct. Foods (2018) 43 70-76. DOI: 10.1016/j.jff.2018.01.019 26. Choe H., Lee H., Lee J., Kim Y.. Protective effect of gamma-aminobutyric acid against oxidative stress by inducing phase II enzymes in C2C12 myoblast cells. J. Food Biochem. (2021) 45 e13639. PMID: 33533516 27. Lu C., Ding J., Park H.K., Feng H.. High intensity ultrasound as a physical elicitor affects secondary metabolites and antioxidant capacity of tomato fruits. Food Control. (2020) 113. DOI: 10.1016/j.foodcont.2020.107176 28. Kou R.W., Xia B., Wang Z.J., Li J.N., Yang J.R., Gao Y.Q., Yin X., Gao J.M.. Triterpenoids and meroterpenoids from the edible Ganoderma resinaceum and their potential anti-inflammatory, antioxidant and anti-apoptosis activities. Bioorg. Chem. (2022) 121. DOI: 10.1016/j.bioorg.2022.105689 29. Sun B., Wu L., Wu Y., Zhang C., Qin L., Hayashi M., Kudo M., Gao M., Liu T.. Therapeutic potential of. Front. Pharmacol. (2020) 11 30. Alsoufi A.S.M., Pączkowski C., Długosz M., Szakiel A.. Influence of selected abiotic factors on triterpenoid biosynthesis and saponin secretion in marigold (. Molecules (2019) 24 2907. DOI: 10.3390/molecules24162907 31. Puttarak P., Panichayupakaranant P.. Factors affecting the content of pentacyclic triterpenes in Centella asiatica raw materials. Pharm. Biol. (2012) 50 1508-1512. DOI: 10.3109/13880209.2012.685946 32. Zhao J., Davis L.C., Verpoorte R.. Elicitor signal transduction leading to production of plant secondary metabolites. Biotechnol. Adv. (2005) 23 283-333. DOI: 10.1016/j.biotechadv.2005.01.003 33. Fletcher R.H., Fairfield K.M.. Vitamins for chronic disease prevention in adults: Clinical applications. JAMA. (2002) 287 3127-3129. DOI: 10.1001/jama.287.23.3127 34. Hayes M., Ferruzzi M.G.. Update on the bioavailability and chemopreventative mechanisms of dietary chlorophyll derivatives. Nutr. Res. (2020) 81 19-37. DOI: 10.1016/j.nutres.2020.06.010 35. Sim U., Sung J., Lee H., Heo H., Jeong H.S., Lee J.. Effect of calcium chloride and sucrose on the composition of bioactive compounds and antioxidant activities in buckwheat sprouts. Food Chem. (2020) 312. DOI: 10.1016/j.foodchem.2019.126075 36. Yu J., Lee H., Heo H., Jeong H.S., Sung J., Lee J.. Sucrose-induced abiotic stress improves the phytochemical profiles and bioactivities of mung bean sprouts. Food Chem. (2023) 400. DOI: 10.1016/j.foodchem.2022.134069 37. Banik N., Bhattacharjee S.. Complementation of ROS scavenging secondary metabolites with enzymatic antioxidant defense system augments redox-regulation property under salinity stress in rice. Physiol. Mol. Biol. Plants (2020) 26 1623-1633. DOI: 10.1007/s12298-020-00844-9 38. Yu J., Engeseth N.J., Feng H.. High Intensity Ultrasound as an abiotic elicitor—effects on antioxidant capacity and overall quality of romaine lettuce. Food Bioprocess Technol. (2016) 9 262-273. DOI: 10.1007/s11947-015-1616-7 39. Gani A., Baba W.N., Ahmad M., Shah U., Khan A.A., Wani I.A., Masoodi F.A., Gani A.. Effect of ultrasound treatment on physico-chemical, nutraceutical and microbial quality of strawberry. LWT - Food Sci. Technol. (2016) 66 496-502. DOI: 10.1016/j.lwt.2015.10.067 40. Malik S., Hossein Mirjalili M., Fett-Neto A.G., Mazzafera P., Bonfill M.. Living between two worlds: two-phase culture systems for producing plant secondary metabolites. Crit. Rev. Biotechnol. (2013) 33 1-22. DOI: 10.3109/07388551.2012.659173 41. González E.A., Nazareno M.A.. Antiradical action of flavonoid–ascorbate mixtures. LWT - Food Sci. Technol. (2011) 44 558-564. DOI: 10.1016/j.lwt.2010.09.017 42. Arora R., Kumar R., Agarwal A., Reeta K.H., Gupta Y.K.. Comparison of three different extracts of. Biomed. Pharmacother. (2018) 105 1344-1352. DOI: 10.1016/j.biopha.2018.05.156 43. Buraphaka H., Putalun W.. Stimulation of health-promoting triterpenoids accumulation in. Ind. Crops Prod. (2020) 146. DOI: 10.1016/j.indcrop.2020.112171 44. Zheng F., Chen L., Zhang P., Zhou J., Lu X., Tian W.. Carbohydrate polymers exhibit great potential as effective elicitors in organic agriculture: A review. Carbohydr. Polym. (2020) 230. DOI: 10.1016/j.carbpol.2019.115637 45. Mendoza-Sánchez M., Guevara-González R.G., Castaño-Tostado E., Mercado-Silva E.M., Acosta-Gallegos J.A., Rocha-Guzmán N.E., Reynoso-Camacho R.. Effect of chemical stress on germination of cv Dalia bean (. Food Chem. (2016) 212 128-137. DOI: 10.1016/j.foodchem.2016.05.110 46. Pansarasa O., Bertorelli L., Vecchiet J., Felzani G., Marzatico F.. Age-dependent changes of antioxidant activities and markers of free radical damage in human skeletal muscle. Free Radic. Biol. Med. (1999) 27 617-622. DOI: 10.1016/s0891-5849(99)00108-2 47. Delrieu L., Martin A., Touillaud M., Pérol O., Morelle M., Febvey-Combes O., Freyssenet D., Friedenreich C., Dufresne A., Bachelot T., Heudel P.-E., Trédan O., Crochet H., Bouhamama A., Pilleul F., Pialoux V., Fervers B.. Sarcopenia and serum biomarkers of oxidative stress after a 6-month physical activity intervention in women with metastatic breast cancer: results from the ABLE feasibility trial. Breast Cancer Res. Treat. (2021) 188 601-613. DOI: 10.1007/s10549-021-06238-z 48. Ryan M.J., Dudash H.J., Docherty M., Geronilla K.B., Baker B.A., Haff G.G., Cutlip R.G., Alway S.E.. Vitamin E and C supplementation reduces oxidative stress, improves antioxidant enzymes and positive muscle work in chronically loaded muscles of aged rats. Exp. Gerontol. (2010) 45 882-895. DOI: 10.1016/j.exger.2010.08.002 49. Oyenihi A., Langa S., Mukaratirwa S., Masola B.. Effects of Centella asiatica on skeletal muscle structure and key enzymes of glucose and glycogen metabolism in type 2 diabetic rats. Biomed. Pharmacother. (2019) 112. DOI: 10.1016/j.biopha.2019.108715
--- title: Clinical impact of healthcare-associated respiratory syncytial virus in hospitalized adults authors: - Alexandra Hill-Ricciuti - Edward E. Walsh - William G. Greendyke - Yoonyoung Choi - Angela Barrett - Luis Alba - Angela R. Branche - Ann R. Falsey - Matthew Phillips - Lyn Finelli - Lisa Saiman journal: Infection Control and Hospital Epidemiology year: 2023 pmcid: PMC10015267 doi: 10.1017/ice.2022.128 license: CC BY 4.0 --- # Clinical impact of healthcare-associated respiratory syncytial virus in hospitalized adults ## Body The impact of respiratory syncytial virus (RSV) in adults is increasingly appreciated. RSV can cause significant morbidity in older adults and in those who are immunocompromised or have cardiopulmonary comorbidities. 1,2 The healthcare costs of RSV-related hospitalizations in adults are similar to those of influenza. 3 Although healthcare-associated (HA) influenza in hospitalized adults has been well described and is known to result in adverse clinical outcomes and increased healthcare utilization, 4–6 less is known about HA-RSV. Although several studies have described RSV outbreaks in a variety of healthcare settings caring for adults, 7–14 risk factors and outcomes associated with HA-RSV in nonoutbreak settings have not been well characterized. To address this knowledge gap, we assessed the demographic characteristics, comorbid conditions, and clinical outcomes of adult patients with HA-RSV compared with adult patients hospitalized with community-onset (CO) RSV infections. As a potential surrogate for severity of illness in those with HA-RSV, we explored escalation of respiratory support associated with detection of RSV, and we compared the characteristics and outcomes of patients with HA-RSV who did and did not have escalation of respiratory support. ## Abstract ### Objective: To describe the clinical impact of healthcare-associated (HA) respiratory syncytial virus (RSV) in hospitalized adults. ### Design: Retrospective cohort study within a prospective, population-based, surveillance study of RSV-infected hospitalized adults during 3 respiratory seasons: October 2017–April 2018, October 2018–April 2019, and October 2019–March 2020. ### Setting: The study was conducted in 2 academically affiliated medical centers. ### Patients: Each HA-RSV patient (in whom RSV was detected by PCR test ≥4 days after hospital admission) was matched (age, sex, season) with 2 community-onset (CO) RSV patients (in whom RSV was detected ≤3 days of admission). ### Methods: Risk factors and outcomes were compared among HA-RSV versus CO-RSV patients using conditional logistic regression. Escalation of respiratory support associated with RSV detection (day 0) from day −2 to day +4 was explored among HA-RSV patients. ### Results: In total, 84 HA-RSV patients were matched to 160 CO-RSV patients. In HA-RSV patients, chronic kidney disease was more common, while chronic respiratory conditions and obesity were less common. HA-RSV patients were not more likely to be admitted to an ICU or require mechanical ventilation, but they more often required a higher level of care at discharge compared with CO-RSV patients ($44\%$ vs $14\%$, respectively). Also, $29\%$ of evaluable HA-RSV patients required respiratory support escalation; these patients were older and more likely to have respiratory comorbidities, to have been admitted to intensive care, and to die during hospitalization. ### Conclusions: HA-RSV in adults may be associated with escalation in respiratory support and an increased level of support in living situation at discharge. Infection prevention and control strategies and RSV vaccination of high-risk adults could mitigate the risk of HA-RSV. ## Study design, sites, and participants We designed a retrospective cohort study within a large, prospective, multicenter, multiseason, population-based, active surveillance study of RSV-associated hospitalization in adults 18 years of age and older. 15 As previously described, active surveillance for RSV took place during 3 successive RSV seasons from October 2017 to March 2020. To identify patients with laboratory-confirmed CO-RSV infection, study staff reviewed infection control databases and clinical virology laboratory logs to ascertain the results of PCR tests ordered as the standard of care for patients admitted with acute respiratory infection. To identify missed patients with CO-RSV and patients with HA-RSV infection, when the surveillance seasons ended, the electronic medical record (EMR) was queried for all positive RSV tests in hospitalized adults during the study period. The study sites were academically affiliated hospital systems in northern Manhattan and Rochester, New York. In the current retrospective study, we compared risk factors and outcomes for adult patients with HA-RSV with hospitalized adults with CO-RSV. The institutional review boards of the study sites approved this study with a waiver of informed consent. ## HA-RSV and CO-RSV case definitions and matching criteria HA-RSV was defined as a patient hospitalized for 4 or more calendar days at a study site prior to RSV detection or transferred from an acute-care hospital with a combined length of hospital stay at either the outside hospital or study site of 4 or more contiguous days prior to RSV detection. Patients transferred from outside hospitals with known RSV infection were excluded. Patients with CO-RSV were selected from the previously described cohort of adults hospitalized with CO-RSV. 15 Eligible patients with CO-RSV were ≥18 years of age, had symptoms consistent with acute respiratory illness, and had RSV detected within 3 calendar days of admission. Each patient with HA-RSV was matched to 2 patients with CO-RSV by age (±5 years), sex, and RSV season. If >2 suitable CO-RSV patients were identified for an HA-RSV patient, those closest in age to the HA-RSV patient were selected. ## Viral detection The northern Manhattan study sites used the FilmArray Respiratory Panel (BioFire Diagnostics, Salt Lake City, UT), which detects influenza types A H3, A H1, B; parainfluenza virus types 1–4; RSV; human metapneumovirus; adenovirus, rhinovirus and enterovirus; and human coronavirus types 229E, HKU1, NL63, OC43; as well as Mycoplasma pneumoniae, Bordatella pertussis, and Chlamydophilia pneumoniae. The Rochester study sites used either the FilmArray Respiratory Panel, Simplexa FLU/RSV Duplex (Diasorin Molecular, Cypress, CA) or Cepheid GeneXpert Flu/RSV Duplex (Cepheid, Sunnyvale, CA). ## Data collection and outcomes For HA-RSV patients, the reasons for hospitalization and for respiratory pathogen testing, such as worsening cough, were extracted from healthcare providers’ progress notes. International Classification of Disease, Tenth Revision (ICD-10) discharge codes that were related to RSV (ie, B97.4, J12.1, J20.5, or J21.0) were extracted from the EMR, and cause(s) of death were abstracted from the death certificate or death notes, when applicable. For both HA-RSV and CO-RSV patients, demographic, and clinical characteristics, including comorbid conditions and living situation at admission, were collected. Living situation was classified as living independently at home, living at home with assistance from family members or home health aide or residing in an assisted living facility, or living in a rehabilitation or skilled nursing facility. 16 Patients transferred from acute-care hospitals or who were homeless were excluded from assessment of living situation at admission. For both HA-RSV and CO-RSV patients, outcomes included length of stay after RSV detection, admission to an intensive care unit (ICU) and/or mechanical ventilation initiated in the 4 days following RSV detection, and in-hospital mortality. For those who survived, living situation at discharge and changes in living situation from admission to discharge that reflected the need for increased support (eg, living independently at admission versus discharge to a skilled nursing facility) were determined as previously described. 16 Patients who died, were transferred to another acute-care hospital, or were not eligible for the analysis of living situation at admission were excluded from the analysis of changes in living situation. Readmission within 30 days of discharge was also assessed. ## Escalation of respiratory support To explore escalation of respiratory support associated with HA-RSV, the type of respiratory support modalities (eg, nasal cannula and/or mechanical ventilation) and degree of support (ie, fraction of inspired oxygen [FiO2] and mean airway pressure [MAP] sustained for ≥1 hour each day) were collected before and after detection of RSV (day of detection = day 0). The daily maximum respiratory support used from day −10 (when available) to day −3 prior to RSV detection was considered the baseline support. This support was compared with the daily maximum respiratory support used from day −2 to day +4. This timeframe was selected because it reflected the potential time course of clinical deterioration from RSV prior to and after providers sent the diagnostic test. Patients who were transferred to the study sites from day −2 to day 0 were excluded from this analysis because a baseline period of support could not be reliably established. Escalation of respiratory support was defined as follows: [1] increase in supplemental oxygen by ≥ 1 liter per minute for ≥ 1 hour while maintaining the same mode of noninvasive respiratory support, [2] increase in modality of noninvasive support such as transition from room air to nasal cannula and/or nasal cannula to bilevel positive airway pressure (BiPAP), [3] transition from noninvasive to invasive support such as BiPAP to mechanical ventilation, or [4] increase in invasive support such as increase in MAP and/or FiO2 while mechanically ventilated. ## Statistical analysis Baseline demographic and clinical characteristics and outcomes of patients with HA-RSV versus CO-RSV were compared using conditional logistic regression. Outcomes with $P \leq .10$ in univariate analysis were then assessed in a multivariable logistic regression model after controlling for comorbid conditions that were significantly different between patients with HA-RSV and CO-RSV. Odds ratios and $95\%$ confidence intervals were also calculated. Length of stay following RSV detection in patients with HA-RSV was compared with patients with CO-RSV but was not included in the bivariate or multivariable analysis due to confounding by comorbid conditions, particularly in patients with HA-RSV. The proportion of patients with HA-RSV with and without escalation of respiratory support was determined, and the types and timing of escalations were characterized. Clinical and demographic characteristics of patients with HA-RSV with and without escalation of respiratory support were compared using the χ2 test or the Fisher exact test, as appropriate, for categorical variables and the Mann-Whitney U test or the Student t test as appropriate for continuous variables. Outcomes with $P \leq .10$ in univariate analysis were then assessed in a multivariable logistic regression model. All analyses were conducted in SAS version 9.4 software (SAS Institute, Cary, NC) and $P \leq .05$ was considered statistically significant. ## Characteristics of patients with HA-RSV versus CO-RSV During the study period, 84 patients met the HA-RSV case definition including 33 patients between October 2017 and April 2018, 34 patients between October 2018 and April 2019, and 17 patients between October 2019 and March 2020 (Fig. 1). Patients with HA-RSV were admitted to the study sites for management of a variety of conditions, most commonly cardiac conditions ($23\%$), neurologic or psychiatric events ($19\%$), and infections ($14\%$). Also, 16 ($19\%$) were transferred from another acute-care hospital. Patients with HA-RSV were hospitalized for a median of 12 days (IQR, 7–16 days) prior to detection of RSV. They underwent testing for respiratory pathogens at the study sites due to new onset or worsening respiratory symptoms or physical findings, most commonly cough ($51\%$), fever ($30\%$), shortness of breath ($13\%$), and/or rhonchi ($13\%$). No RSV outbreaks or clusters were identified by the infection prevention and control teams at the sites during the study period. Fig. 1.Epidemiology of 84 HA-RSV versus 160 CO-RSV cases in 3 respiratory viral seasons: October 2017–April 2018, October 2018–April 2019, and October 2019–March 2020. Due to the coronavirus disease 2019 (COVID-19) pandemic, data collection in the third season ceased in March 2020 due to the onset of the COVID-19 pandemic and cessation of testing for non–severe acute respiratory coronavirus virus 2 (SARS-CoV-2) viruses at the study sites. Note. HA, healthcare-associated; RSV respiratory syncytial virus; CO, community onset. Of the 84 patients with HA-RSV, 76 were matched to 2 patients with CO-RSV and 8 were matched to 1 patient because an appropriate second CO-RSV match could not be found, for a total of 160 patients with CO-RSV. The demographic characteristics were similar between the 2 groups (Table 1). Most patients with HA-RSV ($86\%$) and CO-RSV ($87\%$) had ≥1 comorbid condition, most commonly cardiac comorbidities in both groups. Patients with HA-RSV were more likely to have chronic kidney disease (CKD, $31\%$ vs $26\%$; $$P \leq .04$$) and were less likely to have respiratory comorbidities ($31\%$ vs $46\%$; $$P \leq .03$$) and/or obesity ($23\%$ vs $32\%$; $$P \leq .04$$). Table 1.Characteristics of Patients With Healthcare-Associated (HA) Respiratory Syncytial Virus (RSV) Versus Community-Onset (CO) RSV, Univariate AnalysisCharacteristicHA-RSV($$n = 84$$),No. (%) CO-RSV($$n = 160$$),No. (%) P Value Demographic characteristics Sex, male41 [51]79 [50]1.00 Age group 1.0018–49 y (ref)14 [17]23 [14]…50–64 y29 [35]54 [34]1.00≥65 y41 [49]83 [52]1.00 Race White (ref)38 [46]67 [42]…Black/African American17 [20]28 [18].87Asian1 [1]0 [0].99Unknown/Other28 [33]65 [41].46 Ethnicity Hispanic (ref)18 [21]50 [31]…Non-Hispanic41 [49]47 [29].35Unknown25 [30]63 [39].03 Living situation at admission a Living independently (ref)38 [58]89 [59]…Living at home with assistance of family/friends or aide or in assisted living facility19 [29]52 [34].43Skilled nursing or rehabilitation facility9 [14]11 [7].06 Clinical characteristics, no. (%) Comorbid conditions Respiratory b 26 [31]74 [46].03 Cardiac c 51 [61]77 [48].75Chronic kidney disease26 [31]41 [26].04 Immunosuppressive conditions d 28 [33]54 [34].81Neurologic22 [26]28 [18].92Chronic liver disease6 [7]5 [3].11Diabetes22 [26]56 [35].21Obesity19 [23]51 [32].04 No. of comorbid conditions 0 (ref)12 [14]20 [13]…1–241 [49]85 [53].24≥331 [37]55 [34].17Note. Bold P value indicates statistical significance. a Living status at admission excludes patients who were transferred in from other acute-care hospitals ($$n = 17$$), those who were homeless ($$n = 3$$), and those missing living status at admission ($$n = 6$$). b Respiratory conditions included chronic obstructive pulmonary disease, asthma, pulmonary hypertension, and obstructive sleep apnea. c Cardiac conditions included congestive heart failure, coronary artery disease, hypertension, arrythmias, and valvular heart disease. d Immunosuppressive conditions included HIV, cancer, and transplant recipient. ## Outcomes of patients with HA-RSV versus CO-RSV Following RSV detection, the median length of hospitalization for patients with HA-RSV was longer than that for patients with CO-RSV (10 days [IQR, 5–21] for HA-RSV vs 6 days [IQR, 3–10] for CO-RSV; $P \leq .001$). Although not statistically significant, the proportion of patients who died during hospitalization was higher among those with HA-RSV than those with CO-RSV ($15\%$ vs $6\%$; $$P \leq .25$$) (Table 2). Among those who survived to discharge and had an admission living situation available, patients with HA-RSV were more likely to require an increased level of support in their living situation at discharge compared with their living situation at admission than patients with CO-RSV ($42\%$ vs $14\%$; $$P \leq .01$$). In multivariable analysis, after controlling for comorbid conditions, patients with HA-RSV remained more likely to require an increased level of support at discharge compared with patients with CO-RSV (OR, 6.96; $95\%$ CI, 1.39–34.78; $$P \leq .02$$). Table 2.Outcomes of Patients With Healthcare-Associated (HA) Respiratory Syncytial Virus (RSV) Versus Community-Onset (CO) RSV, Univariate AnalysisOutcomesHA-RSV($$n = 84$$),No. (%) CO-RSV($$n = 160$$),No. (%) P ValueAdmission to ICU within 4 d following RSV detection13 [15]31 [20].60Ventilation initiated within 4 d following RSV detection3 [4] a 16 [18].52In-hospital mortality13 [15]9 [6].25 Living situation at discharge b Living independently (ref)18 [26]72 [50]…Living at home with assistance of family/friends or aide or in an assisted living facility20 [28]48 [34].24Skilled nursing or rehabilitation facility30 [44]23 [16] <.001 Changes in living situation from admission to discharge c, d Unchanged (ref)31 [56]122 [86]…Increased level of support24 [44]20 [14].01 Readmission within 30 d14 [17]15 [9].99Note. ICU, intensive care unit. Bold P value indicates statistical significance. a One patient excluded from analysis of escalation of respiratory support. b Excludes patients who died ($$n = 22$$), were homeless at discharge ($$n = 3$$), were transferred to another acute care hospital ($$n = 2$$), and those for whom data were unavailable ($$n = 6$$). c Excludes patients who died ($$n = 22$$), were transferred to study sites but survived to discharge ($$n = 13$$), those who were homeless at admission and/or discharge ($$n = 3$$), were transferred to another acute care hospital ($$n = 2$$), and those for whom living status at admission and/or discharge were unavailable ($$n = 7$$). d No patients had a decreased level of support at discharge. ## Escalation of respiratory support among patients with HA-RSV Overall, 77 evaluable patients ($92\%$) with HA-RSV were included in the analysis of escalation of respiratory support (Table 3). From day −2 to day +4 relative to RSV detection, 55 ($71\%$) did not have an escalation in respiratory support, including 44 who remained on room air. Of the 22 ($29\%$) who had escalation of respiratory support from their baseline, 11 ($50\%$) were changed from room air to nasal cannula, 4 ($18\%$) had an increase in FiO2 on nasal cannula, 2 ($9\%$) were changed from nasal cannula to a non-rebreather mask, 1 ($5\%$) was changed from room air to BiPAP, 2 ($9\%$) had mechanical ventilation initiated (1 from room air and 1 from nasal cannula), and 2 ($9\%$) had an increase in MAP and/or FiO2 on mechanical ventilation. Of 22 escalations, 15 ($68\%$) occurred on day 0 or day +1 (Fig. 2). Table 3.Comparison of Healthcare-Associated (HA) Respiratory Syncytial Virus (RSV) Cases With and Without Respiratory Support Escalation, Univariate AnalysisCharacteristicEscalation($$n = 22$$),No. (%) No Escalation($$n = 55$$),No. (%) P ValueSex, male12 [55]26 [50].72Age, median y (IQR)73 (65–78)62 (51–70) <.001 Age group18–49 y1 [5]12 [22]…50–64 y5 [23]23 [42].65≥65 y16 [73]20 [36].02 Race White (ref)11 [50]24 [44]…Black/African American5 [23]12 [22]1.00Asian0 [0]0 [0]1.00Unknown/Other6 [27]18 [34].77 Ethnicity Hispanic (ref)4 [18]13 [24]…Non-Hispanic14 [63]26 [47].54Unknown4 [18]16 [29]1.00 Living situation at admission a Living independently (ref)9 [53]29 [62]…Living at home with assistance of family/friends or aide4 [24]13 [28]1.00Skilled nursing or rehabilitation facility4 [24]5 [11].24 Comorbid conditions Respiratory11 [50]10 [18].005 Cardiac12 [55]34 [62].61Chronic kidney disease9 [41]14 [25].18Immunosuppressive conditions9 [41]19 [35].61Neurologic6 [27]15 [27]1.00Chronic liver disease1 [5]4 [7]1.00Diabetes6 [27]16 [29].87Obesity5 [23]12 [22]1.00 No. of comorbid conditions 0 (ref)2 [9]9 [16]…1–211 [50]28 [51].70≥39 [41]18 [33].45 Outcomes ICU admission within 4 days following RSV detection8 [36]2 [4] <.001 Median total hospital length of stay, d (IQR)29 (24–32)20 (13–50).47Median length of stay following RSV detection, d (IQR)15 (8–22)8 (4–21).03 In-hospital mortality6 [27]4 [7].03 Readmission within 30 d5 [23]8 [15].50 Living situation at discharge b Living independently (ref)5 ($18\%$)13 ($27\%$)…Living at home with assistance of family/friends or aide4 [24]16 [33]1.00Skilled nursing or rehabilitation facility9 [59]19 [40].49Note. IQR, interquartile range, ICU, intensive care unit. Bold P value indicates statistical significance. a Living situation on admission excludes patients who were transferred to study sites ($$n = 11$$), homeless ($$n = 1$$), and those for whom data were unavailable ($$n = 1$$). b Living situation at discharge excludes patients who died during admission ($$n = 10$$), those transferred to another other acute care hospital ($$n = 1$$), those homeless ($$n = 1$$), and those for whom data were unavailable ($$n = 1$$). Fig. 2.Timing of respiratory support escalation relative to HA-RSV detection date. During the interval day +2 to day −4, the number of HA-RSV cases with escalation of respiratory support (first day of escalation) is shown. Day 0 is the day of detection of RSV. Note. HA, healthcare-associated; RSV respiratory syncytial virus. ## HA-RSV with and without escalation of respiratory support Patients with HA-RSV who had an escalation of respiratory support were significantly older than those without escalation (median age, 73 years [IQR, 65–78] vs 62 years [IQR, 51–70]; $P \leq .001$) and were more likely to have respiratory comorbidities ($50\%$ vs $18\%$; $$P \leq .005$$) (Table 3). Patients with HA-RSV with escalations were more likely to have been admitted to an ICU ($27\%$ vs $7\%$; $P \leq .001$), to have had a longer length of stay following RSV detection (median, 15 days [IQR, 8–22] vs 8 days [IQR, 4–21]; $$P \leq .03$$), and to have died during hospitalization ($36\%$ vs $4\%$; $$P \leq .03$$). In multivariable analysis, after controlling for age and respiratory conditions, ICU admission ($$P \leq .03$$) and mortality ($$P \leq .002$$) remained associated with escalation of respiratory support, but length of stay after RSV detection was no longer significantly different in those with and without escalation ($$P \leq .48$$). ## Causes of death and analysis of discharge codes for HA-RSV Of the 79 ($94\%$) of 84 patients with HA-RSV with ICD-10 discharge codes available, 14 ($18\%$) had an RSV-related diagnostic code. Only 1 of the 17 patients with an escalation of respiratory support had ICD-10 discharge codes available. Of the 13 patients with HA-RSV who died during admission, none had RSV listed as a cause or contributor to mortality on their death certificates. ## Discussion In this study, we assessed factors associated with HA-RSV in hospitalized adults in nonoutbreak settings, and we noted some interesting differences in patterns of comorbid conditions when we compared our study to similar studies conducted in HA versus CO influenza. Patients with HA influenza, compared with patients with CO influenza, had more chronic medical conditions, including higher rates of chronic lung disease other than asthma, cardiovascular diseases, metabolic disease, and immunosuppressive conditions. 17 In contrast, patients with HA-RSV were less likely than patients with CO-RSV to have respiratory comorbidities. Respiratory comorbidities are well known to be exacerbated by RSV infection, often leading to subsequent hospitalization. 18–21 In the large prospective surveillance study from which the matched patients with CO-RSV were derived, we found that persons with chronic obstructive pulmonary disease and asthma had higher hospitalization rates due to CO-RSV infection than those without these conditions. 15 In the current study, we noted that those with CO-RSV were more likely to be obese; however, previous studies have not identified obesity as a risk factor for severe RSV, including our previous finding that people with and without obesity had similar hospitalization rates for CO-RSV. 15 The lack of association of obesity with severe RSV contrasts with the association of obesity with severe influenza and COVID-19. 22 Finally, in the current study, patients with HA-RSV were more likely have CKD compared to patients with CO-RSV. Similarly, other studies have reported that patients with HA-influenza had significantly higher rates of renal disease compared to those with CO influenza. 17,23 In a post hoc analysis, we found that the LOS for patients with HA-RSV and CO-RSV who had CKD was significantly longer than those without CKD: median, 10 days (IQR, 6–25) versus 8 days (IQR, 3–14), respectively ($$P \leq .004$$). Patients with CKD may have had an increased opportunity for exposure to respiratory viruses while hospitalized. Although most patients with HA-RSV did not have decompensation in their respiratory status, $29\%$ of patients with HA-RSV had an escalation of respiratory support from day −2 to day +4 from RSV detection, $15\%$ required transfer to the ICU, and $4\%$ had initiation of mechanical ventilation temporally associated with RSV detection. Those who had an escalation of respiratory support were older and more likely to have respiratory comorbidities, further underscoring the impact of RSV on older adults and those with chronic respiratory conditions. Those who had an escalation were also more likely to have severe outcomes, including ICU admissions and in-hospital mortality, although we were unable to determine whether these outcomes were attributed to RSV or to other underlying medical conditions. Our findings contribute to an increased understanding of the impact of RSV in hospitalized adults. A substantial proportion of the patients with HA-RSV were frail on admission, evidenced by the finding that $14\%$ lived in skilled nursing facilities prior to hospitalization. Furthermore, patients with HA-RSV were hospitalized for a median of 12 days prior to detection of RSV, suggesting that their admitting diagnoses and comorbid conditions required prolonged hospitalizations. Although ICU admission and mechanical ventilation after RSV detection were similar among patients with HA-RSV and CO-RSV, patients with HA-RSV had overall longer lengths of hospital stay following RSV detection, had a higher proportion of deaths, and were frailer at discharge. A higher proportion of patients in the HA-RSV group required admission to a skilled nursing facility. A combination of underlying conditions, HA-RSV infection, and deconditioning or loss of functional status associated with prolonged hospitalization in older adults could have contributed to these findings, 24 and they highlight the high level of healthcare utilization associated with HA-RSV and the need to prevent HA-RSV in hospitalized adults. Our data suggest that using ICD-10 codes or death certificates would underestimate the burden of HA-RSV in adults because few of the patients with HA-RSV had RSV-related ICD-10 codes and none had RSV noted on their death certificate. This finding could reflect the relative lack of clinical impact of HA-RSV and/or the timing of death relative to detection of RSV infection, but it may also reflect an underappreciation of the potential impact of HA-RSV by providers and diagnostic coders. Similarly, in a study of adults admitted with respiratory illness and RSV detected by PCR, only $51\%$ of patients had an ICD-10 discharge code corresponding to RSV infection. 25 Others have suggested that the use of death certificates alone may significantly underestimate mortality associated with RSV infection. 26 This study had several limitations. Given the relatively small sample size, the study may have been underpowered to detect differences in some outcomes between patients with HA-RSV versus CO-RSV. The study was also conducted in 2 academic centers; thus, these findings may not be generalizable to other settings. Although EMRs were queried for all RSV-positive patients during the study seasons, patients with HA-RSV were likely underestimated because testing for RSV in hospitalized patients with new onset respiratory symptoms was not systematic. Furthermore, patients were not tested and documented to be negative at admission; thus, CO-RSV may have been misclassified as HA because of prolonged detection from an illness prior to admission. Another potential limitation was that criteria for escalation of respiratory support were not standardized but were implemented at the discretion of providers. Finally, clinical outcomes cannot definitively be attributed to RSV versus other underlying medical conditions. In conclusion, this study provides a unique perspective on the impact of HA-RSV in hospitalized adults. Healthcare-associated respiratory viruses are likely underappreciated in hospitalized adults. We found that HA-RSV was associated with escalation of respiratory support and an increased level of support in patients’ living situation at discharge. These outcomes increase the use of healthcare resources and related costs. Although this study cannot determine the mortality rate associated with HA-RSV, available data, such as death certificates and ICD-10 codes, likely underestimate this outcome. Infection control and prevention and RSV vaccines for adults at high risk of HA-RSV could mitigate this risk. ## Financial support L.S., E.E.W., A.R.B. and A.R.F. received grant support, paid to their institutions, from Merck Sharp & Dohme, a subsidiary of Merck & Company, Kenilworth, NJ. ## Conflicts of interest Y.C., M.P., and LF. are employees of Merck Sharp & Dohme Corp., a subsidiary of Merck & Company, Kenilworth, NJ, and may own stock. ## References 1. Falsey AR, Hennessey PA, Formica MA, Cox C, Walsh EE.. Respiratory syncytial virus infection in elderly and high-risk adults. N Engl J Med (2005.0) 352 1749-1759. PMID: 15858184 2. Ackerson B, Tseng HF, Sy LS. Severe morbidity and mortality associated with respiratory syncytial virus versus influenza infection in hospitalized older adults. Clin Infect Dis (2019.0) 69 197-203. PMID: 30452608 3. Ackerson B, An J, Sy LS, Solano Z, Slezak J, Tseng HF.. Cost of hospitalization associated with respiratory syncytial virus infection versus influenza infection in hospitalized older adults. J Infect Dis (2020.0) 222 962-966. PMID: 32300806 4. Parkash N, Beckingham W, Andersson P, Kelly P, Senanayake S, Coatsworth N.. Hospital-acquired influenza in an Australian tertiary centre 2017: a surveillance- based study. BMC Pulm Med (2019.0) 19 79. PMID: 30991976 5. Fullana Barcelo MI, Asensio Rodriguez J, Artigues Serra F. Epidemiological and clinical characteristics of community-acquired and nosocomial influenza cases and risk factors associated with complications: a four season analysis of all adult patients admitted in a tertiary hospital. Influenza Other Respir Viruses (2021.0) 15 352-360. PMID: 33125178 6. Sendi P, Drager S, Batzer B, Walser S, Dangel M, Widmer AF.. The financial burden of an influenza outbreak in a small rehabilitation centre. Influenza Other Respir Viruses (2020.0) 14 72-76. PMID: 31651074 7. Contributing and terminating factors of a large outbreak in an adult hematology and transplant unit. PLoS Curr (2014.0) 6 8. Jensen TO, Stelzer-Braid S, Willenborg C. Outbreak of respiratory syncytial virus (RSV) infection in immunocompromised adults on a hematology ward. J Med Virol (2016.0) 88 1827-1831. PMID: 26990584 9. Kelly SG, Metzger K, Bolon MK. Respiratory syncytial virus outbreak on an adult stem cell transplant unit. Am J Infect Control (2016.0) 44 1022-1026. PMID: 27430734 10. Hababou Y, Taleb A, Recoing A. Molecular investigation of an RSV outbreak in a geriatric hospital. BMC Geriatr (2021.0) 21 120. PMID: 33579210 11. Kestler M, Munoz P, Mateos M, Adrados D, Bouza E.. Respiratory syncytial virus burden among adults during flu season an underestimated pathology. J Hosp Infect (2018.0) 100 463-468. PMID: 29614245 12. French CE, McKenzie BC, Coope C. Risk of nosocomial respiratory syncytial virus infection and effectiveness of control measures to prevent transmission events: a systematic review. Influenza Other Resp Viruses (2016.0) 10 268-290 13. Chu HY, Englund JA, Podczervinski S. Nosocomial transmission of respiratory syncytial virus in an outpatient cancer center. Bio Blood Marrow Transpl (2014.0) 20 844-851 14. Haber N.. Respiratory syncytial virus infections in elderly adults. Med Mal Infect (2018.0) 48 377-382. PMID: 29548714 15. Branche AR, Saiman L, Walsh EE 16. Goldman CR, Sieling WD, Alba LR. (2021.0) 17. Jung MA, D’Mello T, Perez A. Hospital-onset influenza infections. Am J Infect Control (2014.0) 442 7-11 18. Falsey AR, Formica MA, Hennessey PA, Criddle MM, Sullender WM, Walsh EE.. Detection of respiratory syncytial virus in adults with chronic obstructive pulmonary disease. Am J Respir Crit Care Med (2006.0) 173 639-643. PMID: 16387798 19. Kurai D, Saraya T, Ishii H, Takizawa H.. Virus-induced exacerbations in asthma and COPD. Front Microbiol (2013.0) 4 293. PMID: 24098299 20. Lee N, Lui GC, Wong KT. High morbidity and mortality in adults hospitalized for respiratory syncytial virus infections. Clin Infect Dis (2013.0) 57 1069-1077. PMID: 23876395 21. Falsey AR, Walsh EE, Esser MT, Shoemaker K, Yu L, Griffin MP.. Respiratory syncytial virus-associated illness in adults with advanced chronic obstructive pulmonary disease and/or congestive heart failure. J Med Virol (2019.0) 91 65-71. PMID: 30132922 22. Zhao X, Gang X, He G. Obesity increases the severity and mortality of influenza and COVID-19: a systematic review and meta-analysis. Front Endocrinol (2020.0) 11 595109 23. Godoy P, Torner N, Soldevila N. Hospital-acquired influenza infections detected by a surveillance system over six seasons, from 2010/2011 to 2015/2016. BMC Infect Dis (2020.0) 20 80. PMID: 31992207 24. Boyd CM, Landefeld CS, Counsell SR. Recovery of activities of daily living in older adults after hospitalization for acute medical illness. J Am Geriatr Soc (2008.0) 56 2171-2179. PMID: 19093915 25. Datta S, Walsh EE, Peterson DR, Falsey AR.. Can analysis of routine viral testing provide accurate estimates of respiratory syncytial virus disease burden in adults?. J Infect Dis (2017.0) 215 1706-1710. PMID: 28863444 26. Prill MM, Langley GE, Winn A, Gerber SI.. Respiratory syncytial virus-associated deaths in the United States according to death certificate data, 2005 to 2016. Health Sci Rep (2021.0) 4 e428. PMID: 34754948
--- title: Poverty, price and preference barriers to improving diets in sub-Saharan Africa authors: - Derek D. Headey - Olivier Ecker - Andrew R. Comstock - Marie T. Ruel journal: Global Food Security year: 2023 pmcid: PMC10015269 doi: 10.1016/j.gfs.2022.100664 license: CC BY 4.0 --- # Poverty, price and preference barriers to improving diets in sub-Saharan Africa ## Abstract Suboptimal diets are the most important preventable risk factor for the global burden of non-communicable diseases. The EAT-Lancet reference diet was therefore developed as a benchmark for gauging divergence from healthy eating standards. However, no previous research has comprehensively explored how and why this divergence exists in poorer countries undergoing nutrition transitions. This study therefore analyzes dietary patterns and drivers of the demand for nutritious foods using nationally representative household surveys from Ethiopia, Kenya, Tanzania, and Uganda. We show how barriers to dietary convergence stem from combinations of poverty, high relative food prices and weak preferences for some specific healthy foods. The article concludes by discussing interventions for strengthening consumer demand for healthy diets in Africa. ## Highlights •Household data are used to compare diet patterns in Ethiopia, Kenya, Tanzania, and Uganda to the EAT-Lancet reference diet.•Consumption gaps for fruits, vegetables, and pulses and nuts/seeds are large for all consumer groups.•Food budgets of most households in these countries are far too low to afford the EAT-diet.•Consumer preferences for vegetables are weak, but preferences for fruits and animal-sourced foods are strong.•Income growth alone will not solve diet problems in sub-Saharan Africa; both supply and demand side interventions are needed. ## Introduction Globally, suboptimal diets are one of the most important preventable risk factors for non-communicable diseases (NCDs), accounting for $22\%$ of all deaths and $15\%$ of disability-adjusted life years among adults (Afshin et al., 2019). The dimensions of healthy diets are complex and evidence on the health benefits and costs of consuming specific foods or food groups is imperfect given the high degree of dependence on observational evidence. Even so, there is mounting evidence that some foods and food components significantly elevate the risks of NCDs and associated mortality and morbidity (such as processed red meat, saturated fat, salt, and sugar) while others are protective (such as whole grains, vegetables, fruits, pulses, nuts/seeds and fish). The influential report of the EAT-Lancet Commission on healthy diets from sustainable food systems showed that people in most regions of the world overconsume unhealthy foods (e.g. ultra-processed foods rich in sugars, fats, salt and other unhealthy ingredients) and underconsume nutritious foods (e.g., fruits and vegetables), and that a transformation to healthy diets by 2050 will require reducing the global consumption of unhealthy foods by more than $50\%$ and increasing the global consumption of nutritious foods by more than $100\%$ (Willett et al., 2019). While the expected challenges of moving consumers towards consumption of healthy diets are immense, so too are the benefits of preventing early mortality and morbidity from diet-related NCDs and improving the quality of life for millions of people, in addition to achieving substantial environmental benefits (Willett et al., 2019). Despite these potential benefits, the economic and behavioral challenges of achieving dietary change are still poorly understood, especially in low- and middle-income countries (LMICs). For many households in LMICs, there are likely binding demand-side constraints to increasing consumption of nutritious foods and achieving healthy diets. Incomes and expenditures for many households in LMICs may be lower than the cost of a healthy diet (Headey and Alderman, 2019; Hirvonen et al., 2020). However, for the growing middle-classes, food choices are increasingly driven by time constraints and need for convenience, taste, and social status considerations (Baker and Friel, 2016). LMICs also face significant supply-side constraints stemming from low productivity in the value chains for nutritious foods, especially perishable fresh foods. The union of demand and supply conditions determine the relative prices of different foods, which in turn affect food choices and aggregate costs of nutritionally balanced diets, as well as national food policies and the investment choices of agri-food system actors. Aggressive marketing of ultra-processed foods, for example, plays an important role in shifting consumer preferences (Baker and Friel, 2016). In this article we compared household food consumption in four East African countries to the healthy reference diet proposed by the EAT-Lancet Commission and analyzed the drivers of consumer demand for key food groups in these countries (Willett et al., 2019). We analyzed nationally representative survey data from Ethiopia, Kenya, Tanzania, and Uganda—four countries whose economies and food systems are undergoing rapid transformation. We first show that average East African diets are poorly balanced across major food groups, lacking sufficient amounts of nutritious foods in particular. We then explore three factors that might explain consumption gaps for nutritious foods: poverty (low household food budgets relative to the total cost of the EAT-Lancet reference diet), prices (particularly the high prices of nutritious foods), and preferences (identified by low income elasticities for some nutritious foods). We conclude the article by discussing potential interventions to strengthen consumer demand for nutritious foods and healthy diets in LMICs. ## Household survey samples and variable construction The analysis of dietary patterns and drives of the demand for a healthy diet in East Africa uses data from recent household expenditure and consumption surveys from Ethiopia, Kenya, Tanzania, and Uganda. All surveys were conducted between 2014 and 2017 and are representative at the national level and by rural and urban areas. They are the Ethiopia Socioeconomic Survey 2015-16, the Kenya Integrated Household Budget Survey 2015-16, the Tanzania National Panel Survey 2014-15, and the Uganda National Household Survey 2016-17. In all surveys, a 7-day food consumption recall was applied to collect item-level data on food quantities consumed at home, expenditures for purchased foods, and interviewee-estimated “shadow expenditures” (based on sale values) for own-produced and gifted foods. The main variables used in the analysis were constructed from these food consumption modules: Reported food item consumption quantities were summed up to obtain consumption quantities by food group, after converting non-metric quantity units to metric ones in the cases of Ethiopia and Uganda. They were also converted into item-level calorie consumption amounts (using calorie conversion factors and edible portion coefficients from the USDA National Nutrient Database (USDA, 2020) and the Tanzania Food Composition Table for some East Africa-specific food items) and summed up to obtain calorie consumption amounts by food group and in total. Following the same aggregation method, total food expenditures and expenditures by food group were derived from the reported expenditures for purchased items and the estimated value of own-produced and gifted foods. Itemized food prices were approximated from the reported expenditures and consumption quantities only for purchased foods and separately for rural and urban areas of each country. The approximation procedure uses stepwise median calculations at ascending spatial aggregation levels. Food item unit values were averaged at the lowest administrative unit level, if there were at least 10 valid observations for the purchased item (having values larger than 1.5 interquartile range below the 25th percentile value and smaller than 1.5 interquartile range above the 75th percentile value), and if not, the calculation step was repeated at the next higher administrative unit level. The lowest level considered was the “division” (or “county”), followed by the “district” (or “zone”) and then “region.” Household-specific food group prices were derived as the means of the unit values of all consumed food items within the same food group weighted by the household-specific consumption quantity shares of the food items in that food group. Then, a household-specific food price index (used in the food demand system model estimations) was constructed as the mean of food group prices weighted by the consumption quantity shares of the food groups in total food. Food expenditures and various non-food expenditures, which were reported across different survey modules, were summed up to obtain total household expenditures. Following the method by Deaton and Zaidi to construct consumption aggregates (Deaton and Zaidi, 2002), lumpy infrequent expenditures and tax/levy expenditures were excluded, and “user-cost” rates were applied to obtain the present values for durable goods. The user-cost rate for each durable good reported by a household was calculated based on the good's reported purchase value multiplied by the five-year average of the prevailing interest rate in each country leading up to the time of the survey. Total household expenditure is used as proxy for disposable household income. Given different household demographics and individual dietary needs, household-level food consumption quantities and calorie consumption amounts were converted to per adult equivalent (AE) to be able to consistently summarize results across households and compare estimates across samples and with the reference intakes of the EAT-Lancet reference diet. An AE expresses an individual household member as a fraction of an adult person—here, in terms of calories. The reference person is an average adult with daily dietary energy requirements of 2500 kcal. This corresponds to the reference intake level of the EAT-Lancet reference diet for total food. Household-specific AE values were calculated from detailed dietary energy requirements for individuals as provided in the FAO/WHO/UNU report (FAO, WHO, & UNU, 2001). These calculations account for household compositions by sex and age and the dietary energy needs of breastfeeding mothers. In the final samples used in the analysis, the average household member corresponds to about 0.88 AE in Tanzania, 0.89 AE in Uganda, 0.90 AE in Ethiopia, and 0.95 AE in Kenya. Finally, data cleaning was done for consistency across all four surveys, and the procedure included three steps. First, rare, obvious reporting errors in the main variables of the analysis were corrected observation-by-observation. Most of these related to the reported units, such as in the case of implausibly small food quantities that were recorded in kilograms instead of grams, and age of infants recorded in years instead of months. In the second and third steps, entire households were dropped from the samples. Households that did not complete the survey interview, did not report consumption of food at home, or have implausible calorie consumption amounts were dropped. Households were defined as having implausible calorie consumption amounts, if their consumption per AE was below 600 kcal/day or above 6000 kcal/day. Next, for each of the 15 food groups used in the food demand system model estimations, consumption quantities and expenditures were tabulated, and households with implausibly large consumption quantities or expenditures were dropped from the samples. The final samples used in the analysis included 3249 rural households and 1107 urban households in Ethiopia, 12,318 rural households and 7894 urban households in Kenya, 1871 rural households and 1244 urban households in Tanzania, and 9429 rural households and 4337 urban households in Uganda. ## Food demand system model and elasticity calculation The analysis of the demand drivers for a healthy diet uses complete food demand system models to econometrically estimate parameters that are then used to derive income elasticities of demand for 15 distinct food groups. The model estimations were performed separately by rural and urban areas of each country to allow for structurally different food demand curves. They included two separate estimation stages. In the first stage, a Working-Leser model (Leser, 1963; Working, 1943) was estimated to derive the income elasticities of total food demand vis-à-vis the aggregate demand for nonfood consumption. This model is conducive to this analysis because it does not require prices for nonfood expenditures that are mostly unobserved in the household surveys used. However, the two-stage approach relies on the assumption of separability between food and nonfood consumption. It is thus assumed that a household first decides on the allocation of the total budget to food and nonfood expenditures and then allocates the food budget to the individual food groups. Within-food budget allocations were then modeled in the second stage, where full substitutability between all food groups, conditional on the available food budget, is allowed. To estimate food group demand, a quadratic almost ideal demand system (QUAIDS) was used (Banks et al., 1997). The quadratic version was preferred over the more commonly used linear-approximated AIDS (Deaton and Muellbauer, 1980) to allow for the flexibility of a rank-three demand system, which has been shown to be empirically necessary (Buse, 1994; Lewbel, 1991). The standard QUAIDS model specification was augmented to account for censored observations in the dependent variables (food group budget shares of total food expenditure). The two-step procedure used was originally proposed by Shonkwiler and Yen [1999] and later implemented in a QUAIDS framework by Ecker and Qaim [2011]. Censoring occurred in the survey data for a considerable number of observations, because households did not consume all 15 food groups during the recall period (of 7 days) but are assumed to do so over a longer observation period. Ignoring censored dependent variables in demand system estimations yields biased parameter estimates. In conformity with the Working-Leser model specification, the standard QUAIDS model is also augmented to control for household economies of scale in food consumption using a linear translation through the intercept. Assuming weak separability in consumer preferences and low variability of food group prices with income levels, unconditional elasticities were calculated by adding up the conditional elasticities over the two budgeting stages for each household, as suggested by Edgerton [1997]. The household-specific income elasticities were averaged at the rural and urban population means in each country, after dropping extreme estimates. As a measure of accuracy of the mean elasticities, standard errors are calculated using a bootstrap estimator. The estimator accounts for clustering at the village/town or city quarter level to account for the correlation among households living in the same neighborhood (Attanasio et al., 2013). The bootstrapping method was performed on the household-specific, conditional income elasticities derived directly from the QUAIDS estimations. The unconditional, mean income elasticities were cleaned for extreme estimates (cutting off the distributions at 1.5 interquartile ranges below the 25th percentile value or above the 75th percentile value). ## Household consumption patterns and healthy reference intakes Fig. 1 shows average food consumption amounts per adult equivalent (AE) by the major food groups of the EAT-Lancet reference diet, expressed on the basis of calories, in the four East African countries and respective reference intakes of the diet. The EAT-Lancet reference diet consists of a balanced mix of 21 plant-based and animal-source food categories that belong to 7 major food groups. However, we further split the “protein foods” group into those from animal (meat, eggs, fish) and plant sources (legumes, nuts), producing nine food groups. The EAT-Lancet reference diet defines reference intake quantities (with possible intake ranges) for each food category (except for a maximum intake level for added sugars) and provides caloric reference intakes corresponding to these quantities. Reference intakes are expressed for a total caloric intake of about 2500 kilocalories (kcal) per day (which is the daily dietary energy requirement of an average adult), with an allowance of only one third of total calories being derived from starchy staples. While the reference intakes should not be interpreted as strict caloric thresholds that must be achieved, they provide useful benchmarks for a diet that yields sufficient calories from a diverse set of food groups that are likely to also provide adequate amounts of essential macro- and micronutrients for most people. For comparability with the reference intakes, food consumption amounts are scaled to one adult equivalent (AE).Fig. 1Food consumption calories (kcal/day) per adult-equivalent by major food groups in four sub-Saharan African countries relative to the healthy reference intakes of the EAT-Lancet reference dietNote: One adult equivalent (AE) corresponds to an average adult with a dietary energy requirement of 2500 kcal/day. Consumption estimates refer to foods consumed at home only. Starchy staples include cereals, starchy roots/tubers, and plantains. Discretionary foods include snacks, sweets, and beverages and are considered as non-required foods according to the EAT-Lancet reference diet. Fig. 1Source: Authors' estimates from nationally representative surveys, in 2014–2017. The average diet in these four LMICs has several stark different to the EAT-Lancet reference diet. First, it is likely that many people consume too many calories overall. Average total calorie consumption amounts per AE vary only slightly around 2500 kcal per day (Fig. 1), but these estimates exclude food consumed away from home (that amounts to an estimated $4\%$–$13\%$ of total food expenditures, on average). While inadequate calorie intake still affects some people in these surveys, the overconsumption of energy among a sizable proportion of households is plausible given that all four countries report high prevalence of overweight or obesity among adult women, with national rates varying between $8\%$ and $33\%$ in 2014–2016 (ICF International, 2020). Second, the average diet in these countries is starkly lacking in diversity, with an estimated $59\%$–$77\%$ of total calories stemming from starchy staples and mostly from refined grains whose excess consumption is associated with increased risk of weight gain, metabolic abnormalities, and cardiovascular diseases (see Box 1) (Willett et al., 2019). This dependence on starchy staples means that consumption gaps for healthy food groups are immense, although the size of these gaps varies by household location and economic status, as shown in Fig. 2. Across rural and urban areas of the four countries, the largest differences in the distance to the reference intakes between lower and higher income quintiles exist for the two animal-source food groups (meat/fish/eggs and dairy foods) and fruits. For these food groups, the gaps often exceed $50\%$, especially among the lowest and middle-income quintiles. However, even the highest quintiles fall short of consuming the reference intakes, except for dairy foods in rural and urban Kenya and meat/fish/eggs in urban areas of Kenya and Tanzania. Box 1Consumption of whole grains and traditional cerealsThe EAT-Lancet Commission emphasizes the consumption of whole grains as a central component of a healthy diet. High intake of whole grains has been associated with reduced risk of coronary disease, type 2 diabetes, and overall mortality (Willett et al., 2019). Meanwhile, refined grains have been removed of the grains’ bran, resulting in major loss of micronutrients and fiber. The Commission recommends that carbohydrates should be obtained primarily from whole grains with low intake of refined grains. The EAT-Lancet reference diet suggests an intake level of 232 g/day for an average adult. This is considerably larger than the optimal level of intake (125 g/day) identified in the Global Burden of Disease Study 2017 to minimize the risk from all diet-related causes of death (Afshin et al., 2019).The bulk of staple crops consumed in Ethiopia, Kenya, and Tanzania are cereals, where they provide $88\%$–$95\%$ of consumed calories from starchy staples, on average. However, cereal consumption in all four countries mainly consists of refined grains and grain products and is dominated by maize flour in Kenya, Tanzania, and Uganda. Cereal consumption is least diverse in Tanzania and Kenya. Maize accounts nationally for $66\%$ (277 g/day per AE) and $63\%$ (238 g/day per AE) of total cereal consumption quantity in Tanzania and Kenya, respectively, of which most comes from the consumption of non-bran maize flours (corresponding to $91\%$ and $79\%$ of total maize consumption, respectively). In Ethiopia, maize consumption is comparably low, accounting for an average of $24\%$ of total cereal consumption quantity (132 g/day per AE). Teff consumption and millet and sorghum consumption average at $27\%$ and $19\%$ of total cereal consumption quantity (146 g/day and 106 g/day per AE), respectively. In Uganda, cereal consumption is generally low and provides only $34\%$ of all calories obtained from starchy staples, as starchy roots and tubers are the main staple crops. Maize consumption averages $55\%$ of cereal consumption (85 g/day per AE) nationwide; almost all maize is consumed as normal and fine flour ($93\%$). Rice is the second most consumed cereal in Kenya, Tanzania, and Uganda (varying nationally between 29 g/day and 98 g/day per AE and $14\%$ and $23\%$ of total cereal consumption quantities).Across the four countries, there are notable differences in the consumption of traditional cereals (sorghum, millet, and teff) compared to the Green Revolution cereals (maize, wheat, and rice) between rural and urban areas and, within these areas, between the poor and the rich. The consumed quantities of millet and sorghum are consistently larger and provide greater shares of calories from cereal consumption in rural than urban areas. Within rural and urban areas, millet and sorghum consumption also accounts for larger shares of cereal calories among the lower income quintiles than the higher income quintiles in all four countries. Teff consumption in Ethiopia, however, is much higher in urban than rural areas and among the rich than the poor. Alt-text: Box 1Fig. 2Gaps (%) between estimated household consumption and EAT-Lancet reference diet intakes by expenditure quintile in rural and urban areasSource: Authors' estimates from nationally representative survey data in 2014–2017.Note: Consumption gaps measure the different between consumption quantities and reference intakes, both measured on the basis of calories. Q1 denotes the lowest income quintile by rural or urban areas, and Q5 the richest. Fig. 2 Overall, the smallest differences in consumption gaps across income quintiles occur for pulses and nuts/seeds, especially in urban areas, suggesting that demand for this food group does not rise much with total expenditure. In urban Uganda, the average pulse and nut/seed consumption gaps for the richest and poorest income quintiles are somewhat larger than those for the other income quintiles. Average pulse and nut/seed consumption gaps are relatively similar in rural and urban areas, and consumption never exceeds $60\%$ of the reference intake for this food group. On the other hand, Fig. 2 also shows that, in addition to starchy staples, the two highest income quintiles in urban areas of all four countries overconsume added sugars—even without the inclusion of food consumed away from home. In Kenya, the wealthiest of the four countries, average consumption of sugars among all income quintiles in both urban and rural areas are above the maximum intake level of the EAT-Lancet reference diet. The consumption breakdown by income quintile also provides some evidence on general consumer preferences for the different food groups. For example, in both rural and urban areas of all four countries, the consumption of animal-source food groups rises sharply with increasing household expenditure levels, as does consumption of fruits, suggesting these are preferred foods. In contrast, vegetable consumption shows variation across geographies, sometimes rising steeply with total expenditure but in some geographies rising very modestly (e.g. Uganda). Notably, pulse and nut/seed consumption increases only marginally with total expenditure in a number of geographies. ## Explaining consumption gaps for healthy foods What explains the observed large consumption gaps for healthy foods in East Africa? Three key hypotheses are explored: low household food budgets (due to insufficient disposal income), high costs of a healthy diet (due to high prices of nutritious foods in local consumer markets), and weak consumer preferences for healthy foods (as revealed by econometrically derived income elasticities). ## The poverty problem: Are household food budgets too low? Previous research by Hirvonen and colleagues used nationally averaged price data for 159 countries to demonstrate that the EAT-Lancet reference diet is expensive relative to real household incomes in LMICs (Hirvonen et al., 2020). Here we focus on a more granular comparison of localized EAT-Lancet reference diet costs relative to each household's food expenditures by rural and urban areas of each studied country, graphed in Fig. 3. The cheapest diets that meet the reference intakes from locally sourced foods are always less expensive in rural than urban areas, varying between $1.93 in rural Uganda and $2.33 in rural Ethiopia, while in urban areas it varies between $2.25 in urban Uganda and $3.09 in urban Ethiopia (all in 2011 PPP).Fig. 3Distributions of household food expenditure and the costs of the EAT-Lancet reference diet in rural areas (blue lines) and urban areas (red lines)Note: One adult equivalent (AE) corresponds to an average adult with a dietary energy requirement of 2500 kcal/day.. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)Fig. 3Source: Authors' estimates from nationally representative survey data in 2014–2017. However, consistent with Hirvonen and colleagues’ international analysis, we also find that the majority of households cannot afford the EAT-Lancet reference diet, especially in rural areas where median food expenditure is typically about half of the cost of the cheapest reference diet. Indeed, the shares of the population unable to afford the EAT-Lancet reference diet is extremely high in all four countries, particularly in rural areas: $96\%$ of rural and $91\%$ of urban Ethiopians live in households whose food expenditures are below the costs of the reference diet, while the corresponding rural and urban estimates are $88\%$ and $79\%$ for Kenya, $90\%$ and $80\%$ for Tanzania, and $93\%$ and $87\%$ for Uganda. Thus, small household food budgets are a major barrier towards convergence of current diets to the EAT-Lancet reference diet for most of the population in East Africa. ## The price problem: Are the costs of consuming healthy foods too high? To assess whether the cost of nutritious foods was a limiting factor in their consumption, we estimated the median costs of the reference intakes per day (Fig. 4, Panel A) and the median costs per 100 Kcal (Panel B) for each required food group of the EAT-Lancet reference diet. Fig. 4Food group costs of the EAT-Lancet reference diet in rural and urban areas. Note: Reference intakes of the EAT-Lancet reference diet are measured on the basis of calories. Fig. 4Source: Authors' estimates from nationally representative survey data in 2014–2017. Panel A shows that the meat, fish, and eggs group is the most expensive component of the EAT-Lancet reference diet, costing anywhere between $0.50 and $1.09 per day (in 2011 PPP) to meet recommended intakes, or $26\%$–$38\%$ of the total costs of the EAT-Lancet reference diet. Consuming the reference intakes of other nutritious food groups and starchy staples has broadly similar daily costs of anywhere between $0.15 and $0.43 (except for pulses and nuts/seeds in urban Ethiopia). Yet Panel B shows why food-insecure households would find it difficult to choose foods that are nutrient-dense but expensive sources of calories. The costs per 100 kcal are also the highest for meat, fish, and eggs, ranging from $0.33 and $0.72, followed by vegetables ($0.25 to $0.55 per 100 kcal), and fruits and dairy (both varying between $0.10 and $0.25 per 100 kcal). Pulses, nuts and seeds are always much cheaper than meat/eggs/fish, but costs vary sizably over geographies. In contrast to these nutrient-dense food groups, staple foods and added fats and sugars are very inexpensive in caloric terms, costing just a few cents per 100 kcal. For very poor and food-insecure households this makes these foods an attractive source of calories and implies that it is economically quite costly for them to diversify away from starchy staples into nutrient dense foods. ## The preference problem: Are consumer preferences for nutritious foods too weak? Fig. 2 indicated that richer households tended to have smaller consumption gaps for EAT-Lancet reference diet food groups than poorer households; clearly, steeper gradients between consumption of a food and total household expenditure would likely imply strong preferences for that food in a given geography. Here we extend that logic by more rigorously estimating income (expenditure) elasticities for different foods as a means of inferring latent food preferences, given relative food prices and various household characteristics, from a demand systems estimation. Specifically, these income elasticities tell us that if a household increases their consumption of a given food by just $3\%$ in response to a $10\%$ increase in income (i.e., an elasticity of approximately 0.3), then it can be inferred that the prevailing preferences for that food is relatively week and that real income growth is unlikely to substantially narrow an existing consumption gap for that food. In contrast elasticities closer to unity would indicate moderately strong preferences, and elasticities in excess of unity would indicate very strong preferences for a given food. Fig. 5 shows estimated average income elasticities for 15 main food groups, which are a further disaggregation of the major food groups of the EAT-Lancet reference diet implemented to reflect potentially diverse preference for foods within those food groups. For example, we separate bananas (sometimes a staple in Africa) from other fruits, and dark green leafy vegetables from other vegetables. The elasticities are derived from econometric estimates of complete food demand system models for rural and urban areas of each country (see Materials and Methods), and hollow circles represent elasticity estimates that are not statistically significant from zero at the $5\%$ level. Fig. 5Average income elasticities of food demand in rural and urban areasNotes: Hollow circles denote elasticity estimates that are insignificantly different from zero at the $5\%$ level. DGLV = dark green leafy vegetables. Bananas include plantains. Considering local consumption frequencies, “Eggs” includes fish (rarely consumed) in Ethiopia, but “Fish” includes eggs in Kenya, Tanzania, and Uganda. “ Other foods” include snacks, sweets, and other highly processed foods, as well as spices and condiments.. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)Fig. 5Source: Authors' estimates from complete food demand system models, using data from nationally representative surveys in 2014–2017. Overall, the results reveal diversity in the revealed consumer preferences across food groups and household locations, but nevertheless present several common food preference patterns across rural and urban areas in the four countries. First, the income elasticities for starchy staples are often surprisingly high, especially in rural areas. This suggests that most consumers in these countries still have relatively strong preferences for staple foods, likely explained by persistent food insecurity and a resulting preference for cheap and storable sources of calories. Second, the income elasticities for animal-source foods, especially meat, as well as fruits other than bananas are generally high. While meat is an important, nutritious food group especially for children and mothers in poor households, high income responsiveness of meat demand is concerning among wealthier households that already consume enough animal-source foods to support their protein and micronutrient needs. Fish also has high income elasticities in Tanzania and Uganda, where it is an important protein and micronutrient source in rural and urban diets. Third, income elasticities for vegetables are generally modest in both rural and urban areas of the four countries, suggesting that, with rising income, consumers increase vegetable consumption by lower margins than they do for most animal-source foods. In rural areas, in particular, the income elasticities for dark green leafy vegetables are lower than those for other vegetables, despite being nutrient-dense. Fourth, income elasticities for pulses and nuts/seeds are close to unity, particularly in rural areas of the four countries, suggesting that rural income growth translates into relatively uniform increases in consumption of pulses and nuts/seeds, which can contribute to narrowing the consumption gaps for these rich plant-based protein and micronutrient sources in rural East Africa. However, income elasticities for pulses are typically much lower than they are for meat in urban geographies. Finally, income elasticities for potentially unhealthy food groups (including added sugars; other foods such as snacks, sweets, and other highly processed foods; and beverages) consumed at home vary but tend to be quite high, especially for beverages (mainly tea, soft drinks, and alcohol). In addition, separate results (not reported) show that expenditures for food consumed away from home rise at least proportionately with household incomes (especially in urban areas), suggesting that consumer preferences for prepared, possibly unhealthy foods are likely quite strong. ## Discussion The EAT-Lancet Commission report proposed an international healthy reference diet consistent with good health and NCD prevention and further demonstrated that most regions in the world heavily under-consume protective foods and over-consume unhealthy foods (Willett et al., 2019). Economic studies, however, have documented the high costs of nutritious foods (Headey and Alderman, 2019) and healthy diets (Dizon et al., 2019; Mahrt et al., 2019; Raghunathan et al., 2020; FAO, IFAD, UNICEF, WFP, WHO, 2022), including the EAT-diet (Hirvonen et al., 2020). In this study we extended these previous analyses by documenting the large consumption gaps for key nutritious foods among most rural and urban economic groups in four African countries, and systematically exploring why these gaps exist. Low incomes and food expenditures are clearly an overarching constraint in these countries – less than $20\%$ of households can afford the EAT-diet – and poorer and more food-insecure consumers tend to naturally minimize consumption of fruits and vegetables because they are expensive sources of calories (if not of nutrients). However, our demand analysis also reveals instances of low income elasticities for vegetables, which implies that income is not the dominant constraint to increased consumption of these nutritious foods. There are also indications that consumption of pulses, nuts and seeds rises less steeply than it does for animal-sourced foods among better off urban populations. These elasticities imply that even accelerated household income growth over the next decade or two would only close consumption gaps for nutritious foods by a few percentage points. Income elasticities for fruits and animal-sourced foods, on the other hand, are much larger. For some animal sourced foods like meat (red meat especially), which generally has a high income elasticity, rapidly rising incomes can lead to health risks as populations shift to excess consumption (Willett et al., 2019). This study has several strengths, but also limitations. Diets are consumed by individuals, not households, and aggregation to the household level can introduce recall errors, and ignores intra-household distributional issues in food consumption. However, only household surveys provide nationally representative data for the studied countries, as well as detailed expenditures and food consumption quantities. The surveys in question also vary in terms of how extensive their food lists are, and do not report quantities of food items consumed away from home, so consumption of different foods could be underestimated. The EAT-Lancet reference diet has itself also been questioned for having relatively low and option intake of animal-source foods (Adesogan et al., 2020). Achieving a healthy diet without animal-source foods is possible, although these foods are rich sources of bioavailable essential micronutrients for which deficiencies are common especially among children and women in developing countries. Animal-source foods are also culturally important foods in East Africa. Therefore, this analysis considers animal-source foods as a required food group of the reference diet. An important extension to this kind of research is to use national food-based dietary guidelines instead (Herforth et al., 2019), although these too often fail to separate out meat, fish and eggs from pulses and nuts. Finally, demand system estimation also presents many challenges with household surveys that vary in quality; compelling the researcher to make potentially important choices about the treatment of possible data errors and heterogeneity across different types of households, including farm households that consume some of the food they produce. Despite these challenges, the main conclusions of our study on the heterogeneous demand for healthy and unhealthy foods is likely to be robust. The significant under-consumption of healthy foods, largely due to affordability issues, is consistent with a global study demonstrating the high relative prices of nutrient-dense foods in Africa (Headey and Alderman, 2019). Africa-wide research on household food demand (Colen et al., 2018) shows strong demand for animal-sourced foods and sugar-sweetened beverages, moderate demand for fruits, and relatively weak demand for pulses, nuts and vegetables. A more nutrition-focused analysis of the dietary patterns of young children in Africa found highly heterogeneous consumption-wealth gradients (Choudhury et al., 2019). Consumption prevalence of dark green leafy vegetables declined with wealth, suggesting very weak demand compared to very strong gradients for fruits, dairy and animal-sourced foods. Collectively, this body of work suggests potential overconsumption of certain types of nutritious foods as incomes increase even as there are large and potentially persistent gaps for other nutritious foods, such as vegetables and legumes/nuts, that are not income-sensitive. Such heterogeneity in demand suggests the need for multiple types of approaches to achieve healthier diets. Supply-side interventions clearly have some scope to address the high cost and poor accessibility of healthy foods through nutrition-sensitive agricultural programs (Ruel et al., 2018), value chain interventions (Allen and de Brauw, 2018), and cross-cutting investments in efficiency-enhancing infrastructure such as irrigation, electrification, transport and cold storage, and food safety (Muunda et al., 2021). However, the demand analysis in this study also raises concerns that, even as African consumers are increasingly able to afford more nutritious foods and healthy diets, they may not choose to do so. Such findings beg the question of what can be done to foster healthy dietary choices through demand-side interventions. By far the most common interventions used to affect food choices in LMICs are social behavioral change communications (SBCC) interventions. Personal or group-based BCC interventions, however, have typically focused on improving maternal and child diets and nutrition (Kim et al., 2018), but the approaches used for these nutritionally vulnerable groups may not be cost-effective or easily scaled up or adapted to other demographic groups. In high income countries, many nutrition-focused interventions have tried to influence the broader consumer population. Interventions in supermarkets are common, including experimental work altering food positioning (An et al., 2013; Huitink et al., 2020; Hyseni et al., 2017) and food labelling (Andreyeva and Luedicke, 2013; Campos et al., 2011; Cecchini and Warin, 2016; Hyseni et al., 2017; Just and Gabrielyan, 2018), both of which show some evidence of impact. However, their applicability to lower income and rural consumers in sub-Saharan Africa is questionable, especially given their limited access to and use of supermarkets. Efforts to affect rural food choice have relied more on production-based interventions, particularly Enhanced Homestead Food Production (EHFP) programs in which agricultural extension (focused on promotion of increasing production of vegetables, fruits, poultry or fish) is combined with SBCC to raise caregiver knowledge and promote greater consumption of nutrient-rich foods, especially mothers and young children (Haselow et al., 2016). These programs show some evidence of impact (An et al., 2013; Huitink et al., 2020; Hyseni et al., 2017), especially when supply-side and demand-side interventions are combined. It is unclear, however, interventions whether there is a need for supply-side interventions where nutritious foods are available and affordable in local markets (Hirvonen and Headey, 2018), what nutritional tradeoffs exist when women adopt new agricultural practices in the context of significant time constraints (Ruel et al., 2018), and whether fruit and vegetable interventions can be implemented in more water-scarce settings (Hirvonen and Headey, 2018). Consideration of water access, market access and women's time constraints is therefore truly critical in the design of EHFP interventions. A potential alternative or complement to EHFP programs is the use of food/cash transfers that could directly stimulate demand by expanding total food demand and influencing the composition of demand through nutrition BCC. Food/cash transfers have been shown to increase consumption of healthy foods, but also of unhealthy foods in some instances (Ahmed et al., 2019; Almas et al., 2019; Cunha et al., 2018; Hidrobo et al., 2014; Hoddinott & Skoufias, 2004; Leroy et al., 2010). Evidence from Latin America and Egypt (food subsidies) suggests that some of these programs have inadvertently led to rises in overweight/obesity in environments where ultra-processed foods were available, accessible, and affordable (Leroy et al., 2010; Leroy et al., 2019). Potentially, conditional BCC interventions focused on healthy diets and lifestyles for all household members could have synergistic effects with transfers by reducing the risks of both undernutrition and overweight/obesity (Ahmed et al., 2019). School feeding and take-home-rations (THR) programs have also led to evidence of benefits for child diets as well as cognitive development and schooling outcomes (Aurino et al., 2020; Evans et al., 2012; Gelli et al., 2018; Jomaa et al., 2011; Webb, 2015), and potential spillovers for younger (pre-school) siblings (Aurino et al., 2020). However, there is limited evidence on sustained impacts on diets, few interventions that combine nutritional education with school feeding, and no evidence that we are aware of on whether school-based interventions yield long-term dietary or nutritional benefits through adulthood. One recent study from India did report intergenerational benefits in the form of large height-for-age Z scores among children of mothers exposed to school feeding programs in the 1990s (Chakrabarti et al., 2021). Another recent experimental study from Vietnam combined nutrition education with school snacks and found that students retained improved knowledge after 6 months, but only when education was supported by complementary healthy snacks (Nguyen et al., 2021). Despite limited evidence, the fact that LMICs have extremely young populations, and that long-term food preferences may be formed in childhood and adolescence, suggests that school-based interventions might be an important means of shifting preferences at a generational level. Finally, it is clear that taxation policies offer scope to alter the relative prices of healthy and unhealthy foods, and hence to potentially improve diets at scale. However, evidence on this issue is solely confined to middle- and high-income countries, and particularly to taxes on unhealthy foods like added sugars and fats, and sugary beverages (Bartlett et al., 2014; Black et al., 2012; Chakrabarti et al., 2018; Eyles et al., 2012; Hyseni et al., 2017; Jou and Techakehakij, 2012; McFadden et al., 2014; Ni Mhurchu et al., 2010; Powell et al., 2013; Sturm et al., 2013; Thow et al. 2014; Waterlander et al., 2012). The ex ante modelling components of this research suggests that taxes will reduce consumption of unhealthy foods, and some empirical evidence from countries like Mexico has also found evidence of impacts on reduced consumption of taxed foods and beverages (Jou and Techakehakij, 2012; Powell et al., 2013; Thow et al., 2014). There is much less evidence on subsidies or vouchers to promote consumption of healthy foods, as well as less practical experience on how these would be implemented in LMICs. In India, the addition of pulses to the nation's Public Distribution Scheme (PDS) has been effective, but pulses are less perishable and logistically less demanding than fresh fruits, vegetables or animal sourced foods (Chakrabarti et al., 2018). In Indonesia, the country's major rice-in-kind program has transitioned to a food voucher scheme designed to diversify diets (Sembako). But while early evaluation evidence showed increased consumption of eggs (Banerjee et al., 2021), more anecdotal recent evidence suggests that suppliers for the program have reverted back to rice. Still, interventions to favorably alter the set of food prices facing consumers in principle offers a scalable approach to dietary improvement, provided important logistical challenges can be managed. In summary, there are many potential points of intervention for improving diets, but a significant dearth of relevant evidence for LMICs, and even less information on cost effectiveness. There are particularly large knowledge gaps on interventions that are likely to have higher potential in LMICs and in remote rural communities, including ICT-based messaging, interventions in wet markets, school-based nutrition education, nutrition education/BCC programs designed to address dietary risks for all household members (not just mothers and young children) and food voucher or public distribution schemes that offer nutrient-dense non-staple foods. Moreover, there is also an urgent need for double-duty interventions that support consumers in achieving healthier diets (through both supply and demand actions) to simultaneously address multiple forms of malnutrition, including overweight/obesity and multiple micronutrient deficiencies (Hawkes et al., 2020). ## Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Derek Headey reports financial support was provided by $\frac{10.13039}{100000877}$The Rockefeller Foundation. Derek Headey reports financial support was provided by $\frac{10.13039}{100000865}$Bill and Melinda Gates Foundation. Marie Ruel reports financial support was provided by $\frac{10.13039}{100000877}$The Rockefeller Foundation. Olivier Ecker reports financial support was provided by $\frac{10.13039}{100000877}$The Rockefeller Foundation. Andrew Comstock reports financial support was provided by $\frac{10.13039}{100000877}$The Rockefeller Foundation. ## Supplementary data The following is the Supplementary data to this article:Multimedia component 1Multimedia component 1 ## Data availability The authors do not have permission to share data. ## References 1. Adesogan A.T., Havelaar A.H., McKune S.L., Eilittä M., Dahl G.E.. Animal source foods: sustainability problem or malnutrition and sustainability solution? Perspective matters. Global Food Secur. (2020) 25 2. Afshin A., Sur P.J., Fay K.A., Cornaby L., Ferrara G., Salama J.S.. Health effects of dietary risks in 195 countries, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet (2019) 393 1958-1972. PMID: 30954305 3. Ahmed A., Hoddinott J., Roy S.. (2019) 4. Allen S., de Brauw A.. Nutrition sensitive value chains: theory, progress, and open questions. Global Food Secur. (2018) 16 22-28 5. Almas I., Haushofer J., Shapiro J.P.. (2019) 6. An R., Patel D., Segal D., Sturm R.. Eating better for less: a national discount program for healthy food purchases in South Africa. Am. J. Health Behav. (2013) 37 56-61. PMID: 22943101 7. Andreyeva T., Luedicke J.. Federal food package revisions: effects on purchases of whole-grain products. Am. J. Prev. Med. (2013) 45 422-429. PMID: 24050418 8. Attanasio O., Di Maro V., Lechene V., Phillips D.. Welfare consequences of food prices increases: evidence from rural Mexico. J. Dev. Econ. (2013) 104 136-151 9. Aurino E., Gelli A., Adamba C., Osei-Akoto I., Alderman H.. Food for thought? Experimental evidence on the learning impacts of a large-scale school feeding program. J. Hum. Resour. (2020) 1019-1051R1. DOI: 10.3368/jhr.58.3.1019-10515R1 10. Baker P., Friel S.. Food systems transformations, ultra-processed food markets and the nutrition transition in Asia. Glob. Health (2016) 12 80 11. Banerjee A., Hanna R., Olken B.A., Satriawan E., Sumarto S.. (2021) 12. Banks J., Blundell R., Lewbel A.. Quadratic engel curves and consumer demand. Rev. Econ. Stat. (1997) 79 527-539 13. Bartlett S., Klerman J., Olsho L.. Evaluation of the Healthy Incentives Pilot (HIP): Final Report (2014) 14. Black A.P., Brimblecombe J., Eyles H., Morris P., Vally H., O′Dea K.. Food subsidy programs and the health and nutritional status of disadvantaged families in high income countries: a systematic review. BMC Publ. Health (2012) 12 1099 15. Buse A.. Evaluating the linearized almost ideal demand system. Am. J. Agric. Econ. (1994) 76 781-793 16. Campos S., Doxey J., Hammond D.. Nutrition labels on pre-packaged foods: a systematic review. Publ. Health Nutr. (2011) 14 1496-1506 17. Cecchini M., Warin L.. Impact of food labelling systems on food choices and eating behaviours: a systematic review and meta-analysis of randomized studies. Obes. Rev. (2016) 17 201-210. PMID: 26693944 18. Chakrabarti S., Kishore A., Roy D.. Effectiveness of food subsidies in raising healthy food consumption: public distribution of pulses in India. Am. J. Agric. Econ. (2018) 100 1427-1449 19. Chakrabarti S., Scott S.P., Alderman H., Menon P., Gilligan D.O.. Intergenerational nutrition benefits of India's national school feeding program. Nat. Commun. (2021) 12 4248. PMID: 34253719 20. Choudhury S., Headey D.D., Masters W.A.. First foods: diet quality among infants aged 6–23 months in 42 countries. Food Pol. (2019) 88 101762 21. Colen L., Melo P.C., Abdul-Salam Y., Roberts D., Mary S., Gomez Y Paloma S.. Income elasticities for food, calories and nutrients across Africa: a meta-analysis. Food Pol. (2018) 77 116-132 22. Cunha J.M., De Giorgi G., Jayachandran S.. The price effects of cash versus in-kind transfers. Rev. Econ. Stud. (2018) 86 240-281 23. Dataset: ICF InternationalThe Demographic and Health Surveys Program2020ICF InternationalCalverton MDhttps://dhsprogram.com/. (2020) 24. Deaton A., Muellbauer J.. An almost ideal demand system. Am. Econ. Rev. (1980) 70 312-326 25. Deaton A., Zaidi S.. (2002) 26. Dizon F., Herforth A., Wang Z.. The cost of a nutritious diet in Afghanistan, Bangladesh, Pakistan, and Sri Lanka. Global Food Secur. (2019) 21 38-51 27. Ecker O., Qaim M.. Analyzing nutritional impacts of policies: an empirical study for Malawi. World Dev. (2011) 39 412-428 28. Edgerton D.L.. Weak separability and the estimation of elasticities in multistage demand systems. Am. J. Agric. Econ. (1997) 79 62-79 29. Evans C.E., Christian M.S., Cleghorn C.L., Greenwood D.C., Cade J.E.. Systematic review and meta-analysis of school-based interventions to improve daily fruit and vegetable intake in children aged 5 to 12 y. Am. J. Clin. Nutr. (2012) 96 889-901. PMID: 22952187 30. Eyles H., Ni Mhurchu C., Nghiem N., Blakely T.. Food pricing strategies, population diets, and non-communicable disease: a systematic review of simulation studies. PLoS Med. (2012) 9 31. FAO, IFAD, UNICEF, WFP, WHOThe State of Food Security and Nutrition in the World 2022: Repurposing Food and Agricultural Policies to Make Healthy Diets More Affordable2022FAO, IFAD, UNICEF, WFP and WHORome. (2022) 32. FAO, WHO, & UNUHuman energy requirementsReport of a Joint FAO/WHO/UNU Expert Consultation2001United NationsRome. Report of a Joint FAO/WHO/UNU Expert Consultation (2001) 33. Gelli A., Margolies A., Santacroce M., Roschnik N., Twalibu A., Katundu M.. Using a community-based early childhood development center as a platform to promote production and consumption diversity increases children's dietary intake and reduces stunting in Malawi: a cluster-randomized trial. J. Nutr. (2018) 148 1587-1597. PMID: 30204916 34. Haselow N.J., Stormer A., Pries A.. Evidence‐based evolution of an integrated nutrition‐focused agriculture approach to address the underlying determinants of stunting. Matern. Child Nutr. (2016) 12 155-168. PMID: 27187913 35. Hawkes C., Ruel M.T., Salm L., Sinclair B., Branca F.. Double-duty actions: seizing programme and policy opportunities to address malnutrition in all its forms. Lancet (2020) 395 142-155. PMID: 31852603 36. Headey D.D., Alderman H.H.. The relative caloric prices of healthy and unhealthy foods differ systematically across income levels and continents. J. Nutr. (2019) 149 2020-2033. PMID: 31332436 37. Herforth A., Arimond M., Álvarez-Sánchez C., Coates J., Christianson K., Muehlhoff E.. A global review of food-based dietary guidelines. Adv. Nutr. (2019) 10 590-605. PMID: 31041447 38. Hidrobo M., Hoddinott J., Peterman A., Margolies A., Moreira V.. Cash, food, or vouchers? Evidence from a randomized experiment in northern Ecuador. J. Dev. Econ. (2014) 107 144-156 39. Hirvonen K., Headey D.. Can governments promote homestead gardening at scale? Evidence from Ethiopia. Global Food Secur. (2018) 19 40-47 40. Hirvonen K., Bai Y., Headey D., Masters W.A.. Affordability of the EAT-Lancet reference diet: a global analysis. Lancet Global Health (2020) 8 e59-e66. PMID: 31708415 41. Hoddinott John, Skoufias Emmanuel. The impact of PROGRESA on food consumption. Econ. Dev. Cult. Change (2004) 53 37-61 42. Huitink M., Poelman M.P., Seidell J.C., Kuijper L.D.J., Hoekstsra T., Dijkstra C.. Can healthy checkout counters improve food purchases? Two real-life experiments in Dutch supermarkets. Int. J. Environ. Res. Publ. Health (2020) 17 8611 43. Hyseni L., Atkinson M., Bromley H., Orton L., Lloyd-Williams F., McGill R.. The effects of policy actions to improve population dietary patterns and prevent diet-related non-communicable diseases: scoping review. Eur. J. Clin. Nutr. (2017) 71 694-711. PMID: 27901036 44. Jomaa L.H., McDonnell E., Probart C.. School feeding programs in developing countries: impacts on children's health and educational outcomes. Nutr. Rev. (2011) 69 83-98. PMID: 21294742 45. Jou J., Techakehakij W.. International application of sugar-sweetened beverage (SSB) taxation in obesity reduction: factors that may influence policy effectiveness in country-specific contexts. Health Pol. (2012) 107 83-90 46. Just D.R., Gabrielyan G.. Influencing the food choices of SNAP consumers: lessons from economics, psychology and marketing. Food Pol. (2018) 79 309-317 47. Kim S.S., Nguyen P.H., Tran L.M., Sanghvi T., Mahmud Z., Haque M.R.. Large-scale social and behavior change communication interventions have sustained impacts on infant and young child feeding knowledge and practices: results of a 2-year follow-up study in Bangladesh. J. Nutr. (2018) 148 1605-1614. PMID: 30169665 48. Leroy J.L., Gadsden P., Rodríguez-Ramírez S., de Cossío T.G.. Cash and in-kind transfers in poor rural communities in Mexico increase household fruit, vegetable, and micronutrient consumption but also lead to excess energy consumption. J. Nutr. (2010) 140 612-617. PMID: 20089777 49. Leroy J.L., Olney D.K., Ruel M.T.. PROCOMIDA, a food-assisted maternal and child health and nutrition program, contributes to postpartum weight retention in Guatemala: a cluster-randomized controlled intervention trial. J. Nutr. (2019) 149 2219-2227. PMID: 31373374 50. Leser C.E.V.. Forms of engel functions. Econometrica (1963) 31 694-703 51. Lewbel A.. The rank of demand systems: theory and nonparametric estimation. Econometrica (1991) 59 711-730 52. Mahrt K., Mather D., Herforth A., Headey D.. (2019) 53. McFadden A., Green J.M., Williams V., McLeish J., McCormick F., Fox-Rushby J.. Can food vouchers improve nutrition and reduce health inequalities in low-income mothers and young children: a multi-method evaluation of the experiences of beneficiaries and practitioners of the Healthy Start programme in England. BMC Publ. Health (2014) 14 148 54. Muunda E., Mtimet N., Schneider F., Wanyoike F., Dominguez-Salas P., Alonso S.. (2021) 55. Nguyen T., de Brauw A., van den Berg M., Do H.T.P.. (2021) 56. Ni Mhurchu C., Blakely T., Jiang Y., Eyles H.C., Rodgers A.. Effects of price discounts and tailored nutrition education on supermarket purchases: a randomized controlled trial. Am. J. Clin. Nutr. (2010) 91 736-747. PMID: 20042528 57. Powell L.M., Chriqui J.F., Khan T., Wada R., Chaloupka F.J.. Assessing the potential effectiveness of food and beverage taxes and subsidies for improving public health: a systematic review of prices, demand and body weight outcomes. Obes. Rev. (2013) 14 110-128. PMID: 23174017 58. Raghunathan K., Headey D., Herforth A.. (2020) 59. Ruel M.T., Quisumbing A.R., Balagamwala M.. Nutrition-sensitive agriculture: what have we learned so far?. Global Food Secur. (2018) 17 128-153 60. Shonkwiler J.S., Yen S.T.. Two-step estimation of a censored system of equations. Am. J. Agric. Econ. (1999) 81 972-982 61. Sturm R., An R., Segal D., Patel D.. A cash-back rebate program for healthy food purchases in South Africa: results from scanner data. Am. J. Prev. Med. (2013) 44 567-572. PMID: 23683973 62. Thow A.M., Downs S., Jan S.. A systematic review of the effectiveness of food taxes and subsidies to improve diets: understanding the recent evidence. Nutr. Rev. (2014) 72 551-565. PMID: 25091552 63. USDAUSDA Food Composition Database2020United States Department of Agriculture (USDA)Washington DC. (2020) 64. Waterlander W.E., Steenhuis I.H., de Boer M.R., Schuit A.J., Seidell J.C.. Introducing taxes, subsidies or both: the effects of various food pricing strategies in a web-based supermarket randomized trial. Prev. Med. (2012) 54 323-330. PMID: 22387008 65. Webb P.. (2015) 66. Willett W., Rockström J., Loken B., Springmann M., Lang T., Vermeulen S.. Food in the Anthropocene: the EAT–Lancet Commission on healthy diets from sustainable food systems. The Lancet (2019) 16 1-47. DOI: 10.1016/S0140-6736(18)31788-4 67. Working H.. Statistical laws of family expenditure. J. Am. Stat. Assoc. (1943) 38 43-56
--- title: 'Intermediate hyperglycaemia, diabetes and blood pressure in rural Bangladesh: five-year post-randomisation follow-up of the DMagic cluster-randomised controlled trial' authors: - Edward Fottrell - Carina King - Naveed Ahmed - Sanjit Kumer Shaha - Joanna Morrison - Malini Pires - Abdul Kuddus - Tasmin Nahar - Hassan Haghparast-Bidgoli - A.K. Azad Khan - Kishwar Azad journal: The Lancet Regional Health - Southeast Asia year: 2022 pmcid: PMC10015271 doi: 10.1016/j.lansea.2022.100122 license: CC BY 4.0 --- # Intermediate hyperglycaemia, diabetes and blood pressure in rural Bangladesh: five-year post-randomisation follow-up of the DMagic cluster-randomised controlled trial ## Body Research in contextEvidence before this studyThe Bangladesh DMagic trial reported in 2019 that mHealth and Participatory Learning and Action (PLA) community mobilisation interventions were effective at increasing knowledge and awareness of type-2 diabetes and its risk factors in rural Bangladesh. The PLA intervention also resulted in large reductions in population prevalence of diabetes and intermediate hyperglycaemia, and reductions in the incidence of type 2 diabetes among an intermediate hyperglycaemic cohort. Economic evaluation showed the PLA intervention to be highly cost effective and subsequent equity analysis showed that PLA impacts were observed across age, sex and wealth groups. The medium- to long-term impacts of such population-level diabetes prevention and control interventions is unknown though evidence from other settings suggests that diabetes prevention strategies targeting high-risk individuals may require some degree of intervention maintenance. Added value of this studyOur five-year post-randomisation follow-up study shows that PLA effects on knowledge and awareness of diabetes remain but mHealth effects are no longer observed. Positive impacts of PLA on diabetes and intermediate hyperglycaemia outcomes are no longer seen. However, measures of hypertension, hypertension control and exploratory analyses of key risk behaviours and additional measures of blood pressure suggest lasting positive health impacts of PLA.Implications of all the available evidencePopulation-level interventions that seek to address broad cultural and societal influences of cardiometabolic risk may require a strong focus on maintenance of interventions strategies and effect. Though impacts of our PLA community mobilisation intervention on blood glucose are no longer observable 5-years post-randomisation, whole-population, community-based awareness and lifestyle interventions that prevent the onset of diabetes, even if only temporarily, may cumulate and contribute to wider positive impacts on health, including health behaviours and blood pressure, and should remain a priority for populations with a high burden of risk. ## Summary ### Background The DMagic trial showed that participatory learning and action (PLA) community mobilisation delivered through facilitated community groups, and mHealth voice messaging interventions improved diabetes knowledge in Bangladesh and the PLA intervention reduced diabetes occurrence. We assess intervention effects three years after intervention activities stopped. ### Methods Five years post-randomisation, we conducted a cross-sectional survey among a random sample of adults aged ≥30-years living in the 96 DMagic villages, and a cohort of individuals identified with intermediate hyperglycaemia at the start of the DMagic trial in 2016. Primary outcomes were: 1) the combined prevalence of intermediate hyperglycaemia and diabetes; 2) five-year cumulative incidence of diabetes among the 2016 cohort of individuals with intermediate hyperglycaemia. Secondary outcomes were: weight, BMI, waist and hip circumferences, blood pressure, knowledge and behaviours. Primary analysis compared outcomes at the cluster level between intervention arms relative to control. ### Findings Data were gathered from 1623 ($82\%$) of the randomly selected adults and 1817 ($87\%$) of the intermediate hyperglycaemia cohort. 2018 improvements in diabetes knowledge in mHealth clusters were no longer observable in 2021. Knowledge remains significantly higher in PLA clusters relative to control but no difference in primary outcomes of intermediate hyperglycaemia and diabetes prevalence (OR ($95\%$CI) 1.23 (0.89, 1.70)) or five-year incidence of diabetes were observed (1.04 (0.78, 1.40)). Hypertension (0.73 (0.54, 0.97)) and hypertension control (2.77 (1.34, 5.75)) were improved in PLA clusters relative to control. ### Interpretation PLA intervention effect on intermediate hyperglycaemia and diabetes was not sustained at 3 years after intervention end, but benefits in terms of blood pressure reduction were observed. ### Funding $\frac{10.13039}{501100000265}$Medical Research Council UK: MR/M$\frac{016501}{1}$ (DMagic trial); MR/T$\frac{023562}{1}$ (DClare study), under the Global Alliance for Chronic Diseases (GACD) Diabetes and Scale-up Programmes, respectively. ## Evidence before this study The Bangladesh DMagic trial reported in 2019 that mHealth and Participatory Learning and Action (PLA) community mobilisation interventions were effective at increasing knowledge and awareness of type-2 diabetes and its risk factors in rural Bangladesh. The PLA intervention also resulted in large reductions in population prevalence of diabetes and intermediate hyperglycaemia, and reductions in the incidence of type 2 diabetes among an intermediate hyperglycaemic cohort. Economic evaluation showed the PLA intervention to be highly cost effective and subsequent equity analysis showed that PLA impacts were observed across age, sex and wealth groups. The medium- to long-term impacts of such population-level diabetes prevention and control interventions is unknown though evidence from other settings suggests that diabetes prevention strategies targeting high-risk individuals may require some degree of intervention maintenance. ## Added value of this study Our five-year post-randomisation follow-up study shows that PLA effects on knowledge and awareness of diabetes remain but mHealth effects are no longer observed. Positive impacts of PLA on diabetes and intermediate hyperglycaemia outcomes are no longer seen. However, measures of hypertension, hypertension control and exploratory analyses of key risk behaviours and additional measures of blood pressure suggest lasting positive health impacts of PLA. ## Implications of all the available evidence Population-level interventions that seek to address broad cultural and societal influences of cardiometabolic risk may require a strong focus on maintenance of interventions strategies and effect. Though impacts of our PLA community mobilisation intervention on blood glucose are no longer observable 5-years post-randomisation, whole-population, community-based awareness and lifestyle interventions that prevent the onset of diabetes, even if only temporarily, may cumulate and contribute to wider positive impacts on health, including health behaviours and blood pressure, and should remain a priority for populations with a high burden of risk. ## Introduction Diabetes is a priority non-communicable disease (NCD) listed in the UN and WHO Action Plan to address the global burden of NCDs.1 The International Diabetes Federation estimate that approximately 700 million adults ($10.9\%$) will live with diabetes by 2045 and the greatest burden of disease will be in low- and middle-income countries.2 Currently, around $79\%$ of people with diabetes live in low- or middle-income countries, and more than $60\%$ live in Asian countries. The estimated prevalence of diabetes in Bangladesh is 2–$13\%$, depending on study design, methods and location, and the estimated prevalence of intermediate hyperglycaemia (impaired fasting glucose or impaired glucose tolerance) is between 2 and $22\%$.3 Of those living with diabetes in Bangladesh, it is estimated that 50–$75\%$ are undiagnosed and unaware of their condition.4,5 There is a need for effective and sustainable population-level interventions to raise awareness of and to prevent and control diabetes in settings such as Bangladesh, and a need for longitudinal evidence on the impact of these population-level interventions on diabetes and associated cardiometabolic risk.6 The DMagic (Diabetes Mellitus Action through Groups or Information for better Control) cluster randomised controlled trial showed that, after 18 months of a participatory learning and action (PLA) community mobilisation intervention, community awareness and understanding of type-2 diabetes mellitus (T2DM) was greatly increased and the odds of T2DM and intermediate hyperglycaemia was $64\%$ lower in intervention villages than control villages (adjusted odds ratio ($95\%$ confidence interval) 0.36 (0.27, 0.48)).7 Further, among individuals identified with intermediate hyperglycaemia before the intervention, the cumulative two-year incidence of T2DM was $59\%$ lower in intervention villages (0.41 (0.24, 0.67)). This equates with absolute reductions in prevalence and incidence of $21\%$ for T2DM and $9\%$ for intermediate hyperglycaemia. An mHealth intervention which was also tested in the DMagic trial, raised population knowledge and understanding of diabetes but had no effect on blood glucose measures when compared to controls.7 No intervention effects on BMI or other major risk factors for diabetes were observed in the DMagic trial. The DMagic trial ended in 2018, and was the first evidence of population-level community-based interventions for diabetes prevention and control using PLA. Though not directly comparable to our intervention or context, evidence from other settings suggests that diabetes prevention strategies targeting high-risk individuals can achieve reductions in diabetes incidence that last for several years post-intervention, but that often some degree of intervention maintenance is required.8, 9, 10 In the current study we aim to describe the medium-term sustainability of observed intervention effects in the absence of intervention maintenance and, given hyperglycaemia is a key modifiable risk factor for the development of cardiovascular diseases (CVDs), explore possible positive effects on measures of blood pressure three years after the end of all intervention activity. ## Setting This study took place in Faridpur District, south-central Bangladesh. Faridpur has a population of over 1.7 million people in an area of just over 2000 km2 and is situated on the banks of the Padma River. The district has a mainly agricultural economy, with the main crops being jute and rice. The population is mainly Bengali and almost $90\%$ of the population in Faridpur are Muslim, with the remaining population largely Hindu. Administratively, Faridpur District is divided into nine upazillas. Four upazillas in Faridpur District were purposefully selected because they were accessible to the district headquarters of the Diabetic Association of Bangladesh (BADAS) in Faridpur Sadar: these are Boalmari, Saltha, Madhukhali and Nagarkanda. For each of these upazillas, the 2011 Bangladesh Census11 was used to select 96 villages with population size of between 750 and 2500 (total estimated population 125,000). ## DMagic interventions & trial design DMagic was a three-arm, cluster-randomised trial of participatory community mobilisation, mHealth mobile phone voice messaging, and usual care (control) in 96 villages. Community mobilisation involved 18 monthly group meetings, led by salaried lay facilitators, applying a PLA cycle focused on diabetes prevention and control. 122 groups comprised of an average of 27 members each were established across 32 villages. Each group was open to all community members and progressed through a four phase PLA cycle of problem identification and prioritisation, strategy development, strategy implementation and evaluation. Facilitators were locally recruited men and women who each led up to nine male or female groups, respectively, and helped groups to plan and coordinate activities, including wider community meetings that involved sharing learnings and strategies with others in the local area. Group strategies varied between groups and depending on local priorities, but common approaches included awareness raising, group exercises (especially walking groups), and locally organised diabetes screening.12 The mHealth intervention involved free twice-weekly voice messages sent to individual's mobile phones across 32 villages over 14 months promoting awareness and behaviour change to reduce diabetes risk.13 Voice messages were developed based on formative research and behaviour change theory, as described previously.14 Anyone residing in any of the 32 mHealth villages with access to a mobile phone could opt-in to receive the intervention messages free of charge and recipients were encouraged via the messages to share the content of messages with family and friends. Stratified 1:1:1 randomisation of the 96 villages allocated them to the mHealth intervention, the community mobilisation intervention, or control, with each upazilla constituting one stratum. Because of the nature of the interventions being tested, the intervention team and participants could not be masked to allocation. The DMagic trial is registered with the ISRCTN registry (ISRCTN41083256). The current follow-up study was not part of the original trial design. ## Follow-up sample Using the household sampling frame developed for the DMagic endline survey in $\frac{2017}{18}$, a new sample of 20 adults aged ≥30 years was randomly selected from each study village (total $$n = 1920$$) in 2021. This sample size is based on $80\%$ power, with $5\%$ significance, to detect a $30\%$ reduction in the primary outcome of T2DM and intermediate hyperglycaemia between the 32 PLA and 32 control clusters, assuming a $40\%$ prevalence in the control clusters, intracluster correlation coefficient (ICC) of 0.07 and $20\%$ non-response. In addition, we purposively sampled known individuals identified with intermediate hyperglycaemia in the DMagic baseline survey conducted in 2016 who were also located in the DMagic endline survey in 2018 ($$n = 2099$$). Full details of how study clusters were sampled has been previously published.13 To select the current study sample, 20 households with at least one eligible adult were selected using simple random sampling. At the next stage, a single eligible adult from each household was selected for inclusion in the survey using simple random sampling. Eligibility was based on permanent residence of at least the past 6 months in the study village. Pregnancy was an exclusion criteria due to potential for gestational diabetes and other pregnancy-related metabolic, physiologic and anthropometric effects that were beyond the scope of our interventions. ## Pre-specified As in the DMagic trial, we had two primary outcomes. 1) The combined prevalence of intermediate hyperglycaemia and T2DM among adults aged ≥30 years. This uses the same definition as in the DMagic trial and is based on WHO definitions and blood glucose cut-offs for normoglycaemia, impaired fasting glucose, impaired glucose tolerance and T2DM, or a prior diagnosis of T2DM by a medical professional15 (supplementary Table S1). 2) The five-year cumulative incidence of T2DM (defined according to WHO criteria or based on reported medical diagnosis of T2DM) among individuals identified with intermediate hyperglycaemia in the 2016 DMagic baseline (pre-intervention) survey. Pre-specified secondary outcome measures were assessed among the random population sample. These were objective physical assessments of systolic and diastolic blood pressure, hypertension, hypertension control, mean BMI, proportion of overweight or obesity, and abdominal obesity (supplementary Table S1). In addition, we included survey-assessed measures of self-rated health (on scale of 0–100), diabetes knowledge (relating to causes, symptoms, complications, prevention and control) and proportion reporting a minimum of 150 min of physical activity per week. Among individuals reporting a prior diagnosis of diabetes we report diabetic control (defined as blood glucose levels below the diabetic threshold among individuals with a self-reported medical diagnosis of diabetes), self-reported receipt of medical diabetes treatment or advice, self-reported monthly blood glucose monitoring, and self-reported diabetes co-morbidities that respondents had been told by a medical professional were associated with diabetes. Measures of psychological distress using the Self-Rated Health Questionnaire (SRQ-20) and the mean daily number of fruit and vegetable portions consumed were reported in our DMagic trial but were dropped from our 2021 survey. ## Exploratory In addition to the aforementioned outcomes that allow direct comparison with the DMagic trial analysis, we included outcomes that explore possible mechanisms or effects of the DMagic interventions. The selection of exploratory outcomes was based on findings from our process evaluation12 and visual participatory analysis16 of DMagic which indicated certain behaviours, practices and attitudes that had not been pre-specified in our trial analysis but were considered to be particularly important by intervention participants and could plausibly mediate intervention effects. These were: participation in brisk walking activities and time spent engaging in brisk walking, self-reported sugar consumption, salt consumption and oil consumption, 24 hour dietary diversity, depression and anxiety, and median score on the Appraisal of Diabetes Scale (ADS),17 a standardised diabetes-specific tool to evaluate a person's thoughts about coping with diabetes.18 We introduced new measures of depression using the Patient Health Questionnaire 9 (PHQ-9), which is a nine-item questionnaire designed to screen for depression and has been use previously and validated in Bangladesh.19, 20, 21 All participants were screened using the two-item PHQ-2 tool and those who screened positive for possible depressive disorder (a score of 3 or more) completed the full PHQ-9 survey. Anxiety was assessed using the Generalised Anxiety Disorder Assessment (GAD-7), a scale developed to identify probable cases of generalised anxiety and to assess symptom severity, which has previously been validated in Bangladesh.22,23 Considering that vascular biology and epidemiological evidence suggests that better-controlled blood glucose or delayed diabetes may confer cardiovascular disease benefit24 we explored PLA intervention impacts on additional measures known to be independently associated with an increased risk of stroke and ischaemic heart disease.25 These were: isolated systolic blood pressure, isolated diastolic blood pressure, and pulse pressure, which is an indicator of large blood vessel stiffness (supplementary Table S1). ## Procedures Recruitment and training of data collectors took place in July 2021 and data collection took place between August–September 2021. Sampled individuals were visited at their household, informed of the study and consent was obtained. All sampled individuals in a single cluster were informed of the anthropometric, blood glucose, and blood pressure measurement requirements of the study and were requested to attend a local centre on the morning of a specified day following an overnight fast. The centre was established by the field team for the purposes of the study and was at a central, convenient location in the village. Collection of questionnaire data took place at a private outside location near the respondent's home before or after the physical measurements or at the time of physical measurement in the testing centre. Data were linked using a study ID number. Data were collected by 12 teams of fieldworkers comprised of a total of 28 men and women with at least secondary education who were recruited locally and selected through a written assessment and interview. All fieldworkers underwent 10 days training on survey methods and how to take physical measurements followed by one week supervised field practice and daily debriefs in villages in Faridpur that were not included in the study. Data collectors were supervised by four field supervisors with experience in survey methods. Each supervisor was responsible for three data collection teams, spending half a day observing and verifying data within each team at least every two days. Within each village, teams were aided by a village assistant, usually a young male, who received a daily payment to coordinate study participants and assist data collectors in their duties. Questionnaire data were gathered using Samsung Galaxy Grand Prime large screen smartphones using ODK Collect. All survey procedures were conducted in line with COVID-19 safety precautions, including the use of face masks, and were in line with Government of Bangladesh guidance at the time. Detailed information on the sociodemographic characteristics of all sampled individuals were collected using a structured survey instrument adapted from the WHO Stepwise tool26 and the 2014 Bangladesh Demographic and Health Survey.27 This was designed to measure the background demographic and socio-economic characteristics, lifestyle and behavioural risk factors, diabetes awareness indicators and health seeking behaviour and costs of care seeking among study participants. Fieldworkers measured blood pressure, blood glucose concentration, body weight, height, and waist and hip girth using standard methods. Blood pressure was measured using the OMRON HBP 1100 Professional Blood Pressure Monitor (Kyoto, Japan). Two measurements were taken at approximately 5-min intervals and the respondent's blood pressure obtained by averaging these measurements. Measurements of height, weight, and waist and hip girth were taken with light clothes without shoes. The weighing tools were calibrated daily by known weight. For height, the subject stood in erect posture vertically touching the occiput, back, hip, and heels on the wall while gazing horizontally in front and keeping the tragus and lateral orbital margin in the same horizontal plane. Waist girth was measured by placing a plastic tape horizontally midway between 12th rib and iliac crest on the mid-axillary line. Similarly, hip circumference was measured by taking the extreme end posteriorly and the symphysis pubis anteriorly. Blood glucose was measured using the One Touch Varioflex Glucometer (Lifescan, Inc., Milpitas, CA 95035) in whole blood obtained by finger prick from capillaries in the middle or ring finger after an over-night fast. All individuals then received a 75 g glucose load dissolved in approximately 250 ml of water and had a repeat capillary blood test within 5 min of 120 min post ingestion to determine glucose tolerance status and differentiate between individuals with intermediate hyperglycaemia and those with diabetes according to WHO criteria.15 Individuals who reported a prior medical diagnosis of diabetes were not required to provide fasting and 2-h blood glucose measures but instead provided a random blood glucose sample. Although capillary blood glucose concentrations may overestimate blood glucose concentrations compared to venous samples, the method is feasible and acceptable for epidemiological studies and any measurement inaccuracy would be consistent across study arms. Data were transferred from each data collectors’ tablet onto a laptop in the field every two days, by one of the field supervisors and gathered data were transferred from the laptop to the data manager in Dhaka once per week. Detected errors or requests for verification were sent back to the field team in Faridpur. ## Analysis We compared the prevalence of intermediate hyperglycaemia and T2DM between clusters allocated to PLA, mHealth and control arm in the DMagic trial. All analysis was by intention-to-treat at the individual and cluster level, adjusting for clustering where appropriate and wealth quintile derived from principal components analysis (as in the DMagic trial). The intention-to-treat population only includes non-pregnant adults aged ≥30 years who are permanently residing in the village in which they were surveyed. Participants with missing data on the primary outcomes were excluded from primary outcome analysis, in-line with the primary analysis in the DMagic trial. Estimates of the intervention effects are presented with $95\%$ confidence intervals. Analysis of primary outcomes was conducted by EF, who was blinded to intervention allocation, and results were shared with an independent trial steering committee before revealing allocation and proceeding with secondary outcome analyses. Prespecified secondary outcomes and explanatory analyses were based on complete data only, i.e. cases with missing data on any outcome were excluded from that analysis. Comparative analysis used random-effects logistic regression for binary outcomes and mixed-effects linear regression for continuous outcomes, each allowing for clustering and upazilla stratification. Continuous outcome measures with a skewed distribution were log-transformed before regression analysis. Given that distinctive types of hypertension are strongly age-, sex- and BMI-dependent and correlate with hyperglycaemia,28,29 our exploratory analysis of isolated diastolic, isolated systolic and PP were also adjusted for age, sex, BMI and diabetic status. No adjustments were made for the multiple statistical comparisons in this study on the basis that comparisons between trial arms and almost all outcomes replicate our a priori analysis plan for the DMagic trial and reflect the experimental design of the study. The explanatory outcomes, though not pre-specified as part of DMagic, were nonetheless defined as relevant outcomes based on process evaluation findings and prior to analysis of the 2021 data. All conducted comparisons are reported in this paper and our interpretation of results emphasises effect size, confidence intervals and consistency in intervention effects rather than focusing on p-values and arbitrary cut-off values of statistical significance. All analyses were done using STATA/SE version 15.1. ## Sensitivity In view of the clinical relevance of T2DM as an outcome in its own right (i.e., not combined with intermediate hyperglycaemia), we did a post-hoc analysis in which we assessed intervention effects on a diabetes only outcome. ## Ethics Written informed consent was obtained from all participants before data collection, or a thumb print for those unable to write. Ethical approvals for the DMagic trial and for this follow-up study were given by the University College London Research Ethics Committee (ref: $\frac{4766}{002}$ and ref: $\frac{4199}{007}$) and the Ethical Review Committee of the Diabetic Association of Bangladesh (ref: BADAS-ERC/EC/t5100246 and ref: BADAS-ERC/E/$\frac{19}{00276}$). ## Role of the funding source The funder of the study had no role in study design, data collection, analysis, interpretation or writing of this paper. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. ## Response rates Survey and/or anthropometric data were gathered from $\frac{1566}{1920}$ ($82\%$) of the random population sample (Fig. 1). The cross-sectional sample was similar in terms of sociodemographic characteristics between trial arms (Table 1). Non-responders were more likely to be men (213 ($23\%$) of 936 men vs 141 ($14\%$) of 984 women) and a similar pattern was observed across all arms. Reasons for non-response included migration (194 ($54.8\%$)), death (108 ($30.7\%$)), inability to locate (13 ($3.7\%$)), or illness preventing participation (8 ($2.3\%$)). Only 29 ($8.2\%$) of the randomly sampled individuals refused to participate in the study. Reasons for non-response were generally similar across study arms, although death as a reason for non-response was higher in the control arm ($31.3\%$ ($$n = 41$$)) and mHealth arm ($36.4\%$ ($$n = 40$$)) compared to the PLA arm ($23.9\%$ ($$n = 27$$)).Fig. 1DMagic trial & follow-up profile. Table 1Sociodemographic characteristics among random sample between trial arms in 2021.Sociodemographic parameterControlmHealthPLACluster level (based on 2017 data)Villages (Clusters)323232Average village population aged ≥30 years (sd)521 [189]551 [152]548 [225]Average number of households (sd)269 [97]282 [79]285 [112]Individual levelaAge30–39 years40–49 years50–59 years60–69 years70–100 years65 ($12.7\%$)194 ($37.9\%$)127 ($24.8\%$)75 ($14.7\%$)51 ($10.0\%$)64 ($12.6\%$)188 ($35.4\%$)134 ($25.2\%$)85 ($16.0\%$)55 ($10.4\%$)66 ($12.6\%$)184 ($35.2\%$)132 ($25.2\%$)81 ($15.5\%$)56 ($10.7\%$)SexMaleFemale238 ($46.5\%$)274 ($53.5\%$)225 ($42.4\%$)301 ($56.7\%$)260 ($49.7\%$)259 ($49.5\%$)EducationNonePrimarySecondaryTertiary231 ($45.1\%$)130 ($25.4\%$)149 ($29.1\%$)2 ($0.4\%$)238 ($44.8\%$)112 ($21.1\%$)171 ($32.2\%$)5 ($0.9\%$)239 ($45.7\%$)124 ($23.7\%$)155 ($29.6\%$)1 ($0.2\%$)IlliterateLiterateIlliterate218 ($42.6\%$)294 ($57.4\%$)234 ($44.1\%$)292 ($55.0\%$)221 ($42.3\%$)298 ($57.0\%$)Marital statusNot marriedbMarried75 ($14.7\%$)437 ($85.4\%$)95 ($17.9\%$)431 ($81.2\%$)61 ($11.7\%$)437 ($85.4\%$)ReligionOtherMuslim40 ($7.8\%$)472 ($92.2\%$)53 ($10.0\%$)473 ($88.1\%$)50 ($9.6\%$)469 ($89.7\%$)Respondent OccupationNo paid workManual labour/tradeNon-manual labour303 ($57.1\%$)167 ($31.5\%$)61 ($11.5\%$)309 ($58.2\%$)171 ($32.2\%$)46 ($8.7\%$)273 ($52.2\%$)168 ($32.1\%$)78 ($14.9\%$)Wealth quintileMost poorVery poorPoorLess poorLeast poor110 ($21.5\%$)95 ($18.6\%$)101 ($19.7\%$)114 ($22.3\%$)92 ($18.0\%$)122 ($23.0\%$)104 ($19.6\%$)115 ($21.7\%$)97 ($18.3\%$)88 ($16.6\%$)114 ($21.8\%$)91 ($17.4\%$)114 ($21.8\%$)85 ($16.3\%$)115 ($22.0\%$)aData missing for all parameters for 9 respondents ($0.6\%$) (5 individuals in mHealth arm and 4 individuals in the PLA arm) who participated in the physical and anthropometric measurements but not the interview survey.bIncluding never married, widowed, separated & divorced. Among the intermediate hyperglycaemia cohort of 2099 individuals, 1817 ($87\%$) participated in the 2021 interview survey and/or anthropometric measurement (Fig. 1). Individuals lost to follow-up were more likely to be men (136 ($18\%$) of 752 men vs 146 ($11\%$) of 1347 women) and leading reasons for loss to follow-up were death (133 ($47.2\%$)), migration (116 ($41.1\%$)) and inability to locate (11 ($3.9\%$)). Only 11 individuals ($3.9\%$) refused to participate. A similar pattern of loss to follow-up was observed across all arms. ## Primary outcomes No difference in the combined prevalence of T2DM and intermediate hyperglycaemia or the 5-year cumulative incidence of T2DM among the intermediate hyperglycaemia cohort was observed between intervention arms relative to control (Table 2).Table 22021 frequency, proportions and relative (odds ratio) and absolute (coefficient) effects and $95\%$ confidence interval comparing normoglycaemia and intermediate hyperglycaemia and diabetes according to WHO diagnostic criteria21 a) among the random survey population (outcome 1), and b) among the intermediate hyperglycaemia cohort (outcome 2). Results are adjusted for (i) the stratified, clustered design, and (ii) the stratified, clustered design and adjustment for household wealth quintile. Outcome 1: Population prevalence of intermediate hyperglycaemia and diabetesGlycaemic statusControlmHealthPLANormoglycaemic309 ($60.6\%$)312 ($59.1\%$)291 ($55.6\%$)Diabetic or intermediate hyperglycaemic201 ($39.4\%$)216 ($40.9\%$)232 ($44.4\%$)Total510 ($100.0\%$)528 ($100.0\%$)523 ($100.0\%$)Relative difference odds ratio ($95\%$ CI)ControlmHealthPLA(i) adjusted for stratified, clustered designReference1.07 (0.77, 1.48); $$p \leq 0.691.23$$ (0.89, 1.70); $$p \leq 0.21$$(ii) adjusted for (i) plus wealthReference1.08 (0.78, 1.51); $$p \leq 0.641.23$$ (0.87, 1.74); $$p \leq 0.24$$Absolute risk difference ($95\%$ CI)ControlmHealthPLA(i) adjusted for stratified, clustered designReference1.49 (−6.35, 9.33); $$p \leq 0.714.80$$ (−3.13, 12.7); $$p \leq 0.24$$(ii) adjusted for (i) plus wealthReference1.77 (−5.78, 9.33); $$p \leq 0.654.76$$ (−3.25, 12.8); $$p \leq 0.24$$Outcome 2: Five-year cumulative incidence among intermediate hyperglycaemic cohortGlycaemic statusControlmHealthPLANormoglycaemic272 ($43.7\%$)252 ($41.8\%$)263 ($45.2\%$)Intermediate hyperglycaemic237 ($38.0\%$)232 ($35.7\%$)208 ($35.7\%$)Diabetic114 ($18.3\%$)119 ($19.7\%$)111 ($19.1\%$)Total623 ($100.0\%$)603 ($100.0\%$)582 ($100.0\%$)Relative difference odds ratio ($95\%$ CI)ControlmHealthPLA(i) adjusted for stratified, clustered designReference1.07 (0.77, 1.48); $$p \leq 0.691.04$$ (0.78, 1.40); $$p \leq 0.77$$(ii) adjusted for (i) plus wealthReference1.09 (0.78, 1.52); $$p \leq 0.611.04$$ (0.77, 1.41); $$p \leq 0.81$$Absolute risk difference ($95\%$ CI)ControlmHealthPLA(i) adjusted for stratified, clustered designReference0.95 (−4.18, 6.08); $$p \leq 0.720.54$$ (−4.34, 5.42); $$p \leq 0.83$$(ii) adjusted for (i) plus wealthReference1.22 (−4.05, 6.50); $$p \leq 0.650.40$$ (−4.62, 5.42); $$p \leq 0.88$$ ## Secondary outcomes The prevalence of hypertension was lower in PLA clusters relative to control (adjusted odds ratio (aOR) ($95\%$ confidence interval): 0.73 (0.54, 0.97), $$p \leq 0.031$$) and individuals with a diagnosis of hypertension in PLA clusters were more than twice as likely to have controlled blood pressure relative to control clusters (aOR ($95\%$CI): 2.77 (1.34, 5.75), $$p \leq 0.0061$$) (Table 3). There was no evidence of an effect of either intervention on measures of overweight and obesity, self-rated health, or proportion of respondents participating in at least 150 min of physical activity per week (Table 3). Although knowledge and understanding of diabetes in terms of its causes, symptoms, complications, prevention and control was generally high in all arms, it was higher in PLA villages compared to control and improvements in knowledge observed in the mHealth arm at the end of the DMagic trial were no longer statistically different to control. Table 32021 frequency, proportions and relative (odds ratio) and absolute (coefficient) effects and $95\%$ confidence interval comparing pre-specified secondary outcomes adjusted for [1] the stratified, clustered design, and [2] the stratified, clustered design and adjustment for household wealth quintile. OutcomesAllocationCrude1Adjusted2PLAmHealthControlPLA vs ControlmHealth vs ControlPLA vs ControlmHealth vs ControlObjective physical measuresBlood pressureMean diastolic blood pressure (sd)74.2 (10.2)75.2 (10.9)74.6 (11.7)−0.36 (−2.24, 1.51); $$p \leq 0.710.47$$ (−1.34, 2.28); $$p \leq 0.61$$−0.32 (−2.24, 1.59); $$p \leq 0.740.49$$ (−1.37, 2.35); $$p \leq 0.61$$Mean systolic blood pressure (sd)121.4 (16.9)122.9 (18.8)123.2 (20.2)−1.72 (−4.63, 1.19); $$p \leq 0.25$$−0.32 (−3.07, 2.44); $$p \leq 0.82$$−1.68 (−4.66, 1.30); $$p \leq 0.27$$−0.16 (−3.00, 2.68); $$p \leq 0.91$$Hypertension (%)109 ($20.8\%$)135 ($25.6\%$)136 ($26.7\%$)0.72 (0.54, 0.96); $$p \leq 0.0260.94$$ (0.71, 1.24); $$p \leq 0.670.73$$ (0.54, 0.97); $$p \leq 0.0310.96$$ (0.72, 1.26); $$p \leq 0.75$$Hypertension control (%)41 ($66.1\%$)41 ($47.7\%$)34 ($42.5\%$)2.69 (1.33, 5.44); $$p \leq 0.00611.27$$ (0.68, 2.37); $$p \leq 0.462.77$$ (1.34, 5.75); $$p \leq 0.00611.31$$ (0.69, 2.51); $$p \leq 0.41$$Overweight & obesityMean Body Mass Index (BMI) (sd)22.4 (3.8)22.3 (3.8)22.4 (3.7)0.02 (−0.45, 0.48); $$p \leq 0.95$$−0.08 (−0.55, 0.39); $$p \leq 0.730.01$$ (−0.43, 0.45); $$p \leq 0.97$$−0.01 (−0.45, 0.43); $$p \leq 0.97$$Overweight or obese (%)214 ($40.9\%$)214 ($40.5\%$)209 ($41.0\%$)0.99 (0.76, 1.30); $$p \leq 0.960.99$$ (0.77, 1.26); $$p \leq 0.931.00$$ (0.77, 1.31); $$p \leq 0.981.03$$ (0.79, 1.34); $$p \leq 0.84$$Abdominal obesity (%)163 ($62.9\%$)205 ($68.1\%$)193 ($70.7\%$)0.68 (0.44, 1.07); $$p \leq 0.0970.86$$ (0.55, 1.35); $$p \leq 0.520.69$$ (0.43, 1.12); $$p \leq 0.140.90$$ (0.57, 1.43); $$p \leq 0.65$$Interview survey measuresQuality of life & wellbeingMedian Self Rated Health (IQR)80 (70–95)80 (70–95)80 (70–90)0.00 (−0.04, 0.04); $$p \leq 0.99$$−0.03 (−0.08, 0.01); $$p \leq 0.140.00$$ (−0.04, 0.04); $$p \leq 0.99$$−0.03 (−0.08, 0.01); $$p \leq 0.16$$Diabetes knowledgeAbility to report one or more valid causes of diabetes (%)459 ($88.4\%$)409 ($77.8\%$)382 ($74.6\%$)3.95 (1.39, 11.24); $$p \leq 0.0101.14$$ (0.49, 2.62); $$p \leq 0.763.97$$ (1.38, 11.43); $$p \leq 0.0111.17$$ (0.50, 2.70); $$p \leq 0.72$$Ability to report one or more valid symptoms of diabetes (%)473 ($91.1\%$)440 ($83.7\%$)410 ($80.0\%$)4.43 (1.55, 12.62); $$p \leq 0.00541.27$$ (0.61, 2.62); $$p \leq 0.534.42$$ (1.55, 12.62); $$p \leq 0.00551.31$$ (0.63, 2.71); $$p \leq 0.47$$Ability to report one or more valid complications of diabetes (%)447 ($86.1\%$)398 ($75.7\%$)364 ($71.1\%$)5.08 (1.63, 15.79); $$p \leq 0.00501.40$$ (0.59, 3.32); $$p \leq 0.455.19$$ (1.64, 16.35); $$p \leq 0.00491.42$$ (0.60, 3.39); $$p \leq 0.43$$Ability to report one or more valid ways to prevent diabetes (%)472 ($90.9\%$)451 ($85.7\%$)416 ($81.3\%$)4.05 (1.45, 11.30); $$p \leq 0.00751.30$$ (0.66, 2.56); $$p \leq 0.454.02$$ (1.44, 11.22); $$p \leq 0.00801.34$$ (0.68, 2.62); $$p \leq 0.40$$Ability to report one or more valid ways to control diabetes (%)475 ($91.5\%$)466 ($88.6\%$)439 ($85.7\%$)3.37 (1.28, 8.85); $$p \leq 0.0141.24$$ (0.70, 2.19); $$p \leq 0.463.27$$ (1.23, 8.68); $$p \leq 0.0171.27$$ (0.72, 2.23); $$p \leq 0.41$$Physical ActivityAverage of 150 min or more doing physical activity per week (%)291 ($56.1\%$)258 ($49.1\%$)289 ($56.5\%$)0.99 (0.63, 1.56); $$p \leq 0.970.71$$ (0.45, 1.14); $$p \leq 0.160.99$$ (0.63, 1.56); $$p \leq 0.970.72$$ (0.45, 1.14); $$p \leq 0.16$$Among individuals with diabetesDiabetes awareness & careSelf-awareness of diabetic status (%)a39 ($42.4\%$)28 ($37.3\%$)26 ($39.4\%$)1.05 (0.54, 2.03); $$p \leq 0.900.86$$ (0.43, 1.74); $$p \leq 0.681.10$$ (0.55, 2.23); $$p \leq 0.790.85$$ (0.41, 1.73); $$p \leq 0.65$$Diabetes control (%) (random blood glucose<11.1 mmol/l)b22 ($56.4\%$)21 ($72.4\%$)17 ($65.4\%$)0.71 (0.25, 2.01); $$p \leq 0.521.28$$ (0.38, 4.31); $$p \leq 0.690.53$$ (0.17, 1.72); $$p \leq 0.291.40$$ (0.32, 6.06); $$p \leq 0.65$$Receipt of professional treatment or advice for diabetes (%)b37 ($94.9\%$)25 ($89.3\%$)23 ($85.2\%$)3.42 (0.56, 20.81); $$p \leq 0.181.89$$ (0.17, 21.62); $$p \leq 0.613.54$$ (0.54, 23.35); $$p \leq 0.192.18$$ (0.11, 42.74); $$p \leq 0.61$$Minimum monthly blood glucose testing (%)bc19 ($48.7\%$)10 ($35.7\%$)11 ($40.7\%$)1.31 (0.46, 3.80); $$p \leq 0.610.83$$ (0.26, 2.61); $$p \leq 0.750.91$$ (0.28, 2.93); $$p \leq 0.880.68$$ (0.19, 2.41); $$p \leq 0.55$$Diabetes-related complications (%)bd29 ($74.4\%$)21 ($75.0\%$)20 ($74.1\%$)0.76 (0.21, 2.67); $$p \leq 0.660.80$$ (0.07, 9.12); $$p \leq 0.860.35$$ (0.07, 1.70); $$p \leq 0.190.19$$ (0.00, 168.81); $$p \leq 0.63$$aAmong those identified as diabetic by objective blood glucose test ($$n = 233$$).bAmong individuals with self-reported diabetes ($$n = 94$$).cMissing information for 3 individuals (2 mHealth; 1 control).dMissing information on complications for 4 individuals (2 mHealth; 2 control). Among the 233 individuals with blood glucose readings indicating T2DM, 93 ($39.9\%$) reported a prior diagnosis of diabetes, with no difference in awareness between trial arms. Among a total of 94 individuals reporting a prior diagnosis of diabetes (1 had no blood glucose reading), approximately two thirds ($$n = 60$$, ($63.8\%$)) had random blood glucose levels lower than 11.1 mmol/l, indicating diabetic control. The proportion of people with controlled diabetes was lower in the PLA arm ($56.4\%$) compared to mHealth ($72.4\%$) and control ($65.4\%$), though reported receipt of professional treatment and advice, and at least monthly blood glucose monitoring was higher in the PLA arm. None of the observed numerical differences in diabetes care indicators were statistically significant. ## Exploratory outcomes Exploratory analysis of behavioural and health outcomes indicated that whilst individuals living in PLA clusters were no more likely to participate in brisk walking activities compared to individuals in control clusters, they spent on average $31\%$ (approximately 55 min) longer doing this activity per week (Table 4). No significant differences in dietary habits were observed and although mean ADS score was lower in the PLA arm compared to control (indicating more positive appraisal), this was not statistically significant. On average, individuals in the PLA arm scored lower on the PHQ-2 screening tool, however, of those who did screen positive and who completed the PHQ-9 tool, individuals in the PLA arm scored significantly higher than those in the control arm, indicating more severe depressive symptoms. Table 42021 frequency, proportions and relative (odds ratio) and absolute (coefficient) effects and $95\%$ confidence interval comparing exploratory outcomes adjusted for [1] the stratified, clustered design, [2] the stratified, clustered design and adjustment for household wealth quintile, and [3] (hypertensive outcomes only) the stratified, clustered design and adjustment for household wealth quintile, age group, sex, diabetic status and BMI.OutcomesCrude1Adjusted2Adjusted3Community PLAControlPLA vs ControlPLA vs ControlPLA vs ControlWalkingParticipates in brisk walking (%)93 ($17.9\%$)79 ($15.4\%$)1.43 (0.80, 2.55); $$p \leq 0.231.40$$ (0.79, 2.46); $$p \leq 0.23$$Median (IQR) time spent brisk walking per week240 (180–360)180 (120–300)0.27 (0.04, 0.49); $$p \leq 0.0190.27$$ (0.02, 0.51); $$p \leq 0.031$$DietMean 24 Hour Dietary Diversity Score (DDS) (sd)6.72 (1.86)6.61 (1.93)0.10 (−0.22, 0.42); $$p \leq 0.540.08$$ (−0.22, 0.38); $$p \leq 0.59$$No added sugar to foods in previous 24 h178 ($39.6\%$)154 ($37.6\%$)1.07 (0.78, 1.47); $$p \leq 0.651.08$$ (0.79, 1.48); $$p \leq 0.62$$No added salt to foods145 ($27.9\%$)125 ($24.4\%$)1.19 (0.86, 1.65); $$p \leq 0.311.17$$ (0.84, 1.64); $$p \leq 0.34$$Mean (sd) monthly household oil consumption4.62 (1.32)4.74 (1.47)−0.12 (−0.36, 0.11); $$p \leq 0.31$$−0.13 (−0.36, 0.10); $$p \leq 0.25$$Appraisal of Diabetes Scale (ADS)Mean (SD) ADS score among known diabetics ($$n = 97$$)12.2 (3.37)13.9 (4.93)−1.39 (−3.53, 0.76); $$p \leq 0.21$$−0.61 (−2.84, 1.62); $$p \leq 0.59$$Depression & anxietyMedian (IQR) PHQ2 score (depression screening)1 (0–2)1 (0–2)−0.12 (−0.23, −0.01); $$p \leq 0.026$$−0.12 (−0.22, −0.01); $$p \leq 0.028$$Mean (SD) PHQ9 score (among PHQ2 screen positive, $$n = 142$$)13.3 (4.8)11.1 (4.6)2.21 (0.08, 4.34); $$p \leq 0.0422.27$$ (0.12, 4.43); $$p \leq 0.039$$Median (IQR) GAD7 score (anxiety)3 (1–6)3 (1–6)−0.07 (−0.23, 0.09); $$p \leq 0.41$$−0.06 (−0.21, 0.09); $$p \leq 0.44$$Blood pressure measuresIsolated systolic hypertensiona37 ($7.1\%$)51 ($10.0\%$)0.69 (0.42, 1.16); $$p \leq 0.160.70$$ (0.42, 1.17); $$p \leq 0.180.62$$ (0.36, 1.06); $$p \leq 0.081$$Isolated diastolic hypertensiona7 ($1.3\%$)15 ($2.9\%$)0.44 (0.18, 1.10); $$p \leq 0.0800.42$$ (0.17, 1.05); $$p \leq 0.0630.41$$ (0.16, 1.04); $$p \leq 0.060$$Mean pulse pressure (sd)47.2 (12.0)48.6 (13.8)−1.39 (−3.15, 0.38); $$p \leq 0.13$$−1.38 (−3.16, 0.40); $$p \leq 0.13$$−1.78 (−3.34, −0.22); $$p \leq 0.026$$aDenominator is all non-cases. Blood pressure-related measures of cardiovascular risk suggest a possible positive advantage in PLA clusters relative to control, particularly in terms of mean pulse pressure and with adjustment for known correlates of age, sex, BMI and diabetic status. There was no evidence of mHealth intervention effect on any of the exploratory outcomes (supplementary Table S2). As per our DMagic analysis, and in view of the clinical relevance of T2DM as an outcome in its own right (i.e., not combined with intermediate hyperglycaemia), we did a post-hoc analysis in which we assessed intervention effects on a diabetes-only outcome. The adjusted odds of diabetes was $45\%$ higher in PLA clusters compared to control (1.45 (1.03, 2.04); $$p \leq 0.034$$), and no significant effect was observed in mHealth clusters relative to control (1.13 (0.78, 1.64); $$p \leq 0.51$$). ## Discussion Our five-year post-randomisation follow-up of the DMagic cluster randomised controlled trial shows that whilst knowledge about diabetes remains significantly higher in PLA clusters relative to control, intervention effects on blood glucose outcomes are no longer observed. Improvements in knowledge among the mHealth clusters that we measured in 2018 were also no longer observable in 2021. The DMagic interventions did not specifically target high-risk individuals, but rather employed a broad population-level approach to prevention and control and our trial design similarly assessed population level outcomes among individuals residing in study clusters rather than just those directly exposed to and engaged with the interventions. We know from process evaluation that the PLA community mobilisation intervention stimulates change at the individual, household and community levels that enable, reinforce and amplify impacts, such that effects are observed even in those who do not directly engage with the intervention.30 It is therefore challenging to directly compare our findings to those from other targeted diabetes intervention follow-up studies and furthermore, there is a lack of such studies from LMICs. Nevertheless, there is evidence from high-income settings that lifestyle modification interventions among high-risk groups can be promising long-term diabetes prevention strategies. However, the need for some degree of maintenance intervention to observe prolonged effects is noted.10 All DMagic intervention activities ended in 2017 and there has been no further support or maintenance to PLA groups or mHealth since then. While there was a community hand-over process for groups, unpublished data from our surveys indicates that none of the 122 PLA groups established in DMagic met later than 2018, and we are not aware of any other population-based interventions for diabetes prevention and control in our study areas since DMagic. Previous follow-up of PLA for maternal and neonatal health suggests that groups may be sustainable31 and so more understanding is needed on what aspects of our study context, diabetes-focus and handover might have influenced sustainability of DMagic PLA groups. Despite large impacts of PLA on the prevalence of diabetes and intermediate hyperglycaemia and the two-year incidence of diabetes among the intermediate hyperglycaemia cohort in 2018, these primary outcomes did not differ significantly between the three randomised groups in 2021. This finding differs from diabetes intervention follow-up studies in China8 and Finland,9 which showed that reduction in diabetes incidence remained for several years after the period of active intervention. However, these studies were in high-risk individuals. Similar to the Diabetes Prevention Program Outcome Study (DPPOS),32 our findings may be attributable to a fall in the incidence of diabetes and intermediate hyperglycaemia in the control and mHealth arms due to the majority of individuals susceptible to intermediate hyperglycaemia and diabetes developing these outcomes during the initial DMagic trial, leaving a reduced number at risk in subsequent years. Our data might also indicate a rebound effect, whereby our PLA intervention did not prevent intermediate hyperglycaemia and diabetes, but rather delayed the onset of hyperglycaemia in susceptible individuals. We do not have reliable data on the date of onset of intermediate hyperglycaemia or diabetes in our study but can assume a substantial increase in incidence in PLA clusters post-2018 to result in comparable five-year (2016–2021) incidence in all trial arms. Possible delayed diabetes and a potential survival effect of DMagic PLA interventions (i.e. those with later diabetes surviving longer) are also plausible explanations for observed primary outcomes and especially the higher prevalence of diabetes only outcomes in the PLA arm in the absence of observable changes in diabetes risk such as BMI or risk behaviours. Evidence from the Da Qing Diabetes Prevention Outcome Study of lifestyle interventions among high-risk individuals in China indicates that a delay in diabetes onset is associated with fewer cardiovascular events, lower incidence of microvascular complications, fewer cardiovascular disease deaths, fewer all-cause deaths and an average increase of life expectancy.33 Hyperglycaemia is a key modifiable risk factor for CVD risk and so effective reduction of hyperglycaemia, even if temporary, may have a positive effect on CVD risk.6 Further, increased risk of CVD associated with diabetes is augmented with coexistent hypertension, thus lower blood pressure and controlled hypertension promotes vascular health and may be especially important in reducing microvascular and macrovascular complications of diabetes. We observed improvements in hypertension, hypertension control and pulse pressure in PLA clusters relative to control in our follow-up. The exploratory nature of our analysis of isolated pressures and pulse pressure and multiple hypothesis testing notwithstanding, the role of blood pressure lowering to improve prognosis in T2DM is well-established24,29,34 and these observations are potentially important if they translate to lower cardiovascular risk in PLA communities. Existing population attributable risk estimates suggest that even small decreases in population blood pressure can result in large decreases to overall cardiovascular risk in the population.35 Though observed absolute differences in mean diastolic and systoloic blood pressure in our study are relatively small between study arms, the lower mean and smaller standard deviation of blood pressure measures in the PLA arm compared to control infer a significantly decreased probability that individuals in the PLA arm will reach the diastolic or systolic thresholds for hypertension. Indeed, our observed relative reduction of odds of hypertension by $27\%$ and more than two-fold increase in control of blood pressure among individuals with hypertension associated with PLA clusters suggest positive and plausible lasting impacts of the PLA intervention on blood pressure. Further, our observed, though not statistically significant, lower odds of isolated blood pressures and reduction of −1.78 mmHg pulse pressure (when adjusted for age, sex, wealth, BMI and diabetes status) could convey meaningful reductions in population risk of ischaemic heart disease and stroke.36 Though our DMagic interventions did not specifically target blood pressure, raised blood pressure and raised blood glucose share several common risk factors and so many of the possible PLA intervention mechanisms that reduce diabetes risk could plausibly also have beneficial effects on blood pressure. However, despite the large observable effects on blood glucose in 2018, intervention effects on hypertension7 and measures of blood pressure (retrospective analysis in supplementary Table S3) were not observed in our 2018 data. There may be several reasons for this, including a possible inertia in blood pressure that means that changes resulting from intervention mechanisms take longer to have an effect. Alternatively (or in addition), the observed improvements in blood pressure might themselves be mediated by improvements in blood glucose and so may not be expected to occur simultaneously with reductions in intermediate hyperglycaemia and diabetes. We do not have measures of the quantity of salt consumed by study participants or the relative or absolute changes in the amount of salt consumed, but small reductions in salt being added to food were observed in PLA clusters relative to control. Reduction of salt consumption is one of the most effective ways to reduce blood pressure in populations and individuals and may have contributed to the changes in blood pressure we observed.37 Effective reduction of population salt consumption is likely to require a combination of targeted salt reduction strategies as well as community interventions – such as PLA – that address individual and contextual factors influencing dietary salt use. The Framingham Heart Study reported that, with increasing age, a shift from diastolic to systolic hypertension and then to pulse pressure was a predictor of coronary heart disease.38 Isolated diastolic hypertension is a relatively uncommon hypertension phenotype but is associated with increased stroke, heart disease and other sequalae of hypertension.25 Pulse pressure is a marker for increased large arterial stiffness and is a major independent predictor of cardiovascular mortality and atrial fibrillation39 and as little as 10 mmHg increase in pulse pressure can increase cardiovascular risk by approximately $20\%$.40 Although mean pulse pressure observed in our study is within a normal range in PLA and control clusters and the epidemiological and clinical significance of the small observed difference is uncertain, further follow-up would be valuable. As in the original DMagic trial analysis, there are no major differences in behavioural outcomes. The overall number of individuals living with diabetes who were aware of their status was higher than in 2018, but still low and so comparisons of diabetes-specific behaviours lack statistical power. Nevertheless, taken together, the greater knowledge of diabetes symptoms, prevention and control in PLA clusters and numerically (though not statistically significant) higher levels of awareness, receipt of professional treatment/advice and regular blood glucose monitoring, albeit with lower levels of control, and the lower (more positive) ADS score suggest there may be some differences in how people live and experience diabetes in PLA clusters compared to control and mHealth clusters. The fact that, despite higher prevalence of diabetes in PLA clusters relative to control, measures of self-rated health and self-reported complications of diabetes did not differ between arms may further indicate either more recent progression to diabetes or better self-management in PLA clusters, although our measure of diabetes control contradicts the latter. The proportion of the population engaging in an average of at least 150 min of physical activity per week was lower in our 2021 survey compared to 2018, with no difference between study arms. The reason for this decline in physical activity is unknown but may plausible be related to a global decline in physical activity associated with the COVID-19 pandemic and imposed restrictions, as observed in other South Asian settings.41,42 Walking was identified as a critical strategy of the PLA intervention in DMagic, with the intervention addressing socio-cultural barriers to walking for exercise.12,43 We therefore conducted exploratory analysis of brisk walking as an activity and observed that, although the proportion of people engaging in this activity did not differ between PLA and control arms, the time spent brisk walking was almost 1 hour longer per week in PLA villages. Physical activity is known to have favourable effects on cardiometabolic health and, though the optimal frequency and intensity of physical activity is not universally defined, brisk walking has been identified as an appropriate and accessible form a physical activity that can be practiced by individuals and groups without cost and with low risk of injury.44 With the possible exception of time spent engaged in brisk walking, there was no evidence of mHealth intervention effect on any of the exploratory outcomes (supplementary Table S2). This is perhaps not surprising given that primary and secondary outcomes did not change in the mHealth arm in DMagic. The possible effect on walking time may be spurious given it is inconsistent with the absence of other mHealth intervention effects outcomes. A limitation of our study is that our sample size, though large, was designed to assess primary outcomes and several of our analyses of secondary and exploratory outcomes lack statistical power. Like many diabetes intervention studies, our assessment is relatively short-term, lacks intermediary measures of outcomes since the end of intervention and focuses on blood sugar, self-reported behaviours and relatively short-term cardiometabolic risk markers. Nevertheless, high response and follow-up rates, rigorous field methods, including fasting and 75 g oral glucose tolerance tests among a representative population-based sample of people in rural Faridpur are strengths of our study. Further, the robust design of the original DMagic trial, including cluster randomisation of 96 villages with limited contamination between clusters and the intention-to-treat analysis of this follow-up observational study enhance validity of our findings. Finally, it is important to note we did not apply any statistical corrections for multiple hypothesis testing in our analysis on the basis that the comparisons were by in lagre pre-specified and part of our original experimental design. Nevertheless, interpretation of results should focus on effect size and confidence intervals and consistency of intervention effects across outcomes rather than on concepts of absolute statistical significance. Adult metabolic health is a complex interaction of blood glucose, blood pressure, obesity, sex and age. Delayed hyperglycaemia and small changes in physical activity, dietary and diabetes-related behaviours associated with exposure to DMagic PLA community mobilisation intervention may cumulate and contribute to positive impacts on blood pressure, an important markers of cardiovascular risk, five years after randomisation. By addressing broad cultural and societal influences of cardiometabolic risk, whole-population, community-based awareness and lifestyle interventions are likely to be cost-effective strategies to reach large groups of people with potential to affect the entire distribution of disease risk, even if only by a small degree, to affect the proportion of those at risk.6 It is likely that sustained changes in social norms and associated benefits of these require a strong focus on maintenance, which in the context of DMagic, could include strategies for intermittent follow-up, incentivisation to groups and remote support, including using digital health technologies. Interventions that prevent the onset of diabetes, even if only temporarily, should remain a priority for populations with a high burden of risk since even short-term delay may postpone diabetes related complications and costly care. Longer-term follow-up are needed to fully understand lasting intervention effects on diabetes onset, cardiovascular complications and mortality. ## Contributors EF, project Principal Investigator, led in the design of the study, conducted statistical analyses and drafted the manuscript. CK provided technical coordination of the survey and data management process and contributed to analysis and interpretation. NA contributed to intervention development, survey methods and interpretation. SKS coordinated quantitative data collection activities, data management and interpretation. JM led the process evaluation component of the study and contributed to intervention development and interpretation of findings. MP contributed to survey design and interpretation. AK contributed to project management, survey and intervention coordination and interpretation of study findings. TN developed and coordinated the implementation of PLA group activities. HHB contributed to survey design and interpretation. AKAK provided technical oversight of all project activities and facilitated community engagement and intervention development activities. KA coordinated project activities in Bangladesh, co-led the project and contributed to intervention development and study design. All authors have reviewed and contributed to the reporting of study findings in this paper. ## Data sharing statement De-identified data collected for this study and a data dictionary are available from the corresponding author on reasonable request. ## Declaration of interests Recipients of funding for this work were Principal Investigator Fottrell and Co-investigators Azad, Khan, Kuddus, Haghparast-Bidgoli, Morrison and King. Co-authors Ahmed, Shaha, Pires and Nahar were employed on the project using project funding. The authors declare no competing interests. ## Supplementary data Abstract Bangla Supplementary Tables ## References 1. Cousin E., Duncan B.B., Stein C.. Diabetes mortality and trends before 25 years of age: an analysis of the global burden of disease study. Lancet Diabetes Endocrinol (2019) 10 2. Lin X., Xu Y., Pan X.. Global, regional, and national burden and trend of diabetes in 195 countries and territories: an analysis from 1990 to 2025. Sci Rep (2020) 10 3. Akhtar S., Nasir J.A., Sarwar A.. Prevalence of diabetes and pre-diabetes in Bangladesh: a systematic review and meta-analysis. BMJ Open (2020) 10 4. Fottrell E., Ahmed N., Shaha S.K.. Distribution of diabetes, hypertension and non-communicable disease risk factors among adults in rural Bangladesh: a cross-sectional survey. BMJ Global Health (2018) 3 5. Non-communicable diseases: a brief guide to non-communicable diseases (NCDs) and their impact globally and in Bangladesh. (2022) 6. Schwarz P.E.H., Timpel P., Harst L.. Blood sugar regulation for cardiovascular health promotion and disease prevention: JACC health promotion series. J Am Coll Cardiol (2018) 72 1829-1844. PMID: 30286928 7. Fottrell E., Ahmed N., Morrison J.. Community groups or mobile phone messaging to prevent and control type 2 diabetes and intermediate hyperglycaemia in Bangladesh (DMagic): a cluster-randomised controlled trial. Lancet Diabetes Endocrinol (2019) 7 200-212. PMID: 30733182 8. Li G., Zhang P., Wang J.. The long-term effect of lifestyle interventions to prevent diabetes in the China Da Qing Diabetes Prevention Study: a 20-year follow-up study. Lancet (2008) 371 1783-1789. PMID: 18502303 9. Lindström J., Ilanne-Parikka P., Peltonen M.. Sustained reduction in the incidence of type 2 diabetes by lifestyle intervention: follow-up of the Finnish Diabetes Prevention Study. Lancet (2006) 368 1673-1679. PMID: 17098085 10. Haw J.S., Galaviz K.I., Straus A.N.. Long-term sustainability of diabetes prevention approaches: a systematic review and meta-analysis of randomized clinical trials. JAMA Intern Med (2017) 177 1808-1817. PMID: 29114778 11. Bangladesh Bureau of Statistics. (2013) 12. Morrison J., Akter K., Jennings H.M.. Implementation and fidelity of a participatory learning and action cycle intervention to prevent and control type 2 diabetes in rural Bangladesh. Glob Health ResPolicy (2019) 4 19 13. Fottrell E., Jennings H., Kuddus A.. The effect of community groups and mobile phone messages on the prevention and control of diabetes in rural Bangladesh: study protocol for a three-arm cluster randomised controlled trial. Trials (2016) 17 600. PMID: 27993166 14. Jennings H.M., Morrison J., Akter K.. Developing a theory-driven contextually relevant mHealth intervention. Glob Health Action (2019) 12 15. 15World Health OrganizationInternational Diabetes FederationDefinition and diagnosis of diabetes mellitus and intermediate hyperglycaemia: report of a WHO/IDF consultation. Geneva, Switzerland2006. (2006) 16. Mannell J., Davis K., Akter K.. Visual participatory analysis: a qualitative method for engaging participants in interpreting the results of randomized controlled trials of health interventions. J Mix Methods Res (2021) 15 18-36. PMID: 33456409 17. Carey M.P., Jorgensen R.S., Weinstock R.S.. Reliability and validity of the appraisal of diabetes scale. J Behav Med (1991) 14 43-50. PMID: 2038044 18. Nair R., Kachan P.. Outcome tools for diabetes-specific quality of life: study performed in a private family practice clinic. Can Fam Physician (2017) 63 e310-e315. PMID: 28615409 19. Levis B., Sun Y., He C.. Accuracy of the PHQ-2 alone and in combination with the PHQ-9 for screening to detect major depression: systematic review and meta-analysis. JAMA (2020) 323 2290-2300. PMID: 32515813 20. Levis B., Benedetti A., Thombs B.D.. Accuracy of Patient Health Questionnaire-9 (PHQ-9) for screening to detect major depression: individual participant data meta-analysis. BMJ (2019) 365 l1476. PMID: 30967483 21. Naher R., Rabby M.R.A., Sharif F.. Validation of patient health questionnaire-9 for assessing depression of adults in Bangladesh. Dhaka Univ J Biol Sci (2021) 30 275-281 22. Spitzer R.L., Kroenke K., Williams J.B.W., Löwe B.. A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch Intern Med (2006) 166 1092-1097. PMID: 16717171 23. Dhira T.A., Rahman M.A., Sarker A.R., Mehareen J.. Validity and reliability of the Generalized Anxiety Disorder-7 (GAD-7) among university students of Bangladesh. PLoS One (2021) 16 24. Reusch J.E.B., Wang C.C.L.. Cardiovascular disease in diabetes: where does glucose fit in?. J Clin Endocrinol Metab (2011) 96 2367-2376. PMID: 21593112 25. Mahajan S., Zhang D., He S.. Prevalence, awareness, and treatment of isolated diastolic hypertension: insights from the China PEACE million persons project. J Am Heart Assoc (2019) 8 26. Bonita R., de Courten M., Dwyer T., Jamrozik K., W R.. (2001) 27. Research NIoP Associates. (2016) 28. Xie K., Gao X., Bao L., Shan Y., Shi H., Li Y.. The different risk factors for isolated diastolic hypertension and isolated systolic hypertension: a national survey. BMC Publ Health (2021) 21 1672 29. Os I., Gudmundsdottir H., Kjeldsen S.E., Oparil S.. Treatment of isolated systolic hypertension in diabetes mellitus type 2. Diabetes Obes Metabol (2006) 8 381-387 30. Morrison J., Akter K., Jennings H.M.. Participatory learning and action to address type 2 diabetes in rural Bangladesh: a qualitative process evaluation. BMC Endocr Disord (2019) 19 118. PMID: 31684932 31. Sondaal A.E.C., Tumbahangphe K.M., Neupane R., Manandhar D.S., Costello A., Morrison J.. Sustainability of community-based women's groups: reflections from a participatory intervention for newborn and maternal health in Nepal. Community Dev J (2019) 54 731-749. PMID: 31885401 32. Knowler W.C., Fowler S.E., Hamman R.F.. 10-year follow-up of diabetes incidence and weight loss in the Diabetes Prevention Program Outcomes Study. Lancet (2009) 374 1677-1686. PMID: 19878986 33. Gong Q., Zhang P., Wang J.. Morbidity and mortality after lifestyle intervention for people with impaired glucose tolerance: 30-year results of the Da Qing Diabetes Prevention Outcome Study. Lancet Diabetes Endocrinol (2019) 7 452-461. PMID: 31036503 34. Petrie J.R., Guzik T.J., Touyz R.M.. Diabetes, hypertension, and cardiovascular disease: clinical insights and vascular mechanisms. Can J Cardiol (2018) 34 575-584. PMID: 29459239 35. Whelton P.K.. Epidemiology and the prevention of hypertension. J Clin Hypertens (2004) 6 636-642 36. Husnu X., Murat B., Murat C., Ibrahim T.. Current approach to isolated diastolic hypertension. Arch Clin Hypertension (2020) 29-31 37. He F.J., Li J., MacGregor G.A.. Effect of longer term modest salt reduction on blood pressure: cochrane systematic review and meta-analysis of randomised trials. BMJ Br Med J (Clin Res Ed) (2013) 346 f1325 38. Franklin S.S., Larson M.G., Khan S.A.. Does the relation of blood pressure to coronary heart disease risk change with aging? The Framingham Heart Study. Circulation (2001) 103 1245-1249. PMID: 11238268 39. Tokitsu T., Yamamoto E., Hirata Y.. Clinical significance of pulse pressure in patients with heart failure with preserved left ventricular ejection fraction. Eur J Heart Fail (2016) 18 1353-1361. PMID: 27197000 40. Homan T.D., Bordes S., Cichowski E.. (2022) 41. Ghosh A., Arora B., Gupta R., Anoop S., Misra A.. Effects of nationwide lockdown during COVID-19 epidemic on lifestyle and other medical issues of patients with type 2 diabetes in north India. Diabetes Metabol Syndr (2020) 14 917-920 42. Kusuma D., Pradeepa R., Khawaja K.I.. Low uptake of COVID-19 prevention behaviours and high socioeconomic impact of lockdown measures in South Asia: evidence from a large-scale multi-country surveillance programme. SSM Popul Health (2021) 13 43. Morrison J., Jennings H., Akter K.. Gendered perceptions of physical activity and diabetes in rural Bangladesh: a qualitative study to inform mHealth and community mobilization interventions. WHO South East Asia J Public Health (2019) 8 104-111. PMID: 31441446 44. Montesi L., Moscatiello S., Malavolti M., Marzocchi R., Marchesini G.. Physical activity for the prevention and treatment of metabolic disorders. Intern Emerg Med (2013) 8 655-666. PMID: 23657989
--- title: Association of High Normal Body Weight in Youths With Risk of Hypertension authors: - Corinna Koebnick - Margo A. Sidell - Xia Li - Susan J. Woolford - Beatriz D. Kuizon - Poornima Kunani journal: JAMA Network Open year: 2023 pmcid: PMC10015306 doi: 10.1001/jamanetworkopen.2023.1987 license: CC BY 4.0 --- # Association of High Normal Body Weight in Youths With Risk of Hypertension ## Key Points ### Question How are body weight and change in body weight over time associated with the risk of hypertension in youths at the higher end of normal weight? ### Findings In this cohort study including more than 800 000 youths, weight change expressed as a change in the distance from the median BMI for age per year was associated with a change in hypertension risk independent of baseline weight classes. However, the risk associated with weight change was higher in youths living with low to high normal weight and overweight than in youths living with severe obesity, indicating a plateau in the association between weight gain and risk of hypertension. ### Meaning These findings suggest that even modest elevations in the BMI-for-age percentile in the upper range of normal weight may confer an increased risk of hypertension among children, which may further increases with excess weight gain over time. ## Abstract This cohort study of youths in southern California aged 3 to 17 years explores the risk of hypertension associated with high normal weight at baseline and differentiates between the risk of hypertension associated with baseline body weight and the risk associated with additional gain in body weight over time. ### Importance Ample evidence links obesity to hypertension in youths. However, the association of high normal body mass index (BMI) with obesity and the interaction with different weight trajectories are not well understood. ### Objective To examine the hypertension risk associated with high normal BMI for age and different weight trajectories in youths. ### Design, Setting, and Participants This retrospective cohort study assessed 801 019 youths aged 3 to 17 years in an integrated health care system in Southern California from January 1, 2008, to February 28, 2015, with a maximum follow-up of 5 years from January 1, 2008, to February 28, 2020. Data analysis was performed from 2018 to 2022. ### Exposures Youths were compared by first available (baseline) sex-specific BMI for age and change in the distance to the median BMI for age during the 5-year follow-up. ### Main Outcomes and Measures Cox proportional hazards regression models with age as a time scale to assess hypertension risk (based on 2017 Blood Pressure Guidelines by the American Academy of Pediatrics from 3 consecutive independent visits), adjusted for sex, race and ethnicity, socioeconomic status, baseline year, and birth year. ### Results A total of 801 019 youths (mean [SD] age, 9.4 [4.6] years; 409 167 [$51.1\%$] female]; 59 399 [$7.4\%$] Asian and Pacific Islanders, 65 712 [$8.2\%$] Black, and 427 492 [$53.4\%$] Hispanic) were studied. Compared with youths with a baseline BMI for age in the 40th to 59th percentiles, the adjusted hazard ratio (aHR) for hypertension within a maximum of 5 years was 1.26 ($95\%$ CI, 1.20-1.33) for youths between the 60th and 84th percentiles if they maintained their BMI for age. With every 1-unit annual increase in the distance to the median BMI for age, the aHR increased by 1.04 ($95\%$ CI, 1.04-1.05). The aHR was 4.94 ($95\%$ CI, 4.72-5.18) in youths with a baseline BMI for age in the 97th percentile or higher who maintained their body weight. Weight gain increased the risk associated with baseline BMI for age in the 97th percentile or higher with an aHR of 1.04 ($95\%$ CI, 1.04-1.05) per 1-unit annual increase in the distance to the median BMI for age. The risk associated with weight change was higher in youths living with low to high normal weight and overweight than in youths living with severe obesity. ### Conclusions and Relevance In this cohort study of youths, high normal body weight above the 60th percentile of BMI for age was associated with increased risk of hypertension. Weight gain was associated with further increases in hypertension risk. Further research is needed to evaluate the wide range of body weight considered normal in youths and the health risks associated with high normal weight. ## Introduction In 2015 to 2018, the prevalence of hypertension was $4.6\%$ in US children aged 8 to 12 years and $3.7\%$ in US adolescents aged 13 to 17 years.1 Hypertension during youth tracks into adulthood and is associated with cardiac and vascular target organ damage, such as thickening of the arteries, increased arterial stiffness, and decreased endothelial function.2 With increasing evidence that the target organ damage might become irreversible independent of blood pressure control,3,4 preventing sustained hypertension and associated target organ damage in children is essential. Obesity may be the most potent modifiable risk factor for hypertension during childhood.5,6,7,8,9,10,11 The epidemiologic evidence linking childhood obesity to the risk of hypertension is substantial.1,8,10,12,13,14 US data indicate a prevalence ratio for hypertension of 1.54 ($95\%$ CI, 0.64-3.69) for children who are overweight and 3.05 ($95\%$ CI, 1.78-5.20) for children with obesity.1 Hypertension increases disproportionally in youths who are severely obese with a sex- and age-specific body mass index (BMI; calculated as weight in kilograms divided by height in meters squared) above the 97th percentile.15 However, 2 current knowledge gaps exist. First, normal body weight in youths has a wide range from the 5th to the 85th percentiles of BMI for age,16,17,18 and, to our knowledge, no data exist to assess the health risk associated with low or high normal body weight. Second, data are lacking to differentiate between the risk of hypertension during childhood associated with baseline body weight and the risk associated with additional gain in body weight over time. With an increasing proportion of US youths who are heavier than reference youths based on Centers for Disease Control and Prevention (CDC) growth charts,17,18 the determination of the risk associated with excess body weight and weight gain has clinical and public health implications for strategies to prevent and treat pediatric hypertension. We designed the current study to [1] assess the risk of hypertension associated with high normal weight at baseline and [2] differentiate between the risk of hypertension associated with baseline body weight and the risk associated with additional gain in body weight over time. For the current study, we divided normal body weight into low (5th-39th percentiles), medium (40th-59th percentiles), and high (60th-84th percentiles) to provide insight into the risk of hypertension at a weight below the threshold for overweight. We report the incidence of hypertension within 5 years from a large multiethnic cohort of youths in southern California between 3 and 17 years of age who were passively followed up through electronic medical records. ## Study Design and Participants For the current retrospective cohort study, we identified youths between the ages of 3 and 17 years at baseline who were actively enrolled in a health plan between January 1, 2008, and February 28, 2015, with Kaiser Permanente Southern California (KPSC), a large, prepaid, integrated managed health care system.19 The follow-up of the cohort until the end of the study on February 28, 2020, was conducted through passive surveillance of clinical care information using an electronic health record (EHR) system. The first available blood pressure during the study period was used as the baseline date. Youths were ineligible to participate if they had no blood pressure or body weight measurements, preexisting hypertension, medical conditions known to significantly affect growth or blood pressure, or any chronic complex care conditions (eFigure in Supplement 1).20,21 We identified 940 409 eligible youths who were free of hypertension. We then excluded 139 390 youths with missing follow-up information because they left the KPSC health plan, gaps (>90 days) in health insurance coverage, missing valid follow-up blood pressures (blood pressures taken during medical visits indicating fever or pregnancy were not considered valid), or missing follow-up height to calculate blood pressure percentiles. The final analytical cohort comprised 801 019 hypertension-free youths who had their index date defined as the first visit with a BMI between 2008 and 2015 and at least 1 follow-up blood pressure within 5 years of the defined index date. The study protocol was reviewed and approved by the institutional review board of KPSC. The requirement of informed consent was waived. All data were deidentified. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.22 ## Blood Pressure Measurements and Classification Blood pressure was measured routinely at the beginning of almost every outpatient clinical visit, as described in detail elsewhere.15 Nurses and medical assistants were trained following the guidelines of the American Association of Critical Care Nurses for pediatric care.23 Digital devices (Welch Allyn Connex series, Welch Allyn Inc) are the preferred blood pressure measurement devices at KPSC. In some cases, a wall-mounted or portable aneroid sphygmomanometer (Welch Allyn Inc) was used. A full range of different cuff sizes was available at the locations where vital signs, including blood pressure, were measured in the clinics. All staff measuring blood pressure were certified in blood pressure measurement during their initial staff orientation and recertified annually. Blood pressure measures for all outpatient encounters were extracted from the EHR unless the measured body temperature at the time of the encounter was greater than 38.0 °C. We calculated blood pressure percentiles for age, sex, and height using 2017 updated normative tables and definitions based on children with normal body weight by the American Academy of Pediatrics.24 In youths between the ages of 3 and younger than 13 years, blood pressure was defined as stage I hypertension if at the 95th percentile or greater to less than the 95th percentile plus 12 mm Hg (or ≥$\frac{130}{80}$ mm Hg, whichever is lower) and as stage II hypertension if at the 95th percentile plus 12 mm Hg (or ≥$\frac{140}{90}$ mm Hg, whichever is lower).24 For youths 13 years or older, blood pressure was defined as stage I hypertension if $\frac{130}{80}$ to $\frac{139}{89}$ mm Hg and as stage II hypertension if $\frac{140}{90}$ mm Hg or higher.24 ## Definition of Study Outcome The primary study outcome was incident and persistent hypertension. During follow-up, participants were classified as having hypertension if they met 1 of the following conditions: [1] stage I hypertension confirmed during 3 independent medical visits, [2] stage II hypertension confirmed in at least 1 additional independent visit as stage II hypertension or higher, [3] 2 medical visits with a diagnosis of hypertension (defined as International Classification of Diseases, Ninth Revision [ICD-9] codes 401-405 and 362.11; International Statistical Classification of Diseases and Related Health Problems, Tenth Revision [ICD-10] codes I10-I13, I15.0, I15.8, and H35.039), or [4] 1 diagnosis of hypertension and a prescription of antihypertensive drugs.24,25 The date of the first occurrence was used as the event date. If the conditions above were not met (eg, a single high blood pressure), the occurrence was not counted as an event. For a sensitivity analysis, we used any first occurrence as an event: a single blood pressure indicating stage I or II hypertension and a single medical visit with a diagnosis of hypertension. ## Cohort Follow-up Youths were followed up passively for a maximum of 5 years through February 28, 2020, using information extracted from the EHR. We calculated the follow-up time from a child’s index date until the first occurrence of 1 of the following events: incidence of hypertension, pregnancy, end of KPSC health care coverage, death, end of follow-up on February 28, 2020, or 5 years after the index date. ## Body Weight and Height Body weight and height were routinely measured and extracted from the EHR. Biologically implausible BMI was defined as upper and lower 0.01 percentile (lower BMI of 7.9 and upper BMI of 70.8) and omitted before analysis ($$n = 5912$$ BMIs of 28 804 620 measurements). Definitions of overweight and obesity in children and adolescents are based on the sex-specific BMI-for-age growth charts developed by the CDC.16 Youths were categorized as underweight (BMI for age <5th percentile), low normal weight (BMI for age ≥5th to <40th percentiles), medium normal weight (BMI for age ≥40th to <60th percentiles), high normal weight (BMI for age ≥60th to <85th percentiles), overweight (BMI for age ≥85th to <95th percentiles), moderately obese (BMI for age ≥95th to <97th percentiles), and severely obese (BMI for age ≥97th percentile). Change in BMI from baseline was calculated for each follow-up visit as absolute difference in the distance from the median BMI for age and sex.26 ## Covariates We obtained sex, age, and year of birth from EHRs. Data for this study included youths whose parents reported their race/ethnicity as Asian or Pacific Islander, non-Hispanic Black or African American (hereafter, Black), non-Hispanic White (hereafter, White), Hispanic or Latino (regardless of race, hereafter, Hispanic), and other or unknown race or ethnicity based on various sources, such as registration records, clinical visit records, and birth certificates. The category other includes self-reported “other races or ethnicities” and “multiple races or ethnicities.” Data on race and ethnicity were obtained to investigate whether any racial and ethnic patient populations experienced disparities in their health and health care. We used government assistance for health care insurance (yes/no), such as Medicaid, as a proxy for a low socioeconomic status. ## Statistical Analysis Data analysis was performed from 2018 to 2022. Baseline characteristics, including sex, race and ethnicity, and socioeconomic status, were summarized as mean (SD) or numbers (percentages) reported by baseline BMI-for-age category. Incidence rates (IRs) of hypertension were estimated by dividing the number of hypertension cases by the total person-years of follow-up, overall and within each subgroup, reported as IR per 1000 person-years. The occurrences of hypertension followed a Poisson distribution, and $95\%$ CIs were estimated accordingly. To assess the association between baseline BMI for age, absolute change in the distance from median BMI for age, and risk of hypertension, we used Cox proportional hazards regression models with age (in years) as the time scale and birth year (in 5-year intervals, eg, 2000-2004) as strata. Sex, race and ethnicity, state-subsidized health insurance, and BMI-for-age category were treated as fixed effects; absolute change in the distance from the median BMI for age was modeled as a time-varying factor and can be interpreted as change per year (with age as the time scale of the model). We first modeled the effect of BMI-for-age class at baseline with and without adjusting for the change in distance from the median BMI for age. Next, we tested interaction terms in the fully adjusted model between change in distance from the median BMI for age and sex, state-subsidized health insurance, and baseline BMI-for-age category. We determined the best model fit based on the lowest Akaike information criterion value. Our final model included a 2-way interaction (change in distance from the median BMI for age × baseline BMI-for-age category). Adjusted hazard ratios (aHRs), corresponding $95\%$ CIs, and P values for interactions were reported. Sensitivity analysis using only 1 hypertensive blood pressure to define the outcome of hypertension did not result in relevant changes of the risk estimates. All P values were 2-sided, and $P \leq .05$ was considered significant. All statistical analyses were conducted using SAS software, version 9.4 (SAS Institute Inc). ## Cohort Demographic Characteristics and Crude Hypertension IRs A total of 801 019 youths (mean [SD] age, 9.4 [4.6] years; 409 167 [$51.1\%$] female]; 59 399 [$7.4\%$] Asian and Pacific Islanders, 65 712 [$8.2\%$[Black, and 427 492 [$53.4\%$] Hispanic) were studied (Table). During 3 579 994 person-years of follow-up with (mean [SD] follow-up time, 4.47 [1.20] years; mean [SD] number of office visits with blood pressure measurement, 9.9 [7.4]), we identified 24 969 youths with incident hypertension (IR, 6.97 per 1000 person-years; $95\%$ CI, 6.89-7.06) (eTable 1 in Supplement 1). The cohort contributed a total of 7 898 716 data points to the analysis of which 5 815 025 ($74.6\%$) had blood pressure assessments. The IRs per 1000 person-years were higher among boys (8.49; $95\%$ CI, 8.36-8.63) compared with girls (5.52; $95\%$ CI, 5.42-5.63), youths with a state-subsidized health plan (7.91; $95\%$ CI, 7.72-8.11) compared with those without (6.70; $95\%$ CI, 6.61-6.80), and highest among White (7.20; $95\%$ CI, 7.02-7.38) and Hispanic youths (7.19; $95\%$ CI, 7.08-7.32) compared with other youths (IRs ranged from 5.71 to 6.40). The IRs were similar for youths who were underweight and had low normal weight but then increased gradually with increasing BMI-for-age category. Table. \| Characteristic \| Baseline BMI-for-age percentile \| Baseline BMI-for-age percentile.1 \| Baseline BMI-for-age percentile.2 \| Baseline BMI-for-age percentile.3 \| Baseline BMI-for-age percentile.4 \| Baseline BMI-for-age percentile.5 \| Baseline BMI-for-age percentile.6 \| Total \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Characteristic \| Underweight (<5th) \| Low normal (5th-39th) \| Medium normal (40th-59th) \| High normal (60th-84th) \| Overweight (85th-94th) \| Moderate obesity (95th-96th) \| Severe obesity (≥97th) \| Total \| \| Sample sizes \| 25 880 (3.2) \| 166 859 (20.8) \| 128 176 (16.0) \| 215 993 (27.0) \| 132 464 (16.5) \| 41 465 (5.2) \| 90 182 (11.3) \| 801 019 (100.0) \| \| Sex \| \| \| \| \| \| \| \| \| \| Male \| 12 916 (49.9) \| 82 057 (49.2) \| 61 603 (48.1) \| 101 367 (46.9) \| 62 447 (47.1) \| 20 793 (50.1) \| 50 669 (56.2) \| 391 852 (48.9) \| \| Female \| 12 964 (50.1) \| 84 802 (50.8) \| 66 573 (51.9) \| 114 626 (53.1) \| 70 017 (52.9) \| 20 672 (49.9) \| 39 513 (43.8) \| 409 167 (51.1) \| \| Age at index, mean (SD), y \| 7.9 (4.5) \| 8.8 (4.7) \| 9.2 (4.7) \| 9.5 (4.7) \| 10.0 (4.5) \| 10.4 (4.2) \| 9.9 (4.3) \| 9.4 (4.6) \| \| Race and ethnicityb \| \| \| \| \| \| \| \| \| \| Asian or Pacific Islander \| 3498 (13.5) \| 17 335 (10.4) \| 10 625 (8.3) \| 14 948 (6.9) \| 7530 (5.7) \| 2022 (4.9) \| 3441 (3.8) \| 59 399 (7.4) \| \| Black \| 2067 (8.0) \| 13 020 (7.8) \| 10 440 (8.1) \| 18 518 (8.6) \| 10 777 (8.1) \| 3248 (7.8) \| 7642 (8.5) \| 65 712 (8.2) \| \| Hispanic \| 11 018 (42.6) \| 75 862 (45.5) \| 62 508 (48.8) \| 113 562 (52.6) \| 77 716 (58.7) \| 26 069 (62.9) \| 60 757 (67.4) \| 427 492 (53.4) \| \| White \| 7388 (28.5) \| 49 229 (29.5) \| 35 966 (28.1) \| 55 068 (25.5) \| 28 374 (21.4) \| 7638 (18.4) \| 13 317 (14.8) \| 196 980 (24.6) \| \| Other or unknownc \| 1909 (7.4) \| 11 413 (6.8) \| 8637 (6.7) \| 13 897 (6.4) \| 8067 (6.1) \| 2488 (6.0) \| 5025 (5.6) \| 51 436 (6.4) \| \| State-subsidized health plan \| \| \| \| \| \| \| \| \| \| No \| 20 601 (79.6) \| 131 975 (79.1) \| 100 330 (78.3) \| 166 132 (76.9) \| 99 472 (75.1) \| 30 514 (73.6) \| 65 261 (72.4) \| 614 285 (76.7) \| \| Yes \| 5279 (20.4) \| 34 884 (20.9) \| 27 846 (21.7) \| 49 861 (23.1) \| 32 992 (24.9) \| 10 951 (26.4) \| 24 921 (27.6) \| 186 734 (23.3) \| \| Blood pressure, mean (SD), mm Hg \| \| \| \| \| \| \| \| \| \| Systolic \| 96.9 (10.1) \| 99.1 (10.6) \| 100.5 (10.9) \| 101.9 (11.0) \| 104.2 (11.1) \| 106.2 (10.9) \| 106.9 (11.1) \| 102.1 (11.2) \| \| Diastolic \| 58.0 (8.1) \| 58.5 (8.2) \| 58.8 (8.2) \| 59.3 (8.2) \| 60.2 (8.2) \| 61.2 (8.2) \| 61.5 (8.3) \| 59.5 (8.3) \| \| Distance from median BMI for age and sex, mean (SD) \| \| \| \| \| \| \| \| \| \| Absolute \| −5.4 (1.5) \| −2.1 (0.9) \| 0.0 (0.5) \| 2.4 (1.0) \| 6.0 (1.3) \| 9.2 (1.2) \| 15.2 (5.2) \| 3.2 (5.8) \| \| Relative, % \| −24.1 (6.8) \| −9.5 (4.1) \| 0.2 (2.1) \| 10.8 (4.4) \| 26.8 (5.9) \| 41.4 (6.0) \| 67.8 (23.2) \| 14.4 (25.8) \| ## Baseline BMI for Age and Risk of Hypertension Youths at or above high normal weight had a higher risk of hypertension than youths with medium normal weight, independent of change in body weight during follow-up. In youths with high normal weight who maintained their BMI for age (Δ distance from median BMI for age during follow-up was 0), the aHR for incidence of hypertension was 1.26 ($95\%$ CI, 1.20-1.33), after adjusting for race and ethnicity, state-subsidized insurance coverage, and birth year (Figure). The risk of hypertension increased with increasing baseline weight class. The aHR for hypertension was 4.94 ($95\%$ CI, 4.72-5.18) for youths with a BMI for age in the 97th percentile or higher who maintained their BMI for age. Increased baseline weight was associated with a higher risk of hypertension regardless of weight gain and was consistent among youths with stable weight and those who lost or gained weight (eTable 2 in Supplement 1). Similarly, high normal weight was associated with a higher risk of hypertension in children who gained or lost weight compared with their peers with medium normal weight. Figure.: Adjusted Hazard Ratios for Incidence of Hypertension in Youths Aged 3 to 17 Years by Baseline Body Mass Index (BMI) for Age Class If the Distance to the Median BMI for Age Was MaintainedHazard ratios were adjusted for change in distance to the median BMI for age, sex, race and ethnicity, and state-subsidized health plan. Maintaining BMI for age was defined as a 0-unit change in the distance from the median BMI for age at a measurement point from the baseline. BMI was calculated as weight in kilograms divided by height in meters squared. ## Changes in BMI During Follow-up and Risk of Hypertension Weight change expressed as a change in the distance from the median BMI for age per year was associated with a change in hypertension risk independent of baseline weight classes. However, the risk associated with weight change was higher in youths living with low to high normal weight and overweight than in youths living with severe obesity, indicating a plateau in the association between weight gain and risk of hypertension. The aHR per 1-unit change in the distance from the median BMI for age per year was 1.05 ($95\%$ CI, 1.03-1.07) for youths living with a BMI for age in the less than 5th percentile, 1.07 ($95\%$ CI, 1.06-1.08) for youths living with a BMI for age between the 5th and 39th percentiles, 1.08 ($95\%$ CI, 1.07-1.08) for youths living with a BMI for age between the 40th and 59th percentiles, 1.08 ($95\%$ CI, 1.07-1.08) for youths living with a BMI for age between the 60th and 84th percentiles, 1.09 ($95\%$ CI, 1.08-1.09) for youths living with a BMI for age between the 85th and 94th percentiles, 1.06 ($95\%$ CI, 1.05-1.07) for youths living with a BMI for age between the 95th and 96th percentiles, and 1.04 ($95\%$ CI, 1.04-1.05) for youths living withs a BMI for age in the 97th percentile or greater. Although higher baseline body weight is associated with higher hypertension risk at any level of weight gain (eTable 2 in Supplement 1), the additional risk associated with a per 1-unit change in the distance from the median BMI for age per year was lower for youths living with a BMI for age in the 97th percentile or greater. ## Discussion In this large, prospective cohort study of 801 019 youths, a baseline body weight in the upper range of normal weight (60th-84th percentiles of BMI for age) was associated with an increased hypertension risk compared with youths between the 40th and 59th percentiles of BMI for age, with a $26\%$ higher risk of hypertension if their body weight remained stable during follow-up. Our results suggest that the current range of normal weight from the 5th to 84th percentiles of BMI for age in children may be too wide and requires reevaluation with regard to health risks. Obesity has been discussed as the main driver for pediatric hypertension during the last few decades.5,8,10,27 The prevalence of hypertension is currently approximately $3\%$ in youths with normal weight, $5\%$ in youths who are overweight, and nearly $10\%$ in youths who have severe obesity.1 Although several studies8,10,15,28,29,30,31,32 have shown that the risk of hypertension is higher in youths with overweight and obesity, information is lacking on high normal weight in youths as a hypertension risk factor. A recent meta-analysis33 in adults showed a continuous dose-dependent association between BMI and risk of hypertension starting within the normal BMI range. Our study also suggests that a high normal weight below the threshold for overweight is associated with increased hypertension risk in youths. Weight gain is associated with hypertension risk in adults.33 In that meta-analysis study,33 weight gain was associated with a worse prognosis among men than among women. In a prospective cohort study34 of Ethiopian children, weight gain from 48 to 60 months and weight at 60 months were associated with hypertension risk. In children with obesity, further weight gain and failure to reduce body weight were associated with higher blood pressure and a higher risk of developing hypertension compared with weight loss after initiation of obesity treatment.35 Consistent with other studies,34,35 weight gain (defined as an increase in the distance to the median BMI for age) in the current study accelerated and weight loss attenuated the hypertension risk associated with body weight at baseline. As expected, in youth with obesity (BMI for age ≥95th percentile), the escalation or attenuation per 1-unit change in the distance to the median BMI for age was less pronounced than in youths who lived with normal weight. Weight gain and obesity have complex physiologic sequelae mediated through a decrease in insulin sensitivity and the development of insulin resistance.36 The close association between body weight and insulin resistance is partially mediated through inflammatory pathways.36,37,38,39 Weight gain and obesity cause changes in the release of adipokines and cytokines from adipose tissue that manifest in metabolic dysfunctions and lead to compensatory hyperinsulinemia, which may increase blood pressure via multiple mechanisms, including inappropriate activation of the renin-angiotensin-aldosterone system.40,41,42 The metabolic dysfunction also promotes further weight gain, resulting in a vicious cycle of worsening insulin resistance and its metabolic sequelae.43 The dose-dependent association between excess body weight and risk of hypertension in youths described in the current study indicates that obesity-related etiologic pathways leading to hypertension may be activated at an early age and at much lower body weight than suggested by current thresholds for normal weight in youths. ## Strengths and Limitations Our study has several strengths and limitations. Strengths of the study include EHR data from a large, community-based population with a mean follow-up of more than 4 years; the systematic screening of weight, height, and blood pressure at almost every visit; and information available for potential key confounders and covariates. In addition, the study population is generally reflective of Southern California and includes a high proportion of children born to low-income families.44 The cohort study design reduced the chance of possible bias inherent in case-control and hospital-based studies, and the design accounted for factors associated with baseline weight under the assumption of stable weight, weight gain, or weight loss. In addition, the study benefitted from the high frequency of blood pressure and BMI measures throughout childhood, a high quality of weight and height measured due to rigorous training,45 and decision support tools to minimize missing data.46 The large sample size enabled evaluation of smaller weight categories, including 3 categories for low, medium, and high normal weight, with high precision and allowed us to estimate 2-way interactions between baseline weight and weight over time. This approach allowed us to estimate the association with weight under the assumption of stable weight over time. Limitations include the possibility of residual confounding inherent to the observational design, including the possibility of differential distribution of unmeasured or incompletely measured confounders. We can also not exclude limitations in the generalizability of the results in this California cohort with access to health care and the existence of selection bias. ## Conclusions The results of this cohort study indicate a strong association between body weight and the risk of hypertension. A high normal body weight in children, ranging from the 60th to the 84th percentile of BMI for age, was associated with increased hypertension risk. Furthermore, the risk of hypertension increased as additional weight gain occurred over time. Under this evidence, further research should reevaluate the current wide range of body weight considered normal and related health risks of high normal weight. ## References 1. Hardy ST, Sakhuja S, Jaeger BC. Trends in blood pressure and hypertension among US children and adolescents, 1999-2018. JAMA Netw Open (2021) 4. DOI: 10.1001/jamanetworkopen.2021.3917 2. Khoury M, Urbina EM. Hypertension in adolescents: diagnosis, treatment, and implications. Lancet Child Adolesc Health (2021) 5 357-366. DOI: 10.1016/S2352-4642(20)30344-8 3. de Simone G, Devereux RB, Izzo R. Lack of reduction of left ventricular mass in treated hypertension: the strong heart study. J Am Heart Assoc (2013) 2. DOI: 10.1161/JAHA.113.000144 4. Urbina EM, Lande MB, Hooper SR, Daniels SR. Target organ abnormalities in pediatric hypertension. J Pediatr (2018) 202 14-22. DOI: 10.1016/j.jpeds.2018.07.026 5. Falkner B. Children and adolescents with obesity-associated high blood pressure. J Am Soc Hypertens (2008) 2 267-274. DOI: 10.1016/j.jash.2008.01.003 6. Flynn JT, Falkner BE. Obesity hypertension in adolescents: epidemiology, evaluation, and management. J Clin Hypertens (Greenwich) (2011) 13 323-331. DOI: 10.1111/j.1751-7176.2011.00452.x 7. Köchli S, Endes K, Steiner R. Obesity, high blood pressure, and physical activity determine vascular phenotype in young children. Hypertension (2019) 73 153-161. DOI: 10.1161/HYPERTENSIONAHA.118.11872 8. Ferreira S. Obesity and hypertension in children: a worldwide problem. Rev Port Cardiol (Engl Ed) (2018) 37 433-434. DOI: 10.1016/j.repc.2018.04.006 9. Zhang YX, Wang SR, Li SY. Prevalence of severe obesity and its association with elevated blood pressure among children and adolescents in Shandong, China. Blood Press Monit (2017) 22 345-350. DOI: 10.1097/MBP.0000000000000292 10. Brady TM. Obesity-related hypertension in children. Front Pediatr (2017) 5 197. DOI: 10.3389/fped.2017.00197 11. Manios Y, Karatzi K, Protogerou AD. Prevalence of childhood hypertension and hypertension phenotypes by weight status and waist circumference: the Healthy Growth Study. Eur J Nutr (2018) 57 1147-1155. DOI: 10.1007/s00394-017-1398-y 12. Flynn J. The changing face of pediatric hypertension in the era of the childhood obesity epidemic. Pediatr Nephrol (2013) 28 1059-1066. DOI: 10.1007/s00467-012-2344-0 13. Armstrong KR, Cote AT, Devlin AM, Harris KC. Childhood obesity, arterial stiffness, and prevalence and treatment of hypertension. Curr Treat Options Cardiovasc Med (2014) 16 339. DOI: 10.1007/s11936-014-0339-9 14. Woodiwiss AJ, Norton GR. Obesity and left ventricular hypertrophy: the hypertension connection. Curr Hypertens Rep (2015) 17 539. DOI: 10.1007/s11906-015-0539-z 15. Koebnick C, Black MH, Wu J. High blood pressure in overweight and obese youth: implications for screening. J Clin Hypertens (Greenwich) (2013) 15 793-805. DOI: 10.1111/jch.12199 16. Kuczmarski RJ, Ogden CL, Guo SS. 2000 CDC Growth Charts for the United States: methods and development. Vital Health Stat 11 (2002) 11 1-190. PMID: 12043359 17. Ogden CL, Fryar CD, Hales CM, Carroll MD, Aoki Y, Freedman DS. Differences in obesity prevalence by demographics and urbanization in US children and adolescents, 2013-2016. JAMA (2018) 319 2410-2418. DOI: 10.1001/jama.2018.5158 18. Hales CM, Fryar CD, Carroll MD, Freedman DS, Ogden CL. Trends in obesity and severe obesity prevalence in US youth and adults by sex and age, 2007-2008 to 2015-2016. JAMA (2018) 319 1723-1725. DOI: 10.1001/jama.2018.3060 19. Koebnick C, Coleman KJ, Black MH. Cohort profile: the KPSC Children’s Health Study, a population-based study of 920 000 children and adolescents in southern California. Int J Epidemiol (2012) 41 627-633. DOI: 10.1093/ije/dyq252 20. Feinstein JA, Russell S, DeWitt PE, Feudtner C, Dai D, Bennett TD. R package for pediatric complex chronic condition classification. JAMA Pediatr (2018) 172 596-598. DOI: 10.1001/jamapediatrics.2018.0256 21. Feudtner C, Feinstein JA, Zhong W, Hall M, Dai D. Pediatric complex chronic conditions classification system version 2: updated for ICD-10 and complex medical technology dependence and transplantation. BMC Pediatr (2014) 14 199. DOI: 10.1186/1471-2431-14-199 22. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet (2007) 370 1453-1457. DOI: 10.1016/S0140-6736(07)61602-X 23. 23American Association of Critical-Care Nurses. American Association of Critical-Care Nurses Procedure Manual for Pediatric Acute and Critical Care. Saunders; 2007.. American Association of Critical-Care Nurses Procedure Manual for Pediatric Acute and Critical Care (2007) 24. Flynn JT, Kaelber DC, Baker-Smith CM. Clinical Practice Guideline for Screening and Management of High Blood Pressure in Children and Adolescents. Pediatrics (2017) 140. DOI: 10.1542/peds.2017-1904 25. James PA, Oparil S, Carter BL. 2014 Evidence-based guideline for the management of high blood pressure in adults: report from the panel members appointed to the Eighth Joint National Committee (JNC 8). JAMA (2014) 311 507-520. DOI: 10.1001/jama.2013.284427 26. Freedman DS, Woo JG, Ogden CL, Xu JH, Cole TJ. Distance and percentage distance from median BMI as alternatives to BMI. Br J Nutr (2020) 124 493-500. DOI: 10.1017/S0007114519002046 27. Dietz WH. Health consequences of obesity in youth: childhood predictors of adult disease. Pediatrics (1998) 101 518-525. DOI: 10.1542/peds.101.S2.518 28. Flynn J, Zhang Y, Solar-Yohay S, Shi V. Clinical and demographic characteristics of children with hypertension. Hypertension (2012) 60 1047-1054. DOI: 10.1161/HYPERTENSIONAHA.112.197525 29. Wühl E. Hypertension in childhood obesity. Acta Paediatr (2019) 108 37-43. DOI: 10.1111/apa.14551 30. Tran AH, Urbina EM. Hypertension in children. Curr Opin Cardiol (2020) 35 376-380. DOI: 10.1097/HCO.0000000000000744 31. Obarzanek E, Wu CO, Cutler JA, Kavey RE, Pearson GD, Daniels SR. Prevalence and incidence of hypertension in adolescent girls. J Pediatr (2010) 157 461-467, 467.e1-467.e5. DOI: 10.1016/j.jpeds.2010.03.032 32. Lo JC, Chandra M, Sinaiko A. Severe obesity in children: prevalence, persistence and relation to hypertension. Int J Pediatr Endocrinol (2014) 2014 3. DOI: 10.1186/1687-9856-2014-3 33. Jayedi A, Rashidy-Pour A, Khorshidi M, Shab-Bidar S. Body mass index, abdominal adiposity, weight gain and risk of developing hypertension: a systematic review and dose-response meta-analysis of more than 2.3 million participants. Obes Rev (2018) 19 654-667. DOI: 10.1111/obr.12656 34. Wibaek R, Girma T, Admassu B. Higher weight and weight gain after 4 years of age rather than weight at birth are associated with adiposity, markers of glucose metabolism, and blood pressure in 5-year-old Ethiopian children. J Nutr (2019) 149 1785-1796. DOI: 10.1093/jn/nxz121 35. Hagman E, Danielsson P, Elimam A, Marcus C. The effect of weight loss and weight gain on blood pressure in children and adolescents with obesity. Int J Obes (Lond) (2019) 43 1988-1994. DOI: 10.1038/s41366-019-0384-2 36. Kahn SE, Hull RL, Utzschneider KM. Mechanisms linking obesity to insulin resistance and type 2 diabetes. Nature (2006) 444 840-846. DOI: 10.1038/nature05482 37. Kwon H, Pessin JE. Adipokines mediate inflammation and insulin resistance. Front Endocrinol (Lausanne) (2013) 4 71. DOI: 10.3389/fendo.2013.00071 38. Blüher M. Adipose tissue dysfunction in obesity. Exp Clin Endocrinol Diabetes (2009) 117 241-250. DOI: 10.1055/s-0029-1192044 39. Yamauchi T, Kamon J, Waki H. The fat-derived hormone adiponectin reverses insulin resistance associated with both lipoatrophy and obesity. Nat Med (2001) 7 941-946. DOI: 10.1038/90984 40. Manrique C, Lastra G, Gardner M, Sowers JR. The renin angiotensin aldosterone system in hypertension: roles of insulin resistance and oxidative stress. Med Clin North Am (2009) 93 569-582. DOI: 10.1016/j.mcna.2009.02.014 41. Rodríguez A, Ezquerro S, Méndez-Giménez L, Becerril S, Frühbeck G. Revisiting the adipocyte: a model for integration of cytokine signaling in the regulation of energy metabolism. Am J Physiol Endocrinol Metab (2015) 309 E691-E714. DOI: 10.1152/ajpendo.00297.2015 42. DeMarco VG, Aroor AR, Sowers JR. The pathophysiology of hypertension in patients with obesity. Nat Rev Endocrinol (2014) 10 364-376. DOI: 10.1038/nrendo.2014.44 43. Barber TM, Kyrou I, Randeva HS, Weickert MO. Mechanisms of insulin resistance at the crossroad of obesity with associated metabolic abnormalities and cognitive dysfunction. Int J Mol Sci (2021) 22 546. DOI: 10.3390/ijms22020546 44. Koebnick C, Langer-Gould AM, Gould MK. Sociodemographic characteristics of members of a large, integrated health care system: comparison with US Census Bureau data. Perm J (2012) 16 37-41. DOI: 10.7812/TPP/12-031 45. Koebnick C, Mohan YD, Li X, Young DR. Secular trends of overweight and obesity in young southern Californians 2008-2013. J Pediatr (2015) 167 1264-71.e2. DOI: 10.1016/j.jpeds.2015.08.039 46. Koebnick C, Mohan Y, Li X. Failure to confirm high blood pressures in pediatric care-quantifying the risks of misclassification. J Clin Hypertens (Greenwich) (2018) 20 174-182. DOI: 10.1111/jch.13159
--- title: Population Preferences for Primary Care Models for Hypertension in Karnataka, India authors: - Hannah H. Leslie - Giridhara R. Babu - Nolita Dolcy Saldanha - Anne-Marie Turcotte-Tremblay - Deepa Ravi - Neena R. Kapoor - Suresh S. Shapeti - Dorairaj Prabhakaran - Margaret E. Kruk journal: JAMA Network Open year: 2023 pmcid: PMC10015308 doi: 10.1001/jamanetworkopen.2023.2937 license: CC BY 4.0 --- # Population Preferences for Primary Care Models for Hypertension in Karnataka, India ## Key Points ### Question How can government health care services best meet population preferences for ongoing hypertension treatment in Karnataka, India? ### Findings In this cross-sectional study of 1085 adults with hypertension, a discrete choice experiment revealed that $85\%$ of respondents prioritized careful clinical assessment and consistent availability of free medication. Some urban respondents prioritized shorter wait times, and some rural respondents prioritized seeing a physician vs a nurse. ### Meaning The population preferences identified in this study suggest that consistent medication availability and quality of clinical assessment are key priorities for strengthening primary care services for adults with hypertension in urban and rural areas of Karnataka, India. ## Abstract This cross-sectional study uses household surveys and a discrete choice experiment to assess preferences for primary care services among adults with hypertension in Karnataka, India. ### Importance Hypertension contributes to more than 1.6 million deaths annually in India, with many individuals being unaware they have the condition or receiving inadequate treatment. Policy initiatives to strengthen disease detection and management through primary care services in India are not currently informed by population preferences. ### Objective To quantify population preferences for attributes of public primary care services for hypertension. ### Design, Setting, and Participants This cross-sectional study involved administration of a household survey to a population-based sample of adults with hypertension in the Bengaluru Nagara district (Bengaluru City; urban setting) and the Kolar district (rural setting) in the state of Karnataka, India, from June 22 to July 27, 2021. A discrete choice experiment was designed in which participants selected preferred primary care clinic attributes from hypothetical alternatives. Eligible participants were 30 years or older with a previous diagnosis of hypertension or with measured diastolic blood pressure of 90 mm Hg or higher or systolic blood pressure of 140 mm Hg or higher. A total of 1422 of 1927 individuals ($73.8\%$) consented to receive initial screening, and 1150 ($80.9\%$) were eligible for participation, with 1085 ($94.3\%$) of those eligible completing the survey. ### Main Outcomes and Measures Relative preference for health care service attributes and preference class derived from respondents selecting a preferred clinic scenario from 8 sets of hypothetical comparisons based on wait time, staff courtesy, clinician type, carefulness of clinical assessment, and availability of free medication. ### Results Among 1085 adult respondents with hypertension, the mean (SD) age was 54.4 (11.2) years; 573 participants ($52.8\%$) identified as female, and 918 ($84.6\%$) had a previous diagnosis of hypertension. Overall preferences were for careful clinical assessment and consistent availability of free medication; 3 of 5 latent classes prioritized 1 or both of these attributes, accounting for $85.1\%$ of all respondents. However, the largest class ($52.4\%$ of respondents) had weak preferences distributed across all attributes (largest relative utility for careful clinical assessment: β = 0.13; $95\%$ CI, 0.06-0.20; $36.4\%$ preference share). Two small classes had strong preferences; 1 class ($5.4\%$ of respondents) prioritized shorter wait time ($85.1\%$ preference share; utility, β = −3.04; $95\%$ CI, −4.94 to −1.14); the posterior probability of membership in this class was higher among urban vs rural respondents (mean [SD], 0.09 [0.26] vs 0.02 [0.13]). The other class ($9.5\%$ of respondents) prioritized seeing a physician (the term doctor was used in the survey) rather than a nurse ($66.2\%$ preference share; utility, β = 4.01; $95\%$ CI, 2.76-5.25); the posterior probability of membership in this class was greater among rural vs urban respondents (mean [SD], 0.17 [0.35] vs 0.02 [0.10]). ### Conclusions and Relevance In this study, stated population preferences suggested that consistent medication availability and quality of clinical assessment should be prioritized in primary care services in Karnataka, India. The heterogeneity observed in population preferences supports considering additional models of care, such as fast-track medication dispensing to reduce wait times in urban settings and physician-led services in rural areas. ## Introduction Cardiovascular disease accounts for more than $28\%$ of total deaths and $25\%$ of years of life lost among adults older than 50 years in India,1,2 with hypertension affecting more than 200 million individuals and playing a role in at least 1.6 million deaths annually.3,4 In Karnataka state, where hypertension prevalence exceeds $20\%$,5 awareness of disease status is low. Although more than $80\%$ of those diagnosed with hypertension have initiated treatment, only $62\%$ of those receiving treatment have reported consistent medication use.4,6,7 Population-level hypertension control depends on effective primary care.8 Indian policy has prioritized prevention and control of noncommunicable diseases through primary care services since implementation of the 2010 National Programme for Prevention and Control of Cancer, Diabetes, Cardiovascular Diseases and Stroke.9,10 The focus on prevention and control of noncommunicable diseases was further emphasized in the 2018 National Health Protection Mission, which included the planned provision of comprehensive primary care services through 150 000 new health and wellness centers (HWCs).11,12,13,14 However, the extensive primary care system has faced challenges in adapting to the changing needs and preferences of an increasingly educated and rapidly urbanizing population.15,16,17,18,19,20 Disease burden, availability of health care services, and service use patterns differ between rural and urban areas.19,21,22 Only in the past decade has policy begun to be standardized instead of being administered through separate rural and urban national health missions, and differences persist in services and health care use. Private care services are concentrated in urban areas, and public community health centers (CHCs) are predominantly located in rural areas.7 While the bypassing of public primary care has been common across all households with members diagnosed with hypertension, urban respondents have been less likely to use public primary care services than rural respondents ($11\%$ vs $23\%$, respectively).23 Discrete choice experiments (DCEs) are a method of quantifying population preferences to inform health care service delivery.24,25,26,27,28,29 We conducted a DCE to characterize stated preferences for key aspects of health care services among adults with hypertension in an urban district and a rural district in Karnataka, India. ## Methods This cross-sectional study was reviewed and approved by the Harvard Human Research Protection Program, the Public Health Foundation of India Institutional Ethics Committee, and the Indian Institute of Public Health–Bengaluru Campus Institutional Ethics Committee. Permission for study procedures was received from the Indian Council of Medical Research Health Ministry Screening Committee, the Karnataka State Directorate of Health and Family Welfare, and the chief health officers and chief medical officers in participating districts. All participants provided written informed consent. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies. The study reflexivity statement is available in eTable 1 in Supplement 1. ## Study Setting The state of Karnataka, India, is home to more than 66.8 million people, with $39\%$ residing in urban areas.30 This study was conducted in the Bengaluru Nagara district (Bangalore City; 4400 residents per km2, with approximately 10 million total residents) and the Kolar district (384 residents per km2, with approximately 1.5 million total residents). Districts were selected to provide urban (Bengaluru Nagara) and rural (Kolar) study sites that [1] were accessible to the Bangalore-based study team throughout the period of COVID-19–related restrictions, [2] were located in areas in which district health officials were receptive, and [3] had an HWC program that was initiated but not fully operational. Data collection took place from February 2 to February 21, 2021, for the DCE development phase and from June 22 to July 27, 2021, for administration of the household surveys that included the DCE. ## Development of Discrete Choice Experiment This study aimed to assess how government health care services could best meet population preferences for ongoing hypertension treatment. The DCE was designed in accordance with the checklist developed by the Good Research Practices for Conjoint Analysis Task Force31 (full details are provided in eMethods in Supplement 1). In brief, consistent with recommended practice,31 we reviewed published literature and national and state policy to identify an extensive list of attributes that were relevant to ongoing policy innovations in primary care, amenable to intervention at the district health care system level, and pertinent to individuals with hypertension.31,32,33,34 We then conducted 6 focus groups in both urban and rural settings; each focus group comprised 5 to 7 patients receiving care for hypertension to elicit perspectives on their experiences and the relevance of these attributes. The focus groups enabled participants to identify relevant attributes that were not part of the initial list derived from the literature and expert consultations, to rank their own priorities among the proposed attributes, and to suggest attribute levels based on their own experiences. The research team (H.H.L., G.R.B., N.D.S., A.T.T., S.S.S., D.P., and M.E.K.) synthesized focus group findings to select the final attributes and levels (Table 1) to include in the DCE based on priority among patients, relevance to the research question, and independence across attributes. Table 1. \| Attribute \| Levels \| \| --- \| --- \| \| Staff attitudes, including nonclinical personnel such as security \| Clinic staff members are courteous \| \| Staff attitudes, including nonclinical personnel such as security \| Clinic staff members are not always courteous \| \| Total wait time \| 15 min \| \| Total wait time \| 30 min \| \| Total wait time \| 1 h \| \| Total wait time \| 2 h \| \| Total wait time \| 3 h \| \| Total wait time \| 5 h \| \| Clinician type \| Physiciana \| \| Clinician type \| Nurse \| \| Quality of clinical assessment \| Medical staff assess patients carefully \| \| Quality of clinical assessment \| Medical staff do not always assess patients carefully \| \| Availability of free medication \| Free medication is available in this facility \| \| Availability of free medication \| Free medication is not always available in this facility \| ## Experimental Design We designed the DCE with 2 alternatives per choice set and no opt-out option to ensure responses would be provided by all participants. The DCE scenario alternatives were selected to ensure balance and optimize determinant efficiency (a measure of the goodness of a design relative to a hypothetical orthogonal design); 35 choice sets were generated in 5 versions. A choice with 1 option designed to emulate the HWC was added to all versions, resulting in 8 choice sets per respondent (the introductory script and an example of a choice card are available in eFigure 1 in Supplement 1). Data collectors selected 1 version for each respondent, cycling through the 5 versions in turn. Additional survey items addressed demographic characteristics, health status, and previous use of and perspectives on health care services. To ensure consistency, all survey materials, which were originally written in English, were translated into Kannada, then translated back to English by the research team (G.R.B., N.D.S., and D.R.); respondents could select English or Kannada. We pretested all survey materials among 10 respondents. ## Sample Size We calculated the target sample size based on the power required to maximize the efficiency of the DCE. To calculate the minimum sample size, the largest number of levels for an attribute (including interaction terms) was divided by the product of the number of alternatives in each task multiplied by the number of tasks; this quotient was then multiplied by 500.35,36 We calculated the minimum sample size before finalizing the DCE design, using 2 as the number of alternatives in each task, 8 as the number of tasks, and 10 as the product of the largest number of levels of an attribute [5] and a binary interaction term, such as study site or a 2-level attribute. This calculation yielded a minimum of 312 participants per subgroup. To ensure a robust sample size for estimation and enhance the generalizability of findings, we targeted 500 participants per location. ## Sampling We used maps of administrative areas created for the most recent National Health Mission immunization campaign as a sampling framework.37 In the Bengaluru Nagara district, we identified 2 wards (among 198 total wards; mean population, 42 500 per ward) that included both informal settlements and formal areas and could be accessed by the research team. Each ward was divided into approximately 20 units comparable with a city block. The Kolar district includes 6 administrative subdivisions; we selected the main Kolar area, which includes the largest number of village clusters, to represent the rural setting of the district, and we selected 2 of 343 villages using convenience sampling. Field teams visited each unit or village and used systematic random sampling of households until the target of 500 surveys per district was reached. ## Survey Administration Eligibility assessment included a brief informed consent process, a questionnaire, and hypertension screening in accordance with Indian national guidelines. Two blood pressure (BP) measures were obtained (with a third obtained if the difference between the first and second measurements was >5 mm Hg for either systolic or diastolic BP), and the lower BP value was used. Elevated BP was defined as systolic BP of 140 mm Hg or higher or diastolic BP of 90 mm Hg or higher. Eligibility was assessed based on age (≥30 years) and self-reported diagnosis of hypertension and/or elevated BP at the time of the survey; pregnant women were ineligible. All individuals with elevated BP and no previous diagnosis of hypertension received information on government-approved nearby sources of care regardless of study participation. In Bengaluru Nagara, interested adult residents in the households received screening, and 1 resident was selected from those eligible. In Kolar, a Kish grid38 was used to select adults in random order for screening. Eligible adults were invited to participate in the full study after providing a second consent; only 1 respondent was enrolled per household. ## Statistical Analysis Data were collected on tablet computers using the Research Electronic Data Capture (REDCap) database and synced to servers daily. Stata software, version 17.0 (StataCorp LLC), was used for data cleaning and analysis, with additional packages including dcreate, mixlogit, and lclogit2.39,40,41,42 We used R software, version 4.2.1 (R Foundation for Statistical Computing), with packages dplyr, foresplot, tidyverse, readxl, stringr, and haven for the forest plot. Descriptive statistics were used to summarize the demographic characteristics (including gender, age, caste, educational level, and occupation) of the study population; information was collected on caste rather than race and ethnicity in accordance with the Census of India and all major population-based surveys. We conducted robustness checks (details are available in eMethods in Supplement 1) and then fit mixed logit models. These models estimated the likelihood of selecting a clinic as a function of clinic attributes; parameters were allowed to vary randomly across individuals to account for heterogeneity in preferences and scale and to address nonindependence of multiple responses within individuals. The results provided estimates of the mean relative utility of each attribute level within the DCE as well as the SD of the estimated utility. Estimated SDs could be compared with a null hypothesis of 0 variance; direction of the estimate was irrelevant.39 Mean relative utility and SDs were considered significant if their $95\%$ CIs excluded the null. After model testing, we fit the full population mixed logit model on all data with normally distributed parameters, independent covariance structure, robust SEs, and 500 Halton draws. To identify groupings of respondent preferences, we fit a latent class model with up to 8 latent classes41,43; we selected the final model based on bayesian information criterion statistics. We estimated posterior probabilities of class membership for each respondent. We calculated preference shares as the percentage of utility for each attribute among the total utility, multiplying utility for wait time by the defined range for this attribute (4.75 hours). We estimated uptake for each latent class comparing a baseline scenario (staff not always courteous, wait time is 3 hours, primary care clinician is a nurse, clinical assessment not always careful, and free medication not always available) with each of the following 3 scenarios: [1] primary care clinician is a physician (the term doctor was used in the survey) and other attribute levels the same as baseline (physician-led model), [2] wait time is 30 minutes and other attribute levels are the same as baseline (fast-track model), and [3] clinical assessment always careful and free medication always available and other attribute levels are the same as baseline (based on HWC model). We assessed the association of the observed characteristics of location (Bengaluru Nagara or Kolar), gender (identification as female vs male or other), formal education (none vs primary school, secondary school, or college or higher), and awareness of hypertension diagnosis (yes vs no) with the posterior probability of class membership. Findings for location were reported from unadjusted models; adjusted models including location could not be estimated due to the magnitude of class share differences by location. ## Results Of 1927 individuals approached, 1422 ($73.8\%$) consented to receive initial screening. Among those who received screening, 1150 ($80.9\%$) were eligible for study inclusion; of those, 1085 individuals ($94.3\%$) consented to and completed the full survey (eFigure 2 in Supplement 1). The mean (SD) age of respondents was 54.4 (11.2) years; 573 ($52.8\%$) identified as female, 507 ($46.7\%$) identified as male, and 5 ($0.5\%$) identified as other genders (Table 2). A total of 530 respondents were from the Bengaluru Negara district, and 555 were from the Kolar district. The majority of respondents (918 [$84.6\%$]; 510 [$96.2\%$] in Bengaluru Nagara and 408 [$73.5\%$] in Kolar) reported being previously diagnosed with hypertension; of those, nearly all respondents (883 of 913 [$96.2\%$]; 492 of 510 [$96.5\%$] in Bengaluru Nagara and 391 of 408 [$95.8\%$] in Kolar) had previously received treatment for hypertension. Government facilities were the most common source of hypertension care in both Bengaluru Nagara (407 of 510 respondents [$79.8\%$]) and Kolar (347 of 408 respondents [$85.0\%$]). Table 2. \| Characteristic \| Respondents, No. (%) \| Respondents, No. (%).1 \| Respondents, No. (%).2 \| \| --- \| --- \| --- \| --- \| \| Characteristic \| Total (N = 1085) \| Bengaluru Nagara (n = 530) \| Kolar (n = 555) \| \| Age, mean (SD), y \| 54.4 (11.2) \| 57.2 (9.6) \| 51.8 (12.1) \| \| Gender \| \| \| \| \| Female \| 573 (52.8) \| 277 (52.3) \| 296 (53.3) \| \| Male \| 507 (46.7) \| 251 (47.4) \| 256 (46.1) \| \| Othera \| 5 (0.5) \| 2 (0.4) \| 3 (0.5) \| \| Casteb \| \| \| \| \| General category \| 135 (12.4) \| 67 (12.6) \| 68 (12.3) \| \| Scheduled caste or tribe \| 492 (45.3) \| 259 (48.9) \| 233 (42.0) \| \| Other backward caste \| 262 (24.1) \| 124 (23.4) \| 138 (24.9) \| \| Otherc \| 191 (17.6) \| 76 (14.3) \| 115 (20.7) \| \| Missing \| 5 (0.5) \| 4 (0.8) \| 1 (0.2) \| \| Educational level \| \| \| \| \| No formal education \| 261 (24.1) \| 125 (23.6) \| 136 (24.5) \| \| Primary school \| 478 (44.1) \| 255 (48.1) \| 223 (40.2) \| \| Secondary school \| 221 (20.4) \| 103 (19.4) \| 118 (21.3) \| \| College or higher \| 125 (11.5) \| 47 (8.9) \| 78 (14.1) \| \| Occupation \| \| \| \| \| Not employed outside of home \| 328 (30.2) \| 247 (46.6) \| 81 (14.6) \| \| Semiskilled or unskilled \| 444 (40.9) \| 107 (20.2) \| 337 (60.7) \| \| Skilled \| 296 (27.3) \| 163 (30.8) \| 133 (24.0) \| \| Professional \| 4 (0.4) \| 2 (0.4) \| 2 (0.4) \| \| Missing \| 13 (1.2) \| 11 (2.1) \| 2 (0.4) \| \| Ever previously diagnosed with hypertension \| \| \| \| \| No \| 167 (15.4) \| 20 (3.8) \| 147 (26.5) \| \| Yes \| 918 (84.6) \| 510 (96.2) \| 408 (73.5) \| \| Received medication for hypertension \| \| \| \| \| No \| 35 (3.8) \| 18 (3.5) \| 17 (4.2) \| \| Yes \| 883 (96.2) \| 492 (96.5) \| 391 (95.8) \| \| Source of hypertension care \| \| \| \| \| Government facility: primary \| 419 (45.6) \| 207 (40.6) \| 212 (52.0) \| \| Government facility: secondary \| 335 (36.5) \| 200 (39.2) \| 135 (33.1) \| \| Private facility \| 151 (16.4) \| 98 (19.2) \| 53 (13.0) \| \| Otherd \| 13 (1.4) \| 5 (1.0) \| 8 (2.0) \| The 1085 respondents completed 8656 choice tasks within the DCE. Data quality checks identified no concerns with DCE administration or responses. Details on the validity checks and model fitting are provided in the eMethods in Supplement 1. Five attributes were included in the DCE: staff attitudes, total wait time, clinician type, quality of clinical assessment, and availability of free medication. The mixed logit model for the full study population (Table 3) revealed that respondents did not highly value courtesy relative to other attributes (utility, β = 0.04; $95\%$ CI, −0.02 to 0.11), were weakly averse to longer wait times (utility, β = −0.09; $95\%$ CI, −0.12 to −0.05), and preferred being seen by physicians over nurses (utility, β = 0.34; $95\%$ CI, 0.24-0.43). The strongest preferences were for careful clinical assessment (utility, β = 0.67; $95\%$ CI, 0.56-0.78) and availability of free medication (utility, β = 0.68; $95\%$ CI, 0.57-0.80). The SDs revealed population heterogeneity in preferences for all attributes except courtesy. Table 3. \| Attribute (N = 1085)a \| Utility, β (95% CI) \| \| --- \| --- \| \| Mean \| Mean \| \| Total wait time \| −0.09 (−0.12 to −0.05) \| \| Clinic staff members are courteous (vs not always courteous) \| 0.04 (−0.02 to 0.11) \| \| Seen by a physician (vs a nurse)b \| 0.34 (0.24 to 0.43) \| \| Clinicians assess patients carefully (vs do not always assess patients carefully) \| 0.67 (0.56 to 0.78) \| \| Free medication is available in this facility (vs not always available) \| 0.68 (0.57 to 0.80) \| \| SD \| \| \| Total wait time \| 0.26 (0.20 to 0.32) \| \| Clinic staff members are courteous (vs not always courteous) \| 0.02 (−0.18 to 0.22) \| \| Seen by a physician (vs a nurse)b \| 0.96 (0.83 to 1.10) \| \| Clinicians assess patients carefully (vs do not always assess patients carefully) \| 1.24 (1.10 to 1.37) \| \| Free medication is available in this facility (vs not always available) \| 1.25 (1.11 to 1.39) \| Latent class analysis revealed a 5-class solution optimized model fit. Class 5 members ($52.4\%$ of respondents) had significant preferences for careful clinical assessment (utility, β = 0.13; $95\%$ CI, 0.06-0.20) and availability of free medication (utility, β = 0.08; $95\%$ CI, 0.00-0.16), although relative preferences were weakly differentiated (Figure 1). Class 2 members ($16.9\%$ of respondents) strongly prioritized availability of free medication (utility, β = 4.22; $95\%$ CI, 2.65-5.79), while class 1 members ($15.8\%$ of respondents) prioritized careful clinical assessment (utility, β = 6.76; $95\%$ CI, 0.65-12.88) and had a negative preference for availability of free medication (utility, β = −1.09; $95\%$ CI, −2.01 to −0.16) relative to other attributes. Together, these 3 classes with relative preferences for careful clinical assessment and/or availability of free medication composed $85.1\%$ of the total population. Respondents in the remaining classes had strong preferences for being seen by a physician rather than a nurse (class 3 [$9.5\%$ of respondents]; utility, β = 4.01; $95\%$ CI, 2.76-5.25) and for shorter wait time (class 4 [$5.4\%$ of respondents]; utility, β = −3.04; $95\%$ CI, −4.94 to −1.14). In each of the 4 smaller classes, the most preferred attribute within the class comprised at least $50\%$ of total utility ($50.1\%$ for careful clinical assessment in class 1, $68.5\%$ for availability of free medication in class 2, $66.2\%$ for seeing a physician vs a nurse in class 3, and $85.1\%$ for wait time in class 4) (eFigure 3 in Supplement 1). Figure 1.: Latent Class Preferences for Hypertension Services The estimated uptake of services in the scenario in which preferences for careful clinical assessment and consistent availability of free medication were met relative to the baseline scenario of low-quality care was high overall ($72.7\%$) and within each class ($99.7\%$ in class 1, $98.4\%$ in class 2, $77.0\%$ in class 3, and $76.8\%$ in class 4), with the exception of class 5 ($55.2\%$), which had weak relative preferences that yielded little difference in estimated uptake across scenarios (Figure 2). Because class 5 comprised approximately one-half ($52.4\%$) of the study population, weak uptake within this class constrained overall uptake estimates. In the smaller classes that prioritized seeing a physician (class 3) or shorter wait time (class 4), meeting these preferences increased the estimated uptake even more ($98.2\%$ for class 3 under the physician-led scenario and $99.9\%$ for class 4 under the 30-minute wait time scenario) than the scenario in which preferences for careful clinical assessment and consistent availability of free medication were met. Figure 2.: Estimated Uptake of 3 Service Options by Latent ClassEach service option was compared with the baseline scenario, which consisted of the following attributes: staff not always courteous, 3-hour wait time, seen by a nurse, examination not always careful, and free medication not always available. Uptake of 50% represents uptake equivalent to that of the baseline service option. Physician-led indicates a scenario in which the individual is seen by a physician, with other attribute levels unchanged from the baseline scenario. Fast track describes a scenario with 30-minute wait time, with other attribute levels unchanged from the baseline scenario. Assessment and medication indicates a scenario in which clinical assessment is always careful and free medication is always available, with other attribute levels unchanged from the baseline scenario. Class membership differed by location (eTable 2 in Supplement 1), with 3 classes having strong divergence between sites. The mean (SD) posterior probabilities of being in class 1 (which prioritized careful clinical assessment) were 0.28 (0.44) in Kolar vs 0.03 (0.13) in Bengaluru Nagara, and the mean (SD) posterior probabilities of being in class 3 (which prioritized seeing a physician) were 0.17 (0.35) in Kolar vs 0.02 (0.10) in Bengaluru Nagara. Conversely, the mean (SD) share of respondents prioritizing wait time (class 4) was 0.09 (0.26) in Bengaluru Nagara and only 0.02 (0.13) in Kolar. In a model adjusted for gender, level of formal education, and knowledge of hypertension diagnosis (eTable 3 in Supplement 1), individuals without formal education had higher odds of belonging to class 2 (which prioritized availability of free medication; adjusted odds ratio [aOR], 1.88; $95\%$ CI, 1.27-2.77) or class 4 (which prioritized shorter wait time; aOR, 2.22; $95\%$ CI, 1.19-4.13) than belonging to class 5 (which had weak relative preferences). Individuals unaware of their hypertension status had higher odds of belonging to classes prioritizing careful clinical assessment (class 1: aOR, 3.54; $95\%$ CI, 2.21-5.66) or seeing a physician (class 3: aOR, 4.01; $95\%$ CI, 2.34-6.87) than belonging to class 5. ## Discussion This cross-sectional study of population preferences for hypertension care services in Karnataka, India, found overall preferences for careful clinical assessment and/or consistent availability of free medication among the majority of respondents. Approximately one-half ($52.4\%$) of respondents had only weak relative preferences for these attributes. Additional smaller preference classes included primarily rural respondents who prioritized seeing a physician rather than a nurse and primarily urban respondents who prioritized shorter wait times. Given these findings, addressing medication availability and ensuring clinical competence are key priorities in the continued expansion of the HWC primary care model. Previous analyses of national surveys and qualitative studies20,23,44,45,46 found that patients cited long wait times and lack of medication and diagnostic assessment as well as inconsistent clinician availability as major concerns with public primary care clinics. Our study refined this understanding to quantify relative preferences, finding that for most respondents, longer wait times might be acceptable if stronger preferences for consistent medication availability and/or competent care were met. The findings highlighted the importance of resolving supply chain issues that play a role in the inconsistent availability of medications47 and focusing attention on the need for a competent workforce that has time to provide careful assessments.17 Notably, most respondents were willing to be seen by nurses if other preferences were met; only a small proportion of respondents ($9.5\%$), primarily in the rural setting, prioritized physician-led care. While not directly comparable due to differences in DCE design and study population, a DCE among residents of urban slums in Ahmedabad, India, similarly identified competent care as the highest priority overall and found heterogeneity in clinician preference (among traditional, private, and public facilities) by socioeconomic status32; we found that availability of free medication and shorter wait times were particularly important for individuals without formal education. Our findings build on previous work in Karnataka, which found that patients were interested in competent care for noncommunicable diseases48 and suggested that rapid expansion of the nonphysician HWC model in Karnataka may be an appropriate approach to alleviate the overload placed on primary care physicians.18,49 At the same time, physician-led care will remain important for diagnoses and prescription treatment; legally, only physicians can prescribe medication or change treatment regimens, while nurses can manage ongoing treatment. We did not find that courteous treatment was highly valued relative to other attributes, despite the experiences of disrespectful care reported during the focus groups for development of this DCE and in other studies.32,48 This finding may reflect respondents’ willingness to trade discourteous treatment for more competent care or more convenient services, particularly given that most respondents had experience with hypertension treatment. It was also notable that proposed elements of HWC, such as yoga services,48 were not prioritized in focus group discussions. The current cross-sectional study was conducted in a setting of policy innovation in primary care delivery. As of early 2021, more than 2000 HWCs were operational in Karnataka based on data from the state government49; per policy, these facilities are staffed by nonphysician clinicians who directly provide screening and ongoing treatment support for hypertension while overseeing lay health care workers who extend screening into the community. As a safeguard, HWCs are supported by physicians with prescription capabilities at referral facilities. In this context, 3 policy implications were identified. First, continued expansion of HWCs should prioritize medication availability and competent care, including care from nonphysicians, in both urban and rural settings as a primary goal for responding to stated population preferences. Second, subsets of the study population expressed preferences consistent with differentiated models of care, such as fast-track and physician-led services. There is a basis for service delivery innovations in Karnataka through efforts such as evening clinics in informal settlements50; looking forward, the National Digital Health Mission includes ambitious plans for electronic health records to ensure continuity of care across service locations, which could improve consistent access to medications if fully implemented. Third, monitoring of service use and further research on service uptake would help clinicians and policy makers understand the group of comprehensive decision makers who had weak relative preferences for the attributes assessed in this study. ## Limitations This study has several limitations. Administrative areas were selected based on the feasibility of conducting research during the period of COVID-19 pandemic restrictions and the provision of support from district health officials; results may not be generalizable to all districts in Karnataka or to other states in India. The respondent population included a high proportion who were aware of their hypertension diagnosis, potentially due to greater interest in study participation among these individuals. The results provide less opportunity to draw inferences on aspects of service delivery that would help to reach and retain those with currently undiagnosed hypertension. We designed a forced-choice DCE, in which respondents could not opt out of the choice task presented, to optimize internal validity; DCE estimates have been found to be accurate among those who are truly likely to use a potential service or product.29 This design has less external validity in identifying those who are unlikely to use services overall or to use services based on the attributes and levels studied; our uptake estimates should be interpreted as revealing relative trade-offs rather than absolute estimates of population-level use of primary care services. The DCE findings are applicable to the specific attributes and levels assessed; it is possible that the relatively weaker preferences and lower estimated service uptake among the largest class of respondents reflect preferences not captured in this DCE. Some respondents may have considered the attributes of seeing a physician and medication availability collectively due to current policy restricting the authority to provide prescriptions for medication to physicians only. We worded the attributes to focus on independent aspects of care and encouraged respondents to consider the choices as hypothetical to reduce the possibility of conflating attributes. ## Conclusions This cross-sectional study involving population preference assessment found that adults with hypertension prioritized consistent medication availability and quality of clinical assessment in both urban and rural settings within Karnataka, India. Evaluation of additional models of care, such as physician-led services and fast-track medication dispensing to reduce wait times, may be warranted to fully address heterogeneity in population preferences. ## References 1. 1Institute for Health Metrics and Evaluation. GBD Compare data visualization. IHME, University of Washington. October 15, 2020. Accessed November 17, 2021. https://www.healthdata.org/data-visualization/gbd-compare 2. The changing patterns of cardiovascular diseases and their risk factors in the states of India: the Global Burden of Disease Study 1990-2016. Lancet Glob Health. DOI: 10.1016/S2214-109X(18)30407-8 3. Anchala R, Kannuri NK, Pant H. Hypertension in India: a systematic review and meta-analysis of prevalence, awareness, and control of hypertension. J Hypertens (2014.0) 32 1170-1177. DOI: 10.1097/HJH.0000000000000146 4. 4Arokiasamy P, Parasuraman S, Sekher TV, Lhungdim H. Study on Global Ageing and Adult Health (SAGE) wave 1: India national report. International Institute for Population Sciences. September 2013. Accessed November 5, 2018. https://apps.who.int/healthinfo/systems/surveydata/index.php/catalog/65/download/2011. (2013.0) 5. Geldsetzer P, Manne-Goehler J, Theilmann M. Diabetes and hypertension in India: a nationally representative study of 1.3 million adults. JAMA Intern Med (2018.0) 178 363-372. DOI: 10.1001/jamainternmed.2017.8094 6. Prenissl J, Manne-Goehler J, Jaacks LM. Hypertension screening, awareness, treatment, and control in India: a nationally representative cross-sectional study among individuals aged 15 to 49 years. PLoS Med (2019.0) 16. DOI: 10.1371/journal.pmed.1002801 7. 7District Level Household & Facility Survey 4 (DLHS-4). International Institute for Population Sciences. 2013. Accessed November 5, 2018. http://rchiips.org/DLHS-4.html. (2013.0) 8. Kruk ME, Nigenda G, Knaul FM. Redesigning primary care to tackle the global epidemic of noncommunicable disease. Am J Public Health (2015.0) 105 431-437. DOI: 10.2105/AJPH.2014.302392 9. 9Ministry of Health and Family Welfare. National Programme for Prevention and Control of Cancer, Diabetes, Cardiovascular Diseases and Stroke (NPCDCS). Government of India. 2022. Accessed March 2, 2022. https://main.mohfw.gov.in/Major-Programmes/non-communicable-diseases-injury-trauma/Non-Communicable-Disease-II/National-Programme-for-Prevention-and-Control-of-Cancer-Diabetes-Cardiovascular-diseases-and-Stroke-NPCDCS 10. 10National Health Mission. Operational guidelines: prevention, screening and control of common non-communicable diseases. National Health Mission, Government of India. 2016. Accessed October 8, 2020. https://main.mohfw.gov.in/sites/default/files/Operational%20Guidelines%20on%20Prevention%2C%20Screening%20and%20Control%20of%20Common%20NCDs_1.pdf 11. Lahariya C. DOI: 10.1007/s12098-020-03359-z 12. Ved RR, Gupta G, Singh S. India’s health and wellness centres: realizing universal health coverage through comprehensive primary health care. WHO South East Asia J Public Health (2019.0) 8 18-20. DOI: 10.4103/2224-3151.255344 13. Bhargava B, Paul VK. Informing NCD control efforts in India on the eve of Ayushman Bharat. Lancet (2022.0) 399. DOI: 10.1016/S0140-6736(18)32172-X 14. 14Ministry of Electronics and Information Technology. Ayushman Bharat: National Health Protection Mission (AB-NHPM). Government of India. 2020. Accessed July 20, 2021. https://indiaai.gov.in/missions/ayushman-bharat-national-health-protection-mission-ab-nhpm 15. Das J, Holla A, Das V, Mohanan M, Tabak D, Chan B. (2012.0). DOI: 10.1377/hlthaff.2011.1356 16. Mohanan M, Hay K, Mor N. Quality of health care in India: challenges, priorities, and the road ahead. Health Aff (Millwood) (2016.0) 35 1753-1758. DOI: 10.1377/hlthaff.2016.0676 17. Pakhare A, Kumar S, Goyal S, Joshi R. Assessment of primary care facilities for cardiovascular disease preparedness in Madhya Pradesh, India. BMC Health Serv Res (2015.0) 15 408. DOI: 10.1186/s12913-015-1075-x 18. Lall D, Engel N, Devadasan N, Horstman K, Criel B. Challenges in primary care for diabetes and hypertension: an observational study of the Kolar district in rural India. BMC Health Serv Res (2019.0) 19 44. DOI: 10.1186/s12913-019-3876-9 19. Bhojani U, Devedasan N, Mishra A, De Henauw S, Kolsteren P, Criel B. Health system challenges in organizing quality diabetes care for urban poor in South India. PLoS One (2014.0) 9. DOI: 10.1371/journal.pone.0106522 20. Elias MA, Pati MK, Aivalli P. (2018.0). DOI: 10.1136/bmjgh-2017-000519 21. Barua N, Pandav CS. The allure of the private practitioner: is this the only alternative for the urban poor in India?. Indian J Public Health (2011.0) 55 107-114. DOI: 10.4103/0019-557X.85242 22. Gore RJ 23. Kujawski SA, Leslie HH, Prabhakaran D, Singh K, Kruk ME. Reasons for low utilisation of public facilities among households with hypertension: analysis of a population-based survey in India. BMJ Glob Health (2018.0) 3. DOI: 10.1136/bmjgh-2018-001002 24. de Bekker-Grob EW, Ryan M, Gerard K. Discrete choice experiments in health economics: a review of the literature. Health Econ (2012.0) 21 145-172. DOI: 10.1002/hec.1697 25. Mangham LJ, Hanson K, McPake B. How to do (or not to do)...designing a discrete choice experiment for application in a low-income country. Health Policy Plan (2009.0) 24 151-158. DOI: 10.1093/heapol/czn047 26. Lancsar E, Fiebig DG, Hole AR. Discrete choice experiments: a guide to model specification, estimation and software. Pharmacoeconomics (2017.0) 35 697-716. DOI: 10.1007/s40273-017-0506-4 27. Johnson FR, Lancsar E, Marshall D. Constructing experimental designs for discrete-choice experiments: report of the ISPOR Conjoint Analysis Experimental Design Good Research Practices Task Force. Value Health (2013.0) 16 3-13. DOI: 10.1016/j.jval.2012.08.2223 28. Johnson FR, Yang JC, Reed SD. The internal validity of discrete choice experiment data: a testing tool for quantitative assessments. Value Health (2019.0) 22 157-160. DOI: 10.1016/j.jval.2018.07.876 29. Quaife M, Terris-Prestholt F, Di Tanna GL, Vickerman P. How well do discrete choice experiments predict health choices? a systematic review and meta-analysis of external validity. Eur J Health Econ (2018.0) 19 1053-1066. DOI: 10.1007/s10198-018-0954-6 30. 30Office of the Registrar General & Census Commissioner. Census of India. Ministry of Home Affairs, Government of India. 2017. Accessed July 20, 2021. https://censusindia.gov.in/census.website. (2017.0) 31. Bridges JFP, Hauber AB, Marshall D. Conjoint analysis applications in health—a checklist: a report of the ISPOR Good Research Practices for Conjoint Analysis Task Force. Value Health (2011.0) 14 403-413. DOI: 10.1016/j.jval.2010.11.013 32. Černauskas V, Angeli F, Jaiswal AK, Pavlova M. Underlying determinants of health provider choice in urban slums: results from a discrete choice experiment in Ahmedabad, India. BMC Health Serv Res (2018.0) 18 473. DOI: 10.1186/s12913-018-3264-x 33. Fletcher B, Hinton L, McManus R, Rivero-Arias O. Patient preferences for management of high blood pressure in the UK: a discrete choice experiment. Br J Gen Pract (2019.0) 69 e629-e637. DOI: 10.3399/bjgp19X705101 34. Moor SE, Tusubira AK, Akiteng AR. Development of a discrete choice experiment to understand patient preferences for diabetes and hypertension management in rural Uganda. Lancet Glob Health (2020.0) 8 S22. DOI: 10.1016/S2214-109X(20)30163-7 35. Orme B. (2019.0) 36. de Bekker-Grob EW, Donkers B, Jonker MF, Stolk EA. Sample size requirements for discrete-choice experiments in healthcare: a practical guide. Patient (2015.0) 8 373-384. DOI: 10.1007/s40271-015-0118-z 37. 37Ministry of Health and Family Welfare. Karnataka—state program implementation plans: National Health Mission (NHM) PIPs & ROPs. Government of India. 2022. Accessed May 18, 2022. https://nhm.gov.in/index4.php?lang=1&level=0&linkid=61&lid=74 38. 38Kish L. A procedure for objective respondent selection within the household. J Am Stat Assoc. 1949;44(247):380-387. doi:10.1080/01621459.1949.10483314. DOI: 10.1080/01621459.1949.10483314 39. Hole AR. Fitting mixed logit models by using maximum simulated likelihood. Stata J (2007.0) 7 388-401. DOI: 10.1177/1536867X0700700306 40. Gu Y, Hole AR, Knox S. Fitting the generalized multinomial logit model in Stata. Stata J (2013.0) 13 382-397. DOI: 10.1177/1536867X1301300213 41. Yoo HI. Lclogit2: an enhanced command to fit latent class conditional logit models. Stata J (2020.0) 20 405-425. DOI: 10.1177/1536867X20931003 42. Hole A 43. Pacifico D, il Yoo H. Lclogit: a Stata command for fitting latent-class conditional logit models via the expectation-maximization algorithm. Stata J (2013.0) 13 625-639. DOI: 10.1177/1536867X1301300312 44. Rao KD, Sheffel A. Quality of clinical care and bypassing of primary health centers in India. Soc Sci Med (2018.0) 207 80-88. DOI: 10.1016/j.socscimed.2018.04.040 45. Jayanna K, Swaroop N, Kar A. Designing a comprehensive non-communicable diseases (NCD) programme for hypertension and diabetes at primary health care level: evidence and experience from urban Karnataka, South India. BMC Public Health (2019.0) 19 409. DOI: 10.1186/s12889-019-6735-z 46. Bagchi T, Das A, Dawad S, Dalal K. Non-utilization of public healthcare facilities during sickness: a national study in India. J Public Health (Oxf) (2022.0) 30 943-951. DOI: 10.1007/s10389-020-01363-3 47. Thakur J. Key recommendations of high-level expert group report on universal health coverage for India. Indian J Community Med (2011.0) 36 S84-S85. PMID: 22628920 48. Bhojani U, Mishra A, Amruthavalli S. Constraints faced by urban poor in managing diabetes care: patients’ perspectives from South India. Glob Health Action (2013.0) 6 22258. DOI: 10.3402/gha.v6i0.22258 49. 49Health and wellness centres: Karnataka ranks first. The Hindu. May 18, 2021. Accessed November 17, 2021. https://www.thehindu.com/news/national/karnataka/health-and-wellness-centres-karnataka-ranks-first/article34590753.ece. (2021.0) 50. 50National urban health mission. Government of Karnataka. 2013. Accessed December 16, 2021. https://nhm.karnataka.gov.in/page/Programmes/National+Urban+Health+Mission/en
--- title: Intimate Partner Violence, Mental Health Symptoms, and Modifiable Health Factors in Women During the COVID-19 Pandemic in the US authors: - Arielle A. J. Scoglio - Yiwen Zhu - Rebecca B. Lawn - Audrey R. Murchland - Laura Sampson - Janet W. Rich-Edwards - Shaili C. Jha - Jae H. Kang - Karestan C. Koenen journal: JAMA Network Open year: 2023 pmcid: PMC10015312 doi: 10.1001/jamanetworkopen.2023.2977 license: CC BY 4.0 --- # Intimate Partner Violence, Mental Health Symptoms, and Modifiable Health Factors in Women During the COVID-19 Pandemic in the US ## Abstract This cohort study evaluates the risk of depression, anxiety, and posttraumatic stress symptoms and adverse health outcomes among female adults in potentially violent relationships in the early days of pandemic-related restrictions. ## Key Points ### Question Was intimate partner violence (IPV) early in the COVID-19 pandemic in the US associated with adverse mental health symptoms and modifiable health factors in women? ### Findings In this cohort study of 3 nationwide cohorts involving 13 597 female participants, experiencing IPV was associated with higher endorsement of mental health symptoms, shorter sleep duration, poorer sleep quality, and increased use of alcohol or other substances. ### Meaning The findings of this study suggest that IPV during the first 1.5 years of the pandemic in the US was associated with harmful health consequences; screening and interventions for IPV and related health factors are needed to prevent such outcomes. ### Importance During the COVID-19 pandemic, the prevalence and severity of intimate partner violence (IPV) increased. Associations between IPV and mental health symptoms and modifiable health factors early in the pandemic have yet to be explored. ### Objective To prospectively investigate the association of IPV with greater risk of mental health symptoms and adverse health factors during the COVID-19 pandemic in 3 cohorts of female participants. ### Design, Setting, and Participants This cohort study used observational data from 3 prospective, population-based, longitudinal cohorts in the US: the Nurses’ Health Study II, Growing Up Today Study, and Nurses’ Health Study 3. Data analyzed included baseline and follow-up survey responses about IPV experiences early in the pandemic (March-September 2020); mental health domains of depression, anxiety, and posttraumatic stress symptoms (PTSS); and modifiable health factors (May 2020-October 2021). Female participants (both health care professionals and non–health care workers) aged 21 to 60 years from the 3 cohorts were included in the full analytic sample. ### Exposures Experience of IPV measured by the Relationship Assessment Tool and fear of partner. ### Main Outcomes and Measures Mental health symptoms, including depression, anxiety, and PTSS, and modifiable health factors, including sleep duration, sleep quality, physical activity, alcohol use, and use of alcohol or other substances to cope with stress. ### Results The full analytic sample included 13 597 female participants with a mean (SD) age of 44 (10.6) years. Accounting for sociodemographic factors and prepandemic mental health symptoms and correcting for multiple testing, experiencing IPV was associated with higher endorsement of depression (odds ratio [OR], 1.44; $95\%$ CI, 1.38-1.50), anxiety (OR, 1.31; $95\%$ CI, 1.26-1.36), and PTSS (OR, 1.22; $95\%$ CI, 1.15-1.29) in random-effects meta-analyses across the 3 cohorts. The IPV experience was also associated with poorer sleep quality (OR, 1.21; $95\%$ CI, 1.16-1.26), shorter sleep duration (OR, 1.13; $95\%$ CI, 1.08-1.19), increased use of alcohol (OR, 1.10; $95\%$ CI, 1.06-1.14), and use of alcohol or other substances to cope with stress (OR, 1.13; $95\%$ CI, 1.08-1.18) across all cohorts as well as decreased physical activity (OR, 1.17; $95\%$ CI, 1.09-1.26) in the Nurses’ Health Study II only. ### Conclusions and Relevance Results of the study showed that IPV experiences at the start of the pandemic were associated with worse mental health symptoms and modifiable health factors for female participants younger than 60 years. Screening and interventions for IPV and related health factors are needed to prevent severe, long-term health consequences. ## Introduction Intimate partner violence (IPV) is defined as physical, sexual, or psychological harm by a current or former partner.1 Women experience a substantial burden of IPV: approximately one-third of women who have been in a relationship have experienced physical or sexual abuse.2,3,4 Early in the COVID-19 pandemic, IPV experts expressed concern that COVID-19 mitigation actions meant to ensure public safety, such as stay-at-home orders, might further isolate individuals in abusive relationships and increase the prevalence and severity of violence.5,6,7,8 The pandemic played a role in the exacerbated burden of external stressors for many households, such as financial hardship, job loss, and food or housing insecurity.4,9,10,11 Increased isolation of women in potentially violent living situations and external stressors on their partners may be associated with heightened risk of IPV exposure.8 As expected, IPV increased globally in 2020.12 In the US, calls to domestic violence hotlines13,14 increased in the first months of the pandemic. Outside of the pandemic context, IPV is a risk factor for poor health outcomes, including depression and cardiometabolic diseases.15,16,17,18 A potential pathway to adverse physical health outcomes may be changes in modifiable health factors, such as sleep, substance use, or exercise.19,20 It is not yet known how IPV during the pandemic may affect modifiable health factors. A limited number of studies used a cross-sectional21,22,23,24 or qualitative design25 to examine the association between IPV experience and mental health and behavioral outcomes during the first year of the pandemic. These prior studies have found that IPV was associated with mental health problems, such as depression and anxiety symptoms, and with adverse health factors, such as increased substance use and COVID-19 exposure risk-taking (eg, gathering indoors with people not belonging to one’s household during a lockdown).22,23,26 However, to our knowledge, no prospective studies have examined the associations of IPV with health factors and mental health symptoms in large, population-based cohorts. Intimate partner violence affects health across the life course,27 and the implications of IPV may manifest differently for individuals with IPV history. Therefore, a prospective design with information about prepandemic health and IPV history is instrumental in contextualizing the implications of IPV during the pandemic. Herein, we aimed to extend previous studies by prospectively investigating the association of IPV with greater risk of mental health symptoms and adverse health factors during the COVID-19 pandemic in 3 cohort samples of female participants aged 21 to 60 years. We hypothesized that individuals who experienced IPV early in the pandemic would be at higher risk for mental health problems and adverse modifiable health factors, when accounting for prior depression and anxiety. Furthermore, we explored whether the role of IPV in these outcomes differed by prior experience of IPV. ## Study Design and Population This cohort study consisted of participants from 3 prospective cohorts: the Nurses’ Health Study II (NHS II), Growing Up Today Study (GUTS), and Nurses’ Health Study 3 (NHS3).28 The details on sampling and recruitment are provided in the eMethods in Supplement 1. The Brigham and Women's Hospital and the Harvard T.H. Chan School of Public Health Institutional Review Boards approved the study protocol. Completion of the questionnaires by participants was considered to be implied consent. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.29 From April to May 2020, participants who had completed the most recent cohort questionnaire were invited to complete an online COVID-19 survey designed to examine experiences of both health care professionals (ie, those reporting to work in person at patient care institutions) and other individuals during the pandemic. After this initial survey, participants completed 3 monthly and 3 quarterly administered surveys, with the last quarterly questionnaire completed between March and October 2021.30 The month-1 questionnaire, which was mailed 28 days after the baseline questionnaire (May-June 2020), included an assessment of IPV for individuals who reported being in a relationship since March 2020 and were younger than 60 years. The age restriction was implemented because investigators were not able to follow mandatory state reporting guidelines for elder abuse. Month-1 survey returns were accepted through September 2020. While questionnaire return time varied among participants, there was no overlap in exposure and outcome assessment periods at the individual level. The minimum duration between exposure and outcome reporting was 1 month. ## Measures Details about assessment contents and timing are included in the eMethods, eFigure 1, and eTable 1 in Supplement 1. We assessed experiences with IPV, mental health symptoms, modifiable health factors, covariates, and history of IPV. We considered 2 exposures of IPV experiences, which were measured in the month-1 survey for each participant. First, participants who reported having a spouse, partner, or significant other since March 1, 2020, completed an adapted 6-item version of the Relationship Assessment Tool (RAT,31 formerly known as the Women’s Experience with Battering Scale), which measures emotional distress in response to IPV and has been found to have excellent internal consistency and validity. We used a standardized continuous summary score of RAT that ranged from 0 to 36 before standardization, with the highest score indicating the most severe level of emotional distress in response to IPV.15,20 In this sample, the quartile ranges of the RAT score were 0 to 1 for no distress (first quartile), 1 to 2 for very low distress (second quartile), 2 to 6 for mild distress (third quartile), and 6 to 36 for moderate to high distress (fourth quartile) in response to IPV. Second, we included in this survey a single item to assess fear of partner (ie, yes or no to the question, Since March 1, 2020, have you ever felt afraid of your spouse/partner/significant other?).32 Associations between these 2 exposure measures (RAT score and fear of partner) and health were assessed separately in subsequent analyses. We assessed the following domains of mental health: depression, anxiety, and posttraumatic stress symptoms (PTSS), which are common responses to experiencing IPV.33 Current depression and anxiety were assessed with the 4-item Patient Health Questionnaire.34 The time frame of inquiry was adapted to cover the past 7 days. Current PTSS were assessed using the 6-item Impact of Events Scale,35 which was adapted to reflect the current context (eg, I thought about COVID-19 or other current events when I didn’t mean to and I was aware that I had a lot of feelings about COVID-19 or other current events, but I didn’t deal with them). To minimize missing data due to loss to follow-up, we considered the highest level of distress over follow-up (1-17 months after IPV assessment) for each domain separately. Specific modifiable health factors that we examined were reported sleep quality, sleep duration, physical activity, alcohol use, and use of alcohol or other substances to cope with stress in 2020 to 2021. These factors were selected because of their known associations with IPV and long-term physical health36,37 outside of the pandemic context. We adjusted for sociodemographic factors, including age at the baseline survey, race and ethnicity (coded as White [given that >$95\%$ of the samples self-identified as non-Hispanic White] or Other [American Indian or Alaska Native, Asian, Black or African American, Hispanic, Native Hawaiian or Other Pacific Islander, and multiracial]; race and ethnicity were assessed to identify possible disparities, with acknowledgment that race is socially constructed and its measurement can be a proxy for racism38), educational level as a proxy for socioeconomic status (whose information was used was based on data availability: the partner’s was used in NHS II and NHS3, whereas the participant’s was used in GUTS), active health care professional status during the COVID-19 pandemic (yes or no), partnership status (eg, married, separated or divorced, widowed, or single; data from the most recent biennial or modular surveys before 2020 were used), sexual orientation (heterosexual or not heterosexual), and depression or anxiety prior to exposure ascertainment (eMethods in Supplement 1). For all covariates with missing data, missing indicators were used. We were able to examine history of IPV in 2 of the 3 cohorts. Exposure to IPV was measured in 2001 and 2008 for NHS II and in 2007 for GUTS (eMethods in Supplement 1). ## Statistical Analysis All analyses were conducted in R, version 4.1.0 (R Core Team), and 2-sided tests were used. Multiple testing burden was accounted for using a false discovery rate correction. The statistical significance threshold was corrected $P \leq .05.$ Descriptive analyses were performed to characterize the distributions of sociodemographic and health characteristics by IPV exposure in each cohort. Prevalence of IPV, as indicated by the distribution of the continuous RAT score and endorsement of individual items, was also examined. To assess the associations between IPV and each outcome measure, we fitted 16 logistic regression models in each cohort separately (2 exposures [RAT score and fear of partner] and 8 outcomes), resulting in a total of 48 models. After the stratified analysis within each cohort, a random-effects meta-analysis was performed per outcome to summarize the association with IPV across the 3 cohorts. Inverse probability weighting was implemented to address loss to follow-up (eMethods in Supplement 1). In a secondary analysis, to evaluate whether patterns of associations between IPV during the pandemic and health outcomes were different among individuals with vs without prior IPV exposure, we repeated the primary analyses and stratified by IPV history in NHS II and GUTS, for which data on prior IPV exposure were available. All main and secondary analyses were adjusted for the covariates. ## Descriptive Analysis The final analytic sample included 13 597 female participants (3503 from NHS II, 2858 from GUTS, and 7236 from NHS3) with a mean (SD) age of 44 (10.6) years. Most participants identified as being of non-Hispanic White race ($96.3\%$) and being heterosexual ($92.8\%$). Participants of the NHS II, GUTS, and NHS3 had a mean (SD) age of 58 (1.4) years, 33 (3.3) years, and 42 (7.5) years, respectively. At baseline in spring 2020, active health care professionals composed $21.2\%$ of the GUTS, $57.4\%$ of the NHS II, and $76.0\%$ of the NHS3 participants. In the full sample, the proportions of active health care professionals were comparable across 4 quartiles of the RAT score (Table). Prevalence of anxiety and PTSS was high in the samples, with $37.9\%$ of participants in NHS II, $61.9\%$ in GUTS, and $50.5\%$ in NHS3 reporting anxiety and $38.4\%$ in NHS II, $56.7\%$ in GUTS, and $51.3\%$ in NHS3 reporting PTSS at 1 or more of the follow-up interviews. Depressive symptoms were reported by $21.2\%$ of the NHS II, $37.4\%$ of the GUTS, and $32.1\%$ of the NHS3 participants. Table. \| Characteristic \| No. of participants (%) in each RAT score quartile \| No. of participants (%) in each RAT score quartile.1 \| No. of participants (%) in each RAT score quartile.2 \| No. of participants (%) in each RAT score quartile.3 \| \| --- \| --- \| --- \| --- \| --- \| \| Characteristic \| First (mean score = 0.2) \| Second (mean score = 1.5) \| Third (mean score = 3.7) \| Fourth (mean score = 10.7) \| \| No. of participants \| 3400 \| 3399 \| 3399 \| 3399 \| \| Fear of partnera \| 19 (0.6) \| 19 (0.6) \| 18 (0.5) \| 207 (6.1) \| \| Amount of time spent with partner due to COVID-19 pandemic restrictionsa \| \| \| \| \| \| Decreased \| 202 (5.9) \| 223 (6.6) \| 194 (5.7) \| 323 (9.5) \| \| No change \| 923 (27.2) \| 834 (24.6) \| 875 (25.8) \| 797 (23.4) \| \| Increased \| 2258 (66.4) \| 2327 (68.6) \| 2317 (68.2) \| 2259 (66.5) \| \| NA or missing data \| 16 (0.5) \| 10 (0.3) \| 12 (0.4) \| 20 (0.6) \| \| Change in relationship quality \| \| \| \| \| \| Worsened \| 77 (2.3) \| 145 (4.3) \| 280 (8.3) \| 780 (23.0) \| \| No change \| 2252 (66.3) \| 2262 (66.6) \| 2179 (64.2) \| 1918 (56.5) \| \| Improved \| 1058 (31.1) \| 987 (29.1) \| 928 (27.4) \| 685 (20.2) \| \| NA or missing data \| 11 (0.3) \| 3 (0.1) \| 6 (0.2) \| 14 (0.4) \| \| Living arrangement at baseline \| \| \| \| \| \| Alone \| 104 (3.1) \| 96 (2.8) \| 86 (2.5) \| 107 (3.1) \| \| With partnera \| 3147 (92.6) \| 3170 (93.3) \| 3158 (92.9) \| 3149 (92.6) \| \| With children \| 46 (1.4) \| 47 (1.4) \| 39 (1.1) \| 53 (1.6) \| \| With othersb \| 95 (2.8) \| 79 (2.3) \| 108 (3.2) \| 85 (2.5) \| \| With pets \| 8 (0.2) \| 6 (0.2) \| 8 (0.2) \| 5 (0.1) \| \| Age, mean (SD), y \| 46.56 (11.16) \| 42.58 (9.76) \| 43.42 (10.94) \| 45.09 (10.11) \| \| Active health care professional status during the COVID-19 pandemic \| 2046 (60.2) \| 2137 (62.9) \| 1888 (55.5) \| 2043 (60.1) \| \| Non-Hispanic White race \| 3284 (96.6) \| 3271 (96.2) \| 3278 (96.4) \| 3266 (96.1) \| \| Other race and ethnicityc \| 116 (3.4) \| 128 (3.8) \| 121 (3.6) \| 133 (3.9) \| \| Partnership status \| \| \| \| \| \| Married \| 2375 (69.9) \| 2238 (65.8) \| 2287 (67.3) \| 2428 (71.4) \| \| Divorced \| 143 (4.2) \| 129 (3.8) \| 109 (3.2) \| 138 (4.1) \| \| Widowed \| 16 (0.5) \| 5 (0.1) \| 4 (0.1) \| 6 (0.2) \| \| Domestic partnership \| 140 (4.1) \| 164 (4.8) \| 201 (5.9) \| 174 (5.1) \| \| Separated \| 17 (0.5) \| 16 (0.5) \| 28 (0.8) \| 32 (0.9) \| \| Never married or in a relationship \| 622 (18.3) \| 799 (23.5) \| 688 (20.2) \| 547 (16.1) \| \| Missing data \| 87 (2.6) \| 48 (1.4) \| 82 (2.4) \| 74 (2.2) \| \| Spouse, partner, or participant educational level \| \| \| \| \| \| ≤High school diploma \| 261 (7.7) \| 262 (7.7) \| 280 (8.2) \| 311 (9.1) \| \| College degree \| 1668 (49.1) \| 1586 (46.7) \| 1647 (48.5) \| 1696 (49.9) \| \| Graduate school \| 805 (23.7) \| 724 (21.3) \| 813 (23.9) \| 755 (22.2) \| \| NA or missing data \| 666 (19.6) \| 827 (24.3) \| 659 (19.4) \| 637 (18.7) \| \| Sexual orientation \| \| \| \| \| \| Heterosexual \| 3167 (93.1) \| 3131 (92.1) \| 3201 (94.2) \| 3120 (91.8) \| \| Not heterosexual \| 124 (3.6) \| 127 (3.7) \| 112 (3.3) \| 122 (3.6) \| \| Missing data \| 109 (3.2) \| 141 (4.1) \| 86 (2.5) \| 157 (4.6) \| \| Depression or anxiety prior to exposure ascertainment \| \| \| \| \| \| No \| 2729 (80.3) \| 2498 (73.5) \| 2530 (74.4) \| 2261 (66.5) \| \| Yes \| 671 (19.7) \| 899 (26.4) \| 865 (25.4) \| 1136 (33.4) \| \| Missing data \| 0 (0.0) \| 2 (0.1) \| 4 (0.1) \| 2 (0.1) \| \| Depression \| 709 (21.7) \| 848 (26.0) \| 1023 (31.1) \| 1391 (42.8) \| \| Anxiety \| 1314 (40.1) \| 1550 (47.5) \| 1699 (51.6) \| 1919 (59.1) \| \| PTSS \| 1356 (41.4) \| 1518 (46.6) \| 1677 (51.0) \| 1855 (57.2) \| \| Decreased physical activity \| 1068 (35.2) \| 1168 (39.1) \| 1221 (40.6) \| 1239 (42.0) \| \| Poorer sleep quality \| 902 (29.7) \| 1058 (35.5) \| 1192 (39.6) \| 1289 (43.6) \| \| Decreased sleep duration \| 876 (28.9) \| 1038 (34.8) \| 1065 (35.4) \| 1156 (39.1) \| \| Increased alcohol use \| 877 (27.3) \| 992 (30.9) \| 1053 (32.5) \| 1099 (34.5) \| \| Use of alcohol or other substances to cope \| 612 (23.6) \| 693 (27.5) \| 790 (31.0) \| 809 (33.1) \| The Table summarizes the sociodemographic and health characteristics of participants in the full analytic sample. An association between the RAT score and fear of partner was observed: fear of partner was reported by a much higher percentage of participants in the highest RAT score quartile (fourth) than in the first 3 quartiles ($6.1\%$ vs $0.6\%$, $0.6\%$, and $0.5\%$). We summarized the IPV item endorsements in the analytic samples across the 3 cohorts in eFigure 2 in Supplement 1. ## IPV and Mental Health Symptoms Accounting for depression and anxiety before exposure ascertainment, we found in meta-analyses across all 3 cohorts that experiencing IPV was associated with depression, anxiety, and PTSS during follow-up. Experiencing IPV at month 1 was associated with higher odds of depression (odds ratio [OR], 1.44; $95\%$ CI, 1.38-1.50), anxiety (OR, 1.31; $95\%$ CI, 1.26-1.36), and PTSS (OR, 1.22; $95\%$ CI, 1.15-1.29) during follow-up (Figure 1; eTable 2 in Supplement 1). Cohort-specific estimates were comparable. Figure 1.: Associations of the Relationship Assessment Tool Score at Month 1 With Mental Health and Health-Related Behavior During Follow-upEstimates were obtained with random-effects meta-analysis across the 3 cohorts. All models were adjusted for age at baseline, race and ethnicity, educational level as a proxy for socioeconomic status, health care professional status, partnership status, sexual orientation, and depression or anxiety prior to exposure ascertainment. GUTS indicates Growing Up Today Study; IES-6, 6-item Impact of Events Scale; NHS II, Nurses’ Health Study II; NHS3, Nurses’ Health Study 3; OR, odds ratio; PHQ-4, 4-item Patient Health Questionnaire; PTSS, posttraumatic stress symptoms. Fear of one’s partner at month 1 was associated with higher odds of depression (OR, 2.87; $95\%$ CI, 2.16-3.80), anxiety (OR, 2.12; $95\%$ CI, 1.38-3.26), and PTSS (OR, 1.62; $95\%$ CI, 1.13-2.34) during follow-up in a meta-analysis across cohorts (Figure 2; eTable 3 in Supplement 1). There were some differences between cohorts. Feeling afraid was not associated with PTSS in the NHS3 cohort. In the GUTS cohort, fear of one’s partner was not associated with anxiety or PTSS. The $95\%$ CIs for associations between fear of partner and mental health symptoms were relatively wide due to the small numbers of individuals reporting this item. Figure 2.: Associations Between Reported Fear of Spouse, Partner, or Significant Other at Month 1 and Mental Health and Health-Related Behavior During Follow-upEstimates were obtained with random-effects meta-analysis across the 3 cohorts. All models were adjusted for age at baseline, race and ethnicity, educational level as a proxy for socioeconomic status, health care professional status, partnership status, sexual orientation, and depression or anxiety prior to exposure ascertainment. GUTS indicates Growing Up Today Study; IES-6, 6-item Impact of Events Scale; NHS II, Nurses’ Health Study II; NHS3, Nurses’ Health Study 3; OR, odds ratio; PHQ-4, 4-item Patient Health Questionnaire; PTSS, posttraumatic stress symptoms. ## IPV and Modifiable Health Factors Overall, the magnitude of association between IPV and modifiable health factors was smaller than that observed between IPV and mental health symptoms. In a meta-analysis across cohorts, we found that experiencing IPV was associated with poorer sleep quality (OR, 1.21; $95\%$ CI, 1.16-1.26), decreased sleep duration (OR, 1.13; $95\%$ CI, 1.08-1.19), increased alcohol use (OR, 1.10; $95\%$ CI, 1.06-1.14), and use of alcohol or other substances to cope with stress (OR, 1.13; $95\%$ CI, 1.08-1.18). In the NHS II cohort only, experiencing IPV was associated with decreased physical activity (OR, 1.17; $95\%$ CI, 1.09-1.26) (Figure 1; eTable 2 in Supplement 1). Fear of one’s partner was associated with decreased sleep duration (OR, 1.48; $95\%$ CI, 1.13-1.95), poorer sleep quality (OR, 1.75; $95\%$ CI, 1.33-2.31), increased alcohol use (OR, 1.32; $95\%$ CI, 1.01-1.72), use of alcohol or other substances to cope with stress (OR, 1.55; $95\%$ CI, 1.13-2.12), and decreased physical activity (OR, 1.01; $95\%$ CI, 0.77-1.33) in a meta-analysis across cohorts. In cohort-specific estimates, there were differences. In the NHS3 cohort, fear of one’s partner was associated only with poorer sleep quality (OR, 1.82; $95\%$ CI, 1.26-2.63) and use of alcohol or other substances to cope with stress (OR, 1.59; $95\%$ CI, 1.04-2.41). In GUTS, fear was associated only with poorer sleep quality (OR, 1.84; $95\%$ CI, 1.00-3.37) and decreased sleep duration (OR, 1.98; $95\%$ CI, 1.10-3.58). In NHS II, fear was not associated with any modifiable health factor during follow-up (Figure 2; eTable 3 in Supplement 1). Results from minimally adjusted models (ie, only adjusting for age and race and ethnicity) were similar (eTables 4 and 5 in Supplement 1). ## Secondary Analyses We conducted a stratified analysis by prior IPV history in the NHS II and GUTS cohorts. The prevalence of IPV history was $37.2\%$ in NHS II and $40.2\%$ in GUTS. Analyzing the cohorts together, we observed differences in associations with PTSS by IPV history, with ORs that were greater in magnitude among participants with no IPV history (OR, 1.32; $95\%$ CI, 1.05-1.65) compared with individuals with IPV history before the pandemic (OR, 1.13; $95\%$ CI, 0.99-1.30) (Figure 3). The association with use of alcohol or other substances to cope with stress was also greater in magnitude among participants with no IPV history (OR, 1.32; $95\%$ CI, 1.11-1.57) compared with individuals with IPV history (OR, 1.05; $95\%$ CI, 0.95-1.16) (Figure 3). The $95\%$ CIs in these analyses were wide, indicating a lack of precision in estimates. Figure 3.: Stratified Analysis by Prior History of Intimate Partner ViolenceEstimates were obtained with random-effects meta-analysis across 2 cohorts. All models were adjusted for age at baseline, race and ethnicity, educational level as a proxy for socioeconomic status, health care professional status, partnership status, sexual orientation, and depression or anxiety prior to exposure ascertainment. IES-6 indicates 6-item Impact of Events Scale; IPV, intimate partner violence; OR, odds ratio; PHQ-4, 4-item Patient Health Questionnaire; PTSS, posttraumatic stress symptoms. ## Discussion Overall, prevalence of depression in this study was comparable to the prevalence reported in other population-based assessments during the COVID-19 pandemic.39,40 However, anxiety and PTSS were more prevalent, which could be explained by higher distress among active health care professionals41 and the analytic strategy we used to capture the highest level of distress during follow-up. Participants in the full sample who experienced IPV had higher odds of experiencing depression, anxiety, and PTSS over the course of the first 1.5 years of the pandemic. We also found evidence of IPV’s association with modifiable health factors. Specifically, experience of IPV was associated with shorter sleep duration, poorer sleep quality, and increased use of alcohol or other substances in all 3 cohorts. Additionally, IPV was associated with decreased physical activity in the NHS II cohort. When examining the potential implications of IPV history, we found that participants with no IPV history who were exposed to IPV during the pandemic were at higher risk for PTSS and use of alcohol or other substances to cope with stress during the pandemic than participants who were not exposed to IPV. This finding may be attributed to the timing of IPV in that recent traumatic experiences may translate into current symptoms.42,43 The findings of this study have several key implications for both prevention and intervention. Intimate partner violence is a preventable crime associated with harmful health consequences. Routine and well-implemented screening44,45 for IPV, including follow-up and referrals for those receiving positive screening results,46 is critical during a time of increased risk (eg, COVID-19 pandemic). During the first years of the pandemic, access to safe housing and health care became more limited for those living in violent situations, as capacity in shelters and transitional housing decreased and many patients shifted to virtual health care appointments at home, from where it may not have been safe to seek help from clinicians.46 The modifiable health factors examined in the current study were implicated in substantial long-term health problems.47,48,49 Response and referral training to a broad spectrum of clinicians who may interface with women experiencing IPV may allow for quick interventions on the exposure and health risk factors,50 guarding against more severe health consequences in the future. One unique aspect of this study was the assessment of IPV’s implications in the 3 cohorts of participants at different stages of their lives. While consistent patterns of associations emerged across the cohorts, some cohort-specific findings may indicate heightened vulnerability in older participants. Specifically, those who were enrolled in NHS II were more likely to report decreased physical activity after IPV exposure, but no association was identified in the 2 cohorts (GUTS and NHS3) with younger participants. Challenges in seeking support and accessing resources for older women who were in abusive relationships51 may have been exacerbated during the pandemic given the physical confinement at home. The trauma could also be compounded by elder abuse, which we were not able to assess. Promotion of physical activity among older women and increased public awareness and education about IPV are needed in times of collective stress, wherein support may come from formal and informal sources.52 ## Strengths and Limitations The current study had 3 main methodological strengths. First, we leveraged prospective data from over 13 000 female participants in 3 well-characterized longitudinal cohorts, with health assessments both before and during the COVID-19 pandemic. Second, the IPV measure we used focused on cognitive and emotional experiences of IPV, rather than episodic events. This approach aligns well with considerations about characterizing nuanced experiences of trauma during a period of heightened risk: individual responses to prolonged stress and terror may have implications for health beyond the frequency or incidence of violence.53 Third, by examining a range of health factors under a unified analytic framework, rather than focusing on a specific disorder, we provided a comprehensive overview of the outcome of IPV and found associations between domains of health, potentially minimizing the implications of biases present for any specific outcome and generating more relevant public health recommendations.54 Several limitations of this study also must be noted. First, to minimize loss to follow-up within the analytic sample and to capture severe stress response, we considered the highest distress level during follow-up for each domain separately. A future direction is therefore to examine patterns of change over a longer period of the pandemic, with repeatedly collected exposure and confounder information to further disentangle the association of IPV with mental health symptoms and modifiable health factors. Second, this study lacks representation of some groups given that the sample was composed of mostly White heterosexual women who were younger than 60 years, who had a higher socioeconomic status than the general population,28 and about whom we had limited information on IPV type or severity. Health outcomes of IPV may be exacerbated in individuals experiencing severe or certain types of violence, especially in the presence of other acute stressors, such as financial hardship and job insecurity, which we were unable to assess. We did not have information on the sex or gender identity of participants’ partners, and the study was underpowered to pursue any analysis examining effect modification by sexual orientation or by partner sex or gender identity. Additionally, IPV prevalence, severity, and implications for health are affected by racial and ethnic and socioeconomic differences due to health disparities and variations in help-seeking behaviors.55,56 Future work should explore IPV experiences and health consequences during the pandemic among males, persons with nonbinary identity, diverse racial and ethnic groups, and individuals from less advantaged socioeconomic strata. ## Conclusions This cohort study extended previous research by prospectively examining the association of IPV with worse modifiable health factors and mental health symptoms in 3 population-based US cohorts during the first 1.5 years of the COVID-19 pandemic, a unique time of collective stress. Accounting for differences in prepandemic health, the analyses provided insights into health outcomes for female participants younger than 60 years who experienced IPV early in the pandemic. Such IPV exposure had mental and physical health costs for these individuals. Screening and interventions for IPV and related health factors are needed to prevent severe, long-term health consequences. ## References 1. Breiding MJ, Basile KC, Smith SG, Black MC, Mahendra RR. Intimate Partner Violence Surveillance: Uniform Definitions and Recommended Data Elements. Version 2.0 (2015.0) 2. Bacchus LJ, Ranganathan M, Watts C, Devries K. Recent intimate partner violence against women and health: a systematic review and meta-analysis of cohort studies. BMJ Open (2018.0) 8. DOI: 10.1136/bmjopen-2017-019995 3. Smith SG, Zhang X, Basile KC. The National Intimate Partner and Sexual Violence Survey (NISVS): 2015 Data Brief – Updated Release (2018.0) 4. 4World Health Organization. Violence Against Women Prevalence Estimates, 2018. Global, Regional and National Prevalence Estimates for Intimate Partner Violence Against Women and Global and Regional Prevalence Estimates for Non-Partner Sexual Violence Against Women. World Health Organization; 2021.. Violence Against Women Prevalence Estimates, 2018. Global, Regional and National Prevalence Estimates for Intimate Partner Violence Against Women and Global and Regional Prevalence Estimates for Non-Partner Sexual Violence Against Women (2021.0) 5. Evans ML, Lindauer M, Farrell ME. A pandemic within a pandemic—intimate partner violence during Covid-19. N Engl J Med (2020.0) 383 2302-2304. DOI: 10.1056/NEJMp2024046 6. Sánchez OR, Vale DB, Rodrigues L, Surita FG. Violence against women during the COVID-19 pandemic: An integrative review. Int J Gynaecol Obstet (2020.0) 151 180-187. DOI: 10.1002/ijgo.13365 7. Kaukinen C. When stay-at-home orders leave victims unsafe at home: exploring the risk and consequences of intimate partner violence during the COVID-19 pandemic. Am J Crim Justice (2020.0) 45 668-679. DOI: 10.1007/s12103-020-09533-5 8. van Gelder N, Peterman A, Potts A. COVID-19: reducing the risk of infection might increase the risk of intimate partner violence. EClinicalMedicine (2020.0) 21 100348-100348. DOI: 10.1016/j.eclinm.2020.100348 9. Usher K, Bhullar N, Durkin J, Gyamfi N, Jackson D. Family violence and COVID-19: increased vulnerability and reduced options for support. Int J Ment Health Nurs (2020.0) 29 549-552. DOI: 10.1111/inm.12735 10. Fancourt D, Bu F, Mak HW, Paul E, Steptoe A. Covid-19 Social Study Results Release 32 (2021.0) 11. 11World Health Organization. COVID-19 and violence against women: what the health sector/system can do. Policy brief. April 7, 2020. Accessed July 2022. https://apps.who.int/iris/handle/10665/331699. (2020.0) 12. Kourti A, Stavridou A, Panagouli E. Domestic violence during the COVID-19 pandemic: a systematic review. Trauma Violence Abuse (2021.0). DOI: 10.1177/15248380211038690 13. Sorenson SB, Sinko L, Berk RA. The endemic amid the pandemic: seeking help for violence against women in the initial phases of COVID-19. J Interpers Violence (2021.0) 36 4899-4915. DOI: 10.1177/0886260521997946 14. Agüero JM. COVID-19 and the rise of intimate partner violence. World Dev (2021.0) 137. DOI: 10.1016/j.worlddev.2020.105217 15. Mason SM, Wright RJ, Hibert EN. Intimate partner violence and incidence of type 2 diabetes in women. Diabetes Care (2013.0) 36 1159-1165. DOI: 10.2337/dc12-1082 16. Mason SMP, Wright RJMD, Hibert ENMA, Spiegelman D, Forman JPMD, Rich-Edwards JWS. Intimate partner violence and incidence of hypertension in women. Ann Epidemiol (2012.0) 22 562-567. DOI: 10.1016/j.annepidem.2012.05.003 17. Campbell J, Jones AS, Dienemann J. Intimate partner violence and physical health consequences. Arch Intern Med (2002.0) 162 1157-1163. DOI: 10.1001/archinte.162.10.1157 18. Vives-Cases C, Ruiz-Cantero MT, Escribà-Agüir V, Miralles JJ. The effect of intimate partner violence and other forms of violence against women on health. J Public Health (Oxf) (2011.0) 33 15-21. DOI: 10.1093/pubmed/fdq101 19. Pengpid S, Peltzer K. Lifetime spousal violence victimization and perpetration, physical illness, and health risk behaviours among women in India. Int J Environ Res Public Health (2018.0) 15 2737. DOI: 10.3390/ijerph15122737 20. Jun H-J, Rich-Edwards JW, Boynton-Jarrett R, Wright RJ. Intimate partner violence and cigarette smoking: association between smoking risk and psychological abuse with and without co-occurrence of physical and sexual abuse. Am J Public Health (2008.0) 98 527-535. DOI: 10.2105/AJPH.2003.037663 21. Porter C, Favara M, Sánchez A, Scott D. The impact of COVID-19 lockdowns on physical domestic violence: evidence from a list randomization experiment. SSM Popul Health (2021.0) 14. DOI: 10.1016/j.ssmph.2021.100792 22. Raj A, Johns NE, Barker KM, Silverman JG. Time from COVID-19 shutdown, gender-based violence exposure, and mental health outcomes among a state representative sample of California residents. EClinicalMedicine (2020.0) 26. DOI: 10.1016/j.eclinm.2020.100520 23. Gresham AM, Peters BJ, Karantzas G, Cameron LD, Simpson JA. Examining associations between COVID-19 stressors, intimate partner violence, health, and health behaviors. J Soc Pers Relat (2021.0). DOI: 10.1177/02654075211012098 24. Oginni OA, Oloniniyi IO, Ibigbami O. Depressive and anxiety symptoms and COVID-19-related factors among men and women in Nigeria. PLoS One (2021.0) 16. DOI: 10.1371/journal.pone.0256690 25. Sabri B, Hartley M, Saha J, Murray S, Glass N, Campbell JC. Effect of COVID-19 pandemic on women’s health and safety: a study of immigrant survivors of intimate partner violence. Health Care Women Int (2020.0) 41 1294-1312. DOI: 10.1080/07399332.2020.1833012 26. 26Trudell AL, Whitmore E. Pandemic meets pandemic: understanding the impacts of Covid-19 on gender based violence services and survivors in Canada. Ending Violence Association of Canada and Anova. Published August 25, 2020. Accessed February 10, 2023. https://endingviolencecanada.org/wp-content/uploads/2020/08/FINAL.pdf 27. Armour C, Sleath E. Assessing the co-occurrence of intimate partner violence domains across the life-course: relating typologies to mental health. Eur J Psychotraumatol (2014.0) 5 24620. DOI: 10.3402/ejpt.v5.24620 28. Bao Y, Bertoia ML, Lenart EB. Origin, methods, and evolution of the three Nurses’ Health Studies. Am J Public Health (2016.0) 106 1573-1581. DOI: 10.2105/AJPH.2016.303338 29. Vandenbroucke JP, von Elm E, Altman DG. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration. PLoS Med (2007.0) 4. DOI: 10.1371/journal.pmed.0040297 30. Rich-Edwards JW, Ding M, Rocheleau CM. American frontline healthcare personnel’s access to and use of personal protective equipment early in the COVID-19 pandemic. J Occup Environ Med (2021.0) 63 913-920. DOI: 10.1097/JOM.0000000000002308 31. Smith PH, Smith JB, Earp JAL. Beyond the measurement trap: a reconstructed conceptualization and measurement of woman battering. Psychol Women Q (1999.0) 23 177-193. DOI: 10.1111/j.1471-6402.1999.tb00350.x 32. Roberts AL, Lyall K, Rich-Edwards JW, Ascherio A, Weisskopf MG. Maternal exposure to intimate partner abuse before birth is associated with autism spectrum disorder in offspring. Autism (2016.0) 20 26-36. DOI: 10.1177/1362361314566049 33. Lagdon S, Armour C, Stringer M. Adult experience of mental health outcomes as a result of intimate partner violence victimisation: a systematic review. Eur J Psychotraumatol (2014.0) 5 24794. DOI: 10.3402/ejpt.v5.24794 34. Kroenke K, Spitzer RL, Williams JB, Löwe B. An ultra-brief screening scale for anxiety and depression: the PHQ-4. Psychosomatics (2009.0) 50 613-621. PMID: 19996233 35. Thoresen S, Tambs K, Hussain A, Heir T, Johansen VA, Bisson JI. Brief measure of posttraumatic stress reactions: impact of Event Scale-6. Soc Psychiatry Psychiatr Epidemiol (2010.0) 45 405-412. DOI: 10.1007/s00127-009-0073-x 36. Stubbs A, Szoeke C. The effect of intimate partner violence on the physical health and health-related behaviors of women: a systematic review of the literature. Trauma Violence Abuse (2022.0) 23 1157-1172. DOI: 10.1177/1524838020985541 37. Gallegos AM, Trabold N, Cerulli C, Pigeon WR. Sleep and interpersonal violence: a systematic review. Trauma Violence Abuse (2021.0) 22 359-369. DOI: 10.1177/1524838019852633 38. Matsui EC, Perry TT, Adamson AS. An antiracist framework for racial and ethnic health disparities research. Pediatrics (2020.0) 146. DOI: 10.1542/peds.2020-018572 39. Arora T, Grey I, Östlundh L, Lam KBH, Omar OM, Arnone D. The prevalence of psychological consequences of COVID-19: a systematic review and meta-analysis of observational studies. J Health Psychol (2022.0) 27 805-824. DOI: 10.1177/1359105320966639 40. Cénat JM, Blais-Rochette C, Kokou-Kpolou CK. Prevalence of symptoms of depression, anxiety, insomnia, posttraumatic stress disorder, and psychological distress among populations affected by the COVID-19 pandemic: a systematic review and meta-analysis. Psychiatry Res (2021.0) 295. DOI: 10.1016/j.psychres.2020.113599 41. Young KP, Kolcz DL, O’Sullivan DM, Ferrand J, Fried J, Robinson K. Health care workers’ mental health and quality of life during COVID-19: results from a mid-pandemic, national survey. Psychiatr Serv (2021.0) 72 122-128. DOI: 10.1176/appi.ps.202000424 42. Kessler RC, Sonnega A, Bromet E, Hughes M, Nelson CB. Posttraumatic stress disorder in the National Comorbidity Survey. Arch Gen Psychiatry (1995.0) 52 1048-1060. DOI: 10.1001/archpsyc.1995.03950240066012 43. Hartley TA, Violanti JM, Sarkisian K, Andrew ME, Burchfiel CM. PTSD symptoms among police officers: associations with frequency, recency, and types of traumatic events. Int J Emerg Ment Health (2013.0) 15 241-253. PMID: 24707587 44. O’Doherty L, Hegarty K, Ramsay J, Davidson LL, Feder G, Taft A. Screening women for intimate partner violence in healthcare settings. Cochrane Database Syst Rev (2015.0) 2015. DOI: 10.1002/14651858.CD007007.pub3 45. El Morr C, Layal M. Effectiveness of ICT-based intimate partner violence interventions: a systematic review. BMC Public Health (2020.0) 20 1372. DOI: 10.1186/s12889-020-09408-8 46. Rossi FS, Shankar M, Buckholdt K, Bailey Y, Israni ST, Iverson KM. Trying times and trying out solutions: intimate partner violence screening and support for women veterans during COVID-19. J Gen Intern Med (2020.0) 35 2728-2731. DOI: 10.1007/s11606-020-05990-0 47. Huang T, Zeleznik OA, Poole EM. Habitual sleep quality, plasma metabolites and risk of coronary heart disease in post-menopausal women. Int J Epidemiol (2019.0) 48 1262-1274. DOI: 10.1093/ije/dyy234 48. St-Onge M-P, Grandner MA, Brown D. Sleep duration and quality: impact on lifestyle behaviors and cardiometabolic health: a scientific statement from the American Heart Association. Circulation (2016.0) 134 e367-e386. DOI: 10.1161/CIR.0000000000000444 49. Brick J. Medical consequences of alcohol abuse. (2004.0) 50. 50World Health Organization. Strengthening Health Systems to Respond to Women Subjected to Intimate Partner Violence or Sexual Violence: A Manual for Health Managers. World Health Organization; 2017.. Strengthening Health Systems to Respond to Women Subjected to Intimate Partner Violence or Sexual Violence: A Manual for Health Managers (2017.0) 51. Zink T, Jacobson CJ, Regan S, Pabst S. Hidden victims: the healthcare needs and experiences of older women in abusive relationships. J Womens Health (Larchmt) (2004.0) 13 898-908. DOI: 10.1089/jwh.2004.13.898 52. Zapor H, Wolford-Clevenger C, Johnson DM. The association between social support and stages of change in survivors of intimate partner violence. J Interpers Violence (2018.0) 33 1051-1070. DOI: 10.1177/0886260515614282 53. Smith PH, Earp JA, DeVellis R. Measuring battering: development of the Women’s Experience with Battering (WEB) Scale. Womens Health (1995.0) 1 273-288. PMID: 9373384 54. VanderWeele TJ. Outcome-wide epidemiology. Epidemiology (2017.0) 28 399-402. DOI: 10.1097/EDE.0000000000000641 55. Cho H. Racial differences in the prevalence of intimate partner violence against women and associated factors. J Interpers Violence (2012.0) 27 344-363. DOI: 10.1177/0886260511416469 56. Lipsky S, Caetano R, Field CA, Larkin GL. The role of intimate partner violence, race, and ethnicity in help-seeking behaviors. Ethn Health (2006.0) 11 81-100. DOI: 10.1080/13557850500391410
--- title: 'Weight Loss Expectations of Adults With Binge Eating: Cross-sectional Study With a Human-Centered Design Approach' journal: JMIR Formative Research year: 2023 pmcid: PMC10015344 doi: 10.2196/40506 license: CC BY 4.0 --- # Weight Loss Expectations of Adults With Binge Eating: Cross-sectional Study With a Human-Centered Design Approach ## Abstract ### Background People tend to overestimate their expectations for weight loss relative to what is achievable in a typical evidence-based behavioral weight management program, which can impact treatment satisfaction and outcomes. We are engaged in formative research to design a digital intervention that addresses binge eating and weight management; thus, understanding expectations among this group can inform more engaging intervention designs to produce a digital intervention that can achieve greater clinical success. Studies examining weight loss expectations have primarily focused on people who have overweight or obesity. Only one study has investigated weight loss expectations among people with binge eating disorder, a population that frequently experiences elevated weight and shape concerns and often presents to treatment with the goal of losing weight. ### Objective The aim of the study is to investigate differences in weight loss expectations among people with varying levels of binge eating to inform the design of a digital intervention for binge eating and weight management. Such an evaluation may be crucial for people presenting for a digital intervention, given that engagement and dropout are notable problems for digital behavior change interventions. We tested the hypotheses that [1] people who endorsed some or recurrent binge eating would expect to lose more weight than those who did not endorse binge eating and [2] people who endorsed a more severe versus a low or moderate overvaluation of weight and shape would have higher weight loss expectations. ### Methods A total of 760 adults ($$n = 504$$, $66\%$ female; $$n = 441$$, $58\%$ non-Hispanic White) completed a web-based screening questionnaire. One-way ANOVAs were conducted to explore weight loss expectations for binge eating status as well as overvaluation of shape and weight. ### Results Weight loss expectations significantly differed by binge eating status. Those who endorsed some and recurrent binge eating expected to lose more weight than those who endorsed no binge eating. Participants with severe overvaluation of weight or shape expected to lose the most weight compared to those with low or moderate levels of overvaluation of weight and shape. ### Conclusions In the sample, people interested in a study to inform a digital intervention for binge eating and weight management overestimated their expectations for weight loss. Given that weight loss expectations can impact treatment completion and success, it may be important to assess and modify weight loss expectations among people with binge eating prior to enrolling in a digital intervention. Future work should design and test features that can modify these expectations relative to individuals’ intended treatment goals to facilitate engagement and successful outcomes in a digital intervention. ## Introduction Human-centered design is an approach to ground interventions on the needs of the people who will be using them and the contexts in which they will be implemented [1-3]. We are engaged in a program of research applying a human-centered design approach to develop a digital intervention that addresses binge eating and weight management. Digital interventions offer an opportunity to scale effective interventions to those who need them [4], and research supports their efficacy for weight loss among people with overweight and obesity [5]. However, few studies have investigated digital interventions for weight loss among people with binge eating—a group estimated to make up between $13\%$ and $30\%$ of people seeking weight loss treatment [6]. People presenting for weight loss interventions tend to overestimate their expectations for weight loss. Among individuals seeking treatment to lose weight, most endorse a desire to lose more than $10\%$ of their current body weight [7,8]. However, weight loss programs for adults typically aim to help people lose $5\%$-$10\%$ of their body weight [9]. Further, across trials of commercial weight loss interventions (eg, Weight Watchers, Jenny Craig, and Nutrisystem), participants lose, on average, only between $0\%$ and $10\%$ of their body weight [10]. When individuals are asked about their weight loss expectations, the majority report that they would be disappointed to lose between $5\%$ and $10\%$ of their body weight [11,12], and in one study, nearly half ($47\%$) of the participants attained a weight loss that was less than their “disappointed” weight loss goal by the end of treatment [12]. Discrepancies between weight loss expectations and actual weight loss treatment outcomes are problematic because participants’ satisfaction with their weight loss is a determinant of weight loss maintenance, meaning people who are less satisfied with their weight loss are less successful at maintaining the weight that they lost in treatment over time [13,14]. People with obesity presenting for treatment with realistic weight loss expectations also show lower rates of dropout during treatment compared with those who had unrealistically high expectations [14]. Consequently, understanding people’s pretreatment expectations for a weight loss intervention may be important for ensuring that these individuals have greater success during and following treatment and fewer dropouts. People who engage in binge eating have been understudied in the context of weight loss expectations in treatment. Just one study evaluated weight loss expectations among people with binge eating disorder (BED) and showed that people with BED expected to lose more weight than what expert and governmental guidelines deemed a reasonable amount of weight loss [15]. However, this study only investigated weight loss expectations among people meeting full-threshold diagnostic criteria for BED (based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition [DSM-IV], which had a more stringent frequency requirement for diagnosis than currently used in the DSM-5). It is not known what the weight loss expectations are for people with varying levels of binge eating (ie, threshold-level recurrent binge eating compared to subthreshold levels or no binge eating), who encompass a broader sample of the types of individuals who present for weight loss treatment. Understanding weight loss expectations among people who engage in binge eating more broadly is important because people with binge eating frequently experience elevated concerns about their weight and shape [16,17], have heightened eating disorder psychopathology when coupled with a desire to lose weight [18], experience excess weight gain, and commonly present for treatment for binge eating with the goal of losing weight [19], all of which can undermine weight loss treatment outcomes [20-23]. Therefore, understanding weight loss expectations among people with varying levels of binge eating is important to inform the design of interventions that can be targeted at a broader array of people presenting with binge eating. Moreover, evaluating people’s expectations for treatment may be especially crucial for people presenting for a digital intervention. Engagement and dropout are notable problems for digital behavior change interventions [24], which means that digital interventions need designs that are engaging in order to achieve clinical impact. If people presenting to a digital intervention for weight management and binge eating have unrealistic expectations for weight loss, they may lose motivation or interest to engage in the program and be more likely to discontinue treatment early. Therefore, understanding the treatment expectations of this population can inform designs that sustain users’ motivation and engagement while appropriately shaping their expectations. This study investigated differences in weight loss expectations among people who were interested in participating in a study that aimed to inform the design of a digital intervention for binge eating and weight management. The differences in weight loss expectations were compared among 3 groups: people who had not experienced any binge eating, people who endorsed some binge eating, and people who endorsed recurrent binge eating (diagnostic threshold level of ≥12 episodes) in the past 3 months. The association between weight loss expectations and overvaluation of weight and shape was also explored, given that overvaluation of weight and shape may be a mechanism by which people with binge eating overestimate weight loss expectations. The first hypothesis was that people who endorsed recurrent binge eating would expect to lose more weight in an intervention than people who endorsed only some or no binge eating. The second hypothesis was that people who endorsed a more severe overvaluation of weight and shape would have higher expectations for weight loss. This study is part of a larger program of formative research to design a digital intervention that addresses binge eating and weight management [25]. By understanding the needs and expectations of potential consumers of this digital intervention, such as their expectations for weight loss, a tool can be designed that most effectively meets the needs of those who will be engaging with it to ultimately ensure satisfaction and success in the digital intervention. ## Participants and Procedure Study procedures (recruitment, consent, and screening) were administered in dscout. Dscout is a qualitative market research platform with over 100,000 members who respond to surveys to be screened for eligibility and enrolled in research studies [26]. Members use this research platform primarily through their smartphones, which enables them to provide in-the-moment, in-context responses. Interested individuals responded to a web-based advertisement in dscout for a study to “understand self-tracking behaviors in mobile interventions for weight management and binge eating” [27]. After providing web-based informed consent, participants completed a brief web-based screener comprising questions developed by the study team to assess eligibility, including whether these individuals would be “be willing to use an app to help you lose weight and manage your binge eating.” The screening was automatically ended if respondents endorsed that they were currently pregnant or not interested in losing weight. As part of the web-based screen, individuals were asked to self-report their height and weight (to calculate their BMI, kg/m2) and report if they had ever experienced an episode of binge eating, which was defined to the respondents as “when someone eats an unusually large amount of food and feels a sense of loss of control while eating.” Those who indicated “yes” to binge eating were prompted to report how many binge eating episodes they had in the past 3 months. The respondents were also asked to report the amount of weight loss they perceived would be reasonable or achievable in 4 months, which reflects the duration of the intervention that was being designed. Finally, to assess weight and shape concerns as they related to respondents’ interest in engaging in a weight loss intervention, participants were asked: “Do any of the following statements apply to you? Check all that apply. ‘ I struggle with my weight,’ ‘I have an active interest in losing weight,’ ‘I have an active interest in changing my body size,’ ‘None of the above.’” This question was used as a proxy measure of overvaluation of weight and shape given that the length of screening in dscout (ie, limited to 20 items) precluded including longer, established measures of this construct. Of the 818 individuals who initiated the screener, 45 ($5.5\%$) were screened out automatically (ie, did not indicate an active interest in losing weight or indicated they were pregnant). An additional 13 ($1.5\%$) participants who indicated a current weight that would be incompatible with life (ie, ≤50 pounds [22.7 kg]) were excluded. Thus, 760 participants were included in the analyses. ## Ethics Approval This study was approved by the Northwestern University Institutional Review Board (STU00213531). All participating individuals provided web-based informed consent. Participants were not provided compensation for completing the web-based screener. ## Analyses Since weight loss expectations are contingent on people’s current weight, participants’ percent of expected weight loss was calculated as their expected weight loss divided by their current weight. Weight loss expectation values were imputed for 50 participants who reported a weight loss expectation that was reasonably assumed to be their goal weight rather than their expectation for weight loss (eg, an expectation to lose 190 pounds [86.2 kg] of their current weight of 200 pounds [90.7 kg]). For these individuals, their weight loss expectation was imputed as the difference between their current weight and their reported weight loss expectation value (eg, following the above example, the imputed value was 10 pounds [4.5 kg]). Analyses were run with and without these 50 participants, and results remained consistent across groups; therefore, we report on the full sample. The BMI was calculated for each participant. The imperial equation was used: weight in pounds multiplied by 703, divided by height in inches squared. Regarding binge eating, respondents were grouped as those who endorsed no binge eating in the last 3 months, some binge eating episodes (≥1 but <12 episodes in the last 3 months), and recurrent binge eating episodes (≥12 episodes in the last 3 months). Using our proxy measure for overvaluation of weight and shape, the items participants endorsed regarding their struggles with their weight and body size were summed, yielding 3 groups: low (1 statement selected), moderate (2 statements selected), and severe (3 statements selected). Analyses were conducted using SPSS software (version 27; IBM Corp). A 1-way ANOVA was conducted with posthoc comparisons using Tukey honestly significant difference tests to compare differences between binge eating groups on age, BMI, and weight loss expectations. A chi-square test was conducted to explore differences in the overvaluation of weight and shape severity levels by binge eating status. To explore differences in weight loss expectations by the overvaluation of weight and shape severity level, an ANOVA was conducted. A 2-way ANOVA between groups was used to explore the interaction between binge eating status and weight and shape severity levels on weight loss expectations. P values <.05 were considered statistically significant and η2 was calculated to determine effect size. ## Overview Demographic information for the full sample ($$n = 760$$) and binge eating group is presented in Table 1. Participants had a mean age of 34.4 (SD 11.3) years and a mean BMI of 29.33 (SD 7.7) kg/m2. There was no substantial difference between binge eating groups in age. There was a statistically significant difference in BMI between binge eating groups (F2,750=8.37, $P \leq .001$, η2=0.02), with those in the some and recurrent binge eating groups reporting significantly higher BMI scores than those with no binge eating, although this was a small effect. The majority of participants identified as female ($$n = 504$$, $66.3\%$) and non-Hispanic White ($$n = 441$$, $58\%$). Table 1 \| Characteristic \| Characteristic.1 \| Full sample (N=760) \| Recurrent binge eating (n=199) \| Some binge eating (n=318) \| No binge eating (n=243) \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Age (years), mean (SD) \| Age (years), mean (SD) \| 34.4 (11.3) \| 34.3 (11.1) \| 35.0 (11.3) \| 33.9 (11.5) \| \| BMI (kg/m2), mean (SD) \| BMI (kg/m2), mean (SD) \| 29.3 (7.7) \| 30.1 (7.2) \| 30.1 (8.5) \| 27.7 (6.7) \| \| Gender, n (%) \| Gender, n (%) \| Gender, n (%) \| Gender, n (%) \| Gender, n (%) \| Gender, n (%) \| \| \| Female \| 504 (66.3) \| 121 (60.8) \| 222 (69.8) \| 161 (66.2) \| \| \| Male \| 242 (31.8) \| 75 (37.7) \| 89 (28.0) \| 78 (32.1) \| \| \| Nonbinary \| 7 (0.9) \| 1 (0.5) \| 3 (0.9) \| 3 (1.2) \| \| \| Transgender man \| 2 (0.2) \| 0 (0.0) \| 1 (0.3) \| 1 (0.4) \| \| \| Prefer not to say \| 5 (0.7) \| 2 (1.0) \| 3 (0.9) \| 0 (0.0) \| \| Race or ethnicity, n (%) \| Race or ethnicity, n (%) \| Race or ethnicity, n (%) \| Race or ethnicity, n (%) \| Race or ethnicity, n (%) \| Race or ethnicity, n (%) \| \| \| White \| 441 (58.0) \| 109 (54.8) \| 196 (61.6) \| 136 (56.0) \| \| \| Black \| 126 (16.6) \| 35 (17.6) \| 48 (15.1) \| 43 (17.7) \| \| \| Hispanic \| 81 (10.6) \| 23 (11.6) \| 31 (9.7) \| 27 (11.1) \| \| \| Asian \| 71 (9.3) \| 20 (10.1) \| 27 (8.5) \| 24 (9.9) \| \| \| American Indian or Alaskan Native \| 5 (0.7) \| 0 (0.0) \| 4 (1.3) \| 1 (0.4) \| \| \| Middle Eastern or North African \| 10 (1.3) \| 4 (2.0) \| 1 (0.3) \| 5 (2.1) \| \| \| Pacific Islander or Native Hawaiian \| 1 (0.1) \| 1 (0.5) \| 0 (0.0) \| 0 (0.0) \| \| \| Others \| 1 (0.1) \| 0 (0.0) \| 0 (0.0) \| 1 (0.4) \| \| \| Prefer not to say \| 12 (1.6) \| 3 (1.5) \| 4 (1.3) \| 5 (2.1) \| \| \| Did not answer \| 12 (1.6) \| 4 (2.0) \| 7 (2.2) \| 1 (0.4) \| \| Percentage weight loss expectations, mean (SD) \| Percentage weight loss expectations, mean (SD) \| 10.4 (5.9) \| 11.1 (4.8) \| 10.8 (6.1) \| 9.4 (6.2) \| \| Overvaluation of weight and shape, n (%) \| Overvaluation of weight and shape, n (%) \| \| \| \| \| \| \| Low \| 187 (24.6) \| 25 (12.6) \| 67 (21.1) \| 95 (39.1) \| \| \| Moderate \| 166 (21.8) \| 31 (15.6) \| 65 (20.4) \| 70 (28.8) \| \| \| Severe \| 407 (53.6) \| 143 (71.9) \| 186 (58.5) \| 78 (32.1) \| ## Weight Loss Expectations There was a statistically significant difference in weight loss expectations by binge eating status, although the effect size was small (F2,757=6.08, $$P \leq .002$$, η2=0.016). Posthoc comparisons indicated that participants who endorsed recurrent binge eating expected to lose the most weight ($11.1\%$ ±$4.8\%$ of their weight in 4 months), followed by participants who endorsed some binge eating ($10.8\%$ ±$6.1\%$ of their weight), and participants who endorsed no binge eating ($9.4\%$ ±$6.2\%$ of their weight). The mean score for those with no binge eating was significantly different from those with some binge eating ($$P \leq .01$$) and those with recurrent binge eating ($$P \leq .006$$). There was no statistically significant difference in weight loss expectations between those who endorsed some binge eating and those who endorsed recurrent binge eating. Controlling for BMI, the difference in weight loss expectations by binge eating status was slightly attenuated with trend level significance and small effect size (F2,749=2.82, $$P \leq .06$$, η2=0.01). ## Relationship Between Overvaluation of Weight and Shape, Binge Eating, and Weight Loss Expectations There was a statistically significant association between binge eating status and overvaluation of weight and shape, with a medium effect size (χ24 [$$n = 760$$]=78.12, $P \leq .001$, Φ=0.32), indicating more binge eating was associated with greater severity of overvaluation of weight and shape, as shown in Table 1. There also was a statistically significant difference in weight loss expectations by the overvaluation of weight and shape severity, with a medium-to-large effect size (F2,757=42.87, $P \leq .001$, η2=0.11). Posthoc comparisons showed that participants who endorsed severe overvaluation of weight and shape expected to lose more weight ($12.1\%$ ±$5.5\%$ of their body weight in 4 months) than participants who endorsed moderate overvaluation of weight and shape ($9.4\%$ ±$6.3\%$ of their weight) and low overvaluation of weight and shape ($7.7\%$ ±$5.0\%$ of their weight). The mean score for participants with severe overvaluation of weight and shape was significantly different from those with moderate overvaluation of weight and shape ($P \leq .001$) and those with low overvaluation of weight and shape ($P \leq .001$). Similarly, the mean score for participants who endorsed moderate overvaluation of weight and shape differed significantly from participants who endorsed low overvaluation of weight and shape ($$P \leq .02$$). Controlling for BMI, the difference in weight loss expectation by the overvaluation of weight and shape severity remained significant, with a medium effect size (F2,749=23.25, $P \leq .001$, ηp2=0.06). The interaction effect between binge eating status and overvaluation of weight shape on weight loss expectations was not statistically significant, using the more stringent P value of <.01, given that the Levene test of equality of error variances was significant (F2,751=2.47, $$P \leq .04$$, ηp2=0.013). There was a statistically significant main effect for overvaluation of shape and weight, with a medium-to-large effect size (F2,751=31.075, $P \leq .001$, ηp2=0.076). The main effect for binge eating status did not reach statistical significance (F2,751=0.695, $$P \leq .50$$, ηp2=0.002). ## Principal Findings Given that people presenting for weight loss treatment commonly overestimate their expectations for weight loss [7], it is important to understand whether the presence of binge eating contributes to or exacerbates this problem, which could have implications for the design of digital interventions for people with binge eating. This study examined whether people who endorse binge eating, compared to people without binge eating, expect to lose more weight in interventions for weight management and binge eating, and whether more severe overvaluation of weight and shape is associated with higher weight loss expectations. Findings confirmed our hypotheses, as we found that higher weight loss expectations were associated with both increased overvaluation of weight and shape and more frequent binge eating. In addition, the substantial main effect of overvaluation of weight and shape in a test of the interaction between these constructs suggests that overvaluation of weight and shape contributes to weight loss expectations even beyond the presence of binge eating. To date, only one study has investigated weight loss expectations among people with binge eating, and it showed that people with BED are expected to lose more weight than is recommended by experts and governmental guidelines [15]. This study extended the prior work by documenting that individuals who endorsed any binge eating (ie, some binge eating and recurrent binge eating) expected to lose more weight in treatment than people who did not endorse binge eating. In general, people presenting for weight loss treatment often expect to lose more weight than is typically achievable [7], and this study showed that the presence of binge eating may exacerbate this expectation. We also found that these individuals, on average, expected to lose more than $10\%$ of their body weight, which exceeds what is typically achieved in behavioral weight loss programs [9,10]. These findings highlight the importance of shaping the expectations for weight change that individuals with binge eating can likely achieve in treatment while sustaining their interest in pursuing behavior change. Indeed, modifying these expectations may contribute to better overall engagement with a digital intervention. Digital interventions face engagement challenges [24,28], in part because they lack some of the cues and features that face-to-face treatment with a practitioner offers. This means that digital interventions have greater pressure to use designs that “get it right” in meeting consumers’ needs, preferences, and goals. Because unrealistic weight loss expectations have been associated with lower treatment satisfaction and outcomes in face-to-face weight loss programs [13,14], a digital intervention for weight management must be designed effectively to support this misalignment. This could include designs that help people with binge eating adjust their expectations before or at the start of initiating the digital intervention, as well as features within the intervention that consistently promote weight change expectations that are realistic and achievable, such as in the app’s content and self-monitoring tools. As a more detailed example of a potential feature, consider that behavioral interventions often index people’s goals at the beginning of treatment as a way to increase motivation and sustain engagement. Knowing that people with binge eating are likely to select goals that are not aligned with established recommendations for weight loss, one way to reduce later discrepancies is for the intervention to provide users with goal prompts that model realistic weight change and other health outcomes as well as psychoeducation on the intent of the intervention in building sustainable changes to promote long-term health [29]. This feature could facilitate user agency while also ensuring that a user’s goals are aligned with what is typical in a weight loss program for this population. Similarly, psychoeducation around why these prompts were provided may help users understand the clinical rationale behind these goals. Future work will need to investigate these intervention design ideas, as well as others that are effective for modifying these expectations and acceptable to people with binge eating. ## Limitations The strengths of this study include the large sample of people with varying levels of binge eating recruited from across the United States. Though these findings were exploratory as a secondary analysis, they extend our understanding of the impact of binge eating on weight loss expectations, which is important because unrealistic expectations could have implications for individuals’ success in treatment. Study limitations should also be noted. Because this was a secondary analysis of screening data for enrollment into a subsequent study, our screening measure was self-created for the purpose of screening and therefore was not a validated tool, which included a proxy measure of overvaluation of weight and shape. Assessment of binge eating was via self-report, and self-reported binge eating can differ from objective measures [30] as individuals may not be able to correctly assess a binge eating episode relative to the clinical definition or can misjudge the frequency of binge eating behaviors. Self-reported height and weight, as was assessed in our screener, can also differ from objective measurements, although there is generally strong agreement between self-reported weights and weights measured using clinic or electronic scales that transfer data back to researchers [31-33]. Additionally, we did not assess for BED or other psychiatric diagnoses since the presence and frequency of binge eating (not BED) was the focus of the study for which respondents were completing the screener. The study was conducted on the internet, and there have been recent concerns with the validity of participants’ responses in web-based research platforms for eating disorder research [34], although this has not been shown for the platform we used. Lastly, participants were recruited for a study to understand self-tracking behaviors in mobile interventions for binge eating and weight management, but they were not presenting for treatment. Therefore, it may be beneficial to assess weight loss expectations among people with binge eating who are enrolling in digital interventions for binge eating and weight management. ## Conclusions In this study, we showed that weight loss expectations were higher among people with binge eating (some or recurrent) compared to those with no binge eating, and those with severe overvaluation of weight and shape experienced higher weight loss expectations than those with low to moderate levels of overvaluation. As a next step in the design process to create a digital intervention for managing binge eating and weight, findings indicate that future work should now focus on designing and testing strategies and features that can modify weight loss expectations relative to individuals’ intended treatment goals while still maintaining their motivation to change their behavior. Indeed, addressing weight change expectancies explicitly and early on may increase the likelihood that an individual will engage with a digital intervention fully and positively impact treatment outcomes. ## Data Availability The data analyzed for this study are available from the corresponding author on reasonable request. ## References 1. Norman DA, Draper SW. User Centered System Design: New Perspectives on Human-Computer Interaction (1986.0) 2. Norman DA. The Psychology of Everyday Things (1988.0) 3. Graham AK, Wildes JE, Reddy M, Munson SA, Barr Taylor C, Mohr DC. User-centered design for technology-enabled services for eating disorders. Int J Eat Disord (2019.0) 52 1095-1107. DOI: 10.1002/eat.23130 4. Connolly SL, Kuhn E, Possemato K, Torous J. Digital clinics and mobile technology implementation for mental health care. Curr Psychiatry Rep (2021.0) 23 38. DOI: 10.1007/s11920-021-01254-8 5. Hutchesson MJ, Rollo ME, Krukowski R, Ells L, Harvey J, Morgan PJ, Callister R, Plotnikoff R, Collins CE. eHealth interventions for the prevention and treatment of overweight and obesity in adults: a systematic review with meta-analysis. Obes Rev (2015.0) 16 376-392. DOI: 10.1111/obr.12268 6. Wilfley DE, Citrome L, Herman BK. Characteristics of binge eating disorder in relation to diagnostic criteria. Neuropsychiatr Dis Treat (2016.0) 12 2213-2223. DOI: 10.2147/ndt.s107777 7. Daigle KM, Gang CH, Kopping MF, Gadde KM. Relationship between perceptions of obesity causes and weight loss expectations among adults. J Nutr Educ Behav (2019.0) 51 86-90. DOI: 10.1016/j.jneb.2018.08.015 8. Fabricatore AN, Wadden TA, Womble LG, Sarwer DB, Berkowitz RI, Foster GD, Brock JR. The role of patients' expectations and goals in the behavioral and pharmacological treatment of obesity. Int J Obes (Lond) (2007.0) 31 1739-1745. DOI: 10.1038/sj.ijo.0803649 9. Wilfley DE, Hayes JF, Balantekin KN, Van Buren DJ, Epstein LH. Behavioral interventions for obesity in children and adults: evidence base, novel approaches, and translation into practice. Am Psychol (2018.0) 73 981-993. DOI: 10.1037/amp0000293 10. Gudzune KA, Doshi RS, Mehta AK, Chaudhry ZW, Jacobs DK, Vakil RM, Lee CJ, Bleich SN, Clark JM. Efficacy of commercial weight-loss programs: an updated systematic review. Ann Intern Med (2015.0) 162 501-512. DOI: 10.7326/m14-2238 11. Pétré B, Scheen A, Ziegler O, Donneau AF, Dardenne N, Husson E, Albert A, Guillaume M. Weight loss expectations and determinants in a large community-based sample. Prev Med Rep (2018.0) 12 12-19. DOI: 10.1016/j.pmedr.2018.08.005 12. Foster GD, Wadden TA, Vogt RA, Brewer G. What is a reasonable weight loss? Patients' expectations and evaluations of obesity treatment outcomes. J Consult Clin Psychol (1997.0) 65 79-85. DOI: 10.1037/0022-006x.65.1.79 13. Foster GD, Phelan S, Wadden TA, Gill D, Ermold J, Didie E. Promoting more modest weight losses: a pilot study. Obes Res (2004.0) 12 1271-7. DOI: 10.1038/oby.2004.161 14. Dalle Grave R, Calugi S, Molinari E, Petroni ML, Bondi M, Compare A, Marchesini G. Weight loss expectations in obese patients and treatment attrition: an observational multicenter study. Obes Res (2005.0) 13 1961-9. DOI: 10.1038/oby.2005.241 15. Masheb RM, Grilo CM. Weight loss expectations in patients with binge-eating disorder. Obes Res (2002.0) 10 309-14. DOI: 10.1038/oby.2002.44 16. Grilo CM, Hrabosky JI, White MA, Allison KC, Stunkard AJ, Masheb RM. Overvaluation of shape and weight in binge eating disorder and overweight controls: refinement of a diagnostic construct. J Abnorm Psychol (2008.0) 117 414-419. DOI: 10.1037/0021-843x.117.2.414 17. Ortiz SN, Forrest LN, Kinkel-Ram SS, Jacobucci RC, Smith AR. Using shape and weight overvaluation to empirically differentiate severity of other specified feeding or eating disorder. J Affect Disord (2021.0) 295 446-452. DOI: 10.1016/j.jad.2021.08.049 18. De Young KP, Lavender JM, Anderson DA. Binge eating is not associated with elevated eating, weight, or shape concerns in the absence of the desire to lose weight in men. Int J Eat Disord (2010.0) 43 732-6. DOI: 10.1002/eat.20779 19. de Zwaan M. Binge eating disorder and obesity. Int J Obes Relat Metab Disord (2001.0) 25 Suppl 1 S51-5. DOI: 10.1038/sj.ijo.0801699 20. Mitchell JE. Medical comorbidity and medical complications associated with binge-eating disorder. Int J Eat Disord (2016.0) 49 319-23. DOI: 10.1002/eat.22452 21. Meany G, Conceição E, Mitchell JE. Binge eating, binge eating disorder and loss of control eating: effects on weight outcomes after bariatric surgery. Eur Eat Disord Rev (2014.0) 22 87-91. DOI: 10.1002/erv.2273 22. Sarwer DB, Allison KC, Wadden TA, Ashare R, Spitzer JC, McCuen-Wurst C, LaGrotte C, Williams NN, Edwards M, Tewksbury C, Wu J. Psychopathology, disordered eating, and impulsivity as predictors of outcomes of bariatric surgery. Surg Obes Relat Dis (2019.0) 15 650-655. DOI: 10.1016/j.soard.2019.01.029 23. Goldschmidt AB. Are loss of control while eating and overeating valid constructs? A critical review of the literature. Obes Rev (2017.0) 18 412-449. DOI: 10.1111/obr.12491 24. Yardley L, Spring BJ, Riper H, Morrison LG, Crane DH, Curtis K, Merchant GC, Naughton F, Blandford A. Understanding and promoting effective engagement with digital behavior change interventions. Am J Prev Med (2016.0) 51 833-842. DOI: 10.1016/j.amepre.2016.06.015 25. Graham AK, Munson SA, Reddy M, Neubert SW, Green EA, Chang A, Spring B, Mohr DC, Wildes JE. Integrating user-centered design and behavioral science to design a mobile intervention for obesity and binge eating: mixed methods analysis. JMIR Form Res (2021.0) 5 e23809. DOI: 10.2196/23809 26. Diary. dscout 27. Liu J, Munson SA, Chang A, Voss C, Graham AK. Understanding self-monitoring to inform a mobile intervention for binge eating and weight management: A proof-of-concept randomized trial. Int J Eat Disord (2022.0) 55 642-652. DOI: 10.1002/eat.23700 28. Graham AK, Lattie EG, Mohr DC. Experimental therapeutics for digital mental health. JAMA Psychiatry (2019.0) 76 1223-1224. DOI: 10.1001/jamapsychiatry.2019.2075 29. Cardel MI, Newsome FA, Pearl RL, Ross KM, Dillard JR, Miller DR, Hayes JF, Wilfley D, Keel PK, Dhurandhar EJ, Balantekin KN. Patient-centered care for obesity: how health care providers can treat obesity while actively addressing weight stigma and eating disorder risk. J Acad Nutr Diet (2022.0) 122 1089-1098. DOI: 10.1016/j.jand.2022.01.004 30. Berg KC, Peterson CB, Frazier P, Crow SJ. Convergence of scores on the interview and questionnaire versions of the Eating Disorder Examination: a meta-analytic review. Psychol Assess (2011.0) 23 714-24. DOI: 10.1037/a0023246 31. Harvey-Berino J, Krukowski RA, Buzzell P, Ogden D, Skelly J, West DS. The accuracy of weight reported in a web-based obesity treatment program. Telemed J E Health (2011.0) 17 696-9. DOI: 10.1089/tmj.2011.0032 32. Ross KM, Eastman A, Wing RR. Accuracy of self-report versus objective smart-scale weights during a 12-week weight management intervention. Obesity (Silver Spring) (2019.0) 27 385-390. DOI: 10.1002/oby.22400 33. Krukowski RA, Ross KM. Measuring weight with electronic scales in clinical and research settings during the coronavirus disease 2019 pandemic. Obesity (Silver Spring) (2020.0) 28 1182-1183. DOI: 10.1002/oby.22851 34. Burnette CB, Luzier JL, Bennett BL, Weisenmuller CM, Kerr P, Martin S, Keener J, Calderwood L. Concerns and recommendations for using Amazon MTurk for eating disorder research. Int J Eat Disord (2022.0) 55 263-272. DOI: 10.1002/eat.23614
--- title: 'Development of an Individualized Responsive Feeding Intervention—Learning Early Infant Feeding Cues: Protocol for a Nonrandomized Study' journal: JMIR Research Protocols year: 2023 pmcid: PMC10015354 doi: 10.2196/44329 license: CC BY 4.0 --- # Development of an Individualized Responsive Feeding Intervention—Learning Early Infant Feeding Cues: Protocol for a Nonrandomized Study ## Abstract ### Background Responsive infant feeding occurs when a parent recognizes the infant’s cues of hunger or satiety and responds promptly to these cues. It is known to promote healthy dietary patterns and infant weight gain and is recommended as part of the Dietary Guidelines for Americans. However, the use of responsive infant feeding can be challenging for many parents. Research is needed to assist caregivers recognize infant hunger or satiety cues and overcoming barriers to using responsive infant feeding. ### Objective The Learning Early Infant Feeding Cues (LEIFc) intervention was designed to fill this gap by using a validated coaching approach, SS-OO-PP-RR (“super,” Setting the Stage, Observation and Opportunities, Problem Solving and Planning, Reflection and Review), to promote responsive infant feeding. Guided by the Obesity-Related Behavioral Intervention Trials model, this study aims to test the feasibility and fidelity of the LEIFc intervention in a group of mother-infant dyads. ### Methods This pre-post quasi-experimental study with no control group will recruit mothers ($$n = 30$$) in their third trimester (28 weeks and beyond) of pregnancy from community settings. Study visit 1 will occur prenatally in which written and video material on infant feeding and infant hunger and satiety cues is provided. Demographic information and plans for infant feeding are also collected prenatally via self-report surveys. The use of responsive infant feeding via subjective (survey) and objective (video) measures is recorded before (study visit 2, 1 month post partum) and after (study visit 5, 4 months post partum) intervention. Coaching on responsive infant feeding during a feeding session is provided by a trained interventionist using the SS-OO-PP-RR approach at study visits 3 (2 months post partum) and 4 (3 months post partum). Infant feeding practices are recorded via survey, and infant weight and length are measured at each postpartum study visit. Qualitative data on the LEIFc intervention are provided by the interventionist and mother. Infant feeding videos will be coded and tabulated for instances of infant cues and maternal responses. Subjective measures of responsive infant feeding will also be tabulated. The use of responsive infant feeding pre-post intervention will be analyzed using matched t tests. Qualitative data will be examined to guide intervention refinement. ### Results This study initially began in spring 2020 but was halted because of the COVID-10 pandemic. With new funding, recruitment, enrollment, and data collection began in April 2022 and will continue until April 2023. ### Conclusions After refinement, the LEIFc intervention will be tested in a pilot randomized controlled trial. The long-term goal is to implement LEIFc in the curricula of federally funded maternal-child home visiting programs that serve vulnerable populations—those that often have infant feeding practices that do not align with recommendations and are less likely to use responsive infant feeding. ### International Registered Report Identifier (IRRID) DERR1-$\frac{10.2196}{44329}$ ## Introduction Infant feeding practices, which include what and how infants are fed, contribute to infant growth and development, and thus, lifelong health [1]. Specifically, healthy infant feeding practices contribute to a healthy weight gain trajectory, which has been associated with decreased obesity later in life [2,3]. Healthy infant feeding practices include not only what infants are fed but also how infants are fed. The introduction of breastfeeding, breastfeeding until at least 6 months of age, introduction of complementary foods around 6 months and not before 4 months of age, and avoidance of juice and sugar sweetened beverages are recommended examples of what infants should be fed [1,4]. The responsive feeding approach is recommended for how infants should be fed and has been associated with healthy infant weight gain trajectories [4]. Responsive feeding occurs when a parent (or other caregiver) learns and recognizes the infant’s cues of hunger and satiety (Textbox 1) and responds promptly to these cues [5]. Responsive feeding is thought to promote the ability to recognize and respond to internal cues of hunger and satiety [6-10]. This ability to recognize these internal cues and self-regulate intake in response to physiologic need is associated with healthy weight gain during infancy as well as dietary behaviors later in life [8,11]. For these reasons, it is important that parents use the responsive feeding approach when feeding their infant. Three things must be in place for responsive feeding to occur: [1] the child signals hunger; [2] the caregiver recognizes cues and responds promptly; and [3] the child experiences a predictable response to hunger [5]. For this process to occur, caregivers must read their child’s nonverbal cues and respond contingently. The degree to which a caregiver responds contingently to a child’s communication, whether it is to a child’s hunger cues or to bids for social interaction, is widely known to predict a child’s language development [12-14]. As such, coaching caregivers to respond to feeding cues could impact overall caregiver responsiveness to their child’s communication bids, which could support the child’s early language development by extension. Interventions to promote responsive feeding have shown benefit in preventing rapid infant weight gain and promoting healthy dietary patterns in infants and young children [15-22]. Despite the correlation between responsive feeding and healthy infant weight gain, the successful use of responsive feeding by caregivers is still lacking [23]. Caregivers report that recognition of hunger and satiety cues in their infant is challenging [24,25]. Infant cues can be subtle, and it can be difficult for parents to differentiate reflexive movements such as hand to mouth and rooting. Additionally, recent literature has identified maternal characteristics such as infant feeding beliefs, age, mental health, and the mother’s own eating behavior as contributing to the use of responsive feeding [26-29]. Infant temperament and support from health care professionals and family and friends also play a role [23,26]. Therefore, more research is needed to find ways to assist caregivers in recognition of hunger and satiety cues and to overcome additional barriers to promote the use of responsive feeding during infancy. The Learning Early Infant Feeding Cues (LEIFc) intervention was designed to fill this gap by using a validated coaching approach with the mothers of new infants to promote responsive feeding. The proposed study aims to test the feasibility and fidelity of the LEIFc intervention in a group of mother-infant dyads. We aimed to [1] develop, refine, and test a responsive feeding intervention that can be used within real-world settings to serve mother-infant dyads in the community and [2] design the model with a primary emphasis on responsive feeding, but with a secondary focus on early communication to help caregivers understand the dual benefit of noticing and responding to infant cues. ## Ethics Approval The institutional review board at Florida State University approved this study (STUDY00002895). ## Theoretical Framework The Obesity-Related Behavioral Intervention Trials (ORBIT) model is used to guide the proposed study [30]. The ORBIT model uses an iterative process and was chosen because the LEIFc intervention is behavioral in nature, in the early stage of development, and focused on prevention of a chronic disease [30]. Testing fidelity and feasibility of the LEIFc intervention study falls into phase I of the ORBIT model as the intervention has been designed but needs to be refined based on preliminary testing. The findings from this initial study will provide the foundation to move to phase II and conduct a proof-of-concept trail. ## Study Design and Intervention This study uses a pre-post quasi-experimental design with no control group to examine the feasibility, fidelity, and social validity of the LEIFc intervention. The intervention was developed based on existing literature regarding enablers and barriers to the use of responsive feeding by caregivers and communication cues present early in life. The multicomponent LEIFc intervention involves maternal education on responsive feeding practices and mother-infant communication while using a family-guided caregiver coaching approach called SS-OO-PP-RR (“super,” Setting the Stage, Observation and Opportunities, Problem Solving and Planning, Reflection and Review) during feeding sessions [31,32]. The SS-OO-PP-RR approach was developed based on adult learning theory and has been successfully used with parents and childcare providers to promote development across domains in infants and young children [31-33]. For the LEIFc intervention, SS-OO-PP-RR was modified to be specific to communication during feeding to promote responsive feeding (Textbox 2). It is used as a guide for the interventionist during the coaching session but will also be used to evaluate the fidelity of the intervention. The objective measure of responsive feeding will occur through video recording of feeding sessions between mothers and infants. Videos will be coded before the intervention and then after 2 coaching sessions between the interventionist and the mother during a feeding session. The subjective measure of responsive feeding will be collected via a self-report questionnaire completed by mothers before and after the intervention study visits. ## Setting and Participants Mothers of any parity will be recruited from community settings during the third trimester (28 weeks and beyond) of a healthy pregnancy. To be eligible, mothers need to be aged 18 years or older, able to read and speak English, and agree to video recording of an infant feeding in their home. Any medical or congenital condition in the fetus that would interfere with infant feeding or growth (ie, Down syndrome or cleft lip or palate) will be excluded. Mothers will be enrolled in the third trimester and then screened again after the birth of the infant to ensure that mother and infant remain eligible to participate. An interventionist will be recruited to deliver the LEIFc intervention in the homes of new mothers. He or she will ideally have a background in maternal-child health and may be a nurse, social worker, speech-language pathologist, dietician, or early child development specialist. The interventionist will be trained on the SS-OO-PP-RR approach and the LEIFc protocol. Because the purpose of this study is to test the fidelity and feasibility of the intervention, feedback from the interventionist will be sought as part of the research protocol. ## Procedures Eligible mothers will be identified through community settings such as obstetrician offices, prenatal parenting or birth classes, social media forums that pregnant moms may frequent, and word of mouth. If eligible and interested in the study, study personnel will consent mothers for the study. Once enrolled in the study, the interventionist will implement each study visit (Table 1). Study visit 1 is conducted when the mother is in the third trimester of pregnancy. Written material on infant feeding and infant cues of hunger and satiety is provided. Additionally, 2 short videos, one on infant hunger cues and the second on infant satiety cues, are watched with the interventionist and the mother. Any questions on feeding or communication are answered by the interventionist. Study visit 2 is conducted when the infant is approximately 1 month old. At this visit, a video recording of a feeding session between the mother and the infant is taken with no interaction with the interventionist. The feeding can be from the breast or from the bottle (with breastmilk or formula). Study visits 3 (at infant age of 2 months) and 4 (at infant age of 3 months) are the intervention study visits. At each visit, the interventionist will use the SS-OO-PP-RR tool to coach the mother through the feeding to identify hunger and satiety cues and discuss ways to respond. Each of these visits is also video recorded. The final study visit is at infant age of 4 months when another video recording of the infant feeding session is taken with no coaching from the interventionist. At study visit 1, demographic data are collected along with the mother’s prenatal plans for infant feeding. At study visit 2, birth history is collected along with infant feeding practices, and the mother completes the Infant Feeding Questionnaire (IFQ). During study visits 3 and 4, infant feeding practices are collected. At study visit 5, the mother completes the IFQ and provides feedback on the study itself through a series of qualitative questions. Table 1 \| Study visit \| Timing of visit \| Components of the visit \| Visit location \| \| --- \| --- \| --- \| --- \| \| 1 \| Prenatal (28 weeks or beyond of gestation) \| Education provided on communication with your baby during feeding—videos and written material \| Video conferencing platform (Zoom) \| \| 2 \| Infant age approximately 1 month \| Complete Infant Feeding QuestionnaireVideo record the mother and baby during a feeding with no coachingCollect current infant feeding practices \| Family home \| \| 3 \| Infant age approximately 2 months \| Video record the mother and baby during a feedingProvide coaching to the mother during the feedingCollect current infant feeding practices \| Family home \| \| 4 \| Infant age approximately 3 months \| Video record the mother and baby during a feedingProvide coaching to the mother during the feedingCollect current infant feeding practices \| Family home \| \| 5 \| Infant age approximately 4 months \| Video record the mother and baby during a feeding with no coachingComplete Infant Feeding QuestionnaireCollect current infant feeding practices \| Family home \| ## Outcome Measures The primary outcome in this study is the use of responsive feeding. This will be measured before (study visit 2) and after (study visit 5) intervention. Data are obtained subjectively through evaluation of the IFQ and objectively through a video review. The IFQ will provide a subjective measure of responsive feeding. This 20-item self-report tool has been validated for use in mothers of infants [29,34,35]. Likert-type scoring is used; higher scores indicate a stronger measure of the construct. Three subscales will be used to measure responsive feeding: use of food to calm, use of food to soothe, and awareness of infant cues. The objective measure of responsive feeding will be collected via coding of the video session before and after intervention. The research team developed a coding scheme to capture both mother and infant behaviors during the feeding interactions. The coding scheme was adapted from the Responsiveness to Child Feeding Cues Scale [36] to define specific, discrete maternal behaviors that relate to responding to infant hunger and satiety cues as well as early infant communication acts. Infant behaviors are coded for hunger, satiety, and early communication acts. Research assistants will be trained on the coding scheme and will use existing videos to code with a goal of at least $85\%$ agreement achieved to confirm inter-rater reliability. The coding scheme developed by the team will be used to code the feeding interactions for study visits 2 and 5. Videos will be coded for the number and rate of maternal utterances directed toward the child and responsiveness to the child’s vocalizations using the Noldus Observer XT software. The percentage of the interactions spent jointly engaged between the mother and the infant will also be coded. Infant feeding practices, demographic characteristics (ie, maternal age, marital status, education level, and race and ethnicity), and maternal depression will be considered as covariates. Initiation and duration of breastfeeding and age at complementary food introduction will be extracted from data and tabulated into variables: duration of breastfeeding (any and exclusive) calculated in weeks and age at complementary food introduction calculated in weeks. Demographic data are collected at study visit 1, infant feeding practices at study visits 2 through 5, and maternal depression (via the Patient Health Questionnaire-9) at study visits 2 through 5. ## Statistical Analyses Power analysis for the matched-pair t test was conducted. With the lower limit of the medium effect size of 0.5 [37], sample sizes of 27 and 36 provided $80\%$ and $90\%$ power, respectively, under 1-sided directional null hypothesis and the nominal significance rate of.05. We will aim to enroll 30 participants for this feasibility and fidelity study. Once data are clean, descriptive statistics will be computed and examined. For the primary outcome, benefit will be determined by an increase in responsive feeding codes pre-post SS-OO-PP-RR coaching sessions (visits 2 and 5). IFQ subscale scores will be analyzed using matched sample t tests for pre-post coaching sessions to determine the benefit of the intervention. If the assumptions of statistical tests are not clearly met, the Mann-Whitney U test would be considered as an alternative. In a future pilot randomized controlled trial (RCT), analysis of covariance (ANCOVA) will be conducted where the difference between treatment and control groups for the postmeasure is examined with the premeasure as a covariate. The use of ANCOVA provides a statistical control in addition to the design control for the premeasure via random assignment in the RCT. ## Results This study was initially designed to begin recruitment in the spring of 2020. Because of the COVID-19 pandemic, all research activities ceased from March to September 2020. The research team modified the protocol to conduct this study with socially distanced measures in place (ie, using videoconferencing platforms for data collection) and began recruitment in October 2020. Two enrolled participants were unable to complete study visits as scheduled, and data collection was unsuccessful. New funding was sought and obtained; recruitment began again in April of 2022 and data collection soon after. It is anticipated that data collection will be complete in April of 2023. ## Anticipated Findings Although data collection for this study is ongoing, it is anticipated that the LEIFc intervention will be feasible to implement with and socially acceptable to mother-infant dyads by a trained interventionist. Additionally, it is hypothesized that fidelity of the intervention would need to be refined to better meet the needs of mother-infant dyads, especially those from high-risk populations (ie, minority groups and low socioeconomic status). Qualitative data collected from mothers (visit 5) and from the interventionist (visits 3 and 4) will be used to refine the intervention. Finally, it is hypothesized that there may be a change in responsive feeding before or after intervention; however, it cannot be attributed to the intervention until tested with a control group, in the future planned RCT. ## Implications Childhood obesity rates continue to rise. Data from the National Health and Nutrition Examination Survey (NHANES) found that between 2017 and 2020 (prepandemic), $19.7\%$ of children aged 2-19 years in the United States had obesity [38]. This is up from the previous NHANES data reporting $18.5\%$ of children with obesity [39]. This rise in childhood obesity percentage suggests that efforts in recent years are not enough and more strategies are needed to halt this epidemic. Additionally, there are a higher percentage of children who are non-Hispanic Black, Hispanic, or from families living below the poverty line with obesity ($24.8\%$, $26.2\%$, and $25.8\%$, respectively) [38]. Children with obesity are at risk for lifelong cardiometabolic conditions such as hypercholesterolemia, hypertension, and type 2 diabetes [40,41]. Children with obesity are also more likely to have mental health disorders (ie, depression and anxiety) and complications of the pulmonary, orthopedic, and gastrointestinal systems [1,42,43]. The first 1000 days of life, from conception to age 2 years, is a critical period for the development of habits that contribute to one’s weight status later in life, such as dietary preferences and behaviors [3,44-46]. Therefore, interventions implemented during the first 1000 days and targeting infants from high-risk groups (those of minority race and ethnicity and low socioeconomic status) have potential to decrease one’s obesity risk and subsequent complications. The promotion of responsive feeding is one such intervention that could be used to promote healthy infant weight gain and prevent obesity. ## Outcomes Education on and promotion of responsive feeding should begin prenatally and evolve with the developmental of the child [5]. Research has demonstrated that responsive feeding during infancy helps an individual learn to eat in response to their own internal cues opposed to eating on a schedule [8-10]. The development of this eating pattern early in life may promote healthy eating behaviors lifelong, thus contributing to healthy weight status and prevention of cardiometabolic complications [41]. In addition, this early support for maternal-child communication may enhance language development during infancy. Infants’ earliest communication acts relate to signaling hunger cues, and maternal responsiveness to a child’s cues has long been established as an important predictor of later language outcomes [14]. As such, the LEIFc intervention has the potential to support 2 positive outcomes in children—healthy weight gain and positive communication outcomes. Infant feeding in the first months after birth often occurs in the home, making this an ideal environment to promote responsive feeding [44,46]. Should the LEIFc intervention be successful, implementation in maternal-child home visiting programs would be sought. Maternal, infant, and early childhood home visiting programs, such as Early Head Start and Healthy Start, provide services to families from pregnancy up to the age of 5 years in at-risk communities [47,48]. These programs serve families from low socioeconomic backgrounds that often include those of minority status. Children from these families have higher rates of childhood obesity, cardiometabolic complications, and language delays [38,43]. The implementation of a responsive feeding intervention in these programs has potential to prevent several complications in children who receive this support, making implementation of the LEIFc intervention an important component to maternal-child home visiting program curricula. ## Conclusions The current funded study will test the feasibility and fidelity of the LEIFc intervention in a group of mother-infant dyads. The intervention has potential to promote 2 childhood outcomes: healthy weight gain and early language development. The results of this feasibility study will allow the researchers to refine the intervention and move to phase IIa of the ORBIT model, testing proof of concept of the intervention. If successful, the LEIFc intervention would be expanded to include vulnerable mother-infant dyads enrolled in maternal-child home visiting programs. Once implemented into the curricula of such programs, application and sustainability would be strong. ## Data Availability Data are available from the corresponding author upon request. ## References 1. 1AAP Committee on NutritionPediatric Nutrition, 8th ed2020Itasca, USAAmerican Academy of Pediatrics. Pediatric Nutrition, 8th ed (2020.0) 2. Ong KK, Loos RJF. Rapid infancy weight gain and subsequent obesity: systematic reviews and hopeful suggestions. Acta Paediatr (2006.0) 95 904-908. DOI: 10.1080/08035250600719754 3. Woo Baidal JA, Locks LM, Cheng ER, Blake-Lamb TL, Perkins ME, Taveras EM. Risk factors for childhood obesity in the first 1,000 days: a systematic review. Am J Prev Med (2016.0) 50 761-779. DOI: 10.1016/j.amepre.2015.11.012 4. Scientific report of the 2020 Dietary Guidelines Advisory Committee: advisory report to the Secretary of Agriculture and the Secretary of Health and Human Services. Dietary Guidelines Advisory Committee (2020.0) 5. Pérez-Escamilla R, Segura-Pérez S, Lott M. Feeding guidelines for infants and young toddlers. Nutr Today (2017.0) 52 223-231. DOI: 10.1097/NT.0000000000000234 6. Black MM, Aboud FE. Responsive feeding is embedded in a theoretical framework of responsive parenting. J Nutr (2011.0) 141 490-494. DOI: 10.3945/jn.110.129973 7. Hodges EA, Wasser HM, Colgan BK, Bentley ME. Development of feeding cues during infancy and toddlerhood. MCN Am J Matern Child Nurs (2016.0) 41 244-251. DOI: 10.1097/NMC.0000000000000251 8. Spill MK, Callahan EH, Shapiro MJ, Spahn JM, Wong YP, Benjamin-Neelon SE, Birch L, Black MM, Cook JT, Faith MS, Mennella JA, Casavale KO. Caregiver feeding practices and child weight outcomes: a systematic review. Am J Clin Nutr (2019.0) 109 990S-1002S. DOI: 10.1093/ajcn/nqy276 9. Ventura AK. Associations between breastfeeding and maternal responsiveness: a systematic review of the literature. Adv Nutr (2017.0) 8 495-510. DOI: 10.3945/an.116.014753 10. Silva GAP, Costa KAO, Giugliani ERJ. Infant feeding: beyond the nutritional aspects. J Pediatr (Rio J) (2016.0) 92 S2-S7. DOI: 10.1016/j.jped.2016.02.006 11. DiSantis KI, Hodges EA, Johnson SL, Fisher JO. The role of responsive feeding in overweight during infancy and toddlerhood: a systematic review. Int J Obes (Lond) (2011.0) 35 480-492. DOI: 10.1038/ijo.2011.3 12. Chang L, de Barbaro K, Deák G. Contingencies between infants' gaze, vocal, and manual actions and mothers' object-naming: longitudinal changes from 4 to 9 months. Dev Neuropsychol (2016.0) 41 342-361. DOI: 10.1080/87565641.2016.1274313 13. Gros-Louis J, West M, King A. Maternal responsiveness and the development of directed vocalizing in social interactions. Infancy (2014.0) 19 385-408. DOI: 10.111/infa.12054 14. Bornstein MH, Putnick DL, Bohr Y, Abdelmaseh M, Lee CY, Esposito G. Maternal sensitivity and language in infancy each promotes child core language skill in preschool. Early Child Res Q (2020.0) 51 483-489. DOI: 10.1016/j.ecresq.2020.01.002 15. Hohman EE, Paul IM, Birch LL, Savage JS. INSIGHT responsive parenting intervention is associated with healthier patterns of dietary exposures in infants. Obesity (Silver Spring) (2017.0) 25 185-191. DOI: 10.1002/oby.21705 16. Savage JS, Birch LL, Marini M, Anzman-Frasca S, Paul IM. Effect of the insight responsive parenting intervention on rapid infant weight gain and overweight status at age 1 year: a randomized clinical trial. JAMA Pediatr (2016.0) 170 742-749. DOI: 10.1001/jamapediatrics.2016.0445 17. Daniels LA, Mallan KM, Nicholson JM, Thorpe K, Nambiar S, Mauch CE, Magarey A. An early feeding practices intervention for obesity prevention. Pediatrics (2015.0) 136 e40-e49. DOI: 10.1542/peds.2014-4108 18. Daniels LA, Mallan KM, Nicholson JM, Battistutta D, Magarey A. Outcomes of an early feeding practices intervention to prevent childhood obesity. Pediatrics (2013.0) 132 e109-e118. DOI: 10.1542/peds.2012-2882 19. Wen LM, Baur LA, Rissel C, Xu H, Simpson JM. Correlates of body mass index and overweight and obesity of children aged 2 years: findings from the healthy beginnings trial. Obesity (Silver Spring) (2014.0) 22 1723-1730. DOI: 10.1002/oby.20700 20. Paul IM, Savage JS, Anzman SL, Beiler JS, Marini ME, Stokes JL, Birch LL. Preventing obesity during infancy: a pilot study. Obesity (Silver Spring) (2010.0) 19 353-361. DOI: 10.1038/oby.2010.182 21. Fangupo LJ, Heath AL, Williams SM, Somerville MR, Lawrence JA, Gray AR, Taylor BJ, Mills VC, Watson EO, Galland BC, Sayers RM, Hanna MB, Taylor RW. Impact of an early-life intervention on the nutrition behaviors of 2-y-old children: a randomized controlled trial. Am J Clin Nutr (2015.0) 102 704-712. DOI: 10.3945/ajcn.115.111823 22. Savage JS, Hohman EE, Marini ME, Shelly A, Paul IM, Birch LL. INSIGHT responsive parenting intervention and infant feeding practices: randomized clinical trial. Int J Behav Nutr Phys Act (2018.0) 15 64. DOI: 10.1186/s12966-018-0700-6 23. Redsell SA, Slater V, Rose J, Olander EK, Matvienko-Sikar K. Barriers and enablers to caregivers' responsive feeding behaviour: a systematic review to inform childhood obesity prevention. Obes Rev (2021.0) 22 e13228. DOI: 10.1111/obr.13228 24. McNally J, Hugh-Jones S, Hetherington MM. "An invisible map"-maternal perceptions of hunger, satiation and 'enough' in the context of baby led and traditional complementary feeding practices. Appetite (2020.0) 148 104608. DOI: 10.1016/j.appet.2020.104608 25. Appleton J, Laws R, Russell CG, Fowler C, Campbell KJ, Denney-Wilson E. Infant formula feeding practices and the role of advice and support: an exploratory qualitative study. BMC Pediatr (2018.0) 18 12. DOI: 10.1186/s12887-017-0977-7 26. Thompson AL, Wasser H, Nulty A, Bentley ME. Feeding style profiles are associated with maternal and infant characteristics and infant feeding practices and weight outcomes in African American mothers and infants. Appetite (2021.0) 160 105084. DOI: 10.1016/j.appet.2020.105084 27. Guivarch C, Charles MA, Forhan A, Heude B, de Lauzon-Guillain B. Associations between maternal eating behaviors and feeding practices in toddlerhood. Appetite (2022.0) 174 106016. DOI: 10.1016/j.appet.2022.106016 28. Temmen CD, Lipsky LM, Faith MS, Nansel TR. Prospective relations between maternal emotional eating, feeding to soothe, and infant appetitive behaviors. Int J Behav Nutr Phys Act (2021.0) 18 105. DOI: 10.1186/s12966-021-01176-x 29. Bushaw A, Lutenbacher M, Karp S, Dietrich M, Graf M. Infant feeding beliefs and practices: effects of maternal personal characteristics. J Spec Pediatr Nurs (2020.0) 25 e12294. DOI: 10.1111/jspn.12294 30. Czajkowski SM, Powell LH, Adler N, Naar-King S, Reynolds KD, Hunter CM, Laraia B, Olster DH, Perna FM, Peterson JC, Epel E, Boyington JE, Charlson ME. From ideas to efficacy: the ORBIT model for developing behavioral treatments for chronic diseases. Health Psychol (2015.0) 34 971-982. DOI: 10.1037/hea0000161 31. Brown JA, Woods JJ. Effects of a triadic parent-implemented home-based communication intervention for toddlers. J Early Interv (2015.0) 37 44-68. DOI: 10.1177/1053815115589350 32. Romano M, Woods J. Collaborative coaching with early head start teachers using responsive communication strategies. Top Early Child Spec Educ (2018.0) 38 30-41. DOI: 10.1177/0271121417696276 33. Friedman M, Woods J. Coaching teachers to support child communication across daily routines in early head start classrooms. Infants Young Child (2015.0) 28 308-322. DOI: 10.1097/IYC 34. Baughcum AE, Powers SW, Johnson SB, Chamberlin LA, Deeks CM, Jain A, Whitaker RC. Maternal feeding practices and beliefs and their relationships to overweight in early childhood. J Dev Behav Pediatr (2001.0) 22 391-408. DOI: 10.1097/00004703-200112000-00007 35. Bahorski JS, Schneider-Worthington CR, Chandler-Laney PC. Modified eating in the absence of hunger test is associated with appetitive traits in infants. Eat Behav (2020.0) 36 101342. DOI: 10.1016/j.eatbeh.2019.101342 36. Hodges EA, Johnson SL, Hughes SO, Hopkinson JM, Butte NF, Fisher JO. Development of the responsiveness to child feeding cues scale. Appetite (2013.0) 65 210-219. DOI: 10.1016/j.appet.2013.02.010 37. Cohen J. Statistical Power Analysis for the Behavioral Sciences, 2nd ed (1988.0) 38. Stierman B, Afful J, Carroll MD. National Health and Nutrition Examination Survey 2018-March 2020 Prepandemic Data Files: Development of Files and Prevalence Estimates for Selected Health Outcomes 39. Skinner AC, Ravanbakht SN, Skelton JA, Perrin EM, Armstrong SC. Prevalence of obesity and severe obesity in US children, 1999-2016. Pediatrics (2018.0) 141 e20173459. DOI: 10.1542/peds.2017-3459 40. Bhupathiraju SN, Hu FB. Epidemiology of obesity and diabetes and their cardiovascular complications. Circ Res (2016.0) 118 1723-1735. DOI: 10.1161/CIRCRESAHA.115.306825 41. Woo JG. Infant growth and long-term cardiometabolic health: a review of recent findings. Curr Nutr Rep (2019.0) 8 29-41. DOI: 10.1007/s13668-019-0259-0 42. Daniels SR, Hassink SG. The role of the pediatrician in primary prevention of obesity. Pediatrics (2015.0) 136 e275-e292. DOI: 10.1542/peds.2015-1558 43. Oberg C, Colianni S, King-Schultz L. Child health disparities in the 21st century. Curr Probl Pediatr Adolesc Health Care (2016.0) 46 291-312. DOI: 10.1016/j.cppeds.2016.07.001 44. Birch LL, Doub AE. Learning to eat: birth to age 2 y. Am J Clin Nutr (2014.0) 99 723S-728S. DOI: 10.3945/ajcn.113.069047 45. Saavedra JM, Deming D, Dattilo A, Reidy K. Lessons from the feeding infants and toddlers study in North America: what children eat, and implications for obesity prevention. Ann Nutr Metab (2013.0) 62 Suppl 3 27-36. DOI: 10.1159/000351538 46. Thompson AL, Bentley ME. The critical period of infant feeding for the development of early disparities in obesity. Soc Sci Med (2013.0) 97 288-296. DOI: 10.1016/j.socscimed.2012.12.007 47. Home visiting. Administration HRS 48. Head start programs. Services USDoHH
--- title: 'Influences on Patient Uptake of and Engagement With the National Health Service Digital Diabetes Prevention Programme: Qualitative Interview Study' journal: Journal of Medical Internet Research year: 2023 pmcid: PMC10015356 doi: 10.2196/40961 license: CC BY 4.0 --- # Influences on Patient Uptake of and Engagement With the National Health Service Digital Diabetes Prevention Programme: Qualitative Interview Study ## Abstract ### Background Digital diabetes prevention programs (digital-DPPs) are being implemented as population-based approaches to type 2 diabetes mellitus prevention in several countries to address problems with the uptake of traditional face-to-face diabetes prevention programs. However, assessments of digital-DPPs have largely focused on clinical outcomes and usability among those who have taken them up, whereas crucial information on decision-making about uptake (eg, whether a user downloads and registers on an app) and engagement (eg, the extent of use of an app or its components over time) is limited. Greater understanding of factors that influence uptake and engagement decisions may support large-scale deployments of digital-DPPs in real-world settings. ### Objective This study aimed to explore the key influences on uptake and engagement decisions of individuals who were offered the National Health Service Healthier You: Digital Diabetes Prevention Programme (NHS-digital-DPP). ### Methods A qualitative interview study was conducted using semistructured interviews. Participants were adults, aged ≥18 years, diagnosed with nondiabetic hyperglycemia, and those who had been offered the NHS-digital-DPP. Recruitment was conducted via 4 providers of the NHS-digital-DPP and 3 primary care practices in England. Interviews were conducted remotely and were guided by a theoretically informed topic guide. Analysis of interviews was conducted using an inductive thematic analysis approach. ### Results Interviews were conducted with 32 participants who had either accepted or declined the NHS-digital-DPP. In total, 7 overarching themes were identified as important factors in both decisions to take up and to engage with the NHS-digital-DPP. These were knowledge and understanding, referral process, self-efficacy, self-identity, motivation and support, advantages of digital service, and reflexive monitoring. Perceptions of accessibility and convenience of the NHS-digital-DPP were particularly important for uptake, and barriers in terms of the referral process and health care professionals’ engagement were reported. Specific digital features including health coaches and monitoring tools were important for engagement. ### Conclusions This study adds to the literature on factors that influence the uptake of and engagement with digital-DPPs and suggests that digital-DPPs can overcome many barriers to the uptake of face-to-face diabetes prevention programs in supporting lifestyle changes aimed at diabetes prevention. ## Diabetes Prevention Diabetes is a global health priority. The World Health Organization estimates that diabetes was the seventh leading cause of death across the world in 2016 [1]. In the United Kingdom, approximately 3.9 million people are diagnosed with diabetes, of which $90\%$ are diagnosed with type 2 diabetes mellitus (T2DM) [2], and a further 5 million people are estimated to have nondiabetic hyperglycemia (raised blood glucose levels or prediabetes) in England [3]. T2DM is associated with obesity, lack of physical activity, and genetic risk factors such as ethnicity [4]. For many people, T2DM may be preventable by changes to diet and activity [5,6]. There is high-quality international evidence that intensive group-based programs focusing on healthy eating, weight loss, and increased exercise can reduce the risk of progression to T2DM in people with prediabetes [7-10]. In 2016, the National Health Service Healthier You: Diabetes Prevention Programme (NHS-DPP) was established with the aim to prevent or delay the onset of T2DM in adults in England who are identified to be at high risk. The NHS-DPP delivers behavioral interventions that encourage increased physical activity and a healthy diet, in addition to weight loss, for people who are overweight. The face-to-face version of the NHS-DPP (NHS-f2f-DPP) involves individuals attending at least 13 in-person group‐based sessions over a period of at least 9 months. Early outcome data indicate that the NHS-f2f-DPP led to weight loss and glycated hemoglobin reductions among individuals who completed the program [9,11], and it is associated with reduced population incidence of T2DM [12]. However, face-to-face diabetes prevention programs (f2f-DPPs) do not suit everybody, and there are recognized barriers to attendance [13]. People who work or have caring responsibilities may find it difficult to attend in-person programs [14]. Furthermore, f2f-DPPs are usually delivered in groups, meaning that those who do not like groups may find it difficult to participate [15]. Only $56\%$ of those referred to the NHS-f2f-DPP during the first 12 months of the program took up the referral [16]. To expand the reach and access of DPPs, digital diabetes prevention programs (digital-DPPs) have been suggested as complementary alternatives to f2f-DPPs [13]. ## Digital Health Interventions for Diabetes Prevention Digital health interventions (DHIs) have been shown to be effective in increasing physical activity, changing diets, and promoting weight loss in general populations [17-19]. DHIs offer many benefits, for example, being able to integrate principles of persuasive design such as personalization, gamification, and social influence and behavior change techniques such as self-monitoring to encourage users to take up behavior change [20]. They can also capitalize on habitual smartphone and internet use among the general population to deliver intense behavior change support programs that are highly scalable [21]. There is emerging evidence to suggest that DPPs can be delivered effectively using digital technologies and achieve outcomes comparable with f2f-DPP in those who take them up [21-25]. In addition, digital-DPPs may be more acceptable to some people than f2f-DPPs, as they may be easier to fit into busy lifestyles, avoid the perceived stigma associated with attending a group, and have the potential for tailoring and personalization [15,26]. However, little is understood about how to best translate digital health solutions into real-world conditions and ways that engage and meet the needs of diverse stakeholders [20]. Early analyses of the NHS-DPP showed that the uptake was significantly lower for those of working age [27]. To address inequalities in access according to age, a digital pathway was introduced in 2019 [28]. This digital version of the NHS-DPP uses DHIs including apps that allow users to access health coaches and set and monitor goals electronically and access educational material and peer support groups and wearable technologies that monitor levels of physical activity. ## Uptake of and Engagement With Digital-DPPs Uptake and engagement with DHIs are generally agreed to be prerequisites for effectiveness. Low rates of uptake, retention, and program completion represent a major barrier to effective implementation and public health impact of digital interventions [29-33], and a greater understanding of why people take up and engage with digital-DPPs is important in promoting their widespread impact. A recent study of factors influencing decisions to attend the NHS-f2f-DPP identified knowledge and understanding of T2DM, perceptions about illness, and social support as important factors in decisions to attend [34]. A recent synthesis of qualitative studies of barriers to and facilitators of lifestyle change in people with prediabetes identified perceptions of the importance of initiating lifestyle change, strategies, and coping mechanisms for maintaining lifestyle changes and supportive relationships and environments as important [35]. Studies on attendance at other services to which individuals are referred via primary care, including diabetes self-management education [15] and weight management services [36], highlight the importance of the way in which referrals are made by health care professionals (HCPs) in encouraging attendance. There may be additional and unique factors that influence decisions to take up digital-DPPs including digital health literacy [37], technological self-efficacy, and perceived usefulness [38,39]. A recent review of the factors that promote adherence (defined by the authors as “the degree to which the user followed the program as it was designed”) to mobile health apps for a range of health conditions found intervention and patient-related factors to be important. User-friendliness, technically stable app design, customizable push notifications, personalized app content, and passive data tracking were some of the app features that influence adherence. Furthermore, certain user characteristics were associated with low adherence including lack of technical competence, low health literacy, low self-efficacy, and low education level [40]. However, to the best of our knowledge, no study has explored uptake and engagement decisions regarding DHIs in the context of diabetes prevention. ## Goals of This Study This study aimed to explore the key influences on participants’ decisions to take up and engage with the National Health Service Healthier You: Digital Diabetes Prevention Programme (NHS-digital-DPP). ## Design Qualitative study using semistructured interviews was conducted with individuals who were offered the NHS-DPP with a choice of face-to-face or digital delivery. ## Participants Individuals eligible for the NHS-DPP are adults aged ≥18 years and diagnosed with nondiabetic hyperglycemia. This is defined as having at least one glycated hemoglobin reading of 42 to 47 mmol/mol or at least one fasting blood glucose reading of 5.5 to 6.9 mmol/L in the 24 months before referral. People already diagnosed with diabetes and pregnant women are not eligible for the program. In most areas, those eligible to participate in the NHS-DPP were identified from primary care lists or during NHS Health Checks. NHS Health Checks are offered to people aged between 40 and 74 years living in England, who have not previously been diagnosed with certain conditions [41]. Participants were informed that they are at high risk of developing T2DM and offered referral to the NHS-DPP. The COVID-19 pandemic had 2 main impacts on the delivery of the NHS-DPP. First, from March 2020, the NHS-f2f-DPP option was not available, and instead, a remote group option was established, which comprised group sessions conducted via a web-based platform or telephone. Second, in August 2020, the NHS-DPP expanded access by including a self-referral route via a web-based risk tool. Eligibility to self-refer to the program was based on the Diabetes UK risk tool (a validated T2DM risk assessment tool) completed via the Diabetes UK website. Those scoring at or above a risk threshold were guided to self-refer with their local NHS-DPP provider (identified by postcode). Thus, during the time this study was conducted (October 2021 to March 2022), individuals could access the NHS-DPP via referral from primary care or via the self-referral route. Those who took up the referral were offered a choice of delivery mode, which, during the study, was remote groups or the NHS-digital-DPP. There was also the option to defer until a later date. ## Intervention The NHS-digital-DPP is delivered by 4 independent providers commissioned to deliver the 9-month behavior change program. Participants were offered 1 of the 4 service providers’ digital programs, depending on which provider was commissioned to deliver the digital service in their local geographical area. Although based on a common NHS England service specification [28,42], the DHIs vary in terms of their provision of materials, inclusion of wearables (eg, accelerometers and wireless weighing scales), extent of human support provided (ranging from a brief onboarding phone call to weekly coaching phone calls), delivery platform (smartphone app and website), and amount and format of educational materials (websites, emails, etc). Table 1 provides a summary of the providers’ features. Table 1 \| NHS-digital-DPP provider’s features \| Provider A \| Provider B \| Provider C \| Provider D \| \| --- \| --- \| --- \| --- \| --- \| \| Materials provided to service user \| Program app \| Program app and program handbook \| Program app \| Program app, program handbook, recipe book, wireless scales, and activity tracker \| \| Educational content \| 42 web-based articles \| Weekly web-based articles \| Bite-sized videos and written modules to supplement participant learnings—these are assigned by the health coach \| Web-based articles that are unlocked daily and 8 optional 4-week web-based courses \| \| Professional input \| Health coaching via series of scheduled telephone calls and web-based chat \| Access to health coaches via chat function \| Health coaching via initial telephone call, then regular video messages and web-based chat \| Health coaching in a web-based message service with a group of approximately 10 people (access to health coach in group or one-on-one chat) \| \| Peer support \| Not part of service at time of study \| Optional web-based discussion forum \| Optional web-based discussion forum \| Optional web-based discussion forum \| ## Study Sampling and Recruitment Sampling aimed to recruit a maximum variation sample of patients eligible for the NHS-DPP (refer to the previous sections), selected to vary by age, sex, ethnicity, socioeconomic status, geographical area, NHS-digital-DPP provider, and level of engagement with the NHS-digital-DPP. Participants were recruited via the 4 providers of the NHS-digital-DPP and via 3 primary care practices located within one clinical commissioning group in North London between October 2021 and March 2022. Providers contacted 2 groups of participants: those who had taken up the NHS-digital-DPP and those who had declined the NHS-digital-DPP in favor of participating in the remote group delivery mode. As the focus of the study was on decision-making around uptake, sampling was targeted at those who had recently been offered the NHS-digital-DPP. Therefore, on a monthly basis, providers emailed anyone referred in the previous 4 weeks. The aim of the additional recruitment via primary care practices was to capture participants who had been invited to participate in the NHS-digital-DPP but had not taken up the offer. A clinical commissioning group, which had in place a locally enhanced service for prediabetes monitoring, meaning referrals to the NHS-DPP and patient responses to referrals were routinely recorded, and which was only offering the remote or NHS-digital-DPP, was selected. Emails and letters were sent in monthly batches via the practices to any patient with a record of having declined or not responded to an invitation during the recruitment period (October 2021 to March 2022). Interested participants (from both recruitment routes) were asked to contact the research team, who then sent study documents and organized a time for interview. To recruit a maximum variation sample, researchers worked with the local National Institute for Health and Care Research Clinical Research Network to identify and sample primary care practices based on a range of factors, including deprivation scores, ethnic diversity, and diabetes prevalence. Researchers also worked iteratively with providers to increase the representation of participants from ethnic minority backgrounds and regions of high deprivation. At several time points, providers were asked to target recruitment emails to participants from those groups specifically. In total, 2051 participants were contacted by the NHS-digital-DPP providers, and 35 were contacted by primary care practices. In total, $3.17\%$ ($\frac{65}{2051}$) initial expressions of interest were received from the provider recruitment and $11\%$ ($\frac{4}{35}$) were received from primary care. Recruitment from providers ceased when data saturation had been reached, which was deemed to happen at interview 28, when no new themes were emerging. Interviews were conducted with all respondents ($\frac{4}{4}$, $100\%$) recruited via primary care. Despite considerable efforts to recruit those who had not taken up the NHS-digital-DPP, this proved to be challenging, and recruitment was stopped for this group at the end of the study data collection period. At this point, it is believed that data saturation had not been reached (refer to the Discussion section). ## Ethics Approval and Informed Consent Ethics approval for the study was obtained from the North West–Greater Manchester East National Health Service (NHS) research ethics committee (17/NW/0426). Verbal informed consent was recorded via an audio recorder for each participant before participation. ## Data Collection Topic guides were developed based on constructs from the health belief model (HBM) [43]. The HBM is one of the most widely used conceptual frameworks for explaining and changing individual health behavior and posits that individuals’ perceived susceptibility to and severity of a disease influence the perceived threat of the disease, which predicts the likelihood of self-management behaviors [44]. Furthermore, for people to comply with participatory preventive interventions, they need to perceive both the risk of the condition in question and the potential benefit of the intervention. The HBM has been widely applied to studies on prediabetes [45] and the development and evaluation of DHIs [46]. The topic guide included the following 6 domains of the HBM: perceived susceptibility, perceived severity, perceived beneﬁts, perceived barriers, self-eﬃcacy, and cues to action (Multimedia Appendix 1). For example, to examine the perceived susceptibility, we asked the following question: “How likely do you feel it is that you will develop diabetes?” The topic guide also included specific questions related to experiences with the NHS-DPP, and a separate set of questions was asked to those who had taken up and those who had not taken up the NHS-DPP. Topic guides were developed iteratively throughout the study to explore emerging areas of interest. All interviews were semistructured and conducted by the same researcher (JR). Owing to COVID-19–related restrictions, participants were given a choice between phone or Zoom (Zoom Video Communications Inc) interviews. Most participants ($\frac{26}{32}$, $81\%$) opted for Zoom. Interviews were recorded using an audio recorder. Demographic data were collected from participants via a proforma at the end of each interview. Audio recordings were securely transferred to a transcription company for transcription. After transcription, the researcher checked the transcripts for accuracy against the audio recordings and anonymized the transcripts. ## Data Analysis Data collection and analysis were conducted concurrently. Anonymized transcripts were coded using NVivo (version 10; QSR International) software. The analysis was conducted using an inductive thematic analysis approach [47]. This involved 6 phases: data familiarization, coding, identification of candidate themes, review and revision of themes, definition and naming of themes, and analysis and interpretation of patterns across the data. Constant comparative analysis was conducted by reviewing the transcripts and exploring the identified themes in subsequent interviews until data saturation was achieved. Codes and emerging themes were generated by the lead researcher and discussed within the multidisciplinary team to promote rigor and transparency in the analysis. The generated themes were considered through the lens of HBM to aid the interpretation of findings. ## Patient and Public Involvement The study was guided by an expert advisory group comprising 5 people with lived experience of prediabetes. The study was discussed with this group, which provided input (written and verbal) into a draft of the topic guide, particularly focusing on the relevance, importance, clarity, and wording of the questions. Changes to the topic guide suggested by the patient and public involvement group were implemented. Emerging findings were presented to the group that commented on preliminary themes. Ongoing work with the group includes coproducing a video to disseminate the research findings. ## Characteristics of Study Participants In total, 32 interviews were conducted, each lasting 42 minutes on average. Demographic characteristics are summarized in Table 2. Of the 32 participants, 24 ($75\%$) had taken up the offer of the NHS-digital-DPP, 4 ($13\%$) had declined the offer in favor of remote group delivery, and 4 ($13\%$) had been offered participation but had not taken up the offer. It is worth noting that in this study, all remote group participants ($\frac{4}{4}$, $100\%$) were from provider B, which gave participants access to their DHI and web-based group meetings. Of the 28 participants who took up the NHS-DPP, 22 ($79\%$) had been referred directly by their primary care practice, 2 ($7\%$) had been informed about the NHS-DPP by primary care practices but were not provided with a referral and contacted providers themselves, and 4 ($14\%$) had not been informed about the NHS-DPP by their primary care practice and had instead used the web-based risk tool and self-referred. Of the 24 participants who had taken up the NHS-digital-DPP, most ($$n = 19$$, $79\%$) participants reported daily use; however, $21\%$ ($\frac{5}{24}$) of the participants reported less frequent use. Compared with the 3623 participants in a pilot study of the NHS-digital-DPP [24], the participants had the same mean age (58, SD 10.7 years), had similar proportion of men and women (men: $50\%$ vs $49\%$), and were equally ethnically diverse (White British: $68\%$ vs $68\%$). They had more years of education (higher education: $78\%$ vs $29\%$) and were from more affluent areas (from least deprived areas: $25\%$ vs $15\%$). Table 2 \| Characteristics \| Characteristics.1 \| Characteristics.2 \| Characteristics.3 \| Characteristics.4 \| Values \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| All participants (N=32) \| All participants (N=32) \| All participants (N=32) \| All participants (N=32) \| All participants (N=32) \| All participants (N=32) \| \| \| Age (years), mean (SD) \| Age (years), mean (SD) \| Age (years), mean (SD) \| 58.3 (10.7) \| 58.3 (10.7) \| \| \| \| Sex, n (%) \| Sex, n (%) \| Sex, n (%) \| Sex, n (%) \| \| \| \| \| Female \| 16 (50) \| 16 (50) \| \| \| \| \| Male \| 16 (50) \| 16 (50) \| \| \| \| Ethnicity, n (%) \| Ethnicity, n (%) \| Ethnicity, n (%) \| Ethnicity, n (%) \| \| \| \| \| White \| 22 (69) \| 22 (69) \| \| \| \| \| Mixed \| 2 (6) \| 2 (6) \| \| \| \| \| Black \| 2 (6) \| 2 (6) \| \| \| \| \| Asian \| 2 (6) \| 2 (6) \| \| \| \| \| Other \| 4 (13) \| 4 (13) \| \| \| \| Education, n (%) \| Education, n (%) \| Education, n (%) \| Education, n (%) \| \| \| \| \| \| 1 (3) \| 1 (3) \| \| \| \| \| Secondary school education (GCSEa, O levelb, or CSEc) \| 6 (19) \| 6 (19) \| \| \| \| \| Higher \| 25 (78) \| 25 (78) \| \| \| \| Self-rated computer skills, n (%) \| Self-rated computer skills, n (%) \| Self-rated computer skills, n (%) \| Self-rated computer skills, n (%) \| \| \| \| \| Basic \| 6 (19) \| 6 (19) \| \| \| \| \| Intermediate \| 13 (41) \| 13 (41) \| \| \| \| \| Advanced \| 13 (41) \| 13 (41) \| \| \| \| Home internet access, n (%) \| Home internet access, n (%) \| Home internet access, n (%) \| Home internet access, n (%) \| \| \| \| \| Yes \| 32 (100) \| 32 (100) \| \| \| \| Owns smartphone, n (%) \| Owns smartphone, n (%) \| Owns smartphone, n (%) \| Owns smartphone, n (%) \| \| \| \| \| Yes \| 31 (97) \| 31 (97) \| \| \| \| \| No \| 1 (3) \| 1 (3) \| \| \| \| Perceptions of current health, n (%) \| Perceptions of current health, n (%) \| Perceptions of current health, n (%) \| Perceptions of current health, n (%) \| \| \| \| \| Excellent \| 2 (6) \| 2 (6) \| \| \| \| \| Very good \| 10 (31) \| 10 (31) \| \| \| \| \| Good \| 12 (38) \| 12 (38) \| \| \| \| \| Fair \| 5 (16) \| 5 (16) \| \| \| \| \| Poor \| 3 (9) \| 3 (9) \| \| \| \| Current occupational status, n (%) \| Current occupational status, n (%) \| Current occupational status, n (%) \| Current occupational status, n (%) \| \| \| \| \| Paid work \| 20 (63) \| 20 (63) \| \| \| \| \| Unemployed \| 2 (6) \| 2 (6) \| \| \| \| \| Voluntary work \| 1 (3) \| 1 (3) \| \| \| \| \| Not working owing to disability or ill health \| 1 (3) \| 1 (3) \| \| \| \| \| Retired \| 8 (25) \| 8 (25) \| \| \| \| Known someone with T2DMd, n (%) \| Known someone with T2DMd, n (%) \| Known someone with T2DMd, n (%) \| Known someone with T2DMd, n (%) \| \| \| \| \| Yes \| 26 (81) \| 26 (81) \| \| \| \| \| No \| 6 (19) \| 6 (19) \| \| \| \| Parent or sibling with T2DM, n (%) \| Parent or sibling with T2DM, n (%) \| Parent or sibling with T2DM, n (%) \| Parent or sibling with T2DM, n (%) \| \| \| \| \| Yes \| 15 (47) \| 15 (47) \| \| \| \| \| No \| 17 (53) \| 17 (53) \| \| \| \| Live with others, n (%) \| Live with others, n (%) \| Live with others, n (%) \| Live with others, n (%) \| \| \| \| \| Yes \| 25 (78) \| 25 (78) \| \| \| \| \| No \| 7 (22) \| 7 (22) \| \| \| \| Deprivation, n (%) \| Deprivation, n (%) \| Deprivation, n (%) \| Deprivation, n (%) \| \| \| \| \| IMDe 1 (most deprived) \| 4 (13) \| 4 (13) \| \| \| \| \| IMD 2 \| 7 (22) \| 7 (22) \| \| \| \| \| IMD 3 \| 6 (19) \| 6 (19) \| \| \| \| \| IMD 4 \| 7 (22) \| 7 (22) \| \| \| \| \| IMD 5 (least deprived) \| 8 (25) \| 8 (25) \| \| \| \| Referral offer source, n (%) \| Referral offer source, n (%) \| Referral offer source, n (%) \| Referral offer source, n (%) \| \| \| \| \| Primary care \| 26 (81) \| 26 (81) \| \| \| \| \| Self-referral \| 6 (19) \| 6 (19) \| \| \| Offer, n (%) \| Offer, n (%) \| Offer, n (%) \| Offer, n (%) \| Offer, n (%) \| \| \| \| \| Accepted \| 24 (75) \| 24 (75) \| \| \| \| \| Declined in favor of remote group \| 4 (13) \| 4 (13) \| \| \| \| \| Not taken up \| 4 (13) \| 4 (13) \| \| Digital NHS-DPPf group only (n=24) \| Digital NHS-DPPf group only (n=24) \| Digital NHS-DPPf group only (n=24) \| Digital NHS-DPPf group only (n=24) \| Digital NHS-DPPf group only (n=24) \| Digital NHS-DPPf group only (n=24) \| \| \| Use of NHS-digital-DPPg DHIsh, n (%) \| Use of NHS-digital-DPPg DHIsh, n (%) \| Use of NHS-digital-DPPg DHIsh, n (%) \| Use of NHS-digital-DPPg DHIsh, n (%) \| Use of NHS-digital-DPPg DHIsh, n (%) \| \| \| \| \| Daily \| 19 (79) \| 19 (79) \| \| \| \| \| Several times a week \| 2 (8) \| 2 (8) \| \| \| \| \| Once a week \| 2 (8) \| 2 (8) \| \| \| \| \| Less than once a week \| 1 (4) \| 1 (4) \| \| \| NHS-digital-DPP provider, n (%) \| NHS-digital-DPP provider, n (%) \| NHS-digital-DPP provider, n (%) \| NHS-digital-DPP provider, n (%) \| NHS-digital-DPP provider, n (%) \| \| \| \| \| A \| 8 (33) \| 8 (33) \| \| \| \| \| B \| 2 (8)i \| 2 (8)i \| \| \| \| \| C \| 8 (33) \| 8 (33) \| \| \| \| \| D \| 6 (25) \| 6 (25) \| \| \| Time using DHI at the time of interview (weeks), n (%) \| Time using DHI at the time of interview (weeks), n (%) \| Time using DHI at the time of interview (weeks), n (%) \| Time using DHI at the time of interview (weeks), n (%) \| Time using DHI at the time of interview (weeks), n (%) \| \| \| \| \| 1-2 \| 1 (4) \| 1 (4) \| \| \| \| \| 2-3 \| 5 (21) \| 5 (21) \| \| \| \| \| 3-4 \| 9 (38) \| 9 (38) \| \| \| \| \| 4-5 \| 9 (38) \| 9 (38) \| ## Overview Overall, seven overarching themes were identified: [1] knowledge and understanding, [2] referral process, [3] self-efficacy, [4] self-identity, [5] motivation and support, [6] advantages of digital service and response efficacy, and [7] reflexive monitoring. Some themes encompassed factors that were related to uptake of and engagement with the NHS-DPP generally, whereas others were specific to the NHS-digital-DPP. These themes are summarized in Multimedia Appendix 2. ## Knowledge and Understanding This theme relates to participants’ knowledge and understanding about T2DM and NHS-DPP. Participants reported mixed understanding about what prediabetes and T2DM are and how one relates to the other. Several participants had not heard of the term prediabetes before their diagnosis, and there was some confusion about what it meant. For others, especially those with a family history of diabetes, there was greater understanding about the nature of the condition. Participants described prediabetes in terms of “sugar levels higher than they should be” (participant 5), being “at risk from diabetes” (participant 9), and “borderline” (participants 4, 8, 11, 19, 22, and 28) for diabetes and understood that prediabetes could progress to “full-blown diabetes” (participant 19). For many participants, the receipt of their prediabetes diagnosis was unexpected and was described as a “shock” (participants 1, 3, 10, 11, 21, 22, 27, and 30). This was particularly true in several cases where participants had not known they were being “tested” for diabetes risk or had not experienced any symptoms of ill health. The surprise at the diagnosis was more apparent for participants who did not identify as being a “typical” person with diabetes, who were described in terms of being overweight and having poor eating habits: For some participants, the surprising nature of the diagnosis affected the readiness to take up the NHS-DPP. A person who declined the program reported feeling very overwhelmed by the diagnosis to take preventive action (participant 29). Another participant described that despite knowing about the potential serious consequences of diabetes, they were not currently in the right “headspace” (participant 32) to engage with making behavior changes or had other health concerns and prediabetes was not perceived as a priority. However, others described the diagnosis as a “wake up call” (participants 3, 21, and 22), which prompted a desire to become healthy: There were those who, despite being diagnosed with prediabetes, reported that they had not been provided with sufficient (or any) information about what this diagnosis meant. Several participants reported being informed about their risk of diabetes via a letter from their primary care practice and had not spoken to an HCP about it. Some participants commented on the perceived lack of importance given to the diagnosis by their HCP, which in turn influenced their perceptions of risk: Those who had less understanding about diabetes risk were either motivated to join the NHS-DPP to become more educated, or in contrast, reported less inclination to embark on changes to lifestyle behaviors immediately. For example, there was some confusion about disease progression, with some believing that T2DM was a future eventuality that should not cause immediate concern: ## Referral Process This theme includes factors related to the implementation of the NHS-DPP, which affected the participants’ decisions to take it up. There was a lack of information in the offer of the NHS-DPP from primary care. Many participants reported not being sure about what they may get from the program because this had not been made clear to them at the point of referral. There was a sense from participants that HCPs did not know much about what they were offering to patients: Most participants reported not knowing that a digital option for the NHS-DPP was available until they had an initial conversation with the program providers, at which point they were offered a choice of service. Overall, $6\%$ ($\frac{2}{32}$) of the participants had declined the offer from their primary care practice as they felt that they were too busy to embark on a program to support lifestyle changes. When the digital version was discussed with these participants during our interviews, both of them suggested this would have been more appealing and likely taken up (participants 31 and 32): When the NHS-DPP was discussed in more detail with participants in primary care services, it positively affected their decision to take it up: In some cases, participants had not been informed about the NHS-DPP by primary care, and in other cases, although they were informed about their risk of T2DM from their primary care practice, no offer of participation in the NHS-DPP had been received. This prompted participants to search for information on their own. These participants reported finding about the NHS-DPP via Facebook advertisements, by Googling, and from the NHS website and had subsequently gained access by using the web-based risk score: The fact that the NHS-DPP was referred to by and affiliated with the NHS was of critical importance for perceptions of trustworthiness and credibility. This was important to participants who contrasted the program with other digital sources of information, which they had found hard to verify or assess as credible. The NHS affiliation also gave participants confidence that the program would be efficacious in bringing about the desired outcomes: Many participants were offered referral via a letter or SMS text message, which was easily dismissed and did not carry the same seriousness as a conversation with a health professional. A participant reported not accepting the offer immediately because of this reason, and instead, waiting another 6 months before joining: There was also no follow-up from primary care about whether participants had taken up the offer of the NHS-DPP, which contributed to a sense that this was not an important thing to engage with and led to delays with participants taking up the offer: For $50\%$ ($\frac{2}{4}$) of the participants who had not taken up the program, this was because they had no recollection of receiving a letter from their primary care practice and no follow-up meant that they had not had further opportunity to be referred: Even after accepting the referral, there were reports of delays in participants accessing the NHS-digital-DPP. Participants reported being almost ready to “quit” (participant 13) and “left to flounder” (participant 1) without knowing what to do after receiving their diagnosis (participant 12) owing to delays in program providers making contact. There were examples of considerable effort by some participants to get onto the program, with repeated phone calls to providers: ## Self-efficacy An important factor affecting self-efficacy (confidence in one’s ability) for diabetes prevention was the participants’ previous efforts to modify behavior, especially around changing diets and weight loss. Some participants reported having made considerable improvements to their diet and exercise behaviors after receiving the diagnosis of prediabetes and before embarking on the NHS-digital-DPP and thus felt confident that they would be able to keep this up and that an intervention to support this would be useful. However, others reported having previously failed with weight loss efforts or found certain aspects of behavior change to be difficult, and therefore, they were less confident that they would be able to make any sustainable changes: Others, especially those who had family members with T2DM, drew on the genetic nature of risk to explain why they felt less capable of preventing disease progression, explaining that lifestyle modifications would not be sufficient to reduce their risk of developing T2DM. Thus, they were hesitant about how useful the NHS-DPP would be: After participants had embarked on using the program, many reported increased feelings of self-efficacy toward being able to make lifestyle changes and reduce the risk of developing diabetes. Increased self-efficacy was related to increased knowledge and was bolstered by seeing changes related to behavioral modifications such as weight loss: ## Self-identity A considerable barrier for some in deciding to embark on the program was that they did not identify as being the target population for an intervention they perceived as being primarily for weight loss: Other participants did not identify as the target users, as they perceived themselves to already have a good level of knowledge about healthy eating. Many participants questioned whether the NHS-digital-DPP would be of any benefit to them, as they considered themselves to not be in need of the “basic information” (participant 11) that they perceived the program would offer. Again, participants contrasted themselves with the type of person they perceived to need support in terms of knowledge: Another subgroup of participants whose self-identity led them to question the usefulness of the NHS-digital-DPP to support them with lifestyle changes included those who had comorbidities and mobility issues. These participants perceived themselves as having needs that are different from those of other users and reported needing a service that was much more tailored to their individual needs: ## Motivation and Support The digital delivery particularly appealed to those who described themselves as being “self-motivated” (participant 12) to take action to change behaviors, perceiving the DHI as a facilitatory tool rather than as a main driver of change. This was often contrasted with the perception of face-to-face support, which was viewed as better for those who needed external motivation to change. Several people who had opted for digital delivery described not wanting to engage in group sessions as they were perceived to be aimed at people who needed more encouragement, who would not be able to achieve desired outcomes on their own with the support of just the DHI: For those who had access to health coaches, they were perceived positively, and they influenced decisions to take up and engage. Many participants viewed health coaches as a way to instill accountability for their lifestyle changes. Some spoke about having sufficient knowledge to make changes to their lifestyles, but that staying motivated was more difficult. One of the main benefits of the health coaches was to help participants stay “on track” (participant 3) and accountable: However, some participants expressed disappointment regarding the lack of support from the DHIs. Participants discussed feeling demotivated to continue use because of lack of perceived support. This was particularly true for those using the DHI, which did not provide formal health coach support (provider B). Those who had not yet been contacted by the health coaches also reported feeling demotivated after joining. In some cases, participants reported waiting for weeks before having their first session with a health coach. Without access to a health coach, several participants reported feeling that all they were doing was tracking their food, and to make meaningful behavior changes, they required more support and a more structured approach to lifestyle changes: For those who opted for the remote group version, peer support was an important factor in this decision. These participants discussed wanting to feel as part of a community and to be able to draw on others’ experiences and share ideas: Although all the DHIs had an aspect of peer support, few participants reported engaging with these features, which included group messaging, chat rooms, and forums. Several participants stated that one of the main reasons for selecting digital was for the avoidance of having to interact with others, which they likened to self-help groups. This was consistent with the previously mentioned identities of these participants as not being “typical” of the prediabetic population: For the few participants who reported having engaged with peer support features, there was some disappointment regarding the lack of group interactions, and others reported that they did not engage with these features because others were not engaging, suggesting that a critical mass of users had not been reached: ## Advantages of Digital Service Perceptions of acceptability were pivotal in making decisions to engage with the NHS-digital-DPP. Many benefits were discussed in contrast to face-to-face and remote group delivery, particularly around the benefits of the convenience that a digital service could offer. The digital service particularly appealed to those who worked and those who had a dislike for groups: The acceptability of a digital service was also discussed in relation to the COVID-19 pandemic, with participants more willing to engage with a DHI, having become more accustomed to doing things via technology during COVID-19–related restrictions, which made the NHS-digital-DPP more acceptable. Many participants also referenced the current strain on primary care as important in their decision to adopt the app, postulating that they would not be likely to be offered any other form of support for their risk because of the pressures on HCPs and primary care related to COVID-19. Many participants stated having accepted that digital delivery of health care would be the norm, going forward: In addition, because of COVID-19, several participants expressed continued fear of group interactions, and thus, having access to a digital service was viewed as critical for engagement in the NHS-DPP: More generally, convenience and accessibility were the main features of the NHS-digital-DPP that appealed to people, with participants reporting that they could fit their use of the NHS-digital-DPP in around work patterns, caring responsibilities, and other daily activities. Being able to use the DHIs at a time that suited the participants helped them to feel in control of their use: Other advantages of the NHS-digital-DPP included anonymity and privacy (participant 10). For some participants, this was important as they felt stigma about being overweight and valued not having to interact with others in person: Generally, the ability to use and engage with the DHIs was high. A few participants had experienced problems with initially downloading the apps but reported that these issues were resolved quickly with assistance from program providers. Even those who described themselves as having basic computer skills found the DHIs easy to engage with. The features that participants reported most commonly engaging with were tracking features, including food, weight, and exercise physical activity tracking. Most participants ($\frac{19}{24}$, $79\%$) reported using these features daily, and the tracking aspects were conceptualized as the main part of the DHIs by many participants. Tracking features helped participants to feel accountable to the DHIs: However, there was variability across providers regarding the appraisal of the quality of features. Several participants from a provider were dissatisfied with the tracking abilities of the app and even described having to supplement the technologies with other apps that they had installed: Some participants commented that the prediabetes aspect of the DHIs were not always prominent, feeling that primarily this was a “weight loss” program. Many participants made the link between weight loss and risk reduction; however, for some participants, it was not clear why a weight loss program would reduce diabetes risk. Some were disappointed with the lack of content specifically about prediabetes: One of the advantages of the NHS-digital-DPP was the ability for tailoring. The health coaches were perceived to be the main mode of delivering tailored content, for example, by sending suitable articles to participants, delivering tailored information during conversations, and helping people to set individual goals: However, as discussed previously, not all participants believed that the degree of tailoring was sufficient. Much of the variability reported about the adequacy of tailoring came from expectations and perceptions of how good the health coaching features of the apps were: ## Reflexive Monitoring This theme relates to how participants appraised their use of the NHS-digital-DPP. Most participants attributed their prediabetes diagnoses to being overweight. Therefore, most of them spoke about reducing weight as their desired outcome of using the DHIs. Those who were aiming to lose weight cited being motivated to continue use because they could observe changes in weight, for example, by noticing clothes feeling loose and decreasing weight measurements: However, for some participants, especially those who were not overweight, their motivation for adopting the NHS-digital-DPP was to reduce their blood glucose levels and get out of the prediabetic range. For these participants, it was less clear how they could monitor their progress. Participants spoke about not knowing if or how they could obtain another blood glucose reading or how prediabetes and risk of diabetes would be monitored following diagnosis. This was seen as important for some participants to assess DHI efficacy and to maintain “motivation” for use (participant 20 and 21): Generally, those using the DHIs reported being motivated to continue use. Although participants had only been using the DHIs for few weeks, many reported increased feelings of efficacy toward preventing diabetes because of using the DHIs and observing changes: Others reported that over time, they anticipated learning enough from using the DHIs to allow them to embed changes into their lifestyle and thus would not need to continue engaging. This was also true for reducing use. People reported anticipating using the tracking features less frequently when they had learned enough or had reached a weight at which they were happy: However, as the NHS-digital-DPP was only being made available to participants for a fixed period of time (9 months), there were some concerns about how participants would continue the changes they had begun when their access ended: ## Principal Findings DHIs for diabetes prevention are showing considerable potential for behavior change among those who engage with them. However, less is known about the factors that influence uptake and engagement with these interventions in real-world populations. To promote participation in and to inform the development of future digital-DPPs, more evidence about these critical factors is needed. This study explored decision-making around uptake of and engagement with the NHS-digital-DPP, with participants who had been offered referral as part of routine NHS care. The views of those who did and did not take up the NHS-digital-DPP and those who chose a different delivery mode are represented. The findings related to both the NHS-DPP generally and the NHS-digital-DPP specifically. Psychological factors related to beliefs about vulnerability to diabetes, self-efficacy for reducing risk, and self-identify and implementation factors including issues with referrals and lack of engagement from HCPs were barriers to the uptake of the NHS-DPP. Factors that related specifically to the uptake of a digital service included perceptions about usefulness in supporting behavioral modifications, perceptions about accessibility and convenience, and views about participating in a group. Specific features of the DHIs including health coaches and tracking features that promoted accountability and motivation were important for promoting engagement. Many participants perceived the NHS-digital-DPP as an acceptable service to help reduce the risk of developing T2DM, and many reported having been supported to make behavior changes. The benefits of the NHS-digital-DPP were often discussed in contrast to perceptions of face-to-face and group-based services, highlighting the need for a range of delivery options for diabetes prevention to ensure that participants can access a service that meets their specific needs and preferences and thus promote engagement. ## Strengths and Limitations One of the main strengths of this study is the collection of data from a real-world population including those who accepted the NHS-digital-DPP, those who opted for a different delivery mode, and those who did not take up the offer. Studying the reasons for not taking up digital interventions is essential for overcoming proinnovation bias [48]. However, it is likely that given the small number of participants in this subgroup, there are still factors that remain obscured, which could be explored with further studies into nonuptake. Furthermore, despite best efforts, because of the way in which the NHS-digital-DPP is offered to participants (usually not until participants have made contact with providers), it remains difficult to isolate views on nonuptake that relate specifically to the digital delivery mode, as opposed to the NHS-DPP more broadly. Although the sample was diverse in terms of ethnicity, age, and sex, there was less diversity represented in terms of other characteristics including socioeconomic status, education, digital access, and computer skills. Thus, these findings are unlikely to fully represent the experiences of those on the other side of the so-called digital divide, which represents inequalities in accessing and using digital technologies [49]. For example, previous studies have shown that adoption of DHIs may be less among those with low socioeconomic status, [50], less computer experience [51], and less access to social networks [52]. Engagement with and adherence to DHIs may also be less among those with low education levels and socioeconomic status [53]. Future studies with individuals from these groups may highlight additional findings about uptake of and engagement with digital-DPPs. The primary focus of the study was on factors that influenced uptake and initial engagement decisions, and thus, participants had access to the DHIs for a relatively short time frame. Ongoing research by this team is examining users’ patterns of use of the NHS-digital-DPP, which will provide findings on how the DHIs are used longitudinally. Further qualitative studies to assess long-term experiences with the NHS-digital-DPP, especially around decisions to continue or cease use, would complement this ongoing study. ## Comparisons With Previous Literature Participants often contrasted themselves with people that they thought were typical of a diagnosis of diabetes and highlighted ways in which they were different (not overweight, more active, and more knowledgeable). These participants found it difficult to identify themselves as being in an “at-risk state” [4], as this conflicted with their own perceptions of having a healthy lifestyle creating a distance to future risk. Similar findings were reported in our qualitative study with NHS-DPP participants, which found that people with prediabetes resist the notion that they are “candidates” for diabetes as this contradicted their perceived identity as healthy individuals [54]. In this study, not feeling typical of someone who develops diabetes also led participants to question how much they would benefit from the NHS-digital-DPP. Findings emphasized the importance of several constructs of the HBM [43] including perceived susceptibility and severity, cues to action, and self-efficacy. Many participants were shocked by their diagnosis of prediabetes, and their understanding about the diagnosis was mixed. This was often related to the way in which the diagnosis had been delivered, often via letter or SMS text message with limited information. However, most participants reported feeling strongly that if they did not take preventive action, they would be susceptible to developing diabetes. However, there were mixed views about how severe diabetes may be if it developed, which were mediated by the way in which participants had been informed about their risk by their HCP and personal experiences with diabetes. A meta-analysis of barriers to and facilitators of lifestyle changes in people with prediabetes identified the point at which people become aware of being at high risk of developing T2DM and realize the potential threat to their health as a vital facilitator of healthy lifestyle changes [35], which were findings supported by a meta-analysis exploring risk appraisals that showed altered risk perceptions have impact on intentions to change behavior and on changes in behavior itself [55]. In many cases, there seemed to be a missed opportunity to engage participants with their diagnosis and preventable actions because of the way in which this information was communicated to participants. Several previous studies have highlighted that the participant’s assessment of the seriousness of prediabetes may be influenced by HCP’s communication and behavior [21,34,56]. Furthermore, health professionals’ communication around the NHS-DPP was perceived as a critical cue to action, that is, the stimulus needed to trigger the decision-making process to accept a recommended health action. In cases where participants had not taken up the NHS-digital-DPP, this was because of poor referrals, including lack of information about the digital option or offers not being received. Those who had taken up the NHS-digital-DPP reported being unclear about the program’s aims and content, because this had not been communicated adequately at the point of referral. A recent study examining the US National diabetes prevention program also found that cues to action were determinants of enrollment, specifically that clear information about the diagnosis of prediabetes and decision support for joining a lifestyle intervention, especially from a trusted health care provider, were critical [45]. However, improving communication around referrals by HCPs may be exacting. For example, a previous study that examined HCPs’ views of a digital T2DM self-management program highlighted the difficulties of HCPs in absorbing the additional tasks needed to provide adequate referrals to the program into an already overwhelming workload [57], and several other studies have identified resource constraints as barriers to HCPs referring to digital services [58,59]. Self-efficacy for being able to reduce the risk of diabetes was generally high among participants, with many viewing a diagnosis of prediabetes as an opportunity to make healthy lifestyle modifications, and self-efficacy for behavior changes improved after participants started using the NHS-digital-DPP. As with previous studies [35], having former positive experiences with exercise and diet facilitated self-confidence for engaging in these behaviors. Participants’ self-confidence for behavior changes was bolstered by positive feedback, for example, seeing weight loss and feeling healthy as a result of using the NHS-digital-DPP. Features of the DHIs, which helped participants track and visualize their progression toward goal attainment, had a positive impact on perceptions of being able to prevent diabetes. However, participants were disappointed at the lack of feedback on whether behavior changes were having an impact on risk of developing T2DM. There were participants with a family history of diabetes who were more likely to perceive disease progression as an inevitability. Previous studies have emphasized the importance of understanding social constructs including inevitable social norms related to genetic predispositions to diabetes, which can influence decisions to engage with health care advice and implement behavior changes [60]. Proactive health coaching was appraised positively by most participants and was important for their decisions to take up and maintain engagement with the program. Health coaching helped participants to access relevant information, set personal goals, review progress, tailor the DHIs, and provide human contact. Previous studies of digital-DPP interventions have shown that health coaches are valued by users [61], may enhance participation and engagement [62], and may enhance the efficacy of digital-DPP interventions on weight loss [21]. Input from professionals may foster feelings of accountability and a sense of being monitored, which have been shown to facilitate lifestyle changes [35]. Despite an emerging body of literature suggesting that peer support is important for engagement with digital diabetes prevention and outcomes [24,35,63], this study found mixed views regarding participants’ desire for peer support. Those who had opted for remote groups did so because they wanted to interact with others. However, among those who opted for the digital service, many reported not wanting or needing to engage with peers, and those who engaged with peers reported that the peer support features were underused by other users, thus decreasing their motivation to engage with these aspects. ## Implications It is likely that adequate discussion by HCPs about prediabetes and T2DM risk would increase patients’ knowledge about disease severity and emphasize the preventable nature of T2DM, thus increasing self-efficacy for behavior changes. The findings also suggest that better communication may raise awareness about the digital service, provide endorsement, and help to sustain participant engagement. Future studies could focus on HCPs’ perceptions of these factors to identify barriers to referral. For example, recent study by this team on the implementation of the NHS-DPP suggests that individuals responsible for the local commissioning and implementation of NHS-DPP report having minimum knowledge about the NHS-digital-DPP in terms of the content [64]. Therefore, strategies to promote patient uptake could focus on raising awareness about the NHS-digital-DPP among those responsible for referring to it. Thought could also be given to ways to ensure that patients receive adequate information that does not have an impact on HCP resources, for example, through direct-to-patient marketing of the NHS-digital-DPP or peer-led information sessions. In addition, HCPs could be provided with specific tailored materials to provide to certain groups. For example, specific messaging for those with a genetic diabetes risk or for those who are not overweight may promote the value of behavior changes and NHS-DPP. Digital-DPPs are still in their infancy, and it is not yet clear how best to optimize their delivery to enhance the desired clinical outcomes. However, findings from this study suggest that accountability and monitoring affected the participants’ early experiences and encouraged uptake and engagement. Accountability was frequently described in terms of interactions with the health coaches. Motivation for continued use was driven by the ability to monitor progress. The DHIs had specific features to help participants visualize and monitor goals that were perceived as motivating. Thus, these findings suggest that digital-DPP interventions should feature elements of accountability and automated monitoring systems. Furthermore, it is important to consider how users of digital-DPPs who are not overweight can be supported to monitor their progress. For example, future implementation efforts could make it clear to participants about how and when they can repeat blood tests to monitor diabetes risk, and a blood test could be potentially offered as part of the service. Finally, thought could be given to tailoring digital-DPP interventions based on participants’ desire to interact with others, for example, placing those with a desire for peer support together as a group. ## Conclusions This study provides important findings on factors that influence decisions to take up and engage with the NHS-digital-DPP. Findings suggest that participants found the DHIs to be convenient, accessible, and useful in supporting behavior changes. Specific features including health coaches and tracking tools were important for initial motivation and accountability. The study also highlights the importance of communication about diabetes risk and NHS-digital-DPP. ## References 1. Diabetes key facts. World Health Organization (2022) 2. Diabetes prevalence 2019. Diabetes UK (2019) 3. NHS diabetes prevention programme (NHS DPP) non-diabetic hyperglycaemia. National Cardiovascular Intelligence Network (2015) 4. Bansal N. Prediabetes diagnosis and treatment: a review. World J Diabetes (2015) 6 296-303. DOI: 10.4239/wjd.v6.i2.296 5. Lindström J, Ilanne-Parikka P, Peltonen M, Aunola S, Eriksson JG, Hemiö K, Hämäläinen H, Härkönen P, Keinänen-Kiukaanniemi S, Laakso M, Louheranta A, Mannelin M, Paturi M, Sundvall J, Valle TT, Uusitupa M, Tuomilehto J. Sustained reduction in the incidence of type 2 diabetes by lifestyle intervention: follow-up of the Finnish diabetes prevention study. Lancet (2006) 368 1673-9. DOI: 10.1016/S0140-6736(06)69701-8 6. Uusitupa M, Khan TA, Viguiliouk E, Kahleova H, Rivellese AA, Hermansen K, Pfeiffer A, Thanopoulou A, Salas-Salvadó J, Schwab U, Sievenpiper JL. Prevention of type 2 diabetes by lifestyle changes: a systematic review and meta-analysis. Nutrients (2019) 11 2611. DOI: 10.3390/nu11112611 7. Dunkley AJ, Bodicoat DH, Greaves CJ, Russell C, Yates T, Davies MJ, Khunti K. Diabetes prevention in the real world: effectiveness of pragmatic lifestyle interventions for the prevention of type 2 diabetes and of the impact of adherence to guideline recommendations: a systematic review and meta-analysis. Diabetes Care (2014) 37 922-33. DOI: 10.2337/dc13-2195 8. Galaviz KI, Weber MB, Straus A, Haw JS, Narayan KM, Ali MK. Global diabetes prevention interventions: a systematic review and network meta-analysis of the real-world impact on incidence, weight, and glucose. Diabetes Care (2018) 41 1526-34. DOI: 10.2337/dc17-2222 9. Valabhji J, Barron E, Bradley D, Bakhai C, Fagg J, O'Neill S, Young B, Wareham N, Khunti K, Jebb S, Smith J. Early outcomes from the English National Health Service Diabetes Prevention Programme. Diabetes Care (2020) 43 152-60. DOI: 10.2337/dc19-1425 10. A systematic review and meta-analysis assessing the effectiveness of pragmatic lifestyle interventions for the prevention of type 2 diabetes mellitus in routine practice. Public Health England (2015) 11. Marsden AM, Bower P, Howarth E, Soiland-Reyes C, Sutton M, Cotterill S. 'Finishing the race' - a cohort study of weight and blood glucose change among the first 36,000 patients in a large-scale diabetes prevention programme. Int J Behav Nutr Phys Act (2022) 19 7. DOI: 10.1186/s12966-022-01249-5 12. McManus E, Meacock R, Parkinson B, Sutton M. Population level impact of the NHS diabetes prevention programme on incidence of type 2 diabetes in England: an observational study. Lancet Reg Health Eur (2022) 19 100420. DOI: 10.1016/j.lanepe.2022.100420 13. McGough B, Murray E, Brownlee L, Barron E, Smith J, Valabhji J. The healthier you: NHS diabetes prevention programme: digital modes of delivery engage younger people. Diabet Med (2019) 36 1510-1. DOI: 10.1111/dme.14083 14. Ritchie ND, Phimphasone-Brady P, Sauder KA, Amura CR. Perceived barriers and potential solutions to engagement in the national diabetes prevention program. ADCES In Pract (2021) 9 16-20. DOI: 10.1177/2633559x20966275 15. Horigan G, Davies M, Findlay-White F, Chaney D, Coates V. Reasons why patients referred to diabetes education programmes choose not to attend: a systematic review. Diabet Med (2017) 34 14-26. DOI: 10.1111/dme.13120 16. Howarth E, Bower PJ, Kontopantelis E, Soiland-Reyes C, Meacock R, Whittaker W, Cotterill S. 'Going the distance': an independent cohort study of engagement and dropout among the first 100 000 referrals into a large-scale diabetes prevention program. BMJ Open Diabetes Res Care (2020) 8 e001835. DOI: 10.1136/bmjdrc-2020-001835 17. Beleigoli AM, Andrade AQ, Cançado AG, Paulo MN, Diniz MF, Ribeiro AL. Web-based digital health interventions for weight loss and lifestyle habit changes in overweight and obese adults: systematic review and meta-analysis. J Med Internet Res (2019) 21 e298. DOI: 10.2196/jmir.9609 18. Rose T, Barker M, Maria Jacob C, Morrison L, Lawrence W, Strömmer S, Vogel C, Woods-Townsend K, Farrell D, Inskip H, Baird J. A systematic review of digital interventions for improving the diet and physical activity behaviors of adolescents. J Adolesc Health (2017) 61 669-77. DOI: 10.1016/j.jadohealth.2017.05.024 19. Sepah SC, Jiang L, Ellis RJ, McDermott K, Peters AL. Engagement and outcomes in a digital diabetes prevention program: 3-year update. BMJ Open Diabetes Res Care (2017) 5 e000422. DOI: 10.1136/bmjdrc-2017-000422 20. Ryan JC, Wiggins B, Edney S, Brinkworth GD, Luscombe-March ND, Carson-Chahhoud KV, Taylor PJ, Haveman-Nies AA, Cox DN. Identifying critical features of type two diabetes prevention interventions: a Delphi study with key stakeholders. PLoS One (2021) 16 e0255625. DOI: 10.1371/journal.pone.0255625 21. Joiner KL, Nam S, Whittemore R. Lifestyle interventions based on the diabetes prevention program delivered via eHealth: a systematic review and meta-analysis. Prev Med (2017) 100 194-207. DOI: 10.1016/j.ypmed.2017.04.033 22. Van Rhoon LV, Byrne M, Morrissey E, Murphy J, McSharry J. A systematic review of the behaviour change techniques and digital features in technology-driven type 2 diabetes prevention interventions. Digit Health (2020) 6 2055207620914427. DOI: 10.1177/2055207620914427 23. Bian RR, Piatt GA, Sen A, Plegue MA, De Michele ML, Hafez D, Czuhajewski CM, Buis LR, Kaufman N, Richardson CR. The effect of technology-mediated diabetes prevention interventions on weight: a meta-analysis. J Med Internet Res (2017) 19 e76. DOI: 10.2196/jmir.4709 24. Ross JA, Barron E, McGough B, Valabhji J, Daff K, Irwin J, Henley WE, Murray E. Uptake and impact of the English National Health Service digital diabetes prevention programme: observational study. BMJ Open Diabetes Res Care (2022) 10 e002736. DOI: 10.1136/bmjdrc-2021-002736 25. Katula JA, Dressler EV, Kittel CA, Harvin LN, Almeida FA, Wilson KE, Michaud TL, Porter GC, Brito FA, Goessl CL, Jasik CB, Sweet CM, Schwab R, Estabrooks PA. Effects of a digital diabetes prevention program: an RCT. Am J Prev Med (2022) 62 567-77. DOI: 10.1016/j.amepre.2021.10.023 26. Winkley K, Evwierhoma C, Amiel SA, Lempp HK, Ismail K, Forbes A. Patient explanations for non-attendance at structured diabetes education sessions for newly diagnosed type 2 diabetes: a qualitative study. Diabet Med (2015) 32 120-8. DOI: 10.1111/dme.12556 27. Barron E, Clark R, Hewings R, Smith J, Valabhji J. Progress of the healthier you: NHS diabetes prevention programme: referrals, uptake and participant characteristics. Diabet Med (2018) 35 513-8. DOI: 10.1111/dme.13562 28. Service Specification No. 1: Provision of behavioural interventions for people with non-diabetic hyperglycaemia. National Health Service England (2015) 29. Eysenbach G. The law of attrition. J Med Internet Res (2005) 7 e11. DOI: 10.2196/jmir.7.1.e11 30. Kelders SM, Kok RN, Ossebaard HC, Van Gemert-Pijnen JE. Persuasive system design does matter: a systematic review of adherence to web-based interventions. J Med Internet Res (2012) 14 e152. DOI: 10.2196/jmir.2104 31. Short CE, DeSmet A, Woods C, Williams SL, Maher C, Middelweerd A, Müller AM, Wark PA, Vandelanotte C, Poppe L, Hingle MD, Crutzen R. Measuring engagement in eHealth and mHealth behavior change interventions: viewpoint of methodologies. J Med Internet Res (2018) 20 e292. DOI: 10.2196/jmir.9397 32. Kohl LF, Crutzen R, de Vries NK. Online prevention aimed at lifestyle behaviors: a systematic review of reviews. J Med Internet Res (2013) 15 e146. DOI: 10.2196/jmir.2665 33. Murray E, Hekler EB, Andersson G, Collins LM, Doherty A, Hollis C, Rivera DE, West R, Wyatt JC. Evaluating digital health interventions: key questions and approaches. Am J Prev Med (2016) 51 843-51. DOI: 10.1016/j.amepre.2016.06.008 34. Begum S, Povey R, Ellis N, Gidlow C, Chadwick P. Influences of decisions to attend a national diabetes prevention programme from people living in a socioeconomically deprived area. Diabet Med (2022) 39 e14804. DOI: 10.1111/dme.14804 35. Skoglund G, Nilsson BB, Olsen CF, Bergland A, Hilde G. Facilitators and barriers for lifestyle change in people with prediabetes: a meta-synthesis of qualitative studies. BMC Public Health (2022) 22 553. DOI: 10.1186/s12889-022-12885-8 36. Albury CV, Ziebland S, Webb H, Stokoe E, Aveyard P. Discussing weight loss opportunistically and effectively in family practice: a qualitative study of clinical interactions using conversation analysis in UK family practice. Fam Pract (2021) 38 321-8. DOI: 10.1093/fampra/cmaa121 37. Jenkins CL, Imran S, Mahmood A, Bradbury K, Murray E, Stevenson F, Hamilton FL. Digital health intervention design and deployment for engaging demographic groups likely to be affected by the digital divide: protocol for a systematic scoping review. JMIR Res Protoc (2022) 11 e32538. DOI: 10.2196/32538 38. O'Connor S, Hanlon P, O'Donnell CA, Garcia S, Glanville J, Mair FS. Understanding factors affecting patient and public engagement and recruitment to digital health interventions: a systematic review of qualitative studies. BMC Med Inform Decis Mak (2016) 16 120. DOI: 10.1186/s12911-016-0359-3 39. Hall AK, Bernhardt JM, Dodd V, Vollrath MW. The digital health divide: evaluating online health information access and use among older adults. Health Educ Behav (2015) 42 202-9. DOI: 10.1177/1090198114547815 40. Jakob R, Harperink S, Rudolf AM, Fleisch E, Haug S, Mair JL, Salamanca-Sanabria A, Kowatsch T. Factors influencing adherence to mHealth apps for prevention or management of noncommunicable diseases: systematic review. J Med Internet Res (2022) 24 e35371. DOI: 10.2196/35371 41. NHS health check. National Health Service England (2019) 42. NHS diabetes prevention programme national service specification. National Health Service England (2021) 43. Strecher V, Rosenstock I, Ayers S, Baum A, McManus C, Newman S, Wallston K, Weinman J, West R. The health belief model. Cambridge Handbook of Psychology, Health and Medicine (1997) 44. Seehusen DA, Fisher CL, Rider HA, Seehusen AB, Womack JJ, Jackson JT, Crawford PF, Ledford CJ. Exploring patient perspectives of prediabetes and diabetes severity: a qualitative study. Psychol Health (2019) 34 1314-27. DOI: 10.1080/08870446.2019.1604955 45. Joiner KL, McEwen LN, Hurst TE, Adams MP, Herman WH. Domains from the health belief model predict enrollment in the National Diabetes Prevention Program among insured adults with prediabetes. J Diabetes Complications (2022) 36 108220. DOI: 10.1016/j.jdiacomp.2022.108220 46. Naslund JA, Aschbrenner KA, Kim SJ, McHugo GJ, Unützer J, Bartels SJ, Marsch LA. Health behavior models for informing digital technology interventions for individuals with mental illness. Psychiatr Rehabil J (2017) 40 325-35. DOI: 10.1037/prj0000246 47. Clarke V, Braun V, Hayfield N, Smith JA. Thematic analysis. Qualitative Psychology: A Practical Guide to Research Methods (2015) 222-48 48. Greenhalgh T, Hinder S, Stramer K, Bratan T, Russell J. Adoption, non-adoption, and abandonment of a personal electronic health record: case study of HealthSpace. BMJ (2010) 341 c5814. DOI: 10.1136/bmj.c5814 49. McAuley A. Digital health interventions: widening access or widening inequalities?. Public Health (2014) 128 1118-20. DOI: 10.1016/j.puhe.2014.10.008 50. Al-Asadi AM, Klein B, Meyer D. Pretreatment attrition and formal withdrawal during treatment and their predictors: an exploratory study of the anxiety online data. J Med Internet Res (2014) 16 e152. DOI: 10.2196/jmir.2989 51. Nijland N, van Gemert-Pijnen JE, Kelders SM, Brandenburg BJ, Seydel ER. Factors influencing the use of a web-based application for supporting the self-care of patients with type 2 diabetes: a longitudinal study. J Med Internet Res (2011) 13 e71. DOI: 10.2196/jmir.1603 52. Jensen JD, King AJ, Davis LA, Guntzviller LM. Utilization of internet technology by low-income adults: the role of health literacy, health numeracy, and computer assistance. J Aging Health (2010) 22 804-26. DOI: 10.1177/0898264310366161 53. Veinot TC, Mitchell H, Ancker JS. Good intentions are not enough: how informatics interventions can worsen inequality. J Am Med Inform Assoc (2018) 25 1080-8. DOI: 10.1093/jamia/ocy052 54. Howells K, Bower P, Burch P, Cotterill S, Sanders C. On the borderline of diabetes: understanding how individuals resist and reframe diabetes risk. Health Risk Soc (2021) 23 34-51. DOI: 10.1080/13698575.2021.1897532 55. Sheeran P, Harris PR, Epton T. Does heightening risk appraisals change people's intentions and behavior? A meta-analysis of experimental studies. Psychol Bull (2014) 140 511-43. DOI: 10.1037/a0033065 56. Troughton J, Jarvis J, Skinner C, Robertson N, Khunti K, Davies M. Waiting for diabetes: perceptions of people with pre-diabetes: a qualitative study. Patient Educ Couns (2008) 72 88-93. DOI: 10.1016/j.pec.2008.01.026 57. Ross J, Stevenson FA, Dack C, Pal K, May CR, Michie S, Yardley L, Murray E. Health care professionals' views towards self-management and self-management education for people with type 2 diabetes. BMJ Open (2019) 9 e029961. DOI: 10.1136/bmjopen-2019-029961 58. Slevin P, Kessie T, Cullen J, Butler MW, Donnelly SC, Caulfield B. Exploring the barriers and facilitators for the use of digital health technologies for the management of COPD: a qualitative study of clinician perceptions. QJM (2020) 113 163-72. DOI: 10.1093/qjmed/hcz241 59. Ross J, Stevenson F, Dack C, Pal K, May C, Michie S, Barnard M, Murray E. Developing an implementation strategy for a digital health intervention: an example in routine healthcare. BMC Health Serv Res (2018) 18 794. DOI: 10.1186/s12913-018-3615-7 60. Vaja I, Umeh KF, Abayomi JC, Patel T, Newson L. A grounded theory of type 2 diabetes prevention and risk perception. Br J Health Psychol (2021) 26 789-806. DOI: 10.1111/bjhp.12503 61. Moin T, Ertl K, Schneider J, Vasti E, Makki F, Richardson C, Havens K, Damschroder L. Women veterans' experience with a web-based diabetes prevention program: a qualitative study to inform future practice. J Med Internet Res (2015) 17 e127. DOI: 10.2196/jmir.4332 62. Schippers M, Adam PC, Smolenski DJ, Wong HT, de Wit JB. A meta-analysis of overall effects of weight loss interventions delivered via mobile phones and effect size differences according to delivery mode, personal contact, and intervention intensity and duration. Obes Rev (2017) 18 450-9. DOI: 10.1111/obr.12492 63. Ekezie W, Dallosso H, Saravanan P, Khunti K, Hadjiconstantinou M. Experiences of using a digital type 2 diabetes prevention application designed to support women with previous gestational diabetes. BMC Health Serv Res (2021) 21 772. DOI: 10.1186/s12913-021-06791-9 64. Brunton L, Soiland-Reyes C, Wilson P. Implications for future policy implementation: a qualitative evaluation of the national rollout of a diabetes prevention programme in England. Res Sq (2022). DOI: 10.21203/rs.3.rs-1776086/v1
--- title: Adipocyte autophagy limits gut inflammation by controlling oxylipin and IL‐10 authors: - Felix Clemens Richter - Matthias Friedrich - Nadja Kampschulte - Klara Piletic - Ghada Alsaleh - Ramona Zummach - Julia Hecker - Mathilde Pohin - Nicholas Ilott - Irina Guschina - Sarah Karin Wideman - Errin Johnson - Mariana Borsa - Paula Hahn - Christophe Morriseau - Bruce D Hammock - Henk Simon Schipper - Claire M Edwards - Rudolf Zechner - Britta Siegmund - Carl Weidinger - Nils Helge Schebb - Fiona Powrie - Anna Katharina Simon journal: The EMBO Journal year: 2023 pmcid: PMC10015370 doi: 10.15252/embj.2022112202 license: CC BY 4.0 --- # Adipocyte autophagy limits gut inflammation by controlling oxylipin and IL‐10 ## Abstract Lipids play a major role in inflammatory diseases by altering inflammatory cell functions, either through their function as energy substrates or as lipid mediators such as oxylipins. Autophagy, a lysosomal degradation pathway that limits inflammation, is known to impact on lipid availability, however, whether this controls inflammation remains unexplored. We found that upon intestinal inflammation visceral adipocytes upregulate autophagy and that adipocyte‐specific loss of the autophagy gene Atg7 exacerbates inflammation. While autophagy decreased lipolytic release of free fatty acids, loss of the major lipolytic enzyme Pnpla2/Atgl in adipocytes did not alter intestinal inflammation, ruling out free fatty acids as anti‐inflammatory energy substrates. Instead, Atg7‐deficient adipose tissues exhibited an oxylipin imbalance, driven through an NRF2‐mediated upregulation of Ephx1. This shift reduced secretion of IL‐10 from adipose tissues, which was dependent on the cytochrome P450‐EPHX pathway, and lowered circulating levels of IL‐10 to exacerbate intestinal inflammation. These results suggest an underappreciated fat‐gut crosstalk through an autophagy‐dependent regulation of anti‐inflammatory oxylipins via the cytochrome P450‐EPHX pathway, indicating a protective effect of adipose tissues for distant inflammation. Gut inflammation modulates autophagy‐dependent oxylipin metabolism in distant fat cells to trigger the release of anti‐inflammatory signals. ## Introduction Autophagy is an essential cellular recycling pathway that engulfs cellular contents, including organelles and macromolecules, in a double membraned autophagosome and directs them toward lysosomal degradation. Many cell types, including immune cells, are reliant on autophagy during their differentiation and for their functions (Clarke & Simon, 2019). Consequently, autophagy dysfunction is associated with the development of a variety of inflammatory diseases and metabolic disorders (Deretic, 2021; Klionsky et al, 2021). Inflammatory bowel diseases (IBD) including its two predominant manifestations, Crohn's disease (CD) and ulcerative colitis (UC), describe a complex spectrum of intestinal inflammation. Genome‐wide association studies identified autophagy‐related genes as susceptibility alleles in IBD (Hampe et al, 2007; McCarroll et al, 2008; Jostins et al, 2012). Mechanistic studies revealed that ablation of autophagy in immune and epithelial cells promotes intestinal inflammation (Cadwell et al, 2008, 2009; Kabat et al, 2016). In addition to the strong genetic association of autophagy and IBD, patients with CD often present with an expansion of the mesenteric adipose tissue around the inflamed intestine, indicating an active involvement of the adipose tissue in the disease pathology (Sheehan et al, 1992). Adipose tissues represent an important immunological organ harboring a variety of immune cells, which are highly adapted to live in lipid‐rich environments such as adipose tissue macrophages (ATMs) (Trim & Lynch, 2021). Lean adipose tissues are predominantly populated by tissue‐resident M2‐type ATMs, while inflammation, such as induced by obesity, subverts their homeostatic function and promotes pro‐inflammatory M1‐type polarization (Russo & Lumeng, 2018). Polarization and function of ATMs depend on the integration of a variety of inflammatory and metabolic signals. M2‐type macrophages rely on the availability and uptake of lipids and the subsequent metabolization of free fatty acids (FFA) compared to M1‐type macrophages (Huang et al, 2014). In addition, oxygenated polyunsaturated fatty acids, so called oxylipins, which are produced through enzymatic lipid oxidation can be released from adipocytes to modify macrophage cytokine production (Klein‐Wieringa et al, 2013). Oxylipins have been widely described as regulatory lipid mediators that regulate inflammatory processes and resolution (Imig & Hammock, 2009; Gilroy et al, 2016; Edin et al, 2018). It is plausible that, either or both, availability of energy substrates such as FFA and signaling through oxylipin mediators will modulate immune responses. Autophagy contributes to FFA release (Singh et al, 2009) and lipid peroxidation (Cai et al, 2018), however, to‐date, little is known about the impact of autophagy in adipocytes on these metabolic cues and whether these may affect inflammation. Here, we sought to investigate the impact of adipocyte autophagy on the immune system during inflammation of a distant organ, the intestine. We observed that autophagy is induced in mature adipocytes upon dextran sulphate sodium (DSS)‐induced intestinal inflammation, and that loss of autophagy in adipocytes exacerbated gut inflammation. Mechanistically, while autophagy in mature adipocytes is required for the optimal release of FFA during inflammation, this was not causative for increased intestinal inflammation. Instead, loss of adipocyte autophagy stabilized the oxidative stress master transcription factor NRF2 and promoted the oxylipin pathway activity shifting the balance of intratissual oxylipins. Local oxylipin imbalance limited the production of anti‐inflammatory IL‐10 from ATMs, aggravating intestinal inflammation. Taken together, we demonstrate a novel mechanism of autophagy in adipocytes regulating local oxylipins that promote the anti‐inflammatory fat‐gut crosstalk, highlighting the importance of intertissual control of inflammation. ## DSS‐induced intestinal inflammation induces lipolysis and autophagy in adipose tissues To investigate how the adipose tissue is affected by intestinal inflammation, we deployed a mouse model of intestinal inflammation evoked by the administration of DSS in drinking water (Fig 1A). As expected, treatment with DSS led to an increased histopathological inflammation, shortened colon length, enlarged mesenteric lymph nodes, and elevated circulating levels of the pro‐inflammatory cytokine TNFα (Fig EV1A–D). In addition, DSS treatment resulted in a significantly higher infiltration of immune cells in the inflamed colon, predominantly of myeloid origin (Fig EV1E and F). Furthermore, DSS colitis reduced body weight (Fig 1B), and in line with that, visceral adipose tissue mass (Fig 1C), as well as serum FFA levels (Fig 1D). Next, we examined whether DSS‐induced colitis can lead to the activation of key lipolysis enzymes in the adipose tissue, contributing to the decrease in adipose tissue mass. We confirmed that DSS‐induced colitis increased phosphorylation of hormone‐sensitive lipase (HSL) and the expression of adipose triglyceride lipase (ATGL) (Fig 1E), in line with increased lipolytic activity from the adipocytes. Figure 1: Intestinal inflammation promotes lipolysis and autophagy in adipose tissues Schematic of experimental design. Sex‐matched and age‐matched wild‐type mice were treated for 5 days with 1.5–2% DSS in drinking water, before switched to water for two more days. Mice were sacrificed on day 7 post‐DSS induction.Body weight development upon DSS treatment (n = 13/group).Tissue weights measured in mesenteric (mWAT) and collective visceral white adipose tissue (visWAT), consisting of gonadal (gWAT), retroperitoneal and omental white adipose tissue on day 7 after start of DSS regime (n = 8/group).Circulating serum levels of FFA during DSS‐induced colitis on day 7 (n = 15/group).Representative immunoblot for key lipolytic enzymes HSL and ATGL protein expression and quantification (n = 5–6/group).Immunoblot analysis of autophagic flux in mWAT (upper panel) and gWAT (lower panel) adipose tissue stimulated ex vivo with lysosomal inhibitors 100 nM Bafilomycin A1 and 20 mM NH4Cl for 4 h or DMSO (Vehicle) (n = 3–4/group).Representative transmission electron microscopy images from mesenteric adipose tissue 7 days post DSS‐induced colitis induction. Lower panel is showing magnification of selected area. White arrows show autophagosomal structures. Atg8 homologs expression was measured by qPCR in visceral adipocytes fraction (left panel) and stromal vascular fraction (right panel) during DSS‐induced colitis (n = 7–8/group).Representative immunoblot for LC3‐I/‐II protein expression and quantification of autophagic flux in gWAT via ex vivo lysosomal inhibition using 100 nM Bafilomycin A1 and 20 mM NH4Cl for 4 h or DMSO (Vehicle). Mice were initially treated with 500 μg anti‐TNFα antibody or isotype control, before administrating DSS in drinking water for 5 days. Mice were sacrificed on day 7 post‐DSS induction (n = 5–6/group).Representative immunoblot for LC3‐I/‐II and ACTIN protein expression and quantification of autophagic flux in creeping fat tissues (CrF) and adjacent mesenteric adipose tissues (Ad. MAT) of Crohn's disease patients (n = 3/group). Additionally, autophagic flux was determined in the mesenteric adipose tissue (MAT) of a colorectal cancer patient as control (dotted line). Data are represented as mean ± s.e.m. (B) Two‐Way repeated measures ANOVA. (C–E, G) Unpaired Student's t‐test. (I) Two‐Way ANOVA. (J) Paired Student's t‐test. P < 0.05, P < 0.01, P < 0.001. Source data are available online for this figure. Figure EV1: DSS leads to efficient induction of intestinal inflammation Representative H&E staining of colon histology and quantification on day 7 after DSS colitis induction (n = 3/group) from one independent experiment.Colon length measured after 1.5–2% DSS colitis regime on day 7 (n = 10–11/group).Spleen weight and mesenteric lymph node weight after 1.5–2% colitis regime on day 7 (n = 9–10/group).TNFα levels in serum were measured in wild‐type mice on day 7 after water and DSS treatment (n = 5/group).Absolute number of colonic CD45+ immune cells on day 7 post‐DSS treatment (n = 6–7/group).Frequency of CD11b+ myeloid cells, CD3+ T cells and CD19+ B cells in colon on day 7 post‐DSS treatment (n = 5–7/group).Colon length of mice upon DSS‐induced colitis treated with anti‐isotype or anti‐TNFα neutralizing antibody (n = 7/group).Body weight development upon DSS‐induced colitis of mice treated either with anti‐isotype or anti‐TNFα neutralizing antibody (n = 7/group).Tissue weight of mWAT or gWAT upon DSS‐induced colitis of mice treated either with anti‐isotype or anti‐TNFα neutralizing antibody (n = 7/group). Data are represented as mean ± s.e.m. (A–E) Unpaired Student's t‐test. (G, I) Two‐way ANOVA. (H) Repeat‐measure two‐way ANOVA. P < 0.01, P < 0.001, *P < 0.0001. Source data are available online for this figure. Lipids can also be provided via the classical lipolysis or through degradation of the lipid droplet via autophagy. Thus, we assessed the autophagy levels in adipose tissue explants from water‐ or DSS‐treated animals, which were cultured in the absence or presence of lysosomal inhibitors and the accumulation of the lipidated autophagosomal marker LC3 protein (LC3‐II) was quantified. DSS‐induced intestinal inflammation substantially increased autophagic flux in mesenteric and in gonadal white adipose tissue (mWAT and gWAT, respectively) (Fig 1F), indicating that both adipose tissues proximal and distal to the intestine are responsive to the inflammation. To validate that adipocytes, but not other adipose tissue‐resident cell types, contribute to the increased autophagic flux in the adipose tissue, we first prepared adipose tissues for transmission electron microscopy. Autophagosomal double‐membrane structures were readily identified in adipocytes from DSS‐treated mice (Fig 1G). Additionally, enriched adipocytes, but not vascular stromal fractions, increased the expression of several Atg8 homologs upon DSS colitis, further demonstrating an induction of autophagy in this cell type (Fig 1H). TNFα has previously been shown to be a potent inducer of autophagy in in vitro differentiated 3T3‐L1 cells (Ju et al, 2019), prompting the hypothesis that the release of TNFα during DSS‐induced intestinal inflammation augments autophagic flux in the adipose tissue. To test this, we blocked TNFα in vivo using a neutralizing antibody. Mice treated with anti‐TNFα and anti‐Isotype showed similar body weight loss and colon shortening during DSS‐induced inflammation, indicating that the mice were similarly inflamed (Fig EV1G and H). In contrast, loss of adipose tissue mass was partially prevented by neutralization of TNFα (Fig EV1I), possibly indicating a reduced release of lipids from the adipose tissue. Importantly, we found that adipose tissue autophagic flux was reduced upon anti‐TNFα treatment (Fig 1I). Lastly, we wanted to assess whether the increase in autophagic flux also occurs in IBD. For this purpose, we collected creeping fat tissues and adjacent noninflamed mesenteric adipose tissues from CD patients. Interestingly, similar to the DSS‐induced mice, we found an increased autophagic flux in the creeping fat compared to the same patient's adjacent noninflamed mesenteric adipose tissue (Fig 1J). Overall, these results demonstrate that lipolysis and autophagy are induced in adipocytes in response to DSS‐induced intestinal inflammation. ## Loss of adipocyte autophagy exacerbates intestinal inflammation Given the increased adipose autophagy we observed as a reaction to intestinal inflammation, we next investigated whether loss of autophagy in adipocytes affects intestinal inflammation. To exclude developmental effects of autophagy loss (Singh et al, 2009; Zhang et al, 2009), we used a tamoxifen‐inducible knockout mouse model to ablate the essential autophagy gene Atg7 specifically in mature adipocytes (Atg7 Ad) in adult mice (Fig 2A). Tamoxifen administration led to the significant reduction of Atg7 transcript levels in visceral adipocytes (Fig 2B). Importantly, the adipocyte‐specific loss of Atg7 resulted in the interruption of conversion of LC3‐I to LC3‐II in the adipose tissue (Fig 2C), confirming effective disruption of the autophagic process in adipose tissue. Figure 2: Loss of adipocyte autophagy exacerbates DSS‐induced colitis Schematic of experimental design. Sex‐matched and age‐matched littermates were treated at 8–12 weeks of age with tamoxifen for 5 consecutive days before tissues were analyzed 14 days after the last tamoxifen administration (Steady State).Representative quantification of knock‐out efficiency measured on Atg7 transcript level by qRT‐PCR in purified primary visceral adipocyte at 2 weeks post‐tamoxifen treatment (n = 4–11/group).Representative immunoblot for LC3‐I and LC3‐II protein expression and quantification of LC3 conversion ratio (LC3‐II/LC3‐I) (n = 3/group).Schematic of experimental design. Sex‐matched and age‐matched littermates were treated at 8–12 weeks of age with tamoxifen for 5 consecutive days and DSS‐induced colitis was induced after a 2‐week washout phase (DSS Day 7).Body weight development upon DSS treatment (n = 25/group).Colon length after 2 weeks postdeletion (steady state; n = 14/group) and after DSS on day 7 (n = 18–22/group).Representative H&E staining images (10× magnification) of colon sections and quantification of histological score at steady state (n = 9/group) and DSS colitis (n = 18–22/group).Expression of pro‐inflammatory cytokines in colon tissues at 7 days post‐DSS induction (n = 18–22/group). Dotted line represents uninflamed controls.Absolute number CD45+ immune cells from colons at steady state (n = 13–14/group) or at 7 days post‐DSS induction (n = 18–22/group).Frequency of myeloid cell population in colon on day 7 post‐DSS induction (n = 18–25/group).Absolute number of Ly6C+ monocytes discriminated by the absence or presence of MHCII for infiltrating and inflammatory monocytes, respectively (n = 18–25/group). Data are represented as mean ± s.e.m. (B, C, F–I) Unpaired Student's t‐test. (E) Two‐Way repeated measures ANOVA. (J, K) Two‐Way ANOVA. P < 0.05, P < 0.01, P < 0.001, **P < 0.0001. Source data are available online for this figure. Having confirmed efficient deletion of Atg7 and disruption of autophagic flux in adipocytes, we next compared the effects of autophagy loss in adipocytes in steady‐state and DSS‐induced colitis. As tamoxifen is known to induce autophagy, and we found this to be true in this setting, we included tamoxifen treatment for all genotypes and added a 2‐week wash‐out period before treatment with DSS (Fig 2D). In all assessed parameters, loss of adipocyte autophagy in steady‐state mice had no effects on intestinal immune homeostasis. In contrast, Atg7 Ad mice showed an increased loss of body weight in comparison to littermate controls upon DSS‐treatment (Fig 2E). In addition, Atg7 Ad mice treated with DSS had significantly shorter colon when compared to their wild‐type littermates during acute inflammation (Fig 2F). Blinded histopathological assessment (Dieleman et al, 1998) confirmed that DSS‐treated Atg7 Ad mice exhibited more severe tissue damage accompanied by increased inflammation and reduced features of repair throughout the colon (Fig 2G). Consistent with an increased inflammatory response, we found increased gene expression of alarmins such as Il1a and Il33, pro‐inflammatory cytokines Tnfa, Ptx3, Ifng, and the IFNγ‐regulated chemokine Cxcl9 in Atg7 Ad mice (Fig 2H). Although total CD45+ immune cell numbers were comparable between adipocyte autophagy‐deficient mice and littermate controls (Fig 2I), DSS‐inflamed Atg7 Ad mice showed an increased frequency of monocytes infiltrating the intestinal tissue (Fig 2J). In particular, the number of MHCII‐expressing, inflammatory monocytes were increased in the lamina propria of Atg7 Ad mice (Fig 2K). This phenotype is in line with the fact that autophagic flux was induced in adipose tissues upon intestinal damage by DSS (Fig 1F) suggesting an important function of adipocyte autophagy during intestinal inflammation. Taken together, these data demonstrate that loss of adipocyte autophagy exacerbates intestinal inflammation in the acute phase of DSS‐induced colitis. Since intestinal inflammation induced by DSS is self‐resolving, we assessed the impact of adipocyte autophagy loss during resolution of the inflammation (Fig EV2A). Two weeks after initial DSS administration, we did not find any differences in colon length between Atg7 Ad and littermate controls, and equally, there were no significant histopathological differences observed between the groups (Fig EV2B and C). Interestingly, frequencies and total numbers of colonic FOXP3+ regulatory T cells (Tregs) were decreased in adipocyte autophagy‐deficient animals compared to wild‐type animals (Fig EV2D), despite not affecting disease recovery. Intestinal FOXP3+ Tregs are classified into three distinct subsets based on co‐expression of TH2 and TH17 transcription factors GATA3+ and RORgt+, respectively (Whibley et al, 2019). While all populations tended to be diminished in Atg7 Ad mice, only RORgt− FOXP3+ Tregs were significantly reduced (Fig EV2E). These data suggest that adipocyte autophagy is dispensable for the resolution of DSS‐induced inflammation but may affect expansion of intestinal Tregs in response to intestinal tissue injury. Figure 3: Autophagy loss reduces secretion of fatty acids from adipocytes Ex vivo lipolysis measured by released free fatty acid (left, n = 4–5/group) and glycerol (right, n = 7–8/group) in culture supernatant of adipose tissue explants simulated with isoproterenol (10 μM) for 1–2 h. Ex vivo lipolysis measured by released free fatty acid (left, n = 4/group) and glycerol (right, n = 7/group) adipose tissue explants simulated with TNFα (100 ng/ml) for 24 h before replacing with fresh medium in the absence of TNFα for 3 h.Representative immunoblot for key lipolytic enzymes HSL, pHSL (Ser660) and quantification (n = 10–11/group).Serum levels of circulating FFAs measured in wild‐type and Atg7‐deficient mice (n = 13–14/group).Concentration of individual FFA species in serum in water‐treated and DSS‐treated mice as measured by FID‐GC (n = 12–14/group). Data are represented as mean ± s.e.m. (A, B, E) Two‐Way ANOVA. (B, D) Unpaired Student's t‐test. (C) Mann–Whitney test. P < 0.05, P < 0.01, P < 0.001. Source data are available online for this figure. Figure EV2: Expansion of intestinal Treg populations is blunted in adipocyte autophagy‐deficient mice without affecting intestinal resolution Schematic of experimental design. Sex‐matched and age‐matched littermates were treated with DSS for 5 days and mice were sacrificed 14 days after start of DSS treatment.Colon length from noninflamed control mice (n = 8/group), adipocyte autophagy‐sufficient WT mice and adipocyte autophagy‐deficient mice (n = 12/group).Representative H&E staining images (10× magnification) of distal colon sections and quantification of histopathological score (n = 7–13/group).Frequency (left panel) and absolute number (right panel) of CD4+ FOXP3+ cells in the colon on day 14 post‐DSS treatment (n = 8–11/group).Frequency of peripheral and thymic Treg (pTreg and tTreg, respectively) cell populations in colon on day 14 post‐DSS treatment (n = 8–11/group).Data are represented as mean ± s.e.m. (B–D) One‐way ANOVA. (E) Two‐way ANOVA. P < 0.05, P < 0.01, *P < 0.0001. Source data are available online for this figure. ## Adipocyte autophagy promotes FFA secretion Recent reports implicated autophagy in mature adipocytes in the secretion of FFA in response to β‐adrenergic receptor‐mediated lipolysis (Cai et al, 2018; Son et al, 2020). To confirm the importance of adipocyte autophagy for optimal lipolytic output, adipose tissue explants were stimulated with the β‐adrenergic receptor agonist isoproterenol and FFA levels were quantified. As expected, FFA and glycerol release was reduced upon lipolysis stimulation in autophagy‐deficient as compared to autophagy‐proficient adipocytes (Fig 3A). TNFα is a crucial cytokine for human and murine IBD pathologies (Friedrich et al, 2019) and it can affect adipose tissue through inhibition of lipogenesis and by promoting FFA secretion (Cawthorn & Sethi, 2008). Since circulating TNFα levels were elevated during DSS colitis (Fig EV1D), we investigated its effects on adipocyte lipid metabolism. In the presence of TNFα, FFA and glycerol release was significantly blunted in autophagy‐deficient compared to wild‐type adipocytes (Fig 3B). Next, the impact of adipocyte autophagy loss on the induction of lipolysis in the context of DSS‐induced colitis was assessed. Induction of HSL phosphorylation was reduced in adipose tissues of Atg7 Ad mice, suggesting a reduced lipolytic potential of autophagy‐deficient adipocytes (Fig 3C). Consistent with the decreased lipolytic activity of autophagy‐deficient adipocytes, Atg7 Ad mice exhibit reduced serum FFA levels compared to wild‐type littermates upon DSS colitis (Fig 3D). While we established that autophagy could modulate overall FFA release, we next tested whether autophagy affects the production and secretion of specific FFA species. To investigate this, serum samples from water‐ and DSS‐treated animals were analyzed by GC‐FID. Confirming our initial findings, the serum concentration of many FFA species was reduced upon adipocyte autophagy loss, indicating that adipocyte autophagy controls overall FFA levels rather than specific FFAs (Fig 3E). Interestingly, loss of adipose tissue mass was comparable between both genotypes upon DSS‐induced colitis (Fig EV3A). Figure 4: Adipocyte‐specific loss of Atgl was dispensable for regulation of intestinal inflammation Schematic of experimental design. DSS‐induced colitis was induced in sex‐matched and age‐matched littermates.Representative quantification of knockout efficiency measured on Atgl transcript level by qRT‐PCR in purified primary visceral adipocyte (n = 3–8/group). Ex vivo lipolysis assays on Atg7‐deficient adipose tissue explants simulated with isoproterenol (10 μM) for 1–2 h (n = 5–6/group).Body weight development upon DSS treatment (n = 8/group).Tissue weights of mWAT and visWAT on day 7 after start of DSS (n = 3–8/group).Colon length after DSS on day 7 (n = 3–8/group).Quantification of histological score at steady state (n = 3/group) and DSS colitis (n = 6–7/group).Expression of pro‐inflammatory cytokines in colon tissues on 7 days post‐DSS induction (n = 8/group). Dotted line represents noninflamed controls.Absolute number CD45+ immune cells from colons on 7 days post‐DSS induction (n = 3–8/group).Frequency of myeloid cell population in colon on day 7 post‐DSS induction (n = 8/group).Absolute number of Ly6C+ monocytes discriminated by the absence or presence of MHCII for infiltrating and inflammatory monocytes, respectively (n = 8/group). Data are represented as mean ± s.e.m. (B, E–I, K) Unpaired Student's t‐test. (D) Two‐Way repeated measures ANOVA. (C, J) Two‐Way ANOVA. **P < 0.0001. Source data are available online for this figure. Figure EV3: Loss of adipocyte autophagy had no effects on adipose tissue and circulating levels of leptin and adiponectin Adipose tissue mass at steady state and on day 7 post‐DSS induction (n = 7–11/group).Circulating levels of adiponectin (n = 3–12/group).Circulating levels of leptin (n = 4–12/group).Data are represented as mean ± s.e.m. (A) Unpaired Student's t‐test. (B, C) One‐way ANOVA. P < 0.01. Source data are available online for this figure. It has previously been described that the adipokines leptin and adiponectin can influence intestinal inflammation in both preclinical and clinical situations (Siegmund et al, 2002; Weidinger et al, 2018), we, therefore, assessed the impact of adipocyte autophagy loss on circulating levels of these adipokines. The levels of both adipokines were equally reduced in their circulation, paralleling the general loss of adipose tissue mass (Fig EV3B and C). Taken together, our data suggest that adipocyte autophagy fine‐tunes the lipolytic output in an inflammatory setting. ## Adipocyte lipolysis is dispensable for DSS‐induced colitis severity Based on our data, we hypothesized that differences in FFA availability may be responsible for a differential intestinal immune response. We, therefore, sought to determine the importance of adipocyte lipolysis during DSS‐induced colitis (Fig 4A) by deleting the cytoplasmic lipase Pnpla2/Atgl, a rate‐limiting enzyme in the lipolytic pathway (Schweiger et al, 2006). Using adipocyte‐specific Pnpla2/Atgl (Atgl Ad) knockout mice, we first confirmed that Pnpla2/Atgl was efficiently deleted in purified visceral adipocytes (Fig 4B), leading to a strong reduction of isoproterenol‐induced FFA release (Fig 4C). Strikingly, upon DSS‐induced colitis, Atgl Ad mice lost comparable amounts of body weight (Fig 4D), although adipose tissue loss was completely prevented (Fig 4E). These data underline that Atgl‐driven lipolysis is a main driver for adipose tissue loss during DSS‐induced colitis. However, detailed analysis of the colon showed no changes in colon shortening, histopathological scores, and expression of inflammatory cytokines (Fig 4F–H). Similarly, there was no difference in the infiltration and presence of different pro‐inflammatory immune cell population in the colonic lamina propria (Fig 4I–K). In summary, inhibition of bulk FFA release from adipocytes through Atgl loss does not mimic the effects on intestinal inflammation observed in Atg7 Ad mice. This suggests that provision of FFA is unlikely to be the mechanism by which autophagy in adipocytes exerts its anti‐inflammatory role. ## Intestinal inflammation promotes a lipolytic transcriptional profile in primary adipocytes At this point, it remained unclear how adipocytes regulate intestinal inflammation. We hypothesized that visceral adipocytes would alter their transcriptional inflammatory profile during intestinal inflammation. To test this, visceral adipocytes were collected from wild‐type and Atg7 Ad mice treated with water or DSS and subjected to RNA sequencing. Since we anticipated sex‐specific differences in adipocyte transcription profiles, we included the same number of male and female mice in each experimental group. The treatment clearly separated the experimental groups in the principal component analysis (PCA) (Fig EV4A). As expected, sex‐specific transcriptional changes explained ~$33\%$ of the dataset variance (Fig EV4A), in line with previous reports (Oliva et al, 2020). Next, we compared noninflamed to inflamed adipocytes by regressing genotype and sex to identify the impact of intestinal inflammation on the adipocyte transcriptome. More than 4,700 genes were differentially regulated between these states (Fig EV4B), among which 2,415 were significantly upregulated and 2,333 downregulated. Gene ontology analysis of these differentially expressed genes revealed an enrichment in several pathways (Fig EV4C). Confirming our earlier results that adipocyte autophagy is affected by DSS‐induced colitis (Fig 1H), intestinal inflammation led to an enrichment of genes involved in macroautophagy in visceral adipocytes, including an increased expression of several Atg8 homologs (Gabarap, Gabarapl1, Map1lc3a, Map1lc3b) (Fig EV4E and F). In addition, genes related to fatty acid metabolism were enriched in visceral adipocytes during intestinal inflammation (Fig EV4D). Similar to cachexic conditions, an increase in lipolytic genes (Lipe, Pnpla2) and simultaneous decrease in lipogenic genes (Dgat2, Mogat2, Lpl) was observed (Baazim et al, 2021). Overall, intestinal inflammation leads to a broad transcriptional response in visceral adipocytes, altering autophagy and fatty acid metabolism, which is reminiscent of a cachexic response phenotype (Baazim et al, 2021). These findings are in line with our previous observations. Figure EV4: Intestinal inflammation induces distinct transcriptional programs in primary visceral adipocytes Principal component analysis of all mice revealing a strong sex effect in the overall transcriptome.Differential gene expression assessing transcriptional changes associated with DSS‐induced inflammation after regressing effect of sex and genotypes in visceral adipocytes.Pathway enrichment analysis of significantly differentially expressed genes in visceral adipocytes during DSS colitis.Heatmap representing differentially expressed genes associated in fatty acid metabolism during DSS‐induced colitis in visceral adipocytes.Heatmap representing differentially expressed genes associated with macroautophagy during DSS‐induced colitis in visceral adipocytes.Normalized counts of Atg8 homologs in visceral adipocytes (n = 12/group). Data are represented as mean ± s.e.m. (F) Unpaired Student's t‐test. P < 0.05, **P < 0.001 Source data are available online for this figure. ## Adipocyte autophagy loss promotes NRF2‐mediated stress response and alters tissue oxylipin levels To get a better understanding of pathways that may be affected by the loss of autophagy in adipocytes, we further analyzed our transcriptomic data by splitting the dataset based on their condition and genotype. Visceral adipocytes from Atg7 Ad mice had a strong reduction in Atg7 levels and an increase in estrogen receptor 1 (Esr1) expression (Fig 5A and B). The latter was verified to be caused by the overexpression of the Cre‐ERT2 construct which mapped to mouse Esr1. Across both treatment groups, we found a total of 32 genes being differentially regulated between WT and Atg7 Ad visceral adipocytes. Six genes were differentially expressed under both water and DSS treatment conditions (Fig 5C). Using ranked gene set enrichment analysis (GSEA) (Subramanian et al, 2005), we found that the xenobiotic pathway was significantly enriched in Atg7 Ad adipocytes upon DSS‐induced colitis (Figs 5D and EV5A). Enzymes which are known for their role in xenobiotic metabolism, such as the large family of cytochrome P450 monooxygenases and epoxide hydrolases (EPHX), metabolize and detoxify exogenous substrates and mediate the production of oxylipins from endogenous polyunsaturated fatty acids. The expression of many of the key genes involved in these processes are regulated by NRF2, a major transcription factor of the xenobiotic and oxidative stress responses. We found that Ephx1 was consistently upregulated upon Atg7 loss in adipocytes (Fig 5C). Remarkably, Ephx1 expression was also increased in datasets obtained from other studies in which autophagy genes such as Atg3 and Beclin‐1 were specifically deleted in adipocytes (Fig EV5B and C) (Cai et al, 2018; Son et al, 2020). Among the genes that were enriched in Atg7‐deficient adipocytes were several other NRF2‐target genes (Fig 5E). In agreement with an activation of the NRF2 pathway, NRF2 protein abundance was increased in Atg7 Ad visceral adipose tissues (Fig 5F). Specificity for NRF2 activation was further confirmed since only NRF2 target gene Ephx1 was transcriptionally upregulated, whereas Ephx2 which is not controlled by NRF2 remained transcriptionally unchanged in autophagy‐deficient adipocytes (Fig 5G). However, both EPHX1 and EPHX2 protein expression were increased in Atg7 Ad adipose tissues (Fig 5H) suggesting that EPHX2 may be affected by autophagy deletion on a post‐transcriptional level. Figure 5: Adipocyte autophagy loss activates NRF2‐EPHX1 pathway and alters intratissual oxylipin balance Differential gene expression in visceral adipocytes from water‐treated WT and Atg7 Ad animals 2 weeks after tamoxifen treatment.Differential gene expression in visceral adipocytes from DSS‐treated WT and Atg7 Ad animals on day 7 post‐DSS treatment.Venn diagram of commonly regulated genes between Atg7‐deficient and Atg7‐sufficient adipocytes during water or DSS treatment.GSEA enrichment analysis between Atg7‐deficient and Atg7‐sufficient adipocytes during DSS treatment.Fold change expression of NRF2‐target genes in primary visceral adipocytes on day 7 after DSS induction from normalized counts of RNAseq dataset (n = 6/group).Representative immunoblot for NRF2 protein expression and quantification (n = 16–18/group).Transcriptional expression of Ephx1 and Ephx2 in visceral adipocytes on day 7 after DSS induction (n = 20–25/group).Representative immunoblot of EPHX1 and EPHX2 in gonadal adipose tissues on day 7 after DSS induction. Asterix indicating nonspecific band (n = 14–18/group).Schematic overview of cytochrome P450‐EPHX oxylipin pathway.Normalized fold change differences in epoxy fatty acid precursor fatty acids, docosahexaenoic acid (DHA), arachidonic acid (AA), and linoleic acid (LA) in mWAT and gWAT (n = 13–14/group).Normalized ratios of epoxy fatty acid to their corresponding diol fatty acid pairs in mWAT and gWAT (n = 6–8/group). Data are represented as mean ± s.e.m. (E–H, L) Unpaired Student's t‐test. (J, K) Two‐way ANOVA. P < 0.05, P < 0.01, P < 0.001, **P < 0.0001. Source data are available online for this figure. Figure EV5: Loss of autophagy‐related genes results in the induction of epoxy hydrolases in adipocytes GSEA enrichment analysis between Atg7‐deficient and Atg7‐sufficient adipocytes during DSS treatment.Fragments per kilobase of exon per million mapped fragments (FPKM) counts from bulk RNAseq dataset of Cai et al (2018) (n = 4/group)Fragments per kilobase of exon per million mapped fragments (FPKM) counts from bulk RNAseq dataset of Son et al (2020) (n = 3/group).Normalized ratios of epoxy fatty acid precursors to their corresponding diol fatty acids pairs in plasma (n = 8/group).Data are represented as mean ± s.e.m. (B, C) Unpaired Student's t‐test. (D) Two‐way ANOVA. P < 0.05, *P < 0.01. Source data are available online for this figure. EPHX1, together with EPHX2, are central for the enzymatic conversion of cytochrome P450‐derived oxylipins such as epoxy fatty acids (EpFA) to dihydroxy/diol fatty acids (DiolFA). EpFA have strong anti‐inflammatory, analgesic, and hypotensive activity, while DiolFA are less biologically active and are associated with more pro‐inflammatory properties (Fig 5I) (McReynolds et al, 2020). In the mesenteric and gonadal adipose tissues, the abundance of nonesterified linoleic acid was reduced (Fig 5J), possibly reflecting their reduced lipolytic capacity. Since we found predominantly changes in EPHX enzyme expression, we tested whether this would shift the balance of oxylipins in the tissue and, possibly, plasma. Indeed, loss of adipocyte autophagy reduced the EpFA:DiolFA ratio, indicating a lower availability of anti‐inflammatory EpFAs in both mesentery and gonadal adipose tissues during DSS colitis (Fig 5K). This was consistently observed for all analyzed DHA‐derived EpFAs which are important substrates for EPHX1 (Fig 5K) (Edin et al, 2018). Strikingly, these effects appear to be locally restricted to the adipose tissues since no changes in oxylipin levels were observed in the plasma (Fig EV5D). In summary, these data suggest that loss of adipocyte autophagy activates NRF2 and increased the expression of EPHX enzymes promoting a local imbalance of EpFA:DiolFA. We hypothesize that this imbalance, in turn, might alter the local inflammatory response in the adipose tissue to intestinal inflammation. ## Loss of adipocyte autophagy reduces IL‐10 secretion from adipose tissues and systemic IL‐10 levels upon DSS‐induced colitis Since the changes in EpFA:DiolFA appeared to be locally restricted, we next wanted to identify which soluble factors may impact on gut inflammation. Evidence suggests that stimulation of macrophages with EpFA promotes the production of IL‐10, while abundance of DiolFA can quench IL‐10 production (McDougle et al, 2017; Levan et al, 2019). Thus, we next tested whether cytokine production from the adipose tissue may contribute to systemic inflammation during DSS‐induced colitis and assessed the secreted cytokine profile from mesenteric adipose tissues. We found that the mesenteric adipose tissue increases the secretion of several cytokines including the anti‐inflammatory cytokine IL‐10 in response to DSS (Fig 6A). Next, we tested whether IL‐10 secretion from the mesenteric and gonadal adipose tissue was affected by adipocyte autophagy loss. Remarkably, disruption of adipocyte autophagy abolished DSS‐induced IL‐10 secretion from both mesenteric and gonadal adipose tissues (Fig 6B). We found that, upon DSS‐induced colitis, CD11b+ F$\frac{4}{80}$+ adipose tissue macrophages (ATMs) are one of the major cell populations producing IL‐10 in visceral adipose tissues (Fig 6C and D), whereas IL‐10 production in T and B cells were unaltered (Fig 6E). In adipose tissues, ATM frequencies were increased by DSS (Appendix Fig S1A), which resulted in an increased presence of crown‐like structures, however, this was comparable between wild‐type and Atg7 Ad mice. EpFA can alter macrophage polarization and increase tissue‐resident macrophage marker expression such as CD206 (Lopez‐Vicario et al, 2015). In line with reduced EpFA levels in the adipose tissue, Atg7 Ad ATMs had a slightly reduced CD206 expression (Appendix Fig S1B) but remained the predominant type of macrophage in the tissue. In addition, expression of CD36, a lipid scavenging receptor which is commonly found on M2‐type macrophages and induced on ATMs during lipolysis, was increased on the surface of ATMs in wild‐type mice during DSS colitis. However, CD36 expression was not increased in Atg7 Ad mice (Appendix Fig S1C), indicating a distinct adaptation to different lipid availability in the adipose tissue. Figure 6: Reduced adipose tissue‐derived IL‐10 secretion and IL‐10 serum levels in adipocyte autophagy‐deficient mice during DSS‐induced colitis Colitis was induced in mice for 7 days and mesenteric adipose tissue explants were cultured with FBS. Cytokine secretion into the supernatant was measured after 24 h of culture (n = 4–12/group).Colitis was induced in mice for 7 days and adipose tissues were extracted and cultured for 6 h in serum‐starved medium. Secretion of IL‐10 and from mesenteric (left panel) and gonadal adipose tissues (right panel) was measured by ELISA. Shapes identify individual experiments (n = 5–15/group).Representative FACS plots of CD11b+ F4/80+ ATMs in visceral adipose tissue from WT and Atg7 Ad mice upon DSS‐induced colitis on day 7.Quantification of IL‐10‐producing ATMs in visceral adipose tissue from WT and Atg7 Ad upon DSS‐induced colitis on day 7 (n = 3–6/group).Frequencies of IL‐10‐producing immune cells in visceral adipose tissues from WT and Atg7 Ad upon DSS‐induced colitis by flow cytometry (n = 4–6/group).Serum cytokines upon DSS‐induced colitis on day 7 postinduction (n = 17–23/group). Data are represented as mean ± s.e.m. (A) Multiple t‐test. (D–F) Two‐way ANOVA. (B) Two‐way ANOVA with regression for experiment. P < 0.05, P < 0.01, P < 0.001. Source data are available online for this figure. Due to the important role of IL‐10 in immune tolerance, we hypothesized that the reduction of IL‐10 secretion from adipose tissues may translate into a systemic reduction of circulating IL‐10 levels. Indeed, we found that while circulating IL‐10 levels were significantly upregulated in DSS‐treated wild‐type mice compared to noninflamed mice, IL‐10 levels were diminished in Atg7 Ad mice (Fig 6F). Taken together, these data suggest that adipose tissues from adipocyte autophagy‐deficient mice have an impaired production and secretion of anti‐inflammatory IL‐10 in DSS‐induced colitis compared to wild‐type mice. ## Cytochrome P450‐EPHX pathway regulates IL‐10 secretion from autophagy‐deficient adipocytes during intestinal inflammation To establish a more mechanistic link between the increased function of the cytochrome P450‐EPHX pathway and IL‐10 in adipose tissues, we first determined whether EpFA supplementation improves IL‐10 production from macrophages in vitro. Pretreatment of RAW264.7 macrophages with different EpFA increased Il10 transcript levels upon LPS stimulation (Fig 7A), which was further confirmed on protein level in the supernatant (Fig 7B). Cytochrome P450 enzymes are key for the production of EpFA. Inhibition of cytochrome P450 resulted in a marked reduction of IL‐10 secretion from DSS‐induced wild‐type adipose tissues (Fig 7C), suggesting that cytochrome P450 is crucial for adipose tissue‐derived IL‐10 during intestinal inflammation. Lastly, blockade of EPHX1 and EPHX2 in Atg7 Ad adipose tissue explants rescued IL‐10 production (Fig 7D), establishing that EPHX enzyme activity can control IL‐10 in autophagy‐deficient adipose tissues. Collectively, these data indicate that EpFA and enzymes controlling their production and degradation can alter adipose tissue IL‐10 levels during intestinal inflammation and this is dependent on autophagy (Fig 7E). Figure 7: Cytochrome P450‐EPHX pathway regulates IL‐10 secretion from autophagy‐deficient adipose tissues upon DSS‐induced intestinal inflammation Quantification of Il10 transcript levels in RAW264.7 upon stimulation to epoxy fatty acids (n = 3/group).Quantification of IL‐10 protein levels in the supernatant of RAW264.7 upon stimulation to epoxy fatty acids (n = 3/group).Quantification of IL‐10 protein levels in the supernatant of ex vivo cultured adipose tissues from water‐ or DSS‐treated wild‐type mice in the absence or presence of the cytochrome P450 inhibitor 1‐ABT (n = 5–9/group).Quantification of IL‐10 protein levels in the supernatant of ex vivo cultured adipose tissues from DSS‐induced Atg7 Ad mice in the absence or presence of the EPHX1 inhibitor NTPA and EPHX2 inhibitor TPPU (n = 4/group). Dotted line represents IL‐10 secretion from adipose tissues of DSS‐induced wild‐type mice.Graphical summary of the anti‐inflammatory fat‐gut crosstalk during intestinal inflammation. Designed using BioRender. Data are represented as mean ± s.e.m. (A, B) One‐Way ANOVA. (C, D) Paired Student's t‐test. P < 0.05, P < 0.01, P < 0.001, **P < 0.0001. Source data are available online for this figure. ## Discussion Immune cells reside within distinct tissue environments, however, the impact of local metabolic cues on inflammatory processes remains incompletely understood (Richter et al, 2018). Our results indicate that autophagy in mature adipocytes contributes to the balance of intratissual oxylipin levels. Furthermore, we demonstrate that adipocytes autophagy is part of the anti‐inflammatory immune response to intestinal inflammation by promoting the release of IL‐10 from adipose tissues. Autophagy‐dependent secretion from adipose tissues contributes to systemic IL‐10 levels, and limits inflammation at a distant tissue site, the colon. Therefore, our study provides novel insights into a cross‐tissue anti‐inflammatory mechanism, enabling the development of therapeutic approaches to target this crosstalk. While polymorphisms in autophagy genes are well established as genetic risk factors for IBD, little is known about autophagy's role in adipocytes in this disease. Our study highlights that autophagic flux is increased in visceral adipose tissues of wild‐type mice upon DSS‐induced colitis and in the creeping fat tissue of CD patients. Since tamoxifen itself induces autophagy, it is possible that tamoxifen may potentiate some of the effects observed in the DSS‐treated wild‐type and Atg7 Ad mouse model. However, the induction of autophagy was also found in wild‐type mice, which have not received tamoxifen, alongside a transcriptional increase of Atg8 homologs. This increase parallel findings during muscle atrophy, where the expression of Map1lc3b, Gabarapl1, Bnip3, Bnip3l, and Vps34 is regulated via FOXO3 activation, which subsequently controls autophagy levels (Mammucari et al, 2007). It appears plausible that a similar FOXO3‐dependent mechanism occurs in adipocytes, especially since visceral adipocytes showed several transcriptomic and macroscopic changes reminiscent of a “cachexia‐like” phenotype. In line with this, we found that inflammatory cues such as TNFα can promote systemic inflammation which is required for the induction of adipocyte autophagy and cachexia‐like phenotype (Rivera et al, 2019). Early studies found that autophagy is crucial for the normal differentiation of adipose tissues in vivo (Singh et al, 2009; Zhang et al, 2009). However, the significance of autophagy in mature adipocytes remained unexplored until recently. Postdevelopmental ablation of autophagy in mature adipocytes decreased β‐adrenergic receptor‐induced lipolysis (Cai et al, 2018; Son et al, 2020). Conversely, disruption of mTOR by genetic deletion of Raptor increases lipolytic output via autophagy (Zhang et al, 2020). It is likely that adipocyte autophagy controls lipolytic output via the degradation of key proteins involved in the lipolytic machinery such as described for perilipins in fibroblasts and adipocytes (Kaushik & Cuervo, 2016; Ju et al, 2019). In addition, we found that phosphorylation of HSL was reduced in autophagy‐deficient adipocytes, which could indicate a role of autophagy in regulating upstream kinase activity, such as PKA, during intestinal inflammation. Moreover, we discovered that adipocyte autophagy can indeed regulate TNFα‐induced lipolysis and by this may fine‐tune lipolytic output of adipocytes upon inflammatory stress conditions. Adipocyte lipolysis during DSS‐induced colitis is driven predominantly through ATGL. Somewhat surprisingly, loss of adipocyte lipolysis had no impact on body weight loss or colonic inflammation, thus raising the question whether adipocyte lipolysis is beneficial or maladaptive in the context of this disease. These observations are reminiscent of findings during infection‐associated cachexia, where deletion of the cytosolic lipases Atgl and Hsl had no impact on body weight loss (Baazim et al, 2019). In contrast, during cancer‐associated cachexia, loss of these lipases prevents body weight loss suggesting that infection and inflammation models of cachexia act through distinct and yet to be identified biological pathways (Das et al, 2011). Lipid uptake occurs across the intestine with highest levels of lipid absorption in the proximal small intestine. DSS leads to a disruption of the epithelial barrier towards the distal colon. However, some studies report that DSS can alter the morphology of the small intestine such as the jejunum, affecting its function and possibly lower dietary lipid absorption (Yazbeck et al, 2011). This could, therefore, alter uptake of dietary FFA, although FFA absorption in the context of DSS‐induced colitis has not been conclusively determined. The observed decline in serum FFA levels may be connected to a reduced food intake during DSS‐induced colitis (Vidal‐Lletjos et al, 2019), their possible reduced absorption, and the depletion of lipid stores, such as gonadal adipose tissues. The induction of autophagy in the adipose tissue may help to maintain circulating FFA levels in addition to curb inflammation through signaling lipids. Loss of adipocyte autophagy increased NRF2 stability, likely through the sequestration of its regulator KEAP1 (Cai et al, 2018). Here, we demonstrate for the first time that this antioxidant/xenobiotic pathway exacerbates an inflammatory disease. Increased expression of EPHX1 was paralleled by an imbalance in oxylipins leading to decreased levels of EpFA and increased DiolFA. Similar to our findings, EPHX1 was recently found to convert in particularly omega‐3 DHA substrates in adipocytes and liver (Edin et al, 2018; Gautheron et al, 2021). Since our data suggest a broader dysregulation of EpFA:DiolFA, it is likely that EPHX2, which was accumulated on protein level in Atg7 Ad adipose tissues, may also contribute to the conversion of oxylipin substrates. Increasing evidence suggests that macrophages are regulated by oxylipins in their environment. Indeed, increased presence of omega‐3‐derived EpFA achieved either through inhibition of EPHX2 or through supplementation has been shown to promote CD206 expression and IL‐10 secretion (Lopez‐Vicario et al, 2015; McDougle et al, 2017). In line with the reduced presence of EpFA in Atg7‐deficient adipose tissue, we found these two hallmarks of anti‐inflammatory macrophages were equally decreased. Other oxylipin species, produced via the LOX and COX pathway, can also modulate DSS‐induced colitis (Willenberg et al, 2015; Crittenden et al, 2021). However, most genes involved in these pathways were reduced in visceral adipocytes during DSS‐induced colitis. It is plausible, however, that other oxylipins may contribute to the observed phenotype in the Atg7 Ad mouse model. Importantly, this study underscores the importance of adipose tissue‐derived IL‐10 in controlling disease severity. Our findings of increased IL‐10 secretion in visceral adipose tissues upon intestinal inflammation, confirmed findings from the Siegmund laboratory that mesenteric ATMs upregulate expression of IL‐10 during intestinal inflammation in both human and mouse (Batra et al, 2012; Kredel et al, 2013). In line with this, global loss of IL‐10 leads to exacerbation of intestinal inflammation (Li et al, 2014). The disruption in systemic IL‐10 levels may also explain the reduced colonic expansion of FOXP3+ Tregs at resolution, since adequate IL‐10 signaling is required for the expression of FOXP3 in intestinal Tregs (Murai et al, 2009). Recent single‐cell transcriptomic analysis of immune cells resident in human creeping fat tissues revealed an anti‐inflammatory and prorepair role of ATMs, further supporting their beneficial role during intestinal inflammation (Ha et al, 2020). Our data highlight how adipocyte dysfunction can impair this adipocyte‐immune cell crosstalk suggesting that this communication may also exist in human pathology. While ATMs accumulate in in creeping fat tissues of CD patients and in the mesentery of mice upon DSS‐induced colitis (Batra et al, 2012), it remains unclear how these macrophages are regulated during intestinal inflammation. We propose that oxylipins can shift macrophage polarization, in part, through their action as PPAR ligands (Overby et al, 2020), which are important regulators of M2‐type polarization and function (Odegaard et al, 2007, 2008). We found that the PPARγ‐target gene CD36 is upregulated during DSS‐induced colitis on ATMs in wild‐type mice and but not in ATMs from Atg7 Ad mice. Similarly, the expression of CD36 on adipocytes can be controlled by oxylipin levels (Lynes et al, 2017). In addition, it is possible that oxylipin imbalance may also affect other IL‐10‐producing cell types in the adipose tissues such as Tregs which also rely on PPARγ for their accumulation and function (Cipolletta et al, 2012). The resulting reduction in systemic IL‐10 levels prolongs pro‐inflammatory programs at the distal inflammation site. As such, IL‐10 signaling is required for intestinal macrophages to prevent pro‐inflammatory exacerbation during DSS‐induced colitis through inhibition of mTOR signaling which controls macrophage pro‐inflammatory activity (Li et al, 2014; Ip et al, 2017). Overall, this study reveals that metabolically healthy adipose tissues are important regulators to prevent excessive inflammation during colitis. However, the function of adipose tissues in IBD may depend on the overall metabolic and disease state. The expansion of the mesentery during CD may initially be beneficial through prevention of bacterial translocation and signaling pathways poised to promote anti‐inflammatory pathways, as shown here (Batra et al, 2012; Ha et al, 2020). However, sustained inflammation may ultimately subvert the function of the mesentery and lead to adipose tissue fibrosis and intestinal strictures (Mao et al, 2019). Sustained tissue fibrosis results in tissue hypoxia (Zuo et al, 2016) and may impact on tissue oxylipin levels in creeping fat of CD patients. Here, we demonstrate for the first time that adipocyte autophagy contributes to the intratissual balance of oxylipin levels and thus controls the anti‐inflammatory immune response to intestinal tissue injury through regulation of adipose tissue‐derived IL‐10 (as summarized in Fig 7E). It underlines the importance of local adipocyte‐immune cell crosstalk through regulation of lipid mediators. This may present a broader local metabolic regulatory pathway to control immune responses to inflammation and infection. ## Mice Adipoq‐Cre ERT2 mice (Sassmann et al, 2010) were purchased from Charles River, UK (JAX stock number: 025124) and were crossed to Atg7 floxed mice (Komatsu et al, 2005). Experimental cages were sex‐ and age‐matched and balanced for genotypes. Genetic recombination was induced at 8–10 weeks of age by oral gavage of 4 mg tamoxifen per mouse for 5 consecutive days. All experimental procedures were conducted 2 weeks after last tamoxifen administration. DSS‐induced colitis was induced by 1.5–$2\%$ (w/v) DSS (MP Biomedicals, 160110) in drinking water. Mice treated with neutralizing antibodies received 0.5 mg/mouse of either anti‐mouse TNFα (Bio X Cell, BE0058) or IgG1 isotype control (Bio X Cell, BE0088) on day 0 by intraperitoneal injection. Mice were treated with DSS for 5 days and assessed on day 7, a peak inflammation time, or on day 14, a resolution time point. Constitutive Adipoq‐Cre × Pnpla2 floxed mice (Sitnick et al, 2013; Schoiswohl et al, 2015) (JAX stock number: 024278) were kindly provided by Prof. Rudolph Zechner. Wild‐type C57BL/6J mice were purchased from Charles River, UK (JAX stock number: 0000664) or bred in‐house. Mice were housed on a 12 h dark/light cycle and fed ad libitum, under specific pathogen‐free conditions. All animal experimentation was performed in accordance to approved procedures by the Local Review Committee and the Home Office under the project license (PPL$\frac{30}{3388}$ and P01275425). ## Human samples Three individuals with Crohn's disease were recruited for this study. Patients were 30, 31, and 52 years old (all women) and were diagnosed with ileal stenosis at the time of ileocecal resection. At the time of surgery, patients were treated with Prednisolon and/or Azathiporin. Creeping fat was harvested from the ileum, while adjacent noninflamed mesenteric adipose tissues were collected from the caecum. As a further control, mesenteric adipose tissue from the caecum was collected from one colorectal cancer patient (female, 56 years old) undergoing right hemicolectomy. All patients gave informed consent in the framework of the IBDome (SFB‐TRR 241 B01). Tissue collection and ethics were approved by the institutional review board of the Charité‐Universitätsmedizin Berlin (Ethics Approval: EA$\frac{1}{200}$/17) and in line with the principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report. Adipose tissue explants (~50–100 mg) were cultured for DMEM supplemented with $10\%$ FBS (Sigma, F9665) and 100 U/ml Pen‐Strep for 4 h in the absence or presence of lysosomal inhibitors 100 nM Bafilomycin A1 and 20 mM ammonium chloride, then snap frozen until extraction for immunoblotting. ## Histopathology assessment Distal, mid, and proximal colon pieces were fixed in $10\%$ neutral buffered formalin for 24 h before washed and transferred into $70\%$ ethanol. Tissue pieces from each sample were embedded in the same paraffin block and 5 μm sections were subsequently stained with hematoxylin and eosin (H&E). Scoring of histology sections was executed in a blinded fashion according to a previously reported scoring system (Dieleman et al, 1998). In brief, each section was assessed for the degree inflammation, the depth of tissue damage, possible crypt damages, with high scores signifying increased tissue damage. In addition, signs of regeneration (epithelial closure, crypt regeneration) were assessed, with high scores indicating delayed regeneration. Changes were multiplied with a factor classifying the involvement tissue area. Total score was calculated and presented. ## Adipose tissue and colon digestion We collected mesenteric adipose tissue separate from a collective set of visceral adipose tissue depots (including omental, gonadal, and retroperitoneal adipose tissue) to distinguish proximal versus distal effects of intestinal inflammation on adipose tissues. Adipose tissues were collected and digested in DMEM containing $1\%$ fatty acid‐free BSA (Sigma, 126609), $5\%$ HEPES (Gibco, 15630‐056), 0.2 mg/ml Liberase TL (Roche, 5401020001), and 20 μg/ml DNaseI (Roche, 11284932001). Tissues were minced in digestion medium and incubated for 25–30 min at 37°C at 180 rpm. Tissues were further broken down by pipetting using wide bore tips and filtered through a 70 μm mesh. Digestion was quenched by adding medium containing 2 mM EDTA. Adipocyte and stromal vascular fraction were separated by centrifugation (700 g, 10 min) and collected for further downstream analysis. Colon digestions were performed as previously described (Danne et al, 2017). Colons were opened longitudinally and fecal content was removed by washing with PBS. Then colons were washed twice in RPMI containing $5\%$ FBS and 5 mM EDTA at 37°C under agitation. Tissues were minced and digested in RPMI supplemented with $5\%$ FBS, 1 mg/ml collagenase type VIII (Sigma), and 40 μg/ml DNaseI (Roche). Cell suspension was strained through 40 μm mesh and cells were subjected to downstream analysis. ## Flow cytometry Flow cytometry staining was performed as previously described (Riffelmacher et al, 2017). Surface staining was performed by incubating cells with fluorochrome‐conjugated antibodies (Biolegend, BD Bioscience, eBioscience) and LIVE/DEAD Fixable Stains (ThermoFischer) and Fc receptor blockade (ThermoFisher, 14‐0161‐85) for 20 min at 4°C. Cells were fixed with $4\%$ PFA for 10 min at room temperature. For intracellular staining of transcription factors, cells were fixed/permeabilized using the eBioscience™ Foxp3/ Transcription Factor Staining Set (00‐5523‐00, Invitrogen). For cytokine staining, cells were stimulated using Cell Activation cocktail (Biolegend) for 4 h at 37°C in RPMI containing $10\%$ FBS. After surface staining, cells were fixed and stained in Cytofix/CytoPerm (BD Bioscience) following manufacturer protocol. Samples were acquired on LSRII or Fortessa X‐20 flow cytometers (BD Biosciences). Flow cytometry antibodies were purchased from Biolegend or eBioscience or BD Biosciences. The following clones were used for immune phenotyping: CD45 (30‐F11), CD11b (M$\frac{1}{70}$), CD11c (N418), Siglec‐F (S17007L), IA/IE (M$\frac{5}{114.14.2}$), Ly6C (HK1.4), Ly6G (1A8), F$\frac{4}{80}$ (BM8), CD19 (6D5), CD4 (GK1.5), CD8 (53‐6.7), Foxp3 (FJK‐16s), Gata3 (TWAJ), RORgt (Q31‐378), TCRb (H57‐597), IL‐10 (JES5‐16E3), CD206 (C068C2), CD36 (HM36). ## Quantitative PCR Adipocytes and adipose tissue RNA were extracted using TRI reagent (T9424, Sigma). Colon tissue RNA were extracted in RLT buffer containing 1,4‐Dithiothreitol. Tissues were homogenized by lysis in 2 ml tubes containing ceramic beads (KT03961‐1‐003.2, Bertin Instruments) using a Precellys 24 homogenizer (Bertin Instruments). RNA was purified following RNeasy Mini Kit (74104, Qiagen) manufacturer instructions. cDNA was synthesized following the High‐Capacity RNA‐to‐cDNA™ kit protocol (4388950, ThermoFischer). Gene expression was assessed using validated TaqMan probes and run on a ViiA7 real‐time PCR system. All data were collected by comparative Ct method either represented as relative expression (2−ΔCt) or fold change (2−ΔΔCt). Data were normalized to the two most stable housekeeping genes; for adipose tissues Tbp and Rn18s and for colon Actb and Hprt. The following TaqMan probes were used for the quantification: Atg7 (Mm00512209_m1), Map1lc3b (Mm00782868_m1), Gabarap (Mm00490680_m1), Gabarapl1 (Mm00457880_m1), Gabarapl2 (Mm01243684_gH), Il33 (Mm00505403_m1), Tnfa (Mm00443258_m1), Ifng (mm01168134_m1), Cxcl9 (Mm00434946_m1), Il1a (Mm00439620_m1), Ptx3 (Mm00477268_m1), Il10 (Mm00439614_m1), Pnpla2 (Mm00503040_m1), Ephx1 (Mm00468752_m1), Ephx 2 (Mm01313813_m1), Actin (Mm02619580_g1), Hprt (Mm01545399_m1), Rn18s (Mm04277571_s1), Tbp (Mm01277042_m1). ## Bulk RNA sequencing Visceral adipocytes were isolated as floating fraction upon digestion. RNA was extracted and converted to cDNA as described above. PolyA libraries were prepared through end reparation, A‐tailing and adapter ligation. Samples were then size‐selected, multiplexed, and sequenced using a NovaSeq6000. Raw read quality control was performed using pipeline readqc.py (https://github.com/cgat‐developers/cgat‐flow). Resulting reads were aligned to GRCm38/Mm10 reference genome using the pseudoalignment method kallisto (Bray et al, 2016). Differential gene expression analysis was performed using DEseq2 v1.30.1 (Love et al, 2014). Pathway enrichment analysis was performed on differentially expressed genes for “Biological Pathways” using clusterProfiler (v4.0) R package (Wu et al, 2021). DESeq2 median of ratios were used for visualization of expression levels. Heatmaps of selected gene sets were presented as z‐scores using R package pheatmap. Gene enrichment analysis was performed using GSEA software using Hallmark gene sets (Subramanian et al, 2005). Transcriptomic dataset is available on EBI ArrayExpress (Access Code: E‐MTAB‐12498). R code is available under https://github.com/cleete/IBD‐Adipocyte‐Autophagy. ## Lipolysis assays Adipose tissues were collected and washed in PBS before subjected to lipolysis assays. For isoproterenol stimulation, adipose tissues were cut into small tissue pieces and incubated in serum‐free DMEM–High Glucose (Sigma, D5796) with $2\%$ fatty acid‐free BSA (Sigma, 126579) in the absence or presence of 10 μM isoproterenol (Sigma, I6504) for the indicated time. TNFα‐induced lipolysis was induced as previously described (Ju et al, 2019). In brief, small adipose tissue pieces were cultured in DMEM–High Glucose for 24 h in the absence or presence of 100 ng/ml recombinant TNFα (Peprotech, 315‐01A) and then transferred into serum‐free DMEM containing $2\%$ fatty acid‐free BSA for 3 h. Supernatants were collected and FFA concentration normalized to adipose tissue input. ## Adipose tissue explant cultures Gonadal or mesenteric adipose tissue explants were collected from mice at indicated time points. For autophagic flux measurements, explants (~50–100 mg) were cultured for DMEM supplemented with $10\%$ FBS (Sigma, F9665) and 100 U/ml Pen‐Strep for 4 h in the absence or presence of lysosomal inhibitors 100 nM Bafilomycin A1 and 20 mM ammonium chloride. Explants were washed in PBS before collection and then frozen at −80°C until proteins were extracted for immunoblotting. For measurement of cytokine secretion, adipose tissue explants were cultured for 6 h in DMEM/High Modified (D6429, Sigma) with 100 U/ml Pen‐Strep in the absence of FBS. For inhibition of cytochrome P450 ex vivo, medium was supplemented with 1‐ABT (Sigma, A3940) at a final concentration of 1 mM. For inhibition of EPHX1 and EPHX2, medium was supplemented with NTPA (kindly received from Christophe Morriseau and Bruce D. Hammock) and TPPU (Sigma, SML0750) at a final concentration of 100 and 10 μM, respectively. Supernatant was collected, spun down (400 g, 5 min) to remove cell debris and then frozen until further analysis. ## Cell culture RAW264.7 cells (Sigma, 91062702) were cultured in DMEM supplemented with $2\%$ glutamine, $10\%$ FBS, and 100 U/ml Pen‐Strep. Cells were seeded in 24‐well plates (Sarstedt, 83.3922) at 100,000 cells/well. After 4 days, when cells reached confluence, medium was removed, and fresh medium supplemented with corresponding epoxy fatty acids (11,12‐EET Max Spec (Cayman Chemical, 10007262‐100), 17,18‐EpETE (EEQ) Max Spec (Cayman Chemical, 25367‐100), 19,20‐EDP Max‐Spec (Cayman Chemical, 25272‐100)) were added at a final concentration of 1 μM for 3 h, then cells were stimulated with 1 μg/ml LPS (Sigma, L8274) for indicated times. For qPCR measurement, cells were cultured for an additional 3 h after LPS stimulation and 12 h for subsequent ELISA readout. ## Free fatty acid analysis Total supernatant and serum FFA levels were measured using Free Fatty Acid Assay Quantification Kit (ab65341, Abcam). For detailed analysis of FFA species, lipids were extracted by Folch's method (Folch et al, 1957) and subsequently run on a one‐dimensional thin layer chromatography (TLC) using a 10 × 10 cm silica gel G plate in a hexane/diethyl ether/acetic acid (80:20:1, by vol.) solvent system. Separated FFA were used for fatty acid methyl esters (FAMEs) preparation through addition of $2.5\%$ H2SO4 solution in dry methanol/toluene (2:1 (v/v)) at 70°C for 2 h. A known amount of C17:0 was added as an internal standard for quantification. FAMEs were extracted with HPLC grade hexane. A Clarus 500 gas chromatograph with a flame ionizing detector (FID) (Perkin‐Elmer) and fitted with a 30 m × 0.25 mm i.d. capillary column (Elite 225, Perkin Elmer) was used for separation and analysis of FAs. The oven temperature was programmed as follows: 170°C for 3 min, increased to 220°C at 4°C/min, and then held at 220°C for 15 min. FAMEs were identified routinely by comparing retention times of peaks with those of G411 FA standards (Nu‐Chek Prep Inc). TotalChrom software (Perkin‐Elmer) was used for data acquisition and quantification. ## Oxylipin analysis Oxylipins were analyzed by means of liquid chromatography mass spectrometry (Rund et al, 2018; Kutzner et al, 2019). The plasma samples were analyzed following protein precipitation and solid‐phase extraction on reversed phase/anion exchange cartridges (Rund et al, 2018; Kutzner et al, 2019). The adipose tissue was homogenized in a ball mill and oxylipins and nonesterified fatty acids were extracted with a mixture of chloroform and iso‐propanol following solid‐phase extraction on an amino propyl SPE cartridge (Koch et al, 2021, 2022). Oxylipin concentrations in adipose tissue and plasma as well as LA, DHA, and ARA in the tissue were determined by external calibration with internal standards (Rund et al, 2018; Kutzner et al, 2019). ## Immunoblotting Autophagic flux in adipose tissues was measured by incubating adipose tissue explants from experimental animals in RPMI in the absence or presence of lysosomal inhibitors 100 nM Bafilomycin A1 and 20 mM NH4Cl for 4 h. DMSO was used as ‘vehicle’ control. Adipose tissues were collected and snap frozen. Protein extraction was performed as previously described (An & Scherer, 2020). In brief, 500 μl of lysis buffer containing protease inhibitors (04693159001, Roche) and phosphoStop (04906837001, Roche) were added per 100 mg of tissue. Cells were lysed using Qiagen TissueLyser II. Tissues were incubated on ice for 1 h and lipid contamination was removed via serial centrifugation and transfer of internatant into fresh tubes. Protein concentration was determined by BCA Protein Assay Kit (23227, Thermo Scientific). A total of 15–30 μg protein was separated on a 4–$12\%$ Bis‐Tris SDS–PAGE and transferred using BioRad Turbo Blot (BioRad, 1704156) or wet transfer onto PVDF membrane. Human tissues were separated on a 4–$20\%$ Tris–Glycine gel (BioRad, 4561093DC) and transferred onto PVDF membranes. Membranes were blocked in TBST containing $5\%$ BSA. Primary antibodies were used at indicated concentration overnight: LC3 (L8918, Sigma) (1:1,500); β‐ACTIN (8H10D10, Cell Signaling) (1:5,000); EPHX1 (Santa Cruz, sc‐135984) (1:500); EPHX2 (Santa Cruz, sc‐166961) (1:500); NRF2 (GeneTex, GTX103322) (1:1,000); pHSL‐Ser600 (Cell Signaling, 45804) (1:2,000); HSL (Cell Signaling, 18381) (1:2,000); ATGL (Cell Signaling, 2439) (1:1,500). Membranes were visualized using IRDye secondary antibodies (1:10,000) (LICOR). Band quantification of Western blots was performed on ImageStudio software (LICOR). Autophagic flux was calculated using LC3‐II normalized values to β‐ACTIN loading control: (LC3‐II (Inh) – LC3‐II (Veh))/(LC3‐II (Veh)), as previously described (Zhang et al, 2019). ## Transmission electron microscopy Mice were sacrificed by increasing concentrations of CO2. Adipose tissues were excised, cut into small 1–2 mm pieces and immediately fixed in prewarmed (37°C) primary fixative containing $2.5\%$ glutaraldehyde and $4\%$ formaldehyde in 0.1 M sodium cacodylate buffer, pH 7.2 for 2 h at room temperature and then stored in the fixative at 4°C until further processing. Samples were then washed for 2× 45 min in 0.1 M sodium cacodylate buffer (pH 7.2) at room temperature with rotation, transferred to carrier baskets and processed for EM using a Leica AMW automated microwave processing unit. Briefly, this included three washes with 0.1 M sodium cacodylate buffer, pH 7.2, one wash with 50 mM glycine in 0.1 M sodium cacodylate buffer to quench free aldehydes, secondary fixation with $1\%$ osmium tetroxide +$1.5\%$ potassium ferricyanide in 0.1 M sodium cacodylate buffer, six water washes, tertiary fixation with $2\%$ uranyl acetate, two water washes, then dehydration with ethanol from 30, 50, 70, 90, 95 to $100\%$ (repeated twice). All of these steps were performed at 37°C and 15–20 W for 1–2 min each, with the exception of the osmium and uranyl acetate steps, which were for 12 and 9 min, respectively. Samples were infiltrated with TAAB Hard Plus epoxy resin to $100\%$ resin in the AMW and then processed manually at room temperature for the remaining steps. Samples were transferred to 2 ml tubes filled with fresh resin, centrifuged for ~2 min at 2,000 g (to help improve resin infiltration), then incubated at room temperature overnight with rotation. The following day, the resin was removed and replaced with fresh resin, then the samples were centrifuged as above and incubated at room temperature with rotation for ~3 h. This step was repeated and then tissue pieces were transferred to individual Beem capsules filled with fresh resin and polymerized for 48 h at 60°C. Once polymerized, blocks were sectioned using a Diatome diamond knife on a Leica UC7 Ultramicrotome. Ultrathin (90 nm) sections were transferred onto 200 mesh copper grids and then post‐stained with lead citrate for 5 min, washed and air dried. Grids were imaged with a Thermo Fisher Tecnai 12 TEM (operated at 120 kV) using a Gatan OneView camera. ## Extracellular cytokine measurements Serum samples were collected by cardiac puncture and collected in Microtainer tubes (365978, BD Bioscience). Samples were centrifuged for 90 s at 15,000 g and serum aliquots were snap‐frozen until further analysis. Global inflammatory cytokine analysis of supernatants of adipose tissue explant cultures and serum were performed using LEGENDPlex™ Mouse Inflammation Panel (740446, Biolegend). Supernatant IL‐10 levels were measured by IL‐10 Mouse Uncoated ELISA Kit (88‐7105‐86, Invitrogen). Adipose tissue‐derived cytokine levels were normalized to input tissue weight. ## Statistical analysis Data were tested for eventual statistical outliers (ROUT analysis with a $Q = 1$%) and outliers were removed if detected. Next, data were tested normality before applying parametric or nonparametric testing. For two normally distributed groups, unpaired Student's tests were applied. Comparisons across more than two experimental groups were performed using one‐way or two‐way ANOVA with Šídák multiple testing correction. Data were considered statistically significant when $P \leq 0.05$ ($P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, $P \leq 0.0001$). Typically, data were pooled from or representative of at least two experiments, if not otherwise indicated, and presented as mean. Data were visualized and statistics calculated in either GraphPad Prism 9 or R software. ## Author contributions Felix Clemens Richter: Conceptualization; formal analysis; funding acquisition; investigation; visualization; methodology; writing – original draft; writing – review and editing. Matthias Friedrich: Conceptualization; formal analysis; supervision; investigation; methodology; writing – review and editing. Nadja Kampschulte: Formal analysis; investigation; methodology; writing – review and editing. Klara Piletic: Investigation; writing – review and editing. Ghada Alsaleh: Conceptualization; investigation; writing – review and editing. Ramona Zummach: Investigation. Julia Hecker: Resources; investigation. Mathilde Pohin: Conceptualization; investigation; writing – review and editing. Nicholas Ilott: Data curation; formal analysis. Irina Guschina: Investigation; methodology. Sarah Karin Wideman: Investigation; writing – review and editing. Errin Johnson: Investigation; methodology; writing – review and editing. Mariana Borsa: Investigation; writing – review and editing. Paula Hahn: Investigation; writing – review and editing. Christophe Morriseau: Resources; methodology. Bruce D Hammock: Conceptualization; resources. Henk Simon Schipper: Conceptualization; supervision; writing – review and editing. Claire M Edwards: Conceptualization; supervision; writing – review and editing. Rudolf Zechner: Resources; methodology. Britta Siegmund: Conceptualization; resources. Carl Weidinger: Conceptualization; resources; investigation. Nils Helge Schebb: Conceptualization; formal analysis; supervision; investigation; methodology; writing – review and editing. Fiona Powrie: Conceptualization; supervision; funding acquisition; writing – review and editing. Anna Katharina Simon: Conceptualization; supervision; funding acquisition; methodology; writing – original draft; writing – review and editing. ## Disclosure and competing interests statement BDH is founder of EicOsis Human Health developing sEH inhibitors as human pharmaceuticals. FP received research support or consultancy fees from Roche, Janssen, GSK, Novartis and Genentech. AKS received consultancy fees from Calico, Oxford Healthspan, The Longevity Lab. ## Data availability All reagents used in this study are commercially available. Source data underlying the graphs as well as representative western blots have been made available in the Source Data file. Transcriptomic dataset is available on EBI ArrayExpress (Access Code: E‐MTAB‐12498). In addition, RNAseq analysis R scripts can be found on https://github.com/cleete/IBD‐Adipocyte‐Autophagy. ## References 1. An YA, Scherer PE. Mouse adipose tissue protein extraction*. Bio Protoc* (2020) 10 2. Baazim H, Schweiger M, Moschinger M, Xu H, Scherer T, Popa A, Gallage S, Ali A, Khamina K, Kosack L. CD8. Nat Immunol (2019) 20 701-710. PMID: 31110314 3. Baazim H, Antonio‐Herrera L, Bergthaler A. The interplay of immunology and cachexia in infection and cancer. Nat Rev Immunol (2021) 22 309-321. PMID: 34608281 4. Batra A, Heimesaat MM, Bereswill S, Fischer A, Glauben R, Kunkel D, Scheffold A, Erben U, Kuhl A, Loddenkemper C. Mesenteric fat ‐ control site for bacterial translocation in colitis?. Mucosal Immunol (2012) 5 580-591. PMID: 22569302 5. Bray NL, Pimentel H, Melsted P, Pachter L. Near‐optimal probabilistic RNA‐seq quantification. Nat Biotechnol (2016) 34 525-527. PMID: 27043002 6. Cadwell K, Liu JY, Brown SL, Miyoshi H, Loh J, Lennerz JK, Kishi C, Kc W, Carrero JA, Hunt S. A key role for autophagy and the autophagy gene Atg16l1 in mouse and human intestinal Paneth cells. Nature (2008) 456 259-263. PMID: 18849966 7. Cadwell K, Patel KK, Komatsu M, Virgin HWT, Stappenbeck TS. A common role for Atg16L1, Atg5 and Atg7 in small intestinal Paneth cells and Crohn disease. Autophagy (2009) 5 250-252. PMID: 19139628 8. Cai J, Pires KM, Ferhat M, Chaurasia B, Buffolo MA, Smalling R, Sargsyan A, Atkinson DL, Summers SA, Graham TE. Autophagy ablation in adipocytes induces insulin resistance and reveals roles for lipid peroxide and Nrf2 signaling in adipose‐liver crosstalk. Cell Rep (2018) 25 1708-1717. PMID: 30428342 9. Cawthorn WP, Sethi JK. TNF‐alpha and adipocyte biology. FEBS Lett (2008) 582 117-131. PMID: 18037376 10. Cipolletta D, Feuerer M, Li A, Kamei N, Lee J, Shoelson SE, Benoist C, Mathis D. PPAR‐gamma is a major driver of the accumulation and phenotype of adipose tissue Treg cells. Nature (2012) 486 549-553. PMID: 22722857 11. Clarke AJ, Simon AK. Autophagy in the renewal, differentiation and homeostasis of immune cells. Nat Rev Immunol (2019) 19 170-183. PMID: 30531943 12. Crittenden S, Goepp M, Pollock J, Robb CT, Smyth DJ, Zhou Y, Andrews R, Tyrrell V, Gkikas K, Adima A. Prostaglandin E. Sci Adv (2021) 7. PMID: 33579710 13. Danne C, Ryzhakov G, Martinez‐Lopez M, Ilott NE, Franchini F, Cuskin F, Lowe EC, Bullers SJ, Arthur JSC, Powrie F. A large polysaccharide produced by helicobacter hepaticus induces an anti‐inflammatory gene signature in macrophages. Cell Host Microbe (2017) 22 733-745. PMID: 29241040 14. Das SK, Eder S, Schauer S, Diwoky C, Temmel H, Guertl B, Gorkiewicz G, Tamilarasan KP, Kumari P, Trauner M. Adipose triglyceride lipase contributes to cancer‐associated cachexia. Science (2011) 333 233-238. PMID: 21680814 15. Deretic V. Autophagy in inflammation, infection, and immunometabolism. Immunity (2021) 54 437-453. PMID: 33691134 16. Dieleman LA, Palmen MJ, Akol H, Bloemena E, Pena AS, Meuwissen SG, Van Rees EP. Chronic experimental colitis induced by dextran sulphate sodium (DSS) is characterized by Th1 and Th2 cytokines. Clin Exp Immunol (1998) 114 385-391. PMID: 9844047 17. Edin ML, Hamedani BG, Gruzdev A, Graves JP, Lih FB, Arbes SJ, Singh R, Orjuela Leon AC, Bradbury JA, DeGraff LM. Epoxide hydrolase 1 (EPHX1) hydrolyzes epoxyeicosanoids and impairs cardiac recovery after ischemia. J Biol Chem (2018) 293 3281-3292. PMID: 29298899 18. Folch J, Lees M, Sloane Stanley GH. A simple method for the isolation and purification of total lipides from animal tissues. J Biol Chem (1957) 226 497-509. PMID: 13428781 19. Friedrich M, Pohin M, Powrie F. Cytokine networks in the pathophysiology of inflammatory bowel disease. Immunity (2019) 50 992-1006. PMID: 30995511 20. Gautheron J, Morisseau C, Chung WK, Zammouri J, Auclair M, Baujat G, Capel E, Moulin C, Wang Y, Yang J. EPHX1 mutations cause a lipoatrophic diabetes syndrome due to impaired epoxide hydrolysis and increased cellular senescence. Elife (2021) 10. PMID: 34342583 21. Gilroy DW, Edin ML, De Maeyer RP, Bystrom J, Newson J, Lih FB, Stables M, Zeldin DC, Bishop‐Bailey D. CYP450‐derived oxylipins mediate inflammatory resolution. Proc Natl Acad Sci USA (2016) 113 E3240-E3249. PMID: 27226306 22. Ha CWY, Martin A, Sepich‐Poore GD, Shi B, Wang Y, Gouin K, Humphrey G, Sanders K, Ratnayake Y, Chan KSL. Translocation of viable gut microbiota to mesenteric adipose drives formation of creeping fat in humans. Cell (2020) 183 23. Hampe J, Franke A, Rosenstiel P, Till A, Teuber M, Huse K, Albrecht M, Mayr G, De La Vega FM, Briggs J. A genome‐wide association scan of nonsynonymous SNPs identifies a susceptibility variant for Crohn disease in ATG16L1. Nat Genet (2007) 39 207-211. PMID: 17200669 24. Huang SC, Everts B, Ivanova Y, O'Sullivan D, Nascimento M, Smith AM, Beatty W, Love‐Gregory L, Lam WY, O'Neill CM. Cell‐intrinsic lysosomal lipolysis is essential for alternative activation of macrophages. Nat Immunol (2014) 15 846-855. PMID: 25086775 25. Imig JD, Hammock BD. Soluble epoxide hydrolase as a therapeutic target for cardiovascular diseases. Nat Rev Drug Discov (2009) 8 794-805. PMID: 19794443 26. Ip WKE, Hoshi N, Shouval DS, Snapper S, Medzhitov R. Anti‐inflammatory effect of IL‐10 mediated by metabolic reprogramming of macrophages. Science (2017) 356 513-519. PMID: 28473584 27. Jostins L, Ripke S, Weersma RK, Duerr RH, McGovern DP, Hui KY, Lee JC, Schumm LP, Sharma Y, Anderson CA. Host‐microbe interactions have shaped the genetic architecture of inflammatory bowel disease. Nature (2012) 491 119-124. PMID: 23128233 28. Ju L, Han J, Zhang X, Deng Y, Yan H, Wang C, Li X, Chen S, Alimujiang M, Li X. Obesity‐associated inflammation triggers an autophagy‐lysosomal response in adipocytes and causes degradation of perilipin 1. Cell Death Dis (2019) 10 121. PMID: 30741926 29. Kabat AM, Harrison OJ, Riffelmacher T, Moghaddam AE, Pearson CF, Laing A, Abeler‐Dorner L, Forman SP, Grencis RK, Sattentau Q. The autophagy gene Atg16l1 differentially regulates Treg and TH2 cells to control intestinal inflammation. Elife (2016) 5. PMID: 26910010 30. Kaushik S, Cuervo AM. AMPK‐dependent phosphorylation of lipid droplet protein PLIN2 triggers its degradation by CMA. Autophagy (2016) 12 432-438. PMID: 26902588 31. Klein‐Wieringa IR, Andersen SN, Kwekkeboom JC, Giera M, de Lange‐Brokaar BJ, van Osch GJ, Zuurmond AM, Stojanovic‐Susulic V, Nelissen RG, Pijl H. Adipocytes modulate the phenotype of human macrophages through secreted lipids. J Immunol (2013) 191 1356-1363. PMID: 23817431 32. Klionsky DJ, Petroni G, Amaravadi RK, Baehrecke EH, Ballabio A, Boya P, Bravo‐San Pedro JM, Cadwell K, Cecconi F, Choi AMK. Autophagy in major human diseases. EMBO J (2021) 40. PMID: 34459017 33. Koch E, Wiebel M, Hopmann C, Kampschulte N, Schebb NH. Rapid quantification of fatty acids in plant oils and biological samples by LC‐MS. Anal Bioanal Chem (2021) 413 5439-5451. PMID: 34296318 34. Koch E, Kampschulte N, Schebb NH. Comprehensive analysis of fatty acid and oxylipin patterns in n3‐PUFA supplements. J Agric Food Chem (2022) 70 3979-3988. PMID: 35324176 35. Komatsu M, Waguri S, Ueno T, Iwata J, Murata S, Tanida I, Ezaki J, Mizushima N, Ohsumi Y, Uchiyama Y. Impairment of starvation‐induced and constitutive autophagy in Atg7‐deficient mice. J Cell Biol (2005) 169 425-434. PMID: 15866887 36. Kredel LI, Batra A, Stroh T, Kuhl AA, Zeitz M, Erben U, Siegmund B. Adipokines from local fat cells shape the macrophage compartment of the creeping fat in Crohn's disease. Gut (2013) 62 852-862. PMID: 22543156 37. Kutzner L, Rund KM, Ostermann AI, Hartung NM, Galano JM, Balas L, Durand T, Balzer MS, David S, Schebb NH. Development of an optimized LC‐MS method for the detection of specialized pro‐resolving mediators in biological samples. Front Pharmacol (2019) 10 169. PMID: 30899221 38. Levan SR, Stamnes KA, Lin DL, Panzer AR, Fukui E, McCauley K, Fujimura KE, McKean M, Ownby DR, Zoratti EM. Elevated faecal 12,13‐diHOME concentration in neonates at high risk for asthma is produced by gut bacteria and impedes immune tolerance. Nat Microbiol (2019) 4 1851-1861. PMID: 31332384 39. Li B, Alli R, Vogel P, Geiger TL. IL‐10 modulates DSS‐induced colitis through a macrophage‐ROS‐NO axis. Mucosal Immunol (2014) 7 869-878. PMID: 24301657 40. Lopez‐Vicario C, Alcaraz‐Quiles J, Garcia‐Alonso V, Rius B, Hwang SH, Titos E, Lopategi A, Hammock BD, Arroyo V, Claria J. Inhibition of soluble epoxide hydrolase modulates inflammation and autophagy in obese adipose tissue and liver: role for omega‐3 epoxides. Proc Natl Acad Sci USA (2015) 112 536-541. PMID: 25550510 41. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA‐seq data with DESeq2. Genome Biol (2014) 15 550. PMID: 25516281 42. Lynes MD, Leiria LO, Lundh M, Bartelt A, Shamsi F, Huang TL, Takahashi H, Hirshman MF, Schlein C, Lee A. The cold‐induced lipokine 12,13‐diHOME promotes fatty acid transport into brown adipose tissue. Nat Med (2017) 23 631-637. PMID: 28346411 43. Mammucari C, Milan G, Romanello V, Masiero E, Rudolf R, Del Piccolo P, Burden SJ, Di Lisi R, Sandri C, Zhao J. FoxO3 controls autophagy in skeletal muscle. Cell Metab (2007) 6 458-471. PMID: 18054315 44. Mao R, Kurada S, Gordon IO, Baker ME, Gandhi N, McDonald C, Coffey JC, Rieder F. The mesenteric fat and intestinal muscle interface: creeping fat influencing stricture formation in Crohn's disease. Inflamm Bowel Dis (2019) 25 421-426. PMID: 30346528 45. McCarroll SA, Huett A, Kuballa P, Chilewski SD, Landry A, Goyette P, Zody MC, Hall JL, Brant SR, Cho JH. Deletion polymorphism upstream of IRGM associated with altered IRGM expression and Crohn's disease. Nat Genet (2008) 40 1107-1112. PMID: 19165925 46. McDougle DR, Watson JE, Abdeen AA, Adili R, Caputo MP, Krapf JE, Johnson RW, Kilian KA, Holinstat M, Das A. Anti‐inflammatory omega‐3 endocannabinoid epoxides. Proc Natl Acad Sci USA (2017) 114 E6034-E6043. PMID: 28687674 47. McReynolds C, Morisseau C, Wagner K, Hammock B. Epoxy fatty acids are promising targets for treatment of pain, cardiovascular disease and other indications characterized by mitochondrial dysfunction, endoplasmic stress and inflammation. Adv Exp Med Biol (2020) 1274 71-99. PMID: 32894508 48. Murai M, Turovskaya O, Kim G, Madan R, Karp CL, Cheroutre H, Kronenberg M. Interleukin 10 acts on regulatory T cells to maintain expression of the transcription factor Foxp3 and suppressive function in mice with colitis. Nat Immunol (2009) 10 1178-1184. PMID: 19783988 49. Odegaard JI, Ricardo‐Gonzalez RR, Goforth MH, Morel CR, Subramanian V, Mukundan L, Red Eagle A, Vats D, Brombacher F, Ferrante AW. Macrophage‐specific PPARgamma controls alternative activation and improves insulin resistance. Nature (2007) 447 1116-1120. PMID: 17515919 50. Odegaard JI, Ricardo‐Gonzalez RR, Red Eagle A, Vats D, Morel CR, Goforth MH, Subramanian V, Mukundan L, Ferrante AW, Chawla A. Alternative M2 activation of Kupffer cells by PPARdelta ameliorates obesity‐induced insulin resistance. Cell Metab (2008) 7 496-507. PMID: 18522831 51. Oliva M, Munoz‐Aguirre M, Kim‐Hellmuth S, Wucher V, Gewirtz ADH, Cotter DJ, Parsana P, Kasela S, Balliu B, Vinuela A. The impact of sex on gene expression across human tissues. Science (2020) 369. PMID: 32913072 52. Overby H, Yang Y, Xu X, Graham K, Hildreth K, Choi S, Wan D, Morisseau C, Zeldin DC, Hammock BD. Soluble epoxide hydrolase inhibition by t‐TUCB promotes Brown adipogenesis and reduces serum triglycerides in diet‐induced obesity. Int J Mol Sci (2020) 21. PMID: 32987880 53. Richter FC, Obba S, Simon AK. Local exchange of metabolites shapes immunity. Immunology (2018) 155 309-319. PMID: 29972686 54. Riffelmacher T, Clarke A, Richter FC, Stranks A, Pandey S, Danielli S, Hublitz P, Yu Z, Johnson E, Schwerd T. Autophagy‐dependent generation of free fatty acids is critical for normal neutrophil differentiation. Immunity (2017) 47 466-480. PMID: 28916263 55. Rivera ED, Coffey JC, Walsh D, Ehrenpreis ED. The mesentery, systemic inflammation, and Crohn's disease. Inflamm Bowel Dis (2019) 25 226-234. PMID: 29920595 56. Rund KM, Ostermann AI, Kutzner L, Galano JM, Oger C, Vigor C, Wecklein S, Seiwert N, Durand T, Schebb NH. Development of an LC‐ESI(−)‐MS/MS method for the simultaneous quantification of 35 isoprostanes and isofurans derived from the major n3‐ and n6‐PUFAs. Anal Chim Acta (2018) 1037 63-74. PMID: 30292316 57. Russo L, Lumeng CN. Properties and functions of adipose tissue macrophages in obesity. Immunology (2018) 155 407-417. PMID: 30229891 58. Sassmann A, Offermanns S, Wettschureck N. Tamoxifen‐inducible Cre‐mediated recombination in adipocytes. Genesis (2010) 48 618-625. PMID: 20715175 59. Schoiswohl G, Stefanovic‐Racic M, Menke MN, Wills RC, Surlow BA, Basantani MK, Sitnick MT, Cai L, Yazbeck CF, Stolz DB. Impact of reduced ATGL‐mediated adipocyte lipolysis on obesity‐associated insulin resistance and inflammation in male mice. Endocrinology (2015) 156 3610-3624. PMID: 26196542 60. Schweiger M, Schreiber R, Haemmerle G, Lass A, Fledelius C, Jacobsen P, Tornqvist H, Zechner R, Zimmermann R. Adipose triglyceride lipase and hormone‐sensitive lipase are the major enzymes in adipose tissue triacylglycerol catabolism. J Biol Chem (2006) 281 40236-40241. PMID: 17074755 61. Sheehan AL, Warren BF, Gear MW, Shepherd NA. Fat‐wrapping in Crohn's disease: pathological basis and relevance to surgical practice. Br J Surg (1992) 79 955-958. PMID: 1422768 62. Siegmund B, Lehr HA, Fantuzzi G. Leptin: a pivotal mediator of intestinal inflammation in mice. Gastroenterology (2002) 122 2011-2025. PMID: 12055606 63. Singh R, Xiang Y, Wang Y, Baikati K, Cuervo AM, Luu YK, Tang Y, Pessin JE, Schwartz GJ, Czaja MJ. Autophagy regulates adipose mass and differentiation in mice. J Clin Invest (2009) 119 3329-3339. PMID: 19855132 64. Sitnick MT, Basantani MK, Cai L, Schoiswohl G, Yazbeck CF, Distefano G, Ritov V, DeLany JP, Schreiber R, Stolz DB. Skeletal muscle triacylglycerol hydrolysis does not influence metabolic complications of obesity. Diabetes (2013) 62 3350-3361. PMID: 23835334 65. Son Y, Cho YK, Saha A, Kwon HJ, Park JH, Kim M, Jung YS, Kim SN, Choi C, Seong JK. Adipocyte‐specific Beclin1 deletion impairs lipolysis and mitochondrial integrity in adipose tissue. Mol Metab (2020) 39. PMID: 32344065 66. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES. Gene set enrichment analysis: a knowledge‐based approach for interpreting genome‐wide expression profiles. Proc Natl Acad Sci USA (2005) 102 15545-15550. PMID: 16199517 67. Trim WV, Lynch L. Immune and non‐immune functions of adipose tissue leukocytes. Nat Rev Immunol (2021) 22 371-386. PMID: 34741167 68. Vidal‐Lletjos S, Andriamihaja M, Blais A, Grauso M, Lepage P, Davila AM, Gaudichon C, Leclerc M, Blachier F, Lan A. Mucosal healing progression after acute colitis in mice. World J Gastroenterol (2019) 25 3572-3589. PMID: 31367158 69. Weidinger C, Ziegler JF, Letizia M, Schmidt F, Siegmund B. Adipokines and their role in intestinal inflammation. Front Immunol (2018) 9 1974. PMID: 30369924 70. Whibley N, Tucci A, Powrie F. Regulatory T cell adaptation in the intestine and skin. Nat Immunol (2019) 20 386-396. PMID: 30890797 71. Willenberg I, Ostermann AI, Giovannini S, Kershaw O, von Keutz A, Steinberg P, Schebb NH. Effect of acute and chronic DSS induced colitis on plasma eicosanoid and oxylipin levels in the rat. Prostaglandins Other Lipid Mediat (2015) 120 155-160. PMID: 25908302 72. Wu T, Hu E, Xu S, Chen M, Guo P, Dai Z, Feng T, Zhou L, Tang W, Zhan L. clusterProfiler 4.0: a universal enrichment tool for interpreting omics data. Innovation (2021) 2. PMID: 34557778 73. Yazbeck R, Howarth GS, Butler RN, Geier MS, Abbott CA. Biochemical and histological changes in the small intestine of mice with dextran sulfate sodium colitis. J Cell Physiol (2011) 226 3219-3224. PMID: 21351101 74. Zhang Y, Goldman S, Baerga R, Zhao Y, Komatsu M, Jin S. Adipose‐specific deletion of autophagy‐related gene 7 (atg7) in mice reveals a role in adipogenesis. Proc Natl Acad Sci USA (2009) 106 19860-19865. PMID: 19910529 75. Zhang H, Alsaleh G, Feltham J, Sun Y, Napolitano G, Riffelmacher T, Charles P, Frau L, Hublitz P, Yu Z. Polyamines control eIF5A Hypusination, TFEB translation, and autophagy to reverse B cell senescence. Mol Cell (2019) 76 110-125. PMID: 31474573 76. Zhang X, Wu D, Wang C, Luo Y, Ding X, Yang X, Silva F, Arenas S, Weaver JM, Mandell M. Sustained activation of autophagy suppresses adipocyte maturation via a lipolysis‐dependent mechanism. Autophagy (2020) 16 1668-1682. PMID: 31840569 77. Zuo L, Li Y, Zhu W, Shen B, Gong J, Guo Z, Zhang W, Wu R, Gu L, Li N. Mesenteric adipocyte dysfunction in Crohn's disease is associated with hypoxia. Inflamm Bowel Dis (2016) 22 114-126. PMID: 26332309
--- title: A fluorescent probe to simultaneously detect both O-GlcNAcase and phosphatase authors: - Jihyeon Boo - Jongwon Lee - Young-Hyun Kim - Chang-Hee Lee - Bonsu Ku - Injae Shin journal: Frontiers in Chemistry year: 2023 pmcid: PMC10015443 doi: 10.3389/fchem.2023.1133018 license: CC BY 4.0 --- # A fluorescent probe to simultaneously detect both O-GlcNAcase and phosphatase ## Abstract O-GlcNAc modification of proteins often has crosstalk with protein phosphorylation. These posttranslational modifications are highly dynamic events that modulate a wide range of cellular processes. Owing to the physiological and pathological significance of protein O-GlcNAcylation and phosphorylation, we designed the fluorescent probe, βGlcNAc-CM-Rhod-P, to differentially detect activities of O-GlcNAcase (OGA) and phosphatase, enzymes that are responsible for these modifications. βGlcNAc-CM-Rhod-P was comprised of a βGlcNAc-conjugated coumarin (βGlcNAc-CM) acting as an OGA substrate, a phosphorylated rhodol (Rhod-P) as a phosphatase substrate and a piperazine bridge. Because the emission wavelength maxima of CM and Rhod liberated from the probe are greatly different (100 nm), spectral interference is avoided. The results of this study revealed that treatment of βGlcNAc-CM-Rhod-P with OGA promotes formation of the GlcNAc-cleaved probe, CM-Rhod-P, and a consequent increase in the intensity of fluorescence associated with free CM. Also, it was found that exposure of the probe to phosphatase produces a dephosphorylated probe, βGlcNAc-CM-Rhod, which displays strong fluorescence arising from free Rhod. On the other hand, when incubated with both OGA and phosphatase, βGlcNAc-CM-Rhod-P was converted to CM-Rhod which lacked both βGlcNAc and phosphoryl groups, in conjunction with increases in the intensities of fluorescence arising from both free CM and Rhod. This probe was employed to detect activities of OGA and phosphatase in cell lysates and to fluorescently image both enzymes in cells. Collectively, the findings indicate that βGlcNAc-CM-Rhod-P can be utilized as a chemical tool to simultaneously determine activities of OGA and phosphatase. ## Introduction Based on the results of the human genome project (Yang et al., 2016; Aebersold et al., 2018), the number of human proteins encoded by genes is estimated to be around 20,000. This number is increased dramatically by posttranslational modifications with small groups (e.g., Phosphoryl, acetyl or methyl groups) or large biomolecules (e.g., Glycans or Ubiquitin), which give rise to several hundreds of thousands of protein variants (Walsh et al., 2005; Aebersold et al., 2018). Because posttranslational modifications of proteins regulate their activities, structures, interactions, and locations, they play an important role in controlling a broad spectrum of cellular processes (Conibear, 2020). Among these modifications, O-GlcNAc modification of proteins frequently takes place in higher eukaryotes (Hart et al., 2007; Wulff-Fuentes et al., 2021). This glycan modification is a unique type of protein glycosylation in that only a single carbohydrate, N-acetylglucosamine (GlcNAc), is attached to side chains of serine (Ser) or threonine (Thr) residues through the O-linkage. O-GlcNAcylation occurs in cytosolic, nuclear and mitochondrial proteins, and has been suggested to act as a nutrient and stress sensor that modulates various cellular events, including transcription, translation, translocation, cell signaling, and metabolism (Hart et al., 2007; Yang and Qian, 2017). Previous studies have shown that abnormal regulation of protein O-GlcNAc modification is involved in the pathogenesis of diverse human diseases (Slawson and Hart, 2011; Zhu and Hart, 2021). A representative example of diseases caused by dysregulated O-GlcNAcylation is cancer (Ferrer et al., 2016). In addition, aberrant protein O-GlcNAcylation is also associated with type 1 and type 2 diabetes (Ma and Hart, 2013). Furthermore, altered protein O-GlcNAc modification in brain is closely related to the onset of neurodegenerative diseases including Alzheimer’s, Huntington’s and Parkinson’s disorders (Yuzwa, et al., 2012; Lee et al., 2021). Dynamic cycling of protein O-GlcNAcylation is modulated in a nutrient- and stress-responsive manner by the cooperative action of O-GlcNAcase (OGA) and O-GlcNAc transferase (OGT). While OGA promotes removal of the ß-O-GlcNAc moiety from Ser and Thr side chains of cytoplasmic and nuclear proteins, OGT catalyzes the attachment of the GlcNAc monosaccharide to these residues (Dong and Hart, 1994; Gao et al., 2001; Banerjee et al., 2013; Saha et al., 2021). Intriguingly, O-GlcNAc modification of proteins engages in extensive crosstalk with phosphorylation, which is a modification in which a phosphoryl group becomes bonded mainly to the side chains of Ser, Thr and tyrosine (Tyr) of proteins (Hart et al., 2011; van der Laarse et al., 2018). Specifically, O-GlcNAcylation/phosphorylation crosstalk takes place competitively at the same residue within proteins (termed reciprocal crosstalk) or at two different residues that are in close proximity in the protein sequence or are spatially close. Owing to the pathophysiological importance of protein O-GlcNAcylation and phosphorylation, a critical need exists to develop tools for the detection of enzymes involved in these modifications, in particular, OGA and phosphatase. Among the available methods, fluorescence-based detection is highly attractive and powerful because it is greatly sensitive to analytes, inexpensive and does not require sophisticated instrumentation (Chen et al., 2011; Ko et al., 2011; Vahrmeijer et al., 2013; Yuan et al., 2013; Guo et al., 2014; Chen et al., 2016; Gao et al., 2017; Park et al., 2020; Li et al., 2022). Recently, fluorescent probes have been developed to determine the individual activities of OGA and phosphatase in cells. For example, a coumarin-based activity probe and a coumarin-conjugated fluorescein-based probe have been created to capture and detect OGA in cells (Hyun et al., 2019; Jung et al., 2022). While only two fluorescent probes for OGA have been devised to date, numerous fluorescent probes are available for detection of phosphatases (Li et al., 2012; Li et al., 2012; Li et al., 2017; Liu et al., 2017). Although these fluorescent probes have been successfully applied to monitor OGA and phosphatase individually, analysis of the data arising from concurrent measurements using the different types of probes could be complicated and inaccurate because of their different cell permeability and often spectral interference (Yuan et al., 2012; Chen et al., 2016; Zhang et al., 2016). To overcome this limitation, fluorescent probes that can simultaneously detect multiple analytes have been devised (Kolanowski et al., 2018). In the investigation described below, we designed, prepared, and evaluated the new fluorescent probe, βGlcNAc-CM-Rhod-P, which contains βGlcNAc and phosphoryl groups that have different fluorescence responses to respective OGA and phosphatase. The results of this effort showed that this probe can be employed to determine both OGA and phosphatase activities in cell lysates and live cells. ## Results and discussion In considering fluorescent probes for differential detection of OGA and phosphatase, we identified coumarin (CM, λ max,em = 440 nm) and rhodol (Rhod, λ max,em = 540 nm) as possible fluorescent dyes (Supplementary Figure S1), because their emission maxima are separated by 100 nm and, thus, interference between the responses of the two fluorophores will be minimal. Also, since the emission spectrum of CM overlaps to a certain degree with the absorption spectrum of Rhod (Huang et al., 2015; Li et al., 2018; Bai et al., 2019; Ou et al., 2019), fluorescence resonance energy transfer (FRET) from CM to Rhod would be possible (Park et al., 2019; Park et al., 2020; Park et al., 2021). On this basis, we designed the novel fluorescent probe, βGlcNAc-CM-Rhod-P, which is comprised of a βGlcNAc-conjugated CM serving as the OGA substrate, a phosphate-appended Rhod as the phosphatase substrate and a piperazine bridge (Figure 1). FIGURE 1: Fluorescence response of βGlcNAc-CM-Rhod-P to O-GlcNAcase (OGA) and phosphatase (see text for detailed explanation). It was anticipated that βGlcNAc-CM-Rhod-P in which the βGlcNAc and phosphoryl groups are bonded to the hydroxyl groups of respective CM and Rhod would display weak fluorescence. Moreover, we envisaged that treatment of the probe with OGA would lead to a reaction in which the GlcNAc moiety is cleaved from the probe to form CM-Phod-P, thereby increasing the intensity of a fluorescence signal arising from unconjugated CM (Figure 1). On the other hand, addition of phosphatase to the probe would result in cleavage of the phosphoryl group to produce βGlcNAc-CM-Rhod, which would exhibit strong fluorescence from free Rhod. Furthermore, when βGlcNAc-CM-Rhod-P is simultaneously treated with OGA and phosphatase, both βGlcNAc and phosphoryl groups would be cleaved to generate CM-Rhod, an event that would promote increases in the intensities of fluorescence of both CM and Rhod, as well as FRET signals from CM to Rhod. In line with these design principles, the fluorescent probe βGlcNAc-CM-Rhod-P was prepared by using the sequence depicted in Scheme 1. Briefly, 2,4-dihydroxybenzaldehyde [1] was condensed with di-tert-butyl malonate to produce adduct 2 that was then subjected to glycosylation with αGlcNAc(OAc)3-Cl to form glycoside 3 (Park and Shin, 2007; Hyun et al., 2018). Removal of the t-Bu group from 3 generated the corresponding acid 4, which was reacted with rhodol under amide bond forming conditions to yield 5 (Chen, Pacheco and Takano et al., 2016). The phenolic hydroxyl group in 5 was phosphorylated by reaction with diallyl phosphoryl chloride to form phosphate triester 6 (Li et al., 2014). Finally, the target βGlcNAc-CM-Rhod-P was generated by sequential removal of the allyl and O-acetyl groups in 6 using a sequence involving palladium-catalyzed deallylation and treatment with NaOMe, and purification by reversed-phase HPLC (RP-HPLC). In addition, using the pathway shown in Scheme S1, the three analogs βGlcNAc-CM-Rhod, CM-Rhod-P, and CM-Rhod were prepared to confirm the identity of products arising by OGA or/and phosphatase promoted cleavage of βGlcNAc-CM-Rhod-P. All newly synthesized substances were characterized by using NMR and MS methods, and the purities of the final compounds were determined by RP-HPLC. SCHEME 1: Synthesis of βGlcNAc-CM-Rhod-P. We next examined the time-dependent fluorescence responses of βGlcNAc-CM-Rhod-P to OGA or/and phosphatase. The results showed that addition of OGA (100 nM) to the probe (10 μM) in Tris buffer (pH 7.4) gives rise to fluorescence from CM at 450 nm (λ ex = 400 nm) whose intensity increases until ca. 1 h and then reaches saturation (Figure 2; Supplementary Figure S2). The fluorescence intensity of CM generated under these conditions was increased ca. 10 times. However, when the probe was co-incubated with OGA and its inhibitor PUGNAc (50 μM) (Park and Shin, 2007; Hyun et al., 2018; Hyun et al., 2018), the intensity of the fluorescence arising from CM was not increased. Moreover, in contrast to the OGA inhibitor, the phosphatase inhibitor Na3VO4 (1 mM) did not affect the fluorescence response of the probe to OGA (Santos et al., 2010; Elkins et al., 2016). The absorption spectra of the probe treated with OGA displayed an increase in absorbance at 400 nm and a slight decrease at 340 nm in a time-dependent manner (Supplementary Figure S3). FIGURE 2: (A) Time-dependent change in fluorescence spectra after treatment of βGlcNAc-CM-Rhod-P (10 μM) with OGA (100 nM) in 50 mM Tris buffer (pH 7.4) containing 1% DMSO (λex = 400 nm, Δt = 6 min). (B) Time-dependent fluorescence response of βGlcNAc-CM-Rhod-P (10 μM) to OGA (100 nM) in the absence and presence of either 50 μM PUGNAc or 1 mM Na3VO4 (λex = 400 nm, λem = 450 nm, Δt = 2 min). The fluorescence response of βGlcNAc-CM-Rhod-P to phosphatase was also evaluated. Incubation of the probe (10 μM) with alkaline phosphatase (ALP, 100 nM) in Tris buffer (pH 7.4) resulted in enhancement of the intensity of the fluorescence of Rhod at 545 nm (λ ex = 510 nm) up to 25 min (Figure 3). The fluorescence intensity of Rhod produced under these conditions was increased ca. 8 times. However, an increase in fluorescence of Rhod promoted by ALP was not observed when the probe was co-incubated with the inhibitor Na3VO4 (1 mM). On the contrary, the OGA inhibitor PUGNAc (50 μM) had no influence on the fluorescence response of the probe to ALP. The absorption spectra of the probe treated with ALP showed a time-dependent increase in absorbance at 510 nm (Supplementary Figure S4). We also evaluated the fluorescence responses of βGlcNAc-CM-Rhod-P to other phosphatases. The results showed that protein tyrosine phosphatase receptor type O (PTPRO) and dual-specificity phosphatase 15 (DUSP15) induce fluorescence responses of the probe that are similar to but more rapid than ALP (Supplementary Figure S5, S6). FIGURE 3: (A) Time-dependent change in fluorescence spectra after treatment of βGlcNAc-CM-Rhod-P (10 μM) with ALP (100 nM) in 50 mM Tris buffer (pH 7.4) containing 1% DMSO (λex = 510 nm, Δt = 6 min). (B) Time-dependent fluorescence response of βGlcNAc-CM-Rhod-P (10 μM) to ALP (100 nM) in the absence and presence of either 50 μM PUGNAc or 1 mM Na3VO4 (λex = 510 nm, λem = 545 nm, Δt = 2 min). We next assessed the fluorescence response of βGlcNAc-CM-Rhod-P in the presence of both OGA and phosphatase. It was found that addition of both OGA (100 nM) and ALP (100 nM) to the probe (10 μM) results in increases in the intensities of fluorescence associated with CM at 450 nm (λ ex = 400 nm) and Rhod at 545 nm (λ ex = 510 nm) (Figure 4). Moreover, the signal (545 nm with 400 nm excitation) for FRET from CM to Rhod was enhanced. Inspection of the absorption spectrum of the probe after simultaneous treatment with these enzymes showed that increases in absorbance at 400 nm and 510 nm take place in a time-dependent manner (Supplementary Figure S7). The effect of enzyme inhibitors on the fluorescence response of βGlcNAc-CM-Rhod-P to OGA and phosphatase was also evaluated. When the probe (10 μM) was co-treated with both OGA (100 nM) and ALP (100 nM) in the presence of Na3VO4 (1 mM), the fluorescence corresponding to CM at 450 nm (λ ex = 400 nm) was increased while the fluorescence associated with Rhod at 545 nm (λ ex = 510 nm) was not (λ ex = 510 nm) (Figure 5). In a corresponding manner, when the probe (10 μM) was co-treated with both enzymes (100 nM) and PUGNAc (50 μM), the fluorescence arising from Rhod was enhanced but that from CM remained unchanged. Furthermore, the results also showed that the intensities of fluorescence arising from both CM and Rhod remain unaltered by treatment of the probe with both OGA and ALP in the presence of both Na3VO4 and PUGNAc. FIGURE 4: Time-dependent change in fluorescence spectra after treatment of βGlcNAc-CM-Rhod-P (10 μM) with both OGA (100 nM) and ALP (100 nM) in 50 mM Tris buffer (pH 7.4) containing 1% DMSO (upper left: λex = 400 nm, Δt = 6 min, upper right: λex = 510 nm, Δt = 6 min, lower left: red dot line; λex = 400 nm, λem = 545 nm, Δt = 2 min, black dot line; λex = 400 nm, λem = 450 nm, Δt = 2 min, lower right: λex = 510 nm, λem = 545 nm, Δt = 2 min). FIGURE 5: Fluorescence response of βGlcNAc-CM-Rhod-P (10 μM) to OGA (100 nM) and ALP (100 nM) in the absence and presence of 50 μM PUGNAc or/and 1 mM Na3VO4 (‘+’ means the presence of the indicated substance and ‘−’ does the absence of the indicated substance). We also determined the detection limit of βGlcNAc-CM-Rhod-P for enzymes by conducting titration experiments on the probe with OGA or ALP. When 10 μM βGlcNAc-CM-Rhod-P was treated with various concentrations (0–100 nM) of OGA or ALP, respective fluorescence signals at 450 nm (λex = 400 nm) or 545 nm (λex = 510 nm) increased in a concentration-dependent fashion (Figure S8). The fluorescence intensity was plotted against the concentration of each enzyme, showing that it is linearly related to the concentration. The regression equations were determined to be ΔF450 nm = 8.18 × [OGA] + 30.9 (r 2 = 0.99) and ΔF545 nm = 62.5 × [ALP] + 350 (r 2 = 0.99). The detection limits of the probe for OGA and ALP were calculated to be 4.3 and 3.1 nM, respectively, based on a 3σ/slope method (σ: standard deviation) (Wang et al., 2015). The findings provide evidence that βGlcNAc-CM-Rhod-P can be used to sensitively detect the enzymes. To gain more information about the response of βGlcNAc-CM-Rhod-P to OGA or/and ALP, solutions of the probe (10 μM) were treated with each or both enzymes (100 nM) in the absence and presence of the inhibitors, and then analyzed by RP-HPLC. The HPLC profiles demonstrate that upon treatment of the probe with only OGA, βGlcNAc-CM-Rhod-P completely disappears concomitant with production of CM-Rhod-P (Figure 6). However, co-treatment of the probe with OGA and PUGNAc did not result in formation of CM-Rhod-P (Supplementary Figure S8). It was also found that exposure of the probe to ALP promotes generation of βGlcNAc-CM-Rhod, which does not occur when Na3VO4 is present. Moreover, the probe was completely converted to CM-Rhod upon treatment with both enzymes. The results of studies with inhibitors revealed that the βGlcNAc-removed (CM-Rhod-P) or phosphoryl group-cleaved product (βGlcNAc-CM-Rhod) is produced when βGlcNAc-CM-Rhod-P is treated with both enzymes and either Na3VO4 or PUGNAc, respectively. However, βGlcNAc-CM-Rhod-P remained unchanged when it was co-incubated with both enzymes and both inhibitors. The combined results indicate that selective cleavage of the βGlcNAc and phosphoryl groups occurs when βGlcNAc-CM-Rhod-P is treated with OGA and ALP, respectively. Also, the findings support the initial proposal that βGlcNAc-CM-Rhod-P is suitable for simultaneously monitoring activities of both OGA and phosphatase. FIGURE 6: Reversed-phase HPLC analysis of products obtained by treatment of 10 μM βGlcNAc-CM-Rhod-P with OGA (100 nM) or/and ALP (100 nM) (b = βGlcNAc-CM-Rhod ([M + H]+: m/z = 792.2), c = CM-Rhod-P ([M + H]+: m/z = 669.1), d = CM-Rhod ([M + H]+: m/z = 589.1)). In the next phase of this investigation, we assessed the use of βGlcNAc-CM-Rhod-P to monitor OGA and phosphatases in live cells. Prior to beginning this study, AGS cells were treated with several non-cytotoxic concentrations (0–100 μM) of βGlcNAc-CM-Rhod-P for 18 h (Supplementary Figure S9). In addition, these cells were incubated with 100 μM βGlcNAc-CM-Rhod-P for several time periods (0–18 h). Analysis of confocal fluorescence microscopy images revealed that AGS cells treated with 100 μM βGlcNAc-CM-Rhod-P for 18 h display substantial fluorescence associated with CM and Rhod when excited at 405 nm and 488 nm, respectively (Supplementary Figures S10, S11). We next explored the application of βGlcNAc-CM-Rhod-P to imaging OGA and phosphatases in other types of cells. For this purpose, HeLa (human cervical cancer cells) and A549 (human lung adenocarcinoma cells) cells along with AGS cells were independently incubated with 100 μM βGlcNAc-CM-Rhod-P for 18 h. Analysis of the intensities of fluorescence arising from CM and Rhod in the treated cells using confocal fluorescence microscopy indicated that OGA and phosphatases activities are dependent on the cell type (Figure 7). Specifically, the activities of both enzymes in HeLa cells are higher than those in AGS and A549 cells, these enzymes in the latter two cells are similar. FIGURE 7: The indicated cells were incubated with 100 μM βGlcNAc-CM-Rhod-P for 18 h. Cell images were obtained by using confocal fluorescence microscopy (scale bar = 10 μm). Graphs show the fluorescence intensity of Rhod (λex = 488 nm) and CM (λex = 405 nm) in cells (mean ± s.d., n = 3). Finally, we tested the utility of βGlcNAc-CM-Rhod-P to monitor activities of OGA and phosphatases in cell lysates. In this study, lysates of HeLa, AGS and A549 cells were individually exposed to the probe. Fluorescence intensities corresponding to CM and Rhod dyes in cell lysates were then determined using a fluorescence microplate reader. The results revealed that HeLa cell lysates treated with the probe display higher fluorescence intensities associated with CM and Rhod than do the other cell lysates after treatment (Supplementary Figure S12). These results are consistent with those obtained from experiments using intact cells. Taken together, the findings provide evidence that βGlcNAc-CM-Rhod-P is applicable to both fluorescence imaging of OGA and phosphatases in cells and fluorescence detection of these enzymes in cell lysates. ## Conclusion Protein O-GlcNAcylation frequently has crosstalk with phosphorylation at the same or two different residues within a protein. Because protein O-GlcNAcylation and phosphorylation are implicated in a wide range of physiological processes and their aberrant modifications cause various human diseases, fluorescent probes for simultaneous detection of both OGA and phosphatase that are crucial for these modifications are in great demand. To date, fluorescent probes that individually measure OGA and phosphatase activities have been developed. However, in many cases, analysis of data obtained using the individual probes for OGA and phosphatase can be both complicating and inaccurate because of different levels of cell penetration and spectral interference. To circumvent this issue, we designed the fluorescent probe βGlcNAc-CM-Rhod-P for monitoring both OGA and phosphatase at the same time. As described above, incubation of βGlcNAc-CM-Rhod-P with phosphatase produced βGlcNAc-CM-Rhod in association with an increase in the intensity of fluorescence arising from Rhod. On the other hand, the fluorescence intensity corresponding to CM was enhanced by production of CM-Rhod-P when the probe was treated with OGA. Moreover, addition of both OGA and phosphatase to the probe led to production of CM-Rhod, thereby increasing the intensities of fluorescence arising from CM and Rhod. The results of cell studies revealed that βGlcNAc-CM-Rhod-P can be employed to fluorescently detect OGA and phosphatase activities in cell lysates and to image these enzymes in cells. It is anticipated that the strategy utilized to design this probe will provide a foundation for creating new probes that possess the ability to concurrently detect two different enzymes. ## General All solvents and chemicals used in the study were purchased from Sigma-Aldrich, Tokyo Chemical Industry (TCI) and Acros in analytical grade, unless particularly mentioned. Alkaline phosphatase (ALP) was purchased from Sigma-Aldrich and other phosphatases (protein tyrosine phosphatase receptor type O (PTPRO) and dual-specificity phosphatase 15 (DUSP15)) were provided by Dr. Bonsu Ku. NMR spectra were recorded on Bruker Avance lll HD 400 and Avance II 400 instruments. High-resolution mass spectrometry data were obtained using an Ultimate 3000 RS-Q-Exactive Orbitrap Plus. UV/VIS absorption spectra were collected on a JASCO V-650 spectrophotometer and fluorescence emission spectra on a JASCO FP-8500 fluorescence spectrophotometer. ## Synthesis of compound 2 To a solution of 2,4-dihydroxybenzaldehyde (1, 1 g, 7.2 mmol) and di-tert-butyl malonate (2.4 mL, 2.3 g, 10.8 mmol) in tert-butanol (17 mL) was added piperidine (0.5 mL, 431 mg, 5.06 mmol) with stirring at room temperature. The mixture was stirred at reflux for 12 h and then cooled to room temperature. The mixture was concentrated under reduced pressure, and the residue was subjected to flash column chromatography using hexane/ethyl acetate (v/v, 4:1) as the eluent to give 2 (700 mg) as a pale yellow solid: yield $36\%$. 1H NMR (400 MHz, DMSO-d 6) d 8.55 (s, 1 H), 7.72 (d, 1 H, $J = 11.2$ Hz), 6.83 (dd, 1 H, $J = 11.4$, 3.2 Hz), 6.71 (d, 1 H, $J = 2.8$ Hz), 1.51 (s, 9 H). 13C NMR (100 MHz, DMSO-d 6) d 164.1, 162.2, 157.0, 156.6, 148.5, 131.8, 114.0, 113.5, 110.3, 101.8, 81.2, 27.8. High-resolution mass spectrometry (ESI-MS, m/z): [M + Na]+ calcd. for [C14H14O5 + Na]+ 285.0739; found 285.0731. ## Synthesis of compound 3 To a solution of 2 (600 mg, 2.28 mmol) in 1:1 of chloroform and water (12 mL) was sequentially added 2-acetamido-3,4,6-tri-O-acetyl-2-deoxy-α-D-glucopyranosyl chloride (αGlcNAc(OAc)3-Cl, 1 g, 2.74 mmol), tetrabutylammonium hydrogen sulfate (TBAHS, 855 mg, 2.5 mmol) and K2CO3 (632 mg, 4.57 mmol) with stirring at room temperature. After stirring for 16 h, the mixture was diluted with dichloromethane and washed sequentially with saturated NH4Cl solution, water and brine. The organic layer was dried over anhydrous Na2SO4, filtered and concentrated under reduced pressure. The residue was subjected to flash column chromatography using hexane/ethyl acetate (v/v, 3:1) as the eluent to give 3 (890 mg) as a white solid: yield $65\%$. 1H NMR (400 MHz, CDCl3) d 8.30 (s, 1 H), 7.45 (d, 1 H, $J = 8.8$ Hz), 7.00 (s, 1 H), 6.94 (d, 1 H, $J = 8.4$ Hz), 6.58 (d, 1 H, $J = 9.2$ Hz), 5.63 (d, 1 H, $J = 8.0$ Hz), 5.49 (t, 1 H, $J = 9.2$ Hz), 5.13 (t, 1 H, $J = 9.6$ Hz), 4.32–4.07 (m, 4 H), 2.06 (t, 9 H $J = 12.4$ Hz), 1.94 (s, 3 H), 1.57 (s, 9 H). 13C NMR (100 MHz, CDCl3) d 171.1, 170.7, 170.7, 169.5, 161.9, 157.3, 156.7, 147.9, 130.6, 116.0, 115.4, 112.8, 102.8, 97.5, 82.8, 72.2, 72.1, 68.5, 62.0, 54.0, 28.1, 23.2, 20.7, 20.7, 20.6. High-resolution mass spectrometry (ESI-MS, m/z): [M]− calcd. for [C28H33NO13]− 590.1874; found 590.1874. ## Synthesis of compound 4 A mixture of 3 (600 mg, 1.01 mmol) and trifluoroacetic acid (TFA, 2.5 mL) in dichloromethane (7.5 mL) was stirred for 1 h at room temperature. The mixture was concentrated under reduced pressure to give 4 (531 mg) as a white solid: yield $98\%$. The crude product was used for the next reaction without further purification. 1H NMR (400 MHz, DMSO-d 6) d 8.18 (d, 1 H, $J = 8.8$ Hz), 8.08 (s, 1 H), 7.67 (d, 1 H, $J = 8.8$ Hz), 7.04 (s, 1 H), 6.93 (d, 1 H, $J = 8.4$ Hz), 5.48 (d, 1 H, $J = 8.4$ Hz), 5.21 (t, 1 H, $J = 9.6$ Hz), 4.93 (t, 1 H, $J = 9.6$ Hz), 4.20–4.25 (m, 2 H), 4.04–4.18 (m, 2 H), 2.01 (d, 6 H, $J = 3.2$ Hz), 1.95 (s, 1 H), 1.78 (s, 1 H). 13C NMR (100 MHz, DMSO-d 6) d 170.1, 169.8, 169.8, 169.5, 160.3, 158.5, 158.2, 155.3, 130.9, 118.8, 115.8, 113.9, 102.8, 97.2, 72.5, 71.1, 68.4, 61.7, 53.1, 22.7, 20.5, 20.4. Mass spectrometry (ESI-MS, m/z): [M + H]+ calcd. For [C24H25NO13 +H]+ 536.1; found 536.6. High-resolution mass spectrometry (ESI-MS, m/z): [M + H]+ calcd. For [C24H24NO13 + H]+ 534.1253; Found 534.1255. ## Synthesis of compound 5 A mixture of 4 (320 mg, 0.59 mmol), rhodol (287 mg, 0.71 mmol, see Supporting Information for its synthesis), 2-(1H-benzotriazol-1-yl)-1,1,3,3-tetramethyluronium hexafluorophosphate (HBTU, 272 mg, 0.71 mmol), 1-hydroxybenzotriazole (HOBt, 97 mg, 0.71 mmol) and diisopropylethylamine (DIEA, 0.39 mL, 309 mg, 2.39 mmol) in N,N′-dimethylformamide (3 mL) was stirred for 8 h at room temperature. The mixture was diluted with ethyl acetate and washed with water and brine. The organic layer was dried over anhydrous Na2SO4, filtered and concentrated under reduced pressure. The residue was subjected to flash column chromatography using ethyl acetate/methanol (v/v, 70:1) as the eluent to give 5 (293 mg) as a red solid: yield $53\%$. 1H NMR (400 MHz, DMSO-d 6) d 8.20 (s, 1 H), 8.16 (s, 1 H), 7.99 (d, 1 H, $J = 7.6$ Hz), 7.80–7.69 (m, 3 H), 7.25 (d, 1 H, $J = 7.6$ Hz), 7.19 (s, 1 H), 7.04 (d, 1 H, $J = 8.8$ Hz), 6.84 (s, 1 H), 6.76 (d, 1 H, $J = 8.8$ Hz), 6.67 (s, 1 H), 6.57–6.55 (m, 3 H), 5.55 (d, 1 H, $J = 8.0$ Hz), 5.24 (t, 1 H, $J = 10.4$ Hz), 4.96 (t, 1 H, $J = 9.6$ Hz), 4.28–4.19 (m, 2 H), 4.10–4.03 (m, 2 H), 3.73 (s, 2 H), 3.52 (s, 2 H), 3.33 (s, 2 H), 3.25 (s, 2 H), 2.02 (d, 6 H, $J = 4.4$ Hz), 1.96 (s, 3 H), 1.79 (s, 3 H). 13C NMR (100 MHz, DMSO-d 6) d 170.1, 169.8, 169.7,169.4, 168.8, 163.2, 159.9, 157.9, 155.2, 152.3, 152.2, 151.9, 142.7, 135.5, 130.3, 130.1, 129.2, 128.6, 126.5, 124.8, 124.2, 121.9, 114.2, 113.4, 112.9, 112.1, 109.7, 108.8, 103.1, 102.3, 101.6, 97.1, 72.3, 71.1, 68.3, 61.6, 53.0, 47.8, 47.2, 45.9, 41.1, 22.7, 20.5, 20.5, 20.4. High-resolution mass spectrometry (ESI-MS, m/z): [M + H]+ calcd. For [C48H43N3O16 + H]+ 917.2721; Found 917.2720. ## Synthesis of compound 6 To a solution of 5 (130 mg, 0.14 mmol) in anhydrous N,N′-dimethylformamide (2.3 mL) was added trimethylamine (TEA, 79 μL, 57.3 mg, 0.56 mmol) and diallyl phosphoryl chloride (47 μL, 55 mg, 0.28 mmol, see Supporting Information for its synthesis) with stirring at 0 C under a nitrogen atmosphere. The mixture was warmed to room temperature. After stirring for 4 h at room temperature, the mixture was diluted with ethyl acetate and washed with water and brine. The organic layer was dried over anhydrous Na2SO4, filtered and concentrated under reduced pressure. The residue was subjected to flash column chromatography using ethyl acetate/methanol (v/v, 90:1) as the eluent to give 6 (95 mg) as a red solid: yield $62\%$. 1H NMR (400 MHz, DMSO-d 6) d 8.19 (s, 1 H), 8.16 (d, 1 H, $J = 9.2$ Hz), 8.02 (d, 1 H, $J = 7.6$ Hz), 7.80–7.71 (m, 3 H), 7.30 (d, 1 H, $J = 7.6$ Hz), 7.23 (d, 1 H, $J = 1.2$ Hz), 7.18 (d, 1 H, $J = 2.0$ Hz), 7.03 (dd, 1 H, $J = 8.6$, 2.4 Hz), 6.97 (d, 1 H, $J = 1.6$ Hz), 6.87–6.83 (m, 3 H), 6.60 (d, 1 H, $J = 8.8$ Hz), 5.96–5.94 (m, 2 H), 5.53 (d, 1 H, $J = 8.4$ Hz), 5.39 (s, 1 H), 5.35 (s, 1 H), 5.27–5.20 (m, 3 H), 4.94 (t, 1 H, $J = 9.6$ Hz), 4.67 (t, 3 H, $J = 4.0$ Hz), 4.27–4.18 (m, 2 H), 4.09–4.02 (m, 2 H), 3.73 (s, 2 H), 3.52 (s, 2 H), 3.26 (s, 2 H), 2.01 (d, 6 H, $J = 4.8$ Hz), 1.95 (s, 3 H), 1.78 (s, 3 H). 13C NMR (100 MHz, CD3OD) d 172.2, 171.8, 171.3, 171.1, 165.9, 162.1, 159.8, 156.9, 154.1, 153.5, 153.4, 153.0, 144.6, 136.8, 133.4, 133.3, 131.5, 131.4, 130.8, 129.7, 127.7, 126.0, 125.2, 122.8, 119.4, 117.9, 117.2, 115.7, 114.9, 113.8, 112.1, 110.1, 109.7, 104.6, 103.3, 99.0, 84.2, 73.7, 73.3, 70.6, 70.5, 70.0, 63.2, 55.3, 47.8, 42.9, 22.8, 20.8, 20.6.31P NMR (162 MHz, CD3OD) d 6.60. High-resolution mass spectrometry (ESI-MS, m/z): [M + H]+ calcd. For [C54H52N3O19P + H]+ 1078.3011; found 1078.3011. ## Synthesis of βGlcNAc-CM-Rhod-P To a solution of 6 (108 mg, 0.1 mmol) and Pd(PPh3)4 (17 mg, 0.01 mmol) in anhydrous tetrahydrofuran (THF, 3 mL) was added phenylsilane (74.1 μL, 65.1 mg, 0.6 mmol) with stirring at room temperature under an argon atmosphere. After stirring for 20 min, the mixture was diluted with dichloromethane, filtered and washed with dichloromethane. The solvent was removed under reduced pressure to give 7 (100 mg) as a red solid: yield $99\%$. The crude product was used for the next reaction without further purification. The mixture of crude 7 (100 mg, 0.1 mmol) and sodium methoxide (0.5 M in methanol, 1.0 mL, 0.5 mmol) in methanol (2 mL) was stirred for 1 h at room temperature. After neutralization with Amberite IR-120 (H+) ion exchange resins, the mixture was filtered and the resins were washed with methanol thoroughly. The solvent was removed under reduced pressure. The residue was purified by RP-HPLC to give ß-GlcNAc-CM-Rhod-P (10 mg) as a red solid: yield $22\%$. 1H NMR (400 MHz, DMSO-d 6) d 8.18 (s, 1 H), 7.98 (d, 1 H, $J = 6.8$ Hz), 7.87 (d, 1 H, $J = 9.2$ Hz), 7.78–7.68 (m, 3 H), 7.26 (d, 1 H, $J = 7.2$ Hz), 7.20 (s, 1 H), 7.06 (s, 1 H), 6.98 (dd, 1 H, $J = 8.4$, 2.0 Hz), 6.90 (d, 1 H, $J = 8.8$ Hz), 6.83–6.76 (m, 2 H), 6.67 (d, 1 H, $J = 8.0$ Hz), 6.57 (d, 1 H, $J = 8.8$ Hz), 5.16 (d, 1 H, $J = 8.4$ Hz), 3.74–3.61 (m, 4 H), 3.50–3.42 (m, 7 H), 3.30 (s, 2 H), 3.23–3.16 (m, 4 H), 1.81 (s, 3 H). 13C NMR (100 MHz, DMSO-d 6) d 169.5, 168.8, 163.2, 160.9, 158.0, 155.3, 154.1, 152.5, 152.3, 151.7, 151.4, 142.9, 135.8, 130.3, 128.9, 128.6, 126.0, 124.8, 124.1, 121.5, 116.6, 116.6, 114.3, 113.8, 113.1, 112.3, 108.4, 107.9, 103.1, 101.6, 98.7, 82.6, 77.4, 74.0, 70.2, 60.7, 55.3, 47.7, 47.2, 45.9, 41.1, 23.2.31P NMR (162 MHz, DMSO-d 6) d −5.80. High-resolution mass spectrometry (ESI-MS, m/z): [M + Na]+ calcd. For [C42H38N3O16P + Na]+ 894.1888; found 894.1880. ## Data availability statement The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author. ## Author contributions IS designed the study and wrote the manuscript. JB, JL, and C-HL helped with writing the manuscript. JB and Y-HK designed, prepared and characterized the compounds. JL conducted cell studies. C-HL and BK were responsible for providing enzymes. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fchem.2023.1133018/full#supplementary-material ## References 1. Aebersold R., Agar J. N., Amster I. J., Baker M. S., Bertozzi C. R., Boja E. S.. How many human proteoforms are there?. Nat. Chem. Biol. (2018) 14 206-214. DOI: 10.1038/NCHEMBIO.2576 2. Bai Y., Wu M. X., Ma Q.-J., Wang C.-Y., Sun J.-G., Tian M.-J.. A FRET-based ratiometric fluorescent probe for highly selective detection of cysteine based on a coumarin-rhodol derivative. New J. Chem. (2019) 43 14763-14771. DOI: 10.1039/c9nj03375k 3. Banerjee P. S., Hart G. W., Cho J. W.. Chemical approaches to study O-GlcNAcylation. Chem. Soc. Rev. (2013) 42 4345-4357. DOI: 10.1039/c2cs35412h 4. Chen W., Pacheco A., Takano Y., Day J. J., Hanaoka K., Xian M.. A single fluorescent probe to visualize hydrogen sulfide and hydrogen polysulfides with different fluorescence signals. Angew. Chem. Int. Ed. (2016) 55 9993-9996. DOI: 10.1002/anie.201604892 5. Chen X., Tian X., Shin I., Yoon J.. Fluorescent and luminescent probes for detection of reactive oxygen and nitrogen species. Chem. Soc. Rev. (2011) 40 4783-4804. DOI: 10.1039/c1cs15037e 6. Chen X., Wang F., Hyun J. Y., Wei T., Qiang J., Ren X.. Recent progress in the development of fluorescent, luminescent and colorimetric probes for detection of reactive oxygen and nitrogen species. Chem. Soc. Rev. (2016) 45 2976-3016. DOI: 10.1039/c6cs00192k 7. Conibear A. C.. Deciphering protein post-translational modifications using chemical biology tools. Nat. Rev. Chem. (2020) 4 674-695. DOI: 10.1038/s41570-020-00223-8 8. Dong D. L. Y., Hart G. W.. Purification and characterization of an O-GlcNAc selective N-acetyl-beta-D-glucosaminidase from rat spleen cytosol. J. Biol. Chem. (1994) 269 19321-19330. DOI: 10.1016/S0021-9258(17)32170-1 9. Elkins J., Fedele V., Szklarz M., Azeez A. K. R., Salah E., Mikolajczyk J.. Comprehensive characterization of the published kinase inhibitor set. Nat. Biotechnol. (2016) 34 95-103. DOI: 10.1038/nbt.3374 10. Ferrer C. M., Sodi V. L., Reginato M. J.. O-GlcNAcylation in cancer biology: Linking metabolism and signaling. J. Mol. Biol. (2016) 428 3282-3294. DOI: 10.1016/j.jmb.2016.05.028 11. Gao M., Yu F., Lv C., Choo J., Chen L.. Fluorescent chemical probes for accurate tumor diagnosis and targeting therapy. Chem. Soc. Rev. (2017) 46 2237-2271. DOI: 10.1039/c6cs00908e 12. Gao Y., Wells L., Comer F. I., Parker G. J., Hart G. W.. Dynamic O-glycosylation of nuclear and cytosolic proteins: Cloning and characterization of a neutral, cytosolic beta-N-acetylglucosaminidase from human brain. J. Biol. Chem. (2001) 276 9838-9845. DOI: 10.1074/jbc.M010420200 13. Guo Z., Park S., Yoon J., Shin I.. Recent progress in the development of near-infrared fluorescent probes for bioimaging applications. Chem. Soc. Rev. (2014) 43 16-29. DOI: 10.1039/c3cs60271k 14. Hart G. W., Housley M. P., Slawson C.. Cycling of O-linked β-N-acetylglucosamine on nucleocytoplasmic proteins. Nature (2007) 446 1017-1022. DOI: 10.1038/nature05815 15. Hart G. W., Slawson C., Ramirez-Correa G., Lagerlof O.. Cross talk between O-GlcNAcylation and phosphorylation: Roles in signaling, transcription, and chronic disease. Annu. Rev. Biochem. (2011) 80 825-858. DOI: 10.1146/annurev-biochem-060608-102511 16. Huang K., Liu M., Cao D., Hou J., Zeng W.. Ratiometric and colorimetric detection of hydrogen sulfide with high selectivity and sensitivity using a novel FRET-based fluorescence probe. Dyes Pigments (2015) 118 88-94. DOI: 10.1016/j.dyepig.2015.03.007 17. Hyun J. Y., Kang N. R., Shin I.. Carbohydrate microarrays containing glycosylated fluorescent probes for assessment of glycosidase activities. Org. Lett. (2018) 20 1240-1243. DOI: 10.1021/acs.orglett.8b00180 18. Hyun J. Y., Kim S., Lee H. S., Shin I.. A Glycoengineered Enzyme with Multiple Mannose-6-Phosphates is Internalized into diseased cells to restore its activity in lysosomes. Cell Chem. Biol. (2018) 25 1255-1267.e8. DOI: 10.1016/j.chembiol.2018.07.011 19. Hyun J. Y., Park S.-H., Park C. W., Kim H. B., Cho J. W., Shin I.. Trifunctional fluorogenic probes for fluorescence imaging and isolation of glycosidases in cells. Org. Lett. (2019) 21 4439-4442. DOI: 10.1021/acs.orglett.9b01147 20. Jung H., Park S.-H., Yang W. H., Cho J. W., Shin I.. An O-GlcNAcase responsive fluorescent probe for biological applications. Sens. Actuators B Chem. (2022) 367 132093-132100. DOI: 10.1016/j.snb.2022.132093 21. Ko S.-K., Chen X., Yoon J., Shin I.. Zebrafish as a good vertebrate model for molecular imaging using fluorescent probes. Chem. Soc. Rev. (2011) 40 2120-2130. DOI: 10.1039/c0cs00118j 22. Kolanowski J. L., Liu F., New E. J.. Fluorescent probes for the simultaneous detection of multiple analytes in biology. Chem. Soc. Rev. (2018) 47 195-208. DOI: 10.1039/c7cs00528h 23. Lee B., Suh P., Kim J.. O-GlcNAcylation in health and neurodegenerative diseases. Exp. Mol. Med. (2021) 53 1674-1682. DOI: 10.1038/s12276-021-00709-5 24. Li H., Kim Y., Jung H., Hyun J. Y., Shin I.. Near-infrared (NIR) fluorescence-emitting small organic molecules for cancer imaging and therapy. Chem. Soc. Rev. (2022) 51 8957-9008. DOI: 10.1039/d2cs00722c 25. Li L., Ge J., Wu H., Xu Q.-H., Yao S. Q.. Organelle-specific detection of phosphatase activities with two-photon fluorogenic probes in cells and tissues. J. Am. Chem. Soc. (2012) 134 12157-12167. DOI: 10.1021/ja3036256 26. Li L., Shen X., Xu Q.-H., Yao S. Q.. A switchable two-photon membrane tracer capable of imaging membrane-associated protein tyrosine phosphatase activities. Angew. Chem. Int. Ed. (2012) 52 424-428. DOI: 10.1002/anie.201205940 27. Li S. J., Li C. Y., Li Y. F., Fei J., Wu P., Yang B.. Facile and sensitive near-infrared fluorescence probe for the detection of endogenous alkaline phosphatase activity. Anal. Chem. (2017) 89 6854-6860. DOI: 10.1021/acs.analchem.7b01351 28. Li T., Tikad A., Pan W., Vincent S.. β-Stereoselective Phosphorylations applied to the synthesis of ADP- and Polyprenyl-β-Mannopyranosides. Org. Lett. (2014) 16 5628-5631. DOI: 10.1021/ol5026876 29. Li W., Wang X., Zhang Y. M., Zhang S. X.. Single probe giving different signals towards reactive oxygen species and nitroxyl. Dyes Pigments (2018) 148 348-352. DOI: 10.1016/j.dyepig.2017.09.033 30. Liu H.-W., Li K., Hu X. X., Zhu L., Rong Q., Liu Y.. Angew. Chem. Int. Ed. (2017) 56 11788-11792. DOI: 10.1002/anie.201705747 31. Ma J., Hart G. W.. Protein O-GlcNAcylation in diabetes and diabetic complications. Expert Rev. Proteomics (2013) 10 365-380. DOI: 10.1586/14789450.2013.820536 32. Ou P., Zhang R., Liu Z., Tian X., Han G., Liu B.. Gasotransmitter regulation of phosphatase activity in live cells studied by three-channel imaging correlation. Angew. Chem. Int. Ed. (2019) 58 2261-2265. DOI: 10.1002/anie.201811391 33. Park S.-H., Kim S., Lee H. S., Shin I.. Real-time spatial and temporal analysis of the translocation of the apoptosis-inducing factor in cells. ACS Chem. Biol. (2021) 16 2462-2471. DOI: 10.1021/acschembio.1c00565 34. Park S.-H., Ko W., Lee H. S., Shin I.. Analysis of protein-protein interaction in a single live cell by using a FRET system based on genetic code expansion technology. J. Am. Chem. Soc. (2019) 141 4273-4281. DOI: 10.1021/jacs.8b10098 35. Park S-H., Ko W., Park S.-H., Lee H. S., Shin I.. Evaluation of the interaction between Bax and Hsp70 in cells by using a FRET system consisting of a fluorescent amino acid and YFP as a FRET pair. ChemBioChem (2020) 21 59-63. DOI: 10.1002/cbic.201900293 36. Park S.-H., Kwon N., Lee J.-H., Yoon J., Shin I.. Synthetic ratiometric fluorescent probes for detection of ions. Chem. Soc. Rev. (2020) 49 143-179. DOI: 10.1039/c9cs00243j 37. Park S., Shin I.. Profiling of glycosidase activities using coumarin-conjugated glycoside cocktails. Org. Lett. (2007) 9 619-622. DOI: 10.1021/ol062889f 38. Saha A., Bello D., Fernández-Tejada A.. Advances in chemical probing of protein O-GlcNAc glycosylation: Structural role and molecular mechanisms. Chem. Soc. Rev. (2021) 50 10451-10485. DOI: 10.1039/d0cs01275k 39. Santos F. P. S., Kantarjian H. M., Jain N., Manshouri T., Thomas D. A., Garcia-Manero G.. Phase 2 study of CEP-701, an orally available JAK2 inhibitor, in patients with primary or post-polycythemia vera/essential thrombocythemia myelofibrosis. Blood (2010) 115 1131-1136. DOI: 10.1182/blood-2009-10-246363 40. Slawson C., Hart G. W.. O-GlcNAc signalling: Implications for cancer cell biology. Nat. Rev. Cancer. (2011) 11 678-684. DOI: 10.1038/nrc3114 41. Vahrmeijer A. L., Hutteman M., van der Vorst J. R., van de Velde C. J., Frangioni J. V.. Image-guided cancer surgery using near-infrared fluorescence. Nat. Rev. Clin. Oncol. (2013) 10 507-518. DOI: 10.1038/nrclinonc.2013.123 42. van der Laarse S. A. M., Leney A. C., Heck A. J. R.. Crosstalk between phosphorylation and O‐GlcNAcylation: Friend or foe. Febs. J. (2018) 285 3152-3167. DOI: 10.1111/febs.14491 43. Wang Q., Zhang S., Ge H., Tian G, Cao N., Li Y.. A fluorescent turnoff/on method based on carbon dots as fluorescent probes for the sensitive determination of Pb. Sens. Actuators B Chem (2013) 207 25-33. DOI: 10.1016/j.snb.2014.10.096 44. Walsh C. T., Garneau-Tsodikova S., Gatto G. J.. Protein posttranslational modifications: The chemistry of proteome diversifications. Angew. Chem. Int. Ed. (2005) 44 7342-7372. DOI: 10.1002/anie.200501023 45. Wulff-Fuentes E., Berendt R. R., Massman L., Danner L., Malard F., Vora J.. The human O-GlcNAcome database and meta-analysis. Sci. Data (2021) 8 25-11. DOI: 10.1038/s41597-021-00810-4 46. Yang X., Coulombe-Huntington J., Kang S., Sheynkman G. M., Hao T., Richardson A.. Widespread expansion of protein interaction capabilities by alternative splicing. Cell (2016) 164 805-817. DOI: 10.1016/j.cell.2016.01.029 47. Yang X., Qian K.. Protein O-GlcNAcylation: Emerging mechanisms and functions. Nat. Rev. Mol. Cell Biol. (2017) 18 452-465. DOI: 10.1038/nrm.2017.22 48. Yuan L., Lin W., Xie Y., Chen B., Zhu S.. Single fluorescent probe responds to H. J. Am. Chem. Soc. (2012) 134 1305-1315. DOI: 10.1021/ja2100577 49. Yuan L., Lin W., Zheng K., He L., Huang W.. Far-red to near infrared analyte-responsive fluorescent probes based on organic fluorophore platforms for fluorescence imaging. Chem. Soc. Rev. (2013) 42 622-661. DOI: 10.1039/c2cs35313j 50. Yuzwa S. A., Shan X., Macauley M. S., Clark T., Skorobogatko Y., Vosseller K.. Increasing O-GlcNAc slows neurodegeneration and stabilizes tau against aggregation. Nat. Chem. Biol. (2012) 8 393-399. DOI: 10.1038/NCHEMBIO.797 51. Zhang R., Zhao J., Han G., Liu Z., Liu C., Zhang C.. Real-Time discrimination and versatile profiling of spontaneous reactive oxygen species in living organisms with a single fluorescent probe. J. Am. Chem. Soc. (2016) 138 3769-3778. DOI: 10.1021/jacs.5b12848 52. Zhu Y., Hart G. W.. Targeting O-GlcNAcylation to develop novel therapeutics. Mol. Asp. Med. (2021) 79 100885-110899. DOI: 10.1016/j.mam.2020.100885
--- title: 'Coexistence of tmexCD3-toprJ1b tigecycline resistance genes with two novel bla VIM-2-carrying and bla OXA-10-carrying transposons in a Pseudomononas asiatica plasmid' authors: - Qin Li - Qiao Chen - Shuang Liang - Wei Wang - Bingying Zhang - Alberto J. Martín-Rodríguez - Qinghua Liang - Feiyang Zhang - Ling Guo - Xia Xiong - Renjing Hu - Li Xiang - Yingshun Zhou journal: Frontiers in Cellular and Infection Microbiology year: 2023 pmcid: PMC10015498 doi: 10.3389/fcimb.2023.1130333 license: CC BY 4.0 --- # Coexistence of tmexCD3-toprJ1b tigecycline resistance genes with two novel bla VIM-2-carrying and bla OXA-10-carrying transposons in a Pseudomononas asiatica plasmid ## Abstract ### Introduction Tigecycline and carbapenems are considered the last line of defense against microbial infections. The co-occurrence of resistance genes conferring resistance to both tigecycline and carbapenems in Pseudomononas asiatica was not investigated. ### Methods P. asiatica A28 was isolated from hospital sewage. Antibiotic susceptibility testing showed resistance to carbapenem and tigecycline. WGS was performed to analyze the antimicrobial resistance genes and genetic characteristics. Plasmid transfer by conjugation was investigated. Plasmid fitness costs were evaluated in Pseudomonas aeruginosa transconjugants including a Galleria mellonella infection model. ### Results Meropenem and tigecycline resistant P. asiatica A28 carries a 199, 972 bp long plasmid PLA28.4 which harbors seven resistance genes. Sequence analysis showed that the 7113 bp transposon Tn7389 is made up of a class I integron without a 5’CS terminal and a complete tni module flanked by a pair of 25bp insertion repeats. Additionally, the Tn7493 transposon, 20.24 kp long, with a complete 38-bp Tn1403 IR and an incomplete 30-bp Tn1403 IR, is made up of partial skeleton of Tn1403, a class I integron harboring bla OXA-10, and a Tn5563a transposon. Moreover, one tnfxB3-tmexC3.2-tmexD3b-toprJ1b cluster was found in the plasmid and another one in the the chromosome. Furthermore, plasmid PLA28.4 could be conjugated to P. aeruginosa PAO1, with high fitness cost. ### Discussion A multidrug-resistant plasmid carrying tmexCD3-toprJ1b and two novel transposons carrying bla VIM-2 and bla OXA-10 -resistant genes was found in hospital sewage, increasing the risk of transmission of antibiotic-resistant genes. These finding highlight the necessary of controlling the development and spread of medication resistance requires continuous monitoring and management of resistant microorganisms in hospital sewage. ## Introduction Tigecycline is one of the last lines of defense against carbapenem-resistant bacterial infections (Aghapour et al., 2019; Jo and Ko, 2021). The Resistance-Nodulation-Division (RND) MDR efflux pump gene cluster tmexCD1-toprJ1 or the variants such as the tnfxB3-tmexC3.2-tmexD3b-toprJ1b is one of the mechanisms which mediates the tigecycline resistance. Additionally, metallo-β-lactamases (MBLs) and carbapenemase coding genes like bla KPC are the main mechanisms mediating carbapenem resistance (Hu et al., 2021; Huang et al., 2022). Emergence of tigecycline and carbapenem resistant bacteria such as E. coli, Klebsiella spp. and the Pseudomonas spp. from the patients poses great challenges to infection control (Lv et al., 2020; Wang et al., 2021a; Wang et al., 2021b; Gao et al., 2022; Li et al., 2022). The genes encoding resistance determinants such as MBLs are usually found in plasmids or are associated with integrons and transposons (Mann et al., 2022). Integrons are able to capture genes that are part of gene cassettes via a site-specific recombination event and transposons contribute significantly to the transfer and transmission of antibiotic resistance (AR) in bacterial populations (Alavi et al., 2011; Mann et al., 2022). It is commonly believed that the hospital sewage provides a significant platform for the generation of new transposons and many of the novel transposons have been reported from the sewage. Acinetobacter johnsonii M19 isolated from hospital sewage carries a novel transposon Tn6618 containing carbapenem resistant gene bla OXA-23, while Shewanella xiamenensis T17 carries the novel transposon Tn6297 encoding OXA-416 (Yousfi et al., 2017; Zong et al., 2020). P. asiatica, a newly proposed unique species of the genus Pseudomonas, belongs to the Pseudomonas putida group, which is a potential human pathogen that can cause nosocomial illness (Tohya et al., 2019a). Moreover, the most prevalent carbapenem resistance gene in the genome of clinical isolates of P. aeruginosa is the bla VIM-2 Metal -β -lactamase (MβL) gene, which is usually present in part of the cassette repertoire of class 1 integrons/transposons (Botelho et al., 2018). The bla VIM-2 gene has been found in P. asiatica (Brovedan et al., 2021; Tohya et al., 2021), indicating that it is an important reservoir of this gene. Here, we describe a novel plasmid that co-harbors the tigecycline association resistance gene tmexCD3-toprJ1b, a bla VIM-2-carrying novel transposon Tn7389, as well as bla OXA-10-carrying novel transposon Tn7493 from a Pseudomonas asiatica strain. ## Bacterial isolation and identification Wastewater samples were collected from a large tertiary hospital in Luzhou in August 2019. The sewage samples were collected from outflow of the sewage treatment stations of hospital. The samples were collected in sterile glass bottles (200ml) at a set time each time. Sewage samples were mixed and diluted with sterile water in a ratio of 1:10 and subsequently inoculated on a MacConkey agar plate at 37°C for 18-24h in the presence of antibiotics: meropenem (0.5 mg/L). One strain, named A28, was isolated and purified three times on Luria-Bertani (LB) broth agar medium following the repeated plate streaking method. The species was identified by detecting the 16S rRNA gene with universal primers 27F (5′-AGA GTT TGA TYM TGG CTC AG-3′) and 1492R (5′-GGY TAC CTT GTT ACG ACT T-3′), and further confirmed by WGS analysis (Smyth et al., 2020). ## Antimicrobial susceptibility test The minimal inhibitory concentrations (MICs) of A28 to antimicrobial agents were determined by broth microdilution method according to the recommendations of the CLSI 2021 breakpoints. Escherichia coli strain ATCC 25922 was used as quality control. ## Whole-genome sequencing and analysis The whole genome of strain A28 isolate was sequenced using Oxford Nanopore Technologies. The species was identified using JSpecies (http://jspecies.ribohost.com/jspeciesws/#analyse). ARGs were identified using ResFinder v.4.1 (https://cge.cbs.dtu.dk/services/ResFinder). MLST (Multi-Locus Sequence Typing) v.2.0 (https://cge.cbs.dtu.dk/services/MLST/) was used to determine the STs of the strain. RAST server v.2.0 (https://rast.nmpdr.org/rast.cgi) was used for genome annotation. The circular map of plasmids was generated using the BLAST Ring Image Generator (BRIG) tool and compared to highly similar plasmids in the NCBI database. The Transposon Registry assigned a name to the novel transposon (https://transposon.lstmed.ac.uk/). ## Conjugation assay and fitness cost of plasmid carriage Conjugation assays were carried out using sodium azide-resistant E. coli J53, rifampicin-resistant E. coli EC600 (Rifr), and rifampicin-resistant P. aeruginosa PAO1 as recipients. Transconjugants were selected on LB agar plates containing meropenem (0.5 mg/L) and sodium azide (100 mg/L) or rifampicin (100 mg/L). The donor and recipient strains were mixed in ratios of 1:1, then cultured overnight on LB agar plates at 37°C.The resistance genes of bla VIM-2 in transconjugants were validated by PCR. A growth curve assay was used to calculate the fitness of the plasmid between P. aeruginosa transconjugants and P. aeruginosa PAO1 (Zhang et al., 2022). Overnight cultures were diluted 1:50 in LB without antibiotics and measured at OD600 every 15 minutes for 11 hours on a Synergy H1 (Labsystems) instrument, with each sample repeated three times. Student’s t-test was used for statistical analysis, with a significance threshold of $95\%$ ($P \leq 0.05$). ## Biofilm formation The ability of the transconjugant and wild-type strain to generate biofilms was determined using crystal violet staining (Ding et al., 2021). The bacterial suspension was discarded and washed three times with sterile water after standing culture at 37°C for 24 hours. Crystal violet was dissolved in a $33\%$ acetic acid solution, and its OD595 value was determined. ## Galleria mellonella killing assay By using serial dilutions, the transconjugant PAO1-A28 and P. aeruginosa PAO1 were divided into two different amounts of bacterial suspensions ranging from 1×10 5 c.f.u. ml −1 to 1×10 6 c.f.u. ml −1. Using a microsyringe, 10µl of the prepared bacterial suspensions were injected into the body cavity of G. mellonella through the right hind foot. The control group was injected with 10 µl PBS buffer. Ten G. mellonella were injected with bacteria in each group and placed in a Petri dish at 37° C for 72 hours. At 12-hour intervals, G. mellonella was observed to survive. ## Characterization of the strain P. asiatica A28 Bacterium A28 was identified as P. asiatica and was resistant to meropenem, imipenem, tigecycline, gentamicin, ceftazidime, aztreonam, and ciprofloxacin, but susceptible to polymyxin and tetracycline (Table 1). The genome of P. asiatica strain A28 was assembled into two complete circularized contigs, one chromosome (5824126 bp, CP063456.1) with GC content $62.51\%$ and one plasmid PLA28.4 (199972 bp, CP063457.1) with GC content $56.36\%$. Species identification with ANI analysis confirmed that the strain A28 belonged to P. asiatica, A28 and had a $98.75\%$ identity ($89.30\%$ query coverage) to P. asiatica RYU5 strain (accession: SAMN05581751) (Tohya et al., 2019b). MLST analysis revealed that the ST of strain A28 was ST15. Table 1 \| Strain \| MIC (mg/L) \| MIC (mg/L).1 \| MIC (mg/L).2 \| MIC (mg/L).3 \| MIC (mg/L).4 \| MIC (mg/L).5 \| MIC (mg/L).6 \| MIC (mg/L).7 \| MIC (mg/L).8 \| MIC (mg/L).9 \| MIC (mg/L).10 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Strain \| C \| Gn \| CAZ \| TET \| AZT \| IPM \| MEM \| CIP \| CTX \| TIP \| PB \| \| P. asiatica A28 \| 64 \| 64 \| 256 \| 4 \| 32 \| 32 \| 32 \| 4 \| >128 \| 4 \| 0.25 \| \| P. aeruginosa PAO1 \| 32 \| 4 \| 256 \| >512 \| 8 \| 8 \| 1 \| <0.25 \| 16 \| 32 \| 1 \| \| Conjugant PAO1-A28 \| 32 \| 32 \| 256 \| >512 \| 8 \| 16 \| 4 \| 2 \| 16 \| 32 \| 0.5 \| ## Characterization of plasmid PLA28.4 Plasmid PLA28.4 is a 199,972 bp circular plasmid with 233 predicted open reading frames. PLA28.4 does, however, feature a putative replication initiator protein RepA (encoded by bp16,426 to 17,292) that has $100\%$ cover and $93.43\%$ amino acid sequence similarity to RepA from the IncP-7 plasmid pCAR1 (GenBank accession number AB088420.1) in P. resinovorans (Maeda et al., 2003). ParA (encoded by bp 18444 to 18923) and ParB (encoded by bp 19123 to 20256) are partitioning proteins that are $80.62\%$ to $97.07\%$ similar to the partition proteins of the IncP-7 plasmid pCAR1. Besides, plasmid PLA28.4 carried 7 resistance genes, including bla VIM-2, bla OXA-10, aac(6’)-Ib3, aph(3’)-I, sul1, aac(6’)-Ib-cr) and the RND-type efflux pump gene cluster tnfxB3-tmexC3.2-tmexD3b-toprJ1b (Figure 1). Figure 1: Comparative structural analysis of PLA28.4 with other similar plasmids available in the NCBI nr database. Starting from the center: (1) GC content of PLA28.4 with an average of 56.36%. (2) GC skew, with a positive GC skew toward the inside and a negative GC skew toward the outside. (3) The reference plasmid PLA28.4 plasmid sequence (CP063457). (4) Plasmid pNK546b (MN583270). (5) Plasmid pCAR1.3 (AP013069). (6) Plasmid pCAR1 (AB088420). (7) Plasmid pCAR1.2 (AB474758). (8) Plasmid unnamed (CP034538). (8) Plasmid pZXPA-20-602k (CP061724). (9) Gene annotation. The Figure was constructed using BRIG. Sequence analysis showed that PLA28.4 was closely related to the IncP-7 plasmid Pnk546b (GenBank accession number MN583270.1) (Li et al., 2020) with a query coverage rate of $50\%$ and identification rate of $84.74\%$. Additionally, PLA28.4 shares a similar plasmid backbone with the IncP-7-type plasmid pCAR1.3 (GenBank accession number AP013069.1) and pCAR1 (Shintani et al., 2006) from Pseudomonas resinovorans, and an unnamed plasmid (GenBank accession number CP034538.1) from Pseudomonas poae. Including the replication/partition region repA-parW-parA as well as the conjugative transfer system (consisting of traNDLEKBACWUF mobility genes), indicating that the plasmid PLA28.4 is conjugative. Moreover, plasmid PLA28.4 had $23\%$ sequence coverage and $99.27\%$ identity with the megaplasmid pZXPA-20-602k (GenBank accession number: CP061724.1) from P. putida, which has both bla VIM-2 and multidrug resistance efflux pump TmexCD1-ToprJ1-like gene cluster (Li et al., 2021). ## Identification of the novel transposon Tn7389 Tn7389 is a new transposon with a 7113 bp backbone and three accessory modules. A complete Tn402-like tni module showed $99.98\%$ nucleotide sequence similarity with the genes for transposase (tniA), transposase helper proteins (tniB, tniQ) and decomposition enzymes (tniC) of Klebsiella aerogenes Tn5090 (Encoding a consistent sequence of corresponding proteins). The 5’ CS of Tn7389 is an incomplete class 1 integron that lacks the 3’ CS and contains the antibiotic resistance gene cassette(aacA4-bla VIM-2) and lacks the 3’ CS (Figure 2). Tn7389 differs from the In1701 gene cassettes found on P. aeruginosa DMC-27B, and their integrase is one base inconsistent (Jahan et al., 2020). Tn7389 has two resistance genes, bla VIM-2 and aacA4, but In1701 only has one carbapenem resistance gene, bla VIM-5. Tn402-like transposons Tn6635 and Tn6636 harboring bla VIM-2 were also discovered in two P. asiatica strains, and these two transposons carried the same entire Tn402-like tni module, but only the tniA gene had one base mutation (G409A) compared to Tn7389 (Brovedan et al., 2021). Tn7389 has the same structure as the Tn6017 transposon found in P. aeruginosa and P. putida isolated from a Spanish hospital (Juan et al., 2010). However, the similarity of tni modules is only $86.36\%$. Tn7389 displayed an inconsistent arrangement of resistance genes on the gene cassettes compared to the Tn402-like transposon on the plasmid of P. asiatica LD209 (Marchiaro et al., 2014). Compared to the megaplasmid PZXPA-20-602K, Tn7389’s variable region (VR) lacks the dhfrIIc gene, whereas the Tn5090-like transposon of PZXPA-20-602K has a complete type 1 integron 3’ CS region with a size of more than 46 kbp (Li et al., 2021). Figure 2: Genetic environment of the novel Tn402-like transposon Tn7389 in P. asiatica A28. The construction of sequence comparison was performed using BLAST (http://blast.ncbi.nlm.nih.gov). Green arrows, integrases of a class of integrons; Light blue arrow: Tn402 tni module; red arrows, antibiotic resistance genes; purple arrows, Tn5563a-like genes; gray arrows, hypothetical protein. ## Identification of the novel transposon Tn7493 The bla OXA-10 gene locates within a compound Tn1403-like transposon of 20.24 kp length, flanked by a complete 38-bp IR of Tn1403 and an incomplete 30-bp IR of Tn1403, and was named Tn7493 (Figure 3). Two cassettes, aacA4-bla OXA-10, encoding resistance to aminoglycosides and oxacillinase, were found in the class 1 integron. Upstream gene cassettes were 5’ CS of intI1 and IRi, flanked by tnpAR and 38-bp IR of transposon Tn1403 (Stokes et al., 2007), and tnpR, 39-bp-long IRs of Tn5563a. Downstream of aacA4-bla OXA-10 was sul1-type 3’ -CS, orf5-hp, and IRt, almost identical to the transposon Tn6217 reported from P. aeruginosa (Xiong et al., 2013). On the flanks of IRt were two reverse insertion sequences, IS26, with an aph(3’)-I gene in the middle. The 3’ CS is a truncated transposon Tn5563a that contains a mercury resistance operon (merPTR) (Szuplewska et al., 2014), without the 3’ CS of Tn1403 and Tn5393. Figure 3: Genetic environment of the novel Tn1403-like transposon Tn7493 in P. asiatica A28. The extents and directions of genes are shown by arrows labeled with gene names. The construction of sequence comparison was performed using BLAST (http://blast.ncbi.nlm.nih.gov). ## Identified the tmexC3.2- tmexD3b-toprJ1b in P. asiatica A28 Two identical RND-type efflux pump fragments tnfxB2-tmexC3.2-tmexD3b-toprJ1b coexist in the chromosome and plasmid PLA28.4 of P. asiatica A28 (Figure 4). The tnfxB2-tmexC3.2- tmexD3-toprJ1b fragment was $100\%$ identical to the cluster found in other six Pseudomonas spp. from Homo sapiens. ( GenBank accession no. CP045554.1, CP039989.1, CP017073.1, CP064948.1, CP064945.1, CP062218.1) and $99.98\%$ identical (one nucleotide substitution) to another cluster found in P. putida megaplasmid pZXPA-20-602k (GenBank accession number: CP061724.1) (from a migratory bird, Zhejiang, China) (Li et al., 2021). Like the tnfxB2-tmexCD1-toprJ1 cluster of K. pneumoniae AHM7C8I (Lv et al., 2020) (GenBank accession number: MK347425.1), tnfxB3 - tmexC3.2-tmexD3b-toprJ1b is adjacent to recF (encoding AAA family ATPase), two hypothetical genes (hp1 and hp2), and two site-specific integrase genes (int1 and int2). Of these, recF has a single base substitution (A2283G), and hp2 has one base substitution (G1820T). P. asiatica A28 had $100\%$ similarity with the tnfxB3-tmexC3.2-tmexD3b-toprJ1b-recF-hp1-hp2-int1-int2 structure of P. aeruginosa (GenBank accession no. CP039989.1) and P. putida (GenBank accession no. CP062218.1). Figure 4: The genetic context of the multidrug resistant efflux pump tnfxB3-tmexC3.2-tmexD3b-toprJ1b. The extents and directions of genes are shown by arrows labeled with gene names. Black arrows, tnfxB1-tmexCD1-toprJ1-like gene clusters; pink arrows, int and int-like genes, predicted to encode site-specific integrases; blue arrows, umuC and umuD; green arrows, mobile related genes; red arrows, antibiotic resistance genes; yellow arrows, mercury resistance genes; gray arrows, hypothetical protein. Regions of homology between 96% and 100% are shaded. ## Conjugation assay, fitness cost, biofilm formation, and G. mellonella killing assay The plasmid PLA28.4 could not be transferred to the recipient cell E. coli J53/C600 by conjugation but could be transferred to P. aeruginosa PAO1. The transfer frequency of PLA28.4 was (2.039±0.077) × 10-8 per recipient. Consequently, we evaluated the effect of acquiring resistance plasmids on biological fitness and observed significant differences in growth rate related to plasmid acquisition in P. aeruginosa PAO1 from 4h-12h ($P \leq 0.0001$, Figure 5A). Biofilm formation was significantly reduced in the transconjugant strain ($P \leq 0.05$) (Figure 5B). We examined the susceptibility of G. mellonella to the transconjugant PAO1-A28 and P. aeruginosa PAO1, which were injected with 1×10 5 c.f.u. ml −1 to 1×10 6 c.f.u. ml −1 of the strains and incubated in the dark at 37°C for up to 72 h. As shown in Figures 5C, D, compared with PAO1, the transconjugant PAO1-A28 showed significantly reduced virulence against G. mellonella ($P \leq 0.05$). The decreased virulence of transconjugant to G. mellonella might be due to the adaptive cost of plasmids. Figure 5: Fitness costs and stability of PLA28.4 in strain P. aeruginosa PAO1. (A) Growth curve of the transconjugant and recipient PAO1. (B) Biofilm formation of the transconjugant and recipient PAO1. (C, D) Survival of G. mellonella following infection with the transconjugant and recipient PAO1. Statistically significant (p < 0.05), statistically significant (p < 0.01), and statistically significant (p < 0.001). ## Discussion As an important reservoir of ARB and ARG, hospital sewage is an important medium for ARG to spread to other environments. In this study, a tigecycline and carbapenem-resistant culture obtained from hospital sewage belonged to P. asiatica ST15, which is a newly proposed unique species of the genus Pseudomonas, belongs to the Pseudomonas putida group (Tohya et al., 2020). Sequencing analysis revealed that it coharboring carrying a tmexCD3-toprJ1b, a novel Tn5090-like transposon Tn7389 harboring bla VIM-2, and a Tn1403-like transposon Tn7493 harboring bla OXA-10. Tn5090 (also known as Tn402) was discovered on IncP-7 plasmid R751 from K. aerogenes in 1994 (Rådström et al., 1994). In Tn7389, two 25-bp initial reverse repeat (IRi) and terminal reverse repeat (IRt) of Tn5090/Tn402 transposon families were located 171 bp downstream of intI1 and 116bp upstream of tniA, respectively, suggesting that the bla VIM-2 could be mobilized using the tni machinery. The integrase and recombination sites containing class 1 integrons can be inserted and removed in the form of gene cassettes at attI1 (Toleman and Walsh, 2011). Multiple Tn5090-like transposons carrying bla VIM-2 have been found in Pseudomonas in a growing number of investigations, suggesting that Tn5090-like transposons are key mobile components of VIM-2 transmission in Pseudomonas (Santos et al., 2010). The bla VIM-2 gene could be mobile via the tni mechanism, which may promote its transmission among other pathogens in the hospital sewage environment and requires closer monitoring. Tn1403 was discovered on RPL11 plasmids from clinical P. aeruginosa isolates expressing resistance to ampicillin, streptomycin, puromycin, and chloramphenicol (Vézina and Levesque, 1991). Tn1403-like transposons have been found primarily in Pseudomonas spp. and have been shown to carry diverse types of ARGs, suggesting that they may play an important role in ARG and metal resistance gene transmission in Pseudomonas. In addition, disinfectant-sulfanilamide resistance (qacEΔ1-sul1) genes cause bacterial resistance to chlorine-containing disinfectants and allows bacteria to survive in disinfected water, which poses a threat to health care systems. Although there are different variants of the MDR efflux pumps tmexCD1-toprJ1, similar structures have also been found in Aeromonas caviae, Raoultella planticola, and Klebsiella quasipneumoniae, suggesting potential horizontal transfer mechanisms among various species (Wang et al., 2021a; Dong et al., 2022; Gao et al., 2022). The transfer of tnfxB2-tmexCD1-toprJ1 has previously been found to be mobilized by site-specific integrase (Lv et al., 2020). However, it could be linked to umuCD, a neighboring mutant DNA repair system, because integrase can accelerate the excision and integration of umuCD (Peng et al., 2021). The proximity of umuCD to the efflux pump structure in various bacteria revealed that it might help spread tmexCD1-TopRJ1-like gene clusters. The IncP-7 plasmid is a conjugative transfer plasmid with a narrow host range (Shintani et al., 2010). Although most reports suggest that IncP-7 plasmids could only be transmitted in Pseudomonas (Xiong et al., 2013), pCAR1 was discovered to be transferable to Sterotrophomonas-like strains in natural water (Shintani et al., 2008). Moreover, the IncP-7 type plasmid pNK546b in P. aeruginosa NK546 also assisted the transmission of another resistant plasmid pNK546a that could not be self-transmissible (Li et al., 2020). In this study, the IncP-7 plasmid PLA28.4 of P. asiatica could be transferred to P. aeruginosa PAO1, suggesting PLA28.4 has the capacity to transmit numerous resistance genes in hospital sewage, according to this study. Collectively, plasmid fitness cost studies found that transferring the PLA28.4 plasmid into P. aeruginosa PAO1 resulted in a lower growth rate, less biofilm generation, and lower pathogenicity, demonstrating that transmission of the PLA28.4 plasmid caused bacteria to pay a cost of adaptation. We discovered a P. asiatica carrying a plasmid containing the tmexCD1-toprJ1-like gene cluster, and two novel transposons carrying bla VIM-2 and bla OXA-10, respectively. Controlling the development and spread of medication resistance requires continuous monitoring and management of resistant microorganisms in hospital sewage. ## Conclusion We discovered a P. asiatica carrying a plasmid containing the tmexCD1-toprJ1-like gene cluster, and two novel transposon carrying bla VIM-2 and bla OXA-10, respectively. Controlling the development and spread of medication resistance requires continuous monitoring and management of resistant microorganisms in hospital sewage. ## Data availability statement The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ncbi.nlm.nih.gov/genbank/, CP063456.1. ## Author contributions SL, WW, BYZ and QHL collected the data. FYZ and LG performed the bioinformatic analyses. QL, QC, AM-R wrote the initial draft of the manuscript. QL, RJH, LX and YSZ conceived the project, reviewed the articles and extracted the data. XX contributed to the revision of this article. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## References 1. Aghapour Z., Gholizadeh P., Ganbarov K., Bialvaei A. Z., Mahmood S. S., Tanomand A.. Molecular mechanisms related to colistin resistance in enterobacteriaceae. Infection Drug resistance. (2019) 12 965-975. DOI: 10.2147/idr.S199844 2. Alavi M. R., Antonic V., Ravizee A., Weina P. J., Izadjoo M., Stojadinovic A.. An enterobacter plasmid as a new genetic background for the transposon Tn1331. Infection Drug resistance (2011) 4 209-213. DOI: 10.2147/idr.S25408 3. Botelho J., Roberts A. P., León-Sampedro R., Grosso F., Peixe L.. Carbapenemases on the move: It's good to be on ICEs. Mobile DNA (2018) 9 37. DOI: 10.1186/s13100-018-0141-4 4. Brovedan M. A., Marchiaro P. M., Díaz M. S., Faccone D., Corso A., Pasteran F.. Pseudomonas putida group species as reservoirs of mobilizable Tn402-like class 1 integrons carrying bla(VIM-2) metallo-β-lactamase genes. Infection Genet. Evol. J. Mol. Epidemiol. evolutionary Genet. Infect. Dis. (2021) 96. DOI: 10.1016/j.meegid.2021.105131 5. Ding M., Shi J., Ud Din A., Liu Y., Zhang F., Yan X.. Co-Infections of two carbapenemase-producing enterobacter hormaechei clinical strains isolated from the same diabetes individual in China. J. Med. Microbiol. (2021) 2021 70(3). DOI: 10.1099/jmm.0.001316 6. Dong N., Zeng Y., Wang Y., Liu C., Lu J., Cai C.. Distribution and spread of the mobilised RND efflux pump gene cluster tmexCD-toprJ in clinical gram-negative bacteria: A molecular epidemiological study. Lancet Microbe (2022) 2022. DOI: 10.1016/s2666-5247(22)00221-x 7. Gao X., Wang C., Lv L., He X., Cai Z., He W.. Emergence of a novel plasmid-mediated tigecycline resistance gene cluster, tmexCD4-toprJ4, in klebsiella quasipneumoniae and enterobacter roggenkampii. Microbiol. Spectr. (2022) 10. DOI: 10.1128/spectrum.01094-22 8. Hu R., Li Q., Zhang F., Ding M., Liu J., Zhou Y.. Characterisation of bla(NDM-5) and bla(KPC-2) co-occurrence in K64-ST11 carbapenem-resistant klebsiella pneumoniae. J. Glob Antimicrob. Resist. (2021) 27 63-66. DOI: 10.1016/j.jgar.2021.08.009 9. Huang J., Yi M., Yuan Y., Xia P., Yang B., Liao J.. Emergence of a fatal ST11-KL64 tigecycline-resistant hypervirulent klebsiella pneumoniae clone cocarrying bla(NDM) and bla(KPC) in plasmids. Microbiol. Spectr. (2022) 2022. DOI: 10.1128/spectrum.02539-22 10. Jahan M. I., Rahaman M. M., Hossain M. A., Sultana M.. Occurrence of intI1-associated VIM-5 carbapenemase and co-existence of all four classes of β-lactamase in carbapenem-resistant clinical pseudomonas aeruginosa DMC-27b. J. antimicrobial chemotherapy. (2020) 75 86-91. DOI: 10.1093/jac/dkz426 11. Jo J., Ko K. S.. Tigecycline heteroresistance and resistance mechanism in clinical isolates of acinetobacter baumannii. Microbiol. Spectr. (2021) 9. DOI: 10.1128/Spectrum.01010-21 12. Juan C., Zamorano L., Mena A., Albertí S., Pérez J. L., Oliver A.. Metallo-beta-lactamase-producing pseudomonas putida as a reservoir of multidrug resistance elements that can be transferred to successful pseudomonas aeruginosa clones. J. antimicrobial chemotherapy. (2010) 2010;65 474-478. DOI: 10.1093/jac/dkp491 13. Li Z., Cai Z., Cai Z., Zhang Y., Fu T., Jin Y.. Molecular genetic analysis of an XDR pseudomonas aeruginosa ST664 clone carrying multiple conjugal plasmids. J. antimicrobial chemotherapy. (2020) 75 1443-1452. DOI: 10.1093/jac/dkaa063 14. Li R., Peng K., Xiao X., Liu Y., Peng D., Wang Z.. Emergence of a multidrug resistance efflux pump with carbapenem resistance gene blaVIM-2 in a pseudomonas putida megaplasmid of migratory bird origin. J. antimicrobial chemotherapy (2021) 76 1455-1458. DOI: 10.1093/jac/dkab044 15. Li Y., Qiu Y., Gao Y., Chen W., Li C., Dai X.. Genetic and virulence characteristics of a raoultella planticola isolate resistant to carbapenem and tigecycline. Sci. Rep. (2022) 12 3858. DOI: 10.1038/s41598-022-07778-0 16. Lv L., Wan M., Wang C., Gao X., Yang Q., Partridge S. R.. Emergence of a plasmid-encoded resistance-Nodulation-Division efflux pump conferring resistance to multiple drugs, including tigecycline, in klebsiella pneumoniae. mBio (2020) 2020 11(2). DOI: 10.1128/mBio.02930-19 17. Maeda K., Nojiri H., Shintani M., Yoshida T., Habe H., Omori T.. Complete nucleotide sequence of carbazole/dioxin-degrading plasmid pCAR1 in pseudomonas resinovorans strain CA10 indicates its mosaicity and the presence of large catabolic transposon Tn4676. J. Mol. Biol. (2003) 326 21-33. DOI: 10.1016/s0022-2836(02)01400-6 18. Mann R., Rafei R., Gunawan C., Harmer C. J., Hamidian M.. Variants of Tn6924, a novel Tn7 family transposon carrying the bla(NDM) metallo-β-Lactamase and 14 copies of the aphA6 amikacin resistance genes found in acinetobacter baumannii. Microbiol. Spectr. (2022) 10. DOI: 10.1128/spectrum.01745-21 19. Marchiaro P. M., Brambilla L., Morán-Barrio J., Revale S., Pasteran F., Vila A. J.. The complete nucleotide sequence of the carbapenem resistance-conferring conjugative plasmid pLD209 from a pseudomonas putida clinical strain reveals a chimeric design formed by modules derived from both environmental and clinical bacteria. Antimicrobial Agents chemotherapy. (2014) 58 1816-1821. DOI: 10.1128/aac.02494-13 20. Peng K., Wang Q., Yin Y., Li Y., Liu Y., Wang M.. Plasmids shape the current prevalence of tmexCD1-toprJ1 among klebsiella pneumoniae in food production chains. mSystems. (2021) 6. DOI: 10.1128/mSystems.00702-21 21. Rådström P., Sköld O., Swedberg G., Flensburg J., Roy P. H., Sundström L.. Transposon Tn5090 of plasmid R751, which carries an integron, is related to Tn7, mu, and the retroelements. J. bacteriology. (1994) 176 3257-3268. DOI: 10.1128/jb.176.11.3257-3268.1994 22. Santos C., Caetano T., Ferreira S., Mendo S.. Tn5090-like class 1 integron carrying bla(VIM-2) in a pseudomonas putida strain from Portugal. Clin. Microbiol. Infect. (2010) 16 1558-1561. DOI: 10.1111/j.1469-0691.2010.03165.x 23. Shintani M., Fukushima N., Tezuka M., Yamane H., Nojiri H.. Conjugative transfer of the IncP-7 carbazole degradative plasmid, pCAR1, in river water samples. Biotechnol. letters. (2008) 2008;30 117-122. DOI: 10.1007/s10529-007-9519-y 24. Shintani M., Takahashi Y., Yamane H., Nojiri H.. The behavior and significance of degradative plasmids belonging to inc groups in pseudomonas within natural environments and microcosms. Microbes environments. (2010) 2010;25 253-265. DOI: 10.1264/jsme2.me10155 25. Shintani M., Yano H., Habe H., Omori T., Yamane H., Tsuda M.. Characterization of the replication, maintenance, and transfer features of the IncP-7 plasmid pCAR1, which carries genes involved in carbazole and dioxin degradation. Appl. Environ. Microbiol. (2006) 72 3206-3216. DOI: 10.1128/aem.72.5.3206-3216.2006 26. Smyth C., O'Flaherty A., Walsh F., Do T. T.. Antibiotic resistant and extended-spectrum β-lactamase producing faecal coliforms in wastewater treatment plant effluent. Environ. pollut. (Barking Essex 1987) (2020) 262. DOI: 10.1016/j.envpol.2020.114244 27. Stokes H. W., Elbourne L. D., Hall R. M.. Tn1403, a multiple-antibiotic resistance transposon made up of three distinct transposons. Antimicrobial Agents chemotherapy. (2007) 51 1827-1829. DOI: 10.1128/aac.01279-06 28. Szuplewska M., Ludwiczak M., Lyzwa K., Czarnecki J., Bartosik D.. Mobility and generation of mosaic non-autonomous transposons by Tn3-derived inverted-repeat miniature elements (TIMEs). PloS One (2014) 9. DOI: 10.1371/journal.pone.0105010 29. Tohya M., Tada T., Watanabe S., Kuwahara-Arai K., Zin K. N., Zaw N. N.. ) emergence of carbapenem-resistant pseudomonas asiatica producing NDM-1 and VIM-2 metallo-β-Lactamases in Myanmar. Antimicrobial Agents chemotherapy (2019) 2019 63(8). DOI: 10.1128/aac.00475-19 30. Tohya M., Uechi K., Tada T., Hishinuma T., Kinjo T., Ohshiro T.. Emergence of clinical isolates of pseudomonas asiatica and pseudomonas monteilii from Japan harbouring an acquired gene encoding a carbapenemase VIM-2. J. Med. Microbiol. (2021) 2021 70. DOI: 10.1099/jmm.0.001258 31. Tohya M., Watanabe S., Tada T., Tin H. H., Kirikae T.. Genome analysis-based reclassification of pseudomonas fuscovaginae and pseudomonas shirazica as later heterotypic synonyms of pseudomonas asplenii and pseudomonas asiatica, respectively. Int. J. systematic evolutionary Microbiol. (2020) 70 3547-3552. DOI: 10.1099/ijsem.0.004199 32. Tohya M., Watanabe S., Teramoto K., Uechi K., Tada T., Kuwahara-Arai K.. Pseudomonas asiatica sp. nov., isolated from hospitalized patients in Japan and Myanmar. Int. J. systematic evolutionary Microbiol. (2019) 69 1361-1368. DOI: 10.1099/ijsem.0.003316 33. Toleman M. A., Walsh T. R.. Combinatorial events of insertion sequences and ICE in gram-negative bacteria. FEMS Microbiol. Rev. (2011) 35 912-935. DOI: 10.1111/j.1574-6976.2011.00294.x 34. Vézina G., Levesque R. C.. Molecular characterization of the class II multiresistance transposable element Tn1403 from pseudomonas aeruginosa. Antimicrobial Agents chemotherapy. (1991) 35 313-321. DOI: 10.1128/aac.35.2.313 35. Wang C. Z., Gao X., Yang Q. W., Lv L. C., Wan M., Yang J.. ) a novel transferable resistance-Nodulation-Division pump gene cluster, tmexCD2-toprJ2, confers tigecycline resistance in raoultella ornithinolytica. Antimicrobial Agents chemotherapy (2021) 2021 65(4). DOI: 10.1128/aac.02229-20 36. Wang Q., Peng K., Liu Y., Xiao X., Wang Z., Li R.. Characterization of TMexCD3-TOprJ3, an RND-type efflux system conferring resistance to tigecycline in Proteus mirabilis, and its associated integrative conjugative element. Antimicrobial Agents chemotherapy (2021) 65. DOI: 10.1128/aac.02712-20 37. Xiong J., Alexander D. C., Ma J. H., Déraspe M., Low D. E., Jamieson F. B.. Complete sequence of pOZ176, a 500-kilobase IncP-2 plasmid encoding IMP-9-mediated carbapenem resistance, from outbreak isolate pseudomonas aeruginosa 96. Antimicrobial Agents chemotherapy. (2013) 57 3775-3782. DOI: 10.1128/aac.00423-13 38. Yousfi K., Touati A., Lefebvre B., Fournier É, Côté J. C., Soualhine H.. A novel plasmid, pSx1, harboring a new Tn1696 derivative from extensively drug-resistant shewanella xiamenensis encoding OXA-416. Microbial Drug resistance (Larchmont NY) (2017) 23 429-436. DOI: 10.1089/mdr.2016.0025 39. Zhang B., Hu R., Liang Q., Liang S., Li Q., Bai J.. Comparison of two distinct subpopulations of klebsiella pneumoniae ST16 Co-occurring in a single patient. Microbiol. Spectr. (2022) 2022. DOI: 10.1128/spectrum.02624-21 40. Zong G., Zhong C., Fu J., Zhang Y., Zhang P., Zhang W.. The carbapenem resistance gene bla(OXA-23) is disseminated by a conjugative plasmid containing the novel transposon Tn6681 in acinetobacter johnsonii M19. Antimicrobial resistance infection control (2020) 9 182. DOI: 10.1186/s13756-020-00832-4
--- title: Exploring the potential of artificial intelligence in improving skin lesion diagnosis in primary care authors: - Anna Escalé-Besa - Oriol Yélamos - Josep Vidal-Alaball - Aïna Fuster-Casanovas - Queralt Miró Catalina - Alexander Börve - Ricardo Ander-Egg Aguilar - Xavier Fustà-Novell - Xavier Cubiró - Mireia Esquius Rafat - Cristina López-Sanchez - Francesc X. Marin-Gomez journal: Scientific Reports year: 2023 pmcid: PMC10015524 doi: 10.1038/s41598-023-31340-1 license: CC BY 4.0 --- # Exploring the potential of artificial intelligence in improving skin lesion diagnosis in primary care ## Abstract Dermatological conditions are a relevant health problem. Machine learning (ML) models are increasingly being applied to dermatology as a diagnostic decision support tool using image analysis, especially for skin cancer detection and disease classification. The objective of this study was to perform a prospective validation of an image analysis ML model, which is capable of screening 44 skin diseases, comparing its diagnostic accuracy with that of General Practitioners (GPs) and teledermatology (TD) dermatologists in a real-life setting. Prospective, diagnostic accuracy study including 100 consecutive patients with a skin problem who visited a participating GP in central Catalonia, Spain, between June 2021 and October 2021. The skin issue was first assessed by the GPs. Then an anonymised skin disease picture was taken and uploaded to the ML application, which returned a list with the Top-5 possible diagnosis in order of probability. The same image was then sent to a dermatologist via TD for diagnosis, as per clinical practice. The GPs Top-3, ML model’s Top-5 and dermatologist’s Top-3 assessments were compared to calculate the accuracy, sensitivity, specificity and diagnostic accuracy of the ML models. The overall Top-1 accuracy of the ML model ($39\%$) was lower than that of GPs ($64\%$) and dermatologists ($72\%$). When the analysis was limited to the diagnoses on which the algorithm had been explicitly trained ($$n = 82$$), the balanced Top-1 accuracy of the ML model increased ($48\%$) and in the Top-3 ($75\%$) was comparable to the GPs Top-3 accuracy ($76\%$). The Top-5 accuracy of the ML model ($89\%$) was comparable to the dermatologist Top-3 accuracy ($90\%$). For the different diseases, the sensitivity of the model (Top-3 $87\%$ and Top-5 $96\%$) is higher than that of the clinicians (Top-3 GPs $76\%$ and Top-3 dermatologists $84\%$) only in the benign tumour pathology group, being on the other hand the most prevalent category ($$n = 53$$). About the satisfaction of professionals, $92\%$ of the GPs considered it as a useful diagnostic support tool (DST) for the differential diagnosis and in $60\%$ of the cases as an aid in the final diagnosis of the skin lesion. The overall diagnostic accuracy of the model in this study, under real-life conditions, is lower than that of both GPs and dermatologists. This result aligns with the findings of few existing prospective studies conducted under real-life conditions. The outcomes emphasize the significance of involving clinicians in the training of the model and the capability of ML models to assist GPs, particularly in differential diagnosis. Nevertheless, external testing in real-life conditions is crucial for data validation and regulation of these AI diagnostic models before they can be used in primary care. ## Introduction Skin diseases are one of the main reasons for consultation in Primary Care (PC)1. To give an example, in the United States, each person has on average, 1.6 skin diseases per year1–3. Approximately $7.6\%$ of the population of Catalonia consults PC annually for skin lesions4, generating $35\%$ of referrals to dermatology5. However, the diagnostic accuracy of general practitioners in dermatological diseases is highly variable, around 48–$77\%$6,7. TD involves storing and transmitting photographs of skin lesions and text through the Internet. The use of TD as a consultation tool for dermatology services in PC is now common. It is estimated that more than $70\%$ of all people with a skin problem in PC can be seen by TD and do not need to be referred to an in-person dermatologist8,9. This is a good sorting method, particularly for skin cancer10,11. TD has been shown to avoid unnecessary travel, decrease waiting time, provide diagnostic support at the time of the visit, and increase both user and provider satisfaction9,12–16. The 4th industrial revolution17 and the application of artificial intelligence (AI) in the healthcare field open a door to more efficient, individualised and preventive medicine. There are currently several fields of medicine in which these new technologies help in the management of various diseases, such as screening for diabetic retinopathy, reading radiological images, or assisting during endoscopies, among others18,19. Medical images are the most widely used data format in AI development20. In recent years there has been a substantial improvement in this field, especially applied to the automatic classification of medical images, through deep learning techniques using convolutional neural networks (CNN). In some cases, the performances are comparable to those achieved by medical specialists. In dermatology, ML using image recognition is especially developed in skin cancer screening21–24. More recently, its use has been extended to a wider range of skin lesions, such as inflammatory and infectious lesions25–28, and also in the recognition of cutaneous manifestations of COVID-1929. This suggests that its use in PC as a diagnostic support and screening tool for consultations related to skin problems would standardise and improve the effectiveness and efficiency of the professionals working there. Some of these tools generate a list of differential diagnoses that can help the GP to broaden their range of diagnoses and therapeutic approaches to the assessed lesion. The fact that the algorithm can give 5 diagnoses from a single image means that the clinician can not only arrive at the final diagnosis, but can also consider alternative diagnoses that may condition the follow-up to ensure that the lesion is developing correctly. For example, an inflammatory lesion may lead to a diagnosis of dermatitis, ringworm, pityriasis, psoriasis, neurodermatitis. These entities are different in themselves but for some of which the therapeutic approach is similar. Another example is a warty lesion, which can make the differential diagnosis between a viral wart, but also between other entities such as seborrheic keratosis and pathologies with malignant potential such as actinic keratosis and also carcinomas. However, although diagnostic yields are very high in silico, there have been few studies performed in routine clinical practice settings employing non-standardised imaging, so validation of these tools prospectively in real life is imperative. In Europe, the current governing regulation is the Medical Device Regulation (Regulation $\frac{2017}{745}$)30, which has been in vigour since May 2020 and repeals Directive $\frac{93}{4231.}$ This new regulation introduced new responsibilities for the European Medicines Agency (EMA) and national authorities competent in the evaluation of certain categories of medical devices. The new regulation stipulates that manufacturers ensure that devices meet a number of essential requirements that depend on the potential risk of each device and require accreditation by an independent body. Thus, in the case of the application of ML model as a complementary diagnostic tool, different groups of experts around the world have developed guidelines to stipulate the essential requirements to be assessed in this practice. Several studies agree that prospective studies, such as the present study, are necessary to confirm that the application of these algorithms in clinical practice works, and to evaluate their potential impact32–36. Although it is in PC where most consultations related to skin conditions are first received, there have been few studies performed in this setting. Some studies have included PC GPs along with dermatologists as image readers to compare the performance of the models with that of the professionals37. Other studies have concluded that AI tools could be used in PC, resulting in a new tool for diagnostic support, screening, and to extend differential diagnosis by non-expert professionals37,38. However, this has not been widely studied and the proof is insufficient. Autoderm is a Class I CE marked DST in dermatology which uses ML to help diagnose skin lesions in a faster and more accurate way39. The current model can examine 44 different types of skin diseases, including inflammatory diseases, tumours, and genital skin problems, among others, representing $90\%$ of the consultations made by the general population1,3,4. The model can be accessed through an Application Programming Interface (API) that can be integrated into any platform that is connected to the Internet. After examining a photograph, the model generates a ranking of the five skin diseases that have the highest concordance with the lesion shown in the photo, sorted in order of probability. Autoderm uses a set of 3 neural networks: resnet-18, resnet-3440 and squeezenet41, provided by TorchVision (PyTorch)42, which is used for applications such as computer vision and natural language processing. It was trained with an in-housedataset of 55,364 images in the training set and 13,841 for the test set. As for dermoscopic images, it was only trained with approximately 2000 images obtained from the HüD dermatoscope and other Dermlite dermatoscope models. These images were all taken by the layman or a healthcare worker using a smartphone. Data augmentation methods were used during algorithm training. This consists of modifying images in the training set (orientation, brightness, etc.) so that relevant information is not lost, but allowing the algorithm to be exposed to a more general distribution of data. After the data augmentation process, the number of images increased to approximately 120,000. The theoretical diagnostic accuracy of the model tested is $49.3\%$ (Top-1), $70.1\%$ (Top-3) and $81.7\%$ (Top-5). Subsequently, two clinical studies were conducted with Autoderm with earlier models in Sweden on Caucasian skin, and in Uganda on black skin (skin type 6 on the Fitzpatrick scale)43,44. While some of these points suggest that ML dermatology models can improve efficiency in primary care by reducing unnecessary referrals and speeding up diagnoses, additional studies are required to assess their practical use in clinical practice, as foreseen by the Medical Devices regulation in the European Union. ## Objectives The main objective of the study is the prospective validation of an ML model as a diagnostic decision support tool for skin diseases through a feasibility study in a real PC clinical practice setting in a region of Catalonia, Spain. The secondary objectives are: 1) evaluate the diagnostic accuracy and efficacy of the ML model in a clinical setting to determine the possibility of implementing it in a PC setting; 2) detect which skin lesions are missing in the study model; 3) estimate the rate of patients agreeing to participate in the study with the aim of using these data for future related research, 4) assess the PC professionals’ degree of satisfaction with the use of the artificial intelligence model. ## Methods The study protocol is described in detail in a separate publication45; however, key elements are summarised below. ## Design Prospective multicentre observational feasibility study with 100 consecutive patients who consulted PC for a skin lesion in the area of Central Catalonia. Anonymised photographs of the lesions were taken and entered into the Autoderm model interface to obtain the diagnoses through AI and to be able to evaluate the diagnostic accuracy, sensitivity and specificity with respect to that of the GPs and dermatologists of the two referral hospitals (Fig. 1).Figure 1Study design general practitioner (GP) vs teledermatology (TD) vs artificial intelligence (AI). ## Study population The study was conducted in 6 PC Centres managed by the Institut Català de la Salut (main provider of PC services in Catalonia) in Central Catalonia, specifically in the regions of Bages, Berguedà and Moianès, predominantly rural and semi-rural areas. In addition, eleven GPs were invited to participate, and all accepted. The reference population included in the study was 512,050 inhabitants. Inclusion criteria persons ≥ 18 years old consulting PC for a skin disease and signing the informed consent form. Exclusion criteria individuals with a skin lesion that could not be photographed with a smartphone or who had difficulty understanding and complying with the protocol were excluded from the study. Poor quality images were also excluded. ## Sample size and sampling procedure The sample size and sampling procedure is described in detail in a separate publication45; however, key elements are summarised in the Fig. 2.Figure 2Diagram procedure. GPs: general practitioners; PC: primary care; TD: teledermatology; AI: artificial intelligence. As described in the study procedure (Fig. 2), the GP first made his/her diagnosis (Top-3) and then ran the image through the AI model. Likewise, in the three subjective questions on the use of the tool (Table 5), the GP was asked, whether seeing the results of the model (Top-5), had helped they with the diagnosis or differential diagnosis, or whether it had saved they the need for a teledermatology (TD) consultation. Most of the photographs analysed in the study were taken by the GP during the face-to-face consultation ($$n = 93$$), as Fig. 3. The remaining 7 photos were taken by the patient and sent using the eConsultation system (The Telematic Consultation System is an asynchronous telemedicine service between patients and health professionals, integrated into the computerised information systems of the Catalan public health system)46. It is available to all patients and all primary care professionals. Figure 3Autoderm screenshot. Dermatologist 1's diagnoses are described as TD in the study. The gold standard was defined as agreement between the top 1 diagnosis of Dermatologist 1 (Dermatologist of the reference hospital in the area, which assessed the TD according to the usual clinical model) and Dermatologist 2 (independent Dermatologist, which assessed the 100 cases only seeing the images). If both dermatologists agreed, this was considered the gold standard diagnosis for the case. Otherwise (37 cases in total), a third dermatologist reviewed the images and agreed with one of the diagnoses issued by dermatologist 1 or 2. ## Statistical analysis The proposed sample size is based on the sample size calculation used in similar research and taking into account that it is a pilot study to validate the usefulness of the tool44,47,48. The validation dataset includes 100 cases, and 4 assessments: face-to-face assessment by the GP (Top-3), assessment of the 5 differential diagnoses in order of probability from the ML model (Top-5), TD assessment by dermatologist 1 (Top-3), and assessment from the dermatologist 2 (Top-3). The evaluation of the ML model was limited to 44 types of skin diseases, while other diagnoses could be included in the evaluations of both GPs and dermatologists according to medical criteria (category other). Regarding the five suggested diagnoses, the AI is not precise enough to only present the top three. However, with the top five diagnoses, it is estimated that the conditions are represented $95\%$ of the time43. The AI serves as a search engine or analytics engine to provide differential diagnoses for skin diseases, empowering the GP to make informed decisions. A confusion matrix was used to calculate the accuracy, sensitivity and specificity of the overall ML model and for each skin disease. All statistical analyses were performed with R Core Team [2022]. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/. The confidence intervals were $95\%$. ## Ethical approval Primary care GPs’ assessment and decisions were not influenced by this study, as the normal dermatology referral workflow was not affected. This project was approved by the Research Ethics Committee (REC) from the Foundation University Institute for Primary Health Care Research Jordi Gol i Gurina (IDIAPJGol) (P$\frac{20}{159}$-P) and the REC of the Hospital Sant Bernabé de Berga. A collaboration agreement has been established between the collaborating institutions: IDIAPJGol; Salut Catalunya Central, Hospital de Berga, Althaia, Xarxa Assistencial Universitària de Manresa and the company First Derm (iDoc24 Inc). The study was performed in accordance with relevant guidelines/regulations, and informed consent was obtained from all participants. All research have performed in accordance with the Declaration of Helsinki. ## Description of the sample One hundred cases were analysed for external validation of the ML model. The PC consultations were mostly in person ($93\%$); however, it is noteworthy that in $7\%$ of the cases, the patient chose to send a photograph of the skin lesion and have a virtual PC consultation. The patients included in the study were mostly Fitzpatrick phototype III ($$n = 78$$) and phototype II ($$n = 17$$) (Table 1).Table 1Descriptive characteristics of the cases analysed. PC [n (%)]TD [n (%)]GS [n (%)]DifficultyHigh11 [11]20 [20]0 (0.0)Average36 [36]40 [40]0 (0.0)Low53 [53]40 [40]100 [100]CertaintyYes40 [40]––No60 [60]––Image qualityPoor3 [3]3 [3]58 [58]Average45 [45]9 [9]0 (0.0)Excellent52 [52]88 [88]42 [42]Time10.3 (2.74)6.17 (2.26)-PhototypeI1 [1]––II17 [17]––III78 [78]––IV3 [3]––V1 [1]––OrigineConsulta7 [7]––In person93 [93]––ManagementBiopsy–7 [7]–Excision–7 [7]–Dermatology clinical follow-up–28 [28]–PC clinical follow-up–53 [53]–Dermatological treatment–5 [5]–PC Primary care, TD Teledermatology, GS Gold standard. Variables described by relative frequency and percentage n (%).Minutes. Mean and standard deviation. Variables that were not asked to all groups of professionals have been marked with the symbol. Both dermatologists and GP agreed that most of the cases assessed ($80\%$ and $89\%$, respectively) were of low or moderate difficulty. In $88\%$ of the cases, they considered that the quality of the image taken by the GP and evaluated by the dermatologists who resolved the telematic consultation was excellent. The photos taken by the patients it has to take into account that 4 of the 7 images were of excellent quality and 3 were of poor quality. The time needed to resolve the consultation was also evaluated, and this was higher in the case of PC (10.3 min on average) versus the time taken with TD (6.17 min on average) (Table 1). It has to take into account that the time spent on the GP consultation was estimated by each professional. It included the total time spent on a face-to-face visit. In Catalonia, a typical face-to-face visit is allotted 12 min. It is assumed that this time accounts for deductions for other tasks. However, the time spent on medical history and physical examination, as well as capturing the photo and uploading it to the shared clinical history portal for review by the referral hospital dermatologist, was included. The total of 100 cases produced 36 different diseases or diagnoses (Table 2), of which 12 were not included in the 44 diagnoses analysed by the ML model (Online Appendix, Table 1).Table 2Description of the case studies with GS diagnosis, how many cases were studied and whether they were included in the ML model. Diagnosticsn (%)ML modelAcne vulgaris2 [2]YesAngiokeratoma1 [1]Balanitis1 [1]YesCommon wart4 [4]YesBorrelia1 [1]YesBasal cell carcinoma4 [4]YesCutaneous squamous cell carcinoma2 [2]YesCondyloma (genital wart)1 [1]YesChondrodermatitis nodularis helicis1 [1]Lymphocytic dermatitis1 [1]Unspecified dermatitis1 [1]YesDermatofibroma3 [3]YesDyshidrotic eczema4 [4]Palmar hidradenitis1 [1]Scabies1 [1]Fibroma1 [1]Granuloma annulare4 [4]Haemangioma3 [3]YesHidradenitis1 [1]Lentigo2 [2]YesLichen planus1 [1]YesVascular malformation1 [1]Dysplastic nevus (atypical mole)1 [1]YesMelanoma1 [1]YesNevus (benign mole)10 [10]YesIntradermal nevus10 [10]YesOnychodystrophy1 [1]Onychomycosis1 [1]Post-inflammatory hyperpigmentation1 [1]YesPityriasis versicolor1 [1]YesPityriasis rosea1 [1]YesPsoriasis4 [4]YesSeborrheic keratosis17 [17]YesActinic keratosis7 [7]YesRosacea2 [2]YesTinea corporis or dermatophytosis (ringworm)2 [2]Yes The results presented in Table 2 suggest that most of the diagnoses consulted in PC were related to a benign tumour; there were 20 consultations for nevus (including the category of benign mole, dysplastic nevus and intradermal nevus), 17 cases of seborrheic keratosis, and 7 cases of actinic keratosis, among others. It should be noted that for the analysis of this study, actinic keratosis was included in the category of benign tumours, although acknowledging the potential risk of malignancy around $1\%$. The second most frequent diagnostic group was inflammatory diseases with 4 cases of each of the following pathologies: psoriasis, dyshidrotic eczema and granuloma annulare and 2 cases of acne vulgaris and rosacea. This was followed by infectious diseases, with 4 cases of verruca vulgaris and 2 cases of tinea corporis. Seven cases of malignant tumours were evaluated: 1 melanoma, 4 basal cell carcinomas (BCC) and 2 cutaneous squamous cell carcinomas (cSCC). Of the 18 cases in which the diagnosis was not included among the 44 diagnoses in the model (Online Appendix, Table 1), the diagnoses of granuloma annulare ($$n = 4$$) and dyshidrotic eczema ($$n = 4$$) are noteworthy because of the number of cases observed. Diagnoses such as scabies, fibroma, onychodystrophy, onychomycosis and hidradenitis, although only identified in 1 or 2 cases during the study, are usually seen in PC consultations and were not included in the list of diagnoses in the ML model. Of these 18 cases, 3 were histopathologically diagnosed: one haemangioma, one case of granuloma annulare and one case of lymphocytic dermatitis. ## Accuracy and sensitivity (Table 3) Table 3 \| Unnamed: 0 \| Accuracy \| 95% CI \| Sensitivity \| 95% CI.1 \| Specificity \| 95% CI.2 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Top 1 \| Top 1 \| Top 1 \| Top 1 \| Top 1 \| Top 1 \| Top 1 \| \| AI \| 0.39 \| (0.29; 0.49) \| 0.36 \| (0.24; 0.49) \| 0.98 \| (0.97; 0.99) \| \| AI PCD \| 0.28 \| (0.17; 0.43) \| 0.34 \| (0.15; 0.53) \| 0.96 \| (0.94; 0.98) \| \| TD \| 0.72 \| (0.62; 0.80) \| 0.7 \| (0.58; 0.83) \| 0.99 \| (0.98; 0.99) \| \| PC \| 0.64 \| (0.54; 0.73) \| 0.61 \| (0.48; 0.73) \| 0.99 \| (0.98; 0.99) \| \| Top 3 \| Top 3 \| Top 3 \| Top 3 \| Top 3 \| Top 3 \| Top 3 \| \| AI \| 0.61 \| (0.51; 0.71) \| 0.52 \| (0.37; 0.66) \| 0.98 \| (0.96; 1.00) \| \| AI PCD \| 0.61 \| (0.47; 0.75) \| 0.57 \| (0.34; 0.80) \| 0.97 \| (0.92; 1.00) \| \| TD \| 0.90 \| (0.82; 0.95) \| 0.88 \| (0.80; 0.97) \| 0.99 \| (0.99; 1.00) \| \| PC \| 0.76 \| (0.66; 0.84) \| 0.7 \| (0.57; 0.83) \| 0.99 \| (0.98; 1.00) \| \| Top 5 \| Top 5 \| Top 5 \| Top 5 \| Top 5 \| Top 5 \| Top 5 \| \| AI PCD \| 0.75 \| (0.61; 0.86) \| 0.63 \| (0.39; 0.87) \| 0.98 \| (0.95; 1.00) \| \| AI \| 0.72 \| (0.62; 0.80) \| 0.59 \| (0.44; 0.75) \| 0.99 \| (0.98; 1.00) \| The diagnostic accuracy score of the ML model in Top-1 was 0.39 (0.29–0.49) compared to 0.72 (0.62–0.80) for TD and 0.64 (0.54–0.73) for GPs. These values increase significantly when Top-3 is assessed with a diagnostic accuracy of 0.61 (0.51–0.71) for the ML model and reaching 0.72 (0.62–0.80) for Top-5 (Table 3). It should be noted that all the values of the diagnostic accuracy of the ML model are lower than those of the professionals, both for TD dermatologists and PC GPs. However, there were 18 cases in which the model was not able to recognise the disease, as it was not trained for the particular diagnosis. Thus, a subanalysis was performed including only the 82 cases corresponding to any of the 44 diagnoses with which the model was trained, after which the diagnostic accuracy increased to 0.48 (0.37–0.59) in Top-1, to 0.75 (0.66–0.85) in Top-3 and to 0.89 (0.79–0.95) in Top-5 (Table 4).Table 4Accuracy, sensitivity and specificity of the ML model with diagnoses for which it has been trained ($$n = 82$$).Accuracy$95\%$ CISensitivity$95\%$ CISpecificity$95\%$ CITop 1AI0.48(0.37; 0.59)0.56(0.40; 0.72)0.98(0.97; 0.99)Top 3AI0.75(0.66; 0.85)0.79(0.67; 0.91)(0.97; 1.00)Top 5AI0.89(0.79; 0.95)0.9(0.82; 0.98)0.99(0.97; 1.00)AI Artificial intelligence, TD Teledermatology, PC Primary care. The overall sensitivity of the model follows a similar trend to the diagnostic accuracy with 0.36 (0.24–0.49) in Top-1, 0.52 (0.37–0.66) in Top-3 and 0.63 (0.39–0.87) in Top-5. Compared to those of both dermatology and GP, the results are slightly lower, with 0.70 (0.58–0.83) and 0.88 (0.80–0.97) for TD Top-1 and Top-3, respectively, and 0.61 (0.48 -0.73) and 0.7 (0.57–0.83) for PC Top-1 and Top-3, respectively (Table 3). However, it should be noted that the specificity at all levels (AI, TD and PC) is close to 1 (0.96–0.99) (Table 3). A detailed study of sensitivity by disease was conducted (Annex, Table 2), but considering the small number of cases of some diseases, they were grouped by diagnostic groups (Fig. 4).Figure 4Mean sensitivity grouped by disease subgroups, only of the 82 cases recognised by the ML model. It was found that in the Top-3, the mean sensitivity of the model was slightly higher with respect to both PC and TD professionals in benign tumours ($$n = 53$$), where the mean sensitivity of the model was 0.87 (0.72;1.0) in the Top-3 and 0.96 (0.90; 1.0) in the Top-5, compared to 0.76 (0.63;0.89) and 0.84 (0.67;1.0) in the Top-3 for PC and TD professionals respectively. For inflammatory diseases ($$n = 12$$), AI was only superior to GP (Top-3 0.68 (0.24;1.0)) in the Top-3, but in none of the scenarios was its accuracy superior to dermatologists (Top-3 0.96 (0.87;1.0)). For infectious diseases ($$n = 8$$), the diagnostic accuracy of the ML model (Top-3 0.69 (0.09;1.0) and Top-5 0.75 (0, 29;1,0)) was superior to that of GP (Top-3 0.60 (0.0;1, 0)), but not compared to dermatologists (Top-3 0.90 (0.48;1,0)). For malignant tumours, GP had a diagnostic sensitivity of 0.92 (0.56–1.0) in the Top-3, superior to that obtained by the AI, which was 0.67 (0.0;1.0) and 0.83 (0.11;1.0) in the Top-3 and Top-5, respectively. Analysing the diagnoses included in this subgroup individually, we can see that in the case of melanoma ($$n = 1$$) the sensitivity is 1 at all levels (PC, TD and AI). For cSCC ($$n = 2$$), the sensitivity in the Top-5 of the model and the Top-3 of the professionals was 1 in all cases. For BCC ($$n = 4$$), GP have a higher sensitivity in the Top-3 (0.75) compared to the model (0.5), which does not increase in the Top-5 either. In all cases, the gold standard in these 7 cases was the histopathological analysis. For genital diseases, there was only 2 cases with an average sensitivity of 1. During data collection, and following standard clinical practice, the 11 GP could include, if they considered it appropriate for case orientation, a dermoscopic image of the skin lesion (AI PCD), taken with a Dermlite DL100 dermatoscope or a DL200 HR applied manually to the smartphone. This situation occurred in $52\%$ of cases, the vast majority of which corresponded to benign (39 of the 52 cases) and malignant (6 of the 7 cases) tumours. In cases in which the GP also assessed the dermoscopic image of the lesion with the ML model, the diagnostic sensitivity of the ML model with respect to the clinical image of the same lesion increased in the following diseases: verruca vulgaris, cSCC (Top-1, Top-3 and Top-5) and intradermal nevus (Top-3 and Top-5) (Online Appendix, Table 2). ## Degree of satisfaction of the professionals Table 5 shows the satisfaction of GPs evaluated through 3 subjective binary response questions to evaluate the satisfaction with the use of AI as a DST for each case. The $92\%$ of GP responded affirmatively to the question of whether it helped them in the differential diagnosis approach. Table 5Satisfaction and acceptance of the GPs.n (%)Together with your diagnostic criteria, would the use of AI have been sufficient to resolve the consultation without a teledermatology consultation? Yes34 [34] No63 [63] DK/NC3 [3]Did the use of AI help you with the diagnosis? Yes60 [60] No38 [38] DK/NC2 [2]Did the use of AI help you to think about other differential diagnoses? Yes92 [92] No8 [8] In $60\%$ of the cases, the AI tool was helpful in reaching the diagnosis of the lesion. In the $34\%$ of cases, they could have avoided the TD consultation (Table 5). ## Discussion In this study, a pilot external validation test of an ML model that identifies 44 skin diseases that represent a very frequent reason for PC consultation was performed in a PC setting. This is a feasibility study in routine clinical practice and will help us to develop additional studies with a larger sample which may contribute to improve the ML model used in PC. The results have shown that the 100 cases included in the study were predominantly of phototype type III, and to a lesser extent type II. According to the new Medical Device Regulation30, it is imperative to perform proper evaluations of ML models for dermatology imaging applications32, also in all skin phototypes. Thus, more studies are needed in order to ensure that they are trained in an inclusive and balanced way, and thus perform with the same accuracy on any skin phototype to avoid the possibility of disadvantaging certain groups of people. Studies exploring the use of ML models as a diagnostic tool in the medical field are starting to be conducted, primarily in image interpretation. This includes applications in interpreting retinal imaging and chest radiography49–51 The overall diagnostic accuracy of the model in this study is lower than that of both GPs and the TD assessment, as well as the one obtained in the theoretical diagnosis in the proof of concept of the model39. However, the average diagnostic sensitivity improves substantially when analysing the 82 cases in which the gold standard is included in one of the 44 diagnoses for which the model is trained. Thus, the observed results highlight the importance of determining the diagnoses not included in order to train the model and adapt it to routine clinical practice. These results differ from most theoretical and retrospective studies in which AI accuracy is usually equal to or higher than that of clinicians22,25,26,37, and are consistent with the few existing prospective and real-world studies52. In addition, it is of relevance that the specificity of the application of AI in dermatologic imaging was very close to 1, which suggests that it is a useful tool for application in routine clinical practice as a CDST. The AI model was trained using images from an online dermatology service (First Derm), not clinical images, and the patients and images have not been verified in a clinical setting. This may result in a bias in image quality due to the technology used, even with the prevalence of some skin conditions. Moreover, the fact that the diagnostic accuracy metrics increase with the Top-3 and Top-5 assessment is consistent with the usefulness in differential diagnosis, a fact already pointed out by Muñoz-López et al. in their study52. Recent algorithms tend to perform a ranked list of diagnoses. Aiding a differential diagnosis rather than a single diagnosis is particularly important in dermatology, where differential diagnosis is used for diagnostic-therapeutic decision-making. Furthermore, it can improve diagnostic accuracy when all diagnoses are taken into account, which is relevant in PC, where most of the time the most important thing is to know whether or not we are dealing with a potentially malignant lesion in order to assess the need or not for referral and/or prioritisation. The fact that TD has been established for years in the PC environment of Central Catalonia as a screening method for in-person dermatology consultations could influence different variables, such as the high quality of the images collected, the consultation time and the degree of participation acceptance of citizens9. With regard to possible interferences in the quality of the images, in the case of dermoscopic images, it should be noted that the dermatoscopes used in the PC setting are not digital or adapted for smartphones, which could lower its quality and bias the image analysis both by the dermatologists and by the ML model. The results suggest that a diagnostic aid for GPs in the resolution of dermatologic consultations would be a significant time-saver. GP can better orient the consultation at the time it occurs, not having to wait for the response time of the TD consultation (24–48 h), and, on the other hand, for dermatology specialists it would mean being able to focus their experience on cases that are difficult to manage in PC. It is not possible to draw conclusions on the individual diagnostic sensitivity by disease and, therefore, it was represented by groups. However, the small number of cases in the pilot study allowed us to perform a more exhaustive analysis of the different diseases. Nonetheless, about $50\%$ of the cases were encompassed within the same category of benign tumours, with the ML model having an advantage over the clinicians with a diagnostic sensitivity of $96\%$ in the Top-5. In the analysis of the 3 cases in which the model failed to diagnose benign tumours, we can see that in 2 of the 3 cases, when analysing the dermoscopy of both nevi, the model included the diagnosis in the Top-5. Therefore, as far as the resolution of the case in routine clinical practice is concerned, it would have been correctly oriented. In the third case, the gold standard was intradermal nevus and, when analysing the Top-5 diagnosis, the ML model included the diagnosis of nevus, but not intradermal nevus, so in the overall analysis it was considered erroneous despite the fact that in clinical practice it is of no importance to differentiate between the two categories (nevus and intradermal nevus). In future versions of the ML model, these diagnoses should be considered as a single diagnosis (nevus) due to the lack of clinical relevance. Therefore, one could infer that the ML model’s diagnostic sensitivity in routine clinical practice in the Top-5 for benign tumours is $100\%$. For malignant tumours, at a theoretical level the use of the ML model would not imply a diagnostic improvement. However, the results are not statiscally significant since the number of cases analysed was very small ($$n = 7$$) and the average diagnostic sensitivity of the professionals was very high in the Top-3. In the Top-5, an average model sensitivity of $83\%$ was observed. The ML model did not include the diagnosis of the lesion in 2 of the 7 cases of malignant tumours. These cases were one BCC and one cSCC, and the pathology report of the lesion was used as the gold standard. This case also generated diagnostic doubt among PC clinicians, since in the case of cSCC was classified as melanoma, as did the ML model. At this point, we also believe it is important to highlight that the diagnoses included in the Top-5 of the image evaluation in all cases included diagnoses in the category of malignant tumours, thus considering the malignant potential of the lesion, a relevant fact for the diagnostic and referral approach of GP. For infectious diseases, the sensitivity of the model in the Top-5 was $75\%$, failing in 3 of the 9 cases included. In the detailed analysis we see that two of the cases were verruca vulgaris. One on the face, with the clinical image, the ML model diagnosed a benign tumour (nevus, intradermal nevus and seborrheic keratosis), epidermal cyst and herpes simplex, but when including the dermoscopic image, the diagnosis of verruca vulgaris was the Top-1. Therefore, showing another case that would be solved following the clinical practice of the GP who used a dermatoscope to help with the diagnostic. The second case the ML model failed probably because the image taken by the GPs showed several lesions, which may have confused both the AI and TD. The third case was a tinea corporis of the scalp with diagnostic agreement between the 3 clinicians who assessed the image; the model’s Top-5 were seborrheic dermatitis, folliculitis, neurodermatitis, vitiligo and psoriasis. Photographing the scalp is always challenging, as cameras usually focus the hair and not the scalp, where most dermatologic diseases actually reside. Therefore, it is possible that the images used for training the ML model would have incurred this problem, decreasing its diagnostic accuracy53. For inflammatory diseases, the sensitivity of the Top-5 model was $93\%$, failing in 1 of the 11 cases. The case was acne vulgaris, in which different erythematous papular rashes could be seen, some of them with superficial crusting in the beard area. In this case, the 5 diagnoses issued by the model were: rosacea, impetigo, folliculitis, BCC and perioral dermatitis, most of them falling into the inflammatory or infectious disease category. For genital diseases, only 2 cases were included; one of balanitis and one of condyloma, in both cases the model found the correct diagnosis in the Top-1. Despite the small number of cases included in this category, the high diagnostic sensitivity in genital diseases could be explained by the fact that the model was trained at a theoretical level with $30\%$ of genital disease photographs in the dataset. It is difficult to consider the optimisation of the model with the inclusion or exclusion of diagnoses to make it more accurate in routine clinical practice; however, there are diseases documented as absent, such as, for example, dyshidrotic eczema, granuloma annulare, scabies, fibroma and hidradenitis. Taking into account the authors’ clinical experience, we suggest including these diseases in future versions of the model to improve its performance. A terminology review of the terms used by Autoderm was performed, as some of the terms used are obsolete or inaccurate in clinical practice. For example, the term "unspecified dermatitis" has never been used among dermatologists, as it is a very unspecific term. As for vascular malformations, it only takes into account haemangiomas, which would be paediatric vascular malformations, but a case assessed in adulthood was also specified. We also suggest unifying the term "Borrelia" and "erythema migrans" to avoid confusion. A proposal has also been made to improve the subclassification of acquired nevi to: junctional nevus (flat mole), compound nevus (flat mole with central raised area), intradermal nevus (raised mole) and nevus with atypical clinical features (since the diagnosis of atypia is histological). The gold standard in this study was defined as a diagnostic consensus between two or three dermatologists, a fact that may generate, in isolated cases of high diagnostic complexity, a greater difficulty compared to studies in which the histopathological analysis of all lesions is compared. These were isolated cases that, with careful deliberation among experts, were resolved correctly, reinforcing our will to act in routine clinical practice without having to perform biopsies that would imply unnecessary morbidity. As for the technical side of the ML model, it should be noted that one of the main advantages is that it can continue to learn patterns indefinitely as more images are obtained. This is in contrast to the normal training period for a GP. This process takes several years and some of the information and experience gained during the working life is eventually lost. A neural network can learn and work indefinitely. Everything suggests that the ML models’ constant learning could also have a positive impact on the professionals’ continued training, who would use it as a DST. On the other hand, it is important to mention the explainability aspect. Many automatic diagnostic algorithms do not have mechanisms for communicating why a prediction is made. This leaves the observer with only a percentage probability, which is insufficient to assess whether the decision has been made correctly or not. ## Limitations The most relevant limitation of the study is the number of images used ($$n = 100$$) for the performance evaluation of the ML model. Since Autoderm evaluates 44 skin conditions, and considering that the prevalence of a significant number of these conditions represent less than 1–$5\%$, the sample data for each class may be unbalanced and some conditions may not be evaluated, leading to an insufficient confidence level and less conclusive results for these conditions. Secondly, due to the size of the sample and the consecutive collecting of cases, no representative results were obtained for less frequent diseases. However, we have included most of the spectrum of skin lesions that are a common reason for PC consultation, as well as banal lesions to avoid selection bias. Thirdly, it should be taken into account that the GPs who agreed to participate voluntarily in the study show an interest in dermatology. Not all of them have a higher academic training in the subject, but it could explain in part that the diagnostic accuracy was higher than that reported in the literature [6,7]. In this context, the ML model would be at a disadvantage in the comparison of overall diagnostic accuracy and sensitivity, as well as in the analysis by disease subgroups. Fourth, a diagnosis made with a single image may have inherent limitations compared to diagnoses made in a clinical setting. The result of the ML model was based on a single photograph, which differs from other ML models, which consider more than one photograph. Finally, the majority of phototypes in the population where the present study was conducted are type II and III, which could be related to a decrease in diagnostic accuracy, as the other two clinical studies with Autoderm were conducted in Sweden (type I and II) and Uganda (type VI) [44,45]. Finally, although it is a strength of the study to know that all GPs accepted to participate in the study, it must be taken into account that it is not possible to know the number of patients invited to participate in the study because the GPs did not register the patients who did not accept to participate in the study. ## Conclusions This external validation feasibility study provides significant advances with respect to previous studies regarding the application of AI in routine clinical practice in PC. It provides, in first place, the diagnostic accuracy results of the ML model for images taken by different GPs in real conditions, including benign or malignant tumours and inflammatory, infectious and genital diseases. In addition, the degree of satisfaction of the professionals with the use of the AI tool in the diagnosis and also with the usefulness of having the differential diagnosis were also recorded. Despite the fact that the diagnostic accuracy in real conditions was lower than the theoretical accuracy of the ML model itself and of the professionals in most diagnostic categories, the results highlight the need for more prospective studies in clinical practice for external validation of the ML models and to be able to assess their implication in improving clinical practice in a real environment. It is necessary for technicians and clinicians to work together to improve the software and adapt it to the clinical environment. A paradigm shift is needed in the theoretical evaluation metrics of these ML model to include clinical and satisfaction parameters adapted to the real world, as called for in the new European Medical Devices Regulation. Because of its accessibility and proximity to the public, as well as the diagnostic diversity of the diseases, PC is an area to be taken into account in future AI studies. AI as a DST can provide greater diagnostic accuracy for GPs, saving time and money by reducing waiting lists for dermatology and optimising the time that dermatology specialists can devote to the most complex cases, maintaining the quality, safety and satisfaction of professionals and citizens in the resolution of consultations related to skin lesions. ## Supplementary Information Supplementary Information. The online version contains supplementary material available at 10.1038/s41598-023-31340-1. ## References 1. Wilmer EN, Gustafson CJ, Davis SA, Feldman SR, Huang WW. Most common dermatologic conditions encountered by dermatologists and nondermatologists. Cutis (2014.0) 94 285-292. PMID: 25566569 2. Hodge JA, Rohrer TA, Van BMJ, Margolis DJ, Sober AJ, Weinstock MA. The burden of skin disease in the United States. J. Am. Dermatol. (2017.0) 76 958-972.e2. DOI: 10.1016/j.jaad.2016.12.043 3. Kerr OA, Tidman MJ, Walker JJ, Aldridge RD, Benton EC. The profile of dermatological problems in primary care: Clinical dermatology. Clin. Exp. Dermatol. (2010.0) 35 380-383. DOI: 10.1111/j.1365-2230.2009.03586.x 4. 4.Servei Català de la Salut. Activitat assistencial de la xarxa sanitària de Catalunya 2012. Departament de Salut. Generalitat de Catalunya. 2013; Available from: http://www20.gencat.cat/portal/site/salut/menuitem.40dd1b31aa3dd6ec3bfd8a10b0c0e1a0/?vgnextoid=c234906c29f3a310VgnVCM1000008d0c1e0aRCRD&vgnextchannel=c234906c29f3a310VgnVCM1000008d0c1e0aRCRD&vgnextfmt=detall&contentid=6f99ec8747db2410VgnVCM1000008d0c1e0aR. 5. Lowell BA, Catherine W, Kirsner RS, Haven N, Haven W. Dermatology in primary care: Prevalence and patient disposition. J. Am. Acad. Dermatol. (2001.0) 45 24-7. DOI: 10.1067/mjd.2001.114598 6. Federman DG, Kirsner RS. The abilities of primary care physicians in dermatology.pdf. Am. J. Manag. Care (1997.0) 3 1487-92. PMID: 10178455 7. Moreno G, Tran H, Chia ALK, Lim A, Shumack S. Prospective study to assess general practitioners’ dermatological diagnostic skills in a referral setting. Australas. J. Dermatol. (2007.0) 48 77-82. DOI: 10.1111/j.1440-0960.2007.00340.x 8. Porta N, Juan JS, Grasa MP, Simal E, Ara M, Querol I. Diagnostic agreement between primary care physicians and dermatologists in the health area of a referral hospital. Actas Dermo-Sifiliográficas (2008.0) 99 207-12. DOI: 10.1016/S1578-2190(08)70233-6 9. Seguí FL, Parella JF, García XG, Peña JM, Cuyàs FG, Mas CA. A cost-minimization analysis of a medical record-based, store and forward and provider-to-provider telemedicine compared to usual care in Catalonia: More agile and efficient, especially for users. Int. J. Environ. Res. Public Health. (2020.0) 17 2008. DOI: 10.3390/ijerph17062008 10. Börve A, Gyllencreutz JD, Terstappen K, Backman EJ, Alden- A, Danielsson M. Smartphone teledermoscopy referrals : A novel process for improved triage of skin cancer patients. Acta Derm Venereol. (2015.0) 2 186-190. DOI: 10.2340/00015555-1906 11. Taberner Ferrer R, ParejaBezares A, LlambrichMañes A, Vila Mas A, Torné Gutiérrez I, Nadal Lladó C. Fiabilidad diagnóstica de una consulta de teledermatología asíncrona. Aten. Primaria (2009.0) 41 552-557. DOI: 10.1016/j.aprim.2008.11.012 12. Mounessa JS, Chapman S, Braunberger T, Qin R, Lipoff JB, Dellavalle RP. A systematic review of satisfaction with teledermatology. J. Telemed. Telecare (2018.0) 24 263-270. DOI: 10.1177/1357633X17696587 13. Vidal-Alaball J, Álamo-Junquera D, López-Aguilá S, García-Altés A. Evaluation of the impact of teledermatology in decreasing the waiting list in the Bages region (2009–2012). Aten. Primaria (2015.0) 47 320-1. DOI: 10.1016/j.aprim.2014.01.009 14. Vidal-Alaball J, Seguí FL, Domingo JLG, Mateo GF, Valmaña GS, Ruiz-Comellas A. Primary care professionals’ acceptance of medical record-based, store and forward provider-to-provider telemedicine in catalonia: Results of a web-based survey. Int. J. Environ. Res. Public Health (2020.0) 17 1-13. DOI: 10.3390/ijerph17114092 15. Tensen E, van der Heijden JP, Jaspers MWM, Witkamp L. Two decades of teledermatology: Current status and integration in national healthcare systems. Curr. Dermatol. Rep. (2016.0) 5 96-104. DOI: 10.1007/s13671-016-0136-7 16. LópezSeguí F, Vidal-Alaball J, Sagarra Castro M, García-Altés A, García CF. General practitioners’ perceptions of whether teleconsultations reduce the number of face-to-face visits in the catalan public primary care system: retrospective cross-sectional study. J. Med. Internet Res. (2020.0) 22 e14478. DOI: 10.2196/14478 17. Cinteza M. What means fourth industrial revolution for medicine. Maedica A J. Clin. Med. (2021.0) 16 343-344. DOI: 10.26574/maedica.2021.16.3.343 18. Kaul V, Enslin S, Gross SA. History of artificial intelligence in medicine. Gastrointest. Endosc. (2020.0) 92 807-12. DOI: 10.1016/j.gie.2020.06.040 19. Schwalbe N, Wahl B. Artificial intelligence and the future of global health. Lancet (2020.0) 395 1579-86. DOI: 10.1016/S0140-6736(20)30226-9 20. Yu KH, Beam AL, Kohane IS. Artificial intelligence in healthcare. Nat. Biomed. Eng. (2018.0) 2 719-31. DOI: 10.1038/s41551-018-0305-z 21. Gomolin A, Netchiporouk E, Gniadecki R, Litvinov IV, Lee KJ. Artificial intelligence applications in dermatology: Where do we stand?. Front. Med. (2020.0) 7 1-7. DOI: 10.3389/fmed.2020.00100 22. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM. Dermatologist-level classification of skin cancer with deep neural networks. Nat. Publ. Gr. (2017.0) 542 115-118 23. Young AT, Xiong M, Pfau J, Keiser MJ, Wei ML. Artificial intelligence in dermatology: A primer. J. Invest. Dermatol. (2020.0) 140 1504-1512. DOI: 10.1016/j.jid.2020.02.026 24. Goyal M, Knackstedt T, Yan S, Hassanpour S. Artificial intelligence-based image classification methods for diagnosis of skin cancer: Challenges and opportunities. Comput. Biol. Med. (2020.0) 127 104065. DOI: 10.1016/j.compbiomed.2020.104065 25. Liu YY, Jain A, Eng C, Way DH, Lee K, Bui P. A deep learning system for differential diagnosis of skin diseases. Nat. Med. (2020.0) 26 900-8. DOI: 10.1038/s41591-020-0842-3 26. Wu H, Yin H, Chen H, Sun M, Liu X, Yu Y. A deep learning, image based approach for automated diagnosis for inflammatory skin diseases. Ann. Trans. Med. (2020.0) 8 1-8. DOI: 10.21037/atm.2020.04.39 27. Thomsen K, Christensen AL, Iversen L, Lomholt HB, Thomsen K. Deep learning for diagnostic binary classification of multiple-lesion skin diseases. Front. Med. (2020.0) 7 1-7. DOI: 10.3389/fmed.2020.574329 28. Hogarty DT, Su JC, Phan K, Attia M, Hossny M, Nahavandi S. Artificial intelligence in dermatology—where we are and the way to the future: A review. Am. J. Clin. Dermatol. (2020.0) 21 41-7. DOI: 10.1007/s40257-019-00462-6 29. Mathur J, Chouhan V, Pangti R, Kumar S, Gupta S. A convolutional neural network architecture for the recognition of cutaneous manifestations of COVID-19. Dermatol. Ther. (2021.0). DOI: 10.1111/dth.14902 30. REGLAMENTO (UE) 2017/745 DEL PARLAMENTO EUROPEO Y DEL CONSEJO de 5 de abril de 2017 sobre los productos sanitarios. D Of la Unión Eur (2017.0) 2013 175 31. Directiva 93/42/CEE del consejo del parlamento europeo, relativa a los productos sanitarios. Dir 93/42/CEE (1993.0) 120 66 32. Daneshjou R, Barata C, Betz-Stablein B, Celebi ME, Codella N, Combalia M. Checklist for evaluation of image-based artificial intelligence reports in dermatology: CLEAR derm consensus guidelines from the international skin imaging collaboration artificial intelligence working group. JAMA Dermatol. (2022.0) 158 90-96. DOI: 10.1001/jamadermatol.2021.4915 33. Taylor M, Liu X, Denniston A, Esteva A, Ko J, Daneshjou R. Raising the bar for randomized trials involving artificial intelligence: The SPIRIT-artificial intelligence and CONSORT-artificial intelligence guidelines. J. Invest. Dermatol. (2021.0) 141 2109-11. DOI: 10.1016/j.jid.2021.02.744 34. Vasey B, Nagendran M, Campbell B, Clifton DA, Collins GS, Denaxas S. Reporting guideline for the early stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. Nat Med. (2022.0) 28 923-933. DOI: 10.1038/s41591-022-01772-9 35. Jobson D, Mar V, Freckelton I. Legal and ethical considerations of artificial intelligence in skin cancer diagnosis. Australas J. Dermatol. (2022.0) 63 e1-5. DOI: 10.1111/ajd.13690 36. Liu X, Cruz Rivera S, Moher D, Calvert MJ, Denniston AK, Chan AW. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: The CONSORT-AI extension. Nat. Med. (2020.0) 26 1364-1374. DOI: 10.1038/s41591-020-1034-x 37. Tschandl P, Codella N, Akay BN, Argenziano G, Braun RP, Cabo H. Comparison of the accuracy of human readers versus machine-learning algorithms for pigmented skin lesion classification: An open, web-based, international, diagnostic study. Lancet Oncol (2019.0) 20 938-947. DOI: 10.1016/S1470-2045(19)30333-X 38. Du-Harpur X, Watt FM, Luscombe NM, Lynch MD. What is AI? Applications of artificial intelligence to dermatology. Br. J. Dermatol. (2020.0) 183 423-30. DOI: 10.1111/bjd.18880 39. 39.Autoderm [Internet]. [cited 2022 Nov 28]. Available from: https://autoderm.firstderm.com/documentation/ 40. 40.He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition 2016; pp. 770–778. 41. 41.Iandola FN, Han S, Moskewicz MW, Ashraf K, Dally WJ, Keutzer K. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size. 2016;1–13. Available from: http://arxiv.org/abs/1602.07360 42. Paszke A, Gross S, Massa F, Lerer A, Bradbury J, Chanan G. PyTorch: An imperative style, high-performance deep learning library. NeurIPS (2019.0) 32 8024-8035 43. Zaar O, Larson A, Polesie S, Saleh K, Tarstedt M, Olives A. Evaluation of the diagnostic accuracy of an online artificial intelligence application for skin disease diagnosis. Acta Derm. Venereol. (2020.0) 100 1-6. DOI: 10.2340/00015555-3624 44. Kamulegeya LH, Okello M, Bwanika JM, Musinguzi D, Lubega W, Rusoke D, Nassiwa F, Börve A. Using artificial intelligence on dermatology conditions in Uganda: A case for diversity in training data sets for machine learning. BioRxiv (2013.0) 53 1689-99 45. Escalé-Besa A, Fuster-Casanovas A, Börve A, Yélamos O, Fustà-Novell X, EsquiusRafat M. Using artificial intelligence as a diagnostic decision support tool in skin disease: Protocol for an observational prospective cohort study. JMIR Res. Protoc. (2022.0) 11 e37531. DOI: 10.2196/37531 46. Josep Vidal-Alaball FLS. Ha llegado para quedarse: Beneficios e inconvenientes de la eConsulta. Aten. Primaria Práct. (2019.0) 2020 2019-2020 47. Haenssle HA, Fink C, Schneiderbauer R, Toberer F, Buhl T, Blum A. Man against machine: Diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. Ann. Oncol. (2018.0) 29 1836-1842. DOI: 10.1093/annonc/mdy166 48. Brinker TJ, Hekler A, Enk AH, Berking C, Haferkamp S, Hauschild A. Deep neural networks are superior to dermatologists in melanoma image classification. Eur. J. Cancer (2019.0) 119 11-17. DOI: 10.1016/j.ejca.2019.05.023 49. Cuadros J. The real-world impact of artificial intelligence on diabetic retinopathy screening in primary care. J. Diabetes Sci. Technol. (2021.0) 15 664-665. DOI: 10.1177/1932296820914287 50. Vidal-Alaball J, RoyoFibla D, Zapata MA, Marin-Gomez FX, Solans FO. Artificial intelligence for the detection of diabetic retinopathy in primary care: Protocol for algorithm development. JMIR Res. Protoc. (2019.0) 8 e12539. DOI: 10.2196/12539 51. Miró Catalina Q, Fuster-Casanovas A, Solé-Casals J, Vidal-Alaball J. Developing an artificial intelligence model for reading chest X-rays: Protocol for a prospective validation study. JMIR Res. Protoc. (2022.0) 11 e39536. DOI: 10.2196/39536 52. Muñoz-López C, Ramírez-Cornejo C, Marchetti MA, Han SS, Del Barrio-Díaz P, Jaque A, Uribe P, Majerson D, Curi M, Del Puerto C, Reyes-Baraona F, Meza-Romero R, Parra-Cares J, Araneda-Ortega P, Guzmán M, Millán-Apablaza R, Nuñez-Mora M, Llopy NDC, Muñoz-López C, Ramírez-Cornejo C, Marchetti MA, Han SS, Del Barrio-Díaz P. Performance of a deep neural network in teledermatology: A single-centre prospective diagnostic study. J. Eur. Acad. Dermatol. Venereol. (2021.0) 35 546-53. DOI: 10.1111/jdv.16979 53. Pasquali P. Photography in Clinical Medicine (2020.0)
--- title: Vitamin D in pregnancy (GRAVITD) – a randomised controlled trial identifying associations and mechanisms linking maternal Vitamin D deficiency to placental dysfunction and adverse pregnancy outcomes – study protocol authors: - Anna Louise Vestergaard - Martin Christensen - Mette Findal Andreasen - Agnete Larsen - Pinar Bor journal: BMC Pregnancy and Childbirth year: 2023 pmcid: PMC10015530 doi: 10.1186/s12884-023-05484-x license: CC BY 4.0 --- # Vitamin D in pregnancy (GRAVITD) – a randomised controlled trial identifying associations and mechanisms linking maternal Vitamin D deficiency to placental dysfunction and adverse pregnancy outcomes – study protocol ## Abstract ### Background The prevalence of vitamin D deficiency is high among pregnant women. Vitamin D deficiency in pregnancy is associated with increased risk of adverse pregnancy outcomes especially complications related to placental dysfunction and insulin resistance. The objective of this study is to investigate if a higher dose of vitamin D supplementation in pregnancy reduces the prevalence of vitamin D deficiency and prevents adverse pregnancy outcome with special emphasize on preeclampsia, foetal growth restriction and gestational diabetes. ### Methods GRAVITD is a double-blinded randomised trial with parallel groups where all pregnant women attending the free of charge national nuchal translucency scan programme in gestational week 10–14 at Randers Regional Hospital are invited to participate. Enrolment started in June 2020. Participants are randomised in a two armed randomization with a 1:1 allocation ratio into 1) control group – receives 10 µg of vitamin D or 2) intervention group – receives 90 µg of vitamin D. A total of 2000 pregnant women will be included. Maternal blood samples and questionnaires describing life-style habits are collected upon enrolment. For half of the participants blood samples and questionnaires will be repeated again in 3rd trimester. Blood samples will be analysed for 25-hydroxy-vitamin D using high-performance liquid chromatography coupled with tandem mass spectrometry. Upon delivery, placental tissue and umbilicalcord blood will be collected and information on maternal and fetal outcomes will be exstracted from medical records. The primary outcomes are serum levels of 25-hydroxy-vitamin D ≥ 75 nmol/L and the rate of preeclampsia, foetal growth restriction and gestational diabetes. Secondary outcome includes identification and impact on placental functions related to vitamin D. A tertiary outcome is to initiate a cohort of children born from mothers in the trial for future follow-up of the effects of vitamin D on childhood health. ### Discussion Provided that this trial finds beneficial effects of a higher dose of vitamin D supplementation in pregnancies, official recommendations can be adjusted accordingly. This will provide a low-cost and easily implementable adjustment of prenatal care which can improve health for both mother and child during pregnancy and beyond. ### Trial registration ClinicalTrial.gov: NCT04291313. Registered February 17, 2020 ## Background Vitamin D deficiency is common among pregnant women worldwide with a reported prevalence of up to $92\%$ in the most severely affected populations [1, 2]. Common causes include living in Northern countries at higher latitude and a darker skin tone [3–5]. Further, the growing prevalence of pregnant women with adiposity is disturbing in this context, as adiposity increases the risk of vitamin D deficiency [6]. Vitamin D status is a matter of concern in the clinical setting, as vitamin D deficiency appear to affect foetal growth trajectories, just as vitamin D deficiency is associated with an increased risk of pregnancy complications related to placental dysfunction. Vitamin D deficiency hence increases the risk of pre-eclampsia (PE), gestational diabetes (GDM), low birthweight and preterm birth [7–10]. Moreover, insufficient maternal vitamin D levels (25(OH)D) is associated with long-term health risks for the offspring [11]. In utero vitamin D exposure affects bone development and the strength of the tooth enamel [12, 13]. Previous studies also link exposure to low levels of vitamin D during intrauterine life to an increased risk of later disease including increased risk of asthma, type 1 diabetes, autism, schizophrenia, and multiple sclerosis in the offspring [10, 11, 14]. Since only 60–$80\%$ of maternal 25(OH)D is transferred to the foetus [15], the maternal serum 25(OH)D level needs to be ≥ 75 nmol/L to ensure that the foetus reaches a 25(OH)D level of at least 50 nmol/L. In earlier studies from Denmark, around $30\%$ of pregnant women had aa 25(OH)D level < 50 nmol/L and as many as 70–$88\%$ of the pregnant women did not reach a serum 25(OH)D level > 75–80 nmol/L [5, 16] putting the foetus at risk of developing vitamin D deficiency. In a previous cohort study we characterized our present study population, i.e. women giving birth at Randers Regional Hospital, a hospital covering both urban areas, smaller cities and rural areas. Here we found that $10\%$ of the participants had vitamin D deficiency (25(OH)D level < 50 nmol/L) and only $58\%$ reached a sufficient 25(OH)D level of ≥ 75 nmol/L [17]. Notably, this was a study undertaken in the summer months and with a high adherence ($86\%$) to the official Danish recommendations of a daily vitamin D supplement of 10 µg during the entire pregnancy. However, the officially recommended intake of 10 µg vitamin D daily is mainly based on a more than 35 year old small-scale study from Norway [18, 19] and does not take into account the nutritional differences between Denmark and Norway – e.g. a higher prevalence of fish consumption in Norway [20]. PE, one of the pregnancy complications most strongly associated with vitamin D status [7], affects 3–$5\%$ of all pregnancies worldwide and PE is still a major cause for maternal and perinatal morbidity and mortality [21, 22]. Danish registers find that $3\%$ of all pregnancies are complicated by PE [23]. We found a similar but slightly higher prevalence in our previous study conducted in 2016–2017 as $4.1\%$ developed PE [17]. In our previous study, around $10\%$ of the children were born small for gestational age (SGA) (birthweight below the 10th percentile) [17]. If growth restriction occur during pregnancy this increases the risk of complications in the prenatal period [24], and also the risk of diseases later in life [24, 25]. Such increased risks for the child occur when the estimated weight of the foetus is -$15\%$ (i.e. < 10th percentile) below the appropriate weight for the gestational age (i.e. foetal weight between the 10th 90th percentile), the condition coined foetal growth restriction (FGR). The risk is even bigger if the weight of the foetus is -$22\%$ or more below appropriate weight, the condition coined intrauterine growth restriction (IUGR) [24]. In our previous study, we found that average vitamin D levels were reduced in women who had children with PE and IUGR albeit the difference were only significant in case of IUGR [17]. Such associations have also been found by others [9]. Today GDM is described as the most common metabolic disorder of pregnancy with an increasing prevalence, affecting around 7–$10\%$ of all pregnancies worldwide [26]. Moreover, the risk of developing GDM is correlated with increasing maternal age and a high pre-pregnancy BMI. These are factors that are also increasing globally among the pregnant population [27–30]. GDM is considered as an early marker of glucose intolerance, associated with both insulin resistance and impaired insulin secretion, thereby an increased risk of maternal and fetal adverse outcomes such as macrosomia, birth trauma, respiratory distress syndrome, jaundice, and hypoglycemia, an increased rate of delivery with cesarean section, preterm labor, FGR and PE. Furthermore, there is a higher risk of obesity and diabetes in later life both for these women and their children. Some studies suggested that vitamin D play a crucial role in maintaining normal glucose levels and reduces the extent of pathologies associated with insulin resistance [31, 32]. Although it is well known that vitamin D is important in pregnancy, there are still conflicting results on whether vitamin D supplementation can reduce the prevalence of pregnancy complications. In 2012, Hollis & Wagner [33] reported that an increased dose of vitamin D supplementation in pregnancy could improve birth outcomes and lead to a significantly decreased prevalence of hypertensive disorders in pregnancy in a randomised controlled trial. Similar results have since been found by others in relation to PE [34–37], GDM [37, 38] and foetal growth [34, 36], however others found no effect [39, 40]. One of the main weaknesses in previous trials is a small sample size leading to lack of statistical power. Compiling existing data is also troublesome due to a large variance in the method chosen for evaluation of 25(OH)D status in different study populations. Notably, significant differences occur among different methods for measuring 25(OH)D [41] and too few studies use the relatively expensive methods using high-performance liquid chromatography coupled with tandem mass spectrometry (HPLC–MS/MS) despite the fact that these methods are considered to be the gold standard for measuring 25(OH)D [41, 42]. Another challenge in previous studies with vitamin D is the administration and doses of vitamin D and the duration of the intervention, which differs a lot from study to study. Along with this, it is important to keep in mind when comparing different clinical studies that the study design should be comparable in order for the results to be comparable. As vitamin D is generated in the skin, another major challenge when performing clinical studies is sunshine exposure, which varies a lot according to geography but also according to season [43]. This is also enhanced by differences in skin tone, which has a large impact on the ability to produce endogenous vitamin D in the skin. A well designed large randomised clinical trial, covering a long enrolment period to allow for seasonal variance, is lacking. ## Study design/aim The aim of this clinical trial is to investigate if a higher daily dose of vitamin D, than what is currently recommended in Denmark, can reduce the prevalence of the major pregnancy complications related to low vitamin D level in pregnancy, when initiated after pregnancy is determined (week 9 to 14). The trial is a double-blinded randomised clinical trial with parallel groups and a two armed randomization with a 1:1 allocation ratio. Focus is on the pregnancy complications related to placental dysfunction – PE, FGR (≤ -$15\%$ of the appropriate weight for the gestational age) and GDM, in addition effects of placental function will also be investigated. Enrolment to the trial will take place over a 2.5–3-year time period covering both the summer and winter months in order to cover periods of different sunshine exposure and variance from year to year. ## Study settings The trial is conducted at the Department of Gynaecology and Obstetrics, Randers Regional Hospital, Denmark. ## Participants Pregnant women attending nuchal translucency scan in gestational week 11–14 as part of the national prenatal screening program at Randers Regional Hospital are invited to participate in the trial. The nuchal translucency scan is a part of the national prenatal screening program, which is offered to all Danish pregnant women and attended by $94\%$ [44]. The screening program is free of charge. When they attend the nuchal translucency scan at the hospital they receive both orally and written information about the trial by a member of the research group. At Randers Regional Hospital there are around 2,200–2,400 childbirths annually. Women who agree to participate will give informed written consent before participation in agreement with The Declaration of Helsinki. ## Exclusion criteria Age < 18 yearsUnable to give written informed consent > 15 complete weeks pregnant at the time of the nuchal translucency scanDisturbances in calcium metabolismChronic kidney diseaseVitamin D treatment initiated by a physician ## Intervention All pregnant women who agree to participate will be randomised in equal proportions to receive either an extra vitamin D supplement (80 µg) or a placebo tablet. The participants will all receive a prenatal multivitamin (Livol®) containing 10 µg of vitamin D, which is the current national recommendation for pregnant women in Denmark [18, 45]. Thereby, all participants as a minimum gets the recommended vitamin D supplementation and are not in a higher risk of vitamin D deficiency than what they are at baseline and if they do not participate in the trial. There will be two parallel groups in the trial receiving either 90 µg or 10 µg of vitamin D respectively (Fig. 1). The dose chosen for the intervention group (90 µg) is within a dose interval previously proven safe by others [42, 46, 47] and the dose is $10\%$ below the current limit for what can be purchased in Denmark without a prescription. The vitamin D supplement tablet and the placebo tablet are identical in appearance and taste and contain exactly the same ingredients except from the cholecalciferol which is only in the vitamin D supplement tablet. The vitamin D supplement and the placebo tablets are packed in anonymized containers and provided with a unique number. Vitamin D supplements, placebo tablets and the prenatal multivitamin (Livol®) are provided by Orkla Care, Denmark. Vitamins for the remaining part of the pregnancy will be handed out at enrolment and the women will be instructed to take the vitamins daily until delivery. Fig. 1CONSORT flow diagram of the trial with intended numbers ## Adherence In order to monitor adherence in the trial, all participants will receive an email with a three-item adherence questionnaire 168 days after inclusion to the trial, which is around gestational week 36–37. They are asked whether they take the supplements every day or if they have forgotten to take them either a few times or for a longer period of time. Further, they are asked if they take the project supplement and the prenatal multivitamin or just one of them. If the questionnaire is not returned a reminder will be sent out after five days and again after 10 days. ## Concomitant care The participants are not allowed to take other vitamin D containing supplements than those handed out as part of the intervention. At the time of enrolment participants will be thoroughly informed about this. Further, it is recommended that all participants continue to take a 40–50 mg iron supplement from week 10 in pregnancy (not handed out by the study group), as the recommendations for pregnant Danish women prescribe. The prenatal multivitamin handed out in this trial contains 27 mg of iron, but since this multivitamin also contains 400 mg of calcium, the iron in the tablet is poorly absorbed. Therefore, a recommendation to the participants will be to take a separate daily iron supplement and to ingest this supplement at least 3 h apart from the supplements handed out in this trial. ## Primary The incidence of PEThe incidence of FGR (defined as a negative deviation of more than $15\%$ (10th percentile) from the expected foetal weight diagnosed at ultrasonography scans)The incidence of GDM (defined as 2-h plasma glucose ≥ 9 mmol/l following a 75 g oral glucose load)Maternal serum-25(OH)D > 75 nmol/L ## Secondary BirthweightSize related to gestational age (Small for Gestational Age; SGA, Appropriate for Gestational Age; AGA, Large for Gestational Age; LGA)The incidence of preterm birth (birth < 37 weeks of gestation)The incidence of postterm birth (birth > 40 weeks of gestation)The incidence of gestational hypertension (blood pressure > $\frac{140}{90}$)Mode of deliveryThe incidence of infection during deliveryUse of antibiotics during labourAdmission rate to the neonatal wardThe incidence of postpartum haemorrhage (bleeding > 500 ml)APGAR score at 1, 5 and 10 min after deliveryIdentification of placental functions related to maternal vitamin D status especially those also related to vitamin D metabolism and pregnancy-complications ## Tertiary Establishment of a new cohort of children born of mothers in the trial for future follow-up in order to investigate the effect a higher prenatal vitamin D supplementation has on the off-spring. ## Participant timeline Figure 2 shows the participant timeline. Fig. 2SPIRIT timeline table for the GRAVITD trial ## Randomisation and blinding Randomisation is done in a 1:1 ratio, using Research Randomizer (version 4.0). To avoid skewering of data randomisation is done for 200 numbers a time. To assure an even distribution in the two intervention groups trough out the year to secure even distribution of season and hence sunshine exposure, the project supplements are packed in batches of 100 containers with 50 containers with placebo and 50 containers with vitamin D supplements. This is being repeated 10 times in total ($$n = 2$$,000). The containers with vitamin D supplement and those with placebo are identical and they are all marked with a unique number referring back to the randomisation. When a woman is included in the trial she will get a random container with tablets. The intervention is blinded for the participants and for all health professionals treating the women during their pregnancy and delivery. Only certain members of the research group, with no role in treatment and diagnosing the women, have access to the master list of randomisation numbers. Unblinding is permissible if a participant develops signs of Vitamin D intoxication and hypercalcemia. ## Blood samples Venous blood samples will be collected at inclusion. For half of the participants a venous blood sample will be collected in their 3rd trimester as well. Blood samples will be analysed for 25(OH)D at Department of Clinical Biochemistry, Aarhus University Hospital, Denmark using HPLC–MS/MS, which is considered as the gold standard method for measurement of 25(OH)D [41]. Blood samples will also be analysed for calcium, phosphate, iron, zinc and HbA1c at Department of Clinical Biochemistry, Randers Regional Hospital, Denmark. ## Questionnaires At inclusion all participants will fill out a questionnaire together with a member from the study group. The questionnaire contains questions about demographic, different aspects of lifestyle (diet, use of dietary supplements, smoking, caffeine intake and alcohol consumption), use of medication, acute- (including covid-19), chronic and genetic diseases, earlier pregnancies, fertility treatment and some pregnancy discomforts (morning sickness, constipation). The questionnaire has a section devoted to sunshine exposure where participants are asked about travel outside of Denmark, sun tanning, use of sunscreen and use of artificial sun. The participants who return for a blood sample in their 3rd trimester will fill out a new questionnaire containing the same questions about lifestyle, use of medication, diseases and sunshine exposure and some more questions about their current pregnancy. The questionnaires used in the trial can be obtained by contacting the corresponding author. ## Biological samples from delivery At delivery, the midwife will take blood samples from the umbilical cord and store the placenta at 5 °C. A member of the study group will afterwards perform systematic sampling of the placental tissue. After informed consent from participants, placenta samples are either placed in RNAlater or snap frozen and stored at -80 °C for future biological studies on genes and proteins related to vitamin D metabolism. Next Generation RNA Sequencing will be performed on a subgroup of placenta samples. ## Information from medical records The following data will be collected from medical records:- Regarding pregnancy: duration of pregnancy, blood pressure (in 1st, 2nd and 3rd trimester, at delivery and after delivery), traces of glucose and protein in urine samples, the placement of the placenta, fetal growth and pregnancy related diseases.- Regarding delivery: mode of delivery, need of initiation of labour, need of oxytocin stimulation, use of analgesia, pyrexia during labour (defined as ≥ 38.2 °C with epidural, without epidural: ≥ 38 °C), use of antibiotics during labour, estimated volume of blood loss at delivery and 2 h postpartum, placental weight and need for caesarean section or assisted delivery.- Regarding the new-born: birthweight, sex, APGAR score at 1-, 5- and 10 min, admission to the neonatal ward, neonatal infection and malformations. ## Data management All data will be registered in an electronic case report form designed for the trial using the Research Electronic Data Capture (REDCap) database. The data collection form can be obtained by contacting the corresponding author. ## Power calculation There is no current information on what effect an increased vitamin D intake will have on the prevalence of preeclampsia or any of the other pregnancy complications of interest. Hence we have based our power calculation on a hypothesis of a 50–$55\%$ reduction in the prevalence of PE which we except to be the least common of the main targets, PE, FGR and GDM. We expect a prevalence of PE of > $4\%$ among the pregnant women getting the 10 µg vitamin D based on a prevalence of $4.1\%$ in our previous study [17] with inclusion mainly during the summer months, hence the prevalence of PE will probably be a little higher when inclusion is done during both winter and summer months. With a presumed prevalence of PE at > $4\%$, a significance level of 0.05 and a power of $80\%$ then 1,000 women in each group will be sufficient to show a statistically significant difference between the two interventions despite a dropout of $5\%$. The entire study sample will be 2,000 women. To test the hypothesis, that all participants in the intervention group reach a sufficient 25(OH)D level (≥ 75 nmol/L), a power calculation show that 15 women are needed in each group. The calculation is done with a power of $80\%$ and a significance level of 0.05 and a presumed prevalence of $42\%$ vitamin D insufficiency (25(OH)D < 75 nmol/L) before the intervention, based on data from our previous study [17]. Hence, we have chosen to take a second blood sample in the 3rd trimester in half of the cohort, resulting in 500 in each group. ## Statistical analysis Data will be analysed according to intention to treat, but if there is a non-compliance and dropout of more than $10\%$, we will also perform the statistical analysis without those with non-compliance. Demographic data will be presented as counts and percentages for categorical variables and continuous variables will be presented as mean and standard deviation. Data will be tested for normality using QQ-plots and will be log transformed if applicable. For all statistical analysis a level of statistical significance will be defined as 0.05. The primary outcomes PE, FGR and GDM, will be assessed by comparing the event rates in the two groups using a chi-square test and results will be presented as absolute and relative risks along with $95\%$ confidence intervals (Cl). A similar approach will be used for categorical secondary outcomes. A multiple linear regression model will be used to adjust for indifferences in baseline characteristics. The 25(OH)D level will be assessed by comparing the mean in the two groups using students t-test. The same approach will be used for continuous secondary outcomes. Linear regression models will be used to assess continuous variables such as serum levels, blood pressure and biological changes in the placentas. Subgroup analysis will be performed for the following subgroups:25(OH)D level at enrolment (< 50, 50–75, > 75)Parity (nulliparous vs primiparous and multiparous)EthnicityBMI (Body Mass Index) (≥ 20- < 25, ≥ 25- < 30, > 30, > 35, > 40)Extreme weight gain during pregnancySmoking statusRelevant medicationSocio economic status (income and education) ## Harms/side effects The vitamin D dose used in this trial is a safe dose [48], and even higher doses of 100 µg/day (4,000 IE/week) or even 875 µg/week (35,000 IE/week) have previously been tested in pregnant women and were found safe [42, 46, 47]. A vitamin D supplement, containing the same amount of vitamin D as our vitamin D tablet, is in Denmark widely accessible to purchase without a prescription. It lies $10\%$ below the upper limit recommended for pregnant women, so if a participant takes another vitamin containing 10 µg of vitamin D despite being told not to, she would not be in risk of taking too much vitamin D. No side effects are expected with this dose, but the participant will be instructed to discontinue the intervention if she develops symptoms that could be a result of vitamin D intoxication followed by of hypercalcaemia (nausea, vomiting, stomach pain, muscle fatigue, bone pain, tiredness, confusion, depression, loss of consciousness) and there is no other reason for the symptoms to occur or a 25(OH)D level above 220 nmol/L is measured in a blood sample at any time during the study period. ## Discussion The current recommendation regarding vitamin D in pregnancy in Denmark is based on a few small and older observational studies from other Northern countries with a diet that differs from the typical Danish diet when it comes to vitamin D containing food items such as fatty fish and the use of fish oil supplements. Since this recommendation was made a lot of new evidence have indicated that vitamin D has several additional beneficial effects in pregnancy than just bone development. New recommendations are therefore needed to fit the pregnant population today with an emphasize on the dietary habits, ethnicity and health status of today's population. With this randomised trial it will be possible to determine if a higher intake of vitamin D in pregnancy can improve the health for pregnant women and their children, by preventing maternal and neonatal complications also associated with health risks in later life. The trial is designed with a statistical power to determine the effect of a higher vitamin D supplement on the risk of developing the major obstetric diseases related to placental dysfunction – PE, FGR and GDM. Such diseases can possibly be prevented for some women by taking an extra vitamin D supplement in pregnancy – a period where most women are already vigilant when it comes to their diet and use of supplements. Further, the trial is large enough to establish if all pregnant women need to take more vitamin D to avoid deficiency or if only some groups of pregnant women need higher doses of vitamin D in their pregnancy e.g. women suffering from overweight or obesity. Our trial has some limitations. We reach the pregnant women at their first contact to the hospital, hence they are in gestational week 9–14 when enrolled in the trial and starting the intervention. As this is still several weeks before complications like PE, GDM and FGR occur, we would argue that this is not too late in pregnancy to initiate the intervention with a higher dose of vitamin D supplementation. Initiation of the intervention already before conception would most likely lead to a better effect of extra vitamin D. However, such a trial design would demand a vastly larger study population as the drop-out and lost-to-follow-up would be much higher. Furthermore, it would be challenging to reach enough women before conception and have them agree to participation in a trial. Therefore, such a trial design would be much more expensive and difficult to conduct, and a trial design like ours is a good alternative as initiation of the intervention is still in the first third of the pregnancy. Another limitation to the trial is the lack of skin type measurement. Skin type could have been evaluated using the Fitzpatrick visual scale, however as this is a visual scale it has certain limitations. Evaluation using the scale would rely on the print of the scale or light in the screen if used electronically. The best way to measure skin type would be by using a device measuring chromaticity unfortunately such a device was not available in this trial. Ethnicity of the participants are evaluated according to the origin of their parents and this will be used as a proxy of their skin type. The trial is conducted as a single-centre study which leads to both strengths and limitations. A strength is the complete alignment in diagnosing and treatment of the women as they are all followed by the same staff of midwifes and doctors and all give birth at the same labour ward with only a very few exceptions. However, a limitation always related to single-centre studies is the question of whether the study population is representative for a larger population. The trial takes place at a hospital covering both urban and rural areas and people from all socioeconomic groups and represent a broad sample of the entire population in Denmark. Selection bias can of course not be excluded completely. It is possible that people of a lower socioeconomic status with more limited resources might be more prone to decline participation in a clinical trial. On the other hand, it cannot be excluded that the offer of free vitamins will influence the decision towards acceptance of attendance among women with limited economic resources. There is also a risk that we might miss a larger proportion of women with a high interested in health matters e.g. older women who have struggled to become pregnant as these women might wish to remain in full control of their vitamin D intake. However, if the trial shows a beneficial effect of higher doses of vitamin D supplementation, initiated after pregnancy is determined this effect would especially be most beneficial to high risk groups including vulnerable pregnant women with lower socioeconomic status and those who have a higher risk of adverse maternal and neonatal outcomes. As both health care workers and many pregnant women themselves rely on the official recommendations for their health choices in pregnancy, results from this trial can help decision makers to make a new and more accurate recommendation concerning the ideal dose of vitamin D supplementation in pregnancy and in this way reach the majority of the pregnant population. ## References 1. Palacios C, Gonzalez L. Is vitamin D deficiency a major global public health problem?. J Steroid Biochem Mol Biol (2014.0) 144 138-145. DOI: 10.1016/j.jsbmb.2013.11.003 2. 2.Lee CL, Ng BK, Wu LL, Cheah FC, Othman H, Ismail NAM. Vitamin D deficiency in pregnancy at term: risk factors and pregnancy outcomes. Horm Mol Biol Clin Investig. 2017;31(3):/j/hmbci.2017.31.issue-3/hmbci-2017-0005/hmbci-2017-0005.xml. 10.1515/hmbci-2017-0005. 3. Karras SN. Understanding vitamin D metabolism in pregnancy: From physiology to pathophysiology and clinical outcomes. Metab, Clin Exp (2018.0) 86 112-123. DOI: 10.1016/j.metabol.2017.10.001 4. Karras SN, Anagnostis P, Naughton D, Annweiler C, Petroczi A, Goulis DG. Vitamin D during pregnancy: why observational studies suggest deficiency and interventional studies show no improvement in clinical outcomes? A narrative review. J Endocrinol Invest (2015.0) 38 1265-1275. DOI: 10.1007/s40618-015-0363-y 5. Andersen LB, Abrahamsen B, Dalgård C, Kyhl HB, Beck-Nielsen SS, Frost-Nielsen M. Parity and tanned white skin as novel predictors of vitamin D status in early pregnancy: a population-based cohort study. Clin Endocrinol (2013.0) 79 333-341. DOI: 10.1111/cen.12147 6. Arunabh S, Pollack S, Yeh J, Aloia JF. Body Fat Content and 25-Hydroxyvitamin D Levels in Healthy Women. J Clin Endocrinol Metab (2003.0) 88 157-161. DOI: 10.1210/jc.2002-020978 7. 7.O'Callaghan KM, Kiely M. Systematic Review of Vitamin D and Hypertensive Disorders of Pregnancy. Nutrients. 2018;10(3):10.3390/nu10030294. 8. 8.De-Regil LM, Palacios C, Lombardo LK, Pena-Rosas JP. Vitamin D supplementation for women during pregnancy. The Cochrane database of systematic reviews. 2016;(1):CD008873. doi(1):CD008873. 9. Wei S-Q, Qi H-P, Luo Z-C, Fraser WD. Maternal vitamin D status and adverse pregnancy outcomes: a systematic review and meta-analysis. J Mat Fetal Neonatal Med (2013.0) 26 889-899. DOI: 10.3109/14767058.2013.765849 10. Wagner CL, Hollis BW, Kotsa K, Fakhoury H, Karras SN. Vitamin D administration during pregnancy as prevention for pregnancy, neonatal and postnatal complications. Rev Endocr Metab Disord (2017.0) 18 307-322. DOI: 10.1007/s11154-017-9414-3 11. Ponsonby AL, Lucas RM, Lewis S, Halliday J. Vitamin D status during pregnancy and aspects of offspring health. Nutrients (2010.0) 2 389-407. DOI: 10.3390/nu2030389 12. Brustad N, Garland J, Thorsen J, Sevelsted A, Krakauer M, Vinding RK. Effect of High-Dose vs Standard-Dose Vitamin D Supplementation in Pregnancy on Bone Mineralization in Offspring Until Age 6 Years: A Prespecified Secondary Analysis of a Double-Blinded, Randomized Clinical Trial. JAMA pediatrics (2020.0) 174 419-427. PMID: 32091548 13. Nørrisgaard PE, Haubek D, Kühnisch J, Chawes BL, Stokholm J, Bønnelykke K. Association of High-Dose Vitamin D Supplementation During Pregnancy With the Risk of Enamel Defects in Offspring: A 6-Year Follow-up of a Randomized Clinical Trial. JAMA pediatrics. (2019.0) 173 924-930. DOI: 10.1001/jamapediatrics.2019.2545 14. Stubbs G, Henley K, Green J. Autism: Will vitamin D supplementation during pregnancy and early childhood reduce the recurrence rate of autism in newborn siblings?. Med Hypotheses (2016.0) 88 74-78. DOI: 10.1016/j.mehy.2016.01.015 15. Við SS. Maternal and infant vitamin D status during the first 9 months of infant life-a cohort study. Eur J Clin Nutr (2013.0) 67 1022-1028. DOI: 10.1038/ejcn.2013.152 16. Møller UK, Streym S, Heickendorff L, Mosekilde L, Rejnmark L. Effects of 25OHD concentrations on chances of pregnancy and pregnancy outcomes: a cohort study in healthy Danish women. Eur J Clin Nutr (2012.0) 66 862-868. DOI: 10.1038/ejcn.2012.18 17. Vestergaard AL, Justesen S, Volqvartz T, Aagaard SK, Andreasen MF, Lesnikova I. Vitamin D insufficiency among Danish pregnant women-Prevalence and association with adverse obstetric outcomes and placental vitamin D metabolism. Acta Obstet Gynecol Scand (2021.0) 100 480-488. DOI: 10.1111/aogs.14019 18. 18.Vitamin D. In: Nordic Council of M, editor. Nordic Nutrition Recommendations 2012. 5th ed: Norden; 2014. p. 349–84. 19. Markestad T, Ulstein M, Aksnes L, Aarskog D. Serum concentrations of vitamin D metabolites in vitamin D supplemented pregnant women A longitudinal study. Acta Obstetricia et Gynecologica Scandinavica (1986.0) 65 63-7. DOI: 10.3109/00016348609158232 20. Matthiessen JLF, Andersen HE, Barbieri K, Borodulin VK, Knudsen Kørup K, horgeirsdottir HE. Diet. The Nordic Monitoring System 2011–2014: Status and development of diet, physical activity, smoking, alcohol and overweight. Copenhagen: Nordic Council of Ministers, Nordic Council of Ministers Secretariat (2016.0) 37 21. Mol BWJ, Roberts CT, Thangaratinam S, Magee LA, de Groot CJM, Hofmeyr GJ. Pre-eclampsia. Lancet (London, England) (2016.0) 387 999-1011. DOI: 10.1016/S0140-6736(15)00070-7 22. Duley L. Pre-eclampsia and the hypertensive disorders of pregnancy. Br Med Bull (2003.0) 67 161-176. DOI: 10.1093/bmb/ldg005 23. Klemmensen AK, Olsen SF, Osterdal ML, Tabor A. Validity of preeclampsia-related diagnoses recorded in a national hospital registry and in a postpartum interview of the women. Am J Epidemiol (2007.0) 166 117-124. DOI: 10.1093/aje/kwm139 24. Pallotto EK, Kilbride HW. Perinatal outcome and later implications of intrauterine growth restriction. Clin Obstet Gynecol (2006.0) 49 257-269. DOI: 10.1097/00003081-200606000-00008 25. Barker DJ. Adult consequences of fetal growth restriction. Clin Obstet Gynecol (2006.0) 49 270-283. DOI: 10.1097/00003081-200606000-00009 26. Behboudi-Gandevani S, Amiri M, Bidhendi Yarandi R, Ramezani TF. The impact of diagnostic criteria for gestational diabetes on its prevalence: a systematic review and meta-analysis. Diabetol Metab Syndr (2019.0) 11 11. DOI: 10.1186/s13098-019-0406-1 27. Johns EC, Denison FC, Norman JE, Reynolds RM. Gestational Diabetes Mellitus: Mechanisms, Treatment, and Complications. Trends Endocrinol Metab (2018.0) 29 743-754. DOI: 10.1016/j.tem.2018.09.004 28. Matthews TJ, Hamilton BE. First births to older women continue to rise. NCHS Data Brief (2014.0) 152 1-8 29. Laopaiboon M, Lumbiganon P, Intarut N, Mori R, Ganchimeg T, Vogel JP. Advanced maternal age and pregnancy outcomes: a multicountry assessment. BJOG (2014.0) 121 49-56. DOI: 10.1111/1471-0528.12659 30. 30.Statistik D. Statistics Denmark - Population data. Mean age of parturient women 1990–2019 (in Danish) Danmarks statistik. Gennemsnitsalder for fødende kvinder 1990–2019. 2020. [Available from: https://www.statistikbanken.dk/10017. 31. Szymczak-Pajor I, Śliwińska A. Analysis of Association between Vitamin D Deficiency and Insulin Resistance. Nutrients (2019.0) 11 794. DOI: 10.3390/nu11040794 32. Silva-Zolezzi I, Samuel TM, Spieldenner J. Maternal nutrition: opportunities in the prevention of gestational diabetes. Nutr Rev (2017.0) 75 32-50. DOI: 10.1093/nutrit/nuw033 33. 33.Hollis BW, Wagner CL. Vitamin D and pregnancy: skeletal effects, nonskeletal effects, and birth outcomes. Calcif Tissue Int.92(2):128–39. 34. Sablok A, Batra A, Thariani K, Batra A, Bharti R, Aggarwal AR. Supplementation of vitamin D in pregnancy and its correlation with feto-maternal outcome. Clin Endocrinol (2015.0) 83 536-541. DOI: 10.1111/cen.12751 35. Naghshineh E, Sheikhaliyan S. Effect of vitamin D supplementation in the reduce risk of preeclampsia in nulliparous women. Adv Biomed Res (2016.0) 5 7. PMID: 26962509 36. Ali AM, Alobaid A, Malhis TN, Khattab AF. Effect of vitamin D3 supplementation in pregnancy on risk of pre-eclampsia & #x2013 Randomized controlled trial. Clin Nutr (2019.0) 38 557-563. DOI: 10.1016/j.clnu.2018.02.023 37. Rostami M, Tehrani FR, Simbar M, Bidhendi Yarandi R, Minooee S, Hollis BW. Effectiveness of Prenatal Vitamin D Deficiency Screening and Treatment Program: A Stratified Randomized Field Trial. J Clin Endocrinol Metab (2018.0) 103 2936-2948. DOI: 10.1210/jc.2018-00109 38. Asemi Z, Hashemi T, Karamali M, Samimi M, Esmaillzadeh A. Effects of vitamin D supplementation on glucose metabolism, lipid concentrations, inflammation, and oxidative stress in gestational diabetes: a double-blind randomized controlled clinical trial. Am J Clin Nutr (2013.0) 98 1425-1432. DOI: 10.3945/ajcn.113.072785 39. Hossain N, Kanani FH, Ramzan S, Kausar R, Ayaz S, Khanani R. Obstetric and Neonatal Outcomes of Maternal Vitamin D Supplementation: Results of an Open-Label, Randomized Controlled Trial of Antenatal Vitamin D Supplementation in Pakistani Women. J Clin Endocrinol Metab (2014.0) 99 2448-2455. DOI: 10.1210/jc.2013-3491 40. Roth DE, Morris SK, Zlotkin S, Gernand AD, Ahmed T, Shanta SS. Vitamin D Supplementation in Pregnancy and Lactation and Infant Growth. N Engl J Med (2018.0) 379 535-546. DOI: 10.1056/NEJMoa1800927 41. Garg U. 25-Hydroxyvitamin D Testing: Immunoassays Versus Tandem Mass Spectrometry. Clin Lab Med (2018.0) 38 439-453. DOI: 10.1016/j.cll.2018.05.007 42. Kiely ME, Wagner CL, Roth DE. Vitamin D in pregnancy: Where we are and where we should go. J Steroid Biochem Mol Biol (2020.0) 201 105669. DOI: 10.1016/j.jsbmb.2020.105669 43. 43.O'Neill CM, Kazantzidis A, Ryan MJ, Barber N, Sempos CT, Durazo-Arvizu RA, et al. Seasonal Changes in Vitamin D-Effective UVB Availability in Europe and Associations with Population Serum 25-Hydroxyvitamin D. Nutrients. 2016;8(9):10.3390/nu8090533. 44. Ekelund CK, Petersen OB, Jorgensen FS, Kjaergaard S, Larsen T, Olesen AW. The Danish Fetal Medicine Database: establishment, organization and quality assessment of the first trimester screening program for trisomy 21 in Denmark 2008–2012. Acta Obstet Gynecol Scand (2015.0) 94 577-583. DOI: 10.1111/aogs.12581 45. 45.Sundhedsstyrelsen. Graviditet og Fødsel. Kost og Kosttilskud 2019 [Available from: https://www.sst.dk/da/Viden/Foraeldreskab/Graviditet-og-foedsel/Information-til-gravide/Kost-og-kosttilskud. 46. Hollis BW, Johnson D, Hulsey TC, Ebeling M, Wagner CL. Vitamin D supplementation during pregnancy: double-blind, randomized clinical trial of safety and effectiveness. J Bone Mine (2011.0) 26 2341-2357. DOI: 10.1002/jbmr.463 47. 47.Roth DE, Al Mahmud A, Raqib R, Akhtar E, Perumal N, Pezzack B, et al. Randomized placebo-controlled trial of high-dose prenatal third-trimester vitamin D3 supplementation in Bangladesh: the AViDD trial. Nutr J. 2013;12:47–2891–12–47. 48. Hathcock JN, Shao A, Vieth R, Heaney R. Risk assessment for vitamin D. Am J Clin Nutr (2007.0) 85 6-18. DOI: 10.1093/ajcn/85.1.6
--- title: Evolution of smoking rates among immigrants in France in the context of comprehensive tobacco control measures, and a decrease in the overall prevalence authors: - Sarah Mahdjoub - Mégane Héron - Ramchandar Gomajee - Simon Ducarroz - Maria Melchior - Fabienne El-Khoury Lesueur journal: BMC Public Health year: 2023 pmcid: PMC10015536 doi: 10.1186/s12889-023-15339-x license: CC BY 4.0 --- # Evolution of smoking rates among immigrants in France in the context of comprehensive tobacco control measures, and a decrease in the overall prevalence ## Abstract ### Background The evolution of smoking rates according to migrant status has not been examined in France, despite a recent reduction in overall smoking rates. ### Methods DePICT is a two waves (2016: $$n = 4356$$; 2017: $$n = 4114$$) nationwide telephone survey, representative of the French adult population. We compared smoking-related behaviors before and after implementation of tobacco-control measures [2017], according to the geographical region of birth. ### Results Compared to 2016, individuals originating from Africa or the Middle East had a slightly higher smoking prevalence in 2017 ($34.7\%$ vs $31.3\%$), despite a higher intention to quit or attempt in the preceding year (adjusted OR(ORa) = 2.72[1.90; 3.90]) compared to non-immigrants. They were also less likely to experience an unsuccessful quit attempt (ORa = 1.76[1.18; 2.62]). ### Conclusion Tobacco-control measures could have widened smoking inequalities related to migrant status. The evolution of smoking-related behaviors among immigrants should be examined when studying the long-term effects of such policies. ## Background Smoking prevalence has been declining over time in many Western countries, but it remains a leading cause of mortality and morbidity [1, 2]. This decline has been greatest among individuals with a high socio-economic position, making tobacco a major contributor to health inequalities [3, 4]. Some tobacco control policies and interventions are reported to be less effective among socially-disadvantaged individuals, which might contribute to the widening of inequalities with regard to smoking [5]. Therefore, the effect of tobacco control measures on equity should be systematically examined. France has one of the highest smoking prevalence rates in the Western world [6]. After decades of stagnation at high smoking rates (around $30\%$), the country amplified tobacco control policies and introduced comprehensive measures in 2016. These measured consisted of the implementation of plain tobacco packaging, an increase in graphic health warnings on tobacco products, massive public health campaigns encouraging smoking cessation, and a planned increase in tobacco price [7]. These measures were followed by an unprecedented decrease in smoking rates among adults: in 2 years, there were 1.6 million fewer smokers among the French adult population (prevalence of regular smoking rates dropped from $29.4\%$ in 2016 to $25.4\%$ in 2018) [8, 9]. Auspiciously, these policies did not seem to widen socio-economic inequalities in this area. Until the COVID-19 epidemic, the decrease in smoking rates was comparable in individuals with low and high socioeconomic status, as defined by educational level [9, 10]. However, there is mounting evidence that marginalized social status due to an immigrant background, could drive health inequalities independently of education and income, especially due to marginalization and interpersonal and structural discrimination [11, 12]. Being an immigrant or having an immigrant background are now considered social determinants of health [13]. Immigrants and their offspring are often disadvantaged health-wise, compared to the general population. They are more likely to experience mental health problems, and steeper rates of health decline in older age [14, 15]. Several theories such as the acculturative stress – that is stress due to living in a foreign culture – and the cumulative disadvantage theory (migrants suffer from the negative effects of having a relatively low socioeconomic position throughout their life course) have been advanced to explain these differences [16, 17]. Further, migrants usually have low or inadequate health literacy compared to the general population [18]. In France, immigrants born in Africa and the Middle East make up the majority of the immigrant population, [19] and are reported to have worse health compared to individuals born in France [20] despite significantly lower smoking rates [21]. Therefore public health campaigns and tobacco control policies, as other preventive interventions, could have distinct impacts according to immigrant status due to different cultural backgrounds and social norms [22]. Understanding the impact of specific tobacco control measures on health inequalities is therefore important for developing and evaluating population-level public health policy interventions. In this study, we investigated tobacco-related behaviors in France before and after the implementation of specific tobacco control measures, according to immigrant status as determined by the geographical region of birth. ## Methods We conducted DePICT (Description des Perceptions, Images, et Comportements liés au Tabagisme), a nationwide telephone survey of residents of mainland France that took place in two waves one year apart: between the end of August and mid-November in 2016 and 2017. Therefore, the first wave took place before the implementation of several tobacco control measures such as plain packaging (January 1st 2017), and smoking cessation media campaigns. The target population consisted of all French speakers aged 18 to 64 years. Interviews were conducted via landline or mobile telephones by trained interviewers working for a polling institute located in the south of Paris. Randomly generated telephone lists were used to call participants up to 30 times using a computer-assisted telephone interviewing (CATI) system. In households reached by landline, one participant was randomly selected by the CATI system (Kish method) [23]. ## Ethical approval and informed consent DePICT was approved by the ethical review committee of the French National Institute of Health and Medical Research (INSERM, CEEI-IRB 00,003,888). All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standard. Informed consent was obtained from all individual participants included in the study. ## Smoking status, intentions to quit, and quit attempts in the preceding year Participants were asked about their lifetime tobacco use and their current smoking status. Current smokers were asked about the daily number of cigarettes smoked, and whether they wished or tried to quit in the preceding 12 months (Y/N). Former smokers were asked about time since the last smoking cessation. ## Geographical region of birth Participants not born in France were asked about their geographical region of birth. We also asked participants about their parent’s geographic region of birth. We then classified individuals in four categories depending on whether they or their parents were born in: a) France (non-immigrant or direct descendant of immigrants), b) another European country (including Eastern Europe), c) an African or a Middle Eastern country, or d) another region. This categorization was motivated by previous research we conducted, where we showed that first and second-generation immigrants to France from Africa and/or the Middle East have different smoking patterns individuals born elsewhere [24, 25]. It was also motivated by the low number of smoking individuals from these “minority” groups in our study. Due to French regulations, [26] we were unable to ask more direct questions about perceived ethnicity, ethnic origin, or the country of birth. Due to their small effect size, first and second generation immigrants were grouped together. ## Socio-demographic characteristics and other covariables We collected data on sociodemographic characteristics which have previously been linked to smoking: sex, age, educational level, and household situation [8]. Further, we also collected self-reported data on ever cannabis use and whether a participant lives with a smoker. ## Statistical analyses To test the association between participants’ immigrant status and smoking cessation we proceeded as follows. For each study wave, data were weighted based on the probability of being selected through the Kish method (the ratio of the number of eligible individuals to the number of telephone lines in a household), [23] and to match the structure of the French population in 2016 for sex, age, education, region of residency and smoking experimentation rates, using data from the National Institute of Statistics and Economic Studies (INSEE) and the National Health Survey [27, 28]. We used the SAS raking macro to estimate a weight value to each participant, such that the weighted distribution of the overall sample is comparable to that with the listed variables in the 2016 French population [29]. In weighted descriptive analyses, we estimated smoking rates according to the study wave, and the geographic region of birth. We also carried out two distinct multivariable regression models, to examine the adjusted association between the geographic region of birth and two different outcomes among smokers or former smokers. The first model was used to determine the adjusted association (ORa) between the geographic region of birth and the intention or attempt to quit in the preceding year (Yes/No) among smokers, adjusting for covariates, which included characteristics previously linked to smoking, which were significantly associated with the study outcome in bivariate analyses. The second multivariable logistic regression model was limited to smokers who intended to or attempted to quit smoking in the preceding year, and former smokers who quit in the preceding year. We therefore examined factors associated with an unsuccessful quit attempt in the preceding year (Yes: smokers who intended to or attempted to quit in the preceding year vs. No: Former smokers in the preceding year). All statistical analyses were conducted using SAS version 9.4 (SAS Institute Inc), statistical significance was set to 0.05. ## Smoking rates We recruited a total of 8470 participants (2016: $$n = 4356$$; 2017: $$n = 4144$$), with an unweighted mean age of 44 [sd = 13; weighted mean = 42 (sd = 13)]. More than half of the participants were women ($53\%$, weighted percent = $51\%$), and people with no high school diploma were under-represented in the original sample (unweighted percent: $31\%$; weighted percent: $47\%$). Overall, the percentage of smokers significantly decreased between the first and second wave (weighted percent, 2016: $34.7\%$; 2017: $32.3\%$; $$p \leq 0.022$$). However, among individuals born in sub-Saharan Africa, North Africa or in the Middle East (AfrME-origin) the percentage of smokers significantly increased by $6.2\%$ between the two study waves ($37.3\%$ vs $41.5\%$; $$p \leq 0.023$$). There were (non-significantly) more former smokers in the general population in the second study wave compared to the first (weighted %: $23.2\%$ vs. $22.6\%$; $$p \leq 0.7$$), while the proportion of former smokers among participants of AfrME-origin significantly decreased ($18.1\%$ vs $9.0\%$; $p \leq 0.001$) (Fig. 1).Fig. 1Smoking status (weighted prevalence (%)) in 2016 and 2017 according to participants or their parents’ geographical region of birth (total $$n = 8470$$; first study wave [2016]: $$n = 4356$$, second study wave [2017]: $$n = 4114$$) ## Smoking quit attempts in the last year Smokers’ characteristics according to their intention or attempt to quit in the preceding year, are presented in Table 1. Less than half of smokers in our study ($$n = 2269$$) were women (weighted percent: $43.6\%$). The average age of smokers was 39 years (sd = 14.3), and AfrME-origin individuals constituted $12.7\%$ (weighted percent) of the smokers’ population. Table 1Characteristics of smokers participating in the DEPICT study (weighted percent), according to their intention or attempt to quit smoking in the preceding year” ($$n = 2261$$)Total population $$n = 2261$$Quit attempt or desire to quit in the last yearp-value (Chi-Square Test)No (unweighted $$n = 535$$)Yes (unweighted $$n = 1726$$)Region of birth<.0001 France1758 ($74.5\%$)$24.4\%$$75.6\%$ Europe207 ($9.5\%$)$20.1\%$$79.9\%$ Africa or the Middle-East234 ($12.7\%$)$11.5\%$$88.5\%$ Other70 ($3.4\%$)$19.6\%$$80.4\%$Study wave0.52 First [2017]969 ($46.8\%$)$22.7\%$$77.3\%$ Second [2016]1292 ($53.2\%$)$21.7\%$$78.3\%$Sex0.86 Men1201 ($56.4\%$)$22.3\%$$77.7\%$ Women1060 ($43.6\%$)$22.1\%$$77.9\%$Age0.07 < 30525 ($30.2\%$)$25.0\%$$75.0\%$ ≥ 30 et < 45734 ($34.6\%$)$21.0\%$$79.\%$ ≥ 451002 ($35.2\%$)$21.1\%$$78.9\%$Educational level0.001 No High school diploma (< Bac)834 ($55.0\%$)$21.0\%$$79.0\%$ High School or two-year university degree885 ($32.0\%$)$21.4\%$$78.6\%$ At least a three-year university degree542 ($13.0\%$)$29.6\%$$70.4\%$Household situation0.08 Doesn’t live with a smoker696 ($27.7\%$)$23.9\%$$76.1\%$ Lives alone843 ($39.6\%$)$23.1\%$$76.9\%$ Lives with a smoker722 ($32.7\%$)$19.8\%$$80.2\%$Number of cigarettes smoked /day<.0001 < 10876 ($35.1\%$)$29.0\%$$70.1\%$ ≥ 101291 ($64.9\%$)$17.5\%$$82.5\%$ Missing944054Ever cannabis use0.37 No1272 ($55.8\%$)$21.6\%$$78.4\%$ Yes986 ($44.2\%$)$23.02\%$$77.0\%$ Missing312 Quit attempt or desire to quit was especially high among individuals from the AFR-ME group ($88.5\%$ vs $11.5\%$) compared to other groups (other European migrants: $79.9\%$ vs 20.1). The results of the multivariable analysis (Table 2) show that AfrME-origin smokers were more likely to report the intention or attempt to quit in the preceding year (ORa = 2.72 [1.90–3.90]) compared to non-immigrants or direct descendant of immigrants. Table 2Determinant of “quit attempt or desire to quit in the last year” (Yes vs No) among smokers in the DePICT study ($$n = 2164$$): results of the multivariable logistic regression model, OR; $95\%$ CIOR intention or attempt to quit in the preceding year (Yes vs No)Region of origin (ref: France) Europe1.28 (0.92; 1.78) Africa or the Middle-East2.72 (1.90; 3.90) Other1.30 (0.77; 2.19)Study wave (ref: first) Second [2017]1.15 (0.95; 1.38)Sex (ref: men) Women1.08 (0.89; 1.31)Age (ref: < 30) ≥ 30 et < 451.17 (0.93; 1.48) ≥ 451.19 (0.93; 1.53)Educational level (ref: High School or two year university degree) No High school diploma (< Bac)0.81 (0.65; 1.01) At least a three year university degree0.69 (0.51; 0.92)Living situation (ref: doesn’t live with a smoker) Lives alone0.99 (0.79; 1.25) Lives with a smoker1.29 (1.01; 1.64)Number of cigarettes smoked (ref: < 10) ≥ 101.96 (1.61; 2.39)Ever cannabis use (ref: no) Yes0.96 (0.78; 1.17)ref reference category; the p-value is strictly less than 0.05 for ORs ($95\%$CI) in bold characters (confidence interval does not contain the value 1) ## Quit attempt in the preceding year For this model, the sample consisted of participants who quit smoking in the preceding year ($$n = 370$$) and smokers who desired or attempted to quit in the preceding year ($$n = 1734$$). The results of the multivariable analysis (Table 3) show that AfrMe-origin individuals were more likely to have had an unsuccessful smoking attempt in the preceding year compared to participants with France as the region of birth (ORa = 1.76 [1.18—2.62]).Table 3Determinant of smoking cessation in the preceding year: results of the multivariable logistic regression model, OR; $95\%$ CI. Depict study, 2016 and 2017, $$n = 2$$ 104Smokers who desired or attempted to quit in the last year (vs ex-smokers who stopped in the last year)Region of origin (ref: France) European1.29 (0.85; 1.96) African or the Middle-East1.76 (1.18; 2.62) Other1.76 (0.81; 3.83)Study wave (ref: first) Second [2017]0.89 (0.70; 1.12)Sex (ref: men) Women0.88 (0.70; 1.11)Age (ref: < 30) ≥ 30 and" < 450.67 (0.50; 0.91) ≥ 450.75 (0.55; 1.03)Educational level (ref: High School or two year university degree) No High school diploma (< Bac)2.07 (1.59; 2.69) At least a three year university degree0.80 (0.58; 1.10)Living situation (ref: Doesn’t live with a smoker) Lives alone1.91 (1.47; 2.48) Lives with a smoker2.93 (2.16; 3.97)The p-value is strictly less than 0.05 for ORs ($95\%$CI) in bold characters (confidence interval does not contain the value 1) ## Discussion Our results, based on data from a two-wave nationally representative repeated cross-sectional study of 8 470 individuals in France in 2016 and 2017, show that despite an overall decrease in smoking rates after the intensification of tobacco control measures, smoking rates appear to have increased among individuals with an immigration background. In particular, individuals born in Africa or the Middle East, who comprise the largest part of immigrants in France, reported significantly higher levels of quit attempts, but an increased smoking prevalence. There is considerable literature on the association between migrant status and unhealthy behaviors in high-income countries. Generally, migrant groups have lower levels of some healthy behaviors such as access to preventative health services (including cancer screening) and physical activity compared to the general population [30, 31]. Further, longer durations of residence are linked with the acquisition of unhealthy behaviors such as unhealthy diet and smoking among migrants [32, 33]. In France, pre-migration prevalence of smoking is generally lower among African migrants arriving in the country. However, this prevalence tends to increase with time, up to levels beyond those of the native-born for certain male migrant groups, while migrant women tend to have significantly lower smoking prevalence compared to the French female general population [34]. This increase in unhealthy behaviors with time among migrants is likely exacerbated by low socio-economic disadvantage, cumulative exposure to racism, and low health literacy [31, 35]. There is also evidence that some public health interventions, which improve overall population health, could lead to ‘intervention generated inequalities’ [36, 37]. However, there is very little data on effective interventions to improve immigrant health, especially from Europe, with experts calling for more data from natural experiments like changes in policy [38]. We advance this literature by describing how comprehensive tobacco control policies in France, which were successful in decreasing overall smoking rates, did not lower smoking rates among migrants and descendant of immigrants. Tobacco control measures may have had comparable – if not better—effects on the desire to quit among immigrants. However, even if smokers born in Africa or the Middle East reported a higher desire and quit attempts, their success rates seem to be lower compared to the general population. Lower quit rates among immigrants could be explained by low access to smoking cessation services (general practitioners and tobacco cessation and addiction specialists), which is common among individuals with low socioeconomic status [39]. It could also be explained by a poorer mental health, and lower health literacy. Other mechanisms could also explain our results, such as a surge in illicit (and cheaper) cigarettes from African and middle eastern countries being sold on the streets. However, little data is available on this subject. These findings could imply that the prevalence of smoking among some immigrants and descendants of immigrants in France increases with time. This is in accordance with other European studies which also found disparities in smoking rates according to migrant status and acculturation [40]. Our findings suggest that tobacco control strategies should provide specific measures to increase successful quit attempts rates among marginalized populations. Prevention and smoking cessation interventions tailored specifically to first and generation immigrants—such as neighborhood-based and/or culturally tailored programs—are needed. The evaluation of public health interventions should also systematically include effects on migrants and other minority populations. ## Limitations Our study is one of the first to examine the change in smoking rates among immigrants after the implementation of new tobacco control measures. However, some limitations need to be noted. First, selective non-response to our repeated survey could have resulted in selection bias, especially if smokers were less inclined to participate. It is possible that smokers were more reluctant to participate in the second wave compared to the first because of a perceived increase in the stigmatisation of smoking. Nevertheless, we did weigh study data to limit such bias. Second, as in most other epidemiological studies, we use self‐reported data on smoking, which may have resulted in under-estimating smoking rates. Further, language barrier could also be a limitation in this survey targeting solely the French-speaking population. Moreover, merging immigrants and descendant of immigrants due to small effect size is likely to conceal differential subgroups trends. We also did not stratify analysis by sex due to small effect size. ## Conclusions Smoking rates appear to have increased among individuals with an immigration background in France, despite the intensification of tobacco control measures and a decrease in smoking rates among the general population. Our study provides evidence suggesting that the effect of tobacco control measures could have different effects depending on the smokers’ migrant status. ## References 1. Bilano V, Gilmour S, Moffiet T, d’Espaignet ET, Stevens GA, Commar A. Global trends and projections for tobacco use, 1990–2025: an analysis of smoking indicators from the WHO Comprehensive Information Systems for Tobacco Control. The Lancet (2015.0) 385 966-976. DOI: 10.1016/S0140-6736(15)60264-1 2. Giskes K, Kunst AE, Benach J, Borrell C, Costa G, Dahl E. Trends in smoking behaviour between 1985 and 2000 in nine European countries by education. J Epidemiol Community Health (2005.0) 59 395-401. DOI: 10.1136/jech.2004.025684 3. Allanson P, Petrie D. Longitudinal methods to investigate the role of health determinants in the dynamics of income-related health inequality. J Health Econ (2013.0) 32 922-937. DOI: 10.1016/j.jhealeco.2013.07.001 4. Jha P, Peto R, Zatonski W, Boreham J, Jarvis MJ, Lopez AD. Social inequalities in male mortality, and in male mortality from smoking: indirect estimation from national death rates in England and Wales, Poland, and North America. The Lancet (2006.0) 368 367-370. DOI: 10.1016/S0140-6736(06)68975-7 5. Hill S, Amos A, Clifford D, Platt S. Impact of tobacco control interventions on socioeconomic inequalities in smoking: review of the evidence. Tob Control (2014.0) 23 e89-97. DOI: 10.1136/tobaccocontrol-2013-051110 6. Ng M, Freeman MK, Fleming TD, Robinson M, Dwyer-Lindgren L, Thomson B. Smoking Prevalence and Cigarette Consumption in 187 Countries, 1980–2012. JAMA (2014.0) 311 183-192. DOI: 10.1001/jama.2013.284692 7. Wirth N, Béguinot E. Martinet Y [Tobacco control in France: what’s new ?]. Rev Prat (2019.0) 69 653-657. PMID: 31626428 8. El-Khoury F, Bolze C, Gomajee R, White V, Melchior M. Lower smoking rates and increased perceived harm of cigarettes among French adults one year after comprehensive tobacco control measures. Drug Alcohol Depend (2019.0) 201 65-70. DOI: 10.1016/j.drugalcdep.2019.03.025 9. 9.Bourdillon F. 1,6 million de fumeurs en moins en deux ans, des résultats inédits // 1.6 million fewer smokers in two years, unprecedented results. BEH. 2019. 10. 10.Pasquereau A, Andler R, Guignard R, Soullier N, Gautier A, Richard J-B, et al. Consommation de tabac parmi les adultes en 2020 : résultats du Baromètre de Santé publique France. Bull Epidémiologique Hebd - BEH. 2021. 11. Choy B, Arunachalam KSG, Taylor M, Lee A. Systematic review: Acculturation strategies and their impact on the mental health of migrant populations. Public Health Pract (2021.0) 2 100069. DOI: 10.1016/j.puhip.2020.100069 12. Krieger N. Discrimination and Health Inequities. Int J Health Serv (2014.0) 44 643-710. DOI: 10.2190/HS.44.4.b 13. Castañeda H, Holmes SM, Madrigal DS, Young M-ED, Beyeler N, Quesada J. Immigration as a Social Determinant of Health. Annu Rev Public Health (2015.0) 36 375-92. DOI: 10.1146/annurev-publhealth-032013-182419 14. Reus-Pons M, Mulder CH, Kibele EUB, Janssen F. Differences in the health transition patterns of migrants and non-migrants aged 50 and older in southern and western Europe (2004–2015). BMC Med (2018.0) 16 57. DOI: 10.1186/s12916-018-1044-4 15. Bas-Sarmiento P, Saucedo-Moreno MJ, Fernández-Gutiérrez M, Poza-Méndez M. Mental Health in Immigrants Versus Native Population: A Systematic Review of the Literature. Arch Psychiatr Nurs (2017.0) 31 111-121. DOI: 10.1016/j.apnu.2016.07.014 16. Garbarski D. Racial/ethnic disparities in midlife depressive symptoms: The role of cumulative disadvantage across the life course. Adv Life Course Res (2015.0) 23 67-85. DOI: 10.1016/j.alcr.2014.12.006 17. Berry JW, Wong PTP, Wong LCJ. Acculturative Stress. Handbook of Multicultural Perspectives on Stress and Coping (2006.0) 287-298 18. Ward M, Kristiansen M, Sørensen K. Migrant health literacy in the European Union: A systematic literature review. Health Educ J (2019.0) 78 81-95. DOI: 10.1177/0017896918792700 19. 19.INEDHow many immigrants are there in France? Ined - Institut national d’études démographiques2018. How many immigrants are there in France? Ined - Institut national d’études démographiques (2018.0) 20. Malmusi D. Immigrants’ health and health inequality by type of integration policies in European countries. Eur J Public Health (2015.0) 25 293-299. DOI: 10.1093/eurpub/cku156 21. Khlat M, Bricard D, Legleye S. Smoking among immigrant groups in metropolitan France: prevalence levels, male-to-female ratios and educational gradients. BMC Public Health (2018.0) 18 479. DOI: 10.1186/s12889-018-5379-8 22. Davies A, Basten A, Frattini C. Migration: A Social Determinant of the Health of Migrants (2009.0) 23. Kish L. A Procedure for Objective Respondent Selection within the Household. J Am Stat Assoc (1949.0) 44 380-387. DOI: 10.1080/01621459.1949.10483314 24. El-Khoury F, Sutter-Dallay A-L, Waerden JVD, Surkan P, Martins S, Keyes K. Smoking Trajectories during the Perinatal Period and Their Risk Factors: The Nationally Representative French ELFE (Etude Longitudinale Française Depuis l\textquotesingleEnfance) Birth Cohort Study. Eur Addict Res (2017.0) 23 194-203. DOI: 10.1159/000479022 25. Melchior M, Hersi R, van der Waerden J, Larroque B, Saurel-Cubizolles M-J, Chollet A. Maternal tobacco smoking in pregnancy and children’s socio-emotional development at age 5: The EDEN mother-child birth cohort study. Eur Psychiatry (2015.0) 30 562-568. DOI: 10.1016/j.eurpsy.2015.03.005 26. 26.Constitutionnel C. Décision n 2007–557 DC du 15 novembre 2007 sur la loi relative à la maîtrise de l’immigration, à l’intégration et à l’asile. 2007. 27. 27.Pasquereau A, Andler R, Guignard R, Richard J-B, Arwidson P, Nguyen-Thanh V. La consommation de tabac en France : premiers résultats du baromètre santé 2017. BEH. 2018. 28. 28.INSEELa population totale en 2016 − Activité, emploi et chômage en 20162016. La population totale en 2016 − Activité, emploi et chômage en 2016 (2016.0) 29. Izrael D, Hoaglin DC, Battaglia MP. A SAS Macro for Balancing a Weighted Sample: Proceedings of the Twenty-Fifth Annual SAS Users Group International Conference. SAS Inst Inc (2000.0) 275 1350-5 30. Lebano A, Hamed S, Bradby H, Gil-Salmerón A, Durá-Ferrandis E, Garcés-Ferrer J. Migrants’ and refugees’ health status and healthcare in Europe: a scoping literature review. BMC Public Health (2020.0) 20 1039. DOI: 10.1186/s12889-020-08749-8 31. Spadea T, Rusciani R, Mondo L, Costa G, Rosano A. Health-Related Lifestyles Among Migrants in Europe. Access to Primary Care and Preventative Health Services of Migrants (2018.0) 57-64 32. Delavari M, Sønderlund AL, Swinburn B, Mellor D, Renzaho A. Acculturation and obesity among migrant populations in high income countries – a systematic review. BMC Public Health (2013.0) 13 458. DOI: 10.1186/1471-2458-13-458 33. Reeske A, Spallek J, Razum O. Changes in smoking prevalence among first- and second-generation Turkish migrants in Germany – an analysis of the 2005 Microcensus. Int J Equity Health (2009.0) 8 26. DOI: 10.1186/1475-9276-8-26 34. Khlat M, Legleye S, Bricard D. Migration-related changes in smoking among non-Western immigrants in France. Eur J Public Health (2019.0) 29 453-457. DOI: 10.1093/eurpub/cky230 35. Read UM, Karamanos A, Silva MJ, Molaodi OR, Enayat ZE, Cassidy A. The influence of racism on cigarette smoking: Longitudinal study of young people in a British multiethnic cohort. PLoS ONE (2018.0) 13 e0190496. DOI: 10.1371/journal.pone.0190496 36. Thomson K, Hillier-Brown F, Todd A, McNamara C, Huijts T, Bambra C. The effects of public health policies on health inequalities in high-income countries: an umbrella review. BMC Public Health (2018.0) 18 869. DOI: 10.1186/s12889-018-5677-1 37. Lorenc T, Petticrew M, Welch V, Tugwell P. What types of interventions generate inequalities? Evidence from systematic reviews. J Epidemiol Community Health (2013.0) 67 190-193. DOI: 10.1136/jech-2012-201257 38. Diaz E, Ortiz-Barreda G, Ben-Shlomo Y, Holdsworth M, Salami B, Rammohan A. Interventions to improve immigrant health. A scoping review. Eur J Public Health (2017.0) 27 433-9. DOI: 10.1093/eurpub/ckx001 39. van Wijk EC, Landais LL, Harting J. Understanding the multitude of barriers that prevent smokers in lower socioeconomic groups from accessing smoking cessation support: A literature review. Prev Med (2019.0) 123 143-151. DOI: 10.1016/j.ypmed.2019.03.029 40. Reiss K, Lehnhardt J, Razum O. Factors associated with smoking in immigrants from non-western to western countries – what role does acculturation play? A systematic review. Tob Induc Dis (2015.0) 13 11. DOI: 10.1186/s12971-015-0036-9
--- title: An efficient edge/cloud medical system for rapid detection of level of consciousness in emergency medicine based on explainable machine learning models authors: - Nora El-Rashidy - Ahmed Sedik - Ali I. Siam - Zainab H. Ali journal: Neural Computing & Applications year: 2023 pmcid: PMC10015549 doi: 10.1007/s00521-023-08258-w license: CC BY 4.0 --- # An efficient edge/cloud medical system for rapid detection of level of consciousness in emergency medicine based on explainable machine learning models ## Abstract Emergency medicine (EM) is one of the attractive research fields in which researchers investigate their efforts to diagnose and treat unforeseen illnesses or injuries. There are many tests and observations are involved in EM. Detection of the level of consciousness is one of these observations, which can be detected using several methods. Among these methods, the automatic estimation of the Glasgow coma scale (GCS) is studied in this paper. The GCS is a medical score used to describe a patient’s level of consciousness. This type of scoring system requires medical examination that may not be available with the shortage of the medical expert. Therefore, the automatic medical calculation for a patient’s level of consciousness is highly needed. Artificial intelligence has been deployed in several applications and appears to have a high performance regarding providing automatic solutions. The main objective of this work is to introduce the edge/cloud system to improve the efficiency of the consciousness measurement through efficient local data processing. Moreover, an efficient machine learning (ML) model to predict the level of consciousness of a certain patient based on the patient’s demographic, vital signs, and laboratory tests is proposed, as well as maintaining the explainability issue using Shapley additive explanations (SHAP) that provides natural language explanation in a form that helps the medical expert to understand the final prediction. The developed ML model is validated using vital signs and laboratory tests extracted from the MIMIC III dataset, and it achieves superior performance (mean absolute error (MAE) = 0.269, mean square error (MSE) = 0.625, R2 score = 0.964). The resulting model is accurate, medically intuitive, and trustworthy. ## Introduction Emergency medicine (EM) is a rapid-growing specialty which is critical and important for the society. The patients are received in urgent cases in which rapid tests and evaluation of vital signs are very important to obtain an accurate diagnosis and make decisions. Therefore, this field attracts researchers to investigate solutions in it. Artificial intelligence is strongly involved in these investigations including machine learning (ML) and deep learning (DL) [1]. One of the important medical issues is the detection of level of consciousness which is considered as one of the important observations of the patient. For this reason, the care givers should rapidly handle the patients in this case to survive them [2]. Level of consciousness can be obtained by evaluation of Glasgow coma scale (GCS) using several methods such as electroencephalography (EEG) and vital signs [3, 4]. Traditionally, the consciousness of a certain patient can be determined based on his eye opening, verbal and motor responses which are the factors of GCS. It is a dominant method which is scaled from 0 to 15. This method needs medical examination from the medical expert which is not available all the time. Therefore, there is a need to investigate an automatic method for patients based on vital signs such as hypotensive, heart-rate-disordered and hyperthermal patients and laboratory tests such as albumin and hemoglobin in addition to collecting some medical records such as EEG and EPG signals [5]. The field of artificial intelligence is involved in Big Data and Analytics [6], Cloud/Edge Computing-based Big Computing [7] and the Internet of Things (IoT)/Cyber-Physical Systems (CPS) [8, 9] applications. These applications dominate industry and research for the development of various smart-world systems [10, 11]. Large, complicated datasets may now be approximated and reduced into extremely accurate predictions and transformative output using artificial intelligence, making human-centered smart systems much easier to implement [12–14]. Machine learning techniques can be used to any types of data such as visual, auditory, numerical, text or some combination [15]. Therefore, engineers can build their edge-based platforms based on machine learning techniques due to its high performance and low time consumption. Furthermore, the deployment of machine learning algorithms is involved to providing a security environment to ensure the privacy of the data of the patients which is transmitted through the network [16, 17]. Therefore, this study comprises the issue of security and privacy and its importance in the edge communication system [18]. The utilization of ML in medicine has witnessed an explosion in numerous medical applications [19–22], including automated diagnosis, classification of disease severity, development of new therapies [23], analysis of medical record [24] and improving the quality of medical data [25]. The use of ML methods in automated diagnosis has bifurcated in diverse disease types, including corona virus [26–30], kidney disease [31, 32], heart disease [33, 34], cancer [35, 36], diabetes and retinopathy [37, 38], skin lesion [39] and other diseases [40–42].• Motivation and contributions The GCS has been extensively used to objectively describe the extent of impaired consciousness within all types of acute medical and trauma patients. GCS assesses patients based on the following aspects: (i) eye-opening verbal, (ii) motor and (iii) verbal responses. These scoring factors are not obtained automatically, which may cause less detailed description. In addition, there is a need to overcome the difficulty with early detection and diagnosis. Therefore, the objective is to provide a numerical method to evaluate the GCS accurately has become indispensable. The main objective of this work is to introduce a machine learning-based system that is performed through IoT and edge/cloud system to enable automatically measuring the level of consciousness. The proposed system consists of three main phases: (i) vital signs acquisition, (ii) Fog-Assisted Consciousness Management (FACM) and (iii) cloud server and clinical service delivery model. In addition, the proposed system is carried out on the MIMIC dataset. Furthermore, the proposed model maintains the interoperability issue by providing natural language explanation for the developed decisions, in order to provide answers for the medical straightforward inquiries. Contributions of this paper can be illustrated in the following points:Investigate a new method for GCS evaluation for automatic estimation of the resulting score. Build an internet of medical things (IoMT) system through edge/cloud technology. Deploy machine learning techniques for detection of level of consciousness. Maintain the interoperability of the ML model in order to provide explanation to the outcome of the model. Compare the deployed techniques to obtain an optimal one in terms of evaluation metrics. Recommend an optimal system and discuss the limitation of its application. Evaluate the effectiveness of adopting the fog technology on the proposed system. This paper is organized as follows: In Sect. 2, background and related work is reviewed. In Sect. 3, an edge/cloud system for consciousness detection is proposed. In Sect. 4, experimental results are detailed. Section 5 presents a discussion for the results and comparison with the works in the literature. In Sect. 6, the paper is concluded. ## Background and related work This work proposes an edge/cloud medical system whose objective is to estimate the level of consciousness. For this purpose, we deployed a set of ML techniques to predict the value of GCS automatically. This section discusses the works in the literature which are relevant to the proposed system. Firstly, we discuss the GCS. GCS is one of the most utilized scores for responsiveness assessment of inpatients. It was introduced in 1974 to standardize the clinical assessment of level of consciousness in patients with head injuries [43]. GCS is an effective means to compare responses of patients in different coma states and to compare effectiveness of treatments [44]. GCS is a realization of three components: motor response, verbal response and eye opening. The scale originally consisted of fourteen points, four for eye opening and five for each motor response and verbal response. A sixth point was added 2 years later for motor response [45]. The GCS and its score points are shown in Table 1.Table 1Glasgow coma scaleBehaviorResponseScoreEye openingSpontaneous4To speech3To pain2No response1Verbal responseOriented5Confused conversation4Inappropriate words3Incomprehensible sounds2No response1Motor responseObeys commands6Localizes pain5Flex to withdraw from pain4Abnormal flexion3Abnormal extension2No response1 The manual calculation of GCS score involves summing the scores corresponding to the best response for each individual behavior. Hence, the total score has values between 3, being in deep coma or death, and 15, being fully alert. To the best of our knowledge, there are no relevant contributions in the literature that exploit laboratory tests and vital signals to automatically provide a numeric estimate of GCS level using machine learning techniques. Consequently, this study is the first to conduct a similar method and results. However, early studies have shown the feasibility of using GCS levels to automatically determine the functional state of the autonomic nervous system (ANS) in coma patients. However, these studies classify the consciousness level into subgroups: two subgroups (with GCS from 3 to 5 and from 6 to 8) [46], or three subgroups (low, mild and high consciousness) [4]. Estévez et al. [ 46] presented an approach to classify coma patients into two subgroups according to their GCS based on the heart rate variability (HRV). The experiments were conducted on 47 patients in coma. All patients were in ICU and mechanically ventilated. In this approach, ECG signals have been extracted, resampled into 1000 Hz and then processed using Hilbert–Huang transform (HHT) to extract a number of key spectral features. A logistic regression model was implemented to classify the consciousness level of patients into two categories: deep and mild coma, based on their HRV. They reported that their model achieved an overall efficiency of $95.74\%$. Latifoğlu et al. [ 4] proposed an approach for automatic evaluation of the state of consciousness of coma patients based on EEG signals. The state of consciousness is classified into either low, mild or high based on GCS levels. They obtained EEG signals from 34 coma patients in ICU. Features are extracted using power spectral density (PSD) method. The authors adopted various machine learning classifiers to classify the consciousness level, and they obtained an accuracy of $92.5\%$. Furthermore, machine learning algorithms are widely adopted in health care, relying on medical data to predict or classify various health states [47, 48]. Also, ML and DL have wide applications in emergency medicine [49–53]. An ML-based model was presented in [54] to predict the outcome of patients after traumatic brain injury (TBI). GCS level, besides the other thirteen parameters, was involved in predicting the patients’ outcome. Authors have conducted a performance comparison of different nine ML algorithms and reported that the random forest algorithm had achieved the best performance in outcome prediction with an accuracy of $91.7\%$. They also concluded that GCS score, besides age, fibrin/fibrinogen degradation products and glucose are the most important factors for outcome prediction. Tsiklidis et al. [ 55] implemented an ML model based on a gradient boosting classifier to predict the mortality rate of trauma patients at admission. They relied on the GCS and other seven health parameters to train the model. The accuracy of the model was $92.4\%$. The authors remarked that GCS, age and systolic blood pressure had the highest impact on the final decision of the model. In addition, Hall et al. [ 56] implemented a decision tree-based model to identify patients with a potentially modifiable outcome after intracerebral hemorrhage (ICH). They demonstrated that the GCS score is one of the most important predictors to identify the patient outcome. Similar results were concluded in [57] and [58] as the GCS score is the most significant variable in predicting the outcome and mortality of TBI, and subarachnoid hemorrhage (SAH) patients, respectively, using various machine learning models [59, 60]. The term "fog computing" refers to a paradigm that brings cloud computing and its associated services to the edge of a network. In this manner, various issues that are inherently associated with cloud computing, such as latency, lack of mobility support and lack of location awareness, are solved [61]. Fog computing and cloud computing have common and distinct features [62, 63]; Table 2 lists a comparison between the fog and cloud platforms in terms of various technical aspects. With the existence of IoT devices, the requirements for high bandwidth, security and low-latency applications are raised [64]. Therefore, fog computing is fitted here to provide such requirements for IoT networks. Recently, ML has been widely adopted with fog computing to enhance its services. Abdulkareem et al. [ 61] investigated the different roles of ML in the fog computing paradigm and provided diverse improvements in ML techniques associated with fog computing services, such as security, accuracy and resource management. Kishor et al. [ 65] presented an ML-based fog computing approach to minimize latency in healthcare applications. They implemented a multimedia data segregation scheme in fog computing to reduce the total latency resulting from the data transmission, computation and network delay. They employed the random forest model to calculate the total latency. The reported results reveal that $95\%$ reduction in latency is achieved compared to another pre-existing model. Khater et al. [ 66] proposed a lightweight intrusion detection system (IDS) for fog computing using a multilayer perceptron (MLP) model. They evaluated the developed system against two benchmark datasets: Australian Defense Force Academy Linux Dataset (ADFA-LD) and Australian Defense Force Academy Windows Dataset (ADFA-WD), which contain exploits and attacks on various applications. The developed system is implemented using a single hidden layer, and it achieved a $94\%$ accuracy in ADFA-LD and $74\%$ accuracy in ADFA-WD.Table 2A comparison between the old challenges that caused by cloud computing and the added updates using fog computing and SDNParameterCloud computingFog & SDNProcessing operation [62, 67, 68] Very highModerateData processing [67, 68, 69]At cloud serverLocally—at fog and SDNTransmission delay [62, 70, 68]HighVery lowLocation-awareness [67, 68, 70]NoHighSystem reliability [67, 68, 69, 71] SupportSupportSystem scalability [62, 67, 68]SupportSupportGeographical distribution [62, 67, 68]NoHighReal-time interactions [68]Not fully supportSupportPower consumption [68]HighLowUbiquitous services [68]SupportSupportManagement model [68]CentralizedDecentralizedDecision-making [68]RemotelyLocallySecurity [68]It’s hardEasy to apply and maintain ## Proposed edge/cloud system This work proposes an edge/cloud system for consciousness detection. The deployed scenario is based on fog-assigned consciousness managing system. The main objective of this proposed system is to enable the scoring system based on vital signs and laboratory tests such as blood pressure, heart rate, respiratory rate and oxygen flow rate to determine the consciousness level of the patient. A theoretical approach is first introduced. Recent research in describing the subsequent level of consciousness has been concerned with determining GCS, which includes several functions such as eye-opening, verbal response and motor response. Then, the collected information is recorded manually by a therapist in a local system computer-based or paper-based. In some cases, this method used may ignore significant factors such as alcohol intoxication, low blood oxygen and drug use histories. In addition, it may suffer a delay time in decision making. All these issues lead to an inappropriate score that can alter and negatively affect a patient’s level of consciousness. Therefore, introducing a dynamic data exchange environment with a high ability to deal transparently with a large scale of vital functions should be required. The structure of the proposed edge/cloud system, as shown in Fig. 1, consists of three main phases: (i) vital signs acquisition, (ii) Fog-Assisted Consciousness Management (FACM), and (iii) cloud server and clinical service delivery model. The edge/cloud system can be implemented in a wide range of intelligent healthcare sectors that are characterized by a massive infusion of data and a high need for careful and rapid decision-making, such as emergency departments and intensive care units. Fig. 1Conceptual design of the edge/cloud system ## Vital signs acquisition The operation in this face divides into two directions are: (i) collecting information of vital signs from the real sensors that directly connected to the patient; and (ii) getting information from the medical report prepared by specialists. Lately, IoT sensor devices play an essential role in medical construction in which it provides interactive real-time network connection to the users and medical devices through various communication technologies. Wearable devices are a part of IoT devices that allow sensor devices to collect real information from patients about their vital signs from everywhere at any time; this technique is called ubiquitous technology. The cope of these collected vital signs will be through the software-defined network (SDN) and fog computing installed in FACM layer. In the traditional architecture proposed in several types of research, the data will transfer to the cloud server; this technique affects not only network bandwidth but also response time. Table 2 describes old challenges that caused by cloud computing and the added updates using fog computing and SDN. ## Fog-Assisted Consciousness Management (FACM) This second phase includes several operations related to managing the process of data transmission from and to the cloud server and end-user within the treatment and prevention phase. The integration between SDN and fog technology initially enables data communication systems to be more dynamic, secure, and reliable. The security and data reliability achieved by the SDN will establish a private channel with the fog server via its controller in OpenFlow to ensure the level of data privacy. Inside this channel, the fog server applies an access control policy that is predefined by the fog setting. For more data protection, the fog server uses the integrity check process via adding octets/bytes with every data packet sent [62]. This technique is a common Internet technique that is used to notify the fog server about any change in the bytes of data. The operations in this phase are listed as follows:Providing a decentralized data transmission strategy through distributed SDN nodes integrated with other deployed fog nodes. This strategy supports local data processing and avoids unnecessary data traveling to the cloud server, thus reducing network bandwidth usage. Establishing a secure channel for boosting data protection during the operation of data exchange between SDN and IoT sensors and SDN and fog nodes. According to vital signs that sent from the lower layer, the fog computing starts to build a predictive model of consciousness level based on an unsupervised learning technique called the principal component analysis (PCA). As shown in Fig. 2, the vital signs are sent from sensors in an event shape. Each event includes sensor_ID, patient_ID, patient’s location, flow entries that tell the SDN what to do with an incoming packet, as well as the patient’s vital signs. Using OpenFlow (OF) protocol, the events will convert to flow entries for feeding SDN. This technique increases SDN capability in network monitoring and management. Fig. 2The process of data and vital signs exchange from patient to SDN and from SDN to Fog From the network point of view, achieving better network connectivity relies on avoiding communication range violation during the process of data transmission [62]. Therefore, the available communication range for each SDN is determining according to the available predefined range assigned by the fog node. For instance, as seen in Fig. 2, the communication range in coverage area I is determined by Fog1 and thus the SDN1 can only direct connected with sensors deployed in this area. Each event sent from this area has been addressed by a labeled information about sensors and patients connected in this area only. In addition, SDN1 has equipped with cache memory for saving requests and adding security rules for them. Table 3 illustrates a sample of data stored in the SDN1 caching table. Table 3A sample of data stored in the SDN1 caching tableHeader of data packetVital signsArea_IDSen_IDFlow entry#Patient_IDlocPub. KeyPriv. KeyO2 FlowHRPRRRBUNXYI-2Srl \| 186123PHU-147085212AABCID234781cy2E48vEDOLWIFI2eMYdf5421T80–10060–$\frac{10070}{12012}$–166–24 The next operation in SDN is between SDN and Fog node. Each SDN generates its own private and public key for adding them to the transferred data to the fog node. For example, a single fog node can manage multiple SDNs in the same area or even different areas to provide a large scale of data processing and analysis and overcome high response time. Herein, the data in the fog node have become accessible from multiple users, and it needs to protect. The transferred data from the SDN to the fog node will be addressed by some flow entries and SDN_ID for labelling. Moreover, the SDN adds a public key with each packet for protection. Only authorized fog node has the private key to decode this packet. It is worth to point out that only data related to the packet remain confidential, while vital signs are sent without encryption for ensuring the acceleration of fog performance. Table 4 discusses the steps of data transmission between SDN and fog node. Table 4The steps of data transmission between SDN and fog nodeStepDescription1Each SDN stores sensing data received from its coverage area into a flow table. This table contains two types of the information: (i) information about forwarded data packet and (ii) information about vital signs2Each SDN generates its own public key and adds it to every data packet as it is shown in Table 33The traffics between SDN and Fog node in the same area are occurred in a secure channel using SSH protocol4SDN uses a public key to encrypt only the header of the packet, and therefore the data packet has become anonymous or without personally identifiable information. The vital signs only will send clearly5The traffic sent from SDN is directly connected to the fog node located in the same area. This process helps to mitigate communication overhead6Each fog node provides an intelligent model that can predict the consciousness level according to vital signs sent from SDN. Only fog node that has the private key can decrypt messages sent from a certain SDN7Fog in a certain area can send the data packet to another fog node in another area for processing. This technique helps to achieve network load balancing so that if the fog node is busy the message will immediately redirect to another fog node. In our case, the fog node only can communicate with other fog nodes in different areas, while each SDN can establish the communication between its fog node located in the same area According to the proposed edge/cloud system, the deployment of fog nodes in the 2nd layer between the cloud server and IoT sensor devices plays a significant role in carrying out many fundamental operations on the received data from the SDN before passing it to the cloud server layer. One of these operations is related to reducing the consumption of network bandwidth [1, 2]. Fog node performs the processing on received data locally. Therefore, the data does not need to travel to the cloud server for processing. On the other hand, the fog nodes use a tree-based approach for building an intelligent model calculating the scoring conscious based on the vital signs received by SDN. Figure 3 depicts the main steps for building an intelligent model on the fog node. Fig. 3The main steps for calculating the level of consciousness The fog node takes the data preparation file from the previous phase in the 1st step. In the 2nd step, the fog node applies PCA for selecting the most effective features among a huge number of vital signs collected from the patient. This operation decreases the size of data that affects the system performance. In the 3rd step, the fog node builds the consciousness prediction model based on a random forest (RF) algorithm. The use of an RF to predict the GCS using vital signs has returned to several reasons that are: (i) RF gives the ability to measure feature correlation for all features using the Gini index that indicates the impact of each feature in the model [72], (ii) RF compromising the explainability and accuracy issues. Generally, models that have a good performance in terms of classification accuracy as SVM and LDA are not able to provide a clear explanation about the output decision [73], whereas the most tree-based algorithms are very good explainability, but may not be the best algorithm in terms of performance [74], and (iii) RF is a tree-based algorithm that utilized several trees and then combined the final decision using a majority voting algorithm. Information gain is used to split points in each tree. As a result, outliers are ignored by most trees that make RF a more stable algorithm [75]. In the 4th step, the GCS with correlation table will be stored at the cloud server to build historical data about these patients. ## Cloud server and clinical service delivery model In the healthcare industry, cloud computing plays a pivotal role in supporting the shift of conventional storage to the digitalization of healthcare data. The revised vital data collected from the FACM phase will be travelled to cloud computing for saving, computing and analyzing. This technique affects network bandwidth usage and accelerates cloud decisions. With cloud computing, a historical healthcare database will build to wrap up patients' data flowing from FACM. This database intends to create data linkages throughout the healthcare systems, irrespective of where the data originate or are stored. Moreover, cloud-based data analysis can prepare more personalized care reports for patients on an individual level, and thus several healthcare-related functions will be improved in terms of GCS evaluation and detection of the level of consciousness automatically. ## Experiment 1 This section is going to measure the performance of the edge/cloud medical system-based fog technology compared to traditional monitoring system without fog technology. The NS2.35 is used to model the wireless sensor network and fog sensor with varying network range from 100 to 400 sensor, Table 5 shows all network settings. The sensor nodes connect to others via IEEE 802.11p/WAVE. Table 6 depicts the configuration of the IEEE 802.11p interface described as it is in [76], and fog nodes were reconfigured based on the reference guide called “Fog hierarchical deployment model” from OpenFog Reference Architecture [77].Table 5Network settingParameterValueSensor range100–400Simulation time1500 SMAC protocolIEEE 802.11pChannel typeWirelessPhyEnergy modelBatteryAntenna modelOmniAntennaPacket size512 bytesTraffic sourceCBRTable 6IEEE80211.p interface configuration [76]ParameterValueChannel175Bandwidth20 MHzFrequency5.875 GHzAntenna gain2dBiSetup TxPower$\frac{23}{18}$ dBmReceiver sensitivity− 95.2 dBm Reliable communication has typically relied upon the quality of the packet transmission process. To this end, the performance of the proposed edge/cloud medical system in achieving high reliability was tested. In Fig. 4a, the x-axis denotes the network size, and the y-axis denotes the total speed of sensors, while in Fig. 4b, the x-axis denotes the total number of data packets, and the y-axis denotes the system throughput (kb/s). As observed shown in Fig. 4a b, the proposed system-based fog computing outperforms the traditional monitoring system-based cloud computing because the fog computing technology can increase the number of successful packets sent without affecting the network bandwidth, as is evident from Fig. 4b. Fog computing also provides the process of data processing locally, and thus, there is no need to transfer data via the Internet; only the data that need more processing and analysis will send to the cloud server. Fig. 4The performance of edge/cloud medical system-based on fog computing versus the traditional monitoring system-based on cloud computing is shown in figure (a), and the overall system throughput is shown in Figure (b) ## Experiment 2 In this experiment, we exploit the proposed edge/cloud system to improve the efficiency of the consciousness measurement through efficient local data processing. Moreover, an efficient machine learning (ML) model to predict the level of consciousness of a certain patient based on the patient’s demographic, vital signs and laboratory tests is proposed, as well as maintaining the explainability issue using Shapley additive explanations (SHAP) that provides natural language explanation in a form that helps the medical expert to understand the final prediction. ## Data collection Medical Information Mart for Intensive Care III (MIMIC-III) is a publicly available dataset for intensive care units (ICU). It comprises electronic health record (EHR) data extracted from the bedside monitors inside ICU units of the Israel Medical center in Boston, USA [78]. MIMIC III was approved and maintained by the Massachusetts Institute of Technology (MIT), and it is freely available on PhysioNet [79]. MIMIC III includes data for 46,520 different patients and 58,976 different admissions, gathered between 2001 and 2012 [80, 81]. Each patient is associated with minute-by-minute vital signs measurements and laboratory tests. The median age of adult patients is 65.8 years, and $55.9\%$ of patients are males. The sampling interval of records ranges from few seconds to hours, according to the acquired physiological measurements. In the current study, we extracted 10,349 records from the entire MIMIC dataset and are used to train/test the model to predict the level of consciousness. The data were randomly split into $\frac{80}{20}$ ratio for training and testing the ML models, respectively. Appendix A gives some information (normal range, unit of measurement) of physiological measurements and laboratory tests that are adopted in the current study. ## Data pre-processing These steps aim to enhance the quality of the chosen dataset. MIMIC III dataset includes several challenges, including missing values, outliers, etc. This may occur due to sensor or transmission failure, error in saving data, etc. Training a model in such noisy and incomplete data is considered the main reason for a model with poor performance. The following subsections discuss the steps taken to handle such data challenges. ## Irregular time interval In MIMIC III, vital signs are measured at irregular time. Some of them measured every couple of minutes and other measured every few seconds [82, 83. Unless most ML techniques are not prepared to deal with time series data, some of them could handle it when sampled with the same interval. To solve this problem, we aggregated patient’s vital signs observations to provide a single record every 1 h by taking the average of all measurements over that hour. As a result, each record includes consistent values. ## Removing outliers Outliers are values that are too far from the normal range [75, 84]. Normal range specified according to medical expert opinions. The outliers are removed; then, expectation maximization technique is used to impute them [85]. ## Data imputation Medical data usually include missing values. This returns to several reasons, including sensor failure, recording data at different time intervals, etc. The simple way to handle missing values is to remove them. However, this way may lead to losing significant information. Therefore, several algorithms have been developed to impute missing values based on the other records such as hot-deck encoding [86], and expectation maximization [85]. In the preprocessing stage of MIMIC III, large proportion of data (40–$55\%$) in important features are lost, but we could not eliminate them due to their importance in the prediction process. Considering this issue, we decide to choose cases that have at least 2 values in each measurement, then applying expectation maximization [85, 87] to impute other missing values. ## Feature extraction In this step, we mainly depend on medical expert opinions in specifying the most important features that could contribute to predict GCS and assure the ability of predicting GCS from vital signs. As mentioned before, GCS is a medical score that used to specify the consciousness of the patient through three main measures: verbal response, eye opening and motor response [88, 89]. These measurements are highly correlated with changing in patient’s vital signs. For example, low blood pressure level may lead to hypotensive; therefore, patient will not be able to respond correctly through verbal or motor response [90]. This is because decreasing blood pressure will result in generating metabolites that cause problem in circulation and tissue functions as well. The same for heart rate, the rapid or slow heart rate will mainly affect the cardiac functions that may cause cardiac arrest or blockage and subsequently affect the circulation. On the other hand, when the heart rate exceeds normal range, it increases the probability of tachyarrhythmia that may be reflected in fibrillation or atrial flutter [91, 92]. Temperature also affects the response of the patients; if the patient has hyperthermia (Temp > 40 C), the temperature autoregulation centers in the brain will be affected [93, 94]. If patient has hypothermia (Temp < = 35 C), this will affect patient response. Deceasing in O2 saturation may also lead to cardiac arrest or lactic acid accumulation [95]. Low hemoglobin level may also consider a risk indicator, especially with kidney diseases patients, while high level may lead to strokes, clots and heart attacks [91, 96, 97]. There are other features such as PCO2, HCO3, PO2. These features used to specify the percentage of carbon and oxygen in the blood, indicate the level of blood PH and patient’s acid–base balance [98]. All of these are critical situations, affect patient response and may lead to sudden death [99, 100]. Appendix A shows the feature names, Id, normal range and unit of measurement (UoM). ## Results and discussion The simulation results are carried out on the collected data from the patients using a server with NVIDIA GPU, Intel Core i7 CPU and 32 GB RAM. We used this facility to make sure that the proposed framework can be commonly used in the real application. This work proposes several ML techniques, including Linear Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), k-Nearest Neighbor (k-NN) and Ridge Regression (RG). In addition, we deployed some ensemble ML methods, including Random Forest (RF) and Gradient Boosting Regression (GBR). Table 7 illustrates the hyperparameters of each one of the proposed models. These parameters are selected based on several iterations using grid search algorithm in terms of the optimal performance. Table 7ML algorithms hyperparametersAlgorithmCoefficientLRfit_intercept = True, normalize = True, copy_X = True, n_jobs = -1SVRC = 1.0,epsilon = 0.1,kernel = 'rbf'DTcriterion = 'mse’, splitter = ’best’, max_depth = 20, presort = Falsek-NNn_neighbors = 7, algorithm = ’auto’, leaf_size = 30, $$p \leq 2$$, metric = ’minkowski’Ridgealpha = 1.0, fit_intercept = True, normalize = False, copy_X = True,, tol = 0.001, solver = 'auto', random_state = 33RFn_estimators = 100, max_depth = 16, random_state = 33GBRn_estimators = 100,max_depth = 16, learning_rate = 1.5,random_state = 33 ## Evaluation metrics The following metrics are used to evaluate the performance of the proposed model, which are Mean Absolute Error (MAE), Mean Square Error (MSE), Median Absolute Error (MedAE) and R2 score. These metrics are computed as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{MAE}} = \frac{1}{n} \mathop \sum \limits_{$i = 1$}^{n} \left\| {y_{i} - \hat{y}_{i} } \right\|$$\end{document}MAE=1n∑$i = 1$nyi-y^i2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{MSE}} = \frac{1}{n} \mathop \sum \limits_{$i = 1$}^{n} \left({y_{i} - \hat{y}_{i} } \right)^{2}$$\end{document}MSE=1n∑$i = 1$nyi-y^i23\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{MedAE}} = {\text{median}}\left({\left\| {y_{1} - \hat{y}_{1} } \right\|, \left\| {y_{2} - \hat{y}_{2} } \right\|, \ldots, \left\| {y_{n} - \hat{y}_{n} } \right\|} \right)$$\end{document}MedAE=mediany1-y^1,y2-y^2,…,yn-y^n4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R^{2} = 1 - \frac{{\mathop \sum \nolimits_{$i = 1$}^{n} \left({y_{i} - \hat{y}_{i} } \right)^{2} }}{{\mathop \sum \nolimits_{$i = 1$}^{n} \left({y_{i} - \overline{y}} \right)^{2} }}$$\end{document}R2=1-∑$i = 1$nyi-y^i2∑$i = 1$nyi-y¯2where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{i}$$\end{document}yi is the ith true value. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{y}}_{i}$$\end{document}y^i is the corresponding predicted value. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n$$\end{document}n is the number of observations. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{y }$$\end{document}y¯ is the mean of the true values. ## Results without feature selection The proposed models are performed on the dataset without feature selection. This scenario is performed to highlight the performance of the proposed models without any preprocessing. Table 8 illustrates the training and testing scores. In addition, it contains the evaluation metrics including MAE, MSE, MedAE and R2 score. It can be observed that the tree algorithm which is deployed using DT achieved a quite high performance with 0.503, 1.186, 0.348 and 0.8926 for MAE, MSE, MedAE and R2 score, respectively. Moreover, the ensemble method (RF) appears an optimal performance with 0.406, 0.8385, 0.2839 and 0.9229 for MAE, MSE, MedAE and R2 score, respectively. Furthermore, Fig. 5a–g shows the learning curves of the proposed methods containing the train and validation curves. It can be observed that the proposed DT and RF methods achieved their optimal performance at number of iterations of 7000, while the other deployed method achieved theirs at 8000. This means that they have a high performance with a low complexity. So, these techniques (DT and RF) can be considered as acceptable automatic GCS prediction solutions without preprocessing. However, there is a need to enhance their performance with some preprocessing like feature selection which is discussed in the next subsection. Table 8Results of ML models without feature selectionAlgorithmTraining scoreTesting scoreMAEMSEMedAER2 scoreLR0.8190.8160.7011.9950.3810.816SVM0.8500.8490.39261.6420.085910.8491DT0.8970.8960.5031.1680.3480.8926KNN0.8910.8350.5701.7950.1190.835Ridge0.8190.8160.7111.9930.3920.816(GBR)0.8850.8800.5721.30310.25780.880RF0.9290.9220.4060.83850.28390.9229Fig. 5Results of machine learning models without feature selection for both training and testing (a) linear regression model, (b) support vector regression model, (c) decision tree model, (d) k-nearest neighbor, (e) ridge regression, (f) XGBoost, (g) random forest regressor ## Results with feature selection In this section, we perform feature selection algorithm which utilized to magnify the impact of the input features by selecting the most important features to estimate the output value. For this purpose, we used the recursive feature elimination (RFE) algorithm with feature scaling to figure out the most impacted feature by a recursive elimination process which leads to a high performance. Table 9 illustrates the simulation results of the proposed models. It can be observed that DT model achieved 0.39, 0.961, 0.25 and 0.917 for MAE, MSE, MedAE and R2 score, respectively. In addition, RF model achieved 0.269, 0.625, 0.0784 and 0.946 for MAE, MSE, MedAE and R2 score, respectively. Thus, it can be noticed the performance of both DT and RF model is considerably increased by deploying feature selection. Furthermore, the proposed SVM regressor achieved 0.283, 0.813, 0.0844 and 0.929 for MAE, MSE MedAE and R2 score, respectively. So, the proposed models can be considered as efficient solutions for detection of level of consciousness. Figure 6a–g shows the training and testing performance of the developed model performance. Table 9Results of ML models with feature selectionAlgorithmTraining scoreTesting scoreMAEMSEMedAER2 scoreLR0.89120.8880.6521.3040.4080.888SVM0.9340.9290.2830.8130.08440.929DT0.9320.9170.3900.9610.250.917KNN0.9490.9340.3500.7620.00.934Ridge0.89110.8850.6581.3050.4140.888(GBR)0.9270.9160.4750.9740.2400.916RF0.9600.9400.2690.6250.07840.946Fig. 6Results of machine learning models with feature selection for both training and testing (a) linear regression model, (b) support vector regression model, (c) decision tree model, (d) k-nearest neighbor, (e) ridge regression, (f) XGBoost, (g) random forest regressor ## Model explainability ML and deep learning (DL) have been widely used in predicting and diagnosing various diseases such as predicting hypertension [101], diabetes [102], sepsis [103]. Unfortunately, most of these studies concentrated on achieving advances in the overall performance of the developed models, while disregarding the interpretability issues. ML and DL are considered as a black box that is unable to provide answers for the medical straightforward inquiries (i.e., why it developed this decision, what the correlation between patient’s medical features and the developed output, etc.). Therefore, the physicians find it complex to depend on such models without a clear and understandable explanation. As a result, there exists a serious gap between the developed models and their utilization in medical practice. In the latest years, a quite number of studies tried to justifying this issue by explaining the developed models using what is known as explainable artificial intelligence [73, 74, 104]. Explainable artificial intelligence or explainability (XAI) is the ability of ML and DL models to open the black box and provide natural language explanation for the developed decisions [105–107], explain what occurred in the developed model from input features to the final output. It utilized to help non-ML experts to understand the solutions developed by ML models. Therefore, we not only concentrate in developing ML model that could predict GCS based on vital signs and laboratory tests, but also provide an explanation for the developed decision. In this work, we depend on the SHAP library in the explanations issue [108]. SHAP explainers usually depend on a tree-based classifier (i.e., DT, RF, XGBoost, LightGBM, etc.) to calculate the contribution of each feature in the decision [109, 110]. Furthermore, we utilize the internal logic of the RF regressor to discuss the explainability of features as well as cases. ## Explainability of features (globally) Feature importance gives a general view of the rank of all features and the impact of each feature on the final decision. In this work, we depend on RF to calculate the importance of all features. Table 10 shows the feature importance for all features. Unfortunately, we cannot depend on it to specify the direction of each feature. For instance, we cannot specify if increasing the cardiac index will contribute to increasing the overall score of consciousness or not. Table 10Importance of features according to RF modelFeature nameCorrelation using RFHeart rate0.09850SpO20.09146Temperature C (calc)0.04965Arterial blood pressure systolic0.03043CVP alarm [low]0.04972O2 flow0.02934CVP alarm [high]0.02933Respiratory rate0.08911HCO3 (serum)0.02906Hemoglobin0.02817Blood urea nitrogen (BUN)0.07517Creatine kinesis0.06293WBC0.02269Insulin0.02019Cardiac index0.01621Cholesterol, HDL0.0349Respiratory effort0.06302Arterial blood pressure diastolic0.06255Glucose finger stick0.05606Arterial pH0.01021weight0.01822Albumin0.02703Urine out void0.00257 In this section, we utilized SHAP summary plots to show the rank of each feature. As shown in Fig. 7, each line represents one feature, and each dot represents the effect of this feature in a specific instance. Feature correlation is represented by colors (blue for low correlation, and red for high correlation). From the summary plot, we can observe the following: [1] heart rate has a significant impact on the overall decision. [ 2] Increasing heart rate and O2 flow value have a positive impact in increasing the overall score. [ 3] On the contrary, decreasing the value of cholesterol, respiratory effort and CVP alarm [high] have a positive impact on the overall performance of the calculated score. [ 4] Summary plot also allows us to specify the impact of the outliers. For example, the respiratory effort is not the global critical feature, but it has a high negative effect on some cases. This is indicated in the long-tail that is distributed along the left direction. Features that have a long tail in the right direction are likely to have a positive effect on the total decision. Our medical expert reported that this is a medically intuitive issue that increases the confidence in our model. Fig. 7SHAP summary plots for the proposed model To ensure the impact of the chosen feature in the developed model, we extract the feature importance using RF model. Table 10 illustrates the correlation among the whole features and the estimated output. It can observe that there are some features have high correlation such as heart rate, temperature and blood pressure. Others have low correlation values such as cardiac index and arterial PH. ## Explainability of cases (locally) In this section, we will discuss the explainability for each case. For example, as we show in Fig. 8, each example represents a horizontal line. It firstly shows the final prediction for this case (GCS = 11.15). It also shows the features that have a positive impact on the final decision (heart rate = 68, albumin = 6), and features that push the total prediction away from the optimal values (SpO2 = $90.6\%$, O2_flow = $90.83\%$ and BUN = 1.6). This represents the effect of each feature in the final output by colors (red = supported, blue = not supported).Fig. 8SHAP model behavior for cases ## Comparison between model performance before and after feature selection (FS) In this section, we compare the performance of our proposed model in terms of the two-feature list (before and after feature selection). Figure 9 shows comparison between all models in terms of MAE, MSE and R2 score. From this figure we can observe that the performance improved about 2–$6\%$ for all evaluation metrics. This enhancement ensures the importance of this step in the final result. Figure 10a, b shows correlation between the feature list and the output before and after FS stage. Fig. 9Comparison between models performance (before and after FS step)Fig. 10a Correlation before FS, b Correlation after FS ## Study limitations Although our proposed model provides a promising solution for GCS automatic calculation, it still has some limitations that need to be handled. First, there are some situations in which changes in vital signs will not affect GCS, such as small changes in blood pressure that did not reach hypotension, hypothermia that did not affect thermoregulation center. Second, MIMIC dataset was extracted from one institute; therefore, using MIMIC dataset to evaluate the developed model does not guarantee the generalization ability of the model. Third, the imputing process for several important features could negatively affect model performance. Therefore, we intend to investigate several imputation techniques. All of these limitations will be addressed in the future studies. ## Comparison with the works in the literature This paper proposes a GCS prediction system to estimate the level of consciousness of the patients based on their vital signs. For this purpose, several machine learning techniques are deployed to achieve the optimal method. The simulation results are carried out on the MIMIC III dataset with the interest of the vital signs and the level of consciousness. This paper proposes several machine learning models to handle this issue. The simulation results are carried out on the data with and without feature selection. The feature selection is performed based on the importance of the features and their impact on the output according to their correlation with the output. The simulation results reveal that the proposed SVM, KNN, DT and RF models achieved the optimal performance prior to GCS value prediction. To highlight the performance of the proposed system, we illustrate the impact of the proposed system with the works in the literature. The objective of the proposed system is to predict the value of the GCS using regression. The works in the literature focused on the classification as a solution for this issue. As proposed in [4], they categorized the GCS into three ranges and performed machine learning techniques to classify among them. We show the performance of the proposed system with the proposed in [4] in terms of the mutual machine learning techniques, including KNN, SVM and RF. We compare their performance as classifiers and regressors to solve the problem of diagnosis of level of consciousness. The regression method is evaluated by R2 score, while the classification method is evaluated by accuracy. Both of the evaluation metrics are within range of 0–1. As shown in Table 11, the proposed models without feature selection achieve 0.835, 0.849 and 0.923 for KNN, SVM and RF, respectively. On the other hand, these models with feature selection achieved 0.934, 0.929 and 0.946 for KNN, SVM and RF, respectively. Therefore, it can be observed that the regression trend achieved a quite high performance rather than classification trend prior to diagnosis of level of consciousness. Table 11Illustration of the proposed work and the works in the relationWorkMethodModelMetricPerformanceProposedRegressionKNNR2 score0.835SVM0.894RF0.923KNN + feature selection0.934SVM + feature selection0.929RF + feature selection0.946[2]ClassificationKNNAccuracy0.875SVM0.831RF0.925 ## Conclusion The problem of detection of level of consciousness has been discussed in this work. This issue has been handled in the presence of IoT system and cloud/edge environment. The proposed framework is based on deploying machine learning for automatic prediction of the level of consciousness based on some vital signs and laboratory tests. Therefore, several machine learning models including both ensemble and kernel models have been implemented to provide a judgeable comparison and extensive study. The simulation results reveal that the proposed ensemble models present a superior performance prior to prediction of the level of consciousness. Therefore, it can be considered as an efficient solution for consciousness level prediction in IoT and cloud/edge environments. ## Appendix A: Features description Item_IDLabelNormal rangeUnit of measurement619, 224,690Respiratory rate12–16Breaths per minute2381, 220,045Heart rate60–100Beats per minute646, 5820SpO2 > $90\%$470O2 flow80–$100\%$1525Creatine0.74–1.35mg/dL3066,772Albumin0.4–5.4g/dL116, 7610Cardiac index2.5–4.0l/min/m21162,781Blood urea nitrogen (BUN)6–24mg/dL220,050, 220,179Blood pressure$\frac{120}{80}$mmHg676, 223,762Temperature36.1–37.2Celsius440,546, 227,062WBC(4–11)000N/μL225,664Glucose finger stick80–130mg/dLInsulin80–130mg/dL50,909Cortisol6–8 am,10–40 pmmcg/dL50,904Cholesterol, HDL40–59mg/dL616Respiratory effort––780Arterial pH7.25–7.55mmHg40,069Urine out void0.3–0.5 ml for kg per HmLWeightKgAgeY198, 227,015GCS11–15(mg/dl) milligrams/deciliter(g/dl) grams/deciliter ## Appendix B: List of abbreviations TermAbbreviationsDLDeep learningLRLinear regressionSVRSupport vector regressionKNNK nearest neighborMLPMultilayer perceptronRFRandom forestBUNBlood urea nitrogenWBCWhite blood cellsGCSGlasgow coma scaleUoMUnit of measurementXAIExplainability artificial intelligenceSHAPShapley additive explanationsSDNSoftware defined networkIoTInternet of thingsPCAPrinciple component analysisFACMFog-Assisted Consciousness ManagementMAEMean absolute errorMSEMean square error ## References 1. Tang KJW, Ang CKE, Theodoros C. Artificial intelligence and machine learning in emergency medicine. Biocybern Biomed Eng (2020.0) 41 156. DOI: 10.1016/j.bbe.2020.12.002 2. Azimi M, Eslamlou AD, Pekcan G. Data-driven structural health monitoring and damage detection through deep learning: state-of the-art review. Sensor (2020.0) 20 2778. DOI: 10.3390/s20102778 3. 3.Tindall SC (1990) Level of consciousness. In: Walker HK, Hall WD, Hurst JW (eds). Boston 4. 4.Latifoğlu F, Altıntop ÇG, Akın AK, et al (2020) Evaluation of glasgow coma score using electroencephalogram signals. In: 2020 IEEE International Symposium on Medical Measurements and Applications (MeMeA). IEEE, pp 1–6 5. Minami Y, Mishima S, Oda J. Prediction of the level of consciousness using pupillometer measurements in patients with impaired consciousness brought to the emergency and critical care center. Acute Med Surg (2020.0) 7 e537. DOI: 10.1002/ams2.537 6. Yang X, Wang T, Ren X, Yu W. Survey on improving data utility in differentially private sequential data publishing. IEEE Trans Big Data (2017.0) 7 729 7. Liang F, Yu W, An D. A survey on big data market: pricing, trading and protection. IEEE Access (2018.0) 6 15132-15154. DOI: 10.1109/ACCESS.2018.2806881 8. Stankovic JA. Research directions for the internet of things. IEEE Internet Things J (2014.0) 1 3-9. DOI: 10.1109/JIOT.2014.2312291 9. Siam AI, Almaiah MA, Al-Zahrani A. Secure health monitoring communication systems based on IoT and cloud computing for medical emergency applications. Comput Intell Neurosci (2021.0) 2021 1-23. DOI: 10.1155/2021/8016525 10. Sujith AVLN, Sajja GS, Mahalakshmi V. Systematic review of smart health monitoring using deep learning and artificial intelligence. Neurosci Inform (2022.0) 2 100028. DOI: 10.1016/j.neuri.2021.100028 11. Magi N, Prasad BG. Activity monitoring for ICU patients using deep learning and image processing. SN Comput Sci (2020.0) 1 123. DOI: 10.1007/s42979-020-00147-6 12. Chen X-W, Lin X. Big data deep learning: challenges and perspectives. IEEE Access (2014.0) 2 514-525. DOI: 10.1109/ACCESS.2014.2325029 13. Nguyen ND, Nguyen T, Nahavandi S. System design perspective for human-level agents using deep reinforcement learning: a survey. IEEE Access (2017.0) 5 27091-27102. DOI: 10.1109/ACCESS.2017.2777827 14. Siam AI, Abou Elazm A, El-Bahnasawy NA. Smart health monitoring system based on IoT and cloud computing. Menoufia J Electron Eng Res (2019.0) 28 37-42. DOI: 10.21608/mjeer.2019.76711 15. Wu X, Liu C, Wang L, Bilal M. Internet of things-enabled real-time health monitoring system using deep learning. Neural Comput Appl (2021.0). DOI: 10.1007/s00521-021-06440-6 16. 16.Papernot N, McDaniel P, Sinha A, Wellman M (2016) Towards the science of security and privacy in machine learning. arXiv Prepr arXiv:161103814 17. 17.Papernot N, McDaniel P, Sinha A, Wellman MP (2018) Sok: security and privacy in machine learning. In: 2018 IEEE European Symposium on Security and Privacy (EuroS&P). IEEE, pp 399–414 18. Li X, He J, Vijayakumar P. A verifiable privacy-preserving machine learning prediction scheme for edge-enhanced HCPSs. IEEE Trans Ind Inform (2021.0) 18 5494-5503. DOI: 10.1109/TII.2021.3110808 19. Deo RC. Machine learning in medicine. Circulation (2015.0) 132 1920-1930. DOI: 10.1161/CIRCULATIONAHA.115.001593 20. Myszczynska MA, Ojamies PN, Lacoste AMB. Applications of machine learning to diagnosis and treatment of neurodegenerative diseases. Nat Rev Neurol (2020.0) 16 440-456. DOI: 10.1038/s41582-020-0377-8 21. Vellido A. Societal issues concerning the application of artificial intelligence in medicine. Kidney Dis (2019.0) 5 11-17. DOI: 10.1159/000492428 22. Buch VH, Ahmed I, Maruthappu M. Artificial intelligence in medicine: current trends and future possibilities. Br J Gen Pract (2018.0) 68 143-144. DOI: 10.3399/bjgp18X695213 23. Rajula HSR, Verlato G, Manchia M. Comparison of conventional statistical methods with machine learning in medicine: diagnosis, drug development, and treatment. Medicina (B Aires) (2020.0) 56 455. DOI: 10.3390/medicina56090455 24. Siam AI, Sedik A, El-Shafai W. Biosignal classification for human identification based on convolutional neural networks. Int J Commun Syst (2021.0). DOI: 10.1002/dac.4685 25. Huang L, Shea AL, Qian H. Patient clustering improves efficiency of federated machine learning to predict mortality and hospital stay time using distributed electronic medical records. J Biomed Inform (2019.0) 99 103291. DOI: 10.1016/j.jbi.2019.103291 26. 26.Detection D, Infections C-, Sedik A, et al (2020) Deploying machine and deep learning models for efficient data-augmented detection of COVID-19 infections 27. El-Rashidy N, Abdelrazik S, Abuhmed T. Comprehensive survey of using machine learning in the covid-19 pandemic. Diagnostics (2021.0) 11 1155. DOI: 10.3390/diagnostics11071155 28. Alballa N, Al-Turaiki I. Machine learning approaches in COVID-19 diagnosis, mortality, and severity risk prediction: a review. Inform Med Unlocked (2021.0) 24 100564. DOI: 10.1016/j.imu.2021.100564 29. Islam MN, Inan TT, Rafi S. A systematic review on the use of AI and ML for fighting the COVID-19 pandemic. IEEE Trans Artif Intell (2020.0) 1 258-270. DOI: 10.1109/TAI.2021.3062771 30. Alyasseri ZAA, Al-Betar MA, Doush IA. Review on COVID-19 diagnosis models based on machine learning and deep learning approaches. Expert Syst (2022.0). DOI: 10.1111/exsy.12759 31. Qin J, Chen L, Liu Y. A machine learning methodology for diagnosing chronic kidney disease. IEEE Access (2020.0) 8 20991-21002. DOI: 10.1109/ACCESS.2019.2963053 32. Kate RJ, Perez RM, Mazumdar D. Prediction and detection models for acute kidney injury in hospitalized older adults. BMC Med Inform Decis Mak (2016.0) 16 39. DOI: 10.1186/s12911-016-0277-4 33. Safdar S, Zafar S, Zafar N, Khan NF. Machine learning based decision support systems (DSS) for heart disease diagnosis: a review. Artif Intell Rev (2018.0) 50 597-623. DOI: 10.1007/s10462-017-9552-8 34. Ahsan MM, Siddique Z. Machine learning-based heart disease diagnosis: a systematic literature review. Artif Intell Med (2022.0) 128 102289. DOI: 10.1016/j.artmed.2022.102289 35. Mahmood H, Shaban M, Rajpoot N, Khurram SA. Artificial Intelligence-based methods in head and neck cancer diagnosis: an overview. Br J Cancer (2021.0) 124 1934-1940. DOI: 10.1038/s41416-021-01386-x 36. Saxena S, Gyanchandani M. Machine learning methods for computer-aided breast cancer diagnosis using histopathology: a narrative review. J Med Imaging Radiat Sci (2020.0) 51 182-193. DOI: 10.1016/j.jmir.2019.11.001 37. Zou Q, Qu K, Luo Y. Predicting diabetes mellitus with machine learning techniques. Front Genet (2018.0). DOI: 10.3389/fgene.2018.00515 38. Khalil H, El-Hag N, Sedik A. Classification of diabetic retinopathy types based on convolution neural network (CNN). Menoufia J Electron Eng Res (2019.0) 28 126-153. DOI: 10.21608/mjeer.2019.76962 39. Kassem MA, Hosny KM, Damaševičius R, Eltoukhy MM. Machine learning and deep learning methods for skin lesion classification and diagnosis: a systematic review. Diagnostics (2021.0) 11 1390. DOI: 10.3390/diagnostics11081390 40. Orooji A, Kermani F. Machine learning based methods for handling imbalanced data in hepatitis diagnosis. Front Heal Inform (2021.0) 10 57. DOI: 10.30699/fhi.v10i1.259 41. Spann A, Yasodhara A, Kang J. Applying machine learning in liver disease and transplantation: a comprehensive review. Hepatology (2020.0) 71 1093-1105. DOI: 10.1002/hep.31103 42. Alharbey R, Dessouky MM, Sedik A. Fatigue state detection for tired persons in presence of driving periods. IEEE Access (2022.0). DOI: 10.1109/ACCESS.2022.3185251 43. Teasdale G, Jennett B. Assessment of coma and impaired consciousness. A practical scale. Lancet (1974.0) 304 81-84. DOI: 10.1016/S0140-6736(74)91639-0 44. Sternbach GL. The glasgow coma scale. J Emerg Med (2000.0) 19 67-71. DOI: 10.1016/S0736-4679(00)00182-7 45. Teasdale G, Jennett B. Assessment and prognosis of coma after head injury. Acta Neurochir (Wien) (1976.0) 34 45-55. DOI: 10.1007/BF01405862 46. Estévez-Báez M, Machado C, García-Sánchez B. Autonomic impairment of patients in coma with different glasgow coma score assessed with heart rate variability. Brain Inj (2019.0) 33 496-516. DOI: 10.1080/02699052.2018.1553312 47. Young AJ, Hare A, Subramanian M. Using machine learning to make predictions in patients who fall. J Surg Res (2021.0) 257 118-127. DOI: 10.1016/j.jss.2020.07.047 48. 48.Durga S, Nag R, Daniel E (2019) Survey on machine learning and deep learning algorithms used in internet of things (IoT) healthcare. In: 2019 3rd International Conference on Computing Methodologies and Communication (ICCMC). IEEE, pp 1018–1022 49. Tang KJW, Ang CKE, Constantinides T. Artificial intelligence and machine learning in emergency medicine. Biocybern Biomed Eng (2021.0) 41 156-172. DOI: 10.1016/j.bbe.2020.12.002 50. Kim J, Chang H, Kim D. Machine learning for prediction of septic shock at initial triage in emergency department. J Crit Care (2020.0) 55 163-170. DOI: 10.1016/j.jcrc.2019.09.024 51. Liu N, Zhang Z, Wah Ho AF, Ong MEH. Artificial intelligence in emergency medicine. J Emerg Crit Care Med (2018.0) 2 82-82. DOI: 10.21037/jeccm.2018.10.08 52. Raita Y, Goto T, Faridi MK. Emergency department triage prediction of clinical outcomes using machine learning models. Crit Care (2019.0) 23 64. DOI: 10.1186/s13054-019-2351-7 53. Gravesteijn BY, Nieboer D, Ercole A. Machine learning algorithms performed no better than regression models for prognostication in traumatic brain injury. J Clin Epidemiol (2020.0) 122 95-107. DOI: 10.1016/j.jclinepi.2020.03.005 54. Matsuo K, Aihara H, Nakai T. Machine learning to predict in-hospital morbidity and mortality after traumatic brain injury. J Neurotrauma (2020.0) 37 202-210. DOI: 10.1089/neu.2018.6276 55. 55.Tsiklidis EJ, Sims C, Sinno T, Diamond SL (2020) Using the National Trauma Data Bank (NTDB) and machine learning to predict trauma patient mortality at admission. PLoS One 15:166. 10.1371/journal.pone.0242166 56. Hall AN, Weaver B, Liotta E. Identifying modifiable predictors of patient outcomes after intracerebral hemorrhage with machine learning. Neurocrit Care (2021.0) 34 73-84. DOI: 10.1007/s12028-020-00982-8 57. Amorim RL, Oliveira LM, Malbouisson LM. Prediction of early TBI mortality using a machine learning approach in a LMIC population. Front Neurol (2020.0). DOI: 10.3389/fneur.2019.01366 58. de Toledo P, Rios PM, Ledezma A. Predicting the outcome of patients with subarachnoid hemorrhage using machine learning techniques. IEEE Trans Inf Technol Biomed (2009.0) 13 794-801. DOI: 10.1109/TITB.2009.2020434 59. Kamruzzaman MM, Alanazi S, Alruwaili M. Fuzzy-assisted machine learning framework for the fog-computing system in remote healthcare monitoring. Measurement (2022.0) 195 111085. DOI: 10.1016/j.measurement.2022.111085 60. Tuli S, Basumatary N, Gill SS. HealthFog: an ensemble deep learning based smart healthcare system for automatic diagnosis of heart diseases in integrated IoT and fog computing environments. Futur Gener Comput Syst (2020.0) 104 187-200. DOI: 10.1016/j.future.2019.10.043 61. Abdulkareem KH, Mohammed MA, Gunasekaran SS. A Review of fog computing and machine learning: concepts, applications, challenges, and open issues. IEEE Access (2019.0) 7 153123-153140. DOI: 10.1109/ACCESS.2019.2947542 62. Ali ZH, Badawy MM, Ali HA. A novel geographically distributed architecture based on fog technology for improving Vehicular Ad hoc Network (VANET) performance. Peer-to-Peer Netw Appl (2020.0) 13 1539-1566. DOI: 10.1007/s12083-020-00910-9 63. Ali ZH, Hagras S, Ali HA. Distributed computing architecture using fog technology for improving intelligent transportation systems in smart city. Int J Comput Appl (2021.0) 183 42-45. DOI: 10.5120/ijca2021921351 64. Verma P, Tiwari R, Hong W-C. FETCH: a deep learning-based fog computing and IoT integrated environment for healthcare monitoring and diagnosis. IEEE Access (2022.0) 10 12548-12563. DOI: 10.1109/ACCESS.2022.3143793 65. Kishor A, Chakraborty C, Jeberson W. A novel fog computing approach for minimization of latency in healthcare using machine learning. Int J Interact Multimed Artif Intell (2021.0) 6 7. DOI: 10.9781/ijimai.2020.12.004 66. Sudqi Khater B, Abdul Wahab AW, Bin IMYI. A lightweight perceptron-based intrusion detection system for fog computing. Appl Sci (2019.0) 9 178. DOI: 10.3390/app9010178 67. Hu P, Dhelim S, Ning H, Qiu T. Survey on fog computing: architecture, key technologies, applications and open issues. J Netw Comput Appl (2017.0) 98 27-42. DOI: 10.1016/j.jnca.2017.09.002 68. Karakus M, Durresi A. Quality of service (QoS) in software defined networking (SDN): a survey. J Netw Comput Appl (2017.0) 80 200-218. DOI: 10.1016/j.jnca.2016.12.019 69. Shaukat U, Ahmed E, Anwar Z, Xia F. Cloudlet deployment in local wireless networks: motivation, architectures, applications, and open challenges. J Netw Comput Appl (2016.0) 62 18-40. DOI: 10.1016/j.jnca.2015.11.009 70. Zhang P, Zhou M, Fortino G. Security and trust issues in fog computing: a survey. Futur Gener Comput Syst (2018.0) 88 16-27. DOI: 10.1016/j.future.2018.05.008 71. Hou X, Li Y, Chen M. Vehicular fog computing: a viewpoint of vehicles as the infrastructures. IEEE Trans Veh Technol (2016.0) 65 3860-3873. DOI: 10.1109/TVT.2016.2532863 72. Liu H, Cocea M. Nature-inspired framework of ensemble learning for collaborative classification in granular computing context. Granul Comput (2019.0) 4 715-724. DOI: 10.1007/s41066-018-0122-5 73. El-Sappagh S, Alonso JM, Islam SMR. A multilayer multimodal detection and prediction model based on explainable artificial intelligence for Alzheimer’s disease. Sci Rep (2021.0) 11 1-26. DOI: 10.1038/s41598-021-82098-3 74. Vellido A. The importance of interpretability and visualization in machine learning for applications in medicine and health care. Neural Comput Appl (2020.0) 32 18069-18083. DOI: 10.1007/s00521-019-04051-w 75. Fawagreh K, Gaber MM, Elyan E. An outlier ranking tree selection approach to extreme pruning of random forests. Commun Comput Inf Sci (2016.0) 629 267-282. DOI: 10.1007/978-3-319-44188-7_20 76. Ameixieira C, Cardote A, Neves F. Harbornet: a real-world testbed for vehicular networks. IEEE Commun Mag (2014.0) 52 108-114. DOI: 10.1109/MCOM.2014.6894460 77. 77.Openfog reference architecture for fog computing. https://www.openfogconsortium.org/. Accessed 17 Apr 2020 78. Johnson EWA, Pollard TJ, Shen L. Data descriptor: MIMIC-III, a freely accessible critical care database. Thromb Haemost (2016.0) 76 258-262. DOI: 10.1038/sdata.2016.35 79. 79.Johnson A, Pollard T, Mark R MIMIC-III Clinical Database v1.4 80. Caicedo-Torres W, Gutierrez J. ISeeU: visually interpretable deep learning for mortality prediction inside the ICU. J Biomed Inform (2019.0) 98 1-24. DOI: 10.1016/j.jbi.2019.103269 81. Adams RP, Mayaud L, Poincare HR. A physiological time series dynamics-based approach to patient monitoring and outcome prediction. IEEE J Biomed Heal Inform (2015.0) 19 1068-1076. DOI: 10.1109/JBHI.2014.2330827.A 82. El-rashidy N, El-sappagh S, Abuhmed T, Abdelrazek S, El-Bakry HM. Intensive care unit mortality prediction: an improved patient-specific stacking ensemble model. IEEE Access (2020.0). DOI: 10.1109/ACCESS.2020.3010556 83. 83.El-rashidy N, El-sappagh S, Abdelrazik S, El-bakry H (2022) Ensemble machine learning model model for mortality prediction inside intensive care unit. Springer International Publishing 84. Greco L, Luta G, Krzywinski M, Altman N. Analyzing outliers: robust methods to the rescue. Nat Methods (2019.0) 16 275-276. DOI: 10.1038/s41592-019-0369-z 85. Moon TK. The expectation-maximization algorithm. IEEE Signal Process Mag (1996.0) 13 47-60. DOI: 10.1109/79.543975 86. 86.Joenssen DW, Bankhofer U (2015) Hot deck methods for imputing missing data hot deck methods for imputing missing data the effects of limiting donor usage10.1007/0097836.4231.53746 87. Caballero-Ruiz E, García-Sáez G, Rigla M. A web-based clinical decision support system for gestational diabetes: automatic diet prescription and detection of insulin needs. Int J Med Inform (2017.0) 102 35-49. DOI: 10.1016/j.ijmedinf.2017.02.014 88. 88.Wright J (2018) Glasgow coma scale. pp 1–2 89. Cook N. The glasgow coma scale. Crit Care Nurs Clin North Am (2020.0). DOI: 10.1016/j.cnc.2020.10.005 90. Ko J, Deprez D, Shaw K. Stretching is superior to brisk walking for reducing blood pressure in people with high-normal blood pressure or stage I hypertension. J Phys Act Health (2020.0) 18 21-28. DOI: 10.1123/jpah.2020-0365 91. Berntson GG, Bigger JTJ, Eckberg DL. Heart rate variability: origins, methods, and interpretive caveats. Psychophysiology (1997.0) 34 623-648. DOI: 10.1111/j.1469-8986.1997.tb02140.x 92. Andršová I, Hnatkova K, Šišáková M. Influence of heart rate correction formulas on QTc interval stability. Sci Rep (2021.0) 11 1-21. DOI: 10.1038/s41598-021-93774-9 93. van der Ven WH, Schuurmans J, Schenk J. Monitoring, management, and outcome of hypotension in Intensive care unit patients, an international survey of the European Society of intensive care medicine. J Crit Care (2022.0) 67 118-125. DOI: 10.1016/j.jcrc.2021.10.008 94. Al-Rashed F, Sindhu S, Al Madhoun A. Elevated resting heart rate as a predictor of inflammation and cardiovascular risk in healthy obese individuals. Sci Rep (2021.0) 11 13883. DOI: 10.1038/s41598-021-93449-5 95. Chan NC, Li K, Hirsh J. Peripheral oxygen saturation in older persons wearing nonmedical face masks in community settings. JAMA (2020.0) 324 2323-2324. DOI: 10.1001/jama.2020.21905 96. Tapio J, Vähänikkilä H, Kesäniemi YA. Higher hemoglobin levels are an independent risk factor for adverse metabolism and higher mortality in a 20-year follow-up. Sci Rep (2021.0) 11 1-13. DOI: 10.1038/s41598-021-99217-9 97. Lee SH, Kim M, Do HK, Lee JH. Low hemoglobin levels and an increased risk of psoriasis in patients with chronic kidney disease. Sci Rep (2021.0) 11 1-7. DOI: 10.1038/s41598-021-94165-w 98. Forman JP, Rifas-Shiman SL, Taylor EN. Association between the serum anion gap and blood pressure among patients at Harvard vanguard medical associates. J Hum Hypertens (2008.0) 22 122-125. DOI: 10.1038/sj.jhh.1002286 99. Anestis DM, Tsitsopoulos PP, Foroglou NG. Cross-cultural adaptation and validation of the greek version of the “full outline of unresponsiveness score”: a prospective observational clinimetric study in neurosurgical patients. Neurocrit Care (2021.0). DOI: 10.1007/s12028-021-01342-w 100. Schmidt WU, Lutz M, Ploner CJ, Braun M. The diagnostic value of the neurological examination in coma of unknown etiology. J Neurol (2021.0) 268 3826-3834. DOI: 10.1007/s00415-021-10527-4 101. Andalib S, Lattanzi S, Di Napoli M. Blood pressure variability: a new predicting factor for clinical outcomes of intracerebral hemorrhage. J Stroke Cerebrovasc Dis Off J Natl Stroke Assoc (2020.0) 29 105340. DOI: 10.1016/j.jstrokecerebrovasdis.2020.105340 102. Wagner R, Heni M, Tabák AG. Pathophysiology-based subphenotyping of individuals at elevated risk for type 2 diabetes. Nat Med (2021.0) 27 49-57. DOI: 10.1038/s41591-020-1116-9 103. El-Rashidy N, Abuhmed T, Alarabi L. Sepsis prediction in intensive care unit based on genetic feature optimization and stacked deep ensemble learning (2021.0) 104. Zheng Q, Delingette H, Ayache N. Explainable cardiac pathology classification on cine MRI with motion characterization by semi-supervised learning of apparent flow. Med Image Anal (2019.0) 56 80-95. DOI: 10.1016/j.media.2019.06.001 105. Lee H, Yune S, Mansouri M. An explainable deep-learning algorithm for the detection of acute intracranial haemorrhage from small datasets. Nat Biomed Eng (2019.0) 3 173-182. DOI: 10.1038/s41551-018-0324-9 106. Jiménez-Luna J, Grisoni F, Schneider G. Drug discovery with explainable artificial intelligence. Nat Mach Intell (2020.0) 2 573-584. DOI: 10.1038/s42256-020-00236-4 107. Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nat Mach Intell (2019.0) 1 206-215. DOI: 10.1038/s42256-019-0048-x 108. Gulum MA, Trombley CM, Kantardzic M. A review of explainable deep learning cancer detection models in medical imaging. Appl Sci (2021.0). DOI: 10.3390/app11104573 109. Kakogeorgiou I, Karantzalos K. Evaluating explainable artificial intelligence methods for multi-label deep learning classification tasks in remote sensing. Int J Appl Earth Obs Geoinf (2021.0) 103 102520. DOI: 10.1016/j.jag.2021.102520 110. Petch J, Di S, Nelson W. Opening the black box: the promise and limitations of explainable machine learning in cardiology. Can J Cardiol (2021.0). DOI: 10.1016/j.cjca.2021.09.004
--- title: Semaglutide in renal ischemia-reperfusion injury in mice authors: - Al-Tameemi Tiba - Heider Qassam - Najah Rayish Hadi journal: Journal of Medicine and Life year: 2023 pmcid: PMC10015556 doi: 10.25122/jml-2022-0291 license: CC BY 3.0 --- # Semaglutide in renal ischemia-reperfusion injury in mice ## Abstract Ischemia and reperfusion injury (I/R) is a serious condition leading to organ failure, characterized by poor blood supply followed by rapid resuscitation of blood flow and reoxygenation. Renal failure caused by renal ischemia has high mortality and morbidity. This study aimed to explore the potential role of Semaglutide as a novel and effective therapeutic strategy for acute renal failure. Additionally, we aimed to assess the possible protective effect of Semaglutide on kidney I/R injury in mice through modulation of the inflammatory and oxidative pathways via phosphatidylinositol 3-kinase/adenosine triphosphate (PI3K/AKT) activation. We employed twenty-eight albino mice to induce the I/R injury model by clamping the renal artery for 30 min followed by a period of reperfusion for 2 hours. The control group was exposed to I/R injury, while the Semaglutide-treated group was pretreated with the drug 12 hours before induction of ischemia at a dose of 100 nmol/L/kg via the intraperitoneal route (i.p). In addition, the DMSO-treated group was subjected to similar conditions to the Semaglutide-treated group. At the end of the experiments, kidneys and blood samples were collected for investigation. Semaglutide could act as a protective agent against acute kidney injury by reducing inflammatory molecules such as tumor necrosis factor-alpha (TNF-α) and its cognate receptor, TNF-α R, interleukine-6 (IL-6). Furthermore, Semaglutide reduced F8 isoprostane levels, increased PI3K and AKT levels in renal tissues, and mitigated renal damage. Semaglutide had renoprotective effects via modulation of the inflammatory response and oxidative pathway by targeting the PI3K/AKT signaling pathway. ## INTRODUCTION Ischemia and reperfusion injury (I/R) of the kidney is a growing public health concern worldwide, particularly in low-income and middle-income countries. It occurs following a period of poor blood supply followed by rapid resuscitation of blood flow and reoxygenation, which can cause organ damage [1]. The causes of renal I/R injury can include sepsis, infarction, renal transplantation, and unilateral nephrectomy [2]. Several signaling pathways contribute to renal damage following I/R, increasing morbidity and mortality [3]. The PI3K-Akt signaling pathway has emerged as an area of interest among the various molecular mechanisms involved in the pathogenesis of renal ischemia and reperfusion injury. PI3K-Akt axis plays a critical role in multiple biological activities such as inflammation, oxidative stress, and chemotaxis; hence, it can be a key pathway in regulating the biological responses against acute kidney injury [4]. Furthermore, PI3K-Akt signaling has been found to be activated following I/R, resulting in increased proliferation and cell viability in renal tubules [5]. Semaglutide is a well-known antihyperglycemic agent used for treating type 2 diabetes, which binds to the glucagon-like peptide (GLP-1) receptor, resulting in increased insulin release and body weight reduction [6]. However, it is uncertain whether Semaglutide has a role in the PI3K-Akt signaling pathway involved in acute kidney injury following I/R. Therefore, this study aimed to investigate the potential renoprotective effects of Semaglutide against renal I/R. ## MATERIAL AND METHODS A total of twenty Swiss albino male mice, aged 17-18 weeks and weighing 30-40g, were randomly assigned into four groups, with five mice in each group. The first group was the sham group, where mice were anesthetized with ketamine and xylazine and underwent laparotomy without clamping. The second group was the I/R group, in which mice were anesthetized and subjected to bilateral renal ischemia for 30 min and 2 h reperfusion by clamping and releasing the renal arteries, respectively. The third group received a dimethyl sulfoxide (DMSO) (vehicle for Semaglutide) injection via the intraperitoneal route (i.p.) 12 hours before ischemia. These mice underwent laparotomy under anesthesia, followed by bilateral renal arteries clamping for 30 min and 2 hours of reperfusion by releasing the clamps. The fourth group was pretreated with Semaglutide at a dose of 100 nmol/L/kg i.p 12 hours prior to the induction of ischemia and reperfusion. At the end of the experiments, the mice were sacrificed, and blood samples were taken directly from the heart to measure the serum levels of tumor necrosis factor-alpha (TNF-α), interleukin-6 (IL-6), TNF-α receptor, urea, creatinine, and tissue levels of PI3K/AKT and F8-isoprostane. The kidneys were fixed in a $10\%$ formalin solution for further investigation to assess the severity of kidney tissue damage. ## Blood and tissue sampling At the end of the experiment, blood was collected from the heart, and enzyme-linked immunosorbent assay (ELISA) was used to measure TNF-α, TNF-α R, IL-6, urea, and creatinine in the serum according to the manufacturer’s instructions. Blood samples were centrifuged at 3000 rpm for 10 min without the use of an anticoagulant to prepare the serum. Kidneys were washed with phosphate buffer saline (PBS) and weighted in a ratio of 1 to 10 W/V to prepare homogenates. The homogenization was carried out using a high-intensity ultrasonic liquid processor with PBS containing $1\%$ of protease inhibitor cocktail and $1\%$ of Triton X-100 [7]. ## Tissue sampling for histopathology scoring The removed kidney was fixed in $10\%$ formalin, processed using standard histological procedures, and embedded in a paraffin block [7]. Kidney tissues were sliced to a thickness of five micrometers and stained with hematoxylin and eosin (H&E). The tissue sections were investigated by an independent histopathologist using a benchtop microscope. The severity of kidney damage involving changes in the tubular cell, such as necrosis, hypertrophy, glomerular atrophy, and inflammatory cell infiltration, was used to identify the extent of the renal damage. Scores were assigned for these variables: (zero) for the sham group considered normal and not subjected to I/R, $25\%$ for mild damage, $50\%$ for moderate damage, $75\%$ for severe damage, and $75\%$-$100\%$ for extremely severe damage [8]. ## Statistical analysis Statistical analysis was conducted using GraphPad Prism 8 software, and the data were presented as mean ± SEM (standard error of the mean). For multiple comparisons, an analysis of variance (ANOVA) followed by Tukey's test was performed. A significance level of $P \leq 0.05$ was considered statistically significant. ## Effects of renal I/R on renal function Figure 1 shows that serum urea and creatinine levels were significantly elevated in the control and DMSO groups compared to the sham group. However, in the Semaglutide pretreatment group, these levels were significantly reduced, indicating that Semaglutide may have protective properties against renal I/R injury. Figure 1: The mean serum level of urea and creatinine (mg/dl) in the four experimental groups, data are expressed as mean ± SEM; P<0.05 sham vs. control; #P<0.05 Semaglutide vs. I/R mice.* ## Semaglutide inhibits tissue levels of TNF-α, TNF-α R, and IL-6 The impact of renal I/R on inflammatory mediators was assessed in different groups. The serum levels of TNF-α, TNF-α R, and IL-6 were significantly increased ($p \leq 0.05$) in the control group as compared to the sham group (Figure 2). In contrast, these levels were significantly decreased in the Semaglutide-treated group (Figure 2), indicating that Semaglutide may have anti-inflammatory effects in the context of renal I/R injury. Figure 2: The mean serum level of TNF-α, TNF-αR, and IL-6 (pg/ml) among four experimental groups. Data are expressed as mean ± SEM; P<0.05 sham vs. control; #P<0.05 Semaglutide vs. I/R mice.* ## Semaglutide reduces F8 isoprostane (F8IsoP) F8 isoprostane (F8IsoP) is a widely recognized biomarker of oxidative stress and is considered a critical indicator of organ injury. As shown in Figure 3, the levels of F8IsoP in the I/R group were significantly higher than those in the sham group, indicating increased oxidative stress in the kidneys following I/R injury. However, treatment with Semaglutide led to a marked reduction in F8IsoP levels, suggesting that Semaglutide may have protective effects against oxidative stress in the context of renal I/R injury. Figure 3: The mean level of tissue F8 ISOP (pg/ml) in the four experimental groups, data are expressed as mean ± SEM; P<0.05 sham vs. control; # P <0.05 Semaglutide vs. I/R mice.* ## Phosphatidylinositol-3-kinase in a polymerase chain reaction (PCR) To investigate the expression of PI3K in renal tissue, PCR analysis was performed, and the results are shown in Figure 4. The levels of PI3K were significantly higher in the control group than in the sham group, as indicated by ∆CT values. However, treatment with Semaglutide resulted in a marked reduction in PI3K expression levels, with values returning to those observed in the sham group ($p \leq 0.05$) (Figure 4). Figure 4: The mean expression level of renal tissue PI3K in the four experimental groups. Data are expressed as mean ± SEM; P<0.05 sham vs. control; # P<0.05 Semaglutide vs. I/R mice.* ## Immunohistochemistry study The immunoreactivity of Akt antibody was examined to investigate Akt protein expression in renal tissue following I/R injury. As shown in Figure 5, the Akt protein levels were significantly lower in the I/R injury group than in the sham group ($p \leq 0.05$), indicating reduced Akt protein expression in response to I/R injury. However, treatment with Semaglutide resulted in a significant increase in Akt protein levels compared to the I/R injury group ($p \leq 0.05$), as depicted in Figure 5 A–D. Figure 5: Data in are expressed as mean ± SEM; P<0.05 sham vs. control; #P<0.05 Semaglutide vs. I/R mice. A: Representative Akt immunohistochemistry (IHC) staining in renal tissue of mice in the sham group. The result shows the normal expression level of Akt in renal tissue; B: Representative Akt IHC staining in renal tissue of mice in the control group. The result shows the low expression level of this protein in renal tissue; C: Representative Akt IHC staining in renal tissue of mice in the DMSO group. The result shows the low expression level of this protein in renal tissue; D: Representative Akt IHC staining in renal tissue of the Semaglutide group. The result shows the high expression level of this protein in renal tissue.* ## Effects of semaglutide on renal histology The histological analysis of renal tissue following I/R injury revealed severe damage characterized by congestion, thickening of blood vessels, inflammation, and hemorrhage, as compared to the normal structures observed in the sham group (Figures 6 A–D and 7, Table 1). However, treatment with Semaglutide resulted in a significant reduction in the severity of renal damage, as evidenced by a lower damage score ($p \leq 0.05$), as shown in Figures 6 A–D and 7, Table 1. Figure 6: A: The histopathological section of renal tissue in the sham group shows normal glomerular texture (black arrows). The tissue was stained using H&E stain and examined using a light microscope and digital camera at a 10X magnification scale. B: The histopathological section of renal tissue in the control group (I/R) shows clear congestion in the blood vessels (black arrow), thickening of the blood vessel wall, and reduced glomerular size (glomerular atrophy, blue arrows). The section shows the infiltration of pinkish structures and homogenous material (amyloid degeneration, yellow arrow) as well as the infiltration of inflammatory cells, primarily neutrophils, in the renal tissue (red arrow). The tissue was stained using H&E stain and examined using a digital camera and light microscope at a 10X magnification scale. C: The histopathological section of renal tissue in the DMSO group shows clear congestion in the blood vessels (black arrow) and thickening of the blood vessel wall. The section shows inflammatory cell infiltration, mainly neutrophils in the renal tissue (red arrow), and there is apparent hemorrhage in the renal tissue near the blood vessels (yellow arrow). The tissue was stained using H&E stain and examined using a digital camera and light microscope at a 10X magnification scale. D: The histopathological examination of renal tissue in the Semaglutide group shows mild hypertrophy of renal tubules (black arrows) and mild atrophy of the renal glomeruli with an increase in the Bowman's space (blue arrows). The tissue was stained using H&E stain and examined using a light microscope and digital camera at a 10X magnification scale. Figure 7: Renal damage scores in the four groups were examined and scored according to a protocol previously described [8]. All data are expressed as Mean ± SEM ($$n = 5$$), P≤0.05 versus sham, # P≤0.05 Semaglutide versus* I/R mice.* TABLE_PLACEHOLDER:Table 1 ## DISCUSSION Acute kidney injury is a significant clinical problem that affects populations worldwide and is associated with high morbidity and mortality rates. Despite advances in understanding the pathogenesis and biomarkers involved in acute kidney injury, there is currently limited knowledge about the fundamental mechanisms behind the deterioration of renal function following injury [9]. Renal I/R is one of the critical players that contribute to the development of acute kidney injury via its role in various biological responses, including inflammation, cell apoptosis, and free radical accumulation [10]. The present study found that levels of urea and creatinine in the serum were elevated in mice subjected to renal I/R compared to the sham group, consistent with previous research indicating that 30 min of ischemia followed by 2 h of blood resuscitation increased levels of creatinine and alanine aminotransferase [11]. These are critical indicators of renal function that are elevated in response to injury. In contrast, treatment with Semaglutide resulted in a marked decrease in levels of serum urea and creatinine in comparison with the renal I/R injury group suggesting that the drug has a protective influence against kidney injury. These findings are consistent with other research indicating that GLP-1 receptor agonists may have renoprotective effects through diuretic and natriuretic properties, as well as a reduction in renal angiotensin II [12]. The results of this study indicate that serum levels of IL-6, TNF-α, and TNF-α receptors were significantly increased in mice exposed to I/R compared to the sham group. These findings are consistent with earlier studies that reported increased levels of these biomarkers in response to I/R injury [13,14]. It is well known that I/R injury leads to many events, including inflammatory responses characterized by increased inflammatory cytokines and macrophage recruitment leading to renal damage [15]. Treatment with Semaglutide reversed the earlier results and reduced these markers in comparison with the I/R group. In accordance with the present results, previous studies have demonstrated that activating GLP-1 receptor by Liraglutide markedly reduced expression of IL-6, TNF-α, TLR2, and TLR4, and mitigated the renal damage and ameliorated the kidney function suggesting that GLP-1 receptor agonists may be potential candidates against renal injury [16]. In this study, we observed a significant increase in levels of F8-isoprostane in the mice subjected to I/R injury compared to the sham group. This finding is consistent with previous studies that have reported F8-isoprostane as a critical biomarker of oxidative damage and an important contributor to tissue injury following I/R injury [17]. In contrast, the treatment with Semaglutide significantly decreased the levels of F8-isoprostane in the renal tissues, indicating a potential impact on oxidative stress. Previous research has shown that GLP-1 receptor activation can reduce platelet activity, a critical factor in lipid peroxidation and inflammation. Therefore, GLP-1 receptor agonists like Semaglutide may exert antioxidant effects [18]. The most notable finding from this study was the significant reduction in PI3K expression levels in the renal tissues of the I/R injury group compared to the sham group, as indicated by ∆CT values (Figure 4). These findings seem consistent with other research, which revealed that I/R resulted in a modulation in PI3K/Akt signaling by decreasing levels of PI3K expression in the kidney [19]. By contrast, treatment with Semaglutide markedly increased levels of PI3K expression and preserved these levels close to the sham group. These results are consistent with recent studies indicating that activating the GLP-1 receptor can reverse the decreased levels of both PI3K and Akt in rat models of I/R injury, suggesting that GLP-1 may play a crucial role in promoting cell survival via modulating the PI3K/Akt axis [20]. This signaling pathway is essential in maintaining renal function, as documented in previous studies [15, 21]. It has inhibitory influences on a variety of biomarkers, such as proinflammatory cytokines and apoptotic molecules that contribute to the pathogenesis of renal injury [22]. The current study revealed that mice subjected to I/R had significantly higher levels of renal tissue damage when compared to the sham group. These findings are consistent with previous studies that reported similar features of renal tissue damage following I/R injury, including tubular cell swelling, pyknotic nuclei, cellular vacuolization, congestion, and cellular necrosis [23,24]. In contrast, treatment with Semaglutide notably reduced the degree of damage in the renal tissues suggesting its renoprotective impact. The results of this research support earlier observations, which showed that Semaglutide mitigated renal injury and reduced the glomerulosclerosis index in a mouse model of hypertension-induced diabetic kidney disease [25]. ## CONCLUSION The findings of this study suggest that Semaglutide may have renoprotective effects by inhibiting inflammatory and oxidative stress mediators and activating the PI3K/AKT pathway. Further studies are needed to fully elucidate the mechanisms behind these effects and to evaluate the clinical application of Semaglutide in the treatment of kidney injury. ## Conflict of interest The authors declare no conflict of interest. ## Ethical approval This study was approved by the Animal Care and Research Committee of the University of Kufa (approval: AEC: 5 February 2022). ## Authorship TTA contributed to data collection, draft writing, and the final paper. HQ contributed to draft editing, revising, and data analysis. NRH played a key role in developing the main idea for the study, providing critical statistical revision, and contributing to the critical final writing of the paper. ## References 1. Malek M, Nematbakhsh M. Renal ischemia/reperfusion injury; from pathophysiology to treatment. J Renal Inj Prev (2015) 4 20-27. DOI: 10.12861/jrip.2015.06 2. Jang HR, Rabb H. The innate immune response in ischemic acute kidney injury. Clin Immunol (2009) 130 41-50. DOI: 10.1016/j.clim.2008.08.016 3. Li Y, Xu B, Yang J, Wang L, Tan X, Hu X. Liraglutide protects against lethal renal ischemia-reperfusion injury by inhibiting high-mobility group box 1 nuclear-cytoplasmic translocation and release. Pharmacological Research (2021) 173 105867. DOI: 10.1016/j.phrs.2021.105867 4. Barthel A, Klotz L-O. Phosphoinositide 3-kinase signaling in the cellular response to oxidative stress. (2005) 386 207-216. DOI: 10.1515/BC.2005.026 5. Liu H-B, Meng Q-H, Huang C, Wang J-B, Liu X-W. Nephroprotective Effects of Polydatin against Ischemia/Reperfusion Injury: A Role for the PI3K/Akt Signal Pathway. Oxidative Medicine and Cellular Longevity (2015) e362158. DOI: 10.1155/2015/362158 6. Knudsen LB, Lau J. The Discovery and Development of Liraglutide and Semaglutide. Frontiers in Endocrinology (2019) 10. DOI: 10.3389/fendo.2019.00155 7. Wei Q, Zhao J, Zhou X, Yu L. Propofol can suppress renal ischemia-reperfusion injury through the activation of PI3K/AKT/mTOR signal pathway. Gene (2019) 708 14-20. DOI: 10.1016/j.gene.2019.05.023 8. Qiu Y, Wu Y, Zhao H, Sun H, Gao S. Maresin 1 mitigates renal ischemia/reperfusion injury in mice via inhibition of the TLR4/MAPK/NF-κB pathways and activation of the Nrf2 pathway. Drug design development and therapy (2019) 13 739. DOI: 10.2147/DDDT.S188654 9. Han SJ, Lee HT. Mechanisms and therapeutic targets of ischemic acute kidney injury. Kidney Res Clin Pract (2019) 38 427-440. DOI: 10.23876/j.krcp.19.062 10. Moellmann J, Klinkhammer BM, Onstein J, Stöhr R. Glucagon-like peptide 1 and its cleavage products are renoprotective in murine diabetic nephropathy. Diabetes (2018) 67 2410-9. DOI: 10.2337/db17-1212 11. Awad AS, Kamel R, Sherief M-AE. Effect of thymoquinone on hepatorenal dysfunction and alteration of CYP3A1 and spermidine/spermine N-1-acetyl-transferase gene expression induced by renal ischemia–reperfusion in rats. Journal of Pharmacy and Pharmacology (2011) 63 1037-42. DOI: 10.1111/j.2042-7158.2011.01303.x 12. Kim MK, Kim DM. Effects of glucagon-like peptide-1 receptor agonists on kidney function and safety in type 2 diabetes patients. Journal of Diabetes Investigation (2021) 12 914. DOI: 10.1111/jdi.13552 13. Rajan DP. Cellular and molecular derangements in acute tubular necrosis. Current opinion in pediatrics (2005) 17 193-9. DOI: 10.1097/01.mop.0000152620.59425.eb 14. Mohammed TJ, Hadi NR, Al-Yasiri I, Yousif NG. Critical role of Ghrelin in downregulation of the inflammatory response after renal injury. Vascular Investigation and Therapy (2018) 1 68 15. Wei Q, Dong Z. Mouse model of ischemic acute kidney injury: technical notes and tricks. American Journal of Physiology-Renal Physiology (2012) 303 F1487-F94. DOI: 10.1152/ajprenal.00352.2012 16. Choi EK, Jung H, Kwak KH, Yi SJ. Inhibition of oxidative stress in renal ischemia-reperfusion injury. Anesthesia & Analgesia (2017) 124 204-13. DOI: 10.1213/ANE.0000000000001565 17. Legrand M, Mik EG, Johannes T, Payen D, Ince C. Renal hypoxia and dysoxia after reperfusion of the ischemic kidney. Molecular medicine (2008) 14 502-16. DOI: 10.2119/2008-00006 18. Simeone P, Liani R, Tripaldi R, Di Castelnuovo A. Thromboxane-dependent platelet activation in obese subjects with prediabetes or early type 2 diabetes: effects of Liraglutide-or lifestyle changes-induced weight loss. Nutrients (2018) 10 1872. DOI: 10.3390/nu10121872 19. Christensen M, Dalbøge LS, Secher T, Gravesen Salinas C. MO069: Therapeutic Effects of Semaglutide as Mono and Combination Treatment with Lisinopril in a Mouse Model of Hypertension-Accelerated Diabetic Kidney Disease. Nephrology Dialysis Transplantation (2022) 37 gfac063-21 20. Zhai R, Xu H, Hu F, Wu J. Exendin-4, a GLP-1 receptor agonist regulates retinal capillary tone and restores microvascular patency after ischemia–reperfusion injury. British Journal of Pharmacology (2020) 177 3389-3402. DOI: 10.1111/bph.15059 21. Zhang G, Wang Q, Zhou Q, Wang R. Protective effect of tempol on acute kidney injury through PI3K/Akt/Nrf2 signaling pathway. Kidney and Blood Pressure Research (2016) 41 129-38. DOI: 10.1159/000443414 22. Mohammed TJ, Al-Yasiri I, Jasim A, Ahmed AA, Hadi NR. Nephroprotective Potential Effects of Ghrelin in Renal Ischemia-Reperfusion Injury in Rats. World Heart Journal (2017) 9 293-301. DOI: 10.22038/AJP.2022.19620 23. Hussien YA, Abdalkadim H, Mahbuba W, Hadi NR. The Nephroprotective effect of lycopene on renal ischemic reperfusion injury: a mouse model. Indian Journal of Clinical Biochemistry (2020) 35 474-81. DOI: 10.1007/s12291-019-00848-7 24. Ling H, Chen H, Wei M, Meng X. The effect of autophagy on inflammation cytokines in renal ischemia/reperfusion injury. Inflammation (2016) 39 347-56. DOI: 10.1007/s10753-015-0255-5 25. Dalbøge LS, Christensen M, Madsen MR, Secher T. Nephroprotective Effects of Semaglutide as Mono-and Combination Treatment with Lisinopril in a Mouse Model of Hypertension-Accelerated Diabetic Kidney Disease. Biomedicines (2022) 10 1661. DOI: 10.3390/biomedicines10071661
--- title: Influence of Murraya koenigii extract on diabetes induced rat brain aging authors: - Lakshmi Bhupatiraju - Krupavaram Bethala - Khang Wen Goh - Jagjit Singh Dhaliwal - Tan Ching Siang - Shasidharan Menon - Bamavv Menon - Kishore Babu Anchu - Siok Yee Chan - Long Chiau Ming - Abdullah Khan journal: Journal of Medicine and Life year: 2023 pmcid: PMC10015565 doi: 10.25122/jml-2022-0151 license: CC BY 3.0 --- # Influence of Murraya koenigii extract on diabetes induced rat brain aging ## Abstract Food supplements are used to improve cognitive functions in age-related dementia. This study was designed to determine the Murraya koenigii leaves’ effect on Alloxan-induced cognitive impairment in diabetic rats and the contents of oxidative stress biomarkers, catalase, reduced glutathione, and glutathione reductase in brain tissue homogenates. Wistar rats were divided into seven groups (six rats per group). Group I received saline water (1 ml, p.o.), Diabetes was induced in Groups II–VII with Alloxan (120 mg/kg/p.o). Group III was provided with Donepezil HCl (2.5 mg/kg/p.o.), Group IV, V, VI, and VII with Murraya koenigii ethanol extract (200 and 400 mg/kg/p.o.) and aqueous extract (200 and 400 mg/kg/p.o.), respectively, for 30 days. Behavior, acetylcholinesterase (AChE) activity, oxidative stress status, and histopathological features were determined in the hippocampus and cerebral cortex. Administration of Murraya koenigii ethanolic and aqueous extracts significantly ($P \leq 0.05$, $P \leq 0.001$) increased the number of holes crossed by rats from one chamber to another. There was an increase in the [1] latency to reach the solid platform, [2] number of squares traveled by rats on the 30th day, and [3] percentage of spontaneous alternation behavior compared to the control group. Administration for successive days markedly decreased AChE activity ($P \leq 0.05$), decreased TBARS level, and increased catalase, GSH, and GR levels. Murayya koenigii could be a promising food supplement for people with dementia. However, more research into sub-chronic toxicity and pharmacokinetic and pharmacodynamics interactions is essential. ## INTRODUCTION Diabetes mellitus (DM) is a very common chronic disease where there is a gradual deterioration of organs in the body and is associated with microvascular and macrovascular complications. One of the later complications is cognitive decline and dementia, representing a serious problem in the elderly population. Cognitive impairment in diabetic patients requires immediate research solutions. It should also be addressed along with microvascular complications such as neuropathy, nephropathy, retinopathy, and cardiovascular complications. Alloxan is a diabetogenic agent used in diabetes research to induce insulin reduction. It acts by aggregating the pancreatic β-cells via the Glut2 glucose transporter and demolishes them through reactive oxygen species (ROS) and free radicals mechanisms [1]. Oxidative stress plays an important role in the enhancement of diabetes complications, including learning and memory impairments, as a result of the increased generation of free radicals and diminished antioxidant defenses [2]. These free radicals lead to increased neuronal death in several brain areas, including the hippocampus, and DNA damage, through protein oxidation and peroxidation of membrane lipids [3]. Since ancient times medicinal plants and their chemical constituents have been widely used to cure or mitigate diseases. India is well-known for its vast medicinal plant biodiversity. Murraya koenigii, for example, includes several bioactive components [4], as a result based on which the plant has been demonstrated to be medicinally essential, but it has received little or no attention from scientists. Murraya koenigii (L.) Spreng. ( Family: Rutaceae) is known as Curry Leaf in English, Mitha Neem or Kadi Patta in Hindi, Surabhinimba in Sanskrit, and Karuveppilei in Tamil. It is commonly utilized as a spice and condiment in India and has been demonstrated to have natural healing properties [5]. Plant-derived (phyto) carbazole alkaloids are an important class of compounds present in the family of Rutaceae (Genera Murraya, Clausena, Glycosmis, Micromelum, and Zanthoxylum). Due to several significant biological activities, such as antitumor, antibacterial, antiviral, antidiabetic, anti-HIV, and neuroprotective activities of the parent skeleton (3-methylcarbazole), carbazole alkaloids are recognized as an important class of potential therapeutic agents. Since ancient times roots and leaves of Murraya koenigii are traditionally used to treat various GIT disorders. They are known to promote appetite, treat nausea, and control flatulence, diarrhea, and dysentery. Moreover, Murraya koenigii is also used to relieve pain, reduce fever, cancer, and hemorrhoids, and acts as an antidote against animal bites [5]. The Murraya koenigii leaf extracts are used to treat diabetes. Phytoconstituents like carbazole alkaloids, glycosides, flavonoids, minerals, and volatile oil are found in this plant [6]. Murraya koenigii can be used directly or in a variety of forms, including extracts and essential oils. The presence of active constituents such as bismahanine, murrayanine, murrayafoline-A, bi-koeniquinone-A, bismurrayaquinone, mukoenine-A, mukoenine-B, mukoenine-C, murrastifoline, Murrayazolinol, murrayacine, murrayazolidine, murrayazoline, mahanimbine, girinimbine, koenioline, xynthyletin, koenigine-Quinone A and koenigine-Quinone B make it highly valuable [7-10]. Preliminary studies reported that the plant has anti-diabetic and neuroprotective activities [11, 12]. Based on our previous preliminary studies, Murraya koenigii (L.) Spreng leaf extracts have shown significant neuroprotective activity in the aluminum-induced cognitive deficits model [13]. Hence, we have designed the current study to evaluate the cognitive enhancing potential of Murraya koenigii (L.) Spreng leaf extracts in Alloxan-induced cognitive decline and brain tissue oxidative stress via behavioral and biochemical study in rats. ## Drugs and chemicals All the chemicals and reagents used in the study were of analytical grade, respectively Donepezil hydrochloride (Incepta Pharmaceuticals Ltd. Dhaka, Bangladesh) and Alloxan monohydrate (Explicit Chemicals, Pvt. Ltd. Pune, India). A blood glucometer (Bayer Healthcare, India) was used. ## Plant collection & identification Murrayya koengii leaves were procured from the Maisammaguda area, Hyderabad, and the sample was authenticated at the Botanical department, Osmania University, Hyderabad, and voucher specimen: MK0152 was deposited. ## Preparation of ethanolic extract The collected plant material was air-dried and made into a coarse powder using a grinder. 100 g of leaf powder was filled in soxhlet apparatus and subjected to ethanol ($95\%$) extraction. After 72 hours, the mixture was filtered, concentrated under reduced pressure to obtain a semisolid extract, and stored in an air-tight flask in the refrigerator for later use. ## Preparation of aqueous extract The dried plant material of Murraya koenigii was powdered. Chloroform and water were added in 1:9 ratios into the conical flask. The powder was poured into the flask and subjected to frequent agitation at 10 minutes intervals for 48 hours. Later the powder was filtered with the help of a muslin cloth. Liquid filtrates are concentrated and evaporated to dryness using a rotary evaporator under reduced pressure to get the crude extract (10.57 g) stored in the refrigerator for further evaluation. The extracts were subjected to preliminary phytochemical identification by the standard procedures given in “Practical Pharmacognosy” by C.K. Kokate. The doses of donepezil and alloxan were selected according to the literature review [14-16]. The doses of Murraya koenigii extracts were selected according to our previous studies [17, 18]. ## Selection and maintenance of animals Adult Wistar Albino rats of either sex, weighing between 150 and 250 grams, were used. The animals were kept in conventional polypropylene cages at room temperature and given ad libitum access to food and water. ## Induction of experimental Diabetes Mellitus and experimental design Rats fasted overnight with free access to water before the experiment. Alloxan monohydrate (120 mg/kg b.w. in 0.9 % cold normal saline) was given with a single intraperitoneal injection. The rats were given a $5\%$ glucose solution to prevent hypoglycemia [19]. Forty-two Wistar Albino rats weighing between 200gms-250gms were chosen for the study. Animals were divided into seven groups of six animals each. Rats of groups II-VII were injected with alloxan to induce diabetes. Blood glucose levels were analyzed using a digital glucometer, and the blood sample was collected from the animal's tail. Animals with blood glucose levels< 200 mg/dl were considered hypoglycemic. Groups III-VII rats were continued with respective treatments: Group 1: The control group received normal saline water;Group 2: Alloxan monohydrate (120mg/kg, i.p.) was administered to rats;Group 3: Alloxan monohydrate (120mg/kg, i.p.) + Donepezil hydrochloride (2.5mg/kg, p.o.);Group 4: Alloxan monohydrate (120mg/kg, i.p.) + ethanolic extract of Murraya koenigii (200mg/kg, p.o.);Group 5: Alloxan monohydrate (120mg/kg, i.p.) + ethanolic extract of Murraya koenigii (400mg/kg, p.o.);Group 6: Alloxan monohydrate (120mg/kg, i.p.) + aqueous extract of Murraya koenigii (200mg/kg, p.o.);Group 7: Alloxan monohydrate (120mg/kg, i.p.) + aqueous extract of Murraya koenigii (400mg/kg, p.o.). The treatment duration was for 30 days. All the doses were given through the oral route by oral lavage. After completion of respective treatments for 30 days, the rats were subjected to evaluation of the influence of Murraya koenigii ethanolic and aqueous extracts on behavioral parameters for assessment of cognitive decline induced by diabetes. ## Hole cross test The experiment was carried out with a wooden box (30cm×20cm×1cm). A fixed partition in the center of the box and a hole (3 cm in diameter) were available at eight of 7.5 cm from the base. On the day of the experiment, every rat was placed on one side of the apparatus; spontaneous movement from one chamber to another through the hole was observed for 3 minutes on the 28th day [20]. ## Passive Avoidance test The experiment was performed in an apparatus that consisted of one dark and one light chamber and was divided by a wall. On day one, i.e., acquisition trail, every rat was first placed in the light chamber and then into a dark chamber, and an electric shock (40V, 0.5mA for 1 second) was delivered to the feet of the rat through the grid floor. After the training session, the rat was immediately returned to the cage [21]. On day two, the animal was placed again in the light chamber and the time taken to access the dark chamber was recorded as step-through latency. If the animal did not enter the dark chamber within a 5-minutes test period, the test was terminated, and the step-through latency was recorded as 300 seconds. ## Morris water maze test The equipment consisted of a circular tank, and a platform was placed 2 cm below the water level. Milk was added to make the water opaque so that the immersed platform was not visible. The tank was divided into four quadrants. During the training sessions, the animal was trained to find the submerged platform. On the day of the experiment, the latency from immersion into the pool to escape onto the hidden platform (maximum duration of 90 seconds) was recorded [22]. ## Y-maze test The y-maze test is commonly employed to assess short-term memory in rats. The equipment consisted of three uniform arms 40 cm long, 12 cm high, 5 cm wide at the bottom, and 10 cm wide at the top, and the arms were separated by 120°. It also consisted of a central equilateral triangle area for the rat to enter into any of the arms. On the 30th day, each rat was placed in one of the arms and was permitted to explore for 8 minutes. Spontaneous alternation behavior was noted, i.e., entry into all three arms on successive choices. Then the sequence and number of arm entries were recorded manually. The percentage of spontaneous alternation behavior was calculated according to the following formula [23]: Na is the number of alternations, and Nabc is the total number of arm entries. ## Open field test The apparatus consisted of a wooden box, and the floor was divided into 25 (5x5) squares. The rats were placed into one corner of the open field chamber, and their behavior was observed for 5 minutes. The following observations were made- a) the number of squares explored, the number of central nine squares and peripheral 16 adjacent squares to the wall explored, b) the number of grooming, and c) the number of rearing [24]. ## Estimation of oxidative stress biomarkers At the end of the experiment, animals were euthanized on the 30th day by CO2 inhalation through the euthanasia chamber and then the brain was isolated carefully and washed with saline solution. The brain was then minced into small pieces and homogenized with phosphate buffer (0.1M, pH-7.4) using a tissue homogenizer to obtain 1:9 (w/v) ($10\%$) whole homogenate. Then the homogenates were centrifuged at 4000 rpm at 4°C for 20 minutes using Remi cool centrifuge, and the resultant supernatant was collected and stored for estimation of tissue antioxidant parameters like malondialdehyde [25], glutathione reductase, and reduced glutathione [26] catalase [27] and brain acetylcholinesterase activity [28]. The brain was excised after the experiment and fixed in formalin ($10\%$ v/v). The tissue was processed, sections were cut, and the slides were prepared and stained with Haematoxylin and Eosin, examined under high power microscope (100x), (400x), and photomicrographs were taken. ## Statistical analysis Values are expressed as mean ± SEM, using t-test, the intergroup variation between various groups was conducted by graph pad Prism software & data were analyzed by one-way analysis of variance (ANOVA), and $P \leq 0.05$ was considered to be statistically significant. ## Effect of Murraya koenigii extracts on hole cross test in Alloxan-induced diabetic rats Alloxan-induced diabetic group exhibited a significant decrease (^$P \leq 0.0001$) in the number of holes crossed from chamber to chamber in comparison to the control group. Murraya koenigii and donepezil-treated groups also restored the Alloxan-induced cognitive impairment in rats ($P \leq 0.0001$) compared with the diabetic group, which signifies progress in spatial memory and learning (Figure 1). Figure 1:* Effect of Ethanol and Aqueous extract of Murraya koenigii leaves in Alloxan induced cognitive impairment in diabetic rats using hole cross test. Values are given as mean±SEM, n=6. Using t-test, variations between groups were done by graph pad Prism software, & Data was assessed by one-way ANOVA ^P<0.0001 vs. control group, #p<0.001, p<0.05, *p<0.001 vs. Alloxan-induced diabetic group. ## Effect of Murraya koenigii extracts on passive avoidance test in Alloxan-induced diabetic rats Alloxan induced diabetic group showed a significant decrease (^$P \leq 0.0001$) in the time taken to step down from the solid platform onto the grid floor when compared to the control group. Murraya koenigii extract-treated groups showed an increase ($P \leq 0.0001$ and $P \leq 0.0001$) in the time latency when compared with the Alloxan-treated group (Figure 2). Figure 2:* Effect of Ethanol and Aqueous extract of Murraya koenigii leaves on a step down latency in Alloxan-induced cognitive impairment in diabetic rats using Passive avoidance test. Values are given as mean±SEM, n=6. Using t-test variations between groups were done by graph pad Prism software, & Data was assessed by one-way ANOVA ^P<0.0001 vs. control group, #p< 0.001, p<0.05, *p<0.001 vs. Alloxan-induced diabetic group. ## Effect of Murraya koenigii extracts in Morris water maze test in Alloxan-induced diabetic rats Alloxan-induced diabetic group showed a significant increase (^$P \leq 0.0007$) in the time taken to reach the solid platform when compared to the control group. Murraya koenigii-treated groups showed a dose-dependent decrease ($P \leq 0.0001$, $P \leq 0.0001$) as compared with the Alloxan-treated group (Table 1). Table 1* \| Treatments \| Quadrant 1 (Latency in sec) \| Quadrant 2 (Latency in sec) \| Quadrant 3 (Latency in sec) \| Quadrant 4 (Latency in sec) \| \| --- \| --- \| --- \| --- \| --- \| \| Control \| 22.4±0.37 \| 25.6±0.50 \| 28.3±0.38 \| 26.1±0.30 \| \| Alloxan 120 mg/kg, i.p \| 14.2±0.35^ \| 15.4±0.50^ \| 18.2±0.58^ \| 16.2±0.58^ \| \| Donepezil HCL 2.5 mg/kg, p.o. \| 45.1±0.64# \| 42.6±0.34# \| 45.3±0.41# \| 49.7±0.41# \| \| M. koenigii Et.extract 200 mg/kg, p.o. \| 35.7±0.67* \| 37.6±0.69* \| 39.3±0.40* \| 40±0.40** \| \| M. koenigii Et.extract 400 mg/kg, p.o. \| 37.2±0.511* \| 36.4±0.41* \| 40.2±0.15* \| 41.2±0.15** \| \| M. koenigii Aq.extract 200 mg/kg, p.o. \| 34.6±0.73* \| 36.7±0.32* \| 33.2±0.39* \| 36.9±0.26* \| \| M. koenigii Aq.extract 400 mg/kg, p.o. \| 38.2±0.50* \| 37.5±0.45* \| 39.6±0.62* \| 36.9±0.54* \| ## Effect of Murraya koenigii extracts on Y-maze test in Alloxan-induced diabetic rats In Murraya koenigii ($P \leq 0.05$, $P \leq 0.01$) and donepezil ($P \leq 0.001$) treated groups, there was an increase in the percentage of spontaneous alternation behavior as compared with alloxan-induced diabetic rats. This signifies that there is an enhancement in spatial short-term memory and learning (Table 2). Table 2 \| Treatments \| % Spontaneous alterations \| \| --- \| --- \| \| Control \| 69.15±0.40 \| \| Alloxan 120 mg/kg (i.p) \| 35.24±0.31^ \| \| Donepezil HCL 2.5 mg/kg (oral) \| 70.01±0.42# \| \| M. koenigii Et.extract 200 mg/kg (oral) \| 42.32±0.35 \| \| M. koenigii Et.extract 400 mg/kg (oral) \| 75.10±0.56 \| \| M. koenigii Aq.extract 200 mg/kg (oral) \| 47.05±0.35* \| \| M. koenigii Aq.extract 400 mg/kg (oral) \| 74.82±0.55* \| ## Effect of Murraya koenigii extracts on an open field test in Alloxan-induced diabetic rats Alloxan-induced diabetic group showed a significant decrease (^$P \leq 0.0001$) in the activities and number of squares explored when compared to the control group. Donepezil-treated rats showed a significant ($P \leq 0.001$) increase as compared to the Alloxan-induced diabetic rats. Murraya koenigii treated groups showed a significant increase ($P \leq 0.001$ and $P \leq 0.0001$) in the activities and squares explored compared to Alloxan treated group (Table 3). Table 3* \| Treatments \| No. of Central squares crossed \| No. of Peripheral squares crossed \| No. of Groomings \| No. of Rearings \| \| --- \| --- \| --- \| --- \| --- \| \| Control \| 103.2±1.02 \| 107.5±0.35 \| 105.2±0.32 \| 106.6±0.53 \| \| Alloxan treated 120 mg/kg \| 80.2±0.89^ \| 78.6±0.71^ \| 70.5±0.27^ \| 63.2±0.37^ \| \| Donepezil HCL 2.5 mg/kg, p.o. \| 125.5±0.72# \| 130.6±0.33# \| 140.3±0.24# \| 138.2±0.35# \| \| M. koenigii Et.extract 200 mg/kg,p.o \| 105.2±0.66* \| 109±0.30* \| 113.2±0.75* \| 115.6±0.68* \| \| M. koenigii Et.extract 400 mg/kg, p.o \| 107±0.6 \| 117±0.60 \| 119.3±0.51 \| 120.2±0.54 \| \| M. koenigii Aq.extract 200 mg/kg, p.o. \| 103.9±0.21* \| 105.6±0.63* \| 115.6±0.63* \| 119.6±0.82* \| \| M. koenigii Aq.extract 400 mg/kg, p.o. \| 108±0.52 \| 119±0.56 \| 120.5±0.34 \| 123.2±0.49 \| ## Effect of Murraya koenigii extracts on the rats’ tissue antioxidant parameters The oxidative lipid damage was high in the Alloxan-induced diabetic rats, as indicated by significantly increased MDA levels ($P \leq 0.0001$) as compared with the Murraya koenigii extracts treated group. Results presented here indicate that Alloxan-induced cognitive decline was significantly attenuated by Murraya koenigii extracts. There was a significant decrease in catalase GSH and GR levels in Alloxan-treated groups ($P \leq 0.0001$) in comparison to normal control rats. Murraya koenigii ethanol and aqueous 200 and 400mg/kg treated groups showed a significant increase ($P \leq 0.0001$) compared to Alloxan-treated groups (Table 4). Table 4 \| Treatments \| Catalase (nmoles/mg protein) \| MDA (µmoles/mg protein) \| GSH (µmoles/mg protein) \| GR (µmoles/mg protein) \| \| --- \| --- \| --- \| --- \| --- \| \| Control \| 42.9±0.24 \| 42.9±0.32 \| 41.2±0.36 \| 23.0±0.54 \| \| Alloxan 120 mg/kg (i.p) \| 31.6±0.42^ \| 79.60.69±^ \| 22.6±0.47^ \| 16.4±0.37^ \| \| Donepezil HCL 2.5 mg/kg (p.o.) \| 41.6±0.34# \| 38.0±0.31# \| 54.2±0.27# \| 20.4±0.40# \| \| M. koenigii Et.extract 200 mg/kg (p.o) \| 35.9±0.92** \| 64.6±0.31* \| 40±0.44** \| 18.1±0.38* \| \| M. koenigii Et.extract 400 mg/kg (p.o) \| 41.0±0.44* \| 43.9±0.61** \| 53.5±0.42* \| 21.2±0.32* \| \| M. koenigii Aq.extract 200 mg/kg (p.o) \| 33.3±0.38* \| 53.6±0.18* \| 31.4±0.29* \| 17.4±0.28* \| \| M. koenigii Aq.extract 400 mg/kg (p.o) \| 45.6±0.45* \| 45.9±0.59** \| 52.6±0.49* \| 25.9±0.32* \| ## Effects of Murraya koenigii extracts on Acetylcholinesterase Activity (AChE) on rat brain Alloxan-induced cognitive decline was evidenced by a significant increase in AChE activity ($p \leq 0.001$) in the hippocampus and cortex region ($p \leq 0.01$). Treatment with Murraya koenigii extracts significantly prevented the diabetes-induced increase in AChE activity ($p \leq 0.001$), which was comparable to the standard donepezil hydrochloride treated group (Figure 3). Figure 3: Effect of Ethanol and Aqueous extract of Murraya koenigii leaves on Acetylcholinesterase activity in Alloxan induced cognitive impairment in diabetic rats. Values are given as mean±SEM, n=6. Using t-test variations between groups was done by graph pad Prism software, & Data was assessed by one-way ANOVA ^P<0.0001 vs. control group, #p< 0.001 p<0.05, *p<0.001 vs. Alloxan-induced diabetic group. ## Effect of Murraya koenigii extracts on rat brain histopathology In normal control animals, no neuronal changes were observed whereas in Alloxan-induced diabetic rats, there was neurodegeneration and vacuolated cytoplasm. These changes were not observed in both the hippocampus or cerebral cortex regions in the Murraya koenigii extracts treated group (Figures 4 and 5). Figure 4: Effect of Murraya koenigii leaves extracts on the histological structure of the brain (cerebral cortex): A – The control group-it shows Dentate gyrus appeared normal. The normal cortex of cerebral hemispheres. Meninges appeared normal. B – Alloxan-induced diabetic group-it shows foci of necrosis noticed in the cerebral cortex region of the brain. C– Donepezil Hcl 2.5 mg/kg group shows frontal cortex appeared normal. D – Murraya koenigii ethanol 200mg/kg group shows mild degenerative changes noticed near the ventricles. E – Murraya koenigii ethanol 400 mg/kg group-it shows (a) Cerebellum appeared normal, (b) normal Purkinje cells, (c) normal granular layer, (d) white matter normal. F – Murraya koenigii Aq extract 200 mg/kg group-mild meningitis observed in the meningeal surrounding cerebral hemispheres. G – Murraya koenigii Aqueous extract 400 mg/kg-it showed dentate gyrus appeared normal. Tissues were stained with Hematoxylin and Eosin at magnification 100X. Figure 5: Effect of Murraya koenigii leaves extracts on the histological structure of the brain (hippocampus): A – The control group-it shows hippocampus appeared normal. B – Alloxan-induced diabetic group-foci of necrosis along with infiltration of inflammatory cells are noticed. C – Donepezil Hcl 2.5 mg/kg group-hippocampus appeared normal. D – Murraya koenigii ethanol 200 mg/kg group shows mild degenerative changes noticed near the ventricles. E – Murraya koenigii ethanol 400 mg/kg group Mild infiltration glial cells, normal hippocampus. F – Murraya koenigii Aq extract 200 mg/kg group-mild meningitis observed. G – Murraya koenigii Aqueous extract 400 mg/kg-normal ventricles of brain observed, normal hippocampus. Tissues were stained with Hematoxylin and Eosin at magnification 100X. ## DISCUSSION In this research, Murraya koenigii ethanolic and aqueous extracts were given to Alloxan-induced rats with cognitive dysfunction and oxidative stress for 30 days. The results demonstrate improved memory and learning. The movement of the animals when placed in a new chamber indicates spontaneous motor activity, and an increase in spontaneous motor activity indicates the level of CNS activity and hence the nootropic effect. In the Hole cross model, administration of the ethanolic and aqueous extracts increased the entries of a rat from a light area to a dark area. The fact that the frequency of entries has increased shows an enhancement in the learning and spatial memory of rats in comparison to Alloxan-treated cognitive deficit rats. Similar results were reported by Sherman et al. [ 29]. Ethanol extract showed a significant increase in the number of entries as compared to aqueous extract. The results are in correlation to the study conducted by Rao et al. Ethanolic extract showed $80\%$ scavenging activity [30]. Open Field test is a valid model to screen the upgradation of learning and memory in rodent models as evidenced by improvement in movement activity. There was an increase in the number of squares crossed, indicating an enhancement of cognitive activity. Kishore et al. study on Foeniculum vulgare fruit extract in mice model revealed a significant increase in the number of squares crossed [31]. The Morris water maze is a reliable model for assessing memory and learning. Alloxan-induced diabetes resulted in a significant deterioration in cognitive performance. In comparison to normal control animals, there was an increase in latency to reach the hidden platform. The latency is a measurement of the signal's integration and association processes during the acquisition phase, and it is related to the time it takes for the input signal to reach the concerned brain structures (synaptic delays). Murraya koenigii ethanolic and aqueous extracts alleviated diabetic rats' cognitive deficits, resulting in a significant improvement in escape latency and a distinct increase in platform quadrant proportion and similar results reported by Biessels et al. [ 32]. The YM behavioral test is used to investigate short-term memory, general motor activity, and stereotypic behavior [33]. Spatial memory is studied by spontaneous alternation. In this test, the percentage of spontaneous alternation behavior was studied. A rise in the percentage spontaneous alternation behavior of Murraya koenigii ethanolic and aqueous extracts treated rats suggests an improved spatial long-term memory and learning as compared with the illness control group [34]. Ethanolic extract at 400mg/kg showed significant activity comparable to standard donepezil. The results of this research showed notable improvement in spatial working memory, spatial working-reference memory, and spatial reference memory. This indicated a nootropic activity of Murraya koenigii ethanolic and aqueous extracts in diabetes-induced cognitive disability models. This activity may be attributed to the presence of phytoconstituents like flavonoids, anthocyanin glycosides, triterpenoids, and phytosterols. As oxidative stress increases, free radicals interact with neuronal cell membranes and cause lipid peroxidation, and there is increased production of TBARS. This increased lipid peroxidation end products accumulate in the neurons, causing degeneration and leading to Alzheimer's disease and cognitive deficit. There was an improvement in brain activity during treatment with Murraya koenigii ethanolic and aqueous extracts for 30 days, as evidenced by a decrease in lipid peroxidation activity ($P \leq 0.001$) [35]. Later complications of oxidative damage to various brain regions constitute morphological abnormalities and memory impairments. There was a decrease in GSH, GR, and CAT levels in the brains of diabetic-induced rats. Treatment with Murraya koenigii ethanolic and aqueous extracts significantly reduced the levels of TBARS and increased the GSH, GR, and CAT levels. The antioxidant properties of Murraya koenigii ethanolic and aqueous extracts might help to ameliorate cognitive dysfunction in diabetic animals. Similar findings were presented by Husna et al. [ 36]. The acetylcholinesterase levels are high in diabetics; ‘this enzyme hydrolyses acetylcholine present in the brain and results in cognitive decline’ [37]. There was a significant increase in acetylcholinesterase activity in the current study. Chronic administration of Murraya koenigii ethanolic and aqueous extracts prevented an increase in acetylcholinesterase activity. Hence, it can be inferred that chronic administration could prevent cognitive decline [38], cholinergic dysfunction, and reduction in oxidative injury in Alloxan-induced diabetic animals [39]. Notably, Murraya koenigii leaves ethanolic and aqueous extracts had successfully improved cholinesterase action in the brain. Nutriment consisting of Murraya koenigii leaves notably augment memory level and decreases cognitive impairment with dosage concentration stimulated by hyoscine and diazepam in juvenile and elderly female rats [40]. Further studies elucidating the clear mode of action and the safety characteristics of poly-phytonutrient of Murraya koenigii leaves are needed before the development of its botanical drug for the clinical indication of neuroprotection. ## CONCLUSION In the present investigation, it can be inferred that chronic administration could prevent cognitive decline, cholinergic dysfunction, and reduction in oxidative injury in Alloxan-induced diabetic animals. Results are in accordance with the neurological justification for the traditional use of Murraya koenigii extracts in the treatment of neurodegenerative disorders, specifically Alzheimer's disease. ## Conflicts of interest The authors declare no conflict of interest. ## Ethics approval Committee approval was obtained prior to the commencement of the study (MRCP/IAEC//2008/Re/S/10). ## Personal thanks The authors are thankful to the authorities of Malla Reddy College of Pharmacy, Secunderabad, for providing support to this study. ## Author contributions Data curation was performed by LB. Formal analysis was conducted by LB and KWG. KWG, CST, AK, SM, BM,KBA, and LCM were in charge of funding acquisition. LB, TCS, AK, KBA, and S-YC performed the investigation, while LB, KB, KBA, and S-YC developed the methodology. TCS and LCM directed the project administration, while TCS, AK, SM, BM, and KBA were in charge of the resources. KWG, TCS,SM, and KB performed data analysis. S-YC and JSD supervised the research, and LB, KB, and KWG performed validation. LB, KB, AK, SM, S-YC, and JSD contributed to visualization, and LB, KB, KBA, and S-YC wrote the original draft; LB,TCS, AK, SM, BM, S-YC, JSD, and LCM contributed to reviewing and editing the manuscript. ## References 1. Chatzigeorgiou A, Halapas A, Kalafatakis K, Kamper E. The use of animal models in the study of Diabetes Mellitus in vivo. (2009.0) 23 245-258 2. Ceriello A. New insights on oxidative stress and diabetic complications may lead to a “causal” antioxidant therapy. Diabetes Care (2003.0) 26 1589-1596. DOI: 10.2337/diacare.26.5.1589 3. Rahman K. Studies on free radicals, antioxidants, and co-factors. Clinical Interventions in Aging (2007.0) 2 219-236. PMID: 18044138 4. Balakrishnan R, Vijayraja D, Jo SH, Ganesan P. Medicinal Profile, Phytochemistry, and Pharmacological Activities of. Antioxidants (Basel) (2020.0) 9 101. DOI: 10.3390/antiox9020101 5. Kirtikar KR, Basu BD. “Indian Medicinal Plants” Popular Prakashan Dehradun. (1999.0) 1 195-196 6. Math MV, Balasubramaniam P. The hypoglycaemic effect of curry leaves (. Indian J Physiol Pharmacol (2005.0) 49 241-52. PMID: 16170995 7. Jain M, Gilhotra R, Singh RP. Curry leaf (. MOJ Biol Med (2017.0) 2 236-256. DOI: 10.15406/mojbm.2017.02.00050 8. Anwer F, Masaldan AS, Kapil RS. Synthesis of Murrayacine; oxidation with DDQ of the activated aromatic methyl group of the alkaloids of. Indian J Chem (1973.0) 11 1314-1315 9. Gupta P, Nahata A, Vinod K. An update on Murayya Koenigii: a multifunctional Ayurvedic herb. Journal of Chinese integrative medicine (2012.0) 9 824-833 10. Raghunathan K, Mitra R. Pharmacognosy of indigenous drugs. Central Council for Research in Ayurveda and Siddha (1985.0) 1 433 11. Prabhu KA, Tamilanban T. Investigation of anti-diabetic activity of stem of. International Journal of Research in Pharmacology and Pharmacotherapeutics (2012.0) 1 165-168 12. Ahmad Kartini. Chemical constituents of. Malaysia: Universiti Putra Malaysia (1999.0) 1-25 13. Maheswari Reddy B, Dhanpal CK, Lakshmi BVS. Anti-Alzheimer's Activity of aqueous extract of leaves of. Research J. Pharm. and Tech (2019.0) 12 1-8. DOI: 10.5958/0974-360X.2019.00328.8 14. Narimatsu N, Harada N, Kurihara H, Nakagata N. Donepezil improves cognitive function in mice by increasing the production of insulin-like growth factor-I in the hippocampus. The Journal of Pharmacology and Experimental Therapeutics (2009.0) 330 1-10. DOI: 10.1124/jpet.108.147280 15. Wang Y, Wang S, Cui W, He J. Olive leaf extract inhibits lead poisoning-induced brain injury. Neural Regen Res (2013.0) 8 2021-2029. DOI: 10.1002/CHIN.197423423 16. Etuk EU. Animals models for studying diabetes mellitus. AgricBiol J N Am (2010.0) 1 130-134 17. Maheswari Reddy B, Dhanpal CK, Lakshmi BVS. Evaluation of Anti-Alzheimer’s activity of Acorus calamus in aluminium chloride induced neurotoxicity in rats. Journal of Global trends in Pharmaceutical sciences (2020.0) 11 7920-7924. DOI: 10.5958/0974-360X.2019.00323.8 18. Maheswari Reddy B, Dhanpal CK, Lakshmi BVS. Protective effect of aqueous extract of leaves of Murraya loenigii, against Aluminium chloride induced oxidative stress in rat liver and kidney. Asian J Pharm Clinical Res (2020.0) 13 1-5. DOI: 10.22159/ajpcr.2020.v13i6.37383 19. Rahman MM, Gray AI. A benzoisofuranone derivative and carbazole alkaloids from. Phytochemistry (2005.0) 66 1601-6. DOI: 10.1016/j.phytochem.2005.05.001 20. Rees DA, Alcolado JC. Animal models of diabetes mellitus. Diabetic medicine: a journal of the British Diabetic Association (2005.0) 22 359-370. DOI: 10.1111/j.1464-5491.2005.01499.x 21. Takagi K, Watanabe M, Saito H. Studies on the spontaneous movement of animals by the hole cross test: Effect of 2-dimethylaminoethane and its acylates on the central nervous system. Jpn J Pharmacol (1971.0) 21 797. DOI: 10.1254/jjp.21.797 22. Kameyama T, Nabeshima T, Kozawa T. Step-down-type passive avoidance-and escape-learning method. Suitability for experimental amnesia models. J Pharmacol Methods (1986.0) 16 39-45. DOI: 10.1016/0160-5402(86)90027-6 23. Heo H, Shin Y, Cho W, Choi Y. Memory improvement in ibotenic acid induced model rats by extracts of Scutellariabaicalensis. Journal of Ethnopharmacology (2009.0) 122 20-7. DOI: 10.1016/j.jep.200811.026 24. Gupta BD, Dandiya PC, Gupta ML. A psychopharmacological analysis of behavior in rat. Jpn J Pharmacol (1971.0) 21 293. DOI: 10.1016/S0021-5198(19)36218-3 25. Ohkawa H, Ohishi N, Yagi K. Assay for lipid peroxides in animal tissues by thiobarbituric acid reaction. Anal Biochem (1979.0) 95 351-8. DOI: 10.1016/0003-2697(79)90738-3 26. Beutler E, Duron O, Kelly B. Improved method for the determination of blood glutathione. J Lab Clin Med (1963.0) 61 882-8. PMID: 13967893 27. Luck H, Bergmeyer HU. Catalase. Methods of enzymatic analysis (1971.0) 885-93. DOI: 10.1016/B978-0-12-395630-9.50158-4 28. Ellman GL, Courtney KD, Andres V. Feather-Stone RM. A new and rapid colorimetric determination of acetylcholinesterase activity. Biochem Pharmacol (1961.0) 7 88-95. DOI: 10.1016/0006-2952(61)90145-9 29. Sharmen F, Mannan A, Rahman MM, Chowdhury MAU. Investigation of in vivoneuropharmacological effect of Alpinianigra leaf extract. Asian Pac J Trop Biomed (2014.0) 4 137-142. DOI: 10.1016/s2221-1691(14)60222-7 30. Rao LJM, Ramalakshmi K, Borse BB, Raghavan B. Antioxidant and radical-scavenging carbazole alkaloids from the oleoresin of curry leaf (. ) Food Chem (2007.0) 100 742-747. DOI: 10.1016/j.foodchem.2005.10.033 31. Kishore RN, Anjaneyulu N, Ganesh MN, Sravya N. Evaluation of anxiolytic activity of ethanolic extract of Foeniculumvulgare in mice model. Int J Pharm PharmSci (2012.0) 4 584-6. DOI: 10.1186/1472-6882-14-310 32. Biessels G, Kamal A, Ramakers GM, Urban IJ. Place learning and hippocampal synaptic plasticity in streptozotocin-induced diabetic rats. Diabetes. DOI: 10.2337/diab.45.9.1259 33. Kwon SH, Kim HC, Lee SY, Jang CG. Loganin improves learning and memory impairments induced by scopolamine in mice. Eur J Pharmacol (2009.0) 619 44-9. DOI: 10.1016/j.ejphar.2009.06.062 34. Hazim AI, Mustapha M, Mansor SM. The effects on motor behaviour and short-term memory tasks in mice following an acute administration of Mitragynaspeciosa alkaloid extract and mitragynine. Journal of Medicinal Plants Research (2011.0) 5810-7. DOI: 10.1007/s12576-014-0304-0 35. Khodabandehloo F, Hosseini M, Rajaei Z, Soukhtanloo M. Brain tissue oxidative damage as a possible mechanism for the deleterious effect of a chronic high dose of estradiol on learning and memory in ovary ectomized rats. ArqNeuropsiquiatr (2013.0) 71 313-9. DOI: 10.1590/0004-282x20130027 36. Husna F, Suyatna FD, Arozal W, Poerwaningsih EH. Anti-Diabetic Potential of. Drug Res (Stuttg) (2018.0) 68 631-636. DOI: 10.1055/a-0620-8210 37. Dougherty KD, Turchin PI, Walsh TJ. Septocingulate and septohippocampal cholinergic pathways: Involvement in working/episodic memory. Brain Res (1998.0) 810 59-71. DOI: 10.1016/s0006-8993(98)00870-1 38. Vasudevan M, Parle M. Antiamnesic potential of. Phyther. Res (2009.0) 23 308-316. DOI: 10.1002/ptr.2620 39. Tan MA, Sharma N, An SSA. Multi-Target Approach of. Pharmaceuticals (2022.0) 15 188. DOI: 10.3390/ph15020188 40. Vasudevan M, Parle M. Antiamnesic potential of. Phytotherapy Research (2009.0) 23 308-316. DOI: 10.1002/ptr.2620
--- title: 'Nutritional status of children and adolescents with Type 1 Diabetes Mellitus in Baghdad: a case-control study' authors: - Sawsan Ali Hussein - Basma Adel Ibrahim - Wasnaa Hadi Abdullah journal: Journal of Medicine and Life year: 2023 pmcid: PMC10015568 doi: 10.25122/jml-2022-0233 license: CC BY 3.0 --- # Nutritional status of children and adolescents with Type 1 Diabetes Mellitus in Baghdad: a case-control study ## Abstract Diabetes mellitus (DM) is a major life-long non-communicable illness correlated with obesity and chronic undernutrition. It is particularly important to monitor the nutritional status of children with type 1 diabetes mellitus (T1DM), as they are still growing and may be affected by the disease or associated conditions like celiac disease. This study aimed to evaluate the nutritional status of children and adolescents with T1DM in Baghdad city and identify possible risk factors for undernutrition. A single-center, case-control study was conducted in Central Child's Teaching Hospital, Baghdad, Iraq, over 9 months from November 2021 to July 2022. The study included patients with T1DM and healthy controls. Detailed history, clinical examination, and anthropometric measures were performed for all participants in the study. The mean age of the sample was 10.0 ±3.73 years and 8.68±3.1 years in diabetic patients and controls, respectively. Anthropometric measures in patients with type 1 diabetes were significantly lower than those of controls ($P \leq 0.001$). All patients within the undernourished group were from large-size families compared with $75.76\%$ of the normally nourished group, with a significant difference. The mean age of disease onset in the normal nourished group was 6.61 ± 2.78 years which was significantly earlier than that of the undernourished group (8.83 ± 2.89). Weight-for-age and BMI z-score had a significant negative correlation with HbA1c (r=-0.312, $$p \leq 0.004$$, and r=-0.295, $$p \leq 0.006$$, respectively). Patients with T1DM had significantly lower anthropometric measures than the normal population. Older children, female gender, large family size, and disease duration are independent predictors of undernutrition in T1DM. BMI and weight-for-age have a significant negative correlation with metabolic control of diabetes represented by HbA1c. ## INTRODUCTION Diabetes mellitus (DM) is a major life-long non-communicable illness that has been correlated with both obesity and chronic undernutrition [1,2]. Type 1 DM was previously listed as one of the causes of severe growth retardation. In Iraq, the healthcare system has been disrupted by wars and conflicts that affect health services, affecting the glycemic control of children living with diabetes [3]. Diet is a core component of type 1 diabetes management, and poor diet quality may affect metabolic control and other health outcomes. However, there is a lack of research on diet quality in children and adolescents with DM [4]. Nutritional status should be regularly evaluated in type 1 DM children as they are still growing and are at risk of malnourishment due to the chronic and debilitating nature of the illness or the presence of associated celiac disease [5]. Factors that affect growth in DM patients include gender, genes, age at onset, growth hormone levels, disease duration, metabolic control, and puberty [6]. Pediatric undernutrition is described as a disproportion between nutrient demand and intake, leading to accumulative deficits of protein, energy, and micronutrients that may have negative effects on their growth and normal development [7]. It is a substantial issue in children with type 1 DM, and evaluation of growth should be considered during the regular follow-up visits of these patients [8]. Evaluation of undernutrition demands anthropometric measurements of body weight and length/height and plotting these variables on population growth charts for comparison against normal values [9]. However, there is an ongoing debate about the most useful measurement in DM follow-up and the consistency of anthropometric parameters. Therefore, a combination of measurements, including body weight, height, mass index, MUAC, and TSFT, should be considered along with other clinical parameters for nutritional evaluation in DM children [7]. This study aimed to evaluate the nutritional status of children and adolescents with type 1 diabetes mellitus in Baghdad city and identify possible risk factors for undernutrition. ## MATERIAL AND METHODS A case-control study was conducted at Central Child's Teaching Hospital in Baghdad, Iraq, over 9 months from November 1st, 2021, to July 31st, 2022. The study included 84 randomly selected patients with type 1 diabetes mellitus, between the ages of 3 and 18, who were under insulin treatment and had a diabetes duration of at least 1 year. They were compared to 84 age and gender-matched healthy controls. Patients with type 2 diabetes and those with chronic conditions such as celiac disease, hypothyroidism, inflammatory bowel diseases, or any other medical syndromes were excluded from the study. ## Sample size The sample size was calculated according to the following formula: N=Z2P(1−P)/d2, where N represents the sample size, Z corresponds to the level of confidence, P is the expected prevalence, and d represents the precision (corresponding to effect size) [10]. According to previous studies that reported a prevalence of T1DM of approximately $5\%$ among Iraqi children, the sample size was calculated as follows: ## Questionnaire Data were collected using a standardized questionnaire that included age (divided into three categories: preschool [<6 years], school age [6-12 years], and adolescent [13-18 years]), sex, family size (defined as families with three or more children considered as large and those with ≤ 2 children regarded as small) [8], family income (classified as low income [< 100,000 ID per capita], medium income [100,000- 250,000 ID per capita], and high income [> 250,000 ID per capita]) [8]. In addition, for the group of patients with diabetes, collected data included age at the onset of diabetes, duration of diabetes since diagnosis, number of diabetic ketoacidoses throughout the illness, and the most recent HbA1c test result (taken within 3 months). HbA1c values were measured using a direct enzymatic assay and were categorized as good (less than $7.5\%$), intermediate ($7.5\%$ to $9.0\%$), or poor (more than $9.0\%$) glycemic control following the International Society for Pediatric and Adolescent Diabetes website [11]. ## Anthropometric parameters All patients underwent a general and systematic evaluation, including the assessment of anthropometric parameters such as weight, height, body mass index (BMI), triceps skin fold thickness (TSFT), and mid-upper arm circumference (MUAC). Weight and height measurements were taken with participants wearing light clothing and no shoes. Weight measurements were taken on a manual scale, with the participant standing on the scale with arms extended along the side of the body and the researcher ensuring the back was straight. Height was measured with a wall stadiometer with a 0.1 cm precision scale. Height was measured with a wall stadiometer with a 0.1 cm precision scale, with the participant standing barefoot with their back to the wall [12]. Body mass index (BMI) was calculated by dividing weight in kilograms by the square of height in meters [13]. The mid-upper arm circumference was measured using a Gulick tape (Baseline 12-1201) with an accuracy of 0.5 cm, following the guidelines of the National Health and Nutrition Examination Survey [12]. Furthermore, the triceps skin fold thickness was measured using a Saehan Medical Skinfold Caliper (SH5020) with 0.1 mm accuracy. Measurements were taken just below the shoulder blades in a horizontal grip, just above the triceps muscle of the upper arm vertical grip, and on the stomach level in an oblique grip at a quarter the distance between the navel and the iliac on the non-dominant side of the body. The test was repeated 3 times from each location, and then the average was calculated from the results obtained [14]. The growth parameters were corrected for age by converting them to z scores [15]. Z scores were utilized because they allow more accuracy in expressing anthropometric status than traditional placement “near” or “below” a certain percentile curve [7]. Normal measures are within 2 standard deviations (SD) from the mean, while those > 2 SD below the mean indicate malnutrition [16]. ## Statistical analysis Statistical analysis was performed using SPSS software version 25.0 (SPSS, Chicago, USA). The normality of data was tested using the Shapiro-Wilk test, and normally distributed data were presented as mean ± SD. Comparisons for continuous data were made using a Student’s t-test. Data with non-normal distribution were described as median and range and were analyzed using the Mann-Whitney U test. Categorical data were expressed in numbers and percentages and analyzed using a Chi-square/Fischer exact test. The correlation between different national indices with age and disease duration was explored by Pearson’s correlation. Multivariate logistic regression was used to identify independent predictors of undernutrition in T1DM patients according to BMI. A p-value less than 0.05 was considered statistically significant. ## RESULTS In the current study, the mean age of patients with diabetes and controls was 10.0 ± 3.73 and 8.68 ± 3.1 years, respectively, with no significant difference ($$P \leq 0.069$$). Among patients with diabetes, $63.1\%$ were in the school-age group (6-12 years), $14.29\%$ were adolescents (13-18 years), and $9.52\%$ were below 6 years old. Females represented $65\%$ and $75\%$ of patients and controls, respectively, with no significant difference. Low income was reported in more than half ($54.67\%$) of the patient's families, and the large family size was prevalent among the majority ($80.95\%$) of patients. The mean age at onset and disease duration was 7.08±2.93 years and 3.0±2.61 years, respectively, with $76.19\%$ of the patients having a disease duration of less than 5 years. Most patients ($76.19\%$) experienced DKA 0-2 times, while only a minority ($4.67\%$) experienced such complications 6-8 times throughout their illness. The HbA1c level was markedly elevated, with a mean of 10.26±$2.31\%$, and $11.9\%$, $26.19\%$, and $61.9\%$ of patients had HbA1c levels of <7.5, 7.5-9, and >9, respectively. The anthropometric measurements in patients and controls are shown in Table 1, with the weight for age, height for age, BMI, and MUAC z scores significantly lower in patients with T1DM than in controls. However, there was no significant difference between the two groups in the TSFT z score. Table 1 \| Variables \| Patients (n=84) \| Controls (n=84) \| P-value \| \| --- \| --- \| --- \| --- \| \| Weight for age z score \| Weight for age z score \| Weight for age z score \| Weight for age z score \| \| Mean±SD \| -0.28±0.13 \| 0.62±1.23 \| <0.001 \| \| Median \| -0.45 \| 0.57 \| <0.001 \| \| Range \| -2.87-2.55 \| -2.0-3.0 \| <0.001 \| \| Height for age z score \| Height for age z score \| Height for age z score \| Height for age z score \| \| Mean±SD \| -0.66±1.29 \| 0.27±1.3 \| <0.001 \| \| Median \| -0.75 \| 0.32 \| <0.001 \| \| Range \| -3.14-2.37 \| -3.0-3.7 \| <0.001 \| \| BMI for age z score \| BMI for age z score \| BMI for age z score \| BMI for age z score \| \| Mean±SD \| 0.13±1.2 \| 2.19±0.96 \| <0.001 \| \| Median \| 0.09 \| 2.33 \| <0.001 \| \| Range \| -2.25-2.32 \| -0.36-3.8 \| <0.001 \| \| Mid upper arm circumference z score \| Mid upper arm circumference z score \| Mid upper arm circumference z score \| Mid upper arm circumference z score \| \| Mean±SD \| -1.53±1.37 \| 1.57±2.38 \| <0.001 \| \| Median \| -1.5 \| 1.0 \| <0.001 \| \| Range \| -4.86-1.39 \| -1.92-7.0 \| <0.001 \| \| Triceps skinfold thickness z score \| Triceps skinfold thickness z score \| Triceps skinfold thickness z score \| Triceps skinfold thickness z score \| \| Mean±SD \| 0.51±0.94 \| 0.64±1.15 \| 0.282 \| \| Median \| 0.51 \| 0.75 \| 0.282 \| \| Range \| -2.51-2.43 \| -2.0-2.7 \| 0.282 \| Wasting, defined as weight > 2 SD below the mean for age, and short stature, defined as height > 2 SD below the mean for age [17], were observed in 18 ($21.42\%$) and 32 ($38.09\%$) patients with diabetes, respectively. Low BMI was found in 18 ($21.42\%$) patients with diabetes. Table 2 shows the associations between demographic and clinical characteristics of the patients with BMI z score. Adolescent patients were more common among the undernourished group, and the reverse was true for school-age patients, with a highly significant difference. All patients within the undernourished group were from large-size families compared with $75.76\%$ of the normal nourished group, with a significant difference. The mean age of disease onset in the normal nourished group was 6.61 ± 2.78 years which was much earlier than that of the undernourished group (8.83 ± 2.89) with a highly significant difference. However, the patient' s gender, family income, disease duration, number of DKA episodes, and HbA1c levels were not different in the well-nourished and undernourished groups. Table 2 \| Variables \| Normal (n=66) \| Undernourished (n=18) \| P-value \| \| --- \| --- \| --- \| --- \| \| Age, years \| Age, years \| Age, years \| Age, years \| \| Preschool (<6) \| 6 (9.09%) \| 2 (11.11%) \| 0.007 \| \| School-age (6-12) \| 47 (71.21%) \| 6 (33.33%) \| 0.007 \| \| Adolescent (13-18) \| 13 (19.70%) \| 10 (55.56%) \| 0.007 \| \| Gender \| Gender \| Gender \| Gender \| \| Male \| 28 (42.42%) \| 4 (22.22%) \| 0.118 \| \| Female \| 38 (57.58%) \| 14 (77.78%) \| 0.118 \| \| Income \| Income \| Income \| Income \| \| Low \| 22 (33.33%) \| 10 (55.56%) \| 0.197 \| \| Medium \| 42 (63.64%) \| 8 (44.44%) \| 0.197 \| \| High \| 2 (3.03%) \| 0 (0%) \| 0.197 \| \| Family size \| Family size \| Family size \| Family size \| \| Small \| 16 (24.24%) \| 0 (0%) \| 0.020 \| \| Large \| 50 (75.76%) \| 18 (100%) \| 0.020 \| \| Age at onset, years \| Age at onset, years \| Age at onset, years \| Age at onset, years \| \| Mean±SD \| 6.61±2.78 \| 8.83±2.89 \| 0.004 \| \| Disease duration, years \| Disease duration, years \| Disease duration, years \| Disease duration, years \| \| <5 \| 48 (72.72%) \| 16 (88.89%) \| 0.154 \| \| ≥5 \| 18 (27.28%) \| 2 (11.11%) \| 0.154 \| \| Diabetic ketoacidosis, No. \| Diabetic ketoacidosis, No. \| Diabetic ketoacidosis, No. \| Diabetic ketoacidosis, No. \| \| 0-2 \| 52 (78.79%) \| 16 (88.89%) \| 0.148 \| \| 3-5 \| 10 (15.15%) \| 2 (11.11%) \| 0.148 \| \| 6-8 \| 4 (6.06%) \| 0 (0%) \| 0.148 \| \| HbA1c, % \| HbA1c, % \| HbA1c, % \| HbA1c, % \| \| <7.5 \| 8 (12.12%) \| 2 (11.11%) \| 0.738 \| \| 7.5-9 \| 16 (24.24%) \| 6 (33.33%) \| 0.738 \| \| >9 \| 42 (63.64%) \| 10 (55.56%) \| 0.738 \| A multivariate logistic regression test was performed to determine the independent predictors of undernutrition in T1DM patients based on BMI. All variables with a p-value of ≤0.150 in the univariate analysis (shown in Table 3) were included in the model. The results revealed that school age (OR=0.42, $95\%$CI= 0.12-0.87, $$p \leq 0.031$$), adolescent (OR= 3.17, $95\%$ CI= 1.14-32.54, $$p \leq 0.019$$), female gender (OR= 2.6, $95\%$ CI= 1.08-28.45, $$p \leq 0.048$$), family size ≥5 members (OR= 3.22, $95\%$CI= 1.23-22.98, $$p \leq 0.011$$) and disease duration (OR= 2.8, $95\%$CI=1.18-19.67, $$p \leq 0.032$$) were independent predictors of undernutrition in T1DM patients according to BMI (Table 4). Weight-for-age z-score revealed a significant negative correlation with HbA1c (r = -0.312, $$p \leq 0.004$$). Furthermore, the BMI z-score also displayed a significant negative correlation with HbA1c (r = -0.295, $$p \leq 0.006$$), as shown in Table 4, Figures 1 and 2. Figure 1: Scatter plot and regression line between HbA1c and weight-for-age z-score. Figure 2: Scatter plot and regression line between HbA1c and BMI for age z-score. ## DISCUSSION Few studies have focused on assessing the nutritional status of children and adolescents with type 1 DM. The current study emphasizes the importance of periodically evaluating anthropometric parameters and assessing the nutritional status of these patients. This provides a safe, non-invasive means for medical professionals to monitor metabolic control and ensure that type 1 DM patients achieve typical somatic development for their age group [17]. Our findings revealed higher rates of wasting and short stature ($21.42\%$ and $38.09\%$, respectively) among diabetic patients compared to previous studies by Khadilkar et al. [ 18] in India and Aljuhani et al. [ 6] in Saudi Arabia, which reported lower rates of $10.9\%$ and $27.1\%$, and $6.9\%$ and $11.9\%$, respectively. Variations in sample size and population characteristics may account for these differences. The anthropometric measures, including weight for age, height for age, BMI, and MUAC z-scores, were significantly lower in diabetic patients compared to controls, which is consistent with previous studies such as Dohan et al., [ 8] in Iraq, Khadilkar et al., [ 18] in India, and Stipancić et al., [ 19] in Croatia, suggesting that diabetes has a negative impact on nutritional status. Furthermore, this study found that malnourished patients with diabetes were mostly adolescents, and older age at disease onset was significantly associated with malnutrition. This finding may be explained by the fact that older age groups may experience dietary restrictions, while younger age groups may not be as restricted in their diet, which is crucial for their normal growth and development. In addition, the effect of sex steroids on adolescents may counteract the effect of insulin, potentially leading to the derangement of their metabolic control. Regarding the number of family members, this study found that large family size was significantly associated with low BMI in individuals with diabetes, which could be explained by a decrement in nutritional and family care with an increase in the number of siblings. The earlier study by Dohan et al. [ 8] found a non-significant correlation which could be attributed to variation in sample size. The present study identified several independent risk factors for undernutrition in type 1 diabetes, including school age, adolescent age, female gender, large family size, and longer disease duration. These findings are supported by previous studies, including Dohan et al. [ 8] and Mousa et al. [ 20], which also found undernutrition more common in females. However, Hassan et al. [ 21] found no significant correlation between gender and growth parameters. Bonfig et al. [ 22] noted a negative correlation between adult height and the duration of diabetes in a large study of over 22,000 children with type 1 DM in Germany and Austria, while Aljuhani et al. [ 6] found no significant effect of gender or disease duration on undernutrition. The correlation between all growth measures in diabetic patients and HbA1c showed a significant negative correlation between weight for age and BMI with HbA1c, which is consistent with the findings of Aljuhani et al. [ 6] and Dohan et al. [ 8]. Hassan et al.[21] found a negative correlation between metabolic control and all anthropometric measures, while the study by Khadilkar et al. [ 18] from India and Galli-Tsinopoulou et al. [ 23] from Greece observed no correlation between HbA1c and growth parameters. This could be attributed to the variation in sample size and general characteristics of the population. The correlation between glycemic control or the duration of illness and child growth varies among studies. Thus, conducting a systematic review of these studies may be crucial in gaining a deeper insight into the growth status of children with diabetes [6]. The current study has several limitations that should be considered when interpreting the results. First, the sample size was relatively small, which may limit the generalizability of the findings to other populations. Secondly, the study was conducted at a single center, which may limit the external validity of the findings. Another limitation is the lack of multiple readings of HbA1c, which may have led to measurement errors and reduced the accuracy of the results. ## CONCLUSION Patients with type 1 diabetes mellitus had significantly lower anthropometric measures than the general population. Furthermore, older children, female gender, large family size, and disease duration were independent predictors of undernutrition in T1DM. Moreover, there was a significant negative correlation between BMI and weight for age with metabolic control of diabetes as represented by HbA1c. Further multi-centric studies with a larger sample size should be carried out to strengthen the study results. ## Conflict of interest The authors declare no conflict of interest. ## Ethical approval This study was approved by the bioethical committee of the College of Medicine, University of Mustansiriyah (No. 233, 2022). ## Consent to participate All patients included in the study provided informed consent for participation. ## Authorship SAH contributed to study design and research supervision. BAI contributed to data analysis and draft manuscript preparation. WHA contributed to the critical revision of the paper. SAH, WHA, and BAI contributed to supervision and funding. WHA and BAI contributed to the final approval of the version to be published. ## References 1. Uwaezuoke SN. Childhood diabetes mellitus and the “double burden of malnutrition”: an emerging public health challenge in developing countries. J Diabetes Metab (2015) 6 2. DOI: 10.4172/2155-6156.1000597 2. Hu FB. Globalization of diabetes: the role of diet, lifestyle, and genes. Diabetes care (2011) 34 1249-57. DOI: 10.2337/dc11-0442 3. Hadi ZS, Al-Kaseer EA, Al-Zubaidi MA. Growth of diabetic children in post conflict Baghdad, Iraq. Journal of the Faculty of Medicine Baghdad (2018) 60 69-73. DOI: 10.32007/jfacmedbagdad.60155 4. Nansel TR, Haynie DL, Lipsky LM, Laffel LM, Mehta SN. Multiple indicators of poor diet quality in children and adolescents with type 1 diabetes are associated with higher body mass index percentile but not glycemic control. Journal of the Academy of Nutrition and Dietetics (2012) 112 1728-35. DOI: 10.1016/j.jand.2012.08.029 5. Nutrition Recommendation and Principles for People with Diabetes Mellitus. Diabetes Care (2000) 23 Sup1-s436 6. Aljuhani FM, Al-agha AE, Almunami BA, Meftah EA. Growth status of children and adolescents with type 1 diabetes mellitus in Jeddah, Saudi Arabia: A cross-sectional study. Curr Pediatr Res (2018) 22 249-254 7. Mehta NM, Corkins MR, Lyman B, Malone A. Defining pediatric malnutrition: A paradigm shift toward etiology-related definitions. Journal of Parenteral and Enteral Nutrition (2013) 37 460-481. DOI: 10.1177/0148607113479972 8. Dohan BR, Habib S, Abd Khazal A. Nutritional Status of Children and Adolescents with Type1 Diabetes Mellitus in Basra. The Medical Journal of Basrah University (2021) 39 54-60. DOI: 10.33762/mjbu.2021.127780.1027 9. Gomez F, Galvan RR, Cravioto J, Frenk S. Malnutrition in infancy and childhood, with special reference to kwashiorkor. Adv Pediatr (1955) 7 131-169. PMID: 14349775 10. Pourhoseingholi MA, Vahedi M, Rahimzadeh M. Sample size calculation in medical studies. Gastroenterol Hepatol Bed Bench (2013) 6 14-7. PMID: 24834239 11. International society for pediatric and adolescent diabetes. ISPAD (2017) 12. Anthropometry Procedures Manual. (2017) 13. Cole TJ, Flegal KM, Nicholls D, Jackson AA. Body mass index cut offs to define thinness in children and adolescents: international survey. BMJ (2007) 335 194. DOI: 10.1136/bmj.39238.399444.55 14. Eaton-Evans J. Nutritional assessment: Anthropometry. Encyclopedia of Human Nutrition (2013) 227-232. DOI: 10.1016/B978-0-12-375083-9.00197-5 15. Chou JH, Roumiantsev S, Singh R. PediTools Electronic Growth Chart Calculators: Applications in Clinical Care, Research, and Quality Improvement. J Med Internet Res (2020) 22 e16204. DOI: 10.2196/16204 16. Wang Y, Chen H, Preedy VR. Use of Percentiles and Z-Scores in Anthropometry. Handbook of Anthropometry: Physical Measures of Human Form in Health and Disease (2012) 29-46 17. Grabia M, Markiewicz-Żukowska R. Nutritional Status of Pediatric Patients with Type 1 Diabetes Mellitus from Northeast Poland: A Case-Control Study. Diabetes Ther (2021) 12 329-343. DOI: 10.1007/s13300-020-00972-1 18. Khadilkar VV, Parthasarathy LS, Mallade BB, Khadilkar AV. Growth status of children and adolescents with type 1 diabetes mellitus. Indian J Endocrinol Metab (2013) 17 1057-1060. DOI: 10.4103/2230-8210.122623 19. Stipančić G, La Grasta Sabolić L, Jurcic Z. Growth disorders in children with type 1 diabetes mellitus. Coll. Antropol (2006) 30 297-304. PMID: 16848143 20. Mousa WL, Aitte SA, Qasim AK. Nutritional Status of Pediatric Patients with Type 1 Diabetes Mellitus in DhiQar government. Annals of R.S.C.B (2021) 25 370-375 21. Hassan NE, El-Kahky A, Hana MA, Abu Shady MM. Physical growth and body composition of controlled versus uncontrolled type 1 Egyptian diabetic children. Maced J Med Sci (2014) 2 567-572. DOI: 10.3889/oamjms.2014.102 22. Bonfig W, Kapellen T, Dost A, Fritsch M. Growth in children and adolescents with type 1 diabetes. J Pediatr (2012) 160 900-903. DOI: 10.1016/j.jpeds.2011.12.007 23. Galli-Tsinopoulou A, Grammatikopoulou MG, Stylianou C, Kokka P, Emmanouilidou E. A preliminary case-control study on nutritional status, body composition, and glycemic control of Greek children and adolescents with type 1 diabetes. Journal of Diabetes (2009) 1 36-42. DOI: 10.1111/j.1753-0407.2008.00002.x
--- title: 'The relationship between maternal health and neonatal low birth weight in Amman, Jordan: a case-control study' authors: - Amer Sindiani - Ekram Awadallah - Eman Alshdaifat - Shatha Melhem - Khalid Kheirallah journal: Journal of Medicine and Life year: 2023 pmcid: PMC10015569 doi: 10.25122/jml-2022-0257 license: CC BY 3.0 --- # The relationship between maternal health and neonatal low birth weight in Amman, Jordan: a case-control study ## Abstract This study aimed to examine the relationship between maternal health during pregnancy and low birth weight (LBW), as well as the impact of COVID-19 on the socio-economic status of pregnant women and its effect on LBW. The study was conducted in Amman, Jordan, and included 2260 mothers who visited Abu-Nusair comprehensive health center between January and December 2020. A matched case-control design was used with 72 cases and 148 controls selected for data collection through medical records and face-to-face interviews. Results showed that factors such as a monthly income of 400 JD or less, living with an extended family, exposure to passive smoking, maternal weight gain of 6–10 kg, maternal anemia, maternal hypertension, delivery by cesarean section, and previous history of LBW newborns were positively associated with an increased risk of LBW. Conversely, factors such as a monthly income above 700 JD, living with a core family, daily intake of iron, calcium, and vitamin D, prenatal visits, healthy food intake, and planning for pregnancy were associated with a lower risk of LBW. COVID-19 infection and its effects on work, family finances, antenatal care visits, and food supply were also positively linked with LBW. In conclusion, socioeconomic status, maternal health, COVID-19, and its impacts were significant risk factors for LBW. ## INTRODUCTION Low birth weight (LBW) is a major public health burden due to its impacts on neonatal health, development, and survival [1]. Infants with LBW face a much higher risk of mortality, with rates 20 times higher compared to those with normal birth weight (NBW) [2]. Additionally, LBW newborns are at increased risk of experiencing cognitive deficits, metabolic diseases, motor delays, cerebral palsy, and other psychological and behavioral problems [1]. The World Health Organization defines low birth weight (LBW) as "live births weighing less than 2500 grams at birth, regardless of gestational age" [3]. LBW can also be defined as a birth weight below the 10th (or 5th) percentile for gestational age or less than 2 standard deviations below the mean for gestational age [4]. Maternal nutrition, physical and psychological health, social status, and socioeconomic factors all play a role in determining the risk of LBW during pregnancy [5], with a higher prevalence in developing countries compared to developed countries. Low birth weight is caused by a combination of factors, including intrauterine growth restriction (IUGR) and preterm birth. These conditions occur due to placental insufficiency, which impairs fetal nutrition and growth [1]. Maternal factors during pregnancy, such as nutrition, economic stability, and social factors, significantly impact neonatal weight [1]. The COVID-19 pandemic has exacerbated these challenges, leading to increased economic and social stress [6] and reducing access to adequate prenatal care, maternal follow-ups, and essential supplements, resulting in a higher incidence of LBW newborns during the lockdown period [7]. Globally in 2015, $14.56\%$ of newborns had low birth weight, a condition that increases their risk of experiencing complications and mortality by 20 times compared to newborns with normal birth weight [1]. In Jordan, LBW is a significant public health concern as it is the leading cause of morbidity and mortality in newborns, with a prevalence rate of $13.8\%$ in 2012 [8]. Therefore, this study aimed to assess the relationship between maternal health, obstetric outcomes, and LBW in Abu Nusair Comprehensive Center (ANCC) north of Amman between January and December 2020. In addition, we aimed to investigate the impact of COVID-19-related socioeconomic factors on neonatal low birth weight. ## Study design and setting The study employed a matched case-control design in which controls were matched to cases based on age and social class (as indicated by the level of education of parents and employment) among women who visited Abu-Nusair comprehensive health center (ANCC) during the specified period. This study was conducted at ANCC, the second-largest comprehensive center in Amman and the biggest in the north of Amman. The study population was dichotomized into two groups: cases and controls. The cases comprised mothers of term-singleton newborns with a birth weight of less than 2500 grams and without any congenital anomalies or deformities who visited the Abu-Nusair Comprehensive Health Center (ANCC) in 2020. The controls comprised mothers of term-singleton newborns with a birth weight ranging from 2500 grams to less than 4500 grams and without any congenital anomalies or deformities, who also visited the ANCC in 2020. Mothers with newborns with multiple births, a history of preterm deliveries, congenital anomalies, and any deformities were excluded from this study as those are common risk factors for LBW. ## Data collection The initial sample size of all mothers eligible for the study was 2260. Data were obtained from medical files and during face-to-face interviews. Seventy-two cases were randomly selected from the total 166 cases using the last digit of the parent's mobile number, and 148 controls were chosen from 2094 controls based on matching age and social class (parents’ level of education, employment). Participants who failed to attend the interviews were excluded. The interview questionnaire consisted of four main sections: sociodemographic data, maternal health during pregnancy, obstetric outcomes, and COVID-19 infection and its associated socioeconomic impacts. The study method and protocol were approved by the Institutional Review Board (IRB) of Jordan University of Science and Technology and the Ministry of Health. Demographic data included questions about age, academic level, working status, monthly family income, residency, and smoking status. Maternal health was assessed based on maternal weight gain during pregnancy, body mass index (BMI), number of prenatal care visits, multivitamin intake, dietary quality, sleep patterns, and medical conditions such as hypertension, diabetes, asthma, anemia, urinary tract infections, rheumatic diseases, systemic lupus erythematosus, and any other relevant conditions. We also assessed various factors related to obstetric outcomes, including parity, gender of the newborn, pregnancy planning, mode of delivery, history of miscarriages, type of pregnancy, gestational age at delivery, and previous LBW newborns. The last section of the interview focused on COVID-19 and included questions regarding infection status, hospital admission, job stability, food access, medication intake, antenatal care visits, and availability of multivitamins. Figure 1 shows a schematic summary of the sampling approach for this study. Figure 1: A schematic representation of the study sampling approach. ## Statistical analysis Data were analyzed using the Statistical Package of Social Science (SPSS) version 25 (IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY: IBM Corp.). Frequencies and percentages were calculated for the categorical data, and a chi-square test (Fisher exact test) was conducted to compare the proportions of the categorical variables between two groups (case group and control group). Binary logistic regression analysis was performed to determine the associations between risk factors of LBW neonatal outcomes among the Jordanian women (case vs. control for each risk factor). The level of significance was set at $p \leq 0.05.$ ## Socio-demographic characteristics More than $65\%$ of the case ($66.7\%$) and control ($66.2\%$) groups reported being unemployed. Around $23.6\%$ of the participants had a monthly income of less than 400 JD, $47.3\%$ had an income between 401–700 JD, and $29.1\%$ had a monthly income exceeding 700 JD. A higher proportion of participants in the case group ($51.4\%$) had a monthly income of 400 JD or less compared to the control group ($10.1\%$), which was statistically significant ($p \leq 0.001$). Conversely, a greater percentage of participants in the control group ($35.1\%$) had a monthly income greater than 700 JD compared to the case group ($16.7\%$), which was also statistically significant ($p \leq 0.001$). A larger proportion of women in the case group lived with extended family members compared to the control group ($36\%$ vs. $20\%$) ($$p \leq 0.001$$). In addition, a majority ($75\%$) of the participants in the case group were exposed to passive smoking during pregnancy, compared to only $32.4\%$ of the participants in the control group ($p \leq 0.001$). These findings are summarized in Table 1. Table 1 \| Variables \| Case (72) n (%) \| Control (148) n (%) \| Total \| P-value \| \| --- \| --- \| --- \| --- \| --- \| \| Occupational status \| Occupational status \| Occupational status \| Occupational status \| Occupational status \| \| Employed \| 24 (33.3) \| 50 (33.8) \| 74 (33.6) \| 0.947 \| \| Unemployed \| 48 (66.7) \| 98 (66.2) \| 146 (66.4) \| 0.947 \| \| Monthly income (JD) \| Monthly income (JD) \| Monthly income (JD) \| Monthly income (JD) \| Monthly income (JD) \| \| ≤400 \| 37 (51.4) \| 15 (10.1) \| 52 (23.6) \| <0.001 \| \| 401–550 \| 10 (13.9) \| 46 (31.1) \| 56 (25.5) \| <0.001 \| \| 550–700 \| 13 (18.1) \| 35 (23.6) \| 48 (21.8) \| <0.001 \| \| >700 \| 12 (16.7) \| 52 (35.1) \| 64 (29.1) \| <0.001 \| \| Type of families \| Type of families \| Type of families \| Type of families \| Type of families \| \| Core family \| 46 (63.9) \| 118 (79.7) \| 164 (74.5) \| 0.001 \| \| Extended family \| 26 (36.1) \| 30 (20.3) \| 56 (25.5) \| 0.001 \| \| Maternal cigarettes use \| Maternal cigarettes use \| Maternal cigarettes use \| Maternal cigarettes use \| Maternal cigarettes use \| \| Yes \| 7 (9.7) \| 3 (2.0) \| 10 (4.5) \| 0.010 \| \| No \| 65 (90.3) \| 145 (98.0) \| 210 (95.5) \| 0.010 \| \| Maternal waterpipe use \| Maternal waterpipe use \| Maternal waterpipe use \| Maternal waterpipe use \| Maternal waterpipe use \| \| Yes \| 12 (16.7) \| 17 (11.5) \| 29 (13.2) \| 0.287 \| \| No \| 60 (83.3) \| 131 (88.5) \| 191 (86.8) \| 0.287 \| \| Maternal smoking status \| Maternal smoking status \| Maternal smoking status \| Maternal smoking status \| Maternal smoking status \| \| Cigarette only \| 2 (2.8) \| 1 (0.7) \| 3 (1.4) \| 0.084 \| \| Water pipe only \| 7 (9.7) \| 15 (10.1) \| 22 (10.0) \| 0.084 \| \| Cigarette and water pipe \| 5 (6.9) \| 2 (1.4) \| 7 (3.2) \| 0.084 \| \| Non-smoker \| 58 (80.6) \| 130 (87.8) \| 188 (85.5) \| 0.084 \| \| Passive smoking \| Passive smoking \| Passive smoking \| Passive smoking \| Passive smoking \| \| Yes \| 54 (75.0) \| 48 (32.4) \| 102 (46.4) \| <0.001 \| \| No \| 18 (25.0) \| 100 (67.6) \| 118 (53.6) \| <0.001 \| \| Body Mass Index (BMI)* \| Body Mass Index (BMI)* \| Body Mass Index (BMI)* \| Body Mass Index (BMI)* \| Body Mass Index (BMI)* \| \| Underweight \| 3 (4.2) \| 7 (4.8) \| 10 (4.6) \| 0.753 \| \| Normal weight \| 52 (73.2) \| 97 (66.0) \| 149 (68.3) \| 0.753 \| \| Overweight \| 11 (15.5) \| 30 (20.4) \| 41 (18.8) \| 0.753 \| \| Obesity \| 5 (7.0) \| 13 (8.8) \| 18 (8.3) \| 0.753 \| ## Maternal health during pregnancy among mothers of LBW newborns and normal-birth-weight newborns To assess maternal health, the following factors were taken into consideration: supplementation of iron, calcium, and vitamin D, weight gain during pregnancy, prenatal visit frequency, pregnancy-related anemia, utilization of other medications, dietary intake, sleep quality, and various health conditions (Table 2). The results indicated that daily supplementation of iron, calcium, and vitamin D was significantly more prevalent among mothers of normal-weight newborns ($85.8\%$, $$n = 127$$) compared to mothers of low-birth-weight newborns ($30.0\%$, $$n = 41$$) ($p \leq 0.001$). In terms of weight gain during pregnancy, mothers of low birth-weight newborns had a higher prevalence of gestational weight gain of 6–10 kg ($31.9\%$, $$n = 23$$) compared to mothers of normal-weight newborns ($23.6\%$, $$n = 36$$). Conversely, mothers of normal-weight newborns had a higher prevalence of gestational weight gain exceeding 10 kg ($76.3\%$, $$n = 113$$) compared to mothers of low birth-weight newborns ($68.1\%$, $$n = 49$$) ($$p \leq 0.001$$) (Table 2). Table 2 \| Variables \| Case (72) n (%) \| Control (148) n (%) \| Total \| P-value \| \| --- \| --- \| --- \| --- \| --- \| \| Iron, calcium and vitamin D use \| Iron, calcium and vitamin D use \| Iron, calcium and vitamin D use \| Iron, calcium and vitamin D use \| Iron, calcium and vitamin D use \| \| Yes, not daily \| 42 (58.3) \| 21 (14.2) \| 63 (28.6) \| <0.001 \| \| Yes, daily \| 30 (41.7) \| 127 (85.8) \| 157 (71.4) \| <0.001 \| \| Maternal weight gain (kg) \| Maternal weight gain (kg) \| Maternal weight gain (kg) \| Maternal weight gain (kg) \| Maternal weight gain (kg) \| \| 6–10 \| 23 (31.9) \| 35 (23.6) \| 58 (26.4) \| 0.001 \| \| 11–16 \| 30 (41.7) \| 98 (66.2) \| 128 (58.2) \| 0.001 \| \| >16 \| 19 (26.4) \| 15 (10.1) \| 34 (15.4) \| 0.001 \| \| Prenatal visits \| Prenatal visits \| Prenatal visits \| Prenatal visits \| Prenatal visits \| \| 1–3 \| 4 (5.6) \| 3 (2.0) \| 7 (3.2) \| <0.001 \| \| 4–7 \| 28 (38.9) \| 25 (16.9) \| 53 (24.1) \| <0.001 \| \| >7 \| 40 (55.6) \| 120 (81.1) \| 160 (72.7) \| <0.001 \| \| Maternal anemia \| Maternal anemia \| Maternal anemia \| Maternal anemia \| Maternal anemia \| \| Yes \| 45 (62.5) \| 20 (13.5) \| 65 (29.5) \| <0.001 \| \| No \| 27 (37.5) \| 128 (86.5) \| 155 (70.5) \| <0.001 \| \| Regularly medications \| Regularly medications \| Regularly medications \| Regularly medications \| Regularly medications \| \| Yes \| 4 (5.6) \| 4 (2.7) \| 8 (3.6) \| 0.289 \| \| No \| 68 (94.4) \| 144 (97.3) \| 212 (96.4) \| 0.289 \| \| Healthy food intake \| Healthy food intake \| Healthy food intake \| Healthy food intake \| Healthy food intake \| \| Yes \| 19 (26.4) \| 116 (78.4) \| 135 (61.4) \| <0.001 \| \| No \| 53 (73.6) \| 32 (21.6) \| 85 (38.6) \| <0.001 \| \| Sleeping well \| Sleeping well \| Sleeping well \| Sleeping well \| Sleeping well \| \| Yes \| 36 (50.0) \| 80 (54.1) \| 116 (52.7) \| 0.572 \| \| No \| 36 (50.0) \| 68 (45.9) \| 104 (47.3) \| 0.572 \| \| Rupture of Membrane \| Rupture of Membrane \| Rupture of Membrane \| Rupture of Membrane \| Rupture of Membrane \| \| Yes \| 59 (83.1) \| 136 (91.9) \| 24 (11.0) \| 0.051 \| \| No \| 12 (16.9) \| 12 (8.1) \| 195 (89.0) \| 0.051 \| \| Diabetes \| Diabetes \| Diabetes \| Diabetes \| Diabetes \| \| Yes \| 4 (5.6) \| 7 (4.7) \| 11 (5.0) \| 0.792 \| \| No \| 68 (94.4) \| 141 (95.3) \| 209 (95.0) \| 0.792 \| \| Recurrent Urinary Tract Infections (UTI) \| Recurrent Urinary Tract Infections (UTI) \| Recurrent Urinary Tract Infections (UTI) \| Recurrent Urinary Tract Infections (UTI) \| Recurrent Urinary Tract Infections (UTI) \| \| Yes \| 16 (22.5) \| 19 (12.8) \| 35 (16.0) \| 0.067 \| \| No \| 55 (77.5) \| 129 (87.2) \| 184 (84.0) \| 0.067 \| \| Hypertension \| Hypertension \| Hypertension \| Hypertension \| Hypertension \| \| Yes \| 11 (15.3) \| 8 (5.4) \| 19 (8.6) \| 0.014 \| \| No \| 61 (84.7) \| 140 (94.6) \| 201 (91.4) \| 0.014 \| \| Asthma \| Asthma \| Asthma \| Asthma \| Asthma \| \| Yes \| 4 (5.6) \| 3 (2.0) \| 7 (3.2) \| 0.156 \| \| No \| 67 (94.4) \| 145 (98.0) \| 212 (96.8) \| 0.156 \| \| Rheumatic disease \| Rheumatic disease \| Rheumatic disease \| Rheumatic disease \| Rheumatic disease \| \| Yes \| 3 (4.2) \| 8 (5.4) \| 11 (5.0) \| 0.708 \| \| No \| 68 (95.8) \| 140 (94.6) \| 208 (95.0) \| 0.708 \| \| Systemic lupus erythematosus \| Systemic lupus erythematosus \| Systemic lupus erythematosus \| Systemic lupus erythematosus \| Systemic lupus erythematosus \| \| Yes \| 3 (4.2) \| 3 (2.0) \| 6 (2.7) \| 0.361 \| \| No \| 69 (95.8) \| 145 (98.0) \| 214 (97.3) \| 0.361 \| \| Other medical condition \| Other medical condition \| Other medical condition \| Other medical condition \| Other medical condition \| \| Yes \| 2 (2.8) \| 9 (6.1) \| 11 (5.0) \| 0.291 \| \| No \| 70 (97.2) \| 139 (93.9) \| 209 (95.0) \| 0.291 \| The number of prenatal visits was negatively associated with LBW newborns ($p \leq 0.001$). Maternal anemia was higher among mothers of LBW newborns ($62.5\%$, $$n = 45$$) compared to mothers of normal-birth-weight newborns ($13.5\%$, $$n = 20$$) ($p \leq 0.001$). Women who had access to healthy food during pregnancy were higher among mothers of normal-weight newborns ($78.4\%$, $$n = 116$$) than among mothers of LBW newborns ($26.4\%$, $$n = 16$$) ($p \leq 0.001$) (Table 2). Additionally, hypertension was more prevalent among women who had low birth weight newborns ($15.3\%$, $$n = 11$$) compared to women who had normal-birth-weight newborns ($5.4\%$, $$n = 8$$) ($$p \leq 0.014$$) (Table 2). ## Obstetric outcomes among mothers of LBW and normal- birth weight newborns A chi-square test (Fisher's exact test) was performed to analyze the distribution of categorical obstetric outcomes variables between the two groups, including parity, neonatal sex, pregnancy planning, previous miscarriage, gestational age, type of pregnancy, delivery mode, and previous LBW newborn (Table 3). $51.4\%$ ($$n = 37$$) of the LBW newborns were female, while $48.6\%$ ($$n = 35$$) were male. In contrast, the control group comprised $68.3\%$ ($$n = 99$$) males and $31.7\%$ ($$n = 46$$) females (Table 3). The percentage of women who planned their pregnancy in the control group ($87.8\%$, $$n = 130$$) was significantly higher ($$p \leq 0.001$$) than in the case group ($62.5\%$, $$n = 45$$) (Table 3). Moreover, LBW newborns were significantly more likely to be born through vaginal delivery ($63.9\%$, $$n = 46$$) compared to cesarean section ($36.1\%$, $$n = 26$$) ($p \leq 0.004$) (Table 3). Additionally, the number of mothers with a history of LBW was significantly higher in the case group ($33.3\%$, $$n = 24$$) compared to the control group ($7.4\%$, $$n = 11$$) ($p \leq 0.001$) (Table 3). Table 3 \| Variables \| Case (72) n (%) \| Control (148) n (%) \| Total \| P-value \| \| --- \| --- \| --- \| --- \| --- \| \| Parity \| Parity \| Parity \| Parity \| Parity \| \| Primigravida \| 30 (41.7) \| 58 (39.2) \| 88 (40.0) \| 0.725 \| \| Multi-gravida \| 42 (58.3) \| 90 (60.8) \| 132 (60.0) \| 0.725 \| \| Neonatal gender \| Neonatal gender \| Neonatal gender \| Neonatal gender \| Neonatal gender \| \| Boy \| 35 (48.6) \| 99 (68.3) \| 134 (61.8) \| 0.005 \| \| Girl \| 37 (51.4) \| 46 (31.7) \| 83 (38.2) \| 0.005 \| \| Planning for pregnancy \| Planning for pregnancy \| Planning for pregnancy \| Planning for pregnancy \| Planning for pregnancy \| \| Yes \| 45 (62.5) \| 130 (87.8) \| 175 (79.5) \| <0.001 \| \| No \| 27 (37.5) \| 18 (12.2) \| 45 (20.5) \| <0.001 \| \| Previous miscarriage \| Previous miscarriage \| Previous miscarriage \| Previous miscarriage \| Previous miscarriage \| \| Yes \| 10 (13.9) \| 11 (7.4) \| 21 (9.5) \| 0.126 \| \| No \| 62 (86.1) \| 137 (92.6) \| 199 (90.5) \| 0.126 \| \| Type of pregnancy \| Type of pregnancy \| Type of pregnancy \| Type of pregnancy \| Type of pregnancy \| \| Induced \| 68 (94.4) \| 128 (87.1) \| 196 (89.5) \| 0.095 \| \| Spontaneous \| 4 (5.6) \| 19 (12.9) \| 23 (10.5) \| 0.095 \| \| Gestational age (week) \| Gestational age (week) \| Gestational age (week) \| Gestational age (week) \| Gestational age (week) \| \| Preterm (37–38) \| 10 (13.9) \| 11 (7.4) \| 21 (9.5) \| 0.126 \| \| Full-term (38–42) \| 62 (86.1) \| 137 (92.6) \| 199 (90.5) \| 0.126 \| \| Mode of delivery \| Mode of delivery \| Mode of delivery \| Mode of delivery \| Mode of delivery \| \| Cesarean section \| 26 (36.1) \| 27 (18.2) \| 53 (24.1) \| 0.004 \| \| Vaginal delivery \| 46 (63.9) \| 121 (81.8) \| 167 (75.9) \| 0.004 \| \| Previous LBW newborn \| Previous LBW newborn \| Previous LBW newborn \| Previous LBW newborn \| Previous LBW newborn \| \| Yes \| 24 (33.3) \| 11 (7.4) \| 35 (15.9) \| <0.001 \| \| No \| 48 (66.7) \| 137 (92.6) \| 185 (84.1) \| <0.001 \| ## The relationship between COVID-19 infection, socioeconomic impact, and LBW The infection rate among the case group was significantly higher compared to the control group, with $43.1\%$ ($$n = 31$$) and $9.5\%$ ($$n = 14$$), respectively ($p \leq 0.001$). Hospitalization due to COVID-19 was also more frequent in the case group ($16.7\%$, $$n = 12$$) compared to the control group ($1.4\%$, $$n = 2$$) ($p \leq 0.001$). The impact of the COVID-19 pandemic on employment was greater in the case group, with $83.3\%$ of mothers of LBW newborns losing their jobs compared to only $4.7\%$ in the control group ($p \leq 0.004$). Additionally, $91.7\%$ of mothers in the case group reported a financial impact from the pandemic, compared to $25.0\%$ in the control group ($p \leq 0.001$). Antenatal care schedules, monthly medications, iron, calcium, vitamin D supplementation, and food access were also more frequently disrupted in the case group than in the control group due to the pandemic ($p \leq 0.001$) (Table 4). Table 4 \| Variables \| Case (72) n (%) \| Control (148) n (%) \| Total \| P-value \| \| --- \| --- \| --- \| --- \| --- \| \| COVID-19 infection \| COVID-19 infection \| COVID-19 infection \| COVID-19 infection \| COVID-19 infection \| \| Yes \| 31 (43.1) \| 14 (9.5) \| 45 (20.5) \| <0.001 \| \| No \| 41 (56.9) \| 134 (90.5) \| 175 (79.5) \| <0.001 \| \| Hospital admission \| Hospital admission \| Hospital admission \| Hospital admission \| Hospital admission \| \| Yes \| 12 (16.7) \| 2 (1.4) \| 14 (6.4) \| <0.001 \| \| No \| 60 (83.3) \| 146 (98.6) \| 206 (93.6) \| <0.001 \| \| Worrying about having COVID-19 \| Worrying about having COVID-19 \| Worrying about having COVID-19 \| Worrying about having COVID-19 \| Worrying about having COVID-19 \| \| Yes \| 68 (94.4) \| 135 (91.2) \| 203 (92.3) \| 0.400 \| \| No \| 4 (5.6) \| 13 (8.8) \| 17 (7.7) \| 0.400 \| \| Losing work \| Losing work \| Losing work \| Losing work \| Losing work \| \| Yes \| 55 (83.3) \| 7 (4.7) \| 62 (29.0) \| 0.004 \| \| No \| 11 (16.7) \| 141 (95.3) \| 152 (71.0) \| 0.004 \| \| Family members losing work \| Family members losing work \| Family members losing work \| Family members losing work \| Family members losing work \| \| Yes \| 58 (80.6) \| 23 (15.5) \| 81 (36.8) \| <0.001 \| \| No \| 14 (19.4) \| 125 (84.5) \| 139 (63.2) \| <0.001 \| \| Economic ability \| Economic ability \| Economic ability \| Economic ability \| Economic ability \| \| Yes \| 66 (91.7) \| 37 (25.0) \| 103 (46.8) \| <0.001 \| \| No \| 6 (8.3) \| 111 (75.0) \| 117 (53.2) \| <0.001 \| \| ANC visits schedule \| ANC visits schedule \| ANC visits schedule \| ANC visits schedule \| ANC visits schedule \| \| Yes \| 59 (83.1) \| 65 (43.9) \| 124 (56.6) \| <0.001 \| \| No \| 12 (16.9) \| 83 (56.1) \| 95 (43.4) \| <0.001 \| \| Monthly medication \| Monthly medication \| Monthly medication \| Monthly medication \| Monthly medication \| \| Yes \| 50 (69.4) \| 40 (27.0) \| 90 (40.9) \| <0.001 \| \| No \| 22 (30.6) \| 108 (73.0) \| 130 (59.1) \| <0.001 \| \| Iron, calcium and vitamin D supplements \| Iron, calcium and vitamin D supplements \| Iron, calcium and vitamin D supplements \| Iron, calcium and vitamin D supplements \| Iron, calcium and vitamin D supplements \| \| Yes \| 59 (81.9) \| 57 (38.5) \| 116 (52.7) \| <0.001 \| \| No \| 13 (18.1) \| 91 (61.5) \| 104 (47.3) \| <0.001 \| \| Food supply \| Food supply \| Food supply \| Food supply \| Food supply \| \| Yes \| 60 (83.3) \| 15 (10.1) \| 75 (34.1) \| <0.001 \| \| No \| 12 (16.7) \| 133 (89.9) \| 145 (65.9) \| <0.001 \| ## Predictors of LBW Seven out of 24 predictors were statistically significant, including monthly income, daily intake of iron, calcium, and vitamin D supplements, passive smoking, maternal anemia, gestational age, COVID-19 infection, and the impact of the COVID-19 pandemic on food supply (Table 5). The logistic regression analysis showed that a monthly income of 250–400 JD was associated with a 12.047-fold increase in the likelihood of having an LBW newborn compared to mothers with a monthly income of over 700 JD while controlling for other variables in the model [$95\%$ CI:1.680–86.401]. Table 5 \| Variable (in the final model a) \| B \| Exp (B) \| Odds Ratio (95% CI) \| \| --- \| --- \| --- \| --- \| \| Monthly income \| Monthly income \| Monthly income \| Monthly income \| \| ≤400 \| 2.489 \| 12.047* \| (1.680–86.401) \| \| 401-550 \| .719 \| 2.052 \| (0.290–14.506) \| \| 550-700 \| 1.149 \| 3.154 \| (0.524–18.973) \| \| >700 \| - \| - \| - \| \| Passive smoking (yes) \| 2.405 \| 11.078* \| (2.570–47.741) \| \| Iron, calcium and vitamin D supplements (daily) \| -1.857 \| 0.156* \| (0.040–0.610) \| \| Maternal anemia (yes) \| 1.607 \| 4.986* \| (1.287–19.318) \| \| Gestational age \| Gestational age \| Gestational age \| Gestational age \| \| Preterm \| 2.149 \| 8.577* \| (1.004–73.301) \| \| Full-term \| - \| - \| - \| \| COVID-19 infection (yes) \| 2.545 \| 12.745* \| (2.510–64.719) \| \| Food supply affected by the COVID-19 pandemic (yes) \| 2.880 \| 17.806* \| (2.196–144.343) \| Mothers exposed to passive smoking during pregnancy were more likely to have an LBW newborn (OR=2.405; $95\%$ CI: 2.570–47.741). Daily supplementation intake increased the odds of having a normal-weight newborn by 6.41 times compared to those who did not take daily supplements ($95\%$ CI: (0.040–0.610). Pregnant women with anemia were more likely to have a low birth weight newborn compared to those without anemia (OR=4.986; $95\%$CI: 1.287–19.318). Preterm delivery had 8.577 times higher odds of a low birth weight compared to full-term delivery ($95\%$ CI: 1.004–73.301). Mothers infected with COVID-19 had 12.75 times higher odds of having a low birth weight newborn ($95\%$ CI: 2.510–64.719). Moreover, mothers whose food supply was affected by the COVID-19 pandemic had 17.806-fold higher odds of having a low birth weight newborn than those whose food supply was not impacted ($95\%$ CI: 4.93–27.78). ## DISCUSSION This study aimed to investigate the relationship between maternal health factors and COVID-19 infection with low birth weight (LBW) in Jordan. Our results showed that a lower monthly income was linked to a higher likelihood of having an LBW newborn. Specifically, more mothers of LBW newborns had incomes below 400 JD in comparison to mothers of normal birth weight babies. The results are in line with other studies suggesting that pregnant women with low incomes may not be able to receive adequate nutrition and health care [9–12]. Pregnant women exposed to secondhand smoke during pregnancy had an increased risk of delivering low birth weight newborns, which is supported by several studies that link passive smoking to LBW [13–17]. The harmful effects of secondhand smoke on fetal growth and development can be attributed to the chemicals exhaled by smokers. Carbon monoxide, a component of secondhand smoke, reduces oxygen delivery to the fetus by forming carboxyhemoglobin and causing vasoconstriction through nicotine [18, 19]. Our findings indicate that a higher proportion of mothers of LBW newborns live with their extended family. This is probably due to a decrease in monthly income per person, which is associated with poorer diet quality during pregnancy [20]. Furthermore, the intake of iron, calcium, and vitamin D supplements during pregnancy was linked to a decreased probability of having a low birth weight newborn, corresponding to other studies [21–23]. The underlying mechanisms behind this association may include improved gestational weight gain and prevention of anemia, which can positively impact both the mother and the fetus [21, 24]. We found no significant association between LBW and the mother's body mass index (BMI) ($$p \leq 0.753$$), which contradicts previous findings [25, 26] that obesity increases the risk of LBW. However, the study did reveal an association between gestational weight gain and LBW risk. Pregnant women with a gestational weight gain of 6–10 kg had a higher risk of delivering LBW newborns, while those with a gestational weight gain of 10–16 kg had a lower risk. These findings are consistent with previous studies [5, 27–29]. For example, Zhao et al. [ 29] found that pregnant women with a gestational weight gain below the recommended range specified by the American Institute of Medicine (11.3–15.9 kg for those with normal pre-BMI) were at a higher risk of delivering low birth weight newborns, compared to women who had a gestational weight gain within this range. The number of prenatal visits was also found to have a negative association with LBW outcomes. These findings are consistent with previous research, which has demonstrated that pregnant women who did not attend antenatal care at least four times are more susceptible to LBW outcomes [30–32]. Additionally, there was a positive correlation between maternal anemia and LBW outcomes. This association has been previously established in a multitude of studies [33, 34] and was reinforced by the results of a systematic review and meta-analysis, which concluded that maternal anemia is positively associated with LBW outcomes [33]. Access to healthy food intake during pregnancy was a significant risk factor for LBW, with women without access being 2.88 times more likely to have an LBW newborn than those with access. These findings are in agreement with several studies [35–37]. Abubakari and Jahn [36] documented that a balanced dietary intake during pregnancy was positively correlated with a reduced risk of LBW. A higher prevalence of hypertension was observed among mothers of LBW newborns, a correlation substantiated by the findings of Liu et al. [ 38] and Rahman et al. [ 39]. Furthermore, female neonates were identified as being more susceptible to LBW, a result that concurs with the findings of Afaya et al. [ 40], Agorinya et al. [ 41], and Manyeh et al. [ 42]. Afaya et al. [ 40] reported that female neonates had $64\%$ higher odds of LBW than male neonates. Our results showed that the risk for LBW decreased significantly when the pregnancy was planned, supporting the result of other studies where unplanned pregnancies increased the risk of low birth weight by $24\%$ compared to planned pregnancies [43, 44]. A higher rate of LBW was observed within cesarean deliveries compared to vaginal deliveries. The results align with Chen et al. [ 45] and Taha et al. [ 46]. Chen et al. [ 45] found that the rate of cesarean section of LBW was 1.24 times higher than those of normal birth weight newborns. During the last ten years, the rate of cesarean deliveries in Jordan has increased [47–49]. More than fifty percent of women delivered by C-sections before 39 weeks of gestation, which is associated with a higher risk of neonatal complications [49]. Women with a history of LBW were significantly more likely to have recurrent LBW newborns, corresponding with other studies [50–53]. For example, Mvunta et al. [ 52] reported that women with a history of LBW were more likely to have recurrent LBW in late pregnancy compared to those who had a previous normal birth weight baby. This study also demonstrated that LBW was significantly associated with COVID-19 infection, hospital admission, losing work, and financial ability affected by the COVID-19 pandemic. Previous studies have shown that the COVID-19 pandemic adversely affects pregnant women and their newborns [34, 54, 55]. The COVID-19 outbreak was considered a major stress that may have negatively affected intrauterine development and increased preterm birth rates as well as low birth weight rates [54]. Previous studies have shown that the COVID-19 pandemic has negatively impacted pregnant women and newborns, potentially increasing stress levels, anxiety, and depression and leading to higher rates of preterm birth and LBW [56]. As a result, the preterm birth rate may rise, and intrauterine growth restriction, particularly low birth weight, may become more common. ## CONCLUSION The present study aimed to identify the risk factors associated with low birth weight (LBW) among pregnant women in Jordan. Results showed that several factors were significantly associated with LBW, including monthly income, daily intake of iron, calcium, and vitamin D supplements, exposure to passive smoking, maternal anemia, gestation age, COVID-19 infection, and disruptions in food supply and financial ability due to the COVID-19 pandemic. The findings of this study highlight the need for comprehensive educational programs for pregnant women that focus on prenatal care, proper nutrition, and supplement intake. Regular screening tests, including those for LBW, should also be a priority to prevent and manage this serious public health issue. It is important to note that these findings should be further validated through additional research in various hospital sectors across Jordan. Such efforts will contribute to a better understanding of the prevalence and risk factors of LBW, ultimately leading to improved maternal and fetal outcomes. ## Conflict of interest The authors declare no conflict of interest. ## Ethical approval Ethical approval for this study was obtained from the Ministry of Health and the Institutional Review Board (IRB) at Jordan University of Science and Technology ($\frac{148}{147}$/2020). ## Consent to participate Informed consent was obtained from all participants. ## Personal thanks The authors would like to thank all participants in this study and the Research Faculty at Jordan University of Science and Technology. ## Authorship AS and EA contributed to the design, concept, data collection, and manuscript writing. EmA contributed to design, concept, and manuscript writing. SM contributed to manuscript revision and editing. KK contributed to data analysis and manuscript writing. ## References 1. K C A, Basel PL, Singh S. Low birth weight and its associated risk factors: Health facility-based case-control study. PLoS One (2020) 15 e0234907. DOI: 10.1371/journal.pone.0234907 2. Low birthweight: country, regional and global estimates. World Health Organization (2004) 3. Obesity: preventing and managing the global epidemic: report of a WHO consultation. World Health Organization (2000) 4. Kramer MS. Determinants of low birth weight: methodological assessment and meta-analysis. Bull World Health Organ (1987) 65 663-737. PMID: 3322602 5. Gebremedhin M, Ambaw F, Admassu E, Berhane H. Maternal associated factors of low birth weight: a hospital based cross-sectional mixed study in Tigray, Northern Ethiopia. BMC Pregnancy Childbirth (2015) 15 222. DOI: 10.1186/s12884-015-0658-1 6. Rodo M, Singh L, Russell N, Singh NS. A mixed methods study to assess the impact of COVID-19 on maternal, newborn, child health and nutrition in fragile and conflict-affected settings. Confl Health (2022) 16 30. DOI: 10.1186/s13031-022-00465-x 7. Salami VU, Okoduwa SIR, Chris AO, Ayilara SI, Okoduwa UJ. Opinion Review of Socioeconomic Impact of COVID-2019 on Women's Health. Front Glob Womens Health (2021) 2 647421. DOI: 10.3389/fgwh.2021.647421 8. Islam MM, Ababneh F, Akter T, Khan HR. Prevalence and risk factors for low birth weight in Jordan and its association with under-five mortality: a population-based analysis. East Mediterr Health J (2020) 26 1273-1284. DOI: 10.26719/emhj.20.096 9. Martinson ML, Reichman NE. Socioeconomic Inequalities in Low Birth Weight in the United States, the United Kingdom, Canada, and Australia. Am J Public Health (2016) 106 748-54. DOI: 10.2105/AJPH.2015.303007 10. Garfield L, Holditch-Davis D, Carter CS, McFarlin BL. Risk factors for postpartum depressive symptoms in low-income women with very low-birth-weight infants. Adv Neonatal Care (2015) 15 E3-8. DOI: 10.1097/ANC.0000000000000131 11. Shaw SH, Herbers JE, Cutuli JJ. Medical and Psychosocial Risk Profiles for Low Birthweight and Preterm Birth. Womens Health Issues (2019) 29 400-406. DOI: 10.1016/j.whi.2019.06.005 12. Demelash H, Motbainor A, Nigatu D, Gashaw K, Melese A. Risk factors for low birth weight in Bale zone hospitals, South-East Ethiopia: a case-control study. BMC Pregnancy Childbirth (2015) 15 264. DOI: 10.1186/s12884-015-0677-y 13. Abusalah A, Gavana M, Haidich AB, Smyrnakis E. Low birth weight and prenatal exposure to indoor pollution from tobacco smoke and wood fuel smoke: a matched case-control study in Gaza Strip. Matern Child Health J (2012) 16 1718-27. DOI: 10.1007/s10995-011-0851-4 14. Goel P, Radotra A, Singh I, Aggarwal A, Dua D. Effects of passive smoking on outcome in pregnancy. J Postgrad Med (2004) 50 12-6. PMID: 15047992 15. Goel P, Radotra A, Singh I, Aggarwal A, Dua D. Effects of passive smoking on outcome in pregnancy. J Postgrad Med (2004) 50 12-6. PMID: 15047992 16. Xi C, Luo M, Wang T, Wang Y. Association between maternal lifestyle factors and low birth weight in preterm and term births: a case-control study. Reprod Health (2020) 17 93. DOI: 10.1186/s12978-020-00932-9 17. Yasmeen T, Sultana R, Khatoon T, Riaz S. Effect of Passive Smoking during Pregnancy on Birth Weight of Neonates. Pakistan J. Med. Heal. Sci (2022) 16 724-726. DOI: 10.53350/pjmhs22165724 18. Lambers DS, Clark KE. The maternal and fetal physiologic effects of nicotine. Semin Perinatol (1996) 20 115-26. DOI: 10.1016/s0146-0005(96)80079-6 19. von Kries R, Toschke AM, Koletzko B, Slikker W. Maternal smoking during pregnancy and childhood obesity. Am J Epidemiol (2002) 156 954-61. DOI: 10.1093/aje/kwf128 20. French SA, Tangney CC, Crane MM, Wang Y, Appelhans BM. Nutrition quality of food purchases varies by household income: the SHoPPER study. BMC Public Health (2019) 19 231. DOI: 10.1186/s12889-019-6546-2 21. Changamire FT, Mwiru RS, Peterson KE, Msamanga GI. Effect of multivitamin supplements on weight gain during pregnancy among HIV-negative women in Tanzania. Matern Child Nutr (2015) 11 297-304. DOI: 10.1111/mcn.12018 22. Jafari F, Eftekhar H, Pourreza A, Mousavi J. Socio-economic and medical determinants of low birth weight in Iran: 20 years after establishment of a primary healthcare network. Public Health (2010) 124 153-8. DOI: 10.1016/j.puhe.2010.02.003 23. Kalem MN, Kamalak Z, Kosus N, Kosus A, Kalem Z. Prenatal multivitamin supplementation increases birth weight. Int. J. Reprod. Contraception, Obstet. Gynecol (2017) 6 2148. DOI: 10.18203/2320-1770.ijrcog20172305 24. Zhang Q, Ananth CV, Li Z, Smulian JC. Maternal anaemia and preterm birth: a prospective cohort study. Int J Epidemiol (2009) 38 1380-9. DOI: 10.1093/ije/dyp243 25. Devaki G, Shobha R. Maternal anthropometry and low birth weight: A review. Biomed. Pharmacol. J (2018) 11 815-820. DOI: 10.13005/bpj/1436 26. Mohammadi M, Maroufizadeh S, Omani-Samani R, Almasi-Hashiani A, Amini P. The effect of prepregnancy body mass index on birth weight, preterm birth, cesarean section, and preeclampsia in pregnant women. J Matern Fetal Neonatal Med (2019) 32 3818-3823. DOI: 10.1080/14767058.2018.1473366 27. Gizaw B, Gebremedhin S. Factors associated with low birthweight in North Shewa zone, Central Ethiopia: case-control study. Ital J Pediatr (2018) 44 76. DOI: 10.1186/s13052-018-0516-7 28. Waits A, Guo CY, Chien LY. Inadequate gestational weight gain contributes to increasing rates of low birth weight in Taiwan: 2011-2016 nationwide surveys. Taiwan J Obstet Gynecol (2021) 60 857-862. DOI: 10.1016/j.tjog.2021.07.013 29. Zhao R, Xu L, Wu ML, Huang SH, Cao XJ. Maternal pre-pregnancy body mass index, gestational weight gain influence birth weight. Women Birth (2018) 31 e20-e25. DOI: 10.1016/j.wombi.2017.06.003 30. Acharya D, Singh JK, Kadel R, Yoo SJ. Maternal Factors and Utilization of the Antenatal Care Services during Pregnancy Associated with Low Birth Weight in Rural Nepal: Analyses of the Antenatal Care and Birth Weight Records of the MATRI-SUMAN Trial. Int J Environ Res Public Health (2018) 15 2450. DOI: 10.3390/ijerph15112450 31. Cunningham SD, Lewis JB, Shebl FM, Boyd LM. Group Prenatal Care Reduces Risk of Preterm Birth and Low Birth Weight: A Matched Cohort Study. J Womens Health (Larchmt) (2019) 28 17-22. DOI: 10.1089/jwh.2017.6817 32. Pinzón-Rondón ÁM, Gutiérrez-Pinzon V, Madriñan-Navia H, Amin J. Low birth weight and prenatal care in Colombia: a cross-sectional study. BMC Pregnancy Childbirth (2015) 15 118. DOI: 10.1186/s12884-015-0541-0 33. Figueiredo ACMG, Gomes-Filho IS, Batista JET, Orrico GS. Maternal anemia and birth weight: A prospective cohort study. PLoS One (2019) 14 e0212817. DOI: 10.1371/journal.pone.0212817 34. Schwartz DA. The Effects of Pregnancy on Women With COVID-19: Maternal and Infant Outcomes. Clin Infect Dis (2020) 71 2042-2044. DOI: 10.1093/cid/ciaa559 35. Durrani AM, Rani A. Effect of maternal dietary intake on the weight of the newborn in Aligarh city, India. Niger Med J (2011) 52 177-81. DOI: 10.4103/0300-1652.86132 36. Abubakari A, Jahn A. Maternal Dietary Patterns and Practices and Birth Weight in Northern Ghana. PLoS One (2016) 11 e0162285. DOI: 10.1371/journal.pone.0162285 37. Zerfu TA, Umeta M, Baye K. Dietary diversity during pregnancy is associated with reduced risk of maternal anemia, preterm delivery, and low birth weight in a prospective cohort study in rural Ethiopia. Am J Clin Nutr (2016) 103 1482-8. DOI: 10.3945/ajcn.115.116798 38. Liu Y, Li N, An H, Li Z. Impact of gestational hypertension and preeclampsia on low birthweight and small-for-gestational-age infants in China: A large prospective cohort study. J Clin Hypertens (Greenwich) (2021) 23 835-842. DOI: 10.1111/jch.14176 39. Rahman LA, Hairi NN, Salleh N. Association between pregnancy induced hypertension and low birth weight; a population based case-control study. Asia Pac J Public Health (2008) 20 152-8. DOI: 10.1177/1010539507311553 40. Afaya A, Afaya RA, Azongo TB, Yakong VN. Maternal risk factors and neonatal outcomes associated with low birth weight in a secondary referral hospital in Ghana. Heliyon (2021) 7 e06962. DOI: 10.1016/j.heliyon.2021.e06962 41. Agorinya IA, Kanmiki EW, Nonterah EA, Tediosi F. Socio-demographic determinants of low birth weight: Evidence from the Kassena-Nankana districts of the Upper East Region of Ghana. PLoS One (2018) 13 e0206207. DOI: 10.1371/journal.pone.0206207 42. Manyeh AK, Kukula V, Odonkor G, Ekey RA. Socioeconomic and demographic determinants of birth weight in southern rural Ghana: evidence from Dodowa Health and Demographic Surveillance System. BMC Pregnancy Childbirth (2016) 16 160. DOI: 10.1186/s12884-016-0956-2 43. Flower A, Shawe J, Stephenson J, Doyle P. Pregnancy planning, smoking behaviour during pregnancy, and neonatal outcome: UK Millennium Cohort Study. BMC Pregnancy Childbirth (2013) 13 238. DOI: 10.1186/1471-2393-13-238 44. Shah PS, Balkhair T, Ohlsson A, Beyene J. Intention to become pregnant and low birth weight and preterm birth: a systematic review. Matern Child Health J (2011) 15 205-16. DOI: 10.1007/s10995-009-0546-2 45. Chen Y, Wu L, Zhang W, Zou L, Li G, Fan L. Delivery modes and pregnancy outcomes of low birth weight infants in China. J Perinatol (2016) 36 41-6. DOI: 10.1038/jp.2015.137 46. Taha Z, Ali Hassan A, Wikkeling-Scott L, Papandreou D. Factors Associated with Preterm Birth and Low Birth Weight in Abu Dhabi, the United Arab Emirates. Int J Environ Res Public Health (2020) 17 1382. DOI: 10.3390/ijerph17041382 47. Abuhammad S, Mukattash TL, Alazzam SI, Yafawi R. Caesarean section delivery from maternal perspective: An exploratory study in Jordan. Int J Clin Pract (2021) 75 e14349. DOI: 10.1111/ijcp.14349 48. Batieha AM, Al-Daradkah SA, Khader YS, Basha A. Cesarean Section: Incidence, Causes, Associated Factors and Outcomes: A National Prospective Study from Jordan. Gynecol. Obstet. Case Rep (2017) 3 1-11. DOI: 10.21767/2471-8165.1000055 49. Khasawneh W, Obeidat N, Yusef D, Alsulaiman JW. The impact of cesarean section on neonatal outcomes at a university-based tertiary hospital in Jordan. BMC Pregnancy Childbirth (2020) 20 335. DOI: 10.1186/s12884-020-03027-2 50. Boo NY, Lim SM, Koh KT, Lau KF, Ravindran J. Risk factors associated with low birth weight infants in the Malaysian population. Med J Malaysia (2008) 63 306-10. PMID: 19385490 51. Metgud CS, Naik VA, Mallapur MD. Factors affecting birth weight of a newborn--a community based study in rural Karnataka, India. PLoS One (2012) 7 e40040. DOI: 10.1371/journal.pone.0040040 52. Mvunta MH, Mboya IB, Msuya SE, John B. Incidence and recurrence risk of low birth weight in Northern Tanzania: A registry based study. PLoS One (2019) 14 e0215768. DOI: 10.1371/journal.pone.0215768 53. Sutan R, Mohtar M, Mahat AN, Tamil AM. Determinant of Low Birth Weight Infants: A Matched Case Control Study. Open J. Prev. Med (2014) 4 91-99. DOI: 10.4236/ojpm.2014.43013 54. Kirchengast S, Hartmann B. Pregnancy Outcome during the First COVID 19 Lockdown in Vienna, Austria. Int J Environ Res Public Health (2021) 18 3782. DOI: 10.3390/ijerph18073782 55. Kyle MH, Glassman ME, Khan A, Fernández CR. A review of newborn outcomes during the COVID-19 pandemic. Semin Perinatol (2020) 44 151286. DOI: 10.1016/j.semperi.2020.151286 56. Wu Y, Zhang C, Liu H, Duan C. Perinatal depressive and anxiety symptoms of pregnant women during the coronavirus disease 2019 outbreak in China. Am J Obstet Gynecol (2020) 223 240.e1-240.e9. DOI: 10.1016/j.ajog.2020.05.009
--- title: Practices, attitudes and knowledge of midwives and nurses regarding gestational diabetes and pregnancy-induced hypertension authors: - Daniela Stan - Claudia Elena Dobre - Doina Carmen Mazilu - Elvira Brătilă journal: Journal of Medicine and Life year: 2023 pmcid: PMC10015574 doi: 10.25122/jml-2023-0021 license: CC BY 3.0 --- # Practices, attitudes and knowledge of midwives and nurses regarding gestational diabetes and pregnancy-induced hypertension ## Abstract Midwives (M) and obstetric nurses (ON) play a critical role in providing healthcare for pregnant patients at all stages of pregnancy, and ongoing training and education are essential to ensure the best outcomes. This longitudinal quantitative research study aimed to assess the impact of an educational program on the knowledge, attitudes, and practices of 125 midwives and obstetric nurses regarding care for patients with gestational diabetes and pregnancy-induced hypertension. The original questionnaire consisted of 56 items grouped into 3 subscales assessing knowledge (15 items), attitudes (18 items), and practices (23 items). The questionnaire was administered at three distinct intervals during the educational program: pre-test, post-test, and follow-up at three months. The data were analyzed using ANOVA and Pearson correlation coefficients to determine the significance of the differences between the 3 moments of the administration of the questionnaire. There was a significant increase in the level of knowledge, attitudes, and practices of midwives and obstetric nurses following the training module, which was sustained at 3 months after completion compared to pre-training. The comparative analysis of the total scores for every 3 sets of items revealed the positive impact of the educational program on the level of knowledge, attitudes, and practices of midwives and obstetric nurses. ## INTRODUCTION Midwives (M) and obstetric nurses (ON) play a critical role in providing primary health care to pregnant patients, and their role is recognized both internationally and nationally. In order to improve the health outcomes of both mothers and children, it is essential to make sure that the quality of their assessment, care, and treatment is at the highest standard. These healthcare providers are deeply engaged in the local community, which allows them to deliver effective interventions that meet the needs of patients, families, and the community at large [1]. The World Health Organization (WHO) emphasized the main priorities for establishing global strategic directions for M and ON in a material focused on 4 areas of interest. These areas include education, jobs, leadership, and service delivery. Adopting and supporting these public policies can lead to increasing the number of M and ON, securing jobs, managing migration, recruiting and retaining M and ON in the areas where they are most needed, developing and strengthening nursing medical leadership in the health and educational systems, and ensuring the respect, protection, and motivation of these categories of medical personnel to obtain their optimal contribution to the provision of health care [2]. Currently, there is a shortage of midwives in hospitals, ambulatory profile units, and community care units in Romania. Unfortunately, a downward trend is estimated in the future as a result of the reduced number of education units that train these specialists. The COVID-19 pandemic has highlighted the need for midwives in health systems, especially primary health care. The pandemic has underscored the importance of protective measures and investment in all health system specialists involved in healthcare activities, public health services, and the provision of essential medical services [2]. Subsequently, it is essential to implement policies that support, develop, and strengthen the role of midwives in the Romanian health system at the national level. Gestational diabetes (GD) and pregnancy-induced hypertension (PIH) are two common pathologies among pregnant women, especially at extreme ages [3-5]. GD is a transitory metabolic disorder characterized by impaired glucose tolerance during pregnancy, leading to a high glycemic index and serious maternal and fetal complications [6]. The main maternal-fetal complications induced by GD are represented by hydramnios, spontaneous abortion, hypertension, preeclampsia, eclampsia, fetal macrosomia, shoulder dystocia, respiratory distress syndrome, neonatal hypoglycemia and even perinatal mortality [7-8]. Preeclampsia is a hypertensive disorder occurring in the second half of pregnancy characterized by high blood pressure and proteinuria (protein in the urine). This condition affects the main organs in the body, such as the brain, liver, kidneys, or placenta, and can have negative effects on the normal evolution of pregnancy [9]. Screening and preventive care are crucial in minimizing the impact of these pathologies on both the mother and fetus. It is essential that M and ON involved in the care of pregnant women, have the necessary theoretical and practical knowledge to effectively evaluate, diagnose, and manage preventable pathologies during pregnancy. This expertise enables them to intervene effectively in the evaluation, diagnosis, and management of the pregnant patient, providing effective healthcare and reducing infant and maternal mortality [10]. Globally, a 20-year analysis indicated a significant decrease in neonatal deaths from 400 to 210 per 100,000 live births between 1990 and 2010 [10]. Assessing the level of knowledge, attitudes, and practices of M and ON caring for pregnant patients can provide essential directions for nursing leadership to improve current care practices. Specialized literature indicates that a lack of knowledge and limited access to the best evidence of care are the primary factors contributing to the ineffective management of pregnant patients in identifying and preventing pregnancy-related pathologies. A study in eastern South Africa identified deficiencies in midwives' knowledge of hypertensive conditions management during pregnancy [10]. The findings of Utz et al. [ 11] on the knowledge and practices of Moroccan general practitioners, midwives, and obstetric nurses caring for pregnant women with GD showed that they had a basic knowledge of the management of GD, while their reported practices were not uniform and reflected discrepancies with national guidelines in the field [11]. The study concluded that updating the knowledge of professionals in the field can result in more effective management of gestational diabetes (GD) within primary healthcare systems [11]. Another study conducted by Suff et al.[12] in the United Kingdom indicated a low level of theoretical and practical knowledge regarding the diagnosis and management of high blood pressure (Hbp) in pregnancy. Chepulis et al. conducted a study in New Zealand to assess the implementation of national screening guidelines for GD. The study found that the guidelines were not implemented in the sample of midwives included in the study, indicating a knowledge gap that may need to be addressed through education in this area [13]. The scoping review by Garti et al. using the three-step JBI methodology found similar results regarding a widespread lack of knowledge among practicing midwives globally in managing pregnancy-induced hypertension, as seen in the analysis of 29 studies. These findings highlight the need to develop accessible and innovative training programs for midwives and to establish policies that prioritize the professional development of midwives to increase their knowledge and skills in managing hypertension during pregnancy [14]. As per the current national legislation, midwives and nurses have a set of responsibilities which include identifying and planning care needs, patient education, monitoring vital functions, and treatment administration. Midwives and nurses are directly involved in the care of pregnant women with GD and PIH. They should possess comprehensive knowledge and practices to develop and apply an appropriate care plan for patients during antenatal and intrapartum periods. It is essential to screen pregnant women for GD and PIH from the first stage of hospitalization and to document the results of these evaluations in the patient's care plan. Providing health education to a pregnant woman with GD and PIH is necessary for the care given by M and ON and should include elements related to nutritional therapy, physical activity, and periodic monitoring of blood glucose and pressure. Maintaining blood glucose levels within normal limits is the main objective of care for pregnant women with GD. This can be achieved through interdisciplinary consultations with a diabetologist, nutritionist, or physiotherapist and by including the recommendations in the patient's care plan. The lack of a standardized protocol for providing maternal care to patients at risk can represent a significant barrier to the delivery of appropriate healthcare practices. The main objective of our study was to evaluate the level of knowledge, attitudes, and practices of midwives and obstetric nurses at the Obstetrics and Gynecology Clinical Hospital Prof. Dr. Panait Sîrbu in providing care to pregnant women with GD and PIH. Our goal was to identify educational needs for developing a comprehensive training program to improve the care provided by healthcare providers. The second objective of our study was to assess the impact of a training program focused on GD and PIH on the level of knowledge, attitudes, and practices of midwives and obstetric nurses. ## MATERIAL AND METHODS The study was designed as a longitudinal study with two primary objectives. The first objective was to develop an educational program that met the educational needs of M and ON, who provide direct care for pregnant patients with GD and Hbp. The second objective was to evaluate the impact of the educational program on the level of knowledge, attitudes, and practices of M and ON. The study was conducted at the Obstetrics and Gynecology Prof. Dr. Panait Sîrbu Hospital, which provides medical services in obstetrics and gynecology, with an average of 531 hospitalizations per month, comprising both inpatient and outpatient care. The hospital has a capacity of 370 beds for inpatient care. In 2020, the total number of hospitalizations was 15,595, of which 4,836 were outpatient hospitalizations, 11,119 were inpatient hospitalizations, and 5,843 were deliveries. Of the total number of 234 midwives and obstetric nurses working in the hospital, 62 are midwives. To determine the appropriate population sample size for the study, calculations were made based on the total number of 234 professionals, with a confidence level of $95\%$ and a $6\%$ margin of error. As a result, 125 participants were randomly selected from the total number of M and ON who agreed to participate in this research. Other categories of healthcare personnel, such as nurses working in the neonatology ward, physiotherapists, doctors, students, or other care staff, were excluded from the analysis. To minimize the potential for bias, the research team maintained consistency in the study sample throughout all three stages of the study (M1, M2, M3), using the same group of 125 M and ON. The study employed a quantitative research method to collect data on the knowledge, attitudes, and practices of M and ON working at a specialized hospital in Bucharest. The questionnaire was pre-tested and validated within a focus group of 25 experts in the field. We evaluated the questionnaire to assess its adaptability to the practices and competencies of M and ON in Romania, as well as the clarity of the text, readability, answer choice alternatives, contextual expressions, and the degree of difficulty of the items. The questionnaire used in our study consisted of a total of 56 questions that evaluated the practices, attitudes, and level of knowledge of midwives and obstetric nurses (M and ON) in managing gestational diabetes (GD) and pregnancy-induced hypertension (PIH). Specifically, the survey included 23 items about practices, 18 items about attitudes, and 15 items about knowledge of GD and PIH. The research was conducted in 5 main stages. Firstly, the practices, attitudes, and knowledge of M and ON professionals caring for pregnant patients with GD and PIH (April 15, 2020) were assessed. Secondly, an educational program was developed based on the identified educational needs following the initial evaluation (moment 1 - M1) (May 1, 2020, to June 1, 2020). Thirdly, the educational program was delivered to a group of 125 M and ON professionals (July 1, 2020 - July 5, 2020). Fourthly, the practices, attitudes, and knowledge of M and ON professionals were evaluated immediately after completing the training program (moment 2 - M2) (July 5, 2020). Lastly, the practices, attitudes, and knowledge were reassessed 3 months after completion of the training program (moment 3 - M3) (October 1, 2020). The level of knowledge, attitudes, and practices of all M and ON who participated in the study were analyzed. The educational program was drafted based on the educational needs identified among M and ON professionals included in the study at the beginning of the evaluation. The program included information on PIH, hypertension treatment, predictive tests for pathological increases in hypertension in the last trimester of gestation, the role of M and ON professionals in the follow-up and supervision of PIH, HELP syndrome, GD, and pregnancy, and its causes and effects on pregnancy, birth and delivery, the role of the M and ON in the care of patients with GD during pregnancy, during labor, and after delivery. The data processing was performed using SPSS 20.0 program. To determine the statistical significance of the average scores in each of the three domains (practices, attitudes, and knowledge), we used analysis of variance (ANOVA). Additionally, we calculated Pearson correlation coefficients (r) to explore the relationship between the scores of the three domains and socio-professional characteristics such as age, education level, gender, professional experience, and place of work. ## RESULTS The questionnaire was administered to 125 medical M and ON professionals working in the obstetrics and gynecology departments of the Clinical Hospital of Obstetrics and Gynecology Prof. Dr. Panait Sîrbu. The distribution of the sample by various characteristics is shown in Table 1. The average age of participants was 44.7 years, and the average professional experience was 18.2 years (Table 1). Table 1 \| Characteristics \| N \| % \| \| --- \| --- \| --- \| \| Age a \| Age a \| Age a \| \| 20–29 \| 8 \| 6.4 \| \| 30–39 \| 19 \| 15.2 \| \| 40–49 \| 55 \| 44.0 \| \| 50–59 \| 38 \| 30.4 \| \| ≥60 \| 2 \| 1.6 \| \| No answer \| 3 \| 2.4 \| \| Gender \| Gender \| Gender \| \| Female \| 124 \| 99.2 \| \| Male \| 1 \| 0.8 \| \| Education level \| Education level \| Education level \| \| Nurse with sanitary high school \| 5 \| 4.0 \| \| Nurse with post-secondary health school \| 30 \| 24.0 \| \| Nurse with a university education \| 26 \| 20.8 \| \| Midwife \| 57 \| 45.6 \| \| Nurse with master's degree \| 7 \| 5.6 \| \| Professional experience (years) b \| Professional experience (years) b \| Professional experience (years) b \| \| ≤4 \| 19 \| 15.2 \| \| 5–9 \| 10 \| 8.0 \| \| 10–14 \| 11 \| 8.8 \| \| 15–19 \| 23 \| 18.4 \| \| 20–24 \| 21 \| 16.8 \| \| 25–29 \| 22 \| 17.6 \| \| 30–34 \| 15 \| 12.0 \| \| ≥35 \| 4 \| 3.2 \| \| Department \| Department \| Department \| \| Obstetrics-gynecology \| 76 \| 60.8 \| \| Delivery room \| 22 \| 17.6 \| \| Intensive therapy \| 16 \| 12.8 \| \| Outpatient \| 6 \| 4.8 \| \| Emergency room \| 5 \| 4.0 \| The evaluation of healthcare practices revealed a wide range of practices with relatively high variability during the initial phase of the research. The respondents reported rarely or never reading articles, journals, or books related to the research topics. Before participating in the educational program, the respondents recorded average scores for practices, attitudes, and knowledge. This suggests that despite $60.8\%$ of respondents having attended training sessions on GD and PIH prevention and management, the level of knowledge, attitudes, and practices prior to the training was average, with correct answers varying between $52.8\%$ to $92.8\%$. The second phase of the study highlighted the significant impact of the training module on the participant's level of knowledge, attitudes, and practices. Specifically, there was a significant increase in the weights of responses indicating the permanent use of correct practices immediately after the training for all items assessed, and this increase was sustained three months after the training. However, one item related to the use of calibrated and well-maintained devices for measuring blood pressure showed a significant increase in the proportion of respondents who answered "always" immediately after the training. However, this difference was not significant when compared to the pre-training and 3-month post-training results. The score for the practices scale was calculated by summing the correct answers for the 23 questions related to practices, resulting in a score that could theoretically range from 0 to 23. The scores were divided into 3 ranges: 0-8 points (low scores), 9-16 points (medium scores), and 17-23 points (high scores) for the graphic representation. The proportion of respondents who achieved high scores (between 17 and 23 points) on the practice scale increased from approximately $65\%$ pre-training to over $93\%$ immediately after and further increased to over $96\%$ at 3 months follow-up (Figure 1). The average scores on the practice scale differed significantly between the three stages of the study, as indicated by Table 2. Figure 1: Practice scale scores. TABLE_PLACEHOLDER:Table 2 The attitudes of M and ON were evaluated in the three stages of the training program, and the percentage of high scores on the attitude scale significantly increased after the training, from approximately $51\%$ to almost $89\%$. At both the immediate and 3-month follow-up assessments, all participants achieved scores within the high range of 13-18 points, as shown in Figure 2. The attitude score was calculated by summing the correct answers for the 18 attitude questions and could vary between 0 and 18. Scores were divided into three ranges: low (0-6 points), medium (7-12 points), and high (13-18 points) for the graphic representation. Figure 2: Attitude scale scores. The average scores also increased significantly, as confirmed by the analysis of variance (ANOVA) (Table 3). Table 3 \| Unnamed: 0 \| M \| SD \| Confidence Intervals 95% \| df \| ANOVA \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Before training (N=125) \| 12.08 \| 3.5 \| 11.46–12.69 \| 2 \| F=148.26 \| \| Immediately after training (N=125) \| 16.23 \| 2.42 \| 15.80–16.66 \| 2 \| F=148.26 \| \| 3 months after training (N=125) \| 17.21 \| 0.82 \| 17.06–17.36 \| 2 \| F=148.26 \| The comparative assessment of the level of knowledge regarding PIH and GD at the three evaluation stages revealed some correct answers, but the proportion of correct responses significantly increased after completing the training module and remained significantly elevated three months after its completion compared to the pre-training stage, as illustrated in Figure 3. Figure 3:** Knowledge scale scores. Similarly, the score for the knowledge scale was determined by adding up the correct responses to the 15 questions, with possible scores ranging from 0 to 15. The scores were divided into 3 ranges: 0-5 points (low scores), 6-10 points (medium scores), and 11-15 points (high scores) for the graphic representation. Figure 3 shows a significant increase in the percentage of high scores (11-15 points) on the knowledge scale, from $48\%$, before the training, to over $95\%$ in the 2 subsequent stages of the training. This growth is also confirmed by the significant differences in the average scores between the pre-training and the other 2 post-training stages (Table 4). Table 4 \| Unnamed: 0 \| m \| SD \| Confidence Intervals 95% \| df \| ANOVA \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Before training (N=125) \| 10.32 \| 1.96 \| 9.97–10.66 \| 2 \| F=207.61 \| \| Immediately after training (N=125) \| 14.12 \| 1.26 \| 13.89–14.34 \| 2 \| F=207.61 \| \| 3 months after training (N=125) \| 13.5 \| 1.42 \| 13.25–13.75 \| 2 \| F=207.61 \| There is an exception recorded for item 18 regarding the definition of GD, for which the percentage of correct answers recorded is not significant. This result is likely because participants answered this item correctly in the pre-training stage. For the item related to the definition of proteinuria, there was no significant difference between the pre-training and 3-month post-training assessments, as the percentage of correct answers was consistently high (over $90\%$) across all three stages. It is worth noting that for two other items (16 and 25, which concerned the value of blood pressure measurement for preeclampsia diagnosis and identifying non-risk factors for GD, respectively), the percentage of correct answers showed a significant decrease at the 3-month post-training assessment compared to the immediately post-training assessment. However, these percentages remained significantly higher than those at the pre-training stage. Following the training program, all three scales were significantly and positively correlated. In contrast, prior to the training, no significant correlation was found between the attitude and practice scale scores, according to the data presented in Table 5. Table 5** \| Item \| Practice score \| Attitudes score \| Knowledge score \| Age \| Professional experience \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| M1 – before training \| M1 – before training \| M1 – before training \| M1 – before training \| M1 – before training \| M1 – before training \| \| Practice scale score \| - \| 0.21 \| 0.12* \| 0.26 \| 0.20 \| \| Attitudes scale score \| \| - \| 0.19* \| 0.04 \| 0.01 \| \| Knowledge scale score \| \| \| - \| 0.09 \| -0.004 \| \| Age \| \| \| \| - \| 0.74 \| \| Professional experience \| \| \| \| \| - \| \| M2 – immediately after training \| M2 – immediately after training \| M2 – immediately after training \| M2 – immediately after training \| M2 – immediately after training \| M2 – immediately after training \| \| Practice scale score \| - \| 0.41 \| 0.32 \| 0.12 \| 0.24 \| \| Attitudes scale score \| \| - \| 0.38* \| 0.16 \| 0.18* \| \| Knowledge scale score \| \| \| - \| 0.10 \| -0.14 \| \| Age \| \| \| \| - \| 0.74** \| \| Professional experience \| \| \| \| \| - \| \| M3 – 3 months after training \| M3 – 3 months after training \| M3 – 3 months after training \| M3 – 3 months after training \| M3 – 3 months after training \| M3 – 3 months after training \| \| Practice scale score \| - \| 0.24* \| 0.48** \| 0.10 \| 0.16* \| \| Attitudes scale score \| \| - \| 0.26** \| 0.14 \| 0.18* \| \| Knowledge scale score \| \| \| - \| 0.04 \| -0.07 \| \| Age \| \| \| \| - \| 0.74 \| \| Professional experience \| \| \| \| \| - \| Participants reported the most frequently encountered challenges in providing appropriate care to patients with PIH or GD. In all 3 stages of the study, the participants identified the high volume of work and the lack of staff as the most common barriers. However, the lack of healthcare protocols was also considered a significant obstacle to providing good care, particularly in the post-training stages. In contrast, the lack of courses dedicated to this topic was cited more frequently in the pre-training stage than in the post-training stages. ## DISCUSSION The main objective of our study was to evaluate the knowledge, attitudes, and practices of midwives and obstetric nurses at the Obstetrics and Gynecology Clinical Hospital Prof. Dr. Panait Sîrbu in caring for pregnant women with GD and PIH. The evaluation was conducted at the 3 points: before the training, immediately after its completion, and three months later to assess the impact of the training on participants’ knowledge, attitudes, and practice. The comparative and correlational analysis of the three scales (practices, attitudes, knowledge) revealed a positive impact of the training program on M and ON practices, attitudes, and knowledge. This is further supported by the comparative analysis of the total scores of each set of items. In light of the positive outcomes observed across all three evaluation scales (i.e., practices, attitudes, and knowledge) immediately after the training and at the three-month follow-up, the study also examined the correlation between age and professional experience. Before the training, only the practice scale scores had a significant positive correlation with age and professional experience. However, immediately after the training and at 3 months follow-up, both the practice and the attitude scale scores had a positive correlation with work experience, with higher scores reported among those with more experience. On the other hand, the knowledge scale did not show a significant correlation with age or professional experience. These findings highlight the positive impact of the training program on knowledge improvement, regardless of the participant's age or professional experience. Comparative findings with our study have been reported by other researchers who assessed the level of knowledge in managing medical care for patients with PIH [10, 14-16] or GD [11, 13, 17]. A similar study conducted in a Romanian hospital initially revealed limited knowledge about preeclampsia and eclampsia among midwives and resident doctors caring for pregnant patients at different stages of pregnancy with PIH. Consequently, researchers recommended the development and promotion of studies that evaluate the impact of training sessions on this topic [18]. The development and implementation of an intensive training program focused on caring for patients with pregnancy-induced hypertension resulted in a significant improvement in the level of knowledge among the medical staff in the study. This finding is likely to benefit the care process provided by the medical staff [19]. Another study by Anyanti et al.[17] revealed a significant knowledge gap among health workers regarding the diagnosis and treatment of PIH and GD. The authors attributed this deficit to the lack of healthcare protocols tailored to the skills of medical staff in Nigeria [17]. The authors suggested the development of guidelines that highlight the specific role of each professional category involved in the care of patients with Hbp and GD and that specific work procedures should be available at the care points to ensure good treatment results [17]. Our results indicate that participants significantly improved their knowledge, attitudes, and practices regarding the care of patients with Hbp and GD after completing the training program. This improvement was sustained at a high level even after 3 months of completing the training program. These results highlight the need for educational programs for M and ON, focused on the healthcare needs of pregnant patients at risk of developing GD and gestational Hbp. In addition to training programs, it is necessary to develop guidelines and work protocols for each category of personnel tailored to their specific skills and duties in the care process, as suggested by Anyanti et al. [ 17]. Our respondents indicated a lack of specific care protocols based on the best evidence of good practices in the field, which is a concern in Romania, given the differences in healthcare skills between midwives and obstetric nurses. In addition, implementing clinical audit activities on care protocols can represent an effective solution to ensure that the best care practices in this field are implemented in clinical practice. Another critical aspect of care during pregnancy is patient education. It is crucial to increase the parturient awareness of the possible risks that may develop during pregnancy. Patient education regarding GD and Hbp prevention plays an essential role in the care process provided by M and ON. In a systematic review, Gholami et al.[20] investigated the effect of educational interventions on the knowledge of pregnant women regarding hypertensive disorders that may occur during pregnancy. The review included 6 studies on this topic, and the results of the analysis showed that multimodal patient education through educational brochures, mobile applications, combinations of brochures, iconographic images, videos, and PowerPoint presentations had a positive and significant effect on the awareness of pregnant women regarding the complications associated with hypertensive disorders of pregnancy. These findings suggest that providing adequate education for patients could reduce the number of severe complications caused by hypertensive disorders [20]. When implementing the best care practices for patients with gestational diabetes (GD) and hypertension (Hbp), healthcare professionals should consider both the facilitators and barriers that may impact the care provided by M and ON. In our study, participants identified high workload, doubled by the lack of staff and specific care protocols, as the main barriers in the care process. A study in Ontario by Murray-Davis et al. assessed barriers and facilitators that may influence the improvement of midwife care practices for patients with GD or Hbp [21]. Midwives' behavior can be influenced by their knowledge, skills, social and professional role and identity, care context, and resources, especially when collaborating with other care providers. Integrating other specialists in the care of patients with GD or Hbp can improve midwives' experiences when providing specific care to patients with these pathologies [21]. PIH and GD are frequent complications associated with pregnancy at extreme ages and can have a negative effect from a medical, psychological, and social point of view, both for the parturient and the fetus. Periodic continuing medical education programs focused on this topic can improve the ability of midwives and obstetric nurses to identify these pathologies early and provide preventive healthcare. The positive effect of the training program on the practices, attitudes, and knowledge of M and ON identified in our study supports this conclusion. Additionally, implementing a healthcare protocol specifically designed for midwives and obstetric nurses on the care of pregnant patients at risk of developing these pathologies can be a viable solution for improving healthcare practices in obstetrics and gynecology departments across hospitals in Romania. Such protocols can also be valuable for professionals working in independent practice. Furthermore, conducting clinical audits can be an effective way to ensure that healthcare protocols are being appropriately implemented in clinical practice. The study has some limitations due to the relatively small sample size and the fact that it was conducted only at a single specialized hospital. Subsequently, the conclusions of the study cannot be readily extended to all hospitals in Bucharest or the country. ## CONCLUSION The results of our study highlight the importance of developing and implementing care protocols and educational programs for midwives and obstetric nurses to guide them in providing specific care for the prevention of Hbp and GD at all points of care. In addition, developing educational guidelines for pregnant patients can improve patient understanding of the risks associated with these conditions and increase compliance with care recommendations. Furthermore, our findings provide important evidence for healthcare leadership to support the value of continuing medical education in improving the medical practice of midwives and obstetric nurses. ## Conflicts of interest The authors declare no conflict of interest. ## Ethical approval The study was approved by the ethics committee of the Clinical Hospital of Obstetrics and Gynecology Prof. Dr. Panait Sîrbu (protocol no. $\frac{9}{11.03.2020}$). ## Consent to participate Written informed consent was obtained from the participants. Participation was voluntary, and they were enrolled in the study after signing the informed consent form, which outlined the study's purpose, objectives, identity protection measures, and the right to withdraw from the study without justification at any time. ## Personal thanks We express our gratitude to Mrs. Ilona Voicu for her contribution to the statistical analysis of the data and Ms. Simona Rasu for providing English translation support, both of which were instrumental in the successful completion of this study. ## Authorship DS contributed to conducting the educational intervention, collecting the data, and writing the article. DS, CE-D, and DC-M contributed to the study design and data analysis, edited the manuscript, finalized verification, and sent the document to publication. EB verified the research design, assisted with data collection and analysis, and supervised the smooth running of the project. All authors discussed the results and contributed to the final form of the article. ## References 1. State of the world's nursing 2020: investing in education, jobs and leadership. Web Annex. Nursing roles in 21st-century health systems. Geneva: 2020 Licence: CC BY-NC-SA 3.0 IGO 2. Global strategic directions for nursing and midwifery 2021-2025*. Jama* (2021.0) 292 30 3. Claramonte Nieto M, Meler Barrabes E, Garcia Martínez S, Gutiérrez Prat M, Serra Zantop B. Impact of aging on obstetric outcomes: Defining advanced maternal age in Barcelona. BMC Pregnancy Childbirth (2019.0) 19 71-5. DOI: 10.1186/s12884-019-2415-3 4. Stan D, Mazilu DC, Dobre CE, Brătilă E. Mother’s health risks in extreme age pregnancies. Ginecologiaro (2022.0) 2 9. DOI: 10.26416/Gine.36.2.2022.6552 5. Bouzaglou A, Aubenas I, Abbou H, Rouanet S. Pregnancy at 40 years Old and Above: Obstetrical, Fetal, and Neonatal Outcomes. Is Age an Independent Risk Factor for Those Complications?. Frontiers in Medicine (2020.0) 7 208. DOI: 10.3389/fmed.2020.00208 6. Fong E. MBBS Evidence Summary. Gestational diabetes: Management (Diet). JBI EBP database. (2021.0) JBI-ES-4721 7. O’Sullivan EP, Avalos G, O’Reilly M, Dennedy MC. Atlantic Diabetes in Pregnancy (DIP): The prevalence and outcomes of gestational diabetes mellitus using new diagnostic criteria. Diabetologia (2011.0) 54 1670-5. DOI: 10.1007/s00125-011-2150-4 8. Management of diabetes in pregnancy: Standards of Medical Care in Diabetes 2020. Diabetes Care (2020.0) 43 S183-S192. DOI: 10.2337/dc20-S014 9. Whitehorn A, Fong E. Evidence Summary. Pre-eclampsia: Prevention. The JBI EBP Database (2021.0) JBI-ES-2591-2 10. Stellenberg EL, Ngwekazi NL. Knowledge of midwives about hypertensive disorders during pregnancy in primary healthcare. African Journal of Primary Health Care & Family Medicine (2016.0) 8 1-6. DOI: 10.4102/phcfm.v8i1.899 11. Utz B, Assarag B, Essolbi A, Barkat A. Knowledge and practice related to gestational diabetes among primary health care providers in Morocco: Potential for a defragmentation of care?. Primary Care Diabetes (2017.0) 11 389-396. DOI: 10.1016/j.pcd.2017.04.005 12. Suff N, Jatoth P, Khalil A, O’Brien P. Knowledge of hypertensive disorders in pregnancy among obstetricians and midwives. Archives of Disease in Childhood-Fetal and Neonatal (2011.0) 96 Fa107. DOI: 10.1136/adc.2011.300163.35 13. Chepulis L, Morison B, Tamatea J, Paul R. Midwifery awareness of diabetes in pregnancy screening guidelines in Aotearoa New Zealand. Midwifery (2022.0) 106 103230. DOI: 10.1016/j.midw.2021.103230 14. Garti I, Gray M, Tan JY, Bromley A. Midwives’ knowledge of pre-eclampsia management: A scoping review. Women and Birth (2021.0) 34 87-104. DOI: 10.1016/j.wombi.2020.08.010 15. Ramadurg U, Vidler M, Charanthimath U, Katageri G. Community health worker knowledge and management of pre-eclampsia in rural Karnataka State, India. Reprod Health (2016.0) 13 113. DOI: 10.1186/s12978-016-0219-8 16. Indarti J, Prasetyo S. Knowledge of Midwives as a Healthcare Provider About Hypertensive Disorders During Pregnancy. Indonesian Journal of Obstetetrics and Gynecology (2019.0) 7 9. DOI: 10.32771/inajog.v7i1.638 17. Anyanti J, Akuiyibo S, Idogho O, Amoo B, Aizobu D. Hypertension and diabetes management practices among healthcare workers in Imo and Kaduna States, Nigeria: An exploratory study. Risk Management and Healthcare Policy (2020.0) 13 2535-2543. DOI: 10.2147/RMHP.S271668 18. Soggiu-Duță CL, Suciu N. Resident physicians’ and Midwives’ Knowledge of Preeclampsia and Eclampsia Reflected in Their Practice at a Clinical Hospital in Southern Romania. Journal Medicine and Life (2019.0) 12 435-441. DOI: 10.25122/jml-2019-0130 19. Soggiu-Duță CL, Crauciuc DV, Crauciuc E, Dmour A. The impact of an intensive educational program regarding preeclampsia on health professional knowledge. Revista Chimia (2019.0) 70 2245-2251. DOI: 10.37358/RC.19.6.7315 20. Gholami K, Norouzkhani N, Kargar M, Ghasemirad H. Impact of Educational Interventions on Knowledge About Hypertensive Disorders of Pregnancy Among Pregnant Women: A Systematic Review 2022. Front. Cardiovasc. Med (2022.0) 9 886679. DOI: 10.3389/fcvm.2022.886679 21. Murray-Davis B, Berger H, Melamed N, Darling EK. A framework for understanding how midwives perceive and provide care management for pregnancies complicated by gestational diabetes or hypertensive disorders of pregnancy. Midwifery (2022.0) 103498. DOI: 10.1016/j.midw.2022.103498
--- title: Prevalence and risk factors of diabetic retinopathy in Basrah, Iraq authors: - Mohammed Al Ashoor - Ali Al Hamza - Ibrahim Zaboon - Ammar Almomin - Abbas Mansour journal: Journal of Medicine and Life year: 2023 pmcid: PMC10015581 doi: 10.25122/jml-2022-0170 license: CC BY 3.0 --- # Prevalence and risk factors of diabetic retinopathy in Basrah, Iraq ## Abstract This study aimed to measure the prevalence and risk factors of diabetic retinopathy (DR) among patients with diabetes mellitus aged 20 to 82 years attending the Faiha Diabetes, Endocrine, and Metabolism Center (FDEMC) in Basrah. A cross-sectional study was conducted at FDEMC, including 1542 participants aged 20 to 82 from January 2019 to December 2019. Both eyes were examined for evidence of DR by a mobile nonmydriatic camera, and statistical analysis was performed to measure the prevalence rates ($95\%$ CI) for patients with different characteristics. The mean age of participants was 35.9, with 689 males ($44.7\%$; $95\%$ CI: 42.2–$47.2\%$) and 853 females ($55.3\%$; $95\%$ CI: 52.8–$57.8\%$). The prevalence rate of DR was $30.5\%$ ($95\%$ CI: 28.1–$32.8\%$), and $11.27\%$ of cases were proliferative retinopathy. DR significantly increased with age (p-value=0.000), it was higher in females (p-value=0.005), and significantly increased with a longer duration of diabetes (p-value<0.001), hyperglycemia (p-value<0.001), hypertension (p-value=0.004), dyslipidemia (p-value<0.001), nephropathy (p-value<0.001) and smoking (p-value<0.001). There was no statistical association between DR and the type of diabetes or obesity. One-third of the participants in this study had DR. Screening and early detection of DR using a simple tool such as a digital camera should be a priority to improve a person’s health status. ## INTRODUCTION Uncontrolled diabetes mellitus (DM) can lead to a condition called diabetic retinopathy (DR), which affects the blood vessels in the retina of the eye. DR is a major clinical representation of uncontrolled diabetes mellitus and a leading cause of blindness among people with diabetes. The severity of DR is often influenced by the duration of the disease and the level of glucose control [1,2]. Unfortunately, the number of diabetes cases is projected to increase worldwide from 382 million in 2013 to 592 million by 2035 [3]. DR has been identified as the fifth most frequent cause of avoidable blindness and the fifth most frequent cause of moderate to severe visual impairment between 1990 and 2010 [4]. The relationship between hyperglycemia and microvascular complications is not fully understood [5,6]. However, cellular changes can lead to microvascular damage, increased capillary permeability, vascular occlusion, and weakening of supporting structures [1]. In addition, vascular endothelial growth factor (VEGF), which is the basis for treating disorders that could impair vision, may promote the formation of new blood vessels and contribute to vascular leakage [1]. The Early Treatment Diabetic Retinopathy Study (ETDRS – the modified Airlie House classification) classified diabetic retinal disease into two types: non-proliferative diabetic retinopathy (NPDR) and proliferative diabetic retinopathy (PDR) [6,7]. The following are known risk factors for diabetic retinopathy: Duration of diabetes mellitus: The Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR) found a direct association between the duration of diabetes and the prevalence of DR, which can reach up to $99\%$ and $60\%$ in type 1 and 2 diabetes, respectively, after 20 years. Proliferative diabetic retinopathy represents $50\%$ of type 1 diabetes cases after 20 years and $25\%$ of type 2 diabetes cases after 25 years [8-15].Glycemic control: Intensive and early glycemic control (HbA1c<$7\%$) decreases the risk of development or progression of DR in both type 1 DM [16] and type 2 DM [17].Hypertension: Tight control of blood pressure ($\frac{140}{80}$ mmHg) decreases the risk of progression of DR [17,18] and induces a $34\%$ risk reduction in microvascular changes [19]. For every 10mmHg reduction in systolic blood pressure, there was a $13\%$ reduction in microvascular endpoints[19].Diabetic kidney disease: The deterioration or treatment of kidney diseases is associated with the worsening or improvement of DR, respectively [20,21]. Most patients with renal disease, as evidenced by proteinuria and/or elevated serum creatinine, have some degree of retinal changes. On the other hand, $35\%$ of asymptomatic retinopathy patients have diabetic kidney disease[19].Dyslipidemia: Elevated serum lipids are strongly correlated with worsening DR [22,23].Obesity: Some studies have shown that obesity is associated with DR [24].Smoking: Smoking increases blood levels of carbon monoxide, platelet aggregation, and vasoconstriction, all of which can increase the risk of diabetic retinopathy [19,25]. However, there is some controversy about the association between smoking and the progression of DR [26, 27]. Patients with diabetes often experience poor visual acuity due to various retinal signs, which can be visualized and recorded using fundus photography - a baseline tool for diagnosing and monitoring retinal diseases. Recently, the introduction of mobile nonmydriatic fundus cameras has greatly improved the quality of DR screening and follow-up programs. This technology is part of telemedicine, which allows patients with diabetes to have their retinas examined at a location outside of a specialized eye examination unit, such as a diabetes center [28-31]. Fluorescein angiography, optical coherence tomography (OCT), and B-scan ultrasonography are other tools used to diagnose DR [6,31,32]. It is essential to differentiate hypertensive retinopathy and other diseases from DR [19]. Imperative management includes controlling associated risk factors and blood glucose to prevent the onset and progression of DR [1,33,34]. Medical intervention may include the use of fenofibrate [35] and intravitreal anti-vascular endothelial growth factors, such as ranibizumab, which is now widely used to treat macular edema [36,37]. Other management options include laser photocoagulation [6,32,38,39] and surgical treatment, such as pars plana vitrectomy [6,32,40]. This study aimed to assess the prevalence of diabetic retinopathy (DR) and its risk factors among patients with diabetes who were receiving care at the Faiha Diabetes, Endocrine, and Metabolism Center in Basrah, located in Southern Iraq. ## Study design and setting This cross-sectional study assessed DR prevalence and risk factors among diabetic patients attending the Faiha Diabetes, Endocrine, and Metabolism Center (FDEMC) in Basrah. The study was conducted at FDEMC from January 2019 to December 2019. A non-random sample was collected by a simple randomization technique, which consisted of 1542 diabetic patients aged 20 to 80 years. Well-trained medical staff recorded electronic data related to clinical and laboratory tests and demographic measures, followed by a fundoscopic examination of the retina using a nonmydriatic mobile camera. All data and retinal images were uploaded to the FDEMC intranet computers and analyzed by groups of endocrinologists and by a single ophthalmologist, respectively. ## Data collection and examinations For the study requirements, a structured questionnaire was formulated that comprised the following: Patient name and patient FDEMC ID number;Demographic information, including: Age in years was stratified into groups: 20 to <30, 30 to <40, 40 to <50, 50 to <60, 60 to <70, and ≥70;Gender. Clinical history, examinations, and laboratory assessments, including: Type of diabetes mellitus (type 1 or type 2);Duration of diabetes mellitus, which is the period between the age of diagnosis and the time of examination, was categorized into groups: <10 years, 10-30 years, and >30 years;Obesity, defined as BMI (body mass index) greater than or equal to 30 mg/m2, calculated as weight (kg) divided by square height (m2);Smoking history, classified as nonsmokers and smokers (current or past smokers were included in the smoking group);Blood pressure was measured by an automated office blood pressure (AOBP) machine in a quiet room and a seated position. Patients were considered hypertensive if the reading was greater than or equal to $\frac{140}{90}$ mm Hg or if they were already on anti-hypertension therapy; results below that level indicated no hypertension;Diabetes control, defined as HbA1c% equal to or less than $7\%$, measured by high-performance liquid chromatography (Varianttm Hemoglobin Testing System; Bio-Rad Laboratories Inc., Hercules, CA, USA);Serum lipid level, obtained from a fasting blood sample and tested for lipid profile by Integra laboratory diagnostics. Dyslipidemia was defined as an abnormal lipid profile when total cholesterol and triglyceride (TG) levels were equal to or above 200 mg/dl and 150 mg/dl, respectively, and high-density lipoprotein HDL levels were less than 45 mg/dl;Nephropathy, defined as the presence of microalbumin in the urine within the range of 30-300 mg/g of creatinine. The immunoturbidometric assay method is often used to quantify albumin concentrations in urine. Ophthalmological examinations: a detailed anterior segment of both eyes was examined, including visual acuity recording, intraocular pressure measurement, and lens opacity examination. A retinal examination was then performed using a nonmydriatic digital retinal camera (D-EYE Portable Retinal Imaging System, d-eye S.r. L.® Padova PD, Italy), which was easily attached to a smartphone (iPhone 6®), creating a handheld direct ophthalmoscope for vision care screening and evaluation with no mydriatic eye drops used. The camera used a magnetic fundus lens attached to an iPhone and utilized a user-friendly smartphone application and the built-in iPhone camera to take fundus photographs for each patient. Multiple images were taken for each eye centered on the fovea [450], and these images were graded into nonproliferative DR and a proliferative DR by an ophthalmologist according to Early Treatment Diabetic Retinopathy Study (ETDRS). The study excluded individuals under 20 years old, with gestational diabetes mellitus, and ocular media opacity (such as cataracts or corneal opacities) that could interfere with proper fundus examination, urinary tract infections, or abnormal thyroid function. ## Statistical analysis Statistical Package for Social Sciences (SPSS) version 20 was used to analyze qualitative and quantitative data. Age was treated as a quantitative variable. The prevalence rate of DR was calculated with a $95\%$ confidence interval (CI) and compared across different characteristics. A Chi-square test was performed to compare the variables with and without DR. Multiple regression analysis was also analyzed at $95\%$ CI to evaluate the relationship between risk factors and DR. P-values ≤0.05 were considered statistically significant. ## Prevalence of DR A total of 1542 patients with a mean age of 35.9 years (range 20-82 years) participated in this study. Of these, 689 were males ($44.7\%$; $95\%$ CI: $42.2\%$-$47.2\%$), and 853 were females ($55.3\%$; $95\%$ CI: $52.8\%$-$57.8\%$). 470 patients had DR, resulting in an overall prevalence rate of $30.5\%$ ($95\%$ CI: $28.1\%$ to $32.8\%$), with $11.27\%$ having proliferative changes. Table 1 summarizes the frequency and prevalence of DR with corresponding $95\%$ confidence intervals (CI). Figure 1 shows the prevalence of DR in those patients. Table 2 summarizes the frequency and prevalence of DR among different risk factors with corresponding p-values. Table 2 \| Variable \| Variable.1 \| Variable.2 \| Variable.3 \| With DR \| With DR.1 \| Without DR \| Without DR.1 \| Total \| Total.1 \| P-value \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Variable \| Variable \| Variable \| Variable \| Frequency \| Percent \| Frequency \| Percent \| Sum \| Percent \| P-value \| \| Age (years) \| 20–29 \| 20–29 \| 20–29 \| 15 \| 1.0% \| 92 \| 6.0% \| 107 \| 7.0% \| 0.000 \| \| Age (years) \| 30–39 \| 30–39 \| 30–39 \| 29 \| 1.9% \| 144 \| 9.3% \| 173 \| 11.2% \| 0.000 \| \| Age (years) \| 40–49 \| 40–49 \| 40–49 \| 94 \| 6.1% \| 305 \| 19.8% \| 399 \| 25.9% \| 0.000 \| \| Age (years) \| 50–59 \| 50–59 \| 50–59 \| 172 \| 11.2% \| 306 \| 19.8% \| 478 \| 31.0% \| 0.000 \| \| Age (years) \| 60–69 \| 60–69 \| 60–69 \| 133 \| 8.6% \| 202 \| 13.1% \| 335 \| 21.7% \| 0.000 \| \| Age (years) \| ≥70 \| ≥70 \| ≥70 \| 27 \| 1.7% \| 23 \| 1.5% \| 50 \| 3.2% \| 0.000 \| \| Gender \| Male \| Male \| Male \| 212 \| 13.8% \| 477 \| 30.9% \| 689 \| 44.7% \| 0.043 \| \| Gender \| Female \| Female \| Female \| 258 \| 16.7% \| 595 \| 38.6% \| 853 \| 55.3% \| 0.043 \| \| Duration \| <10 years \| Pre-PDR \| 153 \| 162 \| 10.5% \| 738 \| 47.9% \| 900 \| 58.4% \| 0.000 \| \| Duration \| <10 years \| PDR \| 9 \| 162 \| 10.5% \| 738 \| 47.9% \| 900 \| 58.4% \| 0.000 \| \| Duration \| 10–29 years \| Pre-PDR \| 237 \| 276 \| 17.9% \| 318 \| 20.6% \| 594 \| 38.5% \| 0.000 \| \| Duration \| 10–29 years \| PDR \| 39 \| 276 \| 17.9% \| 318 \| 20.6% \| 594 \| 38.5% \| 0.000 \| \| Duration \| ≥30 years \| Pre-PDR \| 27 \| 32 \| 2.1% \| 16 \| 1.0% \| 48 \| 3.1% \| 0.000 \| \| Duration \| ≥30 years \| PDR \| 5 \| 32 \| 2.1% \| 16 \| 1.0% \| 48 \| 3.1% \| 0.000 \| \| Glycemic control \| Uncontrolled \| Pre-PDR \| 365 \| 413 \| 26.8% \| 882 \| 57.2% \| 1295 \| 84.0% \| 0.001 \| \| Glycemic control \| Uncontrolled \| PDR \| 48 \| 413 \| 26.8% \| 882 \| 57.2% \| 1295 \| 84.0% \| 0.001 \| \| Glycemic control \| Controlled \| Pre-PDR \| 52 \| 57 \| 3.7% \| 190 \| 12.3% \| 247 \| 16.0% \| 0.001 \| \| Glycemic control \| Controlled \| PDR \| 5 \| 57 \| 3.7% \| 190 \| 12.3% \| 247 \| 16.0% \| 0.001 \| \| DM type \| Type 1 \| Pre-PDR \| 28 \| 37 \| 2.4% \| 151 \| 9.8% \| 188 \| 12.2% \| 0.000 \| \| DM type \| Type 1 \| PDR \| 9 \| 37 \| 2.4% \| 151 \| 9.8% \| 188 \| 12.2% \| 0.000 \| \| DM type \| Type 2 \| Pre-PDR \| 389 \| 433 \| 8.1% \| 921 \| 59.7% \| 1354 \| 87.8% \| 0.000 \| \| DM type \| Type 2 \| PDR \| 44 \| 433 \| 8.1% \| 921 \| 59.7% \| 1354 \| 87.8% \| 0.000 \| \| Blood pressure \| Hypertensive \| Pre-PDR \| 279 \| 312 \| 20.3% \| 523 \| 33.9% \| 835 \| 54.2% \| 0.000 \| \| Blood pressure \| Hypertensive \| PDR \| 33 \| 312 \| 20.3% \| 523 \| 33.9% \| 835 \| 54.2% \| 0.000 \| \| Blood pressure \| Non-hypertensive \| Pre-PDR \| 138 \| 158 \| 10.2% \| 549 \| 35.6% \| 707 \| 45.8% \| 0.000 \| \| Blood pressure \| Non-hypertensive \| PDR \| 20 \| 158 \| 10.2% \| 549 \| 35.6% \| 707 \| 45.8% \| 0.000 \| \| Renal disease \| Nephropathy \| Pre-PDR \| 122 \| 151 \| 9.8% \| 179 \| 11.6% \| 330 \| 21.4% \| 0.000 \| \| Renal disease \| Nephropathy \| PDR \| 29 \| 151 \| 9.8% \| 179 \| 11.6% \| 330 \| 21.4% \| 0.000 \| \| Renal disease \| Non-nephropathy \| Pre-PDR \| 295 \| 319 \| 20.7% \| 893 \| 57.9% \| 1212 \| 78.6% \| 0.000 \| \| Renal disease \| Non-nephropathy \| PDR \| 24 \| 319 \| 20.7% \| 893 \| 57.9% \| 1212 \| 78.6% \| 0.000 \| \| Lipid profile \| Dyslipidemia \| Pre-PDR \| 280 \| 316 \| 20.5% \| 532 \| 34.5% \| 848 \| 55.0% \| 0.000 \| \| Lipid profile \| Dyslipidemia \| PDR \| 36 \| 316 \| 20.5% \| 532 \| 34.5% \| 848 \| 55.0% \| 0.000 \| \| Lipid profile \| Normal \| Pre-PDR \| 137 \| 154 \| 10.0% \| 540 \| 35.0% \| 694 \| 45.0% \| 0.000 \| \| Lipid profile \| Normal \| PDR \| 17 \| 154 \| 10.0% \| 540 \| 35.0% \| 694 \| 45.0% \| 0.000 \| \| Obesity \| Obese \| Pre-PDR \| 219 \| 246 \| 16.0% \| 551 \| 35.7% \| 797 \| 51.7% \| 0.042 \| \| Obesity \| Obese \| PDR \| 27 \| 246 \| 16.0% \| 551 \| 35.7% \| 797 \| 51.7% \| 0.042 \| \| Obesity \| Non-obese \| Pre-PDR \| 198 \| 224 \| 14.5% \| 521 \| 33.8% \| 745 \| 48.3% \| 0.042 \| \| Obesity \| Non-obese \| PDR \| 26 \| 224 \| 14.5% \| 521 \| 33.8% \| 745 \| 48.3% \| 0.042 \| \| Smoking \| Smoker \| Pre-PDR \| 88 \| 115 \| 7.5% \| 227 \| 14.7% \| 342 \| 22.2% \| 0.019 \| \| Smoking \| Smoker \| PDR \| 27 \| 115 \| 7.5% \| 227 \| 14.7% \| 342 \| 22.2% \| 0.019 \| \| Smoking \| Non-smoker \| Pre-PDR \| 329 \| 355 \| 23.0% \| 845 \| 54.8% \| 1200 \| 77.8% \| 0.019 \| \| Smoking \| Non-smoker \| PDR \| 26 \| 355 \| 23.0% \| 845 \| 54.8% \| 1200 \| 77.8% \| 0.019 \| \| Total \| Sum/percent \| Sum/percent \| Sum/percent \| 470 \| 30.5% \| 1072 \| 69.5% \| 1542 \| 100% \| \| ## Age Most participants ($78.6\%$) were in the 40-69 age range. There was a higher prevalence of DR in the 50-59 age group ($11.2\%$) and a lower rate in the 20 to 29 age group ($1.0\%$). There was no significant difference in the prevalence of DR between the 30-39 age group ($1.9\%$) and the ≥70 age group ($1.7\%$). The prevalence rate of DR was significantly higher with increasing age (p-value < 0.000). ## Gender The frequency of females was higher than that of males, 853 ($55.3\%$) and 689 ($44.7\%$), respectively. The prevalence rate of DR was significantly higher in females ($16.7\%$) than in males ($13.8\%$) (p-value < 0.043). ## Diabetes duration 900 patients ($58.4\%$) had diabetes for 10 years or less, 594 patients ($38.5\%$) for 10 to 30 years, and 48 patients ($3.1\%$) for more than 30 years. Despite these findings, the prevalence of DR was higher in patients with a DM duration of 10-30 years ($17.9\%$), of whom PDR was found in 39 patients (p-value <0.000). ## Glycemic control Blood sugar was uncontrolled in a high percentage of patients ($84.0\%$), among whom DR was found in $26.8\%$, and some had PDR changes detected in 48 patients (p-value <0.001). ## Type of Diabetes Mellitus 1354 participants ($87.8\%$) had T2DM, and 188 had T1D ($12.2\%$). The prevalence rate of DR was significantly higher in T2DM ($28.1\%$) than in T1D ($2.4\%$), and 44 patients with T2DM had PDR changes (p-value< 0.000). ## Blood pressure Among the study participants, 835 ($54.2\%$) had hypertension. The prevalence of DR was higher among hypertensive patients ($20.3\%$) than diabetic nonhypertensive patients ($10.2\%$). Furthermore, 33 patients with DR had PDR changes, and this difference was statistically significant (p-value < 0.000). ## Nephropathy 1212 ($78.6\%$) participants had no nephropathy. DR changes were seen more often in patients without nephropathy, while PDR occurred in 29 nephropathic patients (p-value<0.000). ## Dyslipidemia Lipid tests were abnormal in 848 patients ($55.0\%$). DR occurred in $20.5\%$, compared to $10.0\%$ of reference ranges, and PDR occurred in 36 patients (p-value< 0.000). ## Obesity 797 ($51.7\%$) had obesity, slightly higher than the number of nonobese patients, 745 ($48.3\%$). The prevalence of DR was $16.0\%$ in obese patients, compared to $14.5\%$ in nonobese patients, with an approximately equal number of patients with PDR changes in both groups, 27 obese patients and 26 nonobese patients (p-value<0.042). ## Smoking There were 1200 ($77.8\%$) nonsmoker patients and 342 ($22.2\%$) smoker patients, and the prevalence of DR was higher in nonsmoker patients ($23.0\%$) than smoker patients ($7.5\%$), with an approximately equal number of patients with PDR changes in both groups, 27 smoker patients and 26 nonsmoker patients (p-value < 0.019). ## Multivariate analysis A standard multiple regression was conducted to evaluate the relationship between the identified risk factors and DR, revealing that R2 was 0.610 and the F factor was 239, indicating a significant relationship (p-value=.000). However, gender, obesity, and smoking were not significant predictors of DR (p-value =.412,.367 and.076, respectively). When analyzing the correlation coefficient, smoking was strongly associated with DR and reduced the association with the type of diabetes (p-value.500), as shown in Table 3 and Figure 2. ## DISCUSSION The increasing incidence of diabetes worldwide is a significant global healthcare concern, with an estimated 415 million people (age 20-79 years) currently living with diabetes, of whom approximately $50\%$ remain undiagnosed. Diabetic retinopathy is a common complication of untreated diabetes that can progress to visual impairment and blindness [41]. Generally, the probability of blindness in a person with diabetes is 25 times greater than that in the general population; hence early detection of DR is the first step in preventing vision loss. The American Academy of Ophthalmology and the American Diabetes Association suggest that annual ophthalmic examinations should start from the day of a diabetes diagnosis. However, in the past, the lack of accessible and efficient tools to detect early retinal changes led to delayed DR diagnosis and increased healthcare burden [42]. This study detected DR in $30.5\%$ ($95\%$ CI: $28.1\%$ to $32.8\%$) of patients aged 20 years and above attending FDEMC in Basrah (southern Iraq). Of these patients, $11.27\%$ had proliferative changes. The prevalence rate of DR in other governorates of Iraq was consistent with our study. In two studies conducted in Baghdad, the prevalence rates of DR were $30.2\%$ [43] and $33.1\%$ [44]. One study aimed to assess the prevalence rate and risk factors of visual acuity, retinopathy, cataracts, and glaucoma among a large sample size of diabetic patients aged 20-65 years, which differs from the objective of our study. In the other study, the sample size was small, and they assessed the importance of insulin therapy with the other risk factors. A study conducted in Mosul on proliferative and nonproliferative DR of T1D and T1DM had a small sample size and did not include the risk factors mentioned in our study. However, the prevalence rate was $32.35\%$[45]. Other studies in the region around Iraq reported variable rates of DR, including Jordan ($34.1\%$)[46], Turkey ($42.8\%$)[47], Saudi Arabia ($36.4\%$)[48], Iran ($41.9\%$)[49], Oman ($42.4\%$)[21], United Arab Emirates ($19\%$)[20], Qatar ($23.5\%$)[50], Egypt ($20.5\%$)[51] and Lebanon ($35\%$)[52]. The global prevalence rates of DR and proliferative DR among patients with diabetes are $35.4\%$ and $7.5\%$, respectively [4], confirmed in a pooled meta-analysis of 35 studies from 1980-2008 and consistent with our current study. Internationally, the prevalence rate of DR was studied in different countries, including European countries, with diverse rates ranging from $4\%$ in Finland to $52\%$ in the UK, with an average of nearly $40\%$[53]. Variations due to race- and ethnicity-related differences in the prevalence of DR have been considered an important public health issue. Similar variations have also been reported in the USA, ranging from $3.1\%$ to $48\%$ among ethnic groups [53]. Other countries reported DR prevalence rates as follows: the Russian Federation ($34.2\%$)[53], China ($27.9\%$)[54], Korea ($15.8\%$)[55], Singapore ($28.2\%$)[56], and Australia ($35.5\%$)[53]. In our study, there was a relationship between DR and age group, with higher prevalence occurring in the 50-59 age group ($11.2\%$). These findings are consistent with previous studies [20, 44-46, 48, 51, 57-60]. The association of sex with the development of DR is controversial. Various studies have reported that men are at higher risk than women [20, 58, 61], whereas other studies found no gender differences [46, 48, 55, 59, 62] or a higher risk for women[45, 51]. In our study, women had a greater risk of DR development than men ($16.7\%$ and $13.8\%$, respectively). The duration of diabetes in our study was strongly associated with DR, especially at 10-29 years ($17.9\%$), and DR in patients with a duration greater than 30 years was less common ($2.1\%$), which was due to a lower number of participating patients in those groups ($3.1\%$). These results were consistent with other studies [20, 21, 43-46, 48, 51, 52, 54, 55, 57, 58, 61,63,64]. However, in other studies, the opposite was identified [59]. In most previous studies, hyperglycemia (measured by HbA1c) was considered a significant risk factor for the development of DR [43-46,48, 50, 54, 55, 61, 64], which is consistent with the results of our study. However, other studies did not report these findings [51, 59, 63]. Type 1 diabetes usually has a higher risk of retinopathy due to the longer duration of the disease [20, 51, 64]. Nevertheless, we did not see these results in our study, where T2DM was more commonly associated with DR than T1D in univariate analysis, possibly due to the high participation rate of patients 20 years old and above. Other studies have reported equal rates of DR in both types of diabetes [45], which was shown by multivariate analysis. Hypertensive patients with diabetes were considered a distinctive risk factor for DR in our study, and those results were consistent with other studies [20, 24, 43-45, 48, 51, 52, 54, 55, 57, 61,63,64]. Nephropathy was not a risk factor for DR in our study in univariate analysis. However, there was a significant relationship between nephropathy and DR in the multivariate analysis, and these results were consistent with other studies [20, 48, 52, 54, 63]. There was a strong association between dyslipidemia in diabetic patients and the development of DR in our study, which was proven in several studies [22-24, 44, 54, 57, 59, 63]. These results were not consistent with other studies [65]. Our study demonstrated that obesity was not a risk factor for DR, consistent with other studies [55, 57] but not reported in others [24, 44, 59]. The relationship between smoking and DR development is controversial, as the relationship in our study was negative, which is consistent with other findings [26,27,46,57]. However, other studies have reported the opposite. When analyzed by multiple regression tests, smoking had a strong association with DR, in line with other studies [19, 25, 43]. The risk factors under study play a role in the development of DR apart from obesity and type of diabetes. At first, smoking had a weak association with DR, but regression analysis [26] removed the factors that caused the relationship to be underestimated. It is evident from Table 3 that these risk factors explain $61\%$ of the development of DR, leaving approximately $40\%$ affected by other factors not included in our study, and hyperglycemia is part of the disease process. To date, no published Basrah hospital-based studies have addressed retinopathy in diabetic patients despite its negative effect on the quality of life. The retinopathy screening program has not been fully established in our region. For this reason, the use of nonmydriatic digital cameras, as in other countries [29, 30, 66-70], may reduce the time and effort in the early detection of retinopathy. The easy-to-use nonmydriatic digital camera allows physicians and ophthalmologists to screen for retinopathy efficiently and intervene at an early stage, potentially preventing progression to more advanced disease, which is what we tried to prove in this study. This study holds significant importance as it utilized a large sample size to determine the prevalence of diabetic retinopathy (DR) while prospectively evaluating diabetic patients. This approach enables early referral to a retinal disease specialist, highlighting the importance of timely detection and management of DR. Additionally, this study addressed a group of risk factors associated with DR using modern screening methods. Furthermore, all data for this study have been uploaded via an internet network with easy access by endocrinology specialists for review and analysis. In addition, we assessed retinal changes using a nonmydriatic digital camera, which provided us with 450 fundus images that were ready to be analyzed by an ophthalmologist. There are several limitations to this study that must be considered. First, the study was conducted in a single center, and as a result, the findings cannot be generalized to the broader population. Additionally, the retinal images were interpreted by a single ophthalmologist, which increases the risk of bias. Second, this is a cross-sectional study in which DR prevalence data and risk factor information were derived concurrently rather than a case-control or cohort study. Therefore, the results must be interpreted with caution. Smoking behavior was derived from questionnaires rather than objective measurements; thus, more than three-quarters of diabetic patients in our study were nonsmokers, which affects the estimation of the prevalence rate of DR and the causal association. Finally, the use of a digital camera for retinal changes improves sensitivity and specificity compared to standard fundoscopy [69], but it has a limited ability to detect peripheral lesions. ## CONCLUSION DR was observed in one-third of patients, with one-tenth of those cases being proliferative. Our findings indicate that increasing age, female gender, longer diabetes duration, hyperglycemia, hypertension, dyslipidemia, nephropathy, and smoking had a direct correlation with DR. On the other hand, other risk factors, such as type of diabetes and obesity, had no proven effect. The use of nonmydriatic digital cameras has demonstrated the potential to overcome obstacles in the early diagnosis of DR. Nevertheless, future studies involving larger sample sizes, including the Basrah population, are needed to confirm these findings. ## Conflict of interest The authors declare no conflict of interest. ## Ethical approval This study was approved by the ethics committee of the FDEMC (#51, 22.12.2018). ## Consent to participate Written informed consent was obtained from all participants. ## Authorship All authors contributed equally to this research, including collecting and analyzing data, interpreting results, and drawing conclusions. MAA was the corresponding author and responsible for taking and analyzing retinal pictures, and also contributed to the writing, interpretation of results, discussion, and reference listing. AAH interpreted medical and laboratory examinations, performed most statistical analyses, and participated in writing and modifying the discussion. IZ contributed to writing and grammatical correction and interpreted medical and laboratory examinations. AA performed statistical analyses, interpreted results, and provided medical and laboratory examination interpretations. AM supervised this study, interpreted medical and laboratory examinations, and contributed to the writing and frequent revisions of the entire article. ## References 1. Hendrick AM, Gibson MV, Kulshreshtha A. Diabetic Retinopathy. Prim Care (2015) 42 451-64. DOI: 10.1016/j.pop.2015.05.005 2. Hartnett ME, Baehr W, Le YZ. Diabetic retinopathy, an overview. Vision Res (2017) 139 1-6. DOI: 10.1016/j.visres.2017.07.006 3. Kyari F, Tafida A, Sivasubramaniam S, Murthy GV. Prevalence and risk factors for diabetes and diabetic retinopathy: results from the Nigeria national blindness and visual impairment survey. BMC Public Health (2014) 14 1299. DOI: 10.1186/1471-2458-14-1299 4. Lee R, Wong TY, Sabanayagam C. Epidemiology of diabetic retinopathy, diabetic macular edema and related vision loss. Eye Vis (Lond) (2015) 2 17. DOI: 10.1186/s40662-015-0026-2 5. Fowler MJ. Microvascular and Macrovascular Complications of diabetes. Clinical Diabetes (2011) 29 116-122. DOI: 10.10.2337/diaclin.29.3.116 6. Kanski JJ, Kanski BB. Kanski’s Clinical Ophthalmology: a systematic approach. Clinical Ophthalmology: a systematic approach (2016) 881 7. Wu L, Fernandez-Loaiza P, Sauma J, Hernandez-Bogantes E, Masis M. Classification of diabetic retinopathy and diabetic macular edema. World Journal of Diabetes (2013) 4 290-4. DOI: 10.4239/wjd.v4.i6.290 8. Klein R, Klein BE, Moss SE, Davis MD, DeMets DL. The Wisconsin epidemiologic study of diabetic retinopathy. II Prevalence and risk of diabetic retinopathy when age at diagnosis is less than 30 years. Arch Ophthalmol (1984) 102 520-6. DOI: 10.1001/archopht.1984.01040030398010 9. Klein R, Klein BE, Moss SE, Davis MD, DeMets DL. The Wisconsin epidemiologic study of diabetic retinopathy. III Prevalence and risk of diabetic retinopathy when age at diagnosis is 30 or more years. Arch Ophthalmol (1984) 102 527-32. DOI: 10.1001/archopht.1984.01040030405011 10. Klein R, Klein BE, Moss SE, Davis MD, DeMets DL. The Wisconsin Epidemiologic Study of Diabetic Retinopathy. X Four-year incidence and progression of diabetic retinopathy when age at diagnosis is 30 years or more. Arch Ophthalmol (1989) 107 244-9. DOI: 10.1001/archopht.1989.01070010250031 11. Klein R, Klein BE, Moss SE. The Wisconsin epidemiological study of diabetic retinopathy: a review. Diabetes/metabolism reviews (1989) 5 559-70. DOI: 10.1002/dmr.5610050703 12. Klein R, Klein BE, Moss SE. The Wisconsin epidemiologic study of diabetic retinopathy: an update. Australian and New Zealand Journal of Ophthalmology (1990) 18 19-22. DOI: 10.1111/j.1442-9071.1990.tb00579.x 13. Klein R, Klein BE, Moss SE, Cruickshanks KJ. The Wisconsin Epidemiologic Study of diabetic retinopathy. XIV. Ten-year incidence and progression of diabetic retinopathy. Arch Ophthalmol (1994) 112 1217-28. DOI: 10.1001/archopht.1994.01090210105023 14. Klein R, Klein BE, Moss SE, Cruickshanks KJ. The Wisconsin Epidemiologic Study of Diabetic Retinopathy: XVII. The 14-year incidence and progression of diabetic retinopathy and associated risk factors in type 1 diabetes. Ophthalmology (1998) 105 1801-15. DOI: 10.1016/S0161-6420(98)91020-X 15. Klein R, Knudtson MD, Lee KE, Gangnon R, Klein BE. The Wisconsin Epidemiologic Study of Diabetic Retinopathy: XXII the twenty-five-year progression of retinopathy in persons with type 1 diabetes. Ophthalmology (2008) 115 1859-68. DOI: 10.1016/j.ophtha.2008.08.023 16. Nathan DM. The Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications Study at 30 Years: Overview. Diabetes Care (2014) 37 9-16. DOI: 10.2337/dc13-2112 17. King P, Peacock I, Donnelly R. The UK Prospective Diabetes Study (UKPDS): clinical and therapeutic implications for type 2 diabetes. British Journal of Clinical Pharmacology (1999) 48 643-8. DOI: 10.1046/j.1365-2125.1999.00092.x 18. Scanlon PH. Diabetic retinopathy. Medicine (2010) 38 656-60 19. William Tasman EAJ, Mitchell PR, Parks MM. Duane’s Ophthalmology Revised Edition 2013. (2012) 19878 20. Al-Maskari F, El-Sadig M. Prevalence of diabetic retinopathy in the United Arab Emirates: a cross-sectional survey. BMC Ophthalmol (2007) 7 11. DOI: 10.1186/1471-2415-7-11 21. El Haddad OA, Saad MK. Prevalence and risk factors for diabetic retinopathy among Omani diabetics. Br J Ophthalmol (1998) 82 901-6. DOI: 10.1136/bjo.82.8.901 22. Agroiya P, Philip R, Saran S, Gutch M. Association of serum lipids with diabetic retinopathy in type 2 diabetes. Indian Journal of Endocrinology and Metabolism (2013) 17 S335-7. DOI: 10.4103/2230-8210.119637 23. Chew EY, Klein ML, Ferris FL, Remaley NA. Association of elevated serum lipid levels with retinal hard exudate in diabetic retinopathy. Early Treatment Diabetic Retinopathy Study (ETDRS) Report 22. Arch Ophthalmol (1996) 114 1079-84. DOI: 10.1001/archopht.1996.01100140281004 24. van Leiden HA, Dekker JM, Moll AC, Nijpels G. Blood pressure, lipids, and obesity are associated with retinopathy: the hoorn study. Diabetes Care (2002) 25 1320-5. DOI: 10.2337/diacare.25.8.1320 25. Karamanos B, Porta M, Songini M, Metelko Z. Different risk factors of microangiopathy in patients with type I diabetes mellitus of short versus long duration. The EURODIAB IDDM Complications Study. Diabetologia (2000) 43 348-55. DOI: 10.1007/s001250050053 26. Moss SE, Klein R, Klein BE. Cigarette smoking and ten-year progression of diabetic retinopathy. Ophthalmology (1996) 103 1438-42. DOI: 10.1016/s0161-6420(96)30486-7 27. Muhlhauser I, Bender R, Bott U, Jorgens V. Cigarette smoking and progression of retinopathy and nephropathy in type 1 diabetes. Diabet Med (1996) 13 536-43. DOI: 10.1002/(SICI)1096-9136(199606)13:6<536::AID-DIA110>3.0.CO;2-J 28. Maberley D, Walker H, Koushik A, Cruess A. Screening for diabetic retinopathy in James Bay, Ontario: a cost-effectiveness analysis. CMAJ (2003) 168 160-4. PMID: 12538543 29. Soto-Pedre E, Hernaez-Ortega MC. Screening coverage for diabetic retinopathy using a three-field digital non-mydriatic fundus camera. Prim Care Diabetes (2008) 2 141-6. DOI: 10.1016/j.pcd.2008.04.003 30. Beynat J, Charles A, Astruc K, Metral P. Screening for diabetic retinopathy in a rural French population with a mobile non-mydriatic camera. Diabetes Metab (2009) 35 49-56. DOI: 10.1016/j.diabet.2008.07.002 31. Salz DA, Witkin AJ. Imaging in Diabetic Retinopathy. Middle East African Journal of Ophthalmology (2015) 22 145-50. DOI: 10.4103/0974-9233.151887 32. Basic and Clinical Science Course 2016-2017. (2016) 33. Corcóstegui B, Durán S, González-Albarrán MO, Hernández C. Update on Diagnosis and Treatment of Diabetic Retinopathy: A Consensus Guideline of the Working Group of Ocular Health (Spanish Society of Diabetes and Spanish Vitreous and Retina Society). J Ophthalmol (2017) 2017 8234186. DOI: 10.1155/2017/8234186 34. Duh EJ, Sun JK, Stitt AW. Diabetic retinopathy: current understanding, mechanisms, and treatment strategies. JCI Insight (2017) 2 e93751. DOI: 10.1172/jci.insight.93751 35. Knickelbein JE, Abbott AB, Chew EY. Fenofibrate and Diabetic Retinopathy. Curr Diab Rep (2016) 16 90. DOI: 10.1007/s11892-016-0786-7 36. Diabetic Retinopathy Clinical Research Network; Writing Committee; Aiello LP, Beck RW, Bressler NM, Browning DJ. Rationale for the diabetic retinopathy clinical research network treatment protocol for center-involved diabetic macular edema. Ophthalmology (2011) 118 e5-14. DOI: 10.1016/j.ophtha.2011.09.058 37. Nguyen QD, Brown DM, Marcus DM, Boyer DS. Ranibizumab for diabetic macular edema: results from 2 phase III randomized trials: RISE and RIDE. Ophthalmology (2012) 119 789-801. DOI: 10.1016/j.ophtha.2011.12.039 38. Clinical application of Diabetic Retinopathy Study (DRS) findings, DRS Report Number 8. The Diabetic Retinopathy Study Research Group. Ophthalmology (1981) 88 583-600. PMID: 7196564 39. Royle P, Mistry H, Auguste P, Shyangdan D. Pan-retinal photocoagulation and other forms of laser treatment and drug therapies for non-proliferative diabetic retinopathy: systematic review and economic evaluation. Health Technol Assess (2015) 19 1-247. DOI: 10.3310/hta19510 40. Charles S. Diabetic retinopathy vitrectomy study. Arch Ophthalmol (1986) 104 486-8. DOI: 10.1001/archopht.1986.01050160038003 41. Cavan D, Makaroff L, da Rocha Fernandes J, Sylvanowicz M. The Diabetic Retinopathy Barometer Study: Global perspectives on access to and experiences of diabetic retinopathy screening and treatment. Diabetes Res Clin Pract (2017) 129 16-24. DOI: 10.1016/j.diabres.2017.03.023 42. Tavakoli M, Shahri RP, Pourreza H, Mehdizadeh A. A complementary method for automated detection of microaneurysms in fluorescein angiography fundus images to assess diabetic retinopathy. Pattern Recognition (2013) 46 2740-53. DOI: 10.48550/arXiv.1909.01557 43. Abed BK, Rahim YA, Ali MAH, Abbas MF. Prevalence and Risk Factors for Eye Problems among 20-65 Years Old Iraqi Diabetics Patients. J Fac Med Baghdad (2008) 50 166-74. DOI: 10.32007/jfacmedbagdad.5021273 44. Tawfeeq AS. Prevalence and risk factors of diabetic retinopathy among Iraqi patients with type 2 diabetes mellitus. Iraqi J Med Jan- (2015) 17-21 45. Rajab AY. Frequency of diabetic retinopathy in Mosul. Annals of the College of Medicine (2008) 34 129-34. DOI: 10.33899/mmed.2008.8884 46. Al-Amer RM, Khader Y, Malas S, Abu-Yaghi N. Prevalence and risk factors of diabetic retinopathy among Jordanian patients with type 2 diabetes. Digit J Ophthalmol (2008) 14 42-9. DOI: 10.5693/djo.01.2008.013 47. Sehnaz KZ, Temel YM. Duration of diabetes and prevalence of diabetic retinopathy: Istanbul Diabetic Retinopathy Study-IDRS results. (2007) 43-8 48. Ahmed RA, Khalil SN, Al-Qahtani MA. Diabetic retinopathy and the associated risk factors in diabetes type 2 patients in Abha, Saudi Arabia. J Family Community Med (2016) 23 18-24. DOI: 10.4103/2230-8229.172225 49. Maroufizadeh S, Almasi-Hashiani A, Hosseini M, Sepidarkish M, Omani Samani R. Prevalence of diabetic retinopathy in Iran: a systematic review and Meta-analysis. Int J Ophthalmol (2017) 10 782-9. DOI: 10.18240/ijo.2017.05.21 50. Elshafei M, Gamra H, Khandekar R, Al Hashimi M. Prevalence and determinants of diabetic retinopathy among persons ≥ 40 years of age with diabetes in Qatar: a community-based survey. Eur J Ophthalmol (2011) 21 39-47. DOI: 10.5301/ejo.2010.2699 51. Macky TA, Khater N, Al-Zamil MA, El Fishawy H, Soliman MM. Epidemiology of diabetic retinopathy in Egypt: a hospital-based study. Ophthalmic Res (2011) 45 73-8. DOI: 10.1159/000314876 52. Salti HI, Nasrallah MP, Taleb NM, Merheb M. Prevalence and determinants of retinopathy in a cohort of Lebanese type II diabetic patients. Can J Ophthalmol (2009) 44 308-13. DOI: 10.3129/i09-029 53. Sivaprasad S, Gupta B, Crosby-Nwaobi R, Evans J. Prevalence of diabetic retinopathy in various ethnic groups: a worldwide perspective. Surv Ophthalmol (2012) 57 347-70. DOI: 10.1016/j.survophthal.2012.01.004 54. Zhang G, Chen H, Chen W, Zhang M. Prevalence and risk factors for diabetic retinopathy in China: a multi-hospital-based cross-sectional study. Br J Ophthalmol (2017) 101 1591-5. DOI: 10.1136/bjophthalmol-2017-310316 55. Jee D, Lee WK, Kang S. Prevalence and risk factors for diabetic retinopathy: the Korea National Health and Nutrition Examination Survey 2008-2011. Invest Ophthalmol Vis Sci (2013) 54 6827-33. DOI: 10.1167/iovs.13-12654 56. Tan GS, Gan A, Sabanayagam C, Tham YC. Ethnic Differences in the Prevalence and Risk Factors of Diabetic Retinopathy: The Singapore Epidemiology of Eye Diseases Study. Ophthalmology (2018) 125 529-36. DOI: 10.1016/j.ophtha.2017.10.026 57. Hu Y, Teng W, Liu L, Chen K. Prevalence and risk factors of diabetes and diabetic retinopathy in Liaoning province, China: a population-based cross-sectional study. PLoS One (2015) 10 e0121477. DOI: 10.1371/journal.pone.0121477 58. Taş A, Bayraktar MZ, Erdem Ü, Sobaci G, Ucar M. Prevalence and risk factors for diabetic retinopathy in Turkey. Gulhane Medical Journal (2005) 47 164-74 59. Flores-Mena K, Jara-Tamayo K, Herrera-González P, Gea-Izquierdo E. Prevalence and major risk factors of diabetic retinopathy: A cross-sectional study in Ecuador. Bionatura (2017) 2 427-31. DOI: 10.21931/RB/2017.02.04.3 60. Kizor-Akaraiwe NN, Ezegwui IR, Oguego N, Uche NJ. Prevalence, Awareness and Determinants of Diabetic Retinopathy in a Screening Centre in Nigeria. Journal of Community Health (2016) 41 767-71. DOI: 10.1007/s10900-016-0151-4 61. Zhang X, Saaddine JB, Chou CF, Cotch MF. Prevalence of diabetic retinopathy in the United States 2005-2008. JAMA (2010) 304 649-56. DOI: 10.1001/jama.2010.1111 62. Al-Zuabi H, Al-Tammar Y, Al-Moataz R, Al-Sabti K. Retinopathy in newly diagnosed type 2 diabetes mellitus. Medical principles and practice: international journal of the Kuwait University. Health Science Centre (2005) 14 293-6. DOI: 10.1159/000086925 63. Haddad OA, Saad MK. Prevalence and risk factors for diabetic retinopathy among Omani diabetics. British Journal of Ophthalmology (1998) 82 901-6. DOI: 10.1136/bjo.82.8.901 64. Joanne WY, Yau M, Sophie L, Rogers M. Global Prevalence and Major Risk Factors of Diabetic Retinopathy. Epidemiology/Health Services R esearch (2012) 35 556-64. DOI: 10.2337/dc11-1909 65. Heydari B, Yaghoubi G, Yaghoubi MA, Miri MR. Prevalence and risk factors for diabetic retinopathy: an Iranian eye study. Eur J Ophthalmol (2012) 22 393-7. DOI: 10.5301/ejo.5000044 66. Andonegui J, Zurutuza A, de Arcelus MP, Serrano L. Diabetic retinopathy screening with non-mydriatic retinography by general practitioners: 2-year results. Prim Care Diabetes (2012) 6 201-5. DOI: 10.1016/j.pcd.2012.01.001 67. Qu M, Ni C, Chen M, Zheng L. Automatic diabetic retinopathy diagnosis using adjustable ophthalmoscope and multi-scale line operator. Pervasive and Mobile Computing (2017) 41 490-503. DOI: 10.1080/02713683.2020.1764975 68. Gargeya R, Leng T. Automated Identification of Diabetic Retinopathy Using Deep Learning. Ophthalmology (2017) 124 962-9. DOI: 10.1016/j.ophtha.2017.02.008 69. Martinez J, Hernandez-Bogantes E, Wu L. Diabetic retinopathy screening using single-field digital fundus photography at a district level in Costa Rica: a pilot study. Int Ophthalmol (2011) 31 83-8. DOI: 10.1007/s10792-010-9413-9 70. Zaki WMDW, Zulkifley MA, Hussain A, Halim WHWA. Diabetic retinopathy assessment: Towards an automated system. Biomedical Signal Processing and Control (2016) 24 72-82. DOI: 10.1016/j.bspc.2015.09.011
--- title: Low Rates of Psychosocial Screening and Lifestyle Counseling in Hidradenitis Suppurativa Patients in the USA authors: - Terri Shih - Devea R. De - Jonathan Rick - Vivian Y. Shi - Jennifer L. Hsiao journal: Skin Appendage Disorders year: 2023 pmcid: PMC10015644 doi: 10.1159/000528253 license: CC BY 4.0 --- # Low Rates of Psychosocial Screening and Lifestyle Counseling in Hidradenitis Suppurativa Patients in the USA ## Abstract ### Introduction Although hidradenitis suppurativa (HS) is associated with psychosocial comorbidities such as depression as well as modifiable comorbidities such as obesity, rates of psychosocial screening and lifestyle counseling in the USA have not been characterized. ### Methods This cross-sectional study utilized publicly available data from the National Ambulatory Medical Care Survey (NAMCS) between 2008 and 2018 to identify visits with a diagnosis of HS (ICD-9 code 705.83, ICD-10 code L73.2). T tests and multivariate logistic regressions analyzed trends in rates of screening and counseling while controlling for race, sex, and age. Survey weights are applied to each visit to represent a national sample. ### Results Depression screening was completed in only $2\%$ of reported visits. No visits reported screening for alcohol misuse, substance abuse, or domestic violence. There were low rates of counseling for weight reduction ($7.8\%$), diet and nutrition ($3.3\%$), exercise ($2.4\%$), smoking ($1.0\%$), and substance abuse ($0.7\%$). Black patients and individuals with public health insurance received less screening and counseling overall. ### Conclusion Rates of psychosocial screening and counseling on lifestyle modifications are low in ambulatory clinic visits for HS patients, and there are disparities based on race and insurance status. Implementing strategies to incorporate routine psychosocial screening and lifestyle counseling into visits may improve HS patient outcomes. ## Introduction Hidradenitis suppurativa (HS) is a chronic, debilitating inflammatory skin condition characterized by painful nodules, abscesses, sinus tracts, and scarring that imparts significant physical and psychosocial burdens. Associated comorbidities include smoking, obesity, metabolic syndrome, cardiovascular disease, depression, anxiety, and substance use disorder, among others [1]. Risk of intimate partner violence has been reported to be 2.4 times more likely in individuals with HS as compared to those with acne [2]. Screening for psychosocial conditions such as depression and domestic abuse and counseling for lifestyle modifications such as diet and exercise are important components of a comprehensive care strategy for HS. However, few studies have characterized how frequently this screening or counseling occurs for patients with HS. Herein, we examine characteristics of HS ambulatory visits and the rates of psychosocial screening and counseling in patients with HS in the USA. ## Methods The National Ambulatory Medical Care Survey (NAMCS) is conducted annually by the National Center for Health Statistics from the Centers for Disease Control and Prevention, which utilizes a stratified, random sample of patient visits to nonfederal, ambulatory office-based physicians. Physicians are randomly assigned a 1-week reporting period. A random sample of visits is assessed for data on patient demographics and symptoms and physician diagnoses and management, including screening and counseling, medications prescribed, and procedures completed. Survey weights are applied to each visit to represent a national sample. In this study, we searched publicly available NAMCS data between 2008 and 2018 (2017 was unavailable) for visits with a diagnosis of HS (ICD-9 code 705.83, ICD-10 code L73.2). Descriptive statistics were completed for demographic data and rates of psychosocial screening and lifestyle modification counseling. T tests and multivariate logistic regressions analyzed trends in rates of screening and counseling while controlling for race, sex, and age. Multivariate race comparisons excluded the category of race reported as “other” due to small sample size. Visits with missing data in relevant analyses were excluded. All data analyses were performed using SAS Studio 9.04.01 (SAS Institute, Cary, NC, USA). Variance in the complex survey design is accounted for by utilizing survey weights to create national estimates and confidence intervals (CI). ## Results From the 2008–2018 NAMCS datasets, an estimated 2.33 million visits ($95\%$ CI, 1.95 million–2.71 million) had a diagnosis of HS. Of these, $71.1\%$ of the patients were female, $75.6\%$ were white, and the mean age was 37.9 ± 1.0 (range 12–69) (Table 1). Depression screening was completed in a small minority ($2.0\%$) of visits, none of which were completed in black patients (Table 2). Depression screenings were slightly less likely to be conducted in older patients (OR, 0.94 [$95\%$ CI, 0.91–0.98], $$p \leq 0.003$$). No visits reported screening for alcohol misuse, substance abuse, or domestic violence. Physicians reported overall low rates of counseling for weight reduction ($7.8\%$), diet and nutrition ($3.3\%$), exercise ($2.4\%$), smoking ($1.0\%$), and substance abuse ($0.7\%$) (Table 2). Black patients were more likely to be counseled on weight reduction (OR, 4.95 [$95\%$ CI, 2.02–12.13], $$p \leq 0.003$$) but less likely to receive diet and nutrition counseling (OR, 0.52 [$95\%$ CI, 0.32–0.85], $$p \leq 0.01$$). Of visits that reported counseling on exercise, substance abuse, and tobacco use, none were completed in black patients. Older patients were slightly less likely to receive counseling on diet/nutrition (OR, 0.98 [$95\%$ CI, 0.96–1.00], $$p \leq 0.01$$), exercise (OR, 0.94 [$95\%$ CI, 0.91–0.97], $$p \leq 0.001$$), and weight reduction (OR, 0.90 [$95\%$ CI, 0.87–0.94], $p \leq 0.001$). There was no statistically significant difference in rates of counseling between men and women. Patients with higher BMI were more likely to receive counseling on exercise (OR, 1.24 [$95\%$ CI, 1.09–1.41], $$p \leq 0.002$$), weight reduction (OR, 1.09 [$95\%$ CI, 1.06–1.12], $p \leq 0.001$), and diet/nutrition (OR, 1.07 [$95\%$ CI, 1.05–1.08], $p \leq 0.001$) after controlling for age, sex, and race. Visits funded by public insurance including Medicare and Medicaid less frequently received counseling overall. They were significantly less likely to receive counseling on weight reduction (OR, 0.08 [$95\%$ CI, 0.01–0.79], $$p \leq 0.03$$). ## Discussion Rates of psychosocial screening and lifestyle counseling at ambulatory visits were low among patients with HS. Overall, individuals who were black or had public health insurance received less depression screening and lifestyle counseling. Given rates of depression in HS patients have been found to be as high as $26\%$ and there is an increased risk of suicide in HS patients [3], routine depression screening is warranted [1]. However, the rate of depression screening was found to be only $2\%$ in ambulatory clinic visits for HS patients in our study. The rate of substance use disorder in the USA has been found to be $4\%$ in HS patients versus $2\%$ in control patients [4]. One cross-sectional study interviewed 243 Canadian patients (128 with HS, 115 with acne) and found 2.4 times of risk of intimate partner violence compared to patients with acne [2]. However, none of the visits across the 10-year span of our study reported screenings on substance use and domestic violence, highlighting a potential practice gap. HS is associated with smoking, obesity, and poor cardiovascular outcomes [1, 5]. Though more data are needed, studies have suggested a correlation between smoking status and HS severity and duration [6, 7], and weight reduction has been linked to HS disease improvement [8]. Regardless of impact on HS disease activity, counseling on lifestyle modifications for diet, exercise, and smoking cessation should be performed for the overall health of HS patients. Of note, addressing lifestyle changes after the first establishing rapport with patients is helpful [9]. Racial and socioeconomic disparities were observed in the rates of depression screening and lifestyle counseling in patients with HS. Black patients and individuals with public health insurance received less screening and lifestyle counseling overall. It is imperative that depression screening and lifestyle counseling increase for all patients with HS, with particular attention paid to underserved populations. This is especially noteworthy as black patients and patients with low socioeconomic status are disproportionately affected with HS [10, 11]. Limitations of the NAMCS database include lack of data on HS severity. Given HS is associated with delayed and missed diagnoses [12], the number of HS ambulatory visits may be underrepresented. The NAMCS database may not capture all performed screenings for depression and other psychosocial conditions or counseling of lifestyle modifications. Given overall low-estimated total HS visits and screening and counseling rates, comparisons were not made across provider specialties. In addition, visits with missing data were excluded in our analyses. Underscreening for depression and substance abuse in HS patients may be due to lack of awareness. Additionally, integrating mental health screening and lifestyle modification counseling into time-constrained clinic visits may be challenging. Quick screening measures such as the Patient Health Questionnaire-2 for depression and implementation of streamlined mental health referral pathways may be useful [13]. Providing handouts on lifestyle modifications can increase patient's understanding of their comprehensive management plan in an efficient manner [14]. All specialties caring for HS patients should aim to incorporate psychosocial screening and lifestyle counseling into their care to improve patient outcomes. ## Statement of Ethics Ethical approval and consent were not required as this study was based on publicly available data. The National Center for Health Statistics (NCHS) Ethics Review Board reviews the content of the National Ambulatory Medical Care Surveys to ensure compliance with NCHS practices and procedures. Additional information can be found on www.cdc.gov/nchs/ahcd/index.htm. The National Ambulatory Medical Care Surveys fall under Title 42, US Code, section 242K, which permits data collection for health research. NCHS will not disclose responses in identifiable form without the consent of the individual or establishment in accordance with section 308(d) of the Public Health Service Act and the Confidential Information Protection and Statistical Efficiency Act of 2018. Additional information can be found on www.cdc.gov/nchs/ahcd/index.htm. ## Conflict of Interest Statement Jennifer L. Hsiao is on the board of directors for the Hidradenitis Suppurativa Foundation; has served as a consultant for Boehringer Ingelheim, Novartis, and UCB; and has served as a consultant and speaker for AbbVie. Vivian Y. Shi is on the board of directors for the Hidradenitis Suppurativa Foundation (HSF); is a stock shareholder of Learn Health; and has served as an advisory board member, investigator, speaker, and/or received research funding from Sanofi Genzyme, Regeneron, AbbVie, Eli Lilly, Novartis, SUN Pharma, LEO Pharma, Pfizer, Incyte, Boehringer Ingelheim, Aristea Therapeutics, Menlo Therapeutics, Dermira, Burt's Bees, Galderma, Kiniksa, UCB, WebMD, TARGET Pharmasolutions, Altus Lab, MYOR, Polyfin, GpSkin, and Skin Actives Scientific. There was no financial transaction for the preparation of this manuscript. All other authors report no conflicts of interest. ## Funding Sources This article has no funding source. ## Author Contributions Terri Shih and Jonathan Rick completed data analysis. Terri Shih and Devea R. De drafted the manuscript. Jonathan Rick, Vivian Shi, and Jennifer Hsiao edited and reviewed the manuscript. Vivian Shi and Jennifer Hsiao conceptualized and led the project. ## Data Availability Statement All data files are available from publicly available websites accessible through the CDC website, www.cdc.gov/nchs/ahcd/index.htm. Further inquiries can be directed to the corresponding author. ## References 1. Garg A, Malviya N, Strunk A, Wright S, Alavi A, Alhusayen R. Comorbidity screening in hidradenitis suppurativa evidence-based recommendations from the US and Canadian hidradenitis suppurativa foundations. J Am Acad Dermatol (2022) 86 1092-1101. PMID: 33493574 2. Sisic M, Tan J, Lafreniere KD. Hidradenitis suppurativa intimate partner violence and sexual assault. J Cutan Med Surg (2017) 21 383-387. PMID: 28481644 3. Patel KR, Lee HH, Rastogi S, Vakharia PP, Hua T, Chhiba K. Association between hidradenitis suppurativa and suicidality a systematic review and meta-analysis. J Am Acad Dermatol (2020) 83 737-744. PMID: 31862404 4. Garg A, Papagermanos V, Midura M, Strunk A, Merson J. Opioid and cannabis misuse among patients with hidradenitis suppurativa a population-based analysis in the United States. J Am Acad Dermatol (2018) 79 495.e1-500.e1. PMID: 29499293 5. Egeberg A, Gislason GH, Hansen PR. Risk of major adverse cardiovascular events and all-cause mortality in patients with hidradenitis suppurativa. JAMA Dermatol (2016) 152 429-434. PMID: 26885728 6. Sartorius K, Emtestam L, Jemec GBE, Lapins J. Objective scoring of hidradenitis suppurativa reflecting the role of tobacco smoking and obesity. Br J Dermatol (2009) 161 831-839. PMID: 19438453 7. Schrader AMR, Deckers IE, van der Zee HH, Boer J, Prens EP. Hidradenitis suppurativa a retrospective study of 846 Dutch patients to identify factors associated with disease severity. J Am Acad Dermatol (2014) 71 460-467. PMID: 24880664 8. Kromann CB, Ibler KS, Kristiansen VB, Jemec GB. The influence of body weight on the prevalence and severity of hidradenitis suppurativa. Acta Derm Venereol (2014) 94 553-557. PMID: 24577555 9. Shih T, De DR, Brooks B, Fixsen D, Shi VY, Hsiao JL. Optimizing hidradenitis suppurativa clinic visits patient perspectives. Int J Womens Dermatol (2022) 8 e040. PMID: 36000014 10. Garg A, Kirby JS, Lavian J, Lin G, Strunk A. Sex- and age-adjusted population analysis of prevalence estimates for hidradenitis suppurativa in the United States. JAMA Dermatol (2017) 153 760-764. PMID: 28492923 11. Wertenteil S, Strunk A, Garg A. Association of low socioeconomic status with hidradenitis suppurativa in the United States. JAMA Dermatol (2018) 154 1086-1088. PMID: 30073254 12. Rick JW, Thompson AM, Fernandez JM, Maarouf M, Seivright JR, Hsiao JL. Misdiagnoses and barriers to care in hidradenitis suppurativa a patient survey. Australas J Dermatol (2021) 62 e592-e594. PMID: 34314017 13. Kroenke K, Spitzer RL, Williams JBW. The Patient Health Questionnaire-2 validity of a two-item depression screener. Med Care (2003) 41 1284-1292. PMID: 14583691 14. Thompson AM, Fernandez JM, Shih T, Hamzavi I, Hsiao JL, Shi VY. Improving hidradenitis suppurativa patient education using written action plan a randomized controlled trial. J Dermatolog Treat (2021) 33 2677-2679. PMID: 34579620
--- title: The 15-year national trends of genital cancer incidence among Iranian men and women; 2005–2020 authors: - Gita Shafiee - Amir-hossein Mousavian - Ali Sheidaei - Mehdi Ebrahimi - Fatemeh Khatami - Kimiya Gohari - Mohammad Jabbari - Ali Ghanbari-Motlagh - Afshin Ostovar - Seyed Mohammad Kazem Aghamir - Ramin Heshmat journal: BMC Public Health year: 2023 pmcid: PMC10015665 doi: 10.1186/s12889-023-15417-0 license: CC BY 4.0 --- # The 15-year national trends of genital cancer incidence among Iranian men and women; 2005–2020 ## Abstract ### Background Cancer is a major health problem and cause of mortality worldwide. Despite the prevalence of other cancers in males and females, genital cancers are especially important because of their psychological effects on individuals. Currently, cervical cancer, corpus uteri neoplasm, and ovarian cancer are the most common gynecological cancers in Iran. Prostate cancer has increased in Iranian men in the last decade. Therefore, this study aimed to investigate the 15-year national trend in the incidence of genital cancers in the Iranian population. ### Methods In this study, we used Iranian cancer registration data collected by the Ministry of Health and Medical Education, demographic information from the reports of the Statistics Center of Iran, STEPs (STEPwise approach to non-communicable diseases risk factor surveillance), and Caspian (childhood and adolescence surveillance and prevention of adult non-communicable disease). A list of potential auxiliary variables and secondary variables at all levels of the province-age-sex were evaluated during the years. We used mixed-effects Poisson regression to model the data and calculate the incidence of each cancer. ### Results Our results show an enhancement in the outbreak of all types of male cancers, but the most important are prostate (11.46 in 2005 to 25.67 in 2020 per 100,000 males) and testicular cancers (2.39 in 2005 to 5.05 per 100,000 males). As for female cancers, there has been an increase in ovarian and corpus uteri neoplasm incidence with 6.69 and 4.14 incidences per 100,000 females in 2020, making them the most occurring female genital neoplasms. While the occurrence of cervical cancer has decreased over the years (4.65 in 2005 to 3.24 in 2020). In general, the incidence of genital cancers in men and women has amplified in the last 15 years. ### Conclusions Our study examined the trend of change for each malignant genital neoplasm for 15 years in Iranian men and women in each province. Considering the growing trend of the elderly population in Iran, patient awareness and early screening are essential in reducing mortality and costs imposed on patients and the health care system. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12889-023-15417-0. ## Background Cancer is a major worldwide healthcare concern and the cause of many deaths [1]. Despite medical breakthroughs and technological advancements in the prevention and treatment of cancer, the prevalence of people diagnosed with cancer has been on an upward trend in all countries [1]. In 2020, about 19.3 million new cancer cases and over 10.0 million cancer deaths occurred in the world, using the GLOBOCAN report [2]. According to a study by The Global Burden of Diseases (GBD), number of cancers detected and total number of deaths due to cancer has risen by $24.6\%$ and $20.9\%$ from 2010 to 2019 respectively [3]. With the vast increase in incidences, cancer has become one of Iran’s leading causes of death [4]. Female breast cancer ($11.7\%$), lung ($11.4\%$), and colorectal ($10.0\%$) cancers are the most common cancers in the world [2]. In addition to the common cancers in both sexes, genital cancers are important because of their psychological effects due to loss of genital parts or infertility regardless of gender [2]. The American Cancer Society estimated that approximately 1.9 million new cases of cancer are diagnosed in 2021. Of these, a total 376,970 people have genital cancers, which consists of 260,210 and 116,760 newly diagnosed cases in men and women, respectively [5]. In more detail, cervical cancer accounts for the most common type of genital cancers among women worldwide and according to the global classification in 2020, this cancer with an incidence of $3.1\%$ (604,127 of new cases), ranks as the eighth most common cancer in women [2, 6]. In Iran, the incidence rate of genital cancers has increased from 2.5 to 12.3 per 100,000 women from 1990 to 2016, while Cervical cancer, corpus uteri neoplasm and cancer of ovary are the most common gynecological cancers in Iran [7, 8]. Prostate cancer is the second most dominant type of cancer and the fifth cause of cancer mortality among men in the world [2]. Since the population of men over 65 years is growing, the number of subjects diagnosed with prostate cancer will increase in the near future [9]. In Iran, the incidence of prostate cancer has increased in the past decade and is currently higher than other Asian countries [4]. Based on the findings from 2011 to 2015 in Iran, the mean age of genital cancers in men was greater than women, with peak incidence at the age of 70–80 years in men and 50–60 years in women [7, 10]. In general, according to global and Iran statistics on the growing prevalence of genital cancers and given the fact that the incidence of this types of cancer depends on numerous factors such as age, sex, geographical location, lifestyle and race [11–13], a specific and comprehensive study of these malignancies in both genders, is not available in Iran yet. Moreover, there has been no comparison or evaluation of these types of neoplasms. Therefore, considering the importance of this type of study in preventing and reducing the economic costs of health care and improving quality of life, we investigated this issue. ## Data sources We used data from the Iranian population-based cancer registry, gathered by the Ministry of Health and Medical Education from all medical facilities. Individual data were available for 2008 to 2010, 2014, and 2015. The information in data includes ICD10 codes for neoplasm type, age, sex, and the province of residence. There were a few missing values for each variable and the proportion of missing were less than $5\%$, therefore we imputed them using the multiple imputation bootstrapping-based algorithm by Amelia package in R software [14]. Several scenarios for age groups definition were considered including the length of groups, optimal cut points, the minimum valid age, and the way of definition for the last group. For starting age, we relied on the global burden of diseases (GBD) study and set it at 15 years old [3]. Conducting fivefold cross-validation revealed the 10-years length age group has a lower mean square error than alternative approaches especially 5-years length age groups [15]. In addition, the selection of more than 75 years old as the latest age group showed better model performance comparing with more than 85 years. Therefore, the age groups in this study start from 15 and the groups include 10 years until the last one that is more than 75 years (15–24, 25–34, etc., and 75 + years old). The population data were extracted from reports of the Statistical Center of Iran (SCI) for population and housing census 2001, 2006, 2011, and 2016 [16]. The data set was formed according to age and sex groups for each sub-national division. In order to estimate the population for the years between two consecutive censuses, the growth formula for the population was used [17]. The growth rate was calculated and applied separately for each subgroup of the dataset. For the years between 2017 and 2020, the growth rate of the period 2011 to 2016 was used. The connection between cancer registry information and covariates was not possible in individual level. Therefore, we select an ecological approach rather than a cross sectional study. In this manner subjects are groups of individuals who were living in a same province as geography characteristic and were in the same sex-age group. This approach enabled us to use covariates from other sources of information. We prepared a list of potential covariates for modeling section according to relevancy and availability of data. There were two national survey study that are conducting regularly in Iran health system. Both surveys have representative sample and follow the World Health Organization (WHO) guidelines. STEPwise approach to non-communicable diseases risk factor surveillance known as STEPs focus on risk factors for non-communicable diseases in adults more than 18 years old [18]. We used all 6 phases of this survey conducted in years 2005, 2007, 2008, 2009, 2011 and 2016 [19]. In order to cover all target population, we add the information of childhood and adolescence surveillance and prevention of adult non-communicable disease (CASPIAN study). This survey follows the WHO, global school-based student health survey (GHSH) instructions and cover adolescences population at school age [20]. Data for CASPIAN-III (2009–2010), CASPIAN-IV (2011–2012) and CASPIAN-V [2015] were used [21]. Finally, we entered the urbanization proportion to model as the proxy indicator for differentiation between urban–rural lifestyle. This variable derived from population dataset which is estimated based on census information. We defined it as the ratio of population living in urban areas to population living in rural areas. All the data sources are nationally representative surveys that were based on international health organization guidelines. ## Covariates We extracted a list of potential covariates that could cooperate in modeling. In the first step, we calculated all the covariate values at the individual level, then aggregated them to construct a data set for all the combinations of the province, year, age, and sex. In case of unavailable real data, we estimated the values using a nonparametric smoothing approach, spline. In this manner, we used the spline function in R statistical software and computes a monotone cubic spline using Hyman filtering [22]. The smoothing and estimation of covariates were conducted in all levels of province-age-sex combinations across the year. The BMI was computed as weight in kilograms divided by the square of height in meters. The smoking history is defined as if a person smoked any tobacco products during her/his life. The current smoking status is also defined similarly but at the study time. We extracted the key components of food frequency questionnaires, include the appropriate percentage of using fruit, vegetables, and fish. In this part, we used the prevalence of less than five total servings (400 g) of fruit and vegetables per day and non-weekly fish consumption as the risk factors for non-communicable diseases. Blood pressure measurements enter directly into the models as the means of systolic and diastolic blood pressure. In addition, the prevalence of high blood pressure in the sub-populations was added to the covariates list. The same approach was considered for entering fasting blood glucose. Both glucose level and prevalence of type 2 diabetes mellitus were made for modeling. ## Statistical modeling We used a mixed-effects Poisson regression in order to model the data and estimate the incidence rates [23]. The separate models were fitted for each type of malignant neoplasms. The number of new cases were modeled against the fixed effect of covariates. In addition, the fixed effects of age groups entered the model as dummy variables. The correlation between incident cases across times and unknown causes of variations within the provinces were captured by the random effect of year and provinces respectively. Finally, the population at risk entered as the offset in the model. ## Model building and validation A backward elimination approach was used to select the best subset of covariates that should remain in the model. In order to select the best format of entering fasting blood glucose and blood pressure, we fitted 4 different starting full models and then reduced these models to find the best one. These 4 models considered all other covariates in addition to 1) mean of Fasting Blood Glucose (FBG), Systolic blood pressure (SBP), and Diastolic Blood Pressure (DBP) or 2) mean of FBG, the prevalence of hypertension, or 3) mean of SBP and DBP and the prevalence of diabetes or 4) the prevalence of diabetes and hypertension. In this way, we prevented entering collinear variables into the model. Models were compared using Akaike information criterion (AIC) and Bayesian information criterion (BIC) criteria. The model prediction power and validity were explored using a fivefold cross-validation approach. At first, the dataset was divided randomly into 5 subsets. Then at each step, four-part of these subsets were used to model building and the other one for checking the results. The root means the squared error was used to evaluate the models. We used a similar approach to select the best definition of age groups. ## Ethical consideration This study was authorized by the ethical committee of Tehran University of Medical Sciences (IR.TUMS.VCR.REC.1398.218). Recruited participants’ data is protected by all authors. No individual data is reported since results are created using statistical modeling. Participants also provided informed consents. ## Malignant neoplasms of female genital organs Cervix uteri has the highest age-specific incidence rate in 2005 with 12.22 (10.9–13.54) in the 65–74 years old age group per 100,000 females. Corpus uteri has the highest age-specific incidence rates in 2010, 13.88 (13.5–14.27) and 2015, 15.98 (15.4–16.55) for age groups 55–64 and 65–74 years old respectively. Ovarian neoplasm in the age group 65–74 years with the rate of 20.57 (19.94–21.19) has the highest incidence rate in 2020. Supplementary Fig. 1 depicts the changing trend of age-specific incidence rates across all years for all types of female’s genital malignant neoplasm. The crossing lines is a sign of a changing age pattern of incidence rates across years. For instance, the incidence rate of malignant neoplasm of the vulva was higher in 75 + until 2018 and after this time in 65–74 years. Albeit, the distance between these two groups is going to decrease over time. The neoplasm of cervix uteri shows the highest age-standardized incidence rate of 4.65 (4.23–5.10) in 2005. The incidence of this neoplasm is almost declining over the years (Fig. 1). Such that it becomes the third most occurring neoplasm with an incidence rate of 3.24 (2.99–3.51) in 2020 after ovarian and corpus uteri neoplasm with 6.69 (6.32–7.06) and 4.14 (3.86–4.43) respectively. In addition, vagina neoplasm slightly increases, and placenta neoplasm decreases over the years. The age-standardized incidence rate of vulva and unspecified part of the uterus neoplasms are almost constant. Fig. 1Age Standardized Incidence Rate of Malignant Neoplasms of Female Genital Organs in 100,000 female population Percentages of share for each type of neoplasms from the total female genital organs neoplasms across years are presented in Fig. 2 as a stacked bar plot. The largest and the smallest share of malignant neoplasms in 2005 belong to cervix uteri and other unspecified neoplasms with $26.19\%$ and $4.06\%$ respectively. Ovarian neoplasm share increase from $21.37\%$ corresponding to ranked 2 in 2005 to $32.94\%$ corresponding to the first rank in 2020. On the other hand, corpus uteri placed in the second rank of female genital neoplasm in 2020 with $21.61\%$ of total incidence cases. Fig. 2Percentage of each type of malignant neoplasms from the total female genital organ malignant neoplasms Geographical distribution of incidence rates across provinces in 2005 and 2020 for female genital neoplasms of Iran are available in Fig. 3. All provinces show the increasing trend of incidence rate. Fig. 3Geographical distribution of female genital organs neoplasms incidence rates in 2005 and 2020 Supplementary Fig. 2 shows the age-specific and all ages incidence rates of four male genital malignant neoplasms per 100,000 males for 2005, 2010, 2015, and 2020. The malignant neoplasm of the prostate has the highest incidence rate in the ages after 45 years, while testis neoplasm was responsible for the most incident cases in the earlier age groups. The incidence rate of the prostate, testis, and penis neoplasms increased over time. The estimated incidence rate of prostate neoplasm is 12.15 (11.97–12.34) in 2005 and 31.36 (31.23–31.53) in 2020. Age-standardized time trends of incidence rate per 100,000 male population are depicted in Fig. 4. The highest values and the most increasing rate are related to the prostate neoplasm that increases from 11.46 (10.87–12.07) in 2005 to 25.67 (24.96–26.40) in 2020. The age-standardized incidence rate of penis neoplasm shows the least values in the years of study. The incidence rate in 2020 is 2.19 (1.94–2.45) that is twice the incidence rate in 2005 with a value of 1.08 (0.87–1.32).Fig. 4Age Standardized Incidence Rate of Malignant Neoplasms of Male Genital Organs in 100,000 male population The proportions from the total malignant neoplasms of male genital organs are depicted in Fig. 5. This proportion is almost constant over the years. It varies from $75.56\%$ in 2005 to $78.24\%$ in 2020 for prostate neoplasm. The highest proportion of testis neoplasm is related to 2014 with $12.29\%$ and the lowest one is $11.83\%$ for 2005. The penis neoplasm reaches the highest and the lowest proportion in 2010 and 2020 with $4.99\%$ and $3.29\%$ respectively. Fig. 5Percentage for each type of malignant neoplasms from the total male genital organs malignant neoplasms Finally, the geographical distribution of male genital organs neoplasms incidence rates for the first and last year of the study is depicted in Fig. 6.Fig. 6Geographical distribution of male genital organs neoplasms incidence rates in 2005 and 2020 ## Discussion In this study, we conducted an overview of national and subnational incidence rate combined with trends for each type of gynecological cancers in both men and women from 2005 to 2020 in Iran. The total number of cancer incidences in both men and women has increased over the past 15 years. There is a rising trend in the incidence rate of ovarian and vagina cancer as well as corpus uteri while the incidence rate for cervix uteri has decreased over the years. Our results indicate an increase in the incidence of all male cancer types but most notably prostate and testis cancer. In our study in cancers related to women, Cervix uteri showed a decreasing trend from 2005 to 2017 with a mild increase from then to 2020. It has fall from the first to the third place of cancers with the most numbers of incidence in woman behind Ovary cancer and Corpus Uteri, which can be explained with expanding compliance with pap smear screening test and the decreased age for starting screening in women resulting in early diagnosis and detection of pre-malignant lesions [24]. When comparing our data to global findings, the incidence rate for cervical cancer is decreasing over the past decade but it still has the highest numbers of new cases annually [2, 25]. As for the Ovarian cancer, there is an increasing trend from 2005 to 2011 followed with a mild decrease in incidence until 2016 and a sudden increase afterwards. Thus, making it the most common genital cancer in Iranian women. This is in line with other studies performed in Iran [7, 26, 27]. The incidence for ovarian cancer has increased globally while Asia is accountable for more than half the incidences reported ($51.8\%$) followed by Europe ($22.9\%$) [28, 29]. Obesity is a well-known risk factor for ovarian cancer [30, 31]. There is also a strong correlation between Human Development Index (HDI) and life expectancy index with incidence for ovarian cancer [28], while increase in HDI is resulted in reduced number of incidences for cervical cancer [32]. Ovarian cancer is usually detected too late and at advanced stages, since there is no specific symptoms at early stages or a sensitive screening test [33]. Despite the recent decline in cervical cancer trend in Iran, it is still responsible for $21\%$ of all women genital cancers and is the most common in the world. The highest incidence and mortality are in Africa and incidence rates are 7 to 10 times lower in Northern America, Australia, and Western Asia [2, 34]. Patient awareness and easy and affordable screening for cervical cancer using pap smear tests, have helped a lot with early diagnosis and proper treatment of the disease, resulting in decreased mortality rate of cervical cancer around the world [24]. Unlike developed countries, HPV (Human papillomavirus) vaccination at population level is not an optimal choice for cervical cancer prevention in Iran as it is not cost-effective [35–37]. As for the male genital cancers, prostate cancer has the highest numbers of incidence, being responsible for more than $75\%$ of all male genital cancers followed by testis cancer ($12\%$). Although according to age-specific pattern, prostate cancer is the most common type of cancer in 45 years and more while in earlier group ages, testis cancer is more common. Comparing our results to global findings, the incidence rate of prostate cancer varies from 6.3 to 83.4 per 100,000 men across regions. The highest rates found in Western Europe, and Northern America countries and the lowest rates in Asia and Northern Africa. Elderly, family history of this cancer, lifestyle factors such as smoking, obesity, nutritional status may increase the risk of advanced prostate cancer [2, 38]. This might be due to differences in usage of Prostate-Specific Antigen (PSA) testing in each region [39]. Genetic factors are pivotal in occurrence of prostate cancer [40]. Moreover, Chu et al. [ 41] suggests the incidence for prostate cancer in African-Americans to be 40 times higher comparing with African men. Aside from the possibilities of underdiagnosis or lack of proper healthcare system and valid registries, this indicates the importance of environmental factor along with genetics. Westernized life style and physical inactivity have a positive correlation with prostate cancer incidence [42]. There are some suggestions regarding lifestyle or dietary changes [43, 44], but there are no proven prevention methods for prostate cancer and PSA serum marker is currently the best clinical monitoring method for early diagnosis. The strength of this study is the duration of data collection, demonstrating the incidence trend for each individual neoplasm for both age-standardized rate and age-specific rate groups. In addition, we had access to individual data from the national cancer registry. It facilitated our data process and modeling part. We ensured that the data had the most possible completeness, fewer missing values, and that all the relevant fields were gathered. Also gathering data for each province separately, has helped to identify regions with most incidents. This will benefit health care system and policy makers as to where to use resources that is most needed. One limitation we faced was the incomplete or missing data in our registry. Statistical models were used to extrapolate the missing data. The proportion of missing data is not high and is within acceptable range comparing with similar studies in Iran and worldwide [7, 45, 46]. Another limitation of our work was the inability to specify all malignant neoplasms and cancer types and therefore reporting a portion of malignant neoplasm as “malignant neoplasm of unspecified male /female genital organs” although this was the case for a small portion of samples. ## Conclusions We conducted a study to observe the changing trend for each genital malignant neoplasm for a duration of 15 years in both men and women. Our study contained data for each age subgroups as well as trend for cancer incidence in every province of Iran over 15 years. Prostate cancer and Ovarian cancer were the most common cancer in 2020 in men and women respectively. Considering Iran as an aging population, the incidence rate is estimated to increase over next decades. Patient awareness and early screening are essential in reducing mortality and expenses forced upon patients and health care system. ## Supplementary Information Additional file 1: Supplementary Figure 1. Age Specific Trends of Incidence Rate of Malignant Neoplasms of Female Genital Organs in 100,000 female population. Additional file 2: Supplementary Figure 2. Age Specific Trends of Incidence Rate of Malignant Neoplasms of male Genital Organs in 100,000 male population. Additional file 3: Supplementary Figure 3. Geographical guide of Iranian provinces. ## References 1. Bray F, Laversanne M, Weiderpass E, Soerjomataram I. The ever-increasing importance of cancer as a leading cause of premature death worldwide. Cancer (2021.0) 127 3029-3030. DOI: 10.1002/cncr.33587 2. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin (2021.0) 71 209-249. DOI: 10.3322/caac.21660 3. Kocarnik JM, Compton K, Dean FE, Fu W, Gaw BL, Harvey JD, Henrikson HJ, Lu D, Pennini A, Xu R. Cancer incidence, mortality, years of life lost, years lived with disability, and disability-adjusted life years for 29 cancer groups from 2010 to 2019: a systematic analysis for the global burden of disease study 2019. JAMA Oncol (2022.0) 8 420-444. DOI: 10.1001/jamaoncol.2021.6987 4. Farhood B, Geraily G, Alizadeh A. Incidence and mortality of various cancers in Iran and compare to other countries: a review article. Iran J Public Health (2018.0) 47 309-316. PMID: 29845017 5. Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2021. CA Cancer J Clin (2021.0) 71 7-33. DOI: 10.3322/caac.21654 6. Arbyn M, Weiderpass E, Bruni L, de Sanjose S, Saraiya M, Ferlay J, Bray F. Estimates of incidence and mortality of cervical cancer in 2018: a worldwide analysis. Lancet Glob Health (2020.0) 8 e191-e203. DOI: 10.1016/S2214-109X(19)30482-6 7. Eftekharzadeh S, Ebrahimi N, Samaei M, Mohebi F, Mohajer B, Sheidaei A, Gohari K, SaeediMoghaddam S, Ahmadi N, MohammadiFateh S. National and subnational trends of incidence and mortality of female genital cancers in Iran; 1990–2016. Arch Iran Med (2020.0) 23 434-444. DOI: 10.34172/aim.2020.40 8. Roshandel G, Ghanbari-Motlagh A, Partovipour E, Salavati F, Hasanpour-Heidari S, Mohammadi G, Khoshaabi M, Sadjadi A, Davanlou M, Tavangar SM. Cancer incidence in Iran in 2014: results of the Iranian National Population-based Cancer Registry. Cancer Epidemiol (2019.0) 61 50-58. DOI: 10.1016/j.canep.2019.05.009 9. Stangelberger A, Waldert M, Djavan B. Prostate cancer in elderly men. Rev Urol (2008.0) 10 111-119. PMID: 18660852 10. Ferlay J, Soerjomataram I, Dikshit R, Eser S, Mathers C, Rebelo M, Parkin DM, Forman D, Bray F. Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012. Int J Cancer (2015.0) 136 E359-386. DOI: 10.1002/ijc.29210 11. Özdemir BC, Dotto GP. Racial differences in cancer susceptibility and survival: more than the color of the skin?. Trends Cancer (2017.0) 3 181-197. DOI: 10.1016/j.trecan.2017.02.002 12. Katzke VA, Kaaks R, Kühn T. Lifestyle and cancer risk. Cancer J (2015.0) 21 104-110. DOI: 10.1097/PPO.0000000000000101 13. Kamangar F, Dores GM, Anderson WF. Patterns of cancer incidence, mortality, and prevalence across five continents: defining priorities to reduce cancer disparities in different geographic regions of the world. J Clin Oncol (2006.0) 24 2137-2150. DOI: 10.1200/JCO.2005.05.2308 14. Zhang Z. Multiple imputation for time series data with Amelia package. Ann Transl Med (2016.0) 4 56. PMID: 26904578 15. 15.Hastie T, Tibshirani R, Friedman JH. The elements of statistical learning: data mining, inference, and prediction. 2nd Ed. Stanford, California: Springer; 2009. 16. 16.Census 2016 - general results [website]. Statistical Center of Iran; 2016. https://www.amar.org.ir/english/Population-and-Housing-Censuses/Census-2016-General-Results. Accessed 7 Mar 2023. 17. Shyu E, Caswell H. Calculating second derivatives of population growth rates for ecology and evolution. Methods Ecol Evol (2014.0) 5 473-482. DOI: 10.1111/2041-210X.12179 18. 18.World Health OrganizationNoncommunicable DMental Health CWHO STEPS surveillance manual: the WHO STEPwise approach to chronic disease risk factor surveillance / Noncommunicable Diseases and Mental Health, World Health Organization2005GenevaWorld Health Organization. WHO STEPS surveillance manual: the WHO STEPwise approach to chronic disease risk factor surveillance / Noncommunicable Diseases and Mental Health, World Health Organization (2005.0) 19. Djalalinia S, Modirian M, Sheidaei A, Yoosefi M, Zokaiee H, Damirchilu B, Mahmoudi Z, Mahmoudi N, Hajipour MJ, Peykari N. Protocol design for large-scale cross-sectional studies of surveillance of risk factors of non-communicable diseases in Iran: STEPs 2016. Arch Iran Med (2017.0) 20 608-616. PMID: 29048923 20. Beal T, Morris SS, Tumilowicz A. Global patterns of adolescent fruit, vegetable, carbonated soft drink, and fast-food consumption: a meta-analysis of global school-based student health surveys. Food Nutr Bull (2019.0) 40 444-459. DOI: 10.1177/0379572119848287 21. Kelishadi R, Heshmat R, Farzadfar F, EsmaeilMotlag M, Bahreynian M, Safiri S, Ardalan G, RezaeiDarzi E, Asayesh H, Rezaei F. Prevalence of cardio-metabolic risk factors in a nationally representative sample of Iranian adolescents: the CASPIAN-III study. J Cardiovasc Thorac Res (2017.0) 9 12-20. DOI: 10.15171/jcvtr.2017.02 22. Smith L, Hyndman RJ, Wood SN. Spline interpolation for demographic variables: the monotonicity problem. J Popul Res (2004.0) 21 95-98. DOI: 10.1007/BF03032212 23. 23.Agresti A. Analysis of ordinal categorical data. 2nd Ed. Wiley; 2010. Online ISBN:9780470594001. 10.1002/9780470594001. 24. Yang DX, Soulos PR, Davis B, Gross CP, Yu JB. Impact of widespread cervical cancer screening: number of cancers prevented and changes in race-specific incidence. Am J Clin Oncol (2018.0) 41 289-294. DOI: 10.1097/COC.0000000000000264 25. Lin S, Gao K, Gu S, You L, Qian S, Tang M, Wang J, Chen K, Jin M. Worldwide trends in cervical cancer incidence and mortality, with predictions for the next 15 years. Cancer (2021.0) 127 4030-4039. DOI: 10.1002/cncr.33795 26. Sharifian A, Pourhoseingholi MA, Norouzinia M, Vahedi M. Ovarian cancer in Iranian women, a trend analysis of mortality and incidence. Asian Pac J Cancer Prev (2014.0) 15 10787-10790. DOI: 10.7314/APJCP.2014.15.24.10787 27. Moradi Y, Jafari M, Chaichian S, Khateri S, Akbarian A, Moazzami B, Mansori K, Mahmodi Y, Samie S. Trends in ovarian cancer incidence in Iran. Int J Cancer Manag (2016.0) 9 e5452 28. Khazaei Z, Namayandeh SM, Beiranvand R, Naemi H, Bechashk SM, Goodarzi E. Worldwide incidence and mortality of ovarian cancer and Human Development Index (HDI): GLOBOCAN sources and methods 2018. J Prev Med Hyg (2021.0) 62 E174-e184. PMID: 34322634 29. Zhang Y, Luo G, Li M, Guo P, Xiao Y, Ji H, Hao Y. Global patterns and trends in ovarian cancer incidence: age, period and birth cohort analysis. BMC Cancer (2019.0) 19 984. DOI: 10.1186/s12885-019-6139-6 30. Olsen CM, Green AC, Whiteman DC, Sadeghi S, Kolahdooz F, Webb PM. Obesity and the risk of epithelial ovarian cancer: a systematic review and meta-analysis. Eur J Cancer (2007.0) 43 690-709. DOI: 10.1016/j.ejca.2006.11.010 31. Protani MM, Nagle CM, Webb PM. Obesity and ovarian cancer survival: a systematic review and meta-analysis. Cancer Prev Res (2012.0) 5 901-910. DOI: 10.1158/1940-6207.CAPR-12-0048 32. Singh GK, Azuine RE, Siahpush M. Global inequalities in cervical cancer incidence and mortality are linked to deprivation, low socioeconomic status, and human development. Int J MCH AIDS (2012.0) 1 17-30. DOI: 10.21106/ijma.12 33. Patni R. Screening for ovarian cancer: an update. J Midlife Health (2019.0) 10 3-5. PMID: 31001049 34. Hwang JY, Lim WY, Tan CS, Lim SL, Chia J, Chow KY, Chay WY. Ovarian cancer incidence in the multi-ethnic Asian city-state of Singapore 1968–2012. Asian Pac J Cancer Prev (2019.0) 20 3563-3569. DOI: 10.31557/APJCP.2019.20.12.3563 35. Khatibi M, Rasekh HR, Shahverdi Z, Jamshidi HR. Cost-effectiveness evaluation of quadrivalent human papilloma virus vaccine for HPV-related disease in Iran. Iran J Pharm Res (2014.0) 13 225-234. PMID: 24711850 36. Majidi A, Ghiasvand R, Hadji M, Nahvijou A, Mousavi A-S, Pakgohar M, Khodakarami N, Abedini M, AmouzegarHashemi F, RahnamayeFarzami M. Priority setting for improvement of cervical cancer prevention in Iran. Int J Health Policy Manag (2015.0) 5 225-232. DOI: 10.15171/ijhpm.2015.201 37. Yaghoubi M, Nojomi M, Vaezi A, Erfani V, Mahmoudi S, Ezoji K, Zahraei SM, Chaudhri I, Moradi-Lakeh M. Cost-effectiveness analysis of the introduction of HPV vaccination of 9-year-old-girls in Iran. Value Health Reg Issues (2018.0) 15 112-119. DOI: 10.1016/j.vhri.2018.03.001 38. Ha Chung B, Horie S, Chiong E. The incidence, mortality, and risk factors of prostate cancer in Asian men. Prostate Int (2019.0) 7 1-8. DOI: 10.1016/j.prnil.2018.11.001 39. Quinn M, Babb P. Patterns and trends in prostate cancer incidence, survival, prevalence and mortality. Part I: international comparisons. BJU Int (2002.0) 90 162-173. DOI: 10.1046/j.1464-410X.2002.2822.x 40. Langeberg WJ, Isaacs WB, Stanford JL. Genetic etiology of hereditary prostate cancer. Front Biosci (2007.0) 12 4101-4110. DOI: 10.2741/2374 41. Chu LW, Ritchey J, Devesa SS, Quraishi SM, Zhang H, Hsing AW. Prostate cancer incidence rates in Africa. Prostate Cancer (2011.0) 2011 947870. DOI: 10.1155/2011/947870 42. Baade PD, Youlden DR, Krnjacki LJ. International epidemiology of prostate cancer: geographical distribution and secular trends. Mol Nutr Food Res (2009.0) 53 171-184. DOI: 10.1002/mnfr.200700511 43. Mills PK, Beeson WL, Phillips RL, Fraser GE. Cohort study of diet, lifestyle, and prostate cancer in Adventist men. Cancer (1989.0) 64 598-604. DOI: 10.1002/1097-0142(19890801)64:3<598::AID-CNCR2820640306>3.0.CO;2-6 44. Wilson KM, Giovannucci EL, Mucci LA. Lifestyle and dietary factors in the prevention of lethal prostate cancer. Asian J Androl (2012.0) 14 365-374. DOI: 10.1038/aja.2011.142 45. Danaei G, Finucane MM, Lin JK, Singh GM, Paciorek CJ, Cowan MJ, Farzadfar F, Stevens GA, Lim SS, Riley LM. National, regional, and global trends in systolic blood pressure since 1980: systematic analysis of health examination surveys and epidemiological studies with 786 country-years and 5·4 million participants. Lancet (2011.0) 377 568-577. DOI: 10.1016/S0140-6736(10)62036-3 46. Finucane MM, Stevens GA, Cowan MJ, Danaei G, Lin JK, Paciorek CJ, Singh GM, Gutierrez HR, Lu Y, Bahalim AN. National, regional, and global trends in body-mass index since 1980: systematic analysis of health examination surveys and epidemiological studies with 960 country-years and 9·1 million participants. Lancet (2011.0) 377 557-567. DOI: 10.1016/S0140-6736(10)62037-5
--- title: 'Vitamin D3 supplementation as an adjunct in the management of childhood infectious diarrhea: a systematic review' authors: - Samuel N Uwaezuoke - Chioma L Odimegwu - Ngozi R Mbanefo - Chizoma I Eneh - Ijeoma O Arodiwe - Uzoamaka V Muoneke - Francis N Ogbuka - Chibuzo O Ndiokwelu - Anthony T Akwue journal: BMC Infectious Diseases year: 2023 pmcid: PMC10015675 doi: 10.1186/s12879-023-08077-3 license: CC BY 4.0 --- # Vitamin D3 supplementation as an adjunct in the management of childhood infectious diarrhea: a systematic review ## Abstract ### Background Some studies have reported the possible role of vitamin D3 in ameliorating disease outcomes in childhood infectious diarrhea. However, findings about its effectiveness and the association of serum vitamin D levels with diarrhea risk appear inconsistent. We aimed to determine the efficacy of oral vitamin D3 as an adjunct in managing childhood infectious diarrhea and the relationship between vitamin D status and the disease. ### Methods We searched the PubMed and Google Scholar electronic databases for relevant articles without limiting their year of publication. We selected primary studies that met the review’s inclusion criteria, screened their titles and abstracts, and removed duplicates. We extracted data items from selected studies using a structured data-extraction form. We conducted a quality assessment of randomized controlled trials (RCTs) and non-randomized studies with the Cochrane collaboration tool and the Newcastle Ottawa Scale, respectively. We assessed the strength of the relationship between serum vitamin D levels and diarrhea using the correlation model. We estimated the I2 and tau2 values to assess between-study heterogeneity. ### Results Nine full-text articles were selected, consisting of one RCT, three cross-sectional studies, two cohort studies, two longitudinal/prospective studies, and one case-control study. A total of 5,545 participants were evaluated in the nine studies. Six non-randomized studies provided weak evidence of the relationship between vitamin D levels and diarrhea risk as there was no correlation between the two variables. The only RCT failed to demonstrate any beneficial role of vitamin D3 in reducing the risk of recurrent diarrhea. The calculated I2 and tau2 values of $86.5\%$ and 0.03, respectively suggested a high between-study heterogeneity which precluded a meta-analysis of study results. ### Conclusion Oral vitamin D3 may not be an effective adjunct in managing childhood infectious diarrhea. Additionally, the relationship between vitamin D status and infectious diarrhea appears weak. We recommend more adequately-powered RCTs to determine the effectiveness of vitamin D3 as an adjunct therapy in infectious diarrhea. ## Background Diarrhea is one of the top-four infectious causes of childhood morbidity and mortality in tropical developing countries: the remainder comprising pneumonia, malaria, and human immunodeficiency virus (HIV)/acquired immune deficiency syndrome (AIDS) [1, 2]. The World Health Organization (WHO) statistics reveal the enormous health burden associated with childhood infectious diarrhea. For instance, it is the second leading cause of mortality in under-five children, as more than half a million succumb to diarrhea-related deaths yearly [3]. Also, there is global documentation of nearly 1.7 billion diarrhea cases annually [3]. The incidence rate is high in low-income countries (LICs), where each child under three years experiences three episodes of diarrhea on average every year [3]. Contaminated water from poor sanitary hygiene is a significant source of contracting the disease in these settings. Water contamination with fecal matter (due to the high rate of open defecation) constitutes a public health challenge. Thus, rotavirus and Escherichia coli are the most common etiologic agents of moderate-to-severe diarrhea in LICs [3]. There is a standard management protocol that mitigates the adverse consequences associated with childhood diarrhea. The protocol comprises supplemental zinc and low osmolarity oral rehydration solution (ORS) and has increased patients’ survival rates over the years [4–6]. Zinc is critical in modulating the host’s resistance to infectious agents and reducing diarrhea risk, severity, and duration [6]. Its precise mechanism of ameliorating diarrhea-related morbidity is largely unresolved. However, the micronutrient enhances the absorption of water and electrolytes, stimulates intestinal neo-epithelialization, and increases the levels of brush border enzymes [7]. Also, zinc promotes a better clearance of etiologic pathogens by increasing T lymphocytes and macrophage levels [7]. Thus, zinc deficiency negatively impacts the immune system’s maturation [8] and may explain why children with reduced serum zinc levels experience either severe diarrhea or higher episodes of diarrhea [9]. Similarly, vitamin A is another micronutrient considered an adjunct in treating childhood diarrhea. Oral vitamin A is associated with decreased incidence rates of diarrhea and its related mortality [10]. Nevertheless, there is no consensus yet on this beneficial effect. For instance, a randomized controlled trial (RCT) demonstrated that oral vitamin A supplementation did not affect the duration of diarrhea during an acute episode in well-nourished infants aged between 6 and 12 months [11]. Recently, there has been a renewed interest in using oral vitamin D3 to improve the outcomes of childhood infectious diarrhea. Given the pleiotropic nature of vitamin D, it modulates immunologic function: particularly the enhancement of innate immunity, such as the production of gut antimicrobial peptides [12–14]. Because of this link with enteric immunologic function, its role in infectious diarrheas is now a research subject. For instance, some investigators reported that low serum 25-hydroxyvitamin D level was associated with increased intensity of diarrhea and poor disease outcomes in Bulgarian toddlers [15]. Furthermore, a cohort study in Iranian children revealed a negative correlation between serum 25-hydroxyvitamin D level and acute bacterial diarrhea; thus, the authors suggested that vitamin D could be involved in the pathogenesis of diarrhea [16]. However, other researchers in Afghanistan noted that oral vitamin D3 failed to reduce the risk for recurrent diarrhea in a population of infants they studied [17]. We initiated this systematic review because of these inconsistent findings, focusing on controlled intervention studies that utilized vitamin D3 to prevent or treat infectious diarrheas and studies that evaluated the relationship between serum 25-hydroxyvitamin D levels and incident diarrhea in children. We thus aimed to determine the efficacy of oral vitamin D3 as an adjunct in managing childhood infectious diarrhea, and the relationship between vitamin D status and the disease. We conducted and reported the systematic review in conformity with the Preferred Reporting Items for Systematic reviews and Meta-analyses (PRISMA) guidelines [18]. ## Protocol and registration There was no review protocol for the present systematic review. ## Literature search strategy We searched the PubMed and Google Scholar electronic databases for relevant articles without limiting their year and language of publication. Based on the title of the systematic review, we used the following descriptors in PubMed in multiple combinations (as MeSH terms or not) with Boolean operators (AND/OR): (“cholecalciferol“[MeSH Terms] OR “cholecalciferol“[All Fields]) AND “childhood“[All Fields] AND (“dysentery“[MeSH Terms] OR “dysentery“[All Fields] OR (“infectious“[All Fields] AND “diarrhea“[All Fields]) OR “infectious diarrhea“[All Fields]). The date of the last search was 31 August 2022. We also used descriptors like ‘infectious diarrhea,’ ‘cholecalciferol,’ ‘childhood,’ and ‘adjunct therapy’ to search the Google Scholar database for related articles. ## Inclusion and exclusion criteria We selected primary studies which met the inclusion criteria. These criteria include cohort studies or randomized controlled trials (RCTs) on human subjects, cross-sectional or case-control studies that evaluated the association of vitamin D status (serum 25-hydroxyvitamin D level or vitamin D-binding protein [DBP] level as a surrogate marker) in children with episodes of diarrhea, and full-text articles with these study designs published in or translated into the English language. Excluded articles comprised abstracts, reviews, editorials, commentaries, conference proceedings, and studies without primary data. ## Study selection We screened the titles and abstracts of retrieved articles from the two electronic databases and independently assessed potentially eligible full-text articles for selection and inclusion in the final list of papers for review. We resolved possible disagreements on selected studies by consensus. We excluded duplicates and primary studies whose objectives were not in tandem with the aim of the present systematic review. The search of PubMed and Google Scholar databases yielded 1,287 and 24,100 articles, respectively: giving a total of 25,387. After removing duplicates and articles unrelated to the topic, the remaining papers were 1,344. Screening for their relevance to the present systematic review resulted in the exclusion of more records ($$n = 1178$$) - including abstracts ($$n = 26$$), editorials ($$n = 10$$), commentaries ($$n = 6$$), and conference proceedings ($$n = 49$$) - which scaled down the number of papers to 75. Following the assessment of the 75 full-text articles for eligibility, further exclusion of narrative reviews ($$n = 43$$), systematic review/meta-analysis ($$n = 2$$), and studies with secondary data ($$n = 21$$) yielded nine papers. We finally selected nine full-text articles for the present systematic review (Fig. 1). Fig. 1The Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) algorithm for inclusion of studies on the relationship between vitamin D3 and childhood infectious diarrhea ## Quality assessment We assessed the methodological quality of each selected study using Newcastle-Ottawa Scale (NOS) [19] and Cochrane collaboration’s tool [20] for non-randomized studies and RCTs, respectively. The Newcastle Ottawa Scale consists of the following criteria for evaluating case-control or cross-sectional studies: ‘selection’ (maximum of 5 stars), ‘comparability’ (maximum of 2 stars), and ‘exposure/outcome’ (maximum of 3 stars). We rated the quality of each study high if the assigned score is ≥ 7 stars or low if the score is ≤ 7 stars. The Cochrane collaboration’s tool assesses the risk of bias in RCTs based on seven parameters: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other biases. For each parameter, we adjudged a study as having a low risk of bias (+), high risk of bias [-], or unclear risk of bias (?). ## Data extraction and data items We extracted the following data items from the selected articles using a structured data-extraction form: author’s name, year of publication, study setting, design, population, country of study, sample size, and patient demographics such as age and sex. Other extracted items were the pharmacologic interventions in the form of oral vitamin D3 or the estimation of serum 25-hydroxyvitamin D (the most reliable reflection of vitamin D level in the body) or DBP level (as a surrogate marker), and the assessment of diarrhea outcomes or incident disease in the study population. We also retrieved the risk of bias in the RCTs. ## Data synthesis We qualitatively synthesized the extracted data to determine if there were differences in disease incidence or outcomes (between the intervention and the control groups) that had statistical significance. We also synthesized data on the serum 25-hydroxyvitamin D or DBP level (as a surrogate marker), to determine the strength of the relationship between vitamin D status and episodes of diarrhea. We assessed the strength of this relationship using the correlation model. We estimated the I2 and tau2 values to assess between-study heterogeneity, with a focus on the differences in the composition of the study populations. We did not conduct a quantitative synthesis (meta-analysis) of study results to provide an overall estimate of the effect of vitamin D status on diarrhea episodes because of differences in study design and endpoints and the estimated values of heterogeneity: I2 = $86.5\%$; $p \leq 0.001$; tau2 = 0.03 (indicating a high heterogeneity across studies). Also, we conducted a subgroup analysis to identify the possible factors responsible for this significant heterogeneity. ## Study characteristics As shown in Table 1, the nine selected full-text articles consist of one RCT [17], three cross-sectional studies [16, 22, 25], two cohort studies [15, 26], two longitudinal/prospective studies [23, 24], and one case-control study [21]. The countries of the studies are located in the Middle East [16, 22], Asia [17, 26], Europe [15, 25], South America [23, 24], and Africa [21]. One study was both hospital-and community-based [17], six studies were hospital-based [15, 16, 21–23, 25], while two were community-based [24, 26]. We evaluated 5,545 participants in the nine studies. They had a variable age and sex distribution. The majority of the participants were under-five children [15, 17, 21, 25, 26], and their ages ranged from 1 to 11 months [17], 12–42 months [15], 6–24 months [26], and 1–5 years [25]. In one study, the mean age was 17.01 ± 14.8 months [21]. Three studies that evaluated school-age children reported mean ages of 8.9 ± 1.6 years [23] and 8 ± 1.6 years [24], and an age range of 2 months-12 years [16]. One study reported equal sex distribution [17]. In contrast, three studies documented male predominance [16, 21, 22], while one noted a female predominance [24]. We applied star rating to items under parameters like selection and comparability of cases and controls (or cohorts) and assessment of outcome/ascertainment of exposure. In Table 2, the quality assessment of eight of the nine studies using the Newcastle-Ottawa Scale shows a star rating of > 7 (high quality) for six studies [15, 16, 21, 22, 25, 26]. We noted a rating of < 7 (low quality) for only two studies [23, 24]. These two studies were longitudinal in design and had no controls. Assessment parameters like selection and comparability of cases and controls (or cohorts) did not apply to these studies, thus precluding star rating on items under these parameters. The adjudged high-quality studies were cohort, cross-sectional, and case-control in design. As shown in Table 2, the methodological quality of the only RCT [17], assessed with the Cochrane Collaboration tool, reveals a low risk of bias under five parameters: random sequence generation, allocation concealment, blinding of participants, and personnel, incomplete outcome data, and selective reporting. For instance, the authors documented evidence of randomization and masking in the trial. Unique identification numbers were individually randomized in fixed blocks of 20 to the vitamin D3 group, while the investigators applied randomization using a computer-generated list to the placebo group. Table 1Characteristics of studies that reported the relationship between vitamin D3 status and childhood infectious diarrheaStudy (first author’s name and year of publication)Country of studyStudy settingStudy population (sample size and age/sex distribution)Study designAluisio et al. [ 17], 2013AfghanistanFive inner-city districts of Kabul/Passive surveillance center at Maiwind Teaching Hospital-$$n = 3046$$-1-11 months-Equal M/F distributionDouble-blind, placebo-controlled, randomized trialThornton et al. [ 23], 2013ColombiaHospital-based setting in Bogotá-$$n = 475$$-Mean (± SD) age: 8.9 ± 1.6 yearsLongitudinal/Prospective studyMileva et al. [ 15], 2014BulgariaDepartment of Infectious Diseases, Medical University of Varna-$$n = 77$$ ($$n = 30$$, group A patients† & $$n = 47$$, group B patients‡)-12-42 monthsCohort studyTalachian et al. [ 22], 2015IranDepartment of Pediatrics, Hazrat-e-Rasoul Akram Hospital, Tehran-$$n = 50$$ ($$n = 25$$ with acute infectious diarrhea & $$n = 25$$ as controls)-6 months-15 years-Mean (± SD) age: 25.9 ± 25.6 monthsM/F ratio: 1.7:1 * & 1.5:1 §Cross-sectional studyBucak et al. [ 25], 2016TurkeyDepartment of Pediatrics, Adıyaman University School of Medicine, Adıyaman-$$n = 137$$ ($$n = 70$$ with rotaviral diarrhea & $$n = 67$$ as healthy controls)-1-5 yearsCross-sectional studyAhmed et al. [ 26], 2016BangladeshCommunity-based setting in the urban community of Mirpur, Dhaka-$$n = 912$$ ($$n = 446$$ normal-weight children & $$n = 466$$ underweight children)-6-24 monthsCohort studyPalframan et al. [ 24], 2018ColombiaCommunity-based setting in the context of Bogotá School Children Cohort-$$n = 540$$-Mean ± SD age: 8 ± 1.6 years-M/F:$48\%$/$52\%$Longitudinal studyMahyar et al. [ 16], 2019IranQazvin Children Hospital, affiliated with Qazvin University of Medical Sciences (Qazvin, Iran)-$$n = 120$$ ($$n = 60$$ with acute bacterial diarrhea & $$n = 60$$ as controls)-2 months-12 years-M/F: $63.3\%$/$36.7\%$ * & $51.6\%$/$48.4\%$§Cross-sectional studyHassam et al. [ 21], 2019TanzaniaMuhimbili National hospital, Dar es Salaam-$$n = 188$$ under-five children ($$n = 47$$, cases $$n = 94$$, sick controls & $$n = 47$$, healthy controls)-Mean ± SD age: 17.01 ± 14.8 months-M/F: $70.2\%$/$29.8\%$ * & $53.2\%$/$46.8\%$§Unmatched case-control studyM, male F, female SD, standard deviation * Case group §Control group †Patients with risk factors for severe diarrhea ‡Patients without risk factors for severe diarrhea Table 2The methodological quality of the nine selected studies using the Newcastle-Ottawa Scale and Cochrane Collaboration toolStudy(Study design)Selection (max. of 5 stars)Comparability (max. of 2 stars)Exposure/outcome (max. of 3 stars)Total (ten stars)†RSG AC BPP BOA IOD SR OB Thornton et al. [ 23](Longitudinal/prospective study)2 stars-1 star3 starsN/AN/AN/AN/AN/AN/AN/AMileva et al. [ 15] (Cohort study)4 stars1 star3 stars8 starsN/AN/AN/AN/AN/AN/AN/ATalachian et al. [ 22], (Cross-sectional study)4 stars1 star2 stars7 starsN/AN/AN/AN/AN/AN/AN/ABucak et al. [ 25] (Cross-sectional study)4 stars2 stars2 stars8 starsN/AN/AN/AN/AN/AN/AN/AAhmed et al. [ 26] (Cohort study)3 stars1 star2 stars7 starsN/AN/AN/AN/AN/AN/AN/APalframan et al. [ 24], (Longitudinal study)2 stars-1 star3 starsN/AN/AN/AN/AN/AN/AN/AMahyar et al. [ 16] (Cross-sectional study)3 stars2 stars2 stars7 starsN/AN/AN/AN/AN/AN/AN/AHassam et al. [ 21] (Case-control study)3 stars2 stars2 stars7 starsN/AN/AN/AN/AN/AN/AN/AAluisio et al. [ 17], (Randomized control trial)N/AN/AN/AN/A (+) (+) (+) (?) ( +) (+) (?) †Total rating of ≥ 7 stars and < 7 stars suggests high methodological quality and low methodological quality, respectivelyN/A, not applicable max., maximum Parameters of the Cochrane Collaboration tool for randomized control trialsRSG, Random Sequence Generation AC, Allocation Concealment BPP, Blinding of Participants and Personnel BOA, Blinding of Outcome Assessment IOD, Incomplete Outcome Data SR, Selective Reporting OB, Other BiasKey to the risk of bias assessment: low risk of bias (+), high risk of bias [-], or unclear risk of bias (?) The study personnel and participants’ families were blinded to the treatment group to which the participants were assigned. There was an unclear risk of bias under parameters like blinding of outcome assessment and other biases (such as attrition bias). Specifically, the evaluation of diarrheal outcome involved caregivers’ recall of defecation history based on the 24 h preceding each outcome assessment visit, whereas the estimation of 25-hydroxyvitamin D levels was based on samples collected from randomly selected subsets of participants [17]. Additionally, $82.3\%$ of the participants after the trial remained in follow-up with no significant difference in attrition between the vitamin D3 and placebo arms. ## Study findings Table 3 A and 3B summarize the key findings of the nine reviewed studies. In the RCT by Aluisio et al., the authors aimed to evaluate the effects of quarterly supplementation with 100 000 IU of vitamin D3 (cholecalciferol) on the risk for recurrent diarrheal illnesses among children [17]. They randomized 3046 infants who received either oral vitamin D3 ($$n = 1524$$) or placebo ($$n = 1522$$) at 3-month intervals and followed up for 18 months. The study endpoints were diarrhea episodes (based on the WHO definition of diarrhea of ≥ 3 loose/liquid stools in 24 h). They noted incidences of diarrheal episodes of 3.43 ($95\%$ CI, 3.28–3.59) and 3.59 per child-year ($95\%$ CI, 3.44–3.76) in the placebo and oral vitamin D3 arms, respectively. Furthermore, the authors observed no effect on the risk for recurrent diarrheal disease in either intention-to-treat or per-protocol analyses (Table 3 A). Thus, they concluded that quarterly supplementation with vitamin D3 conferred no reduction in the risk of recurrent diarrheal disease [17]. The longitudinal study by Thornton et al. investigated the association of vitamin D status with gastrointestinal and ear infections in school-age children [23]. The authors determined the baseline vitamin D status of randomly selected children ($$n = 475$$) by estimating their plasma 25-hydroxyvitamin D levels and followed them up for an academic year. Interestingly, they found that vitamin D deficiency was associated with increased rates of diarrhea with vomiting (adjusted incidence rate ratio: 2.05; $95\%$ CI: 1.19, 3.53) and earache/discharge with fever (adjusted incidence rate ratio: 2.36; $95\%$ CI: 1.26, 4.44). These findings suggest an inverse relationship between vitamin D status and gastrointestinal/ear infections (Table 3 A). In another longitudinal study by Palframan et al., the investigators evaluated the associations between vitamin D binding protein (DBP) and gastrointestinal/respiratory infections in 540 school-age children [24]. DBP is a surrogate marker of vitamin D. They also examined whether such associations could be mediated through 25-hydroxyvitamin D (Table 3B). Plasma DBP and 25 hydroxyvitamin D were estimated at participants’ enrolment, followed by daily documentation of the infectious morbidity symptoms during the school year. The study endpoints were the rates of gastrointestinal and respiratory morbidity (i.e., the number of days of diarrhea with vomiting, cough with fever, and earache/ear discharge with fever divided by the number of days of observation). The authors found that DBP was inversely associated with the rates of diarrhea with vomiting and earache/ear discharge with fever. However, DBP-morbidity associations were not mediated through 25-hydroxyvitamin D. The two cohort studies by Mileva et al. [ 15] (Table 3 A) and Ahmed et al. [ 26] (Table 3B) reported divergent findings. The former aimed to determine the vitamin D status in toddlers with acute diarrhea and to assess its relationship with diarrhea severity. The authors assayed circulating 25-hydroxyvitamin D3 levels in two groups of patients: Group A, with risk factors for severe diarrhea ($$n = 30$$), and Group B, without risk factors ($$n = 47$$). Diarrhea severity (i.e., more than 20 diarrheal stools per day) was the study outcome. They noted that patients in Group A were vitamin-D insufficient (median = 53.63 nmol/L) compared to those in Group B (median = 66.09 nmol/L). Vitamin D deficiency (median = 49.20 nmol/L) occurred in children with severe diarrhea (> 20 diarrheal stools) compared to vitamin D status in children (median = 64.93 nmol/L) with less severe diarrhea [15]. On the other hand, Ahmed et al. evaluated the association of vitamin D status with diarrhea episodes caused by Enterotoxigenic Escherichia coli (ETEC), Enteropathogenic Escherichia coli (EPEC), and Enteroaggregative Escherichia coli (EAEC) among underweight and normal-weight children (after controlling for other micronutrients status and household/socio-economic variables). At the enrolment of 912 study participants ($$n = 446$$ normal-weight children and $$n = 466$$ underweight children), the authors determined their serum vitamin D and another micronutrient status and isolated and characterized the causative organisms in stool samples collected during a diarrheal episode. ETEC, EPEC, and EAEC in diarrheal stool samples tested during five months of follow-up constituted the study outcomes. They found that vitamin D status was not independently associated with the risk of incident ETEC, EPEC, and EAEC diarrhea in underweight children. However, insufficient vitamin D status and moderate-to-severe retinol deficiency were associated with $44\%$ and $38\%$ reduced risk of incident EAEC diarrhea among normal-weight children [26]. The three cross-sectional studies by Mahyar et al. [ 16] (Table 3B), Talachian et al. [ 22], and Bucak et al. [ 25] (Table 3 A) reported similar findings. The study by Mahyar et al. aimed to determine the correlation between serum 25-hydroxyvitamin D and acute bacterial diarrhea in children [16]. The researchers estimated serum 25-hydroxyvitamin D levels in children with diarrhea ($$n = 60$$) and the control group ($$n = 60$$). They observed a significant difference between the mean ± SD of 25-hydroxyvitamin D levels in children with acute bacterial diarrhea (19.3 ± 7.8 ng/ml) and the control group (22.4 ± 7.3 ng/ml). Talachian et al. compared the serum levels of zinc, vitamins A, and D in children with infectious diarrhea with a control group by measuring and comparing baseline serum vitamin A, 25-hydroxyvitamin D3, and zinc levels in 25 children admitted with acute diarrhea and 25 children without the infection [22]. They found significantly lower 25-hydroxyvitamin D3 levels in the diarrhea group but no significant difference in vitamin A and zinc levels between the diarrhea and the control groups. In their study, Bucak et al. also compared serum 25-hydroxyvitamin D3 levels of hospitalized preschool children with rotaviral diarrhea ($$n = 70$$) with healthy controls ($$n = 67$$) [25]. The study interventions involved measuring and comparing serum levels of 25-hydroxyvitamin D3, parathormone, calcium, phosphate, alkaline phosphatase, complete blood count parameters, and C-reactive protein of the preschool children with rotaviral diarrhea and the controls without the infection. Using serum levels of 25-hydroxyvitamin D3 as their study endpoint, they noted significant differences between the mean serum 25-hydroxyvitamin D3 levels (14.6 ± 8.7 ng/mL) of rotaviral diarrhea patients and healthy controls (29.06 ± 6.51 ng/mL). Finally, the case-control study by Hassam et al. aimed to determine the association between vitamin D levels and diarrhea in children under five years old [21]. The authors estimated serum vitamin D levels in children with diarrhea ($$n = 47$$), sick controls ($$n = 94$$), and healthy controls ($$n = 47$$). They categorized vitamin D status as vitamin D sufficient, insufficient, or deficient. Association between vitamin D status and diarrhea was taken as the primary outcome, while associations between diarrhea and independent variables were the secondary outcome. Despite the high prevalence of vitamin D deficiency in the participants, sick controls were 3.2 times and 5.03 times more likely to be vitamin D deficient than healthy controls. They also found that children with vitamin D deficiency were less likely to have diarrhea than those without vitamin D deficiency (Table 3B). Table 3(A) Major findings of the studies reporting the relationship between vitamin D3 and childhood infectious diarrhea (studies published between 2013 and 2016)Study (first author’s name and year of publication)Study aimsStudy interventionsStudy outcomes/endpointsMajor findingsAluisio et al. [ 17], 2013-To assess the effects of quarterly supplementation with 100 000 IU of vitamin D3 (cholecalciferol) on children’s risk for recurrent diarrheal illnesses.-Randomization of recruited infants to receive either oral vitamin D3 ($$n = 1524$$) or placebo ($$n = 1522$$) at 3-month intervals and followed for 18 months-Diarrhea episodes -The incidences of diarrheal episodes of 3.43 ($95\%$ CI, 3.28–3.59) and 3.59 per child-year ($95\%$ CI, 3.44–3.76) in the placebo and intervention arms, respectively.-No effect on the risk for recurrent diarrheal disease in either intention-to-treat or per-protocol analysesThornton et al. [ 23], 2013-To investigate the association of vitamin D status with gastrointestinal and ear infections in school-age children-Measurement of plasma 25-hydroxy-vitamin D levels in a random sample of children ($$n = 475$$) to determine their baseline vitamin D status. They were followed up for an academic year-Incidence rate ratios & $95\%$ CI for days with diarrhea, vomiting, diarrhea with vomiting, cough with fever, and earache or discharge with fever. †-Vitamin D deficiency,‡ associated with increased rates of diarrhea with vomiting (adjusted incidence rate ratio: 2.05; $95\%$ CI: 1.19, 3.53) and earache/discharge with fever (adjusted incidence rate ratio: 2.36; $95\%$ CI: 1.26, 4.44)Mileva et al. [ 15], 2014-To determine the vitamin D status in toddlers with acute diarrhea and evaluate its relationship with diarrhea severity-Assay of circulating 25-hydroxyvitamin D levels in two groups of patients: Group A, with risk factors for severe diarrhea ($$n = 30$$), and Group B, without risk factors ($$n = 47$$)-Diarrhea severity§-Patients in Group A were vitamin-D insufficient (median = 53.63 nmol/L), compared to those in Group B (median = 66.09 nmol/L).-Vitamin D deficiency (median = 49.20 nmol/L) was detected in children with severe diarrhea compared to vitamin D status in children (median = 64.93 nmol/L) with less severe diarrheaTalachian et al. [ 22], 2015-To compare the serum levels of zinc, vitamins A, and D in children with infectious diarrhea with a control group-Measurement and comparison of baseline serum vitamin A, 25-hydroxyvitamin D, and zinc levels in 25 children admitted with acute diarrhea and 25 children without the infection-Serum levels of 25-hydroxyvitamin D, vitamin A, and zinc-Significantly lower 25-hydroxyvitamin D levels in the diarrhea group-No significant difference in the levels of vitamin A and zinc between diarrhea and control groupsBucak et al. [ 25], 2016-To compare serum 25-hydroxyvitamin D level of hospitalized preschool children with rotaviral diarrhea with that of healthy controls-Measurement and comparison of serum levels of 25-hydroxyvitamin D, parathormone, calcium, phosphate, alkaline phosphatase, complete blood count parameters, and C-reactive protein of preschool children with rotaviral diarrhea and controls without the infection-Serum levels of 25-hydroxyvitamin D-Significant differences between the mean serum 25-hydroxyvitamin D levels of rotaviral diarrhea patients (14.6 ± 8.7 ng/mL) and healthy controls (29.06 ± 6.51 ng/mL).¶ Based on the WHO definition of diarrhea (≥ 3 loose/liquid stools in 24 h) CI, confidence interval †Estimates adjusted for child’s age, sex, and household socio-economic status ‡Vitamin D status classified according to 25 hydroxyvitamin D3 levels as deficient (< 50 nmol/L), insufficient (≥ 50 and < 75 nmol/L) or sufficient (≥ 75 nmol/L) §Above 20 diarrheal stools were considered severe. ¶Serum 25-hydroxyvitamin D3 < 20 ng/mL was associated with rotaviral diarrheaN/B: 1 ng/mL is equivalent to 2.5 nmol/L Table 3(B) Major findings of the studies reporting the relationship between vitamin D3 and childhood infectious diarrhea (studies published between 2016 and 2019)Study (first author’s name and year of publication)Study aimsStudy interventionsStudy outcomes/endpointsMajor findingsAhmed et al. [ 26], 2016-To evaluate the association of vitamin D status (controlling for other micronutrients status and household/socio-economic variables) with ETEC, EPEC, and EAEC diarrhea episodes among underweight and normal-weight children-Determination of serum vitamin D and another micronutrient status at enrolment-Isolation and characterization of causative organisms in stool samples collected during a diarrheal episode-ETEC, EPEC, and EAEC in diarrheal stool samples tested during five months of follow-up-Vitamin D status was not independently associated with the risk of incident ETEC, EPEC, and EAEC diarrhea in underweight children, but moderate-to-severe retinol deficiency was associated with reduced risk for EPEC diarrhea (upon adjustment).-Insufficient vitamin D status and moderate-to-severe retinol deficiency were independently associated with $44\%$ and $38\%$ reduced risk of incident EAEC diarrhea, respectively, among normal-weight childrenPalframan et al. [ 24], 2018-To investigate the associations between DBP andinfectious morbidity among school-age children -To examine whether any associations between DBP and morbidity could be mediated through 25-hydroxyvitamin D-Estimation of plasma DBP and 25-hydroxyvitamin D at enrolment of subjects-Daily documentation of infectious morbidity symptoms during the school year-Rates of gastrointestinal and respiratory morbidity†-DBP was inversely associated with the rates of diarrhea with vomiting and ear-ache/ear discharge with fever-DBP–morbidity associations were not mediated through 25-hydroxyvitamin D.Mahyar et al. [ 16], 2019-To determine the correlation between serum 25-hydroxyvitamin D and acute bacterial diarrhea in children-Estimation of serum 25-hydroxyvitamin D levels in children with diarrhea and control group-Vitamin D status of study participants-Significant difference between the mean ± SD of 25-hydroxyvitamin D levels in the case group (19.3 ± 7.8 ng/ml) and control group (22.4 ± 7.3 ng/ml)Hassam et al. [ 21], 2019-To determine the association between vitamin D levels and diarrhea in under-five children.-Estimation of serum vitamin D levels in children with diarrhea‡-Association between vitamin D status and diarrhea (primary outcome)-Associations between diarrhea and independent variables (secondary outcome)-Children with vitamin D deficiency were less likely to have diarrhea as compared to children without vitamin D deficiencyETEC, Enterotoxigenic Escherichia coliEPEC, Enteropathogenic Escherichia coliEAEC, Enteroaggregative Escherichia coliDBP, vitamin D-binding protein Gastrointestinal and respiratory infections †The number of days of diarrhea with vomiting, cough with fever and earache/ear discharge with fever divided by the number of days of observation ‡Categorized as vitamin D sufficient, insufficient or deficientN/B: 1 ng/mL is equivalent to 2.5 nmol/L Given the inconsistencies noted in the findings of studies that evaluated the relationship between vitamin D status and diarrhea episodes, the strength of this relationship was assessed using a correlation graph. In Fig. 2, the scatter graph shows no significant correlation between the two variables in six studies [15, 16, 21–23, 25]. Assuming the covariance (X, Y) = 0 (from the pattern of the scatter graph), the Pearson correlation coefficient (r) was thus estimated to be 0, underscoring the absence of correlation between the two variables. All the six studies assessed vitamin D status by the quantitative estimation of serum vitamin D levels and adopted the conventional classification of vitamin D status: normal status (75–125 nmol/L), insufficiency (50–75 nmol/L), and deficiency (< 50 nmol/L). Fig. 2Scatter graph showing the nature of the correlation between vitamin D status and diarrhea in children ## Subgroup analysis on association of vitamin D status with diarrhea risk We categorized the study participants into age groups and analyzed the diarrhea risk with respect to their vitamin D status or vitamin D supplementation. In Table 4, vitamin D supplementation in infants showed no effect in reducing diarrhea risk. In the same age group, vitamin D deficiency and insufficiency were associated with increased and reduced diarrhea risk, respectively. Whereas vitamin D deficiency was associated with increased and decreased diarrhea risk in preschoolers, vitamin D insufficiency was associated with decreased diarrhea risk in the same age group. In school-aged children and adolescence, vitamin D deficiency was associated with increased diarrhea risk. Thus, the different outcomes of vitamin D status among infants and preschool-age children may partly explain the apparent non-correlation of vitamin D deficiency (independent variable) with diarrhea disease (dependent variable). Table 4Effect of age group on the relationship between vitamin D status and diarrhea riskAge groupVitamin D status/Vitamin D supplementationDiarrhea riskStudy-Infancy (1–12 months)-Vitamin D supplementation-Vitamin D deficiency -Vitamin D insufficiency†-No effect in reducing risk-Increased diarrhea* risk-Increased diarrhea risk-Decreased diarrhea risk§-Aluisio et al. [ 17]-Talachian et al. [ 22]-Mahyar et al. [ 16]-Ahmed et al. [ 26]-Preschool age (1–5 years)-Vitamin D deficiency -Vitamin D insufficiency †-Increased diarrhea* risk-Increased diarrhea risk-Increased diarrhea risk-Increased diarrhea risk-Decreased diarrhea risk-Decreased diarrhea risk§-Mileva et al. [ 15]-Talachian et al. [ 22]-Bucak et al. [ 25]-Mahyar et al. [ 16]-Hassam et al. [ 21]-Ahmed et al. [ 26]-School age(6–12 years)-Vitamin D deficiency -Increased diarrhea* risk-Increased diarrhea risk-Increased diarrhea risk-Thornton et al. [ 23]-Talachian et al. [ 22]-Mahyar et al. [ 16]-Adolescence (13–18 years)-Vitamin D deficiency -Increased diarrhea* risk-Talachian et al. [ 22]* Serum vitamin D level < 50 nmol/L †Serum vitamin D level = 50–75 nmol/L §in normal-weight subjects ## Discussion Some studies over the past decade report that oral vitamin D3 may ameliorate diarrhea-associated morbidity in children. Others have documented a possible correlation between low serum vitamin D levels and diarrhea episodes. Furthermore, there is a paucity of systematic reviews/meta-analyses on the role of vitamin D3 as a therapeutic adjunct in childhood infectious diarrhea. We initiated the present systematic review because of the lack of consensus in the literature. In this review, we found that most of the studies indicate that vitamin D deficiency was associated with an increased risk of infectious diarrhea [15, 16, 22, 23, 25]. In contrast, DBP level was inversely related to rates of infectious diarrhea and respiratory infections [24]. These findings are consistent with several other studies that indicate a potential protective effect of vitamin D on infectious morbidity [27–32]. We suggest that these observations are predicated on the mechanistic actions of vitamin D3 in innate immunity. Calcitriol (active vitamin D3) levels are regulated by the antagonistic activities of the enzymes CYP27B1 and CYP24A1, which respectively increase and decrease calcitriol levels [33]. Once pathogens come in contact with the gut mucosa, they are recognized by toll-like receptors on macrophages resulting in the receptors’ immunologic activation: aiding intracellular expression of CYP27B1 and vitamin-D receptor (VDR) genes [34]. CYP27B1 produces calcitriol from adequate levels of 25-hydroxyvitamin D in the cytoplasmic matrix. The binding of calcitriol to VDR triggers the production of several endogenous antimicrobial peptides (AMPs), such as cathelicidin and β-defensin, which are widely expressed in the gastrointestinal tract [35, 36]. This calcitriol-VDR interaction also up-regulates nitric oxide (NO) synthase [37]. This pathophysiologic cascade of events explains why vitamin D deficiency may be associated with deranged innate immunity and thus increased susceptibility to intracellular pathogens etiologically linked to diarrhea. Whereas AMPs inhibit bacterial, viral, and fungal infections [38, 39], NO synthase complements bactericidal activity by up-regulating the oxidative burst in macrophages [40]. The clinical-practice implication for this finding is that improving the vitamin D status of children can serve as an ‘immunologic boost’ for them to withstand infectious diarrheas. We also found that the only interventional study in our systematic review failed to demonstrate any beneficial role of vitamin D3 supplementation in reducing diarrhea morbidity [17]. Three-monthly supplementation of high-dose vitamin D3 (100,000 IU) did not confer protection against the risk of recurrent diarrhea. Similarly, an observational analytical study of two cohorts (underweight and normal-weight children) showed no relationship between vitamin D status and the risk of incident ETEC, EPEC, and EAEC diarrhea in underweight children [26]. However, the investigators noted that vitamin D insufficiency was associated with a reduced risk of incident EAEC diarrhea in children with normal weight [26]. Again, a case-control study observed that serum vitamin D levels were not explicitly associated with diarrhea in a population of under-five children [21]. These findings are in tandem with those of a previous systematic review of four trials which did not establish apparent differences between vitamin D-supplemented and-unsupplemented children regarding episodes of diarrhea [41]. The review concluded that vitamin D supplementation was not beneficial in reducing the incidence of childhood diarrhea. Although these observations are inconsistent with the findings of the previously-mentioned related studies [15, 16, 22–25, 27–32], some unidentified factors can explain this disparity. Our subgroup analysis identified age group as a possible factor. Age categorization on vitamin D status and diarrhea risk revealed divergent study outcomes in infants and preschoolers unlike in school-age and adolescent children. Specifically, vitamin D insufficiency was associated with decreased diarrhea risk in infancy and preschool age group. In contrast, vitamin D deficiency was associated with both increased and reduced diarrhea risk in the same age groups. Although the reason for these heterogeneous outcomes is not clear, we speculate that the age-related changes in the gut microbiota may be contributory. The diversity of gut microbiota is higher in adulthood than in childhood although interpersonal differences are higher in the latter than in the former [42]. Again, the gut microbiota assumes adult-like configuration during the first three years of life by which time the gut epithelium and mucosal barrier that it secretes provides a barrier against pathogenic micro-organisms [43, 44]. Dietary alteration may lead to changes in both the composition and diversity of gut microbiota [45]. For instance, formula feeding (and other factors like antibiotic use and caesarean section) may disrupt the composition of the gut microbiota [46]. In fact, the gut microbiota of formula-fed infants are more diverse than those of their breastfed counterparts [47], while children treated with antibiotics have less stable and less diverse flora [48]. Interestingly, some authors report that with age and in obesity, the metabolic activation of vitamin D3 (with the production of calcitriol) is reduced by hepatic steatosis and dysbiosis of the microbiota [49, 50]. The activation process by 25-hydroxylation occurs in the liver via the cytochrome P450 system and in the gut microbiome [51]. Thus, the reduced diarrhea risk reported among vitamin D-insufficient under-five children may be attributed to the protective effect of the diverse composition of their gut microbiota. On the other hand, the increased diarrhea risk noted among their vitamin D-deficient cohorts may be due to the reduced bioavailability of calcitriol. Decreased calcitriol levels follow poor vitamin D activation as a result of dysbiosis of the gut microbiota. The hypothesis appears validated by the fact that the study that observed the association of vitamin D insufficiency with decreased diarrhea risk reported this finding among normal-weight children [26]. Given the less mature and less diverse gut microbiota in malnourished than in normal-weight children [52, 53], it is not surprising that the latter’s gut microbiota composition could have been protective against diarrhea pathogens. Although some authors suggest that a strong relationship between vitamin D status and diarrhea does exist, it may be masked by several other variables identified in interventional studies [54]. Firstly, the serum level of 25-hydroxyvitamin D required for calcium homeostasis and innate immunity varies. While there are existing standard recommendations of daily vitamin D needed to achieve calcium homeostasis, it is still challenging to predict the dose and duration of vitamin D that would optimize its non-calcemic or immunologic effects [55]. Although vitamin D administered in different frequencies (i.e., daily, weekly, or monthly) can maintain similar serum levels of 25-hydroxyvitamin D over an equivalent time frame [56], there is a strong possibility that poor adherence with daily vitamin D administration may result in insufficient vitamin D levels and suboptimal effects. Worse still, some children’s pre-morbid vitamin D status in some settings is deficient. For instance, a recent systematic review and meta-analysis comparing the pooled prevalence of vitamin D deficiency among poor and sick children in sub-Saharan Africa revealed a higher prevalence among healthy children [57]. Again, administering vitamin D2 (ergocalciferol) is adjudged less effective than vitamin D3 (cholecalciferol) at raising the serum levels of 25-hydroxyvitamin D [58]. Thus, differences in dosing strategies and the type of vitamin D may contribute to the disparities in the outcomes of trials on its effectiveness in childhood infectious diarrhea. Secondly, genetic variations of DBP (the major carrier protein for serum 25-hydroxyvitamin D) may play a role in the inconsistencies in study findings [59]. Some authors report that DBP polymorphisms may determine the amount of bioavailable serum 25-hydroxyvitamin D and therefore be more reflective of actual vitamin D status than total serum 25-hydroxyvitamin D [60]. Interestingly, in one of the studies evaluated in the present systematic review, DBP was inversely associated with gastrointestinal and respiratory infections, whereas these morbidity associations were not mediated through 25-hydroxyvitamin D [24]. Likely, these genetic variations could also mask the effects of vitamin D in some populations [54]. Finally, baseline 25-hydroxyvitamin level and VDR polymorphisms in study participants are also possible contributors to the disparities in the present review’s findings. The effectiveness of vitamin D in deficient subjects may be partly related to the inverse relationship between baseline 25-hydroxyvitamin D level and response to vitamin D administration [58]. Baseline vitamin D- sufficient individuals achieve a lesser elevation in 25-hydroxyvitamin D level than their deficient counterparts receiving vitamin D supplementation. Thus, studies with participants whose vitamin D status falls outside the range where the effects on infectious outcomes are obtainable may fail to show an improvement following supplementation [54]. Furthermore, some investigators have demonstrated that variants of VDR can affect response to vitamin D supplementation [61]. Their observation underscores the fact that VDR polymorphisms can also explain the inconsistent findings regarding the effectiveness of vitamin D supplementation as a therapeutic adjunct in infectious diarrheas. The present systematic review has some limitations. The high between-study heterogeneity across the included studies precluded a quantitative synthesis (meta-analysis) of the overall effect of the study results. Additionally, most of our selected studies were non-interventional in nature, as there was no direct assessment of the impact of vitamin D supplementation on serum 25-hydroxyvitamin D levels. The studies evaluated relationships between participants’ vitamin D status and diarrhea morbidity outcomes. The high prevalence of vitamin D deficiency among healthy children in some settings [57] may be a confounder to the association of vitamin D status with infectious morbidity. Thus, non-recognition of this confounding variable will affect the generalizability of the study findings linking vitamin D deficiency with increased diarrhea risk. ## Conclusion This systematic review has shown that vitamin D supplementation is not effective in reducing the risk of childhood infectious diarrhea. Although the association of vitamin D deficiency with infectious diarrhea risk (as demonstrated in three cross-sectional studies [16, 22, 25], one cohort study [15], and one longitudinal study [23]) suggested a possible relationship between vitamin D status and risk of gastrointestinal infections, another longitudinal study [24], one cohort study [26], and the case-control study [21] reported contrary findings. Over all, evaluating the strength of this relationship by correlation model showed a weak association between the two variables in six of the non-randomized studies [15, 16, 21–23, 25]. Nevertheless, the possibility of a strong relationship is supported by the well-documented role of calcitriol in innate immunity. When this non-calcemic action is attenuated, gut AMPs are not produced, resulting in the risk of infectious diarrhea. For future research direction, we recommend more adequately-powered RCTs on oral vitamin D’s role in reducing diarrhea risk. Such interventional studies should also control for potential confounding variables in the study population such as age group, DBP and VDR polymorphisms. ## References 1. Streatfield PK, Khan WA, Bhuiya A. Cause-specific childhood mortality in Africa and Asia: evidence from INDEPTH health and demographic surveillance system sites. Glob Health Action (2014.0) 7 25363. DOI: 10.3402/gha.v7.25363 2. 2.Bhutta ZA, Saeed MA. Childhood infectious diseases: overview. Int Encycloped Pub Health. 2008:620–40 3. 3.World Health Organization. Diarrheal disease. Available from: https://www.who.int/news-room/fact-sheets/detail/diarrhoeal-disease (Accessed 15 August 2022) 4. 4.World Health Organization. (2009). Stuart MC, Kouimtzi M, Hill SR, editors. WHO Model Formulary 2008. World Health Organization. pp. 349 – 51 5. Munos MK, Walker CL, Black RE. The effect of oral rehydration solution and recommended home fluids on diarrhea mortality. Int J Epidemiol (2010.0) 39 75-87. DOI: 10.1093/ije/dyq025 6. Bajait C, Thawani V. Role of zinc in pediatric diarrhea. Indian J Pharmacol (2011.0) 43 232-5. DOI: 10.4103/0253-7613.81495 7. 7.Zinc supplementation helps diarrhea symptoms. Available from: http://www.newsmedical.net/news/2008/02/04/ 34888.aspx (Accessed 19 August 2022) 8. Shankar AH, Prasad AS. Zinc and immune function: the biological basis of altered resistance to infection. Am J Clin Nutr (1998.0) 68 447S-63S. DOI: 10.1093/ajcn/68.2.447S 9. Bahl R, Bhandari N, Hambidge KM, Bahn MK. Plasma zinc as a predictor of diarrheal and respiratory morbidity in children in an urban slum setting. Am J Clin Nutr (1998.0) 68 414S-17S. DOI: 10.1093/ajcn/68.2.414S 10. 10.Imdad A, Mayo-Wilson E, Herzer K, Bhutta ZA. Vitamin A supplementation for preventing morbidity and mortality in children from six months to five years of age. Cochrane Database Syst Rev. 2017; 3. Art. No.: CD008524 11. Yurdakök K, Ozmert E, Yalçin SS, Laleli Y. Vitamin a supplementation in acute diarrhea. J Pediatr Gastroenterol Nutr (2000.0) 31 234-7. DOI: 10.1097/00005176-200009000-00006 12. Bikle D. Non-classic actions of vitamin D. J Clin Endocrinol Metab (2009.0) 94 26-34. DOI: 10.1210/jc.2008-1454 13. Bartley J, Vitamin D. Emerging roles in infection and immunity. Expert Rev Anti Infect Ther (2010.0) 8 1359-69. DOI: 10.1586/eri.10.102 14. Gudmundsson GH, Bergman P, Andersson J, Raqib R, Agerberth B. Battle and balance at the mucosal surfaces-the story of Shigella and antimicrobial peptides. Biochem Biophys Res Commun (2010.0) 396 116-9. DOI: 10.1016/j.bbrc.2010.03.081 15. Mileva S, Galunska B, Gospodinova M, Gerova D, Svinarov D. Vitamin D3 status in children with acute diarrhea. Integr Food Nutr Metab (2014.0) 1 98-9 16. Mahyar A, Ayazi P, Saffari Rad M. The correlation between vitamin D and bacterial diarrhea in children. Arch Pediatr Infect Dis (2019.0) 7 e84382 17. 17.Aluisio AR, Maroof Z, Chandramohan D, et al. Vitamin D3 supplementation and childhood diarrhea: a randomized controlled trial. Pediatrics. 2013;132(4):e832–40. 18. 18.Moher D, Liberati A, Tetzlaff J, Altman DG, the PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PloS Med. 2009;6:e1000097. 19. 19.Wells G, Shea B, O’Connell D et al. The Newcastle-Ottawa Scale (NOS). Disponibile all’indirizzo (2000). Available from: http://www.ohri.ca/programs/clinical_epidemiology/oxford. 20. Higgins JP, Altman DG, Gøtzsche PC. The Cochrane collaboration’s tool for assessing the risk of bias in randomized trials. BMJ (2011.0) 343 d5928. DOI: 10.1136/bmj.d5928 21. Hassam I, Kisenge R, Aboud S, Manji K. Association of vitamin D and diarrhea in children aged less than five years at Muhimbili national hospital, Dar es Salaam: an unmatched case-control study. BMC Pediatr (2019.0) 19 237. DOI: 10.1186/s12887-019-1614-4 22. Talachian E, Bidari A, Noorbakhsh S, Tabatabaei A, Salari F. Serum levels of vitamins a and D, and zinc in children with acute diarrhea: a cross-sectional study. Med J Islam Repub Iran (2015.0) 29 207. PMID: 26157725 23. Thornton KA, Marín C, Mora-Plazas M, Villamor E. Vitamin D deficiency associated with increased incidence of gastrointestinal and ear infections in school-age children. Pediatr Infect Dis J (2013.0) 32 585-93. DOI: 10.1097/INF.0b013e3182868989 24. Palframan KM, Robinson SL, Mora-Plazas M, Marin C, Villamor E. Vitamin D-binding protein is inversely associated with the incidence of gastrointestinal and ear infections in school-age children. Epidemiol Infect (2018.0) 146 1996-2002. DOI: 10.1017/S0950268818002066 25. Ahmed AM, Soares Magalhaes RJ, Long KZ. Association of vitamin D status with incidence of Enterotoxigenic, Enteropathogenic, and Enteroaggregative. Trop Med Int Health (2016.0) 21 973-84. DOI: 10.1111/tmi.12731 26. Zacharioudaki M, Messaritakis I, Galanakis E. Vitamin D receptor, vitamin D binding protein, and CYP27B1 single nucleotide polymorphisms and susceptibility to viral infections in infants. Sci Rep (2021.0) 11 13835. DOI: 10.1038/s41598-021-93243-3 27. Wong KK, Lee R, Watkins RR, Haller N. Prolonged. J Parenter Enteral Nutr (2016.0) 40 682-7. DOI: 10.1177/0148607114568121 28. Abdelfatah M, Nayfe R, Moftakhar B. Low vitamin D level and impact on severity and recurrence of. J Investig Med (2015.0) 63 17-21. DOI: 10.1097/JIM.0000000000000117 29. Bucak IH, Ozturk AB, Almis H. Is there a relationship between low vitamin D and rotaviral diarrhea?. Pediatr Int (2016.0) 58 270-3. DOI: 10.1111/ped.12809 30. Urashima M, Segawa T, Okazaki M, Kurihara M, Wada Y, Ida H. Randomized trial of vitamin D supplementation to prevent seasonal influenza A in school children. Am J Clin Nutr (2010.0) 91 1255-60. DOI: 10.3945/ajcn.2009.29094 31. Wayse V, Yousafzai A, Mogale K, Filteau S. Association of subclinical vitamin D deficiency with severe acute lower respiratory infection in indian children under 5 years. Eur J Clin Nutr (2004.0) 58 563-7. DOI: 10.1038/sj.ejcn.1601845 32. Zhou J, Du J, Huang L, Wang Y, Shi Y, Lin H. Preventive effects of vitamin D on seasonal influenza A in infants: a multicenter, randomized, open, controlled clinical trial. Pediatr Infect Dis J (2018.0) 37 749-54. DOI: 10.1097/INF.0000000000001890 33. Adams JS, Hewison M. Update in vitamin D. J Clin Endocrinol Metab (2010.0) 95 471-8. DOI: 10.1210/jc.2009-1773 34. Vieth R. How to optimize vitamin D supplementation to prevent cancer, based on cellular adaptation and hydroxylase enzymology. Anticancer Res (2009.0) 29 3675-84. PMID: 19667164 35. Liu PT, Stenger S, Li H, Wenzel L, Tan BH, Krutzik SR. Toll-like receptor triggering of a vitamin D-mediated human antimicrobial response. Science (2006.0) 311 1770-3. DOI: 10.1126/science.1123933 36. Wehkamp J, Schauber J, Stange EF. Defensins and cathelicidins in gastrointestinal infections. Curr Opin Gastroenterol (2007.0) 23 32-8. DOI: 10.1097/MOG.0b013e32801182c2 37. Rockett KA, Brookes R, Udalova I, Vidal V, Hill AV, Kwiatkowski D. 1,25-dihydroxy vitamin D3 induces nitric oxide synthase and suppresses the growth of Mycobacterium tuberculosis in a human macrophage-like cell line. Infect Immun (1998.0) 66 5314-21. DOI: 10.1128/IAI.66.11.5314-5321.1998 38. Ramanathan B, Davis EG, Ross CR, Blecha F. Cathelicidins: microbicidal activity, mechanisms of action, and roles in innate immunity. Microbes Infect (2002.0) 4 361-72. DOI: 10.1016/S1286-4579(02)01549-6 39. Klotman ME, Chang TL. Defensins in innate antiviral immunity. Nat Rev Immunol (2006.0) 6 447-56. DOI: 10.1038/nri1860 40. Sly LM, Lopez M, Nauseef WM, Reiner NE. 1alpha,25-Dihydroxyvitamin D3-induced monocyte antimycobacterial activity is regulated by phosphatidylinositol 3-kinase and mediated by the NADPH-dependent phagocyte oxidase. J Biol Chem (2001.0) 276 35482-93. DOI: 10.1074/jbc.M102876200 41. Yakoob MY, Salam RA, Khan FR, Bhutta ZA. Vitamin D supplementation for preventing infections in children under five years of age. Cochrane Database Syst Rev (2016.0) 11 CD008824. PMID: 27826955 42. Yatsunenko T, Rey FE, Manary M. Human gut microbiome viewed across age and geography. Nature (2012.0) 486 222-27. DOI: 10.1038/nature11053 43. Sommer F, Bäckhed F. The gut microbiota- masters of host development and physiology. Nat Rev Microbiol (2013.0) 11 227-38. DOI: 10.1038/nrmicro2974 44. Faderl M, Noti M, Corazza N, Mueller C. Keeping bugs in check: the mucus layer as a critical component in maintaining intestinal homeostasis. IUBMB Life (2015.0) 67 275-85. DOI: 10.1002/iub.1374 45. Alcock J, Maley CC, Aktipis CA. Is eating behavior manipulated by gastrointestinal microbiota? Evolutionary pressures and potential mechanisms. BioEssays (2014.0) 36 940-9. DOI: 10.1002/bies.201400071 46. Mueller NT, Bakacs E, Combellick J, Grigoryan Z, Dominguez-Bello MG. The infant microbiome development: mom matters. Trends Mol Med (2015.0) 21 109-17. DOI: 10.1016/j.molmed.2014.12.002 47. Fanaro S, Chierici R, Guerrini P, Vigi V. Intestinal microflora in infancy: composition and development. Acta Paediatr (2007.0) 92 48-55. DOI: 10.1111/j.1651-2227.2003.tb00646.x 48. Yassour M, Vatanen T, Siljander H. Natural history of the infant gut microbiome and impact of antibiotic treatment on bacterial strain diversity and stability. Sci Trans Med (2016.0) 8 343-81. DOI: 10.1126/scitranslmed.aad0917 49. Wortsman J, Matsuoka LY, Chen TC, Lu Z, Holick MF. Decreased bioavailability of vitamin D in obesity. Amer J Clin Nutr (2000.0) 72 690-3. DOI: 10.1093/ajcn/72.3.690 50. Ogrodnik M, Jurk D. Senescence explains age-and obesity-related liver steatosis. Cell Stress (2017.0) 1 70-2. DOI: 10.15698/cst2017.10.108 51. Ogrodnik M, Miwa S, Tchkonia T. Cellular senescence drives age-dependent hepatic steatosis. Nat Commun (2017.0) 8 15691. DOI: 10.1038/ncomms15691 52. Jonkers D. Microbial perturbations and modulation in conditions associated with malnutrition and malabsorption. Best Pract Res Clin Gastroenterol (2016.0) 30 161-72. DOI: 10.1016/j.bpg.2016.02.006 53. Million M, Diallo A, Raoult D. Gut microbiota and malnutrition. Microb Pathog (2017.0) 106 127-38. DOI: 10.1016/j.micpath.2016.02.003 54. Kearns MD, Alvarez JA, Seidel N, Tangpricha V. The impact of vitamin D on infectious disease: a systematic review of controlled trials. Am J Med Sci (2015.0) 349 245-62. DOI: 10.1097/MAJ.0000000000000360 55. Shapses SA, Manson JE. Vitamin D and prevention of cardiovascular disease and diabetes: why the evidence falls short. JAMA (2011.0) 305 2565-6. DOI: 10.1001/jama.2011.881 56. Ish-Shalom S, Segal E, Salganik T, Raz B, Bromberg IL, Vieth R. Comparison of daily, weekly, and monthly vitamin D3 in ethanol dosing protocols for two months in elderly hip fracture patients. Clin Endocrinol Metab (2008.0) 93 3430-5. DOI: 10.1210/jc.2008-0241 57. 57.Shaka MF, Hussen Kabthymer R, Meshesha MD, Borde MT. Vitamin D deficiency among apparently healthy children and children with common medical illnesses in Sub-Saharan Africa: A systematic review and meta-analysis. Ann Med Surg (Lond). 2022; 24; 75:103403 58. Trang HM, Cole DE, Rubin LA, Pierratos A, Siu S, Vieth R. Evidence that vitamin D3 increases serum 25-hydroxyvitamin D more efficiently than does vitamin D2. Am J Clin Nutr (1998.0) 68 854-8. DOI: 10.1093/ajcn/68.4.854 59. Fu L, Yun F, Oczak M, Wong BY, Vieth R, Cole DE. Common genetic variants of the vitamin D-binding protein (DBP) predict differences in the response of serum 25-hydroxyvitamin D [25 (OH) D] to vitamin D supplementation. Clin Biochem (2009.0) 42 1174-7. DOI: 10.1016/j.clinbiochem.2009.03.008 60. Powe CE, Karumanchi SA, Thadhani R. Vitamin D-binding protein and vitamin D in blacks and whites. N Engl J Med (2014.0) 370 880-1. PMID: 24571762 61. Wang TJ, Zhang F, Richards JB. Common genetic determinants of vitamin D insufficiency: a genome-wide association study. Lancet (2010.0) 376 180-8. DOI: 10.1016/S0140-6736(10)60588-0
--- title: 'Clinical characteristics and prognostic characterization of endometrial carcinoma: a comparative analysis of molecular typing protocols' authors: - Zihui Yang - Xi Yang - Xinyu Liu - Ke Ma - Yi-Ting Meng - Hong-Fang Yin - Jia Wen - Jiang-Hui Yang - Zeng Zhen - Zong-Hao Feng - Qin-Ping Liao journal: BMC Cancer year: 2023 pmcid: PMC10015692 doi: 10.1186/s12885-023-10706-8 license: CC BY 4.0 --- # Clinical characteristics and prognostic characterization of endometrial carcinoma: a comparative analysis of molecular typing protocols ## Abstract ### Background Endometrial carcinoma (EC) is one of the most common gynecological malignancies in China and globally, accounting for the fourth-prevalent cancer in women. Although numerous studies have confirmed prognostic value of The Cancer Genome Atlas (TCGA) molecular subgroups, it is unclear how they are combined with histological features. The main objective of this study was to compare ProMisE and TCGA classification for the rapid and accurate prediction of prognosis within EC patients, together with the provision of a revised strategy for individualized diagnosis and treatment of patients. ### Methods Within this study, 70 patients with EC from Beijing Tsinghua Changgeng Hospital (affiliated to Tsinghua University) were retrospectively examined between July 2015 and December 2021. Samples were processed for determination of clinical markers, together with ProMisE and TCGA classification. ### Results Comparative analysis across four TCGA types (POLE, Low-CN, High-CN, and MSI-H) and age, was statistically significant (χ²= 7.000, $$p \leq 0.029$$). There was no significant difference observed among the four TCGA types and FIGO stage, vascular invasion and depth of invasion, or lymph node metastasis and tumor area. There was no significant association between the expression of Vimentin, Ki-67, PTEN, MSH2, PAX-8, β-catenin, CD10, ER, PR, P16, MLH1, and PMS2 with the four TCGA types. In addition, p63 expression (χ²= 11.09, $$p \leq 0.029$$) and p53 expression (χ²= 11.585, $$p \leq 0.005$$) were statistically significant. Numerous models demonstrated that patients with POLE mutations and low-CN had higher progression free survival (PFS) and overall survival (OS), whereas those with high-CN had lowest values. The log-rank test revealed that the survival rate of PR-positive and ER-positive patients was significantly higher ($p \leq 0.001$). ### Conclusion Overall, these results can be of additional benefit for clinical applications, in comparison to the ProMisE classification method. In addition, PR, ER, vascular infiltration, hyperlipidemia and atherosclerosis were found to be the key factors affecting EC prognosis. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12885-023-10706-8. The online version contains supplementary material available at 10.1186/s12885-023-10706-8. ## Highlights TCGA molecular classification has advantages over ProMisE classification for accurate diagnosis of endometrial carcinoma. Comparison among the four TCGA types (POLE, Low-CN, High-CN, and MSI-H) and age was statistically significant. Patients with POLE mutations and low-CN had higher PFS and overall survival, whereas those with high-CN had the lowest overall survival periods. ## Background Endometrial carcinoma (EC) is one of the most common gynecological cancers, with statistics demonstrating that EC incidence and mortality rates have increased over the past decade globally. In 2022 alone, 65,950 new cases and 12,550 deaths were recorded in the United States of America (USA), highlighting the alarming increase in this under-studied malignancy [1]. In addition, data from the National Cancer Health Organization ranked EC incidence rate as being second-highest in female reproductive system malignant tumors within China, and reported a persistent increase in its incidence rate. The risk stratification of patients with EC by the national comprehensive cancer network (NCCN) is mainly based upon tumor stage / grade, and histological type [2]. According to the International Federation of Gynecology and Obstetrics (FIGO) staging system, the main treatment for EC patients with high-risk factors of recurrence is surgery, which can be combined with adjuvant chemotherapy or radiotherapy post-surgery [3]. Another promising treatment for patients with EC is immunotherapy, which targets several biological pathways by blocking immune checkpoints. Moreover, programmed cell death-1 (PD-1) and its ligands (PD-L1 or B7-H1) are targeted in EC patients with high microsatellite instability (MSI-H) or mismatch repair (MMR)-deficient tumors, which account for approximately $30\%$ of primary EC patients [4]. However, the clinical application of immunotherapy is relatively limited, and additional clinical trials are required to confirm its therapeutic effect. Despite progress in EC treatment, the survival periods and quality-of-life of patients with EC have not been significantly improved. Patients with similar clinicopathological characteristics occasionally manifest differing disease outcomes, which could reflect the molecular heterogeneity of tumor invasion and metastasis. Therefore, novel disease-typing methods are necessary for clinicians in order to enhance accuracy in diagnosis, treatment, and prognosis. In 2013, a multi-agency project, initiated by the National Cancer Institute and the National Human Genome Institute, analyzed bioinformatics for the cancer genome atlas (TCGA) database, and reported the genomic, transcriptomic and proteomic datasets collected from DNA sequencing, a combined DNA Methylation Reverse Phase Protein Array, together with data from microsatellite instability analysis, among 373 EC cases (306 cases of endometrioid carcinoma and 66 cases of serous / mixed carcinoma) [5]. TCGA study is the largest comprehensive genomic study of EC so far. According to the integrated results of somatic gene mutations, based upon microsatellite instability and somatic copy number change, EC is divided into four genomic types [6]: The ‘super mutation’ group, which has a mutated polymerase and exonuclease domain within the POLE gene, is characterized by a high mutation rate [6]. A recently published study conducted by Jiang and colleagues focused upon specific immunology-based biomarker signatures for varying EC sub-types [7]. Among this study’s outcomes, it was revealed that the POLE-mutant EC sub-type demonstrated peak T-effector and interferon-gamma expression signatures, together with having the least innate anti-PD1 resistance expression profile among all EC sub-types. This sub-type is perceived as clinically indolent [8].The ‘high mutation’ group is characterized by microsatellite instability (MSI-H), typically caused by MutL homolog1 (MLH1) promoter methylation, high mutation rates and a few copy number changes (also known as sporadic MSI-H) [6, 9]. Interestingly, the study conducted by Bellone and colleagues revealed that, in comparison to Lynch-like MSI-H, sporadic MSI-H EC patients experienced reduced CD68 + macrophage infiltrations within tumor mass / stromal regions on administration of pembrolizumab, with such patients also experiencing reduced objective response rate (ORR), progression free survival (PFS) and overall survival (OS) timeframes in this study [9].The ‘low copy number’ (low-CN) group encompass most microsatellite stable grade 1 and grade 2 endometrioid carcinomas and have a low mutation rate [6]. This group is also known as having ‘no specific molecular profile (NSMP), and typically leads to prognoses that are deemed within intermediate levels between POLE-mutation and the high-CN group [10]. However, this EC group can also have widely ranging clinical manifestations – ranging from mild to highly aggressive EC within patients [10].The ‘high-CN group’ have extensive copy number abnormalities, low mutation rates, and recurrent p53 mutations [6]. Among all four TCGA-stratified EC groups, the high-CN group is inherently associated with the poorest prognosis evaluations [11]. This was also reflected during the TCGA study, whereby high-CN EC group manifested across all serous carcinoma patients included in the study, together with approximately $25\%$ of participating patients having International Federation of Gynecology and Obstetrics Grade 3 endometrioid carcinoma [12]. Recently, a novel, validated algorithm-based analytical methodology for screening potential EC cases was developed, formally known as the Proactive Molecular Risk Classifier for Endometrial Cancer (ProMisE) [13]. Such an algorithm-derived protocol utilizes datasets from p53 / mismatch repair (MMR) proteomic immunohistochemistry (IHC) together with DNA polymerase epsilon (POLE) mutation evaluation, and devised in an effort to reduce costings (and consequent patient accessibility) for POLE analyses [13]. However, although ProMisE was found to be effective and with the capacity to be rapidly adopted within medical centers, further validation-based evaluations are warranted prior to incorporation of such an algorithm-based analytical protocol for screening purposes (particularly for molecular stratifications within fertility-sparing clinical scenarios) [14, 15]. Consequently, the authors deem fit to place further focus on specific comparative analyses between the currently adopted TCGA-based molecular typing protocol and the novel algorithm-based ProMisE methodology. Within this study, 70 EC patients admitted to Beijing Tsinghua Changgeng Hospital (affiliated to Tsinghua University) were retrospectively examined (from July 2015 to December 2021). In order to analyze epidemiological data and prognostic survival of patients with various types of EC, the collected paraffin sections were subjected to high-throughput TCGA-based molecular classification. The main aim of this study was to compare possible advantages of TCGA-based molecular classification over the algorithm-based ProMisE classification, thus performing a comprehensive evaluation and consequently providing a theoretical basis for in-depth clinical prognostic analysis for EC patients. The study also comprehended the possible molecular basis of EC treatment / research as a potential clinical guide for treatment. ## Study participants and data collection This study collected the clinical data of 376 patients from the Beijing Tsinghua Changgeng Hospital, suspected of having EC. The data was obtained between July 2015 to December 2021. Following data screening - based on inclusion and exclusion parameters − 70 individuals with diagnosed EC were included in the study. The selection criteria for recruited patients were based upon patients who underwent curettage or total hysterectomy at Beijing Tsinghua Changgeng Hospital, and consequently diagnosed with EC through post-surgery pathological procedures. The histological type (Grade 1–3) of endometrioid carcinoma, serous carcinoma, clear-cell carcinoma, or mixed-cell adenocarcinoma was determined for each participant. The participants did not receive any neo-adjuvant therapy or drug interference which could affect the collection of clinical information. Moreover, specimens with complete immunohistochemical and sequencing information (and without any influencing factors) formed part of study. All participants gave informed consent, while patients receiving neo-adjuvant therapy prior to surgical procedures, or patients who were suffering from other combined pelvic malignancies, were excluded from this study. Participants with insufficient information and data, specimens with damaged immunohistochemical and sequencing information, together with participants reluctant to communicate with post-match medical staff, were all excluded from the study. The data was collected in strict compliance with standard data collection protocols. ## Sample collection and processing DNA extraction from collected tissue samples was performed using QIAamp MinElute® DNA Extraction Kits (QIAGEN™), following manufacturer guidelines. DNA quality was assessed through Qubit® and Nanodrop®. ## Sequencing analysis Genomic DNA (200 ng) from each sample was used in library construction. Post-fragmentation, DNA fragments were end-repaired via an A-Tailing reaction. Subsequently, T4 DNA ligase was used to link the index adapter and insert DNA fragment. Following the manufacturer’s protocol, Streptavidin Magnetic Beads (Streptavidin Sepharose®, Cytiva 17-5113-01, Sigma Aldrich™) were used for purification. ## Library amplification The purified linkage product was amplified using 2X HotStart ReadyMix® and 10X Library Amplification Primer Mix® (KAPA Biosystems™, Massachusetts USA). Amplicons were purified using Streptavidin Magnetic Beads and the quality of the genomic DNA library was assessed on an Agilent™ 2100 Bioanalyzer®, following adjusting of concentration to 3ng/µL. Hybridization was performed according to the standard hybridization protocol [16]. Briefly, a reaction mixture was prepared and subjected to PCR. Streptavidin Magnetic Beads (SA Beads) were used to capture and purify hybridized products, following manufacturer guidelines. The captured library prep was amplified using 2X HotStart ReadyMix® solution and 10X Library Amplification Primer Mix®, following measurement of library concentration. ## TCGA molecular classification The DNA library was sequenced to cover the whole-exon region and partial-fusion mutation-related intron region of 32 genes that are closely related to EC, including four DNA mismatch repair regions, and five specific microsatellite regions. Sequencing was performed using a next generation sequencing (NGS) platform, the Illumina™ HiSeq 2000® Sequencing System, and the reference genome GRCh37 - hg19. Analysis was performed using the standard TCGA Molecular Typing Method, by categorizing the samples into four groups: POLE super-mutation, MSI-H, low-CN, and high-CN. The specific classification process included the detection of mutation status for POLE gene. The presence of mutations within the POLE gene was classified as POLE super-mutation. MSI was sequenced at four gene loci, including MLH1, MSH2, MSH6, and PMS2 within wild-type sample for POLE gene. The presence of mutations in one of these genes was defined to be MSI-H. The presence of p53 gene mutations was regarded as high-CN, while the absence of mutations was classified as low-CN (Fig. 1). Fig. 1Flow chart of ProMisE and TCGA classification / sample processing ## Sample preparation Fresh, dissected tissues (< 3 mm thick) were collected and fixed with $2\%$ paraformaldehyde. Each dissected tissue was embedded in paraffin blocks, which were sliced into 5–8 μm segments. Samples were stained using a standard staining procedure and stored at room temperature until further analysis. Immunohistochemical markers, together with the associated detection criteria, are described in Table S1. ## ProMisE classification Molecular classification involving the ProMisE method included the detection of MMR protein/s by immunohistochemistry, in order to identify patients and screen for Lynch syndrome (Fig. 1). The results directed the patients onto either surgical or treatment decisions. The polymerase enzyme for the individual tumor was evaluated- ɛ (POLE) through exonuclease domain mutation (EDM) sequencing results. Finally, protein 53 (p53) was evaluated through immunohistochemistry, resulting in four subgroups of mutations that included MMR-d, POLE, p53 wild-type (wt), and p53 null / missense mutation (abn). ## Statistical analysis Statistical analysis was performed using a two-sided t-test, and statistical significance was determined by α = 0.05, and p ≤ 0.05. All datasets were expressed as the mean standard deviation (x ± s). Two independent sample t-tests were used for comparison between different groups, and IBM™ SPSS® statistics 26. 0 (Chicago, USA) was used for statistical analyses. ## Baseline clinical characteristics Patient age ranged between 38 and 83 (mean age 60 ± 1.41) years; three with POLE super-mutations, nine with MSI-H, 45 with low-CN, and 13 with high-CN. Patient age within POLE super-mutation, MSI-H, low-CN and high-CN groups ranged between 50 and 79 (mean age 63.00 ± 14.73) years, 45–82 (mean age 61.20 ± 12.13) years, 40–80 (60.02 ± 9.59) years, and 38–83 (62.17 ± 12.14) years, respectively. Patient age among all four groups was significantly different (χ²= 7.000, $$p \leq 0.029$$, Table 1). Table 1Comparison of differing age groups in endometrial cancer (EC) patients, with four TCGA molecular typesCategoryTotal number of casesPole mutantMSI-H typeLow CN typeHigh CN typeχ² valueP valueAge (years)707.0000.029≤ 603425234> 603614229 Menopausal status, clinical manifestations, complications, and other data for each patient ($$n = 70$$) is shown in Table S2. The youngest postmenopausal patient was 39 years old and the oldest was 60 years old (mean age 51.62 ± 4.26 years). The age of post-menopausal patients within MSI-H, low-CN and high-CN groups ranged from 51 to 57 (mean age 54.17 ± 2.48) years, 39–60 (mean age 50.76 ± 4.28) years and 45–60 (mean age 52.33 ± 4.58) years, respectively. Menopausal status of patients was not significantly different among the four groups (χ²= 1.987, $$p \leq 0.624$$). There was no significant difference within clinical manifestations, including abnormal uterine bleeding, abnormal imaging manifestations, vaginal drainage, cervicitis, and hyperlipidemia observed among EC patients with differing TCGA groups ($p \leq 0.05$). The majority of patients with EC had complications, including hypertension, diabetes, atherosclerosis, uterine leiomyoma, latent syphilis, and human papillomavirus. No significant difference between EC patients and the differing TCGA groups was observed ($p \leq 0.05$). Any significant difference between patient body mass index (BMI) (< 28 and ≥ 28), treatment route (laparotomy versus laparoscopy), therapy type (adjuvant versus chemotherapy), carcinoembryonic antigen (CEA) levels, and levels of cancer antigens (i.e., 19 − 9, 15 − 3 and 12 − 5 between the four differing TCGA groups) was not observed. The CEA level was increased in one patient ($2.44\%$) who was in the low-CN group. There was no significant difference in CEA indexes between the two groups (χ²= 3.055, $$p \leq 1.000$$). Similarly, cancer antigens CA-1-5, CA-15-3 and CA-19-0 demonstrated increased levels in several patients. However, there was no difference in CEA indexes between the groups of patients with elevated and normal CA levels. The specific analysis of clinical baseline characteristic data is shown in Table S2. ## Pathological classification Based on pathological classification, the 70 samples were divided into type I EC ($$n = 64$$) and type II EC ($$n = 6$$), according to morphological manifestations and immunohistochemical staining results. The type II EC group included four cases of serous carcinoma, one case of clear cell carcinoma, and one case of endometrial serous papillary carcinoma. The TCGA classification of Type I EC and Type II EC groups is shown in Table S3 and indicated statistical significance between the two groups, based upon pathological classification. Among the 70 patients, there were 64 cases of EC, four cases of serous carcinoma, one case of clear cell carcinoma, and one case of serous papillary carcinoma of the endometrium, which were all classified as type II endometrial carcinoma. Three people died of the disease ($4.29\%$), including one case of endometrioid carcinoma, one case of clear cell carcinoma, and one case of endometrial serous papillary carcinoma. There were 63 cases of progression-free patients ($92.20\%$), and seven cases of recurrence ($6.20\%$), including three cases of endometrioid carcinoma, two cases of serous carcinoma, one case of clear cell carcinoma, and one case of serous papillary carcinoma of endometrium. There were significant differences in the number of mortalities and recurrences due to the disease. Excluding non-diseased mortalities or life-accidents, the four molecular types were compared by log-rank method. The overall survival (OS) of type I EC and type II EC were $98.4\%$ and $66.7\%$ respectively ($p \leq 0.001$). In addition, there was a significant difference between type I endometrial cancer and type II endometrial cancer. The progression free survival (PFS) of type II endometrial carcinoma were $96.9\%$ and $33.3\%$, respectively ($p \leq 0.001$). The survival curve for type II EC was significantly lower, and prognosis was worse than within type I EC (Fig. 2A, B). Fig. 2Overall survival curve ($p \leq 0.001$) (A) and progression free survival curve ($p \leq 0.001$) (B) for patients with type I and type II endometrial cancer ## Pathological grading Based on the TCGA classification of EC, the 70 cases were divided into G1 ($$n = 24$$), G2 ($$n = 36$$) and G3 ($$n = 9$$) grade. One case did not belong to any category and was referred to as an ‘unknown grade’. EC samples were categorized as either POLE super-mutation ($$n = 1$$), MSI-H ($$n = 8$$), low-CN ($$n = 43$$), or high-CN ($$n = 8$$). G3 cases were categorized into POLE super-mutation ($$n = 1$$), MSI-H ($$n = 1$$), low-CN ($$n = 2$$), and high-CN ($$n = 5$$). Statistically significant differences in pathological grades between the two groups (G1 ~ G2, and G3) were observed (χ²= 11.098, and $$p \leq 0.006$$, Table S4). ## Pathological stage FIGO staging was used to determine the pathological stage of EC within participants (Table 2). One case could not be categorized and was regarded as an ‘unknown’. There was no significant difference in FIGO stage and the four differing TCGA molecular types in patients with EC (χ²= 2.947, $$p \leq 0.380$$). Table 2Comparison of differing pathological stages and immunohistochemical markers in endometrial cancer (EC) patients, with four TCGA molecular typesCategoryTotal Number of CasesPole MutantMSI-H TypeLow CN TypeHigh CN TypeΧ² ValueP Value FIGO Staging 692.7680.383I58383512II-IV1100101 PTEN expression 651.8440.613Positive90252Negative5636389 P53 Expression 6511.090.029Positive130247Negative5236394 MLH1 Expression 653.0520.029Positive52253510Negative131381 MSH2 Expression 652.7201.000Positive64384211Negative20010 PMS2 Expression 652.7960.410Positive51253410Negative141391 MSH6 Expression 65--Positive65384311 Pax-8 Expression 651.9150.646Positive2014123Negative4524318 β- Catenin Expression 650.8090.895Positive4326278Negative2212163 CD10 Expression 650.7651.000Positive1602113Negative4936328 ER Expression 652.4470.513Positive57283710Negative81061 PR Expression 654.1140.171Positive5638388Negative91053 P16 Expression 654.5460.193Positive2533154Negative4005287 P63 Expression 6511.5850.005Positive81421Negative57244110 ## Immunohistochemistry Staining results of markers related to the pathological occurrence and EC development were analyzed. These markers included Vimentin, Ki-67, PTCN-2, p53, MLH1, MSH2, PMS2, MSH6, Pax-8, β- Catenin, CD10, p16, ER, PR, p63. The number of samples being positive for each marker, together with positivity rate and corresponding TCGA classification for each sample, is depicted in Table 2. There were no significant variations in Vimentin, Ki-67, PTCN-2, MSH2, PMS2, MSH6, Pax-8, β- Catenin, CD10, ER, and p16 indexes of patients with differing TCGA typing. However, there was a significant difference observed between p53, MLH1, and p63 indexes of patients with differing TCGA typing (Table 3). Table 3Comparison of differing immunohistochemical expression for various tumor markers in endometrial cancer patients, with four TCGA molecular typesCategoryTotal Number of CasesPole MutantMSI-H TypeLow CN TypeHigh CN TypeΧ² ValueP Value Vimentin Expression 651.9080.559Positive5327368Negative121173 Ki-67 Expression 651.1050.837Positive4626299Negative1912142 ## Histopathology The tumor area, vascular infiltration, depth of infiltration, as well as the presence of myometrial invasion, was determined for each case. Individual patient tumor size was measured and classified into TCGA groups (Table S5). There was no significant difference between tumor size and the TCGA groups within EC patients(χ²= 0.536, $$p \leq 1.000$$). Each patient was examined for lymph node metastasis and classified into TCGA groups (Table S6). There was no significant difference in lymph node metastasis with TCGA groups among EC patients (χ²= 1.271, $$p \leq 1.000$$). Among all analyzed samples, 13 patients had vascular infiltration. The pathological data for 12 patients were not displayed or obtained. There was no significant difference between the presence of vascular infiltration and TCGA groups among EC patients (χ²= 1.499, $$p \leq 0.742$$). Myometrial invasion was detected within 46 patients. Seventeen patients had myometrial invasion (> $50\%$). Two cases had unknown information. There was no significant difference in lymph node metastasis between the four TCGA groups of EC patients (χ²= 3.715, $$p \leq 0.277$$). ## TCGA molecular typing of genes with non-pathogenic functions Prognostic analysis was based upon 65 samples, as several samples were not found during immunohistochemistry. The median follow-up time was 1234 days, with no loss of follow-up for all patients. During this time, three patients died ($4.62\%$); two in the high-CN group ($66.67\%$) and one in the low-CN group ($33.33\%$). There were six patients with recurrence ($9.23\%$); one in the MSI-H group ($16.67\%$), one in the low-CN group ($16.67\%$), and four in the high-CN group ($66.67\%$). Excluding non-diseased mortality or life-accidents, the four molecular typing groups were compared through log-rank method. The survival curve was obtained by Kaplan Meier method. The overall survival (OS) of patients with POLE super-mutation, MSI-H, low-CN, and high-CN were $100\%$, $93.3\%$, $100\%$, and $80.0\%$, respectively. There was a significant difference between the four typing groups ($$p \leq 0.005$$). The progression free survival (PFS) of patients with POLE super-mutation, MSI-H, low-CN and high-CN were $92.3\%$, $86.7\%$, $100\%$ and $70.0\%$, respectively ($$p \leq 0.007$$). Patients within POLE mutation and low-CN groups had higher PFS and OS, while patients in the high-CN group had the lowest OS timeframes (Figure S1, S2). ## TCGA molecular typing The four molecular typing groups were compared through log-rank method. The OS of patients with POLE super-mutation, MSI-H, low-CN and high-CN were $100\%$, $100\%$, $97.60\%$ and $84.60\%$, respectively. There was no significant difference between the four TCGA groups ($$p \leq 0.380$$). The PFS of POLE mutant, MSI-H, low-CN, and high-CN were $100\%$, $87.50\%$, $97.50\%$ and $69.20\%$ respectively, and a significant difference among the four groups ($p \leq 0.001$) was observed. Patients within POLE super-mutation and low-CN groups had higher PFS and OS, while patients in the high-CN group had lowest PFS and OS timeframes (Figure S3, S4). ## ProMisE typing Among the 65 patients with TCGA molecular typing, 18 cases were ranked in the MMR-D group ($23.08\%$), two cases in the POLE mutation group ($4.62\%$), 11 cases in the p53abn group ($18.46\%$), and 34 cases in the p53wt group ($53.85\%$). Overall, three people died of the disease; one from the MMR-D group ($6.20\%$), and two from the p53abn group ($6.20\%$). Six patients experienced recurrence ($9.23\%$): one from the MMR-D group ($16.67\%$), four from the p53abn group ($66.67\%$) and one from the p53wt group ($16.67\%$). Excluding non-diseased mortality or life-accidents, the four molecular types were compared through log-rank method. The OS for MMR-D, POLE mutation, p53abn and p53wt groups were $94.4\%$, $100.0\%$, $81.8\%$ and $100.00\%$, respectively. There was no significant difference among the four groups ($$p \leq 0.394$$). In addition, the PFS for MMR-D, POLE mutation, p53abn, and p53wt groups were $94.4\%$, $100.00\%$, $63.60\%$ and $97.10\%$, respectively. Differences among the four groups was statistically significant ($$p \leq 0.010$$). The patients in POLE mutation and p53wt groups had higher PFS and OS, while EC patients in the p53abn group had the lowest survival rate (Fig. 3). Fig. 3Overall survival curve ($$p \leq 0.394$$) (A) and progression free survival curve ($$p \leq 0.010$$) (B) for patients with endometrial cancer, as classified by ProMisE. ## Correlation analysis of associated factors Three of the 70 patients died, and the median follow-up time was 41 months. The 3-year, 5-year and 6-year survival rate was $100.00\%$, $98.57\%$, and $95.71\%$, respectively. In terms of recurrence in patients with EC, six of the 70 patients relapsed, with a 1-year, 3-year and 5-year recurrence rate of $1.43\%$, $5.71\%$ and $8.33\%$, respectively. The log-rank test was used to analyze prognostic factors (Fig. 4A-B). The results showed that the survival rate of progesterone receptor (PR) positive in patients was significantly higher than for PR negative patients ($$p \leq 0.003$$). Similarly, the survival rate of estrogen receptor (ER) positive patients was significantly higher than for ER negative patients ($$p \leq 0.001$$). Fig. 4Overall survival curve of endometrial cancer patients positive for progesterone receptor ($$p \leq 0.003$$) (A), and estrogen receptor ($$p \leq 0.001$$) (B) The survival rate of patients with vascular infiltration was lower than that of patients without vascular infiltration ($$p \leq 0.011$$), while survival rate of patients with hyperlipidemia was lower than that of patients without this disease ($$p \leq 0.001$$). The survival rate of patients with atherosclerosis was significantly lower than that of non-atherosclerotic patients ($p \leq 0.001$, Fig. 5A-D). Fig. 5Survival proportions of patients with atherosclerosis (A), comparison of health status of patients with atherosclerosis ($p \leq 0.001$) (B), survival of patients with hyperlipidemia ($$p \leq 0.001$$) (C), and survival proportions of patients with vascular invasion ($$p \leq 0.011$$) (D) FIGO stage ($$p \leq 1.000$$), age ($$p \leq 0.251$$), BMI index ($$p \leq 0.190$$), family history of tumor ($$p \leq 0.369$$), degree of differentiation ($$p \leq 0.369$$), surgical approach ($$p \leq 0.982$$), degree of myometrial infiltration ($$p \leq 0.140$$), tumor size diameter ($$p \leq 0.154$$), menopausal status ($$p \leq 0.184$$), radiotherapy ($$p \leq 0.868$$), chemotherapy ($$p \leq 0.426$$), imaging findings ($$p \leq 0.405$$), uterine leiomyoma ($$p \leq 0.881$$), diabetes mellitus ($$p \leq 0.874$$), high blood pressure ($$p \leq 0.793$$), abnormal uterine bleeding ($$p \leq 1.000$$), human papillomavirus ($$p \leq 1.000$$), cervicitis ($$p \leq 0.766$$), vaginal drainage ($$p \leq 1.000$$), negative syphilis ($$p \leq 0.306$$), CA153 ($$p \leq 1.000$$), CA125 ($$p \leq 0.480$$), P16 ($$p \leq 0.298$$),Vimentin ($$p \leq 0.592$$), β- Catenin ($$p \leq 0.251$$), Pax-8 ($$p \leq 0.370$$), p53 ($$p \leq 0.692$$), Ki-67 ($$p \leq 0.572$$), and PTEN ($$p \leq 0.584$$) all had no significant correlation with prognosis. ## Discussion The clinical value of prognostic / predictive tools can hardly be underestimated for ensuring the accurate potential disease progression within individual patients, and thus allowing for the implementation of bespoke therapeutic strategies for maximizing successful outcomes (including extended OS and PFS timeframes) within such patients. Such predictive tools are invariable more important for specific tumors such as EC, whereby the presence of multiple risk factors can affect EC manifestation and clinical aggressiveness [17–25]. Moreover, the ability to characterize tumors on a genomic level can have potentially crucial clinical value, as – with such knowledge – oncologists and clinical staff will have enhanced awareness and clinical foresight in order to perform more informed and bespoke therapeutic and patient management decisions, in order to ensure successful treatments. Consequently, on focusing upon EC, accurate genomic characterization concerning EC molecular typing / groupings can pivot therapeutic outcomes within patients towards a more favorable clinical endpoint. One milestone achievement was the EC molecular typing prediction tool developed by the Cancer Genome Atlas Research Network in 2013, whereby a total of 373 EC samples were mapped in an integrated manner for proteomic / transcriptomic / genomic expression profile identification [26]. Such great research efforts by the TCGA research network led to the above-described gold-standard EC molecular typing classification, consisting of four major EC molecular-type subgroups, currently still utilized by oncologists on a global scale [26]. Although the prognostic value of TCGA subgroups has been confirmed within many studies, it is unclear how they are combined with histological features, such as tumor grade and histological type. Moreover, the prognostic factors in pathology are always disturbed in clinical diagnosis and evaluation, including repeatability of histological and FIGO classifications. There is often overlap between histological subtypes and grading, which complicates clinical decision-making. Therefore, the diagnostic consistency between observers is still not ideal, especially within high-grade histotypes and frozen paraffin specimens [27]. In addition to TCGA, ProMisE typing has shown its clinical significance in the treatment and diagnosis of EC patients [28]. ProMisE divides EC patients into four prognostic groups: POLE mutation (POLE MT), mismatch repair defect (mmr-d), p53 abnormal (p53 abn), and p53 wild type (p53 wt)[28]. The POLE MT group includes EC with the best prognosis and the highest mutation load. This group is characterized by polymerase ε (POLE) and is the only group that can be fully identified by sequencing. The prognosis of the MMR-d group is moderate, the mutation load is high, and the microsatellite is unstable. This group can be identified by the defective immunohistochemical expression of the mismatch repair protein (MMR). The p53 abn group has the worst prognosis, low mutation load, a high rate of somatic copy number variation and a high mutation rate of TP53. This group can be identified by the abnormal immunohistochemical expression of p53. The p53 wt group has a moderate prognosis, low mutation load, low rate of somatic copy number variation, and no molecular characteristics. This group is typically identified by excluding molecular characteristics of other groups [29]. Consequently, the TCGA and ProMisE classifications were comparatively analyzed across a cohort of 70 EC patients, in order to evaluate the optimal classification system for molecular-type-based EC patient classification. In this study, the only variable that reached statistical significance when comparing the four TCGA subgroups was age. Patients > 60 years also tended to have high-CN and less POLE super-mutants. Following the TCGA-based classification, this study observed a significant difference between type I and type II EC in pathological classification (χ²= 9.437, $$p \leq 0.013$$) among EC patients. There was a significant difference in pathological grade between the G1 / G2 and G3 group (χ²= 11.098, $$p \leq 0.006$$). Moreover, IHC analysis demonstrated insignificant dysregulated expression of vimentin, Ki-67, PTEN, MSH2, Pax-8, and β- catenin. There was no significant difference between the four subgroups of TCGA typing and dysregulated expression of CD10, ER, PR, and p16. Only p63 expression (χ²= 11.585, $$p \leq 0.005$$), and p53 expression (χ²= 11.090, $$p \leq 0.029$$) were significant. Additionally, MSH6 was upregulated across all patients, possibly suggesting that p63, p53 and MSH6 proteins within the MSI family can play important roles within EC. Similar to such results, MSH6 is a high-risk factor affecting patient prognosis, with such expression levels being linked to patient clinicopathological parameters [30]. Being a crucial gene, p53 plays a role within many tumors, providing that wild-type TP53 can activate natural cellular immune response of cells, while mutant TP53 can lead to the immune-escape of tumor cells by negatively regulating cellular natural-immune-signaling, thus promoting the recurrence and metastasis of tumors [31]. MSI-H mutations are common in EC, and the frequency of benign endometrial lesions is often high. Within this study, MSI-H mutations in EC patients were associated with increased tumor grade, severe myometrial infiltration, together with increased clinical stage, as expected. In this study, the only variable that reached statistical significance when comparing the four TCGA subgroups was age. Patients > 60 years also tended to have high-CN and less POLE super-mutants. Comparison of high-throughput sequencing results indicated that p53 immunostaining had elevated accuracy in predicting TP53 mutations within EC, with a consistency of $67.14\%$ (the accuracy of MMR and MSI-H mutations was $46.88\%$). Such observations suggest that there exists a potential correlation between EC patients and pathological features, which needs to be discussed further. It is reported that somatic mutations within the POLE gene are found in 6–$10\%$ of EC patients, and are associated with improving relapse-free survival and germline mutations in patients with grade 3 endometrioid EC [29][36]. Within this study, POLE mutations accounted for $4.82\%$ of all study participants. Tumor suppressor phosphatase and tensin homologue (PTEN) is a phospholipid phosphatase, which counteracts the activity of phosphatidylinositol 3-kinase by dephosphorylating phosphatidylinositol 3-kinase by phosphorylating the D3 position of the inositol ring of phosphatidylinositol. Similarly, TP53 acts in the case of DNA damage and inhibits the cell cycle in the G1 / S phase, thereby activating the repair mechanism. The dysfunction of p53 in malignant tumors is mainly due to the inactivation of the p53 protein through binding protein or TP53 mutation. Within type I EC, TP53 was found to be upregulated, together with PTEN downregulation within higher EC grades [32]. Consequently, the immune expression evaluation of TP53 is helpful to the diagnosis and treatment of various EC types of EC. Secondly, TP53 is a prognostic biomarker for this type of tumor, with TP53 genomic mutations accounting for $7.88\%$ of participants in this study. Pathological classification was based upon TCGA typing, where molecular types were compared through log-rank method. The survival curve using Kaplan Meier method demonstrated that OS for type I EC was higher than that of type II EC. Overall, survival rate of patients with type II EC was significantly lower, together with worse prognosis when compared to patients having type I EC. Finally, ProMisE typing was performed to analyze the EC patient prognoses. There was no significant difference in OS across all four groups ($$p \leq 0.585$$). However, PFS among the four groups was significantly different ($$p \leq 0.012$$). The PFS time and prognosis for the p53wt group were significantly lower than for MMR-d and POLE mutation groups. In addition, within ProMisE typing, it was found that results obtained by high-throughput sequencing of MMR IHC exhibited MLH1 and PMS2 downregulation. This phenomenon was previously linked to MLH1 promoter hypermethylation [37–38]. Another possible explanation for differing results of IHC and high-throughput sequencing is that existed an abundance of normal nuclei within selected EC tissue, with mutant MSI alleles not being detected through high-throughput sequencing. Following comparative analyses for TCGA and ProMisE molecular typing protocol performance in assessing the investigated EC cohort within this study, the TCGA revealed to have increased resolution and robustness in predicting OS and PFS timeframes, in comparison to the ProMisE molecular typing protocol. The added resolution for TCGA molecular protocol lies in the premise that also additional risk factors are taken into consideration when performing predictive evaluations for EC patients with the TCGA protocol. This study does have several limitations. Firstly, the patient cohort of 70 EC cases admitted to our institute, across a period of 6.5 years, is relatively small and can possibly not be fully representative for study outcomes. Secondly, the lack of an external validation cohort in our study is due to the fact that the use of ProMisE typing is still not widespread among most clinicians in China. Importantly, our study was conducted as a retrospective study, highlighted by its clinical relevance analysis in EC patients. In the future, we will supplement ProMisE subtyping analysis with large samples based on Chinese population. However, a follow-up study utilizing a second patient cohort is contemplated, in order to perform temporal validation of such study results. Thirdly, during the NGS investigation, a large number of samples were still not identified correctly due to the inherent challenges involved within immunostaining biopsy specimens using prolonged fixation time and the presence of differing tumor cell structures (extracted from IHC and DNA-extracted sections). This could also be linked to the difficulty of amplicon deep-sequencing method applied to all samples, in order to detect stop-acquisition / splicing mutations and large deletions / insertions, especially within endometrial biopsy samples. Finally, possible errors in establishing the difference between MMR and MSI could be due to the presence of too many normal nuclei (normal endometrium and a large number of tumor infiltrating lymphocytes) in order to detect microsatellite unstable mutant alleles through NGS. In essence, the main outcomes and inferences from this study are outlined below, namely: A)TCGA molecular classification has advantages over ProMisE classification for accurate diagnosis of endometrial carcinoma. B)Comparison among the four TCGA types (POLE, Low-CN, High-CN, and MSI-H) and age was statistically significant. C)Patients with POLE mutations and low-CN had increased PFS and OS, whereas those with high-CN had the lowest overall survival periods. ## Conclusion Presently, diagnostic and classification protocols for EC are still evolving. Modernized, rapid and accurate molecular diagnostic features certainly aid in the integration of molecular subtype diagnosis within clinical practice. Moreover, molecular subtypes are of undoubted predictive importance. Hence, it is particularly important to conduct auxiliary research on POLE mutation, MMR / MSI and abnormal p53 expression. The advantage of TCGA molecular subtype diagnosis method is that it can generate all datasets required by molecular subtypes and other genomic information apart from molecular subtypes, including β- Catenin mutation status, tumor mutation load (TMB) and other parameters, which can guide treatment. The survival analysis demonstrated that TCGA model is more beneficial than ProMisE typing when predicting patient prognosis. Nevertheless, ProMisE typing is less costly, easier, and provides rapid data access. The combination of these two methods provides additional unique information for the diagnosis and prognosis of prose subtypes, though are more valuable in the molecular classification of prose subtypes. Finally, TCGA molecular typing for EC has feasibility and application value within a clinical setting. ## Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1: Table S1: Panel of immunohistochemical markers and associated diagnostic criteria ## References 1. Huang Y, Chen Y, Zhu Y, Wu Q, Yao C, Xia H. Postoperative systemic Immune-Inflammation index (SII): a Superior Prognostic factor of Endometrial Cancer. Front Surg (2021.0) 8 704235. DOI: 10.3389/fsurg.2021.704235 2. Wang M, Hui P. A timely update of immunohistochemistry and molecular classification in the diagnosis and Risk Assessment of Endometrial Carcinomas. Arch Pathol Lab Med (2021.0) 145 1367-78. DOI: 10.5858/arpa.2021-0098-RA 3. Mitchard J, Hirschowitz L. Concordance of FIGO grade of endometrial adenocarcinomas in biopsy and hysterectomy specimens. Histopathology (2003.0) 42 372-8. DOI: 10.1046/j.1365-2559.2003.01603.x 4. Nguyen M, Han L, Pua T, Mares A, Karsy M, LaFargue C. Comparison of FIGO grade 3 endometrioid endometrial carcinomas with type 2 uterine cancers. Can grade 3 tumors be classified as type 2 cancers? A clinicopathological and immunohistochemical analysis. Gynecol Oncol (2013.0) 130 e91-2. DOI: 10.1016/j.ygyno.2013.04.273 5. Tomczak K, Czerwińska P, Wiznerowicz M. The Cancer Genome Atlas (TCGA): an immeasurable source of knowledge. Contemp Oncol (Pozn) (2015.0) 19 A68-77. PMID: 25691825 6. Alexa M, Hasenburg A, Battista MJ. The TCGA Molecular classification of Endometrial Cancer and its possible impact on adjuvant treatment decisions. Cancers (Basel) (2021.0) 13 1478. DOI: 10.3390/cancers13061478 7. Jiang F, Jiang S, Cao D, Mao M, Xiang Y. Immunologic signatures across Molecular Subtypes and potential biomarkers for Sub-Stratification in Endometrial Cancer. Int J Mol Sci (2023.0) 24 1791. DOI: 10.3390/ijms24021791 8. Leon-Castillo A, Horeweg N, Peters EEM, Rutten T, Ter Haar N, Smit VTHBM. Prognostic relevance of the molecular classification in high-grade endometrial cancer for patients staged by lymphadenectomy and without adjuvant treatment. Gynecol Oncol (2022.0) 164 577-86. DOI: 10.1016/j.ygyno.2022.01.007 9. Bellone S, Roque DM, Siegel ER, Buza N, Hui P, Bonazzoli E. A phase II evaluation of pembrolizumab in recurrent microsatellite instability-high (MSI-H) endometrial cancer patients with Lynch-like versus MLH-1 methylated characteristics (NCT02899793). Ann Oncol (2021.0) 32 1045-6. DOI: 10.1016/j.annonc.2021.04.013 10. Momeni-Boroujeni A, Nguyen B, Vanderbilt CM, Ladanyi M, Abu-Rustum NR, Aghajanian C. Genomic landscape of endometrial carcinomas of no specific molecular profile. Mod Pathol (2022.0) 35 1269-78. DOI: 10.1038/s41379-022-01066-y 11. Zhang C, Zheng W. High-grade endometrial carcinomas: morphologic spectrum and molecular classification. Semin Diagn Pathol (2022.0) 39 176-86. DOI: 10.1053/j.semdp.2021.11.002 12. Hussein YR, Broaddus R, Weigelt B, Levine DA, Soslow RA. The genomic heterogeneity of FIGO Grade 3 Endometrioid Carcinoma Impacts Diagnostic Accuracy and Reproducibility. Int J Gynecol Pathol (2016.0) 35 16-24. DOI: 10.1097/PGP.0000000000000212 13. Talhouk A, Jamieson A, Crosbie EJ, Taylor A, Chiu D, Leung S. Targeted Molecular Testing in Endometrial Carcinoma: validation of a clinically driven selective ProMisE testing protocol. Int J Gynecol Pathol (2022.0). DOI: 10.1097/PGP.0000000000000898 14. Timmerman S, Van Rompuy AS, Van Gorp T, Vanden Bempt I, Brems H, Van Nieuwenhuysen E. Analysis of 108 patients with endometrial carcinoma using the PROMISE classification and additional genetic analyses for MMR-D. Gynecol Oncol (2020.0) 157 245-51. DOI: 10.1016/j.ygyno.2020.01.019 15. Falcone F, Normanno N, Losito NS, Scognamiglio G, Esposito Abate R, Chicchinelli N. Application of the proactive Molecular Risk Classifier for Endometrial Cancer (ProMisE) to patients conservatively treated: outcomes from an institutional series. Eur J Obstet Gynecol Reprod Biol (2019.0) 240 220-5. DOI: 10.1016/j.ejogrb.2019.07.013 16. McKenzie R, Scott RJ, Otton G, Scurry J. Early changes of endometrial neoplasia revealed by loss of mismatch repair gene protein expression in a patient diagnosed with Lynch syndrome. Pathology (2016.0) 48 78-80. DOI: 10.1016/j.pathol.2015.11.003 17. Santoro A, Angelico G, Travaglino A, Inzani F, Arciuolo D, Valente M. New Pathological and Clinical Insights in Endometrial Cancer in View of the updated ESGO/ESTRO/ESP Guidelines. Cancers (Basel) (2021.0) 13 2623. DOI: 10.3390/cancers13112623 18. Morrison J, Baldwin P, Buckley L, Cogswell L, Edey K, Faruqi A. British Gynaecological Cancer Society (BGCS) vulval cancer guidelines: recommendations for practice. Eur J Obstet Gynecol Reprod Biol (2020.0) 252 502-25. DOI: 10.1016/j.ejogrb.2020.05.054 19. Tung H-J, Huang H-J, Lai C-H. Adjuvant and post-surgical treatment in endometrial cancer. Best Pract Res Clin Obstet Gynaecol (2022.0) 78 52-63. DOI: 10.1016/j.bpobgyn.2021.06.002 20. Zhao J, Hu Y, Zhao Y, Chen D, Fang T, Ding M. Risk factors of endometrial cancer in patients with endometrial hyperplasia: implication for clinical treatments. BMC Womens Health (2021.0) 21 312. DOI: 10.1186/s12905-021-01452-9 21. Lortet-Tieulent J, Ferlay J, Bray F, Jemal A. International patterns and Trends in Endometrial Cancer incidence, 1978–2013. J Natl Cancer Inst (2018.0) 110 354-61. DOI: 10.1093/jnci/djx214 22. Raglan O, Kalliala I, Markozannes G, Cividini S, Gunter MJ, Nautiyal J. Risk factors for endometrial cancer: an umbrella review of the literature. Int J Cancer (2019.0) 145 1719-30. DOI: 10.1002/ijc.31961 23. Shuning C, Weimin K, Dan L. Research Progress on the relationship between obesity, metabolic abnormalities and endometrial cancer. Chin J Clin Nutr (2022.0) 50 40-3 24. 24.Padilla-Iserte P, Lago V, Tauste C, Díaz-Feijoo B, Gil-Moreno A, Oliver R et al. Impact of uterine manipulator on oncological outcome in endometrial cancer surgery. Am J Obstet Gynecol. 2021;224:65.e1-65.e11. 25. Parazzini F, Di Martino M, Candiani M, Viganò P. Dietary components and uterine leiomyomas: a review of published data. Nutr Cancer (2015.0) 67 569-79. DOI: 10.1080/01635581.2015.1015746 26. Kandoth C, Schultz N, Cherniack AD, Akbani R, Liu Y. Integrated genomic characterization of endometrial carcinoma. Nature (2013.0) 497 67-73. DOI: 10.1038/nature12113 27. Rafaniello-Raviele P, Betella I, Rappa A, Vacirca D, Tolva G, Guerrieri-Gonzaga A. Microsatellite instability evaluation: which test to use for endometrial cancer?. J Clin Pathol (2023.0) 76 29-33. DOI: 10.1136/jclinpath-2021-207723 28. Huvila J, Orte K, Vainio P, Mettälä T, Joutsiniemi T, Hietanen S. Molecular subtype diagnosis of endometrial carcinoma: comparison of the next-generation sequencing panel and proactive molecular risk classifier for Endometrial Cancer classifier. Hum Pathol (2021.0) 111 98-109. DOI: 10.1016/j.humpath.2021.02.006 29. Kommoss S, McConechy MK, Kommoss F, Leung S, Bunz A, Magrill J. Final validation of the ProMisE molecular classifier for endometrial carcinoma in a large population-based case series. Ann Oncol (2018.0) 29 1180-8. DOI: 10.1093/annonc/mdy058 30. Zhao Z, Liu H, Zhou X, Fang D, Ou X, Ye J. Necroptosis-related lncRNAs: Predicting Prognosis and the distinction between the Cold and Hot Tumors in gastric Cancer. J Oncol (2021.0) 2021 6718443. DOI: 10.1155/2021/6718443 31. An HJ, Kim KI, Kim JY, Shim JY, Kang H, Kim TH. Microsatellite instability in endometrioid type endometrial adenocarcinoma is associated with poor prognostic indicators. Am J Surg Pathol (2007.0) 31 846-53. DOI: 10.1097/01.pas.0000213423.30880.ac 32. Kim N, Kim Y-N, Lee K, Park E, Lee YJ, Hwang SY. Feasibility and clinical applicability of genomic profiling based on cervical smear samples in patients with endometrial cancer. Front Oncol (2022.0) 12 942735. DOI: 10.3389/fonc.2022.942735
--- title: Astrocytes display ultrastructural alterations and heterogeneity in the hippocampus of aged APP-PS1 mice and human post-mortem brain samples authors: - Marie-Kim St-Pierre - Micaël Carrier - Fernando González Ibáñez - Mohammadparsa Khakpour - Marie-Josée Wallman - Martin Parent - Marie-Ève Tremblay journal: Journal of Neuroinflammation year: 2023 pmcid: PMC10015698 doi: 10.1186/s12974-023-02752-7 license: CC BY 4.0 --- # Astrocytes display ultrastructural alterations and heterogeneity in the hippocampus of aged APP-PS1 mice and human post-mortem brain samples ## Abstract The past decade has witnessed increasing evidence for a crucial role played by glial cells, notably astrocytes, in Alzheimer’s disease (AD). To provide novel insights into the roles of astrocytes in the pathophysiology of AD, we performed a quantitative ultrastructural characterization of their intracellular contents and parenchymal interactions in an aged mouse model of AD pathology, as aging is considered the main risk factor for developing AD. We compared 20-month-old APP-PS1 and age-matched C57BL/6J male mice, among the ventral hippocampus CA1 strata lacunosum-moleculare and radiatum, two hippocampal layers severely affected by AD pathology. Astrocytes in both layers interacted more with synaptic elements and displayed more ultrastructural markers of increased phagolysosomal activity in APP-PS1 versus C57BL6/J mice. In addition, we investigated the ultrastructural heterogeneity of astrocytes, describing in the two examined layers a dark astrocytic state that we characterized in terms of distribution, interactions with AD hallmarks, and intracellular contents. This electron-dense astrocytic state, termed dark astrocytes, was observed throughout the hippocampal parenchyma, closely associated with the vasculature, and possessed several ultrastructural markers of cellular stress. A case study exploring the hippocampal head of an aged human post-mortem brain sample also revealed the presence of a similar electron-dense, dark astrocytic state. Overall, our study provides the first ultrastructural quantitative analysis of astrocytes among the hippocampus in aged AD pathology, as well as a thorough characterization of a dark astrocytic state conserved from mouse to human. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12974-023-02752-7. ## Introduction Alzheimer’s disease (AD), the most common type of dementia, in which aging is a predominant risk factor [1], is associated clinically with progressive brain atrophy, as well as neuronal and synaptic loss, leading over the years to cognitive decline [2–6]. Notable pathological hallmarks of AD include the build-up of intracellular hyperphosphorylated tau forming neurofibrillary tangles (NFTs) and the extracellular accumulation of amyloid beta (Aß) compacting into fibrillar Aß plaques [2]. AD is now considered a brain manifestation of metabolic disorder: signs of AD that start early on during its progression include reduced brain energy metabolism, resulting from alterations in lipid [7–10] and glucose [11] metabolism, as well as amino acids, and other tricarboxylic acid cycle (TCA) metabolites [9, 12], all of which are important to maintain adequate brain energy levels. A particular feature of AD is the brain region-dependent vulnerability to pathology, starting to affect early on key regions such as the entorhinal cortex and hippocampus [13, 14]; the latter being mainly involved in the regulation of emotions (ventral hippocampus) as well as learning and memory processes (dorsal hippocampus) [15, 16]. The hippocampus, and in particular its cornu ammonis (CA)1, have been extensively investigated in the context of AD, due to the drastic atrophy observed [3–5], along with the functional impairment associated with this region throughout the pathogenesis of AD [17, 18]. The CA1 can be further separated into various layers, each defined by distinct functional, structural, and ultrastructural characteristics. For instance, the CA1 pyramidal neurons project their apical dendrites to the stratum radiatum, which contains their proximal branches while the stratum lacunosum-moleculare contains their distal branches [17]. The stratum lacunosum-moleculare also presents numerous large blood vessels [17], the latter being highly vulnerable to AD-related damage most likely due to a reduced blood flow at baseline compared to other brain regions [19]. Astrocytes, representing roughly 20–$40\%$ of all glial cells in the cerebral gray matter [20, 21], first originate during embryonic development from progenitor cells or radial glia which then mature throughout early postnatal development into astrocytes [22]. Defined morphologically by their star-like appearance given their numerous and complex processes [23], astrocytes are characterized, at the ultrastructural level, by their angular and thin processes often interacting with synaptic elements, their intermediate filaments, and the accumulation of glycogen granules [24]. Astrocytes can help protect brain homeostasis through the clearance of parenchymal metabolic waste through the glymphatic system [25–28], the regulation of cerebral blood flow, the maintenance of the blood–brain barrier [29–32], and are highly involved in synaptic activity and plasticity (synaptogenesis, maintenance, maturation, and elimination) [33–37]. Known for their key role in neuronal metabolic support, astrocytes can help maintain brain functions in numerous ways; for instance, by transforming the exocytotoxic glutamate released by post-synaptic dendritic spines into glutamine, which can be recycled back by synaptic elements [38, 39]. In addition, astrocytic glycogen plays a crucial role in the metabolic neuronal support as the current hypothesis suggests that astrocytes can break down this carbohydrate storage into lactate, which can be shuttled to neurons for their energetic needs [40, 41]. The last decades have provided increasing evidence that glial cells, including astrocytes, are critical players in the pathogenesis of AD. This emerging role of astrocytes in AD is corroborated notably by genome-wide association studies highlighting several astrocytic gene variants connected with a higher risk of developing late-onset AD in humans [42, 43]. While the functions of astrocytes in the pathogenesis of AD remain to be fully demystified, studies depleting astrocytes (using transgenic or pharmacological strategies) to investigate their role in AD pathogenesis are pointing toward beneficial outcomes [44]. In particular, increased Aß levels were measured upon astrocytic ablation in two models of AD pathology: organotypic brain cultures from postnatal day 7 5xFAD mice and hippocampal sections from 9-month-old APP23/glial fibrillary acidic protein (GFAP)-thymidine kinase mice [45, 46], highlighting a potential role of astrocytes in the clearance of Aß. Astrocytes near Aß plaques were also shown to release neprilysin, an enzyme capable of degrading Aß [47], via protein kinase A and C [48], as well as insulin [49], a hormone crucial for the regulation of glucose metabolism [50]. At the ultrastructural level, astrocytic processes were shown to penetrate inside the Aß plaque core [51], suggested to be associated with plaque fragmentation to help with its degradation in human post-mortem brain samples of patients with AD [52]. Astrocytes were also confirmed to engulf dystrophic neurites, often found accumulated nearby Aß plaques in 6- and 12-month-old APP-PS1 mice, a model of AD pathology [53]. A more recent ultrastructural investigation in aged human post-mortem brain samples of individuals with AD further demonstrated that the astrocytic density near Aß plaques did not correlate to plaque size, and hypothesized that their close interaction with the plaque microenvironment could be due to neuritic damage rather than the Aß plaque itself [54]. Studies using single-cell and -nucleus RNA sequencing further demonstrated the highly heterogeneous nature of astrocytes in response to AD pathology, with a myriad of transcriptomic signatures reported such as the disease-associated astrocytes [55–57]. This signature was uncovered in 1.5–2, 4–5, 7–8, 10, 13–14 and 20-month-old male and female 5xFAD mice and presented an upregulation of specific genes such as apoe and clu, both involved in Aß clearance [55]. However, the phenotypic alterations and heterogeneity of astrocytes in AD have not been examined yet at the ultrastructural level using electron microscopy, an approach which provides in-depth knowledge at the nanoscale on the structure of organelles and the cellular interactions among the parenchyma [56]. Understanding the structural alterations of astrocytic organelles, their intracellular contents (notably the nature and quantity of phagosomes), as well as their interaction with AD hallmarks will aid in our understanding of their roles in AD pathology. In addition, as morphological and transcriptomic studies have reported a plethora of astrocytic signatures with varying functions, investigating the heterogeneity of astrocytes on an ultrastructural level will complement previous studies and help mend the gap in unraveling in situ the diverse responses of astrocytes to AD pathology. This study aimed to provide quantitative data on the ultrastructure of astrocytes and assess qualitatively their heterogeneity among the ventral hippocampus CA1 strata lacunosum-moleculare and radiatum, layers highly affected by AD pathology [17, 58]. APP-PS1 and control C57BL/6J male mice were examined at 20 months of age to focus on aged AD pathology specifically. Astrocytes from the two examined layers showed increased interactions with synaptic elements (dendritic spines and axon terminals), along with an increased phagolysosomal pathway activity (more phagosomes and/or mature lysosomes within their cytoplasm). In addition, we uncovered electron-dense, dark astrocytic cells for the first time in aging and AD pathology, possessing ultrastructural features of astrocytes and markers of cellular stress, similar to the dark microglia [59, 60] and similar to dark astrocytic states observed in human post-mortem brain samples of brain injury [61–63] and brain tumors [62, 64] resected following surgery, in rat models of brain injuries (concussive and compressive head injuries [65] and electroshock [66], as well as in spinal cord cultures of embryonic mice [67]. These dark glial cells were positive for the ‘reactive’ astrocytic marker GFAP [68] and were observed throughout the parenchyma often in juxtaposition with large blood capillaries. Moreover, our observations highlight the presence of dark astrocytes in the hippocampal head of an aged human post-mortem brain sample, examined as a case study, similarly to the dark astrocytes previously observed in the parietal cortex of patients with traumatic brain injury and brain tumors [62, 63]. These findings confirmed the conservation across species of dark astrocytes as these cells were encountered in human post-mortem brain samples, thus showcasing similarities in the astrocytic ultrastructural features observed upon aging between mouse and human. ## Animal housing, euthanasia, and perfusion with aldehydes All experiments were performed according to the guidelines of the Institutional Animal Ethics committees, the Canadian Council on Animal Care, as well as the Animal Care Committee of Université Laval. C57BL/6J and age-matched APPSwe-PS1Δe9 male mice on a C57BL/6J background [69] (No. 34832-JAX, Jackson Laboratory, Maine, USA) at 3–4, and 20 months of age ($$n = 3$$–4), were housed under a 12 h light–dark cycle at 22–25 °C with free access to food and water. All experiments were performed on males, for this first study on the topic, considering that previous studies investigated glial heterogeneity in 14- and 20-month-old C57BL/6J and APP-PS1 mice used males [70, 71]. Mice were injected with 10 g/kg Methoxy-X04 (Tocris Biosciences, cat# 4920, Bristol, United Kingdom) 24 h prior to their euthanasia to visualize fibrillar Aß plaques at the light microscopy level [72]. Mice were injected intraperitoneally with sodium pentobarbital (80 mg/kg), then perfused transcardially with $3.5\%$ acrolein [diluted in phosphate buffer (PB): 100 mM at pH 7.4] and $4\%$ paraformaldehyde [PFA, diluted in phosphate-buffered saline (PBS): 50 mM at pH 7.4], followed by a 2-h post-fixation in $4\%$ PFA. Coronal brain sections were cut using a vibratome (Leica VT1000S) at 50 µm of thickness and kept in a cryoprotectant solution [$20\%$ glycerol (v/v), $20\%$ (v/v) ethylene glycol in PBS] at − 20 °C until further processing. ## Processing of human post-mortem brain samples As a case study, sections from a human brain (female, 81 years old; 18 h post-mortem delay, cause of death: asphyxia) were obtained from the CERVO Brain Research Center (QC, Canada). Collecting, storage and handling procedures were approved by the Ethics Committee of the Institut Universitaire en Santé Mentale de Québec and Université Laval. Written and informed consent was obtained for the use of human post-mortem brain tissues and all the experiments were performed in line with the Code of Ethics of the World Medical Association. The brain was first separated in halves trough the midline and hemibrains were cut coronally in 2-cm-thick blocks. They were then fixed in $4\%$ PFA for 3 days at 4 °C before being stored in $15\%$ sucrose and $0.1\%$ sodium azide at 4 °C until further processing. The hippocampal head region of the right hemibrain was then cut using a vibratome (VT1000s) to obtain 50-µm-thick coronal sections which were kept at − 20 °C in a cryoprotectant solution until further processing, in preparation for scanning electron microscopy (SEM) experiments. ## Processing of mouse samples for anti-GFAP immunohistochemistry Brain sections containing the ventral hippocampus CA1 from 20-month-old APP-PS1 male mice (Bregma 2.92 to 3.64 mm [73]) were selected for further processing. Selected sections were quenched with $0.3\%$ H2O2 (Fisher Scientific, Ottawa, lot# 202762) in PBS for 5 min. Afterward, the sections were incubated in $0.1\%$ NaBH4 in PBS for 30 min followed by 3 washes of 10 min in PBS. Brain sections were then incubated in a blocking buffer solution containing $5\%$ normal goat serum (Jackson ImmunoResearch Labs, Baltimore, USA cat# 005-000-121), $5\%$ bovine albumin serum (Sigma-Aldrich, Oakville, cat# 9048-46-8,), and $0.01\%$ Triton X-100 in PBS for 1 h at room temperature (RT). They were then incubated overnight in a blocking buffer solution with the primary rabbit polyclonal anti-GFAP antibody (1:5000; Abcam, Cambridge, MA, USA, Ab7260) at 4 °C. The following day, the sections were washed in $0.01\%$ PBS-Triton (PBS-T) and incubated with a biotinylated goat anti-rabbit polyclonal secondary antibody (1:300; Jackson ImmunoResearch, Baltimore, USA, cat# 111-066-046) in Tris-buffered saline (TBS; 50 mM, pH 7.4) for 2 h at RT. Afterward, the sections were washed in PBS-T and incubated for 1 h at RT in an avidin–biotin complex solution (ABC; 1:100; Vector Laboratories, Newark, USA, cat# PK-6100) in TBS. The staining was revealed with a solution containing $0.05\%$ 3,3′-diaminobenzidine (DAB; Millipore Sigma, Oakville, USA, cat# D5905-50TAB) and $0.015\%$ H2O2 diluted in Tris buffer (TB; 0.05 M, pH 8.0). The samples were washed 3 times in PBS and then further processed with unstained sections for SEM. ## Preparation of mouse and human samples for SEM Mouse brain sections containing the ventral hippocampus CA1 (Bregma 2.92 to 3.64 mm [73]) from 3–4- and 20-month-old C57BL/6J mice and age-matched APP-PS1 mice, both unstained for quantitative analysis and stained for GFAP to confirm the astrocytic identity, were selected for SEM processing. As a case study, post-mortem human brain samples containing the hippocampal head from an aged individual were also processed for SEM. Selected sections were first washed with PB, then incubated for 1 h in a PB solution containing equal volumes of $3\%$ potassium ferrocyanide (Sigma-Aldrich, Ontario, Canada, cat# P9387) and $4\%$ osmium tetroxide (EMS, Pennsylvania, USA, cat# 19190). The brain tissues were next incubated for 20 min in a filtered and heated $1\%$ thiocarbohydrazide solution (diluted in double-distilled water; Sigma-Aldrich, Ontario, Canada, cat# 223220) and for 30 min in $2\%$ aqueous osmium tetroxide. The samples were dehydrated in increasing concentrations of ethanol for 10 min each (2 × $35\%$, 1 × $50\%$, 1 × $70\%$, 1 × $80\%$, 1 × $90\%$ 3 × $100\%$) followed by 3 washes of 10 min in propylene oxide (Sigma-Aldrich, Ontario, Canada, #cat 110205-18L-C). The dehydrated tissues were embedded overnight in Durcupan resin (20 g component A, 20 g component B, 0.6 g component C, 0.4 g component D; Sigma Canada, Toronto, cat# 44,610) and flat-embedded between fluoropolymer films (ACLAR®, Pennsylvania, USA, Electron Microscopy Sciences, cat# 50425–25). Resin-embedded sections between films were kept in the oven for 5 days at 55 °C to allow the resin to polymerize. Regions of interest (containing the hippocampal head for post-mortem human brain and the ventral hippocampus CA1 strata lacunosum-moleculare and radiatum for mouse brain samples) were excised from the resin-embedded sections and glued onto resin blocks for ultramicrotomy. Using a Leica ARTOS 3D ultramicrotome, 73-nm-thick sections were cut with multiple levels obtained from each block (2–6 levels, ~ 6 µm apart) to obtain sufficient images of astrocytes for quantitative ultrastructural analysis. The ultrathin sections were placed on silicon wafers for SEM imaging, performed using a Zeiss Crossbeam 540 microscope. Images from mouse samples were first acquired at 25 nm per pixel for the density and distribution analysis of astrocytes [70]. All samples were imaged at a resolution of 5 nm per pixel for the ultrastructural analysis of typical astrocytes and characterization of dark astrocytes. GFAP-positive (+) typical astrocytes and dark astrocytes were further imaged with a Zeiss Crossbeam 350 scanning electron microscope using SmartSEM software (Fibics). GFAP + dark and typical astrocytic cell bodies were imaged at a resolution of 5 nm and 1 nm per pixel and exported as TIFF files using the Zeiss ATLAS Engine 5 software (Fibics). ## Density and distribution analysis of astrocytic states in mouse samples Parenchymal images (2–6 levels, ~ 6 µm apart) from the ventral hippocampus CA1 stratum lacunosum-moleculare from 4 animals per group were blinded to the genotype and age, then analyzed to investigate the density and distribution of astrocytic states. A distinction was made between dark and non-dark astrocytes (referred to as typical astrocytes in this manuscript) based on our ultrastructural observations. The density of typical and dark astrocytes in APP-PS1 vs C57BL/6J mice was determined, together with the ratio of dark astrocytes over all astrocytes imaged in each genotype using the 25 nm per pixel resolution images. In addition, we investigated the distribution of astrocytes based on their association with the vasculature or parenchyma (with or without any direct contact with the basement membrane of blood vessels, respectively). Typical astrocytes were positively identified based on their electron-lucent cyto- and nucleoplasm, granular nuclear pattern, angular processes interacting with parenchymal elements, as well as the presence of intermediate filaments [24, 60, 74]. A dark astrocytic state, termed dark astrocytes, was also identified based on the similar ultrastructural defining features of typical astrocytes, such as the angular processes and granular pattern of the nucleus, as well as presence of intermediate filaments, and previous EM observations made in organotypic cultures of spinal cord from embryonic mice [67], rat models of brain injury (compressive head injury, concussive head injury), pentylenetetrazole and kainic acid treatment [65], as well as electroshock [66]. The dark astrocytes that we observed often possessed a high accumulation of glycogen granules, ultrastructural markers of cellular stress such as the dilation of the endoplasmic reticulum (ER) and Golgi apparatus cisternae, a partial to total loss of their nuclear heterochromatin pattern, and an electron-dense cyto- and nucleoplasm [65–67]. Similar ultrastructural features were previously described in dark neurons [75–79] and dark microglia [59, 60, 70, 80], particularly the loss of nuclear heterochromatin pattern, electron-dense cytoplasm and nucleoplasm, and markers of cellular stress [60, 71, 72, 80, 81]. The ultrastructural density analysis protocol we performed for typical and dark astrocytes is based on previously published ultrastructural work examining microglia [60, 70]. ## Ultrastructural analysis of typical astrocytes in mouse samples For the ultrastructural analysis of typical and dark astrocytes, quantitative and qualitative, respectively, SEM images captured with a resolution of 5 nm per pixel were used. This analysis was conducted in the ventral hippocampus CA1 strata lacunosum-moleculare and radiatum from 20-month-old C57BL/6J and APP-PS1 mice. In each genotype ($$n = 3$$ animals/group) and localization (near vs far Aß plaques/dystrophic neurites in the case of the stratum lacunosum-moleculare), pictures of 31–38 astrocytes were acquired. Of note, in the stratum radiatum, we investigated astrocytes far from Aß plaques/dystrophic neurites only as little to no plaques were observed in this layer in our ultrathin samples. All the images were blinded to the experimental conditions. In the stratum lacunosum-moleculare, we analyzed a total of 102 astrocytic cell bodies per group, a sample size sufficient to obtain statistical power based on the software GPower V3.1 (effect size of 0.4; power of 0.95 estimated at 102 astrocytes). In the stratum radiatum, we analyzed a total of 59 astrocytic cell bodies per genotype to obtain sufficient statistical power (effect size of 0.9; power of 0.9 estimated at 60 astrocytes) [70]. These effect sizes were previously used to assess the ultrastructural heterogeneity of other glial cells, such as microglia [70, 82]. As we wanted to examine possible glycogen granules within the astrocytic cytoplasm as well as the electron density of their nucleoplasm and cytoplasm in our analysis of their ultrastructure, we did not perform immunostaining which could have masked these features. While no quantitative ultrastructural analysis of astrocytes had been performed yet, the identification and analysis of microglial intracellular contents and parenchymal interactions were previously described in detail [24, 60, 80, 83]. In the current study, the parenchymal interactions of astrocytes with myelinated axons, axon terminals, dendritic spines, and both elements of excitatory synapses were assessed. Myelinated axons were characterized by electron-dense sheaths surrounding the axonal cytoplasm [84]. Axon terminals contained several synaptic vesicles and sometimes juxtaposed dendritic spines recognized by their post-synaptic density [24, 74, 83]. Axon terminals that were or were not in direct contact with one or more dendritic spines were analyzed. Direct contacts with axon terminals, dendritic spines, and both elements of the same excitatory synapse were counted. Immature (primary, secondary) and mature (tertiary) lysosomes were identified by their homogenous or heterogeneous appearance, respectively [60, 71, 74]. The presence of phagosomes, both fully and partially digested, was often recognized among tertiary lysosomes, alongside large lipid droplets [60, 71]. The latter possessed a homogenous interior (either electron-lucent or dense) and an electron-dense outline [24, 60, 74, 80]. Fully or partially digested phagosomes were characterized by a defined membrane delineating a circular or oval shape, electron-lucent interior with (partially digested) or without (fully digested) cellular content [60, 70]. Likewise, autophagosomes possessed a circular double membrane, with an electron-lucent appearance in between the latter, and an interior with the same electron density as the cell’s cytoplasm [24, 60, 74]. Ultrastructural markers of cellular stress were assessed including the presence of altered mitochondria, as well as dilated ER and Golgi apparatus* cisternae. The width of ER and Golgi cisternae, together with the length of mitochondria were measured using ImageJ. ER were identified by their long and narrow stretches, while dilation of their cisternae was positively confirmed when swollen electron-lucent pockets measured at least 100 nm in diameter [59, 60, 71, 85]. Similarly, Golgi apparatuses, characterized by their beehive shape, were considered to have dilated cisternae when displaying swollen electron-lucent pockets larger than 100 nm [60, 70]. Mitochondria were defined as electron-dense double-membraned organelles possessing several cristae [60]. Mitochondria were considered to be ultrastructurally altered when their outer and/or inner membranes were degraded, if their cristae were deteriorated resulting in electron-lucent space, or if they had a “holy shape” indicative of mitochondria wrapping around themselves, a feature associated with impaired mitochondrial membrane potential and structural alterations thought to be associated with oxidative stress [60, 80, 86]. Mitochondria were also defined as elongated if their length measured over 1 µm [85]. The mitochondrial network, defined by the cytoplasmic area occupied by the mitochondrial area, was assessed [87]. Each mitochondrion was traced using the “freehand tool” in Image J, and the sum of all mitochondrial area was divided by the area of the cytoplasm to obtain the mitochondrial network [87]. The presence of glycogen granules, recognized as 22–40 nm electron-dense puncta contained within the astrocytic cytoplasm, was identified [88]. Lastly, nuclear indentations, a phenomenon associated with cell morphology remodeling [89] and observed as hollowed-out portions of the nucleus [90] were noted. Shape descriptors of astrocytes, i.e., area, perimeter, solidity, aspect ratio (AR), and circularity, were further measured using the software Image J. AR and circularity provide information on the elongation of the cells (AR is the ratio of height over width; circularity is 4π times the area over the perimeter squared) [84, 91]. The closer the value of the circularity to 0, the more elongated the cell body is [70, 71, 84, 92]. Solidity, a measurement of irregularity, is defined by the area of the cell body over the convex area (the closer the value to 0, the more irregular the shape is) [84, 91]. ## Qualitative analysis of typical and dark astrocytes in human samples The presence of typical and dark astrocytes in the hippocampal head of human post-mortem brain samples from an aged individual (female; post-mortem delay of 18 h; cause of death: asphyxia) was investigated as a case study, using similar identifying features for mouse astrocytes, and others described for astrocytes in the parietal and frontal cortical regions of human post-mortem samples resected following surgery of brain injury investigating qualitatively their ultrastructure [61]. In brief, astrocytes were positively identified by their angular processes protruding from the cell body, granular nucleus, and presence of intermediate filaments [24, 60–62, 74]. Dark astrocytes possessed similar ultrastructural characteristics alongside an electron-dense cytoplasm and nucleoplasm, as well as markers of cellular stress (e.g., dilated ER and altered mitochondria). Previous studies investigating human post-mortem cerebral cortex samples with brain injuries or cerebellar samples with hemangioblastoma resected following surgery identified similar dark astrocytes, which were termed “dark hypertrophic astrocytes” [61–63] and “dark astrocytes”, respectively [64]. These dark astrocytes were previously described as electron-dense cells with swollen mitochondria, abundant glycogen granules, and dilated ER cisternae [61–64]. The intracellular contents (e.g., mitochondria, fully and partially digested phagosomes, dilated ER, lysosomes) and parenchymal interactions (e.g., axon terminals, dendritic spines, myelinated axons) of these dark astrocytes were identified for the first time during aging among the hippocampal head based on similar criteria as in the mouse samples and those described in the aforementioned studies [61, 62, 64, 65, 67]. ## Statistical analysis Statistical analysis was performed using the software Prism 9 (v.9.2.0 GraphPad). For all quantitative data obtained (ultrastructure and cellular density in mice), the normality of the data distribution was first assessed using a Shapiro–Wilk test. The ultrastructural data of typical astrocytes in the stratum radiatum of C57BL/6J vs APP-PS1 mice were compared with a Mann–Whitney non-parametric test. The ultrastructural data of typical astrocytes in the stratum lacunosum of C57BL/6J vs APP-PS1 mice (far vs near Aß plaques/dystrophic neurites) were analyzed with a Kruskal–Wallis one-way ANOVA followed by a Dunn’s post hoc test. The density data of dark and typical astrocytes in the stratum lacunosum-moleculare of APP-PS1 mice vs C57BL/6J mice passed normality and were analyzed with a Welsh t-test. Data are expressed as mean ± standard error of mean (SEM). The sample size (n) refers to individual animals for the density analysis and individual astrocytes for ultrastructural analysis as performed in previous EM studies studying other glial cell types such as microglia to account for the ultrastructural heterogeneity between individual cells [70, 71, 84, 93–97]. Statistically significant differences are reported as $p \leq 0.05$, $p \leq 0.01$, $p \leq 0.001$, and *$p \leq 0.0001.$ ## Typical astrocytes in the hippocampal CA1 stratum radiatum of aged APP-PS1 vs age-matched C57BL/6J mice exhibit altered parenchymal interactions and intracellular contents The ventral (or anterior) hippocampus CA1 displays severe atrophy [3, 5, 98–101], as well as astrocytic morphological and molecular alterations [39, 102, 103], in samples from mouse models of AD pathology and patients with AD. We thus analyzed the ultrastructural features of astrocytes in these two layers of the ventral hippocampus CA1. This region is of particular interest as previous studies which were conducted in middle-aged and aged APP-PS1 mice revealed ultrastructural alterations together with an increased heterogeneity of microglia, another glial cell type highly affected by AD pathology and known to play a role in its pathogenesis [59, 70]. Across the study, 20-month-old APP-PS1 were compared with age-matched C57BL/6J male mice. We first started our ultrastructural investigation with the analysis of typical astrocytes in the stratum radiatum. We focused on areas located far from Aß plaques/dystrophic neurites (designated as ‘Far AD’) to have a sufficient sample size for this analysis, as little to no plaques were observed in this hippocampal layer among our samples. In the stratum radiatum (Fig. 1A, B), we observed a significant increase in the direct contacts between typical astrocytes and all synaptic elements in the APP-PS1 mice compared to age-matched C57BL/6J controls (Control 20.68 ± 2.244 contacts per astrocyte vs Far AD 29.06 ± 2.587 contacts per astrocyte, $$p \leq 0.0250$$) (Fig. 1C). When we further investigated which part of the synapses was contacted by astrocytes, we found increased interactions of astrocytes from APP-PS1 mice with axon terminals (Control 16.25 ± 1.729 contacts per astrocyte vs Far AD 22.29 ± 2.089 contacts per astrocyte, $$p \leq 0.0447$$) and dendritic spines (Control 1.429 ± 0.2020 contacts per astrocyte vs Far AD 2.194 ± 0.2384 contacts per astrocyte, $$p \leq 0.0243$$), and a tendency for both elements of a same excitatory synapse to be contacted (Control 3.357 ± 0.5400 contacts per astrocyte vs Far AD 4.581 ± 0.5344 contacts per astrocyte, $$p \leq 0.0931$$) (Fig. 1D–F). We confirmed that this increased structural relationship with synaptic elements was not due to a change in either the astrocytic area (Control 57.65 ± 3.380 µm2 vs Far AD 64.01 ± 5.421 µm2, $$p \leq 0.7574$$) or perimeter (Control 62.65 ± 3.603 µm vs Far AD 66.83 ± 3.841 µm, $$p \leq 0.4196$$) (Fig. 1G–H). These results highlight the preferential contacts with synapses made by astrocytes in the stratum radiatum of APP-PS1 mice compared to age-matched C57BL/6J controls. Fig. 1Parenchymal interactions of typical astrocytes in the stratum radiatum. Representative 5 nm per pixel of resolution scanning electron microscopy images acquired in the ventral hippocampus CA1 stratum radiatum of 20-month-old APP-PS1 (far from Aß plaques/dystrophic neurites) and age-matched C57BL/6J male mice (A, B). Quantitative graphs represent the number of direct contacts with C all synaptic elements, D axon terminals, E dendritic spines, and F both elements of the same synapse (axon terminals and dendritic spines). In G and H, the graphs represent, respectively, the area and perimeter of the astrocytic cell body. Data are shown as individual dots and are expressed as mean ± S.E.M. $p \leq 0.05$, using a non-parametric Mann–Whitney test. Statistical tests were performed on $$n = 8$$–11 astrocytes per animal in $$n = 3$$ mice/group, for a total of 59 cell bodies analyzed. red outline = plasma membrane, yellow outline = nuclear membrane, orange pseudo-coloring = dendritic spine, blue pseudo-coloring = axon terminals Moreover, in the stratum radiatum (Fig. 2A, B), intracellular investigation of astrocytes further revealed a tendency for a decreased presence of primary lysosomes in APP-PS1 mice compared to C57BL/6J controls (Control 1.179 ± 0.2523 primary lysosomes per astrocyte vs Far AD 0.6129 ± 0.1950 primary lysosomes per astrocyte, $$p \leq 0.0508$$) (Fig. 2C), while the APP-PS1 mice exhibited an increased number of tertiary lysosomes (Control 1.464 ± 0.3313 tertiary lysosomes per astrocyte vs Far AD 2.742 ± 0.3934 tertiary lysosomes per astrocyte, $$p \leq 0.0163$$) (Fig. 2E). This finding suggests a shift in the phagolysosomal pathway, more precisely an increased maturation of lysosomes resulting in more tertiary lysosomes and less primary lysosomes in the APP-PS1 mice. We observed similar tendencies in the relative percentage of astrocytes (cells positive for the presence of the organelle analyzed) for both primary lysosomes (Control 57.14 ± $9.524\%$ of astrocytes vs Far AD 32.36 ± $8.535\%$ of astrocytes, $$p \leq 0.0695$$) in the C57BL/6J control mice and tertiary lysosomes (Control 53.57 ± $9.598\%$ of astrocytes vs Far AD 77.42 ± $7.634\%$ of astrocytes, $$p \leq 0.0616$$) in the APP-PS1 mice (Fig. 2D, F). Therefore, differences in the number of lysosomes per astrocyte could result from more cells possessing at least one of these organelles, rather than an increased number of lysosomes per astrocytic cell. In addition, we observed an increased number of lipid bodies (Control 1.429 ± 4.161 lipids per astrocyte vs Far AD 4.161 ± 0.7706 lipids per astrocyte, $$p \leq 0.0009$$), and percentage of astrocytes containing at least one lipid body (Control 57.14 ± $9.524\%$ of astrocytes vs Far AD 83.87 ± $6.715\%$ of astrocytes, $$p \leq 0.0424$$) in the APP-PS1 mice compared to C57BL/6J control mice (Fig. 2G–H). Overall, these findings indicate that astrocytes in the stratum radiatum of 20-month-old APP-PS1 male mice exhibit more mature lysosomes, accumulated lipid bodies, and increased interactions with synaptic elements compared to age-matched C57BL/6J mice (see Tables 1 and 2).Fig. 2Intracellular contents of typical astrocytes in the stratum radiatum. Representative 5 nm per pixel of resolution scanning electron microscopy images acquired in the ventral hippocampus CA1 stratum radiatum of 20-month-old APP-PS1 (far from Aß plaques/dystrophic neurites) and age-matched C57BL/6J male mice (A, B). Quantitative graphs representing the number of primary lysosomes (C), tertiary lysosomes (E), and lipid bodies (G) are provided. Quantitative graphs represent the relative proportion of cells positive for primary lysosomes (D), tertiary lysosomes (F), and lipid bodies (H). Data are shown as individual dots and are expressed as mean ± S.E.M. $p \leq 0.05$, *$p \leq 0.001$, using a non-parametric Mann–Whitney test. Statistical tests were performed on $$n = 8$$–11 astrocytes per animal in $$n = 3$$ mice/group, for a total of 59 cell bodies analyzed. Red outline = plasma membrane, yellow outline = nuclear membrane, blue arrow = primary lysosomes, green arrow = secondary lysosomes, orange arrow = tertiary lysosomes, orange pseudo-coloring = lipid bodiesTable 1Absolute ultrastructural analysis of typical astrocytes far from Aß plaques/dystrophic neurites and in aged APP-PS1 mice compared to age-matched C57BL/6 mice in the stratum radiatum of the ventral hippocampus CA1ControlMean ± SEM(Min–Max)ADMean ± SEM(Min–Max)Primary lysosomes (n)1.179 ± 0.2523(0.000–5.000)0.6129 ± 0.1950(0.000–4.000)Secondary lysosomes (n)1.607 ± 0.4435(0.000–10.00)1.032 ± 0.2475(0.000–5.000)Tertiary lysosomes (n)1.464 ± 0.3313(0.000–6.000)2.742 ± 0.3934 (0.000–8.000)All lysosomes (n)4.179 ± 0.6384(0.000–12.00)4.032 ± 0.5893(0.000–12.00)Lipid bodies (n)1.429 ± 0.4224(0.000–11.00)4.161 ± 0.7706 **(0.000–19.00)Altered mitochondria* (n)2.356 ± 0.2780(0.000–6.000)2.387 ± 0.2847(0.000–6.000)Elongated mitochondria (n)3.286 ± 0.4903(0.000–10.00)3.419 ± 0.5283(0.000–11.00)Partially digested phagosomes (n)2.857 ± 0.5213(0.000–12.00)3.710 ± 0.5230(0.000–11.00)Fully digested phagosomes (n)2.929 ± 0.4482(0.000–7.000)4.258 ± 0.7139(0.000–14.00)All phagosomes (n)5.786 ± 0.7226(0.000–15.00)7.968 ± 1.160(0.000–24.00)Association with myelinated axons (n)1.464 ± 0.4755(0.000–9.000)1.065 ± 0.3437(0.000–9.000)Axon terminals (n)16.25 ± 1.729(4.000–42.00)22.29 ± 2.089 (6.000–43.00)Dendritic spines (n)1.429 ± 0.2020(0.000–4.000)2.194 ± 0.2384 (0.000–4.000)All synaptic contacts (n)20.68 ± 2.244(0.000–54.00)29.06 ± 2.587 (8.000–54.00)Dilated ER (n)4.143 ± 0.5767(1.000–13.00)5.129 ± 0.6255(0.000–16.00)Autophagosomes (n)0.5000 ± 0.1585(0.000–3.000)0.7742 ± 0.1518(0.000–3.000)Cell area* (µm2)57.65 ± 3.380(35.58–107.1)64.01 ± 5.421(26.18–160.1)Cytoplasm area (µm2)31.31 ± 3.143(12.81–84.33)38.54 ± 4.795(9.553–142.5)Nucleus area (µm2)26.34 ± 1.598(10.05–41.75)25.47 ± 1.973(8.942–53.39)Cell perimeter (µm)62.65 ± 3.603(33.26–111.6)66.83 ± 3.841(27.89–104.0)Nucleus perimeter (µm)21.80 ± 1.036(12.30–36.71)20.65 ± 0.9511(12.01–28.53)Circularity (a.u.)0.2114 ± 0.01728(0.08100–0.4420)0.1977 ± 0.01601(0.08200–0.4680)AR (a.u.)2.178 ± 0.1097(1.163–3.610)2.467 ± 0.1657(1.206–4.273)Solidity (a.u.)0.6219 ± 0.02208(0.4170–0.8290)0.6157 ± 0.02169(0.3600–0.8690)n number, a.u. arbitrary unit, ER endoplasmic reticulum, and p-values of statistically significant tests are highlighted with an asterisk symbolData reported are shown as number per cell and expressed as means ± SEM in addition to the minimum and maximum value obtained$p \leq 0.05$, $p \leq 0.001$ using a non-parametric Mann–Whitney test. Statistical tests were performed on $$n = 8$$–11 astrocytes per animal in $$n = 3$$ mice/group, for a total of 59 cell bodies analyzedTable 2Relative ultrastructural analysis of typical astrocytes far from Aß plaques/dystrophic neurites in aged APP-PS1 mice compared to age-matched C57BL/6 mice in the stratum radiatum of the ventral hippocampus CA1ControlMean ± SEMFar ADMean ± SEMPrimary lysosomes (%)57.14 ± 9.52432.36 ± 8.535Secondary lysosomes (%)46.43 ± 9.59851.61 ± 9.124Tertiary lysosomes (%)53.57 ± 9.59877.42 ± 7.634Lipid bodies (%)57.14 ± 9.52483.87 ± 6.715Altered mitochondria* (%)85.71 ± 6.73490.32 ± 5.398Elongated mitochondria (%)96.43 ± 3.57190.32 ± 5.398Glycogen granules (%)75.00 ± 8.33387.10 ± 6.121Dilated ER (%)100.0 ± 0.00096.77 ± 3.226Nuclear indentation (%)10.71 ± 5.95216.13 ± $6.715\%$ percent, a.u. arbitrary unit, ER endoplasmic reticulumData reported are shown as % of cells positive for at least one of the elements analyzed for each category and expressed as means ± SEM. The statistical test performed was a non-parametric Mann–Whitney test with $p \leq 0.05.$ Statistical tests were performed on $$n = 8$$–11 astrocytes per animal in $$n = 3$$ mice/group, for a total of 59 cell bodies analyzed ## Typical astrocytes in the hippocampal CA1 stratum lacunosum-moleculare of aged APP-PS1 vs age-matched C57BL/6J mice present increased synaptic contacts and phagolysosomal activity We next pursued our ultrastructural investigation of typical astrocytes in the stratum lacunosum-moleculare (Fig. 3A–C). As in the stratum radiatum, we observed an increased prevalence of direct contacts between astrocytes and dendritic spines in the APP-PS1 mice compared to C57BL/6J control mice (Control 1.342 ± 0.2394 contacts per astrocyte vs AD 2.529 ± 0.3829 contacts per astrocyte, $$p \leq 0.0055$$) (Fig. 3D). When discriminating further the astrocytes based on their proximity to Aß plaques/dystrophic neurites [far (Far AD) vs near (Near AD) Aß plaques/dystrophic neurites], astrocytes near Aß plaques/dystrophic neurites were found to be mainly responsible for these increased contacts with dendritic spines (Control 1.342 ± 0.2394 contacts per astrocyte vs Near AD 2.5821 ± 0.4219 contacts per astrocyte, $$p \leq 0.0240$$) (Fig. 3E). In addition, we observed an overall reduction in the direct contacts with synapses for astrocytes located far versus near Aß plaques/dystrophic neurites (Far AD 13.45 ± 1.261 contacts per astrocyte vs Near AD 19.35 ± 2.071 contacts per astrocyte, $$p \leq 0.0340$$) (Fig. 3E).Fig. 3Parenchymal interactions of typical astrocytes and shape descriptors in the stratum lacunosum-moleculare. Representative 5 nm per pixel of resolution scanning electron microscopy images acquired in the ventral hippocampus CA1 stratum lacunosum-moleculare of 20-month-old C57BL/6J male mice (A) and APP-PS1 male mice far (B) and near (C) Aß plaques/dystrophic neurites. Quantitative graphs represent the number of direct contacts with dendritic spines (D) per genotype (APP-PS1 vs C57BL/6J), and contacts with dendritic spines (E) and with synaptic elements (F) when spatially separating astrocytes between locations near vs far Aß plaques/dystrophic neurites. Quantitative graphs represent the shape descriptors of the astrocytic cell bodies, including G cytoplasmic area, H nucleus area, I cytoplasmic perimeter, and J nucleus perimeter. Data are shown as individual dots and are expressed as means ± S.E.M. $p \leq 0.05$, *$p \leq 0.01$, using a non-parametric Mann–Whitney test for the comparison of dendritic spines in D, and a Kruskal–Wallis test with a Dunn’s post hoc for all other graphs shown. Statistical tests were performed on $$n = 8$$–12 astrocytes per animal in $$n = 3$$ mice/group, for a total of 102 cell bodies analyzed. red outline = plasma membrane, yellow outline = nuclear membrane, blue pseudo-coloring = axon terminals, orange pseudo-coloring = dendritic spines However, unlike the astrocytes analyzed in the stratum radiatum, we measured in the current layer an increase in the area and perimeter of both the cytoplasm and nucleus for astrocytes located near versus far from Aß plaques/dystrophic neurites, which could at least partly explain their increased prevalence of synaptic interactions (cytoplasmic area without nucleus—Far AD 18.02 ± 1.731 µm2 vs Near AD 29.91 ± 3.601 µm2, $$p \leq 0.0083$$; nuclear area—Far AD 18.64 ± 1.822 µm2 vs Near AD 26.23 ± 2.486 µm2, $$p \leq 0.0289$$; cytoplasmic perimeter—Far AD 42.30 ± 2.750 µm vs Near AD 62.73 ± 6.112 µm, $$p \leq 0.0057$$; nucleus perimeter—Far AD 17.72 ± 0.9569 µm vs Near AD 23.07 ± 1.685 µm, $$p \leq 0.0135$$) (Fig. 3G–J). These differences are in line with the findings from previous studies that highlight an atrophy of astrocytes observed far from Aß plaques compared to their hypertrophy near Aß plaques in mouse models of AD pathology [104–106], a morphological shift suggested to be associated with the appearance of Aß plaques within their microenvironment [104]. In terms of intracellular contents, our analysis of typical astrocytes located in the stratum lacunosum-moleculare (Fig. 4A–C) further revealed a tendency for an increase in all phagosomes (fully and partially digested phagosomes) in the APP-PS1 mice compared to C57BL/6J control mice (Control 5.368 ± 0.6191 phagosomes per astrocyte vs AD 8.5588 ± 1.364 phagosomes per astrocyte, $$p \leq 0.0590$$) (Fig. 4D). When we investigated the driving force behind this tendency (i.e., near vs far from Aß plaques/dystrophic neurites), we found a significant increase in all phagosomes (fully and partially digested phagosomes) only in astrocytes near Aß plaques/dystrophic neurites, compared to both astrocytes far from Aß plaques/dystrophic neurites in APP-PS1 mice and astrocytes in C57BL/6J control mice (Control 5.368 ± 0.6191 phagosomes per astrocyte vs Near AD 10.74 ± 1.573 phagosomes per astrocyte, $$p \leq 0.0019$$; Far AD 5.970 ± 0.7010 phagosomes per astrocyte vs Near AD 10.74 ± 1.573 phagosomes per astrocyte, $$p \leq 0.0160$$) (Fig. 4E). This increased number of phagosomes per astrocyte located near Aß plaques/dystrophic neurites was identified specifically for the fully digested phagosomes (Control 2.842 ± 0.3781 phagosomes per astrocyte vs Near AD 6.258 ± 1.017 phagosomes per astrocyte, $$p \leq 0.0016$$; Far AD 3.061 ± 0.4766 phagosomes per astrocyte vs Near AD 6.258 ± 1.017 phagosomes per astrocyte, $$p \leq 0.0046$$) (Fig. 4F). In short, both the strata lacunosum-moleculare and radiatum showed an increased activity of the phagolysosomal pathway in aged 20-month-old APP-PS1 male mice compared to age-matched C57BL/6J controls, resulting in an increased prevalence of mature lysosomes and fully digested phagosomes, respectively. Fig. 4Intracellular contents of typical astrocytes in the stratum lacunosum-moleculare. Representative 5 nm per pixel of resolution scanning electron microscopy images acquired in the ventral hippocampus CA1 stratum lacunosum-moleculare of 20-month-old C57BL/6J male mice (A) and APP-PS1 male mice far (B) and near (C) Aß plaques/dystrophic neurites. Quantitative graphs representing the number of phagosomes D per genotype (APP-PS1 vs C57BL/6J) and E per proximity to Aß plaques/dystrophic neurites. The number of fully digested phagosomes per astrocytic cell body based on the proximity to Aß plaques/dystrophic neurites is shown in F. Quantitative graphs represent the number of cells positive for glycogen granules per genotype (G) and per proximity to Aß plaques/dystrophic neurites (H). Data are shown as individual dots and are expressed as mean ± S.E.M. $p \leq 0.05$, $p \leq 0.01$, $p \leq 0.001$, **$p \leq 0.0001$ using a non-parametric Mann–Whitney test for the comparison of phagosomes (D) and glycogen granules (H), and a Kruskal–Wallis test with a Dunn’s post hoc for all other graphs shown. Statistical tests were performed on $$n = 8$$–12 astrocytes per animal in $$n = 3$$ mice/group, for a total of 102 cell bodies analyzed. Red outline = plasma membrane, yellow outline = nuclear membrane, red arrow = glycogen granules, yellow pseudo-coloring = fully digested phagosomes In our analysis, we also examined glycogen granules, a carbohydrate storage that can be broken down to glucose through glycolysis, and which was shown to be crucial in astrocytes for learning and memory [107, 108], and associated with aging in human brain samples [109]. Glycogen granules were shown to be located within astrocytic processes, notably those in proximity to dendritic spines and axon terminals in the hippocampus and sensorimotor cortex of rodents [110–112]. In the current study, while there were no differences detected in the stratum radiatum, more astrocytes in the stratum lacunosum-moleculare were found to contain glycogen granules in the APP-PS1 mice compared to C57BL/6J control mice (Control 7.895 ± $4.433\%$ of astrocytes vs AD 47.06 ± $8.689\%$ of astrocytes, $$p \leq 0.0002$$). Indeed, close to half of all astrocytes in APP-PS1 mice were positive for glycogen granules compared to nearly $8\%$ in C57BL/6J control mice (Fig. 4G). When astrocytes were spatially separated between locations near vs far Aß plaques/dystrophic neurites, the majority of astrocytes with glycogen granules were found near versus far from Aß plaques/dystrophic neurites (Control 7.895 ± $4.433\%$ of astrocytes vs Near AD 64.52 ± $8.736\%$ of astrocytes, $p \leq 0.0001$; Far AD 12.12 ± $5.770\%$ of astrocytes vs Near AD 64.52 ± $8.736\%$ of astrocytes, $p \leq 0.0001$) (see Tables 3 and 4).Table 3Absolute ultrastructural analysis of typical astrocytes near vs far from Aß plaques/dystrophic neurites in aged APP-PS1 mice compared to age-matched C57BL/6 mice in the stratum lacunosum-moleculare of the ventral hippocampus CA1C57BL/6JAPP-PS1ControlMean ± SEM(Min–Max)Far ADMean ± SEM(Min–Max)Near ADMean ± SEM(Min–Max)Primary lysosomes (n)0.6316 ± 0.1432(0.000–3.000)0.6061 ± 0.1737(0.000–4.000)0.5484 ± 0.1379(0.000–3.000)Secondary lysosomes (n)1.211 ± 0.2776(0.000–7.000)0.6364 ± 0.1837(0.000–4.000)0.8065 ± 0.2384(0.000–6.000)Tertiary lysosomes (n)0.4737 ± 0.1398(0.000–4.000)0.6061 ± 0.1737(0.000–3.000)0.6774 ± 0.2194(0.000–4.000)All lysosomes (n)2.316 ± 0.3580(0.000–9.000)1.848 ± 0.3202(0.000–6.000)2.032 ± 0.3724(0.000–9.000)Lipid bodies (n)2.184 ± 0.4663(0.000–11.00)2.909 ± 0.7049(0.000–14.00)3.839 ± 0.9860(0.000–22.00)Altered mitochondria* (n)0.8947 ± 0.1545(0.000–3.000)0.6061 ± 0.1625(0.000–4.000)1.065 ± 0.2172(0.000–4.000)Elongated mitochondria (n)2.579 ± 0.4869(0.000–17.00)1.939 ± 0.2818(0.000–6.000)3.129 ± 0.6077(0.000–13.00)All mitochondria (n) 13.39 ± 1.095(1.000–40.00)12.79 ± 1.004(2.000–26.00) 19.55 ± 2.270(1.000–57.00)Partially digested phagosomes (n)2.526 ± 0.3533(0.000–10.00)2.909 ± 0.4091(0.000 ± 8.000)4.484 ± 0.7212(0.000–17.000)Fully digested phagosomes (n)2.842 ± 0.3781(0.000–9.000)3.061 ± 0.4766(0.000–10.00)6.258 ± 1.017 &&!!( 0.000–27.00)All phagosomes (n)5.368 ± 0.6191(0.000–19.00)5.970 ± 0.7010(1.000–15.00)10.74 ± 1.573 &&!( 1.000–44.00)Association with myelinated axons (n)1.763 ± 0.4953(0.000–14.00)1.576 ± 0.4013(0.000–9.000)1.258 ± 0.4121(0.000–10.00)Axon terminals (n)9.579 ± 0.8614(2.000–31.00)8.727 ± 1.026(1.000–31.00)12.26 ± 1.392(2.000–33.00)Dendritic spines (n) 1.342 ± 0.2394(0.000–6.000)1.848 ± 0.2579(0.000–6.000)2.581 ± 0.4219 &(0.000–10.00)All synaptic contacts (n)14.50 ± 1.245(4.000–46.00)13.45 ± 1.261(3.000–38.00)19.35 ± 2.071!(4.000–51.00)Dilated ER (n)2.368 ± 0.4967(0.000–16.00)1.303 ± 0.2769(0.000–7.000)3.129 ± 0.9120(0.000–20.00)Non-dilated ER (n)16.82 ± 1.976(2.000–73.00)12.30 ± 1.253(4.000–29.00)18.06 ± 2.313(2.000–48.00)Dilated Golgi apparatus (n)1.000 ± 0.2444(0.000–8.000)0.2727 ± 0.007873(0.000–1.000)0.7097 ± 0.1552(0.000–3.000)Non-dilated Golgi apparatus* (n)2.184 ± 0.4265(0.000–13.00)1.636 ± 0.2254 #(0.000–4.000)2.968 ± 0.4436(0.000–9.000)Autophagosomes (n)0.2105 ± 0.09359(0.000–3.000)0.5758 ± 0.1742(0.000–4.000)0.2903 ± 0.09497(0.000–2.000)Cell area (µm2)*42.53 ± 2.774(20.19–120.0)36.66 ± 2.803(14.73 ± 73.60)56.15 ± 4.641!!!(16.77–130.2)Cytoplasmic area* (µm2) *20.32 ± 2.584(8.028–106.5)18.02 ± 1.731(5.983–51.94)29.91 ± 3.601!!(5.633–89.16)Nucleus area* (µm2)22.21 ± 1.290(10.12–41.16)18.64 ± 1.822(4.045–46.22)26.23 ± 2.486!(4.264–61.13)Cell perimeter (µm)49.50 ± 2.905(24.79–130.6)42.30 ± 2.750(21.78–83.58)62.73 ± 6.112!!(22.71–184.5)Nucleus perimeter (µm)21.35 ± 0.6547(13.53–28.32)17.72 ± 0.9569(7.461–30.47)23.07 ± 1.685(7.796–44.02)AR (a.u.)2.270 ± 0.1555(1.051–5.068)2.217 ± 0.1752 #(1.087–5.509)2.295 ± 0.1771!(1.204–5.750)Circularity (a.u.)0.2468 ± 0.01565(0.08800–0.4980)0.2940 ± 0.02280(0.1090–0.6330)0.2483–0.02722(0.04800–0.5960)Solidity (a.u.)0.6741 ± 0.02028(0.3550–0.8990)0.7146 ± 0.02278(0.4140–0.8980)0.6930 ± 0.02512(0.4070–0.9560)n number, a.u. arbitrary unit, ER endoplasmic reticulum, and p-values of statistically significant tests are highlighted with various symbols (!, #, &) Data reported are shown as number per cell and expressed as means ± SEM in addition to the minimum and maximum value obtained$p \leq 0.05$, $p \leq 0.01$, $p \leq 0.001$ using a Kruskal–Wallis test with a Dunn’s multiple comparisons post hoc test. p value summary,! Near vs Far AD, & Far AD vs C57BL/6J, # Far AD vs C57BL/6J. Statistical tests were performed on $$n = 8$$–12 astrocytes per animal in $$n = 3$$ mice/group, for a total of 102 cell bodies analyzedTable 4Relative ultrastructural analysis of typical astrocytes near vs far from Aß plaques/dystrophic neurites in aged APP-PS1 mice compared to age-matched C57BL/6 mice in the stratum lacunosum-moleculare of the ventral hippocampus CA1C57BL/6JAPP-PS1ControlMean ± SEMFar ADMean ± SEMNear ADMean ± SEMPrimary lysosomes (%)39.47 ± 8.03633.33 ± 8.33341.94 ± 9.009Secondary lysosomes (%)50.00 ± 8.22036.36 ± 8.50445.16 ± 9.086Tertiary lysosomes (%)31.58 ± 7.64230.30 ± 8.12429.03 ± 8.287Lipid bodies (%)55.26 ± 8.17451.52 ± 8.83561.29 ± 8.893Altered mitochondria (%)55.26 ± 8.17442.42 ± 8.73761.29 ± 8.893Elongated mitochondria (%)84.21 ± 5.99581.82 ± 6.81877.42 ± 7.634Glycogen granules (%) ***7.895 ± 4.43312.12 ± 5.77064.52 ± 8.736 &&&&!!!! Dilated ER (%)71.05 ± 7.45663.64 ± 8.50480.65 ± 7.213Nuclear indentation (%)13.16 ± 5.5573.030 ± 3.03019.35 ± $7.213\%$ percent, a.u. arbitrary unit, ER endoplasmic reticulum, p-values of statistically significant tests are highlighted with various symbols (!, &) with indicating p value summaryData reported are shown as % of cells positive for at least one of the elements analyzed for each category and expressed as means ± SEM***$p \leq 0.0001$ using a Kruskal–Wallis test with Dunn’s multiple comparisons post hoc test. p value summary,! Near vs Far AD, & Far AD vs C57BL/6J. Statistical tests were performed on $$n = 8$$–12 astrocytes per animal in $$n = 3$$ mice/group, for a total of 102 cell bodies analyzed ## Dark astrocytes in the hippocampal CA1 stratum lacunosum-moleculare of aged APP-PS1 vs age-matched C57BL/6J mice present similar densities and interactions with the vasculature While imaging in aged APP-PS1 and C57BL/6J mice, we identified an electron-dense astrocytic state based on their distinct ultrastructural features and located often near the vasculature. We confirmed that the dark astrocytes in the ventral hippocampus CA1 were also immunopositive for GFAP, a marker generally associated with astrocytes termed ‘reactive’ (Fig. 5A–C). Dark astrocytes were previously observed both in rodents (e.g., rat models of brain injury, kainic and pentylenetetrazole treatments, electroshock; mouse embryonic spinal cord culture) [65–67] and human post-mortem brain samples (e.g., brain tumors, brain injury) [61–64]. These cells were described as having hypertrophic electron-dense cell bodies and processes often containing altered mitochondria and glycogen granules [61, 62, 65]. While their roles have remained largely elusive, we further confirmed the presence of a similar electron-dense astrocytic state in the ventral hippocampus CA1 of 20-month-old APP-PS1 and C57BL/6J male mice. We then performed quantitative analysis of their distribution (Fig. 6A–C), and examined whether dark astrocytes interacted more or less often with blood vessels in the stratum lacunosum-moleculare of APP-PS1 vs C57BL/6J mice, as vascular dysfunction was previously noted in the hippocampus during aging and AD pathology, using human and mouse samples [113–117].Fig. 5Immunostaining for GFAP in typical and dark astrocytes of the stratum lacunosum-moleculare. Representative 5 nm per pixel and 1 nm per pixel of resolution scanning electron microscopy images showing a typical (A) and dark astrocyte (B, C) immunostained with glial fibrillary acidic protein (GFAP) in the ventral hippocampus CA1 stratum lacunosum-moleculare of 20-month-old APP-PS1 male mice. In A a typical astrocyte, denoted by its electron-lucent cyto- and nucleoplasm, is immunopositive for GFAP. In B an electron-dense dark astrocyte with hyper-ramifications and several tertiary lysosomes is immunostained for GFAP. In C a close-up of the dark astrocyte where the GFAP staining is indicated with an orange arrow. Yellow outline = nuclear membrane, purple outline = dark astrocytic cytoplasm, red outline = typical astrocytic cytoplasm, orange arrow = GFAP immunostaining in dark astrocyte, pink pseudo-coloring = dystrophic neurites, purple pseudo-coloring = amyloid beta plaques, 3rd = tertiary lysosomesFig. 6Density of dark and typical astrocytes in the stratum lacunosum-moleculare. Representative 25 nm per pixel (A) and 5 nm per pixel (B, C) of resolution scanning electron microscopy images of dark astrocytes associated with blood vessels (B) and with the parenchyma (C) from a 20-month-old APP-PS1 male mouse. Quantitative graphs represent the astrocytic density defined ultrastructurally (e.g., via their intermediate filaments, angular processes) and their electron-dense ultrastructure (dark) or electron-lucent (typical) appearance (D) in 20-month-old C57BL/6J vs APP-PS1 male mice. Typical and dark astrocytes in these mice were further categorized based on their association (E) or lack of association (F) with blood vessels in the plane of view. The ratio of dark astrocytic cells associated with a blood vessel and overall astrocytes (typical and dark) associated with the vasculature is represented (G), while the ratio of all dark astrocytes not associated with blood vessels over all astrocytes not associated with blood vessels (typical and dark) is provided (H). Data are shown as individual dots and are expressed as mean ± S.E.M using a Welsh test. Statistical tests were performed on $$n = 4$$ mice/group (2–6 levels per animal). Green pseudo-coloring = dark astrocyte associated with blood vessels, purple pseudo-coloring = dark astrocyte not associated with blood vessels Dark astrocytes were not found exclusively in aged APP-PS1 mice, as they were also observed in age-matched C57BL/6J controls (Control 5.518 ± 1.546 cells per mm2 vs AD 14.190 ± 4.861 cells per mm2, $$p \leq 0.1721$$) (Fig. 6D). Typical astrocytes also did not display significant differences in their density between the two genotypes (Control 207.1 ± 12.89 cells per mm2 vs AD 229.6 ± 8.802 cells per mm2, $$p \leq 0.2056$$). In addition, most of the dark astrocytes observed in both conditions were in direct contact with blood vessels (Control 4.265 ± 1.976 cells per mm2 vs AD 7.666 ± 2.528 cells per mm2, $$p \leq 0.3323$$) while the density of dark astrocytes not touching a blood vessel in the plane of view was lower (Control 1.757 ± 1.230 cells per mm2 vs AD 3.795 ± 1.550 cells per mm2, $$p \leq 0.3448$$) (Fig. 6E–F). Typical astrocytes contacting the basement membrane of a blood vessel were similarly abundant in APP-PS1 mice and C57BL/6J controls (Control 50.02 ± 6.584 cells per mm2 vs AD 51.26 ± 7.142 cells per mm2, $$p \leq 0.9028$$), and the same finding was obtained for typical astrocytes that did not contact a blood vessel in the plane of view (Control 157.0 ± 7.386 cells per mm2 vs AD 176.0 ± 12.92 cells per mm2, $$p \leq 0.2614$$). When we looked at the ratio of dark astrocytes over all astrocytes in direct contact with a blood vessel, this dark state presented equivalent ratios in APP-PS1 mice and C57BL/6J controls (Control 6.309 ± $3.415\%$ of dark astrocytes vs AD 12.28 ± $4.489\%$ of dark astrocytes, $$p \leq 0.3332$$). Similar results were obtained for dark astrocytes that were not directly contacting a blood vessel (Control 0.9455 ± $0.5998\%$ of dark astrocytes vs AD 2.690 ± $1.278\%$ of dark astrocytes, $$p \leq 0.2802$$) (Fig. 6G–H). Overall, there were no significant differences in the density and ratio of dark astrocytes interacting vs non-interacting with a blood vessel between aged APP-PS1 and age-matched C57BL/6J mice, indicating that the distribution of these cells at the vasculature and throughout the parenchyma is shared between aging and AD pathology. Moreover, we also observed dark astrocytes in 3- to 4-month-old C57BL/6J mice within the same region, the ventral hippocampus CA1 stratum lacunosum-moleculare (Fig. 7D). While the abundance of these dark astrocytes remains to be quantified over time to determine whether they become more abundant during aging, our results suggest that these cells are not exclusive to aging while their appearance is not driven by AD pathology. Fig. 7Ultrastructural characterization of dark and typical astrocytes. Representative 5 nm per pixel of resolution scanning electron microscopy images of dark and typical astrocytes acquired in the ventral hippocampus CA1 stratum lacunosum-moleculare of 3- to 4-month-old C57BL/6J male mice (A and C) and stratum lacunosum-moleculare of aged APP-PS1 20-month-old male mice (B–G). In A′, red arrows identify the electron-dense interface between two typical astrocytic elements filled with gap junctions. In A″, black arrow identifies intermediate filaments. In B′, a typical astrocyte makes direct contact with dendritic spines and axon terminals. An angular protuberance is identified with a yellow arrow. In B″, the red arrows identify the electron-dense interface. In C′, the electron-dense interface between two dark astrocytic elements is highlighted with a red arrow. In C″, direct contact of dark astrocytes with a dendritic spine is shown with a blue arrow, comparable to the interaction of synaptic elements and typical astrocytes in B″. In D′, the electron-dense interface between two dark astrocytic end-feet is shown with the red arrows. In D″, glycogen granules identified by white arrows, as well as several contacts with dendritic spines and axon terminals. In E and F, G typical and dark astrocytes, respectively, are located near amyloid beta plaques and dystrophic neurites. Dilated Golgi apparatus cisternae identified by a blue arrow are observed. Several lysosomes identified with an asterisk and internalized dystrophic neurites and dendritic spines are shown. Yellow outline = nuclear membrane, purple outline = dark astrocytic cytoplasm, red outline = typical astrocytic cytoplasm, green outline = basement membrane, red arrow = interface between two astrocytic elements, black arrow = intermediate filaments, white arrow = glycogen granules, blue arrow = dilated Golgi apparatus, yellow arrow = angular processes, white asterisk = lysosomes, orange pseudo-coloring = dendritic spines, pink pseudo-coloring = dystrophic neurites, blue pseudo-coloring = axon terminals, purple pseudo-coloring = amyloid beta plaques ## Dark astrocytes in the strata lacunosum-moleculare and radiatum of 3- to 4-month-old C57BL/6J and 20-month-old APP-PS1 male mice exhibit similar ultrastructural features as typical astrocytes while displaying distinct characteristics We next performed an ultrastructural characterization of the dark astrocytes among the ventral hippocampus CA1 strata lacunosum-moleculare and radiatum. We examined the ultrastructural features of dark and typical astrocytes in young (3- to 4-month-old) C57BL/6J male mice and aged (20-month-old) APP-PS1 mice (Fig. 7). This qualitative analysis revealed numerous similarities between the two astrocytic states, which also displayed distinct characteristics, in homeostatic and pathological conditions. Typical astrocytes possessed angular processes [24, 74, 118], often interacting with synaptic elements, and where a dark interface could be observed between two astrocytic end-feet (electron-dense due to being filled with gap junctions) [119–122] (Fig. 7A, B). We observed similar features in the dark astrocytes associated with blood vessels and other parenchymal elements, both for locations near and far from Aß plaques/dystrophic neurites. Indeed, dark astrocytes displayed the same angular processes, which often inserted themselves between pre- and post-synaptic elements of the same synapse (Fig. 7C, D). In addition, we found the same electron-dense interface between the end-feet of two dark perivascular astrocytes, together with a high accumulation of glycogen granules dispersed throughout their cytoplasm (Fig. 7C, D). Dark astrocytes, similar to dark microglia, were further characterized by their electron-dense cytoplasm and nucleoplasm, alongside a partial to total loss of their chromatin pattern, both in 3- to 4-month-old C57BL/6J (Fig. 7C) and 20-month-old APP-PS1 mice (Fig. 7D). In the parenchyma of aged APP-PS1 mice, dark astrocytes, much like their typical counterparts, were seen interacting extensively with axon terminals, dystrophic neurites, and dendritic spines (Fig. 7E–G). These dark cells were also seen internalizing dystrophic neurites, a feature that was previously observed in typical astrocytes near Aß plaques/dystrophic neurites in 6- and 12-month-old APP-PS1 mice [53]. In addition, several phagosomes, mostly containing axon terminals and dendritic spines, were observed within the dark astrocytic cytoplasm, alongside tertiary lysosomes and lipid bodies (Fig. 7E–G). In our imaging, we captured a single dark astrocyte associated with a blood vessel in the stratum lacunosum-moleculare of an aged APP-PS1 mouse at multiple levels, with each image acquired at a distance of 5–6 µm (Additional file 1: Fig. S1). We identified several phagosomes in the dark astrocytic cytoplasm, most of which were dendritic spines and axon terminals, a feature we previously observed in the dark astrocytes found within the parenchyma. Extensive interactions between the dark astrocyte and synaptic elements were observed in the serial images, a feature attributed in part to their thin and angular processes extending among the parenchyma. Like all dark astrocytes imaged, the presence of glycogen granules was observed throughout the images taken of this particular dark astrocytic cell. Dark astrocytes were further characterized by their markers of cellular stress. Notably, they contained altered mitochondria with swollen cristae that were identified ultrastructurally by an electron-lucent enlargement, mitochondria with a degraded outer membrane, as well as dilated ER and Golgi apparatus cisternae (Fig. 7E–G). The increased electron density within microglial cells was previously hypothesized to be due to the condensation of their cytoplasm related to cellular stress [123]. In addition, Tòth et al. hypothesized that the electron density begins at a specific point in the dark astrocytes where it propagates thereafter throughout the cell [65]. Our observations could support this idea as we found that some astrocytic compartments in direct contact with an Aß plaque and containing fibrillar Aß possessed a more electron-dense appearance (Fig. 8A, B). Therefore, this data could support the view that the electron density of dark astrocytes starts at a specific point which could then spread to the rest of the cell. Fig. 8Dark astrocytes are associated with AD hallmarks. Representative 5 nm per pixel of resolution scanning electron microscopy images of a dark astrocyte (A, B) in the ventral CA1 hippocampus stratum lacunosum-moleculare of 20-month-old APP-PS1 male mice. In A and B the astrocyte is directly interacting with an Aß plaque where a specific segment (shown in B with a green bar) is becoming electron-dense compared to the rest of the cell. Yellow outline = nuclear membrane, purple outline = dark astrocytic cytoplasm, red outline = typical astrocytic cytoplasm, pink pseudo-coloring = dystrophic neurites, purple pseudo-coloring = amyloid beta plaques, green bar = electron-dense area Intriguingly, we often identified dark astrocytes next to a blood vessel interacting with other dark astrocytic cell bodies (Fig. 7C), typical astrocytic bodies (Fig. 7A), in addition to microglial cell bodies (Fig. 9A). While it is still unknown why dark astrocytes often come in close contact with microglial and astrocytic cell bodies, occupying satellite positions, typical astrocytes are known to interact with juxtavascular microglia [95, 124], in addition to contacting neighboring astrocytes notably through their complex and branched processes [119]. Overall, we examined the dark astrocytic state for the first time in adulthood, as well as in aged AD pathology, and found that it is characterized by the presence of glycogen granules, several markers of cellular stress, increased phagocytic capabilities (e.g., abundance of mature lysosomes and numerous phagosomes), a unique electron-dense cytoplasm and nucleoplasm, and a partial to total loss of the nuclear chromatin pattern. Fig. 9Blood vessel-associated dark astrocyte in the stratum lacunosum-moleculare. Representative 5 nm per pixel of resolution scanning electron microscopy images of dark astrocytes acquired in the ventral hippocampus CA1 stratum lacunosum-moleculare of 20-month-old APP-PS1 male mice. In A–A″, a dark astrocyte associated with the vasculature (pseudo-colored in red) is directly interacting with a typical microglial cell body. The red arrows in A″ further indicate the close interaction between the two glial cells. Dark blue outline = dark astrocytic cytoplasm, light blue outline = microglial cytoplasm, yellow = nuclear membrane, red pseudo-coloring = blood vessel, red arrow = direct interaction between a microglial cell body and a dark astrocyte ## Dark astrocytes are observed in the hippocampal head of an aged human post-mortem brain sample Previous studies revealed in the human post-mortem brain following brain injury and brain tumors the presence of a dark astrocytic state [61–64], much like the cells described in the spinal cord cultures of embryonic mice [67] and rat models of electroshock [66], kainic and pentylenetetrazole injection, as well as brain injury [65]. As we observed dark astrocytes in 3- to 4-month-old and 20-month-old C57BL/6J mice and APP-PS1 mice, we further investigated their conservation across species by examining an aged human post-mortem brain sample (female, 81 years old, post-mortem delay 18 h) in the hippocampal head, a region shown to have significant age-related atrophy [98, 100, 101]. Similar to what we uncovered in the mouse brain, we denoted the presence of dark astrocytes that possessed an electron-dense cytoplasm and nucleoplasm in this sample. To the best of our knowledge, this is the first case report that identifies and characterizes this dark astrocytic state among the human hippocampal head in the context of aging. The dark astrocytic cell bodies were seen contacting axon terminals, and their processes were interacting with numerous synapses (both axon terminals and dendritic spines at the same excitatory synapse). Much like typical astrocytes (Fig. 10A), the dark astrocytic cell bodies (Fig. 10B) and their processes (Fig. 10C) also possessed angular protuberances contacting the parenchymal elements and the vasculature. In addition, human dark astrocytes contained several altered mitochondria and dilated ER cisternae, ultrastructural markers of cellular stress which were also previously identified in non-dark astrocytes from human post-mortem parietal cortex samples of patients with AD [125] and in dark astrocytes from human post-mortem samples of brain injury and brain tumors [61–63]. Moreover, in the human dark astrocytes we have examined, several fully digested phagosomes were identified inside the cell body and processes. Fig. 10Typical and dark astrocytes in human post-mortem brain samples. Representative 5 nm per pixel of resolution scanning electron microscopy images of a typical (in A) and dark astrocytes (B, C) in the hippocampal head of an aged female (81-year-old, cause of death—asphyxia, post-mortem delay of 18 h). In A, a typical astrocyte interacts with several axon terminals (pseudo-colored in orange) and myelinated axons (pseudo-colored in yellow). The astrocyte possesses several fully digested phagosomes (pseudo-colored in pink) and altered mitochondria (pseudo-colored in blue). In B a dark astrocyte with several angular processes is making direct contacts with axon terminals (pseudo-colored in orange) and displaying several signs of cellular stress such as altered mitochondria (pseudo-colored in blue) and dilated endoplasmic reticulum (pseudo-colored in purple). In C a dark astrocytic process interacts with axon terminals (pseudo-colored in orange) and dendritic spine (pseudo-colored in green). The dark process contains several fully digested phagosomes (pseudo-colored in pink), altered mitochondria (pseudo-colored in blue), and healthy mitochondria (pseudo-colored in red). Yellow outline = nuclear membrane, green outline = typical astrocytic cytoplasm, purple outline = dark astrocytic cytoplasm, yellow pseudo-coloring = myelinated axons, orange pseudo-coloring = axon terminals, green pseudo-coloring = dendritic spines, blue pseudo-coloring = altered mitochondria, red pseudo-coloring = non-altered mitochondria, pink pseudo-coloring = fully digested phagosomes, purple pseudo-coloring = dilated endoplasmic reticulum ## Discussion Astrocytes which are notably involved in impaired glutamine synthesis, but beneficial for their ability to clear and degrade Aß, and phagocytose dystrophic neurites, were shown to be key players in AD pathology [38, 45, 46, 53, 55, 104, 126, 127]. While investigations of astrocytes in the pathogenesis of AD have gained traction in the last decade, few studies investigated their ultrastructure and to the best of the authors’ knowledge, this is the first quantification of astrocytic intracellular contents and parenchymal interactions by electron microscopy in an aged mouse model of AD pathology. As aging is the most predominant risk factor to developing AD [1], it is crucial to further explore the astrocytic ultrastructure in this context. In addition, as previous studies identified morphological and molecular heterogeneity of astrocytes based on their proximity to Aß plaques and dystrophic neurites [104–106], it is also important to take into account their location to AD hallmarks. In the current study, we first performed an in situ ultrastructural investigation of typical astrocytes, notably based on their distance to Aß plaques/dystrophic neurites, in the ventral hippocampus CA1 strata lacunosum-moleculare and radiatum of 20-month-old APP-PS1 and age-matched C57BL/6J male mice. In a previous study, Sanchez-Mico et al. observed a decrease in phagolysosomal digestion of dystrophic neurites by astrocytes near Aß plaques in the hippocampus of 12-month-old APP751sl mice, which was suggested to result from a reduced astrocytic expression of proteins associated with phagocytosis (Megf10, MerTK) [128]. In our aged mouse model of AD pathology, in the stratum radiatum, typical astrocytes contained more mature tertiary lysosomes but fewer primary lysosomes far from Aß plaques, indicating a shift in maturation of their lysosomal pathway. Yet, while the lysosomes shifted from an immature to a mature appearance, the numbers of fully and partially digested phagosomes within the astrocytic cytoplasm were relatively unchanged between groups. Interestingly, typical astrocytes in the APP-PS1 possessed far more lipid bodies, a feature previously shown to protect neurons against neurotoxicity [129–132]. Indeed, several studies have demonstrated that neurons accumulate unstable lipotoxic elements in the presence of elevated levels of reactive oxygen species (ROS) and altered mitochondria, which are then shuttled to nearby glial cells. This was notably shown in primary mixed glial cells from the olfactory bulb of Apoe−/− male mice, a model used to investigate the function of APOE, followed by an injection of rotenone to increase ROS levels, and in primary astrocytic cultures from ApoE knockout (KO) mice [129, 132]. In inflammatory conditions such as the chronic exposition to noradrenaline or hypoxic stress, primary astrocytic cultures from the neocortex of rats as well as organotypic brain slices from 2- to 4-month-old rats also presented a similar accumulation of lipid droplets, which was suggested to be associated with the protection of neurons from lipotoxicity [133]. We also found that typical astrocytes located in the stratum radiatum of APP-PS1 mice vs C57BL/6J mice interacted more with dendritic spines and axon terminals. Similarly, typical astrocytes near Aß plaques/dystrophic neurites in the stratum lacunosum-moleculare of aged APP-PS1 mice vs C57BL/6J mice contacted more synaptic elements, specifically dendritic spines. In AD pathology, astrocytes were previously reported to negatively influence synaptic numbers, notably via mechanisms that include complement-mediated phagocytosis of synaptic elements [134]. In 6-month-old 5XFAD mice which were exposed to contextual fear conditioning, astrocytes in the dentate gyrus showed a reduced colocalization between PSD95, a marker of post-synaptic density, and GFAP [135], which labels a subset of astrocytes, including ‘reactive’ ones [136, 137]. Synaptic loss near Aß plaques was also reduced in 7- to 13-month-old PS2APP mice, a model of AD pathology, crossed with mice KO for complement 3 (C3) [138], a molecule largely expressed by astrocytes [139]. Similarly, in the hippocampus of 16-month-old APP-PS1 C3 KO mice, levels of synaptic proteins (synapsin-1, synaptophysin, GluR1, PSD95 and Homer1) and pre- and post-synaptic puncta density (in the CA3 specifically, measured using staining for VGlut2 and GluR1, respectively) increased compared to APP-PS1 mice [140], highlighting the astrocytic impact on synaptic loss in AD pathology. As we observed an increase in phagosomes within astrocytes near Aß and dystrophic neurites in the stratum lacunosum-moleculare, it is a possibility that these astrocytes interact more with synaptic elements to phagocytose them. Future studies will be required to confirm this hypothesis, as well as investigate the impact of aging on astrocytic phagocytosis over the course of AD pathology. Another possible explanation for the increase in astrocyte–synapse interactions that we measured in the stratum lacunosum-moleculare of APP-PS1 mice could be the increase in the cytoplasmic perimeter of astrocytes near Aß plaques/dystrophic neurites compared to the ones far from these hallmarks, a morphological difference that was previously reported in the hippocampus of 6- compared to 18-month-old TgF344-AD rats as well as in the dentate gyrus and CA1 of 18-month-old 3xTg mice, both models of AD pathology [105, 106]. Indeed, morphological atrophy (observed far from Aß plaques) vs hypertrophy (in proximity to Aß plaques) was denoted in various brain regions (e.g., hippocampus, cerebral cortex) [104–106]. Future studies will be required, however, to determine the functional implications of these morphological changes. Another interesting feature of typical astrocytes that we found near Aß plaques/dystrophic neurites is their accumulation of glycogen granules. Preferentially located in the astrocytic processes nearby synapses [110], glycogen granules were shown to be involved in learning and memory processes in 3-month-old C57Bl/6N male mice injected in the hippocampus with 1,4-dideoxy-1,4-imino-d-arabinitol, a glycogen phosphorylase inhibitor which blocks glycogenolysis [107, 108]. Both in humans and primates, glycogen accumulation was seen following reperfusion in ischemic stroke, where it was associated with a dysfunctional glycogenolytic pathway, the latter being responsible for the breaking down of glycogen [141]. This increase in glycogen granules was previously associated with the presence of intracellular Aß in astrocytes from post-mortem brain samples of AD patients [142]. This is in line with our observations which identified a high presence of glycogen granules specifically near fibrillar Aß plaques and dystrophic neurites. Astrocytic lactate was shown to be reduced in 6- to 7-month-old female 3xTg mice compared to age-matched controls and was associated with synaptic deficits [143]. A decrease in astrocytic TCA metabolites coupled with functional neuronal excitatory signaling alterations was also previously noted in slices from the hippocampus CA1 of 2- and 4-month-old 5xFAD male mice [144]. Therefore, investigating the glycolytic metabolism disturbances in astrocytes could help better understand their impact on the synaptic dysfunction observed across AD pathology. The concept of glial heterogeneity, notably in neuropathological conditions such as AD, has gained momentum in recent years [55, 56, 145–156, 156–162]. An exponential number of studies using single-cell/nucleus RNA sequencing which aimed to identify unique molecular signatures of glial cells have come out, all pointing toward various clusters of glial cells up- and down-regulating specific gene signatures. Similar techniques were applied to elucidate the transcriptomic heterogeneity of astrocytes in AD pathology, both in mouse models and human post-mortem brain samples, notably identifying the disease-associated astrocytes in mouse models of AD pathology and the reactive astrocytic state in human post-mortem brain samples of patients with AD [55, 56, 159]. These studies have investigated the heterogeneity in astrocytic transcriptomic signatures, leaving an important gap in knowledge pertaining to their ultrastructural heterogeneity in AD pathology. In our in situ investigation of astrocytic heterogeneity, we have identified a unique astrocytic state, the dark astrocytes, combining astrocytic identification criteria with similar ultrastructural features as the dark microglia previously identified in middle-aged and aged APP-PS1 male mice [59, 70]. Dark astrocytes were previously observed in rat models of electroshock [66], compressive and concussive head injury, pentylenetetrazole or kainic acid treatment [65], and spinal cord culture from embryonic mice [67]. Interestingly, unlike the findings of Gallyas et al. [ 66] and Tóth et al. [ 65] which did not observe dark astrocytes in control animals, we observed these electron-dense cells in young and aged C57BL/6J mice (3–4 and 20-month-old) as well as in aged APP-PS1 mice (20-month-old). However, future studies are warranted to quantify these cells over time and determine whether they become more abundant with aging and pathology. Interestingly, these cells have been observed in conditions associated with (neuro)inflammation, such as in AD pathology, as well as following kainic acid intraperitoneal injections and brain injury [65]. In-depth investigation analyzing the effect of the brain’s microenvironment on the appearance of dark astrocytes will be important to perform. Much like the previous observations of dark astrocytes in rodents [65–67] as well as dark microglia, a microglial state associated with an electron-dense cytoplasm and nucleoplasm [59, 60, 81, 163], we observed several signs of oxidative stress such as altered mitochondria, dilated ER and Golgi apparatus, in dark astrocytes. Dark astrocytes were shown to internalize dystrophic neurites, highlighting a potential role for these cells in the pathogenesis of AD. This feature was also identified in typical astrocytes from 6- and 12-month-old APP-PS1 mice [53], alongside several pre- and post-synaptic elements and fibrillar Aß. A full quantification of their intracellular content will help determine if these cells phagocytose more or less of these elements compared to their typical counterparts. Indeed, Sanchez-Mico et al. demonstrated that Aß impaired the ability of astrocytes to phagocytose dystrophic synapses in the hippocampus of 12-month-old APP751sl mice, a model of AD pathology [128]. It remains to be determined if dark astrocytes’ ability to phagocytose dystrophic neurites is also impaired in aged APP-PS1 mice and if this dysfunctional ability is conserved in human post-mortem brain samples. We further observed the presence of dark astrocytes in the hippocampal head of an aged individual, similar to dark astrocytes in the cerebral cortex of male and female post-mortem samples of brain injury and brain tumors [61–63], as well as in brain samples of male and female patients with hemangioblastoma [64]. These dark astrocytes, much like the ones uncovered in mice, possessed signs of cellular stress (altered mitochondria and dilated ER). This conservation, of both the electron-dense state and the oxidative stress markers, across species, denotes similarities between mice and humans: uncovering the mechanism behind the appearance of the dark astrocytes and their function would be key to better understand the diverse, contextually dependent astrocytic response to aging and AD pathology. ## Conclusion We investigated in situ using nanoscale-resolution SEM the ultrastructural alterations in cellular contents and parenchymal interactions of typical astrocytes in aged APP-PS1 and age-matched C57BL/6J male mice. In both examined layers of the hippocampus, we observed increased interactions with synaptic elements along with increased signs of phagolysosomal activity, identifying astrocytic changes linked to AD pathology and their proximity to Aß plaques. Moreover, this ultrastructural study examining astrocytic heterogeneity in aging and AD pathology further characterized a unique astrocytic state, the dark astrocytes, in mice and human post-mortem brain samples. The dark astrocytes displayed markers of cellular stress (e.g., dilated ER and Golgi apparatus), internalized dystrophic neurites (in aged APP-PS1 mice), accumulated glycogen granules within their cytoplasm, and were often located near the vasculature. In addition, we confirmed the conservation of this state in aged human post-mortem brain samples, more specifically among the hippocampal head, highlighting key similarities between species. In short, this study underlines novel ultrastructural alterations of astrocytes in the hippocampus of aged AD pathology, while identifying a dark astrocytic state both in mice and humans. ## Supplementary Information Additional file 1: Figure S1. A blood vessel-associated dark astrocyte imaged serially distances displays several phagosomes. In A–A″, pictures of a dark astrocyte taken serially at a distance of 5–6 µm were acquired. The dark astrocytic cell body, associated with a blood vessel, shows numerous contacts with dendritic spines (pseudo-colored in orange) and axon terminals (pseudo-colored in blue) and contains several partially digested phagosomes (pseudo-colored in purple), notably axon terminals and dendritic spines. Yellow = nuclear membrane, red pseudo-coloring = blood vessel, orange pseudo-coloring = dendritic spine, blue pseudo-coloring = axon terminals, green pseudo-coloring = mitochondria, purple pseudo-coloring = partially digested phagosomes ## References 1. Guerreiro R, Bras J. The age factor in Alzheimer’s disease. Genome Med (2015) 20 106. DOI: 10.1186/s13073-015-0232-5 2. Spires-Jones TL, Hyman BT. The intersection of amyloid beta and tau at synapses in Alzheimer’s disease. Neuron (2014) 82 756-771. DOI: 10.1016/j.neuron.2014.05.004 3. Halliday G. Pathology and hippocampal atrophy in Alzheimer’s disease. Lancet Neurol (2017) 16 862-864. DOI: 10.1016/S1474-4422(17)30343-5 4. Marino S, Bonanno L, Lo Buono V, Ciurleo R, Corallo F, Morabito R. Longitudinal analysis of brain atrophy in Alzheimer’s disease and frontotemporal dementia. J Int Med Res (2019) 47 5019-5027. DOI: 10.1177/0300060519830830 5. Pini L, Pievani M, Bocchetta M, Altomare D, Bosco P, Cavedo E. Brain atrophy in Alzheimer’s disease and aging. Ageing Res Rev (2016) 30 25-48. DOI: 10.1016/j.arr.2016.01.002 6. Terry RD, Masliah E, Salmon DP, Butters N, DeTeresa R, Hill R. Physical basis of cognitive alterations in Alzheimer’s disease: synapse loss is the major correlate of cognitive impairment. Ann Neurol (1991) 30 572-580. DOI: 10.1002/ana.410300410 7. Varma VR, Oommen AM, Varma S, Casanova R, An Y, Andrews RM. Brain and blood metabolite signatures of pathology and progression in Alzheimer disease: a targeted metabolomics study. PLoS Med (2018) 15 e1002482. DOI: 10.1371/journal.pmed.1002482 8. Toledo JB, Arnold M, Kastenmüuller G, Chang R, Baillie RA, Han X. Metabolic network failures in Alzheimer’s disease: a biochemical road map. Alzheimers Dement (2017) 13 965-984. DOI: 10.1016/j.jalz.2017.01.020 9. Trushina E, Dutta T, Persson XMT, Mielke MM, Petersen RC. Identification of altered metabolic pathways in plasma and CSF in mild cognitive impairment and Alzheimer’s disease using metabolomics. PLoS ONE (2013) 8 e63644. DOI: 10.1371/journal.pone.0063644 10. Costa AC, Joaquim HPG, Forlenza OV, Gattaz WF, Talib LL. Three plasma metabolites in elderly patients differentiate mild cognitive impairment and Alzheimer’s disease: a pilot study. Eur Arch Psychiatry Clin Neurosci (2020) 270 483-488. DOI: 10.1007/s00406-019-01034-9 11. Herholz K. Cerebral glucose metabolism in preclinical and prodromal Alzheimer’s disease. Expert Rev Neurother (2010) 10 1667-1673. DOI: 10.1586/ern.10.136 12. van der Velpen V, Teav T, Gallart-Ayala H, Mehl F, Konz I, Clark C. Systemic and central nervous system metabolic alterations in Alzheimer’s disease. Alzheimers Res Ther (2019) 28 93. DOI: 10.1186/s13195-019-0551-7 13. DeTure MA, Dickson DW. The neuropathological diagnosis of Alzheimer’s disease. Mol Neurodegener (2019) 14 32. DOI: 10.1186/s13024-019-0333-5 14. Mrdjen D, Fox EJ, Bukhari SA, Montine KS, Bendall SC, Montine TJ. The basis of cellular and regional vulnerability in Alzheimer’s disease. Acta Neuropathol (2019) 138 729-749. DOI: 10.1007/s00401-019-02054-4 15. Fanselow MS, Dong HW. Are the dorsal and ventral hippocampus functionally distinct structures?. Neuron (2010) 65 7. DOI: 10.1016/j.neuron.2009.11.031 16. Lee AR, Kim JH, Cho E, Kim M, Park M. Dorsal and ventral hippocampus differentiate in functional pathways and differentially associate with neurological disease-related genes during postnatal development. Front Mol Neurosci (2017). DOI: 10.3389/fnmol.2017.00331 17. Masurkar AV. Towards a circuit-level understanding of hippocampal CA1 dysfunction in Alzheimer’s disease across anatomical axes. J Alzheimers Dis Parkinsonism (2018) 8 412. DOI: 10.4172/2161-0460.1000412 18. Su L, Hayes L, Soteriades S, Williams G, Brain SA, Firbank MJ. Hippocampal stratum radiatum, lacunosum and moleculare sparing in mild cognitive impairment. J Alzheimers Dis (2018) 61 415-424. DOI: 10.3233/JAD-170344 19. Shaw K, Bell L, Boyd K, Grijseels DM, Clarke D, Bonnar O. Neurovascular coupling and oxygenation are decreased in hippocampus compared to neocortex because of microvascular differences. Nat Commun (2021) 12 3190. DOI: 10.1038/s41467-021-23508-y 20. Herculano-Houzel S. The glia/neuron ratio: how it varies uniformly across brain structures and species and what that means for brain physiology and evolution. Glia (2014) 62 1377-1391. DOI: 10.1002/glia.22683 21. Verkhratsky A, Butt AM. The history of the decline and fall of the glial numbers legend. Neuroglia (2018) 1 188-192. DOI: 10.3390/neuroglia1010013 22. Akdemir ES, Huang AYS, Deneen B. Astrocytogenesis: where, when, and how. F1000Res (2020) 9 F1000 Faculty Rev-233. DOI: 10.12688/f1000research.22405.1 23. Şovrea AS, Boşca AB. Astrocytes reassessment—an evolving concept part one: embryology, biology, morphology and reactivity. J Mol Psychiatry. (2013) 1 18. DOI: 10.1186/2049-9256-1-18 24. Nahirney PC, Tremblay ME. Brain ultrastructure: putting the pieces together. Front Cell Dev Biol (2021). DOI: 10.3389/fcell.2021.629503/full 25. Wang F, Xu S, Pan F, Verkhratsky A, Huang JH. Editorial: Natural products and brain energy metabolism: astrocytes in neurodegenerative diseases. Front Pharmacol (2022) 3 1039904. DOI: 10.3389/fphar.2022.1039904 26. Mestre H, Mori Y, Nedergaard M. The brain’s glymphatic system: current controversies. Trends Neurosci (2020) 43 458-466. DOI: 10.1016/j.tins.2020.04.003 27. Louveau A, Smirnov I, Keyes TJ, Eccles JD, Rouhani SJ, Peske JD. Structural and functional features of central nervous system lymphatics. Nature (2015) 523 337-341. DOI: 10.1038/nature14432 28. Iliff JJ, Lee H, Yu M, Feng T, Logan J, Nedergaard M. Brain-wide pathway for waste clearance captured by contrast-enhanced MRI. J Clin Invest (2013) 123 1299-1309. DOI: 10.1172/JCI67677 29. Mulligan SJ, MacVicar BA. Calcium transients in astrocyte endfeet cause cerebrovascular constrictions. Nature (2004) 431 195-199. DOI: 10.1038/nature02827 30. Zonta M, Angulo MC, Gobbo S, Rosengarten B, Hossmann KA, Pozzan T. Neuron-to-astrocyte signaling is central to the dynamic control of brain microcirculation. Nat Neurosci (2003) 6 43-50. DOI: 10.1038/nn980 31. Gordon GRJ, Mulligan SJ, MacVicar BA. Astrocyte control of the cerebrovasculature. Glia (2007) 55 1214-1221. DOI: 10.1002/glia.20543 32. Verkhratsky A, Parpura V, Li B, Scuderi C. Astrocytes: the housekeepers and guardians of the CNS. Adv Neurobiol (2021) 26 21-53. DOI: 10.1007/978-3-030-77375-5_2 33. Ullian EM, Sapperstein SK, Christopherson KS, Barres BA. Control of synapse number by glia. Science (2001) 291 657-661. DOI: 10.1126/science.291.5504.657 34. Barker AJ, Koch SM, Reed J, Barres BA, Ullian EM. Developmental control of synaptic receptivity. J Neurosci (2008) 28 8150-8160. DOI: 10.1523/JNEUROSCI.1744-08.2008 35. Hama H, Hara C, Yamaguchi K, Miyawaki A. PKC signaling mediates global enhancement of excitatory synaptogenesis in neurons triggered by local contact with astrocytes. Neuron (2004) 41 405-415. DOI: 10.1016/S0896-6273(04)00007-8 36. Augusto-Oliveira M, Arrifano GP, Takeda PY, Lopes-Araújo A, Santos-Sacramento L, Anthony DC. Astroglia-specific contributions to the regulation of synapses, cognition and behaviour. Neurosci Biobehav Rev (2020) 118 331-357. DOI: 10.1016/j.neubiorev.2020.07.039 37. Chung WS, Allen NJ, Eroglu C. Astrocytes control synapse formation, function, and elimination. Cold Spring Harb Perspect Biol (2015) 7 a020370. DOI: 10.1101/cshperspect.a020370 38. Andersen JV, Christensen SK, Westi EW, Diaz-delCastillo M, Tanila H, Schousboe A. Deficient astrocyte metabolism impairs glutamine synthesis and neurotransmitter homeostasis in a mouse model of Alzheimer’s disease. Neurobiol Dis (2021) 148 105198. DOI: 10.1016/j.nbd.2020.105198 39. Verkhratsky A, Nedergaard M. Physiology of astroglia. Physiol Rev (2018) 98 239-389. DOI: 10.1152/physrev.00042.2016 40. Tsacopoulos M, Magistretti PJ. Metabolic coupling between glia and neurons. J Neurosci (1996) 16 877-885. DOI: 10.1523/JNEUROSCI.16-03-00877.1996 41. Pellerin L, Bouzier-Sore AK, Aubert A, Serres S, Merle M, Costalat R. Activity-dependent regulation of energy metabolism by astrocytes: an update. Glia (2007) 55 1251-1262. DOI: 10.1002/glia.20528 42. Wang Z, Zhang Q, Lin JR, Jabalameli MR, Mitra J, Nguyen N. Deep post-GWAS analysis identifies potential risk genes and risk variants for Alzheimer’s disease, providing new insights into its disease mechanisms. Sci Rep (2021) 11 20511. DOI: 10.1038/s41598-021-99352-3 43. Smith AM, Davey K, Tsartsalis S, Khozoie C, Fancy N, Tang SS. Diverse human astrocyte and microglial transcriptional responses to Alzheimer’s pathology. Acta Neuropathol (2022) 143 75-91. DOI: 10.1007/s00401-021-02372-6 44. St-Pierre MK, VanderZwaag J, Loewen S, Tremblay MÈ. All roads lead to heterogeneity: the complex involvement of astrocytes and microglia in the pathogenesis of Alzheimer’s disease. Front Cell Neurosci (2022) 16 932572. DOI: 10.3389/fncel.2022.932572 45. Katsouri L, Birch AM, Renziehausen AWJ, Zach C, Aman Y, Steeds H. Ablation of reactive astrocytes exacerbates disease pathology in a model of Alzheimer’s disease. Glia (2020) 68 1017-1030. DOI: 10.1002/glia.23759 46. Davis N, Mota BC, Stead L, Palmer EOC, Lombardero L, Rodríguez-Puertas R. Pharmacological ablation of astrocytes reduces Aβ degradation and synaptic connectivity in an ex vivo model of Alzheimer’s disease. J Neuroinflammation (2021) 18 73. DOI: 10.1186/s12974-021-02117-y 47. Apelt J, Ach K, Schliebs R. Aging-related down-regulation of neprilysin, a putative β-amyloid-degrading enzyme, in transgenic Tg2576 Alzheimer-like mouse brain is accompanied by an astroglial upregulation in the vicinity of β-amyloid plaques. Neurosci Lett (2003) 339 183-186. DOI: 10.1016/S0304-3940(03)00030-2 48. Yamamoto N, Nakazawa M, Nunono N, Yoshida N, Obuchi A, Tanida M. Protein kinases A and C regulate amyloid-β degradation by modulating protein levels of neprilysin and insulin-degrading enzyme in astrocytes. Neurosci Res (2021) 166 62-72. DOI: 10.1016/j.neures.2020.05.008 49. Yamamoto N, Ishikuro R, Tanida M, Suzuki K, Ikeda-Matsuo Y, Sobue K. Insulin-signaling pathway regulates the degradation of amyloid β-protein via astrocytes. Neuroscience (2018) 10 227-236. DOI: 10.1016/j.neuroscience.2018.06.018 50. Norton L, Shannon C, Gastaldelli A, DeFronzo RA. Insulin: the master regulator of glucose metabolism. Metabolism (2022) 1 155142. DOI: 10.1016/j.metabol.2022.155142 51. Wegiel J, Wang KC, Tarnawski M, Lach B. Microglia cells are the driving force in fibrillar plaque formation, whereas astrocytes are a leading factor in plague degradation. Acta Neuropathol (2000) 100 356-364. DOI: 10.1007/s004010000199 52. Wisniewski HM, Wegiel J. Spatial relationships between astrocytes and classical plaque components. Neurobiol Aging (1991) 12 593-600. DOI: 10.1016/0197-4580(91)90091-W 53. Gomez-Arboledas A, Davila JC, Sanchez-Mejias E, Navarro V, Nuñez-Diaz C, Sanchez-Varo R. Phagocytic clearance of presynaptic dystrophies by reactive astrocytes in Alzheimer’s disease. Glia (2018) 66 637-653. DOI: 10.1002/glia.23270 54. Serrano-Pozo A, Muzikansky A, Gómez-Isla T, Growdon JH, Betensky RA, Frosch MP. Differential relationships of reactive astrocytes and microglia to fibrillar amyloid deposits in Alzheimer disease. J Neuropathol Exp Neurol (2013) 72 462-471. DOI: 10.1097/NEN.0b013e3182933788 55. Habib N, McCabe C, Medina S, Varshavsky M, Kitsberg D, Dvir-Szternfeld R. Disease-associated astrocytes in Alzheimer’s disease and aging. Nat Neurosci (2020) 23 701-706. DOI: 10.1038/s41593-020-0624-8 56. Morabito S, Miyoshi E, Michael N, Shahin S, Martini AC, Head E. Single-nucleus chromatin accessibility and transcriptomic characterization of Alzheimer’s disease. Nat Genet (2021) 53 1143-1155. DOI: 10.1038/s41588-021-00894-z 57. Muñoz-Castro C, Noori A, Magdamo CG, Li Z, Marks JD, Frosch MP. Cyclic multiplex fluorescent immunohistochemistry and machine learning reveal distinct states of astrocytes and microglia in normal aging and Alzheimer’s disease. J Neuroinflammation (2022) 19 30. DOI: 10.1186/s12974-022-02383-4 58. Cabinio M, Saresella M, Piancone F, LaRosa F, Marventano I, Guerini FR. Association between hippocampal shape, neuroinflammation, and cognitive decline in Alzheimer’s disease. J Alzheimers Dis (2018) 66 1131-1144. DOI: 10.3233/JAD-180250 59. Bisht K, Sharma KP, Lecours C, Gabriela Sánchez M, El Hajj H, Milior G. Dark microglia: a new phenotype predominantly associated with pathological states. Glia (2016) 64 826-839. DOI: 10.1002/glia.22966 60. St-Pierre MK, Carrier M, Lau V, Tremblay MÈ. Investigating microglial ultrastructural alterations and intimate relationships with neuronal stress, dystrophy, and degeneration in mouse models of Alzheimer’s disease. Methods Mol Biol (2022) 2515 29-58. DOI: 10.1007/978-1-0716-2409-8_3 61. Castejón OJ. Biopathology of astrocytes in human traumatic and complicated brain injuries. Review and hypothesis. Folia Neuropathol (2015) 53 173-192. DOI: 10.5114/fn.2015.54419 62. Castejón O. Astrocyte subtypes in the gray matter of injured human cerebral cortex: a transmission electron microscope study. Brain Inj (1999) 13 291. DOI: 10.1080/026990599121665 63. Castejón OJ. Morphological astrocytic changes in complicated human brain trauma. A light and electron microscopic study. Brain Inj (1998) 12 409-427. DOI: 10.1080/026990598122539 64. Shimura T, Hirano A, Llena JF. Ultrastructure of cerebellar hemangioblastoma. Some new observations on the stromal cells. Acta Neuropathol (1985) 67 6-12. DOI: 10.1007/BF00688119 65. Tóth Z, Séress L, Tóth P, Ribak CE, Gallyas F. A common morphological response of astrocytes to various injuries: “dark” astrocytes. A light and electron microscopic analysis. J Hirnforsch (1997) 38 173-186. PMID: 9176730 66. Gallyas F, Horváth Z, Dávid K, Liposits Z. An immediate morphopathologic response of a subpopulation of astrocytes to electroshock: “dark” astrocytes. Neurobiology (Bp) (1994) 2 245-253. PMID: 7881403 67. Munoz-Garcia D, Ludwin SK. Gliogenesis in organotypic tissue culture of the spinal cord of the embryonic mouse. I. Immunocytochemical and ultrastructural studies. J Neurocytol (1986) 15 273-290. DOI: 10.1007/BF01611431 68. Hol EM, Pekny M. Glial fibrillary acidic protein (GFAP) and the astrocyte intermediate filament system in diseases of the central nervous system. Curr Opin Cell Biol (2015) 32 121-130. DOI: 10.1016/j.ceb.2015.02.004 69. Borchelt DR, Ratovitski T, van Lare J, Lee MK, Gonzales V, Jenkins NA. Accelerated amyloid deposition in the brains of transgenic mice coexpressing mutant presenilin 1 and amyloid precursor proteins. Neuron (1997) 19 939-945. DOI: 10.1016/S0896-6273(00)80974-5 70. St-Pierre MK, Carrier M, González Ibáñez F, Šimončičová E, Wallman MJ, Vallières L. Ultrastructural characterization of dark microglia during aging in a mouse model of Alzheimer’s disease pathology and in human post-mortem brain samples. J Neuroinflammation (2022) 19 235. DOI: 10.1186/s12974-022-02595-8 71. El Hajj H, Savage JC, Bisht K, Parent M, Vallières L, Rivest S. Ultrastructural evidence of microglial heterogeneity in Alzheimer’s disease amyloid pathology. J Neuroinflammation (2019) 16 87. DOI: 10.1186/s12974-019-1473-9 72. Bisht K, El Hajj H, Savage JC, Sánchez MG, Tremblay MÈ. Correlative light and electron microscopy to study microglial interactions with β-amyloid plaques. J Vis Exp (2016) 1 e54060 73. Paxinos G, Franklin KBJ. Paxinos and Franklin’s the Mouse brain in stereotaxic coordinates (2012) 360 74. Peters A, Palay SL, Webster H. The fine structure of the nervous system: neurons and their supporting cells (1991) 534 75. Turmaine M, Raza A, Mahal A, Mangiarini L, Bates GP, Davies SW. Nonapoptotic neurodegeneration in a transgenic mouse model of Huntington’s disease. Proc Natl Acad Sci USA (2000) 97 8093-8097. DOI: 10.1073/pnas.110078997 76. Kherani ZS, Auer RN. Pharmacologic analysis of the mechanism of dark neuron production in cerebral cortex. Acta Neuropathol (2008) 116 447-452. DOI: 10.1007/s00401-008-0386-y 77. Colbourne F, Sutherland GR, Auer RN. Electron microscopic evidence against apoptosis as the mechanism of neuronal death in global ischemia. J Neurosci (1999) 19 4200-4210. DOI: 10.1523/JNEUROSCI.19-11-04200.1999 78. Dietrich WD, Alonso O, Halley M, Busto R. Delayed posttraumatic brain hyperthermia worsens outcome after fluid percussion brain injury: a light and electron microscopic study in rats. Neurosurgery (1996) 38 533-541. PMID: 8837806 79. Kuroiwa T, Nagaoka T, Ueki M, Yamada I, Miyasaka N, Akimoto H. Different apparent diffusion coefficient: water content correlations of gray and white matter during early ischemia. Stroke (1998) 29 859-865. DOI: 10.1161/01.STR.29.4.859 80. St-Pierre MK, Bordeleau M, Tremblay MÈ. Visualizing Dark Microglia. Methods Mol Biol (2019) 2034 97-110. DOI: 10.1007/978-1-4939-9658-2_8 81. St-Pierre MK, Šimončičová E, Bögi E, Tremblay MÈ. Shedding light on the dark side of the microglia. ASN Neuro (2020) 12 1759091420925335. DOI: 10.1177/1759091420925335 82. Bordeleau M, Lacabanne C, Fernández de Cossío L, Vernoux N, Savage JC, González-Ibáñez F. Microglial and peripheral immune priming is partially sexually dimorphic in adolescent mouse offspring exposed to maternal high-fat diet. J Neuroinflammation (2020) 17 264. DOI: 10.1186/s12974-020-01914-1 83. Tremblay MÈ, Majewska AK. Ultrastructural analyses of microglial interactions with synapses. Methods Mol Biol (2019) 2034 83-95. DOI: 10.1007/978-1-4939-9658-2_7 84. Bordeleau M, Fernández de Cossío L, Lacabanne C, Savage JC, Vernoux N, Chakravarty M. Maternal high-fat diet modifies myelin organization, microglial interactions, and results in social memory and sensorimotor gating deficits in adolescent mouse offspring. Brain Behav Immun Health. (2021) 15 100281. DOI: 10.1016/j.bbih.2021.100281 85. Hui CW, St-Pierre MK, Detuncq J, Aumailley L, Dubois MJ, Couture V. Nonfunctional mutant Wrn protein leads to neurological deficits, neuronal stress, microglial alteration, and immune imbalance in a mouse model of Werner syndrome. Brain Behav Immun (2018) 1 450-469. DOI: 10.1016/j.bbi.2018.06.007 86. Miyazono Y, Hirashima S, Ishihara N, Kusukawa J, Nakamura KI, Ohta K. Uncoupled mitochondria quickly shorten along their long axis to form indented spheroids, instead of rings, in a fission-independent manner. Sci Rep (2018) 8 350. DOI: 10.1038/s41598-017-18582-6 87. Decoeur F, Picard K, St-Pierre MK, Greenhalgh AD, Delpech JC, Sere A. N-3 PUFA deficiency affects the ultrastructural organization and density of white matter microglia in the developing brain of male mice. Front Cell Neurosci (2022). DOI: 10.3389/fncel.2022.802411 88. Prats C, Graham TE, Shearer J. The dynamic life of the glycogen granule. J Biol Chem (2018) 293 7089-7098. DOI: 10.1074/jbc.R117.802843 89. Versaevel M, Braquenier JB, Riaz M, Grevesse T, Lantoine J, Gabriele S. Super-resolution microscopy reveals LINC complex recruitment at nuclear indentation sites. Sci Rep (2014) 4 7362. DOI: 10.1038/srep07362 90. Henry MS, Bisht K, Vernoux N, Gendron L, Torres-Berrio A, Drolet G. Delta opioid receptor signaling promotes resilience to stress under the repeated social defeat paradigm in mice. Front Mol Neurosci (2018) 11 100. DOI: 10.3389/fnmol.2018.00100 91. Hart ML, Lauer JC, Selig M, Hanak M, Walters B, Rolauffs B. Shaping the cell and the future: recent advancements in biophysical aspects relevant to regenerative medicine. J Funct Morphol Kinesiol (2018) 3 2. DOI: 10.3390/jfmk3010002 92. Leyh J, Paeschke S, Mages B, Michalski D, Nowicki M, Bechmann I. Classification of microglial morphological phenotypes using machine learning. Front Cell Neurosci (2021) 15 241. DOI: 10.3389/fncel.2021.701673 93. Hui CW, St-Pierre A, El Hajj H, Remy Y, Hébert SS, Luheshi GN. Prenatal immune challenge in mice leads to partly sex-dependent behavioral, microglial, and molecular abnormalities associated with Schizophrenia. Front Mol Neurosci (2018) 11 13. DOI: 10.3389/fnmol.2018.00013 94. Lecours C, St-Pierre MK, Picard K, Bordeleau M, Bourque M, Awogbindin IO. Levodopa partially rescues microglial numerical, morphological, and phagolysosomal alterations in a monkey model of Parkinson’s disease. Brain Behav Immun (2020) 90 81-96. DOI: 10.1016/j.bbi.2020.07.044 95. Mondo E, Becker SC, Kautzman AG, Schifferer M, Baer CE, Chen J. A developmental analysis of juxtavascular microglia dynamics and interactions with the vasculature. J Neurosci (2020) 40 6503-6521. DOI: 10.1523/JNEUROSCI.3006-19.2020 96. Savage JC, St-Pierre MK, Carrier M, El Hajj H, Novak SW, Sanchez MG. Microglial physiological properties and interactions with synapses are altered at presymptomatic stages in a mouse model of Huntington’s disease pathology. J Neuroinflammation (2020) 17 98. DOI: 10.1186/s12974-020-01782-9 97. Yasumoto Y, Stoiljkovic M, Kim JD, Sestan-Pesa M, Gao XB, Diano S. Ucp2-dependent microglia-neuronal coupling controls ventral hippocampal circuit function and anxiety-like behavior. Mol Psychiatry (2021) 26 2740-2752. DOI: 10.1038/s41380-021-01105-1 98. Malykhin NV, Bouchard TP, Camicioli R, Coupland NJ. Aging hippocampus and amygdala. NeuroReport (2008) 19 543-547. DOI: 10.1097/WNR.0b013e3282f8b18c 99. Russo ML, Molina-Campos E, Ybarra N, Rogalsky AE, Musial TF, Jimenez V. Variability in sub-threshold signaling linked to Alzheimer’s disease emerges with age and amyloid plaque deposition in mouse ventral CA1 pyramidal neurons. Neurobiol Aging (2021) 1 207-222. DOI: 10.1016/j.neurobiolaging.2021.06.018 100. Veldsman M, Nobis L, Alfaro-Almagro F, Manohar S, Husain M. The human hippocampus and its subfield volumes across age, sex and APOE e4 status. Brain Commun (2021) 3 219. DOI: 10.1093/braincomms/fcaa219 101. Driscoll I, Hamilton DA, Petropoulos H, Yeo RA, Brooks WM, Baumgartner RN. The aging hippocampus: cognitive, biochemical and structural findings. Cereb Cortex (2003) 13 1344-1351. DOI: 10.1093/cercor/bhg081 102. Schitine C, Nogaroli L, Costa MR, Hedin-Pereira C. Astrocyte heterogeneity in the brain: from development to disease. Front Cell Neurosci (2015) 20 76 103. Zhou B, Zuo YX, Jiang RT. Astrocyte morphology: diversity, plasticity, and role in neurological diseases. CNS Neurosci Ther (2019) 25 665-673. DOI: 10.1111/cns.13123 104. Li KY, Gong PF, Li JT, Xu NJ, Qin S. Morphological and molecular alterations of reactive astrocytes without proliferation in cerebral cortex of an APP/PS1 transgenic mouse model and Alzheimer’s patients. Glia (2020) 68 2361-2376. PMID: 32469469 105. Olabarria M, Noristani HN, Verkhratsky A, Rodríguez JJ. Concomitant astroglial atrophy and astrogliosis in a triple transgenic animal model of Alzheimer’s disease. Glia (2010) 58 831-838. PMID: 20140958 106. Mampay M, Velasco-Estevez M, Rolle SO, Chaney AM, Boutin H, Dev KK. Spatiotemporal immunolocalisation of REST in the brain of healthy ageing and Alzheimer’s disease rats. FEBS Open Bio (2020) 11 146-163. DOI: 10.1002/2211-5463.13036 107. Vezzoli E, Calì C, De Roo M, Ponzoni L, Sogne E, Gagnon N. Ultrastructural evidence for a role of astrocytes and glycogen-derived lactate in learning-dependent synaptic stabilization. Cereb Cortex (2020) 30 2114-2127. DOI: 10.1093/cercor/bhz226 108. Alberini CM, Cruz E, Descalzi G, Bessières B, Gao V. Astrocyte glycogen and lactate: new insights into learning and memory mechanisms. Glia (2018) 66 1244-1262. DOI: 10.1002/glia.23250 109. Gertz HJ, Cervos-Navarro J, Frydl V, Schultz F. Glycogen accumulation of the aging human brain. Mech Ageing Dev (1985) 31 25-35. DOI: 10.1016/0047-6374(85)90024-7 110. Calì C, Tauffenberger A, Magistretti P. The strategic location of glycogen and lactate: from body energy reserve to brain plasticity. Front Cell Neurosci (2019) 6 82. DOI: 10.3389/fncel.2019.00082 111. Mohammed H, Al-Awami AK, Beyer J, Cali C, Magistretti P, Pfister H. Abstractocyte: a visual tool for exploring nanoscale astroglial cells. IEEE Trans Vis Comput Graph (2018) 24 853-861. DOI: 10.1109/TVCG.2017.2744278 112. Calì C, Baghabra J, Boges DJ, Holst GR, Kreshuk A, Hamprecht FA. Three-dimensional immersive virtual reality for studying cellular compartments in 3D models from EM preparations of neural tissues. J Comp Neurol (2016) 524 23-38. DOI: 10.1002/cne.23852 113. Apátiga-Pérez R, Soto-Rojas LO, Campa-Córdoba BB, Luna-Viramontes NI, Cuevas E, Villanueva-Fierro I. Neurovascular dysfunction and vascular amyloid accumulation as early events in Alzheimer’s disease. Metab Brain Dis (2022) 37 39-50. DOI: 10.1007/s11011-021-00814-4 114. Farkas E, Luiten PG. Cerebral microvascular pathology in aging and Alzheimer’s disease. Prog Neurobiol (2001) 64 575-611. DOI: 10.1016/S0301-0082(00)00068-X 115. Shabir O, Berwick J, Francis SE. Neurovascular dysfunction in vascular dementia, Alzheimer’s and atherosclerosis. BMC Neurosci (2018) 17 62. DOI: 10.1186/s12868-018-0465-5 116. Klohs J. An integrated view on vascular dysfunction in Alzheimer’s disease. Neurodegener Dis (2019) 19 109-127. DOI: 10.1159/000505625 117. Solis E, Hascup KN, Hascup ER. Alzheimer’s disease: the link between amyloid-β and neurovascular dysfunction. J Alzheimers Dis (2020) 76 1179-1198. DOI: 10.3233/JAD-200473 118. Calì C, Agus M, Kare K, Boges DJ, Lehväslaiho H, Hadwiger M. 3D cellular reconstruction of cortical glia and parenchymal morphometric analysis from serial block-face electron microscopy of juvenile rat. Prog Neurobiol (2019) 1 101696. DOI: 10.1016/j.pneurobio.2019.101696 119. Aten S, Kiyoshi CM, Arzola EP, Patterson JA, Taylor AT, Du Y. Ultrastructural view of astrocyte arborization, astrocyte–astrocyte and astrocyte–synapse contacts, intracellular vesicle-like structures, and mitochondrial network. Prog Neurobiol (2022) 213 102264. DOI: 10.1016/j.pneurobio.2022.102264 120. Nagy JI, Rash JE. Astrocyte and oligodendrocyte connexins of the glial syncytium in relation to astrocyte anatomical domains and spatial buffering. Cell Commun Adhes (2003) 10 401-406. DOI: 10.1080/cac.10.4-6.401.406 121. Quigley HA. Gap junctions between optic nerve head astrocytes. Invest Ophthalmol Vis Sci (1977) 16 582-585. PMID: 405347 122. Quillen S, Schaub J, Quigley H, Pease M, Korneva A, Kimball E. Astrocyte responses to experimental glaucoma in mouse optic nerve head. PLoS ONE (2020) 15 e0238104. DOI: 10.1371/journal.pone.0238104 123. Bisht K, Sharma K, Lacoste B, Tremblay MÈ. Dark microglia: why are they dark?. Commun Integr Biol. (2016) 9 e1230575. DOI: 10.1080/19420889.2016.1230575 124. Joost E, Jordão MJC, Mages B, Prinz M, Bechmann I, Krueger M. Microglia contribute to the glia limitans around arteries, capillaries and veins under physiological conditions, in a model of neuroinflammation and in human brain tissue. Brain Struct Funct (2019) 224 1301-1314. DOI: 10.1007/s00429-019-01834-8 125. Baloyannis SJ, Larrivee D. Mitochondria and Alzheimer’s disease an electron microscopy study. Redirecting Alzheimer strategy (2019) 126. Verkhratsky A, Rodrigues JJ, Pivoriunas A, Zorec R, Semyanov A. Astroglial atrophy in Alzheimer’s disease. Pflugers Arch (2019) 471 1247-1261. DOI: 10.1007/s00424-019-02310-2 127. Spanos F, Liddelow SA. An overview of astrocyte responses in genetically induced Alzheimer’s disease mouse models. Cells (2020) 9 E2415. DOI: 10.3390/cells9112415 128. Sanchez-Mico MV, Jimenez S, Gomez-Arboledas A, Muñoz-Castro C, Romero-Molina C, Navarro V. Amyloid-β impairs the phagocytosis of dystrophic synapses by astrocytes in Alzheimer’s disease. Glia (2021) 69 997-1011. DOI: 10.1002/glia.23943 129. Liu L, MacKenzie KR, Putluri N, Maletić-Savatić M, Bellen HJ. The glia-neuron lactate shuttle and elevated ROS promote lipid synthesis in neurons and lipid droplet accumulation in glia via APOE/D. Cell Metab (2017) 26 719-737.e6. DOI: 10.1016/j.cmet.2017.08.024 130. Liu L, Zhang K, Sandoval H, Yamamoto S, Jaiswal M, Sanz E. Glial lipid droplets and ROS induced by mitochondrial defects promote neurodegeneration. Cell (2015) 160 177-190. DOI: 10.1016/j.cell.2014.12.019 131. Moulton MJ, Barish S, Ralhan I, Chang J, Goodman LD, Harland JG. Neuronal ROS-induced glial lipid droplet formation is altered by loss of Alzheimer’s disease-associated genes. Proc Natl Acad Sci USA (2021) 118 e2112095118. DOI: 10.1073/pnas.2112095118 132. Ioannou MS, Jackson J, Sheu SH, Chang CL, Weigel AV, Liu H. Neuron-astrocyte metabolic coupling protects against activity-induced fatty acid toxicity. Cell (2019) 177 1522-1535.e14. DOI: 10.1016/j.cell.2019.04.001 133. Smolič T, Tavčar P, Horvat A, Černe U, Halužan Vasle A, Tratnjek L. Astrocytes in stress accumulate lipid droplets. Glia (2021) 69 1540-1562. DOI: 10.1002/glia.23978 134. Hulshof LA, van Nuijs D, Hol EM, Middeldorp J. The role of astrocytes in synapse loss in Alzheimer’s disease: a systematic review. Front Cell Neurosci (2022) 16 899251. DOI: 10.3389/fncel.2022.899251 135. Choi M, Lee SM, Kim D, Im HI, Kim HS, Jeong YH. Disruption of the astrocyte–neuron interaction is responsible for the impairments in learning and memory in 5XFAD mice: an Alzheimer’s disease animal model. Mol Brain (2021) 10 111. DOI: 10.1186/s13041-021-00823-5 136. Xu J. New insights into GFAP negative astrocytes in calbindin D28k immunoreactive astrocytes. Brain Sci (2018) 8 143. DOI: 10.3390/brainsci8080143 137. Tatsumi K, Isonishi A, Yamasaki M, Kawabe Y, Morita-Takemura S, Nakahara K. Olig2-lineage astrocytes: a distinct subtype of astrocytes that differs from GFAP astrocytes. Front Neuroanat (2018) 14 8. DOI: 10.3389/fnana.2018.00008 138. Wu T, Dejanovic B, Gandham VD, Gogineni A, Edmonds R, Schauer S. Complement C3 is activated in human AD brain and is required for neurodegeneration in mouse models of amyloidosis and tauopathy. Cell Rep (2019) 28 2111-2123.e6. DOI: 10.1016/j.celrep.2019.07.060 139. Lian H, Litvinchuk A, Chiang ACA, Aithmitti N, Jankowsky JL, Zheng H. Astrocyte-microglia cross talk through complement activation modulates amyloid pathology in mouse models of Alzheimer’s disease. J Neurosci (2016) 36 577-589. DOI: 10.1523/JNEUROSCI.2117-15.2016 140. Shi Q, Chowdhury S, Ma R, Le KX, Hong S, Caldarone BJ. Complement C3 deficiency protects against neurodegeneration in aged plaque-rich APP/PS1 mice. Sci Transl Med. (2017) 9 eaaf6295. DOI: 10.1126/scitranslmed.aaf6295 141. Cai Y, Guo H, Fan Z, Zhang X, Wu D, Tang W. Glycogenolysis is crucial for astrocytic glycogen accumulation and brain damage after reperfusion in ischemic stroke. Science (2020) 23 101136 142. Kurt MA, Davies DC, Kidd M. β-amyloid immunoreactivity in astrocytes in Alzheimer’s disease brain biopsies: an electron microscope study. Exp Neurol (1999) 158 221-228. DOI: 10.1006/exnr.1999.7096 143. Le Douce J, Maugard M, Veran J, Matos M, Jégo P, Vigneron PA. Impairment of glycolysis-derived l-serine production in astrocytes contributes to cognitive deficits in Alzheimer’s disease. Cell Metab (2020) 31 503-517.e8. DOI: 10.1016/j.cmet.2020.02.004 144. Andersen JV, Skotte NH, Christensen SK, Polli FS, Shabani M, Markussen KH. Hippocampal disruptions of synaptic and astrocyte metabolism are primary events of early amyloid pathology in the 5xFAD mouse model of Alzheimer’s disease. Cell Death Dis (2021) 12 1-13. DOI: 10.1038/s41419-021-04237-y 145. Hu Y, Fryatt GL, Ghorbani M, Obst J, Menassa DA, Martin-Estebane M. Replicative senescence dictates the emergence of disease-associated microglia and contributes to Aβ pathology. Cell Rep (2021) 35 109228. DOI: 10.1016/j.celrep.2021.109228 146. Krasemann S, Madore C, Cialic R, Baufeld C, Calcagno N, El Fatimy R. The TREM2-APOE pathway drives the transcriptional phenotype of dysfunctional microglia in neurodegenerative diseases. Immunity (2017) 47 566-581.e9. DOI: 10.1016/j.immuni.2017.08.008 147. Clayton K, Delpech JC, Herron S, Iwahara N, Ericsson M, Saito T. Plaque associated microglia hyper-secrete extracellular vesicles and accelerate tau propagation in a humanized APP mouse model. Mol Neurodegener (2021) 16 18. DOI: 10.1186/s13024-021-00440-9 148. Srinivasan K, Friedman BA, Etxeberria A, Huntley MA, van der Brug MP, Foreman O. Alzheimer’s patient microglia exhibit enhanced aging and unique transcriptional activation. Cell Rep (2020) 31 107843. DOI: 10.1016/j.celrep.2020.107843 149. Keren-Shaul H, Spinrad A, Weiner A, Matcovitch-Natan O, Dvir-Szternfeld R, Ulland TK. A unique microglia type associated with restricting development of Alzheimer’s disease. Cell (2017) 169 1276-1290.e17. DOI: 10.1016/j.cell.2017.05.018 150. Delizannis AT, Nonneman A, Tsering W, De Bondt A, Van den Wyngaert I, Zhang B. Effects of microglial depletion and TREM2 deficiency on Aβ plaque burden and neuritic plaque tau pathology in 5XFAD mice. Acta Neuropathol Commun (2021) 9 150. DOI: 10.1186/s40478-021-01251-1 151. Rothman SM, Tanis KQ, Gandhi P, Malkov V, Marcus J, Pearson M. Human Alzheimer’s disease gene expression signatures and immune profile in APP mouse models: a discrete transcriptomic view of Aβ plaque pathology. J Neuroinflammation (2018) 15 256. DOI: 10.1186/s12974-018-1265-7 152. McFarland KN, Ceballos C, Rosario A, Ladd T, Moore B, Golde G. Microglia show differential transcriptomic response to Aβ peptide aggregates ex vivo and in vivo. Life Sci Alliance (2021) 4 e202101108. DOI: 10.26508/lsa.202101108 153. Lodder C, Scheyltjens I, Stancu IC, Botella Lucena P, Gutiérrez de Ravé M, Vanherle S. CSF1R inhibition rescues tau pathology and neurodegeneration in an A/T/N model with combined AD pathologies, while preserving plaque associated microglia. Acta Neuropathol Commun (2021) 9 108. DOI: 10.1186/s40478-021-01204-8 154. Natunen T, Martiskainen H, Marttinen M, Gabbouj S, Koivisto H, Kemppainen S. Diabetic phenotype in mouse and humans reduces the number of microglia around β-amyloid plaques. Mol Neurodegener (2020) 15 66. DOI: 10.1186/s13024-020-00415-2 155. Romero-Molina C, Navarro V, Sanchez-Varo R, Jimenez S, Fernandez-Valenzuela JJ, Sanchez-Mico MV. Distinct microglial responses in two transgenic murine models of TAU pathology. Front Cell Neurosci (2018) 12 421. DOI: 10.3389/fncel.2018.00421 156. Sobue A, Komine O, Hara Y, Endo F, Mizoguchi H, Watanabe S. Microglial gene signature reveals loss of homeostatic microglia associated with neurodegeneration of Alzheimer’s disease. Acta Neuropathol Commun (2021) 9 1. DOI: 10.1186/s40478-020-01099-x 157. Gerrits E, Brouwer N, Kooistra SM, Woodbury ME, Vermeiren Y, Lambourne M. Distinct amyloid-β and tau-associated microglia profiles in Alzheimer’s disease. Acta Neuropathol (2021) 141 681-696. DOI: 10.1007/s00401-021-02263-w 158. Olah M, Menon V, Habib N, Taga MF, Ma Y, Yung CJ. Single cell RNA sequencing of human microglia uncovers a subset associated with Alzheimer’s disease. Nat Commun (2020) 11 6129. DOI: 10.1038/s41467-020-19737-2 159. Xu J, Zhang P, Huang Y, Zhou Y, Hou Y, Bekris LM. Multimodal single-cell/nucleus RNA sequencing data analysis uncovers molecular networks between disease-associated microglia and astrocytes with implications for drug repurposing in Alzheimer’s disease. Genome Res (2021) 31 1900-1912. DOI: 10.1101/gr.272484.120 160. Sala Frigerio C, Wolfs L, Fattorelli N, Thrupp N, Voytyuk I, Schmidt I. The major risk factors for Alzheimer’s disease: age, sex, and genes modulate the microglia response to Aβ plaques. Cell Rep (2019) 27 1293-1306.e6. DOI: 10.1016/j.celrep.2019.03.099 161. Sierksma A, Lu A, Mancuso R, Fattorelli N, Thrupp N, Salta E. Novel Alzheimer risk genes determine the microglia response to amyloid-β but not to TAU pathology. EMBO Mol Med (2020) 12 e10606. DOI: 10.15252/emmm.201910606 162. Marschallinger J, Iram T, Zardeneta M, Lee SE, Lehallier B, Haney MS. Lipid-droplet-accumulating microglia represent a dysfunctional and proinflammatory state in the aging brain. Nat Neurosci (2020) 23 194-208. DOI: 10.1038/s41593-019-0566-1 163. St-Pierre MK, Carrier M, Lau V, Tremblay MÈ, Jahani-Asl A. Investigating microglial ultrastructural alterations and intimate relationships with neuronal stress, dystrophy, and degeneration in mouse models of Alzheimer’s disease. Neuronal cell death (2022)
--- title: 'A natural experiment to assess how urban interventions in lower socioeconomic areas influence health behaviors: the UrbASanté study' authors: - Hélène Charreire - Benoit Conti - Lucile Bauchard - Ndèye Aïta Cissé - Marlène Perignon - Pascaline Rollet - Coline Perrin - Sophie Blanchard - Céline Roda - Thierry Feuillet - Malika Madelin - Vincent Dupuis - Anne-Sophie Evrard - Anne-Peggy Hellequin - Isabelle Coll - Corinne Larrue - Sophie Baudet-Michel - Gabrielle Vernouillet - Fernande Ntsame-Abegue - Isabelle Fabre - Caroline Méjean - Jean-Michel Oppert journal: BMC Public Health year: 2023 pmcid: PMC10015725 doi: 10.1186/s12889-023-15388-2 license: CC BY 4.0 --- # A natural experiment to assess how urban interventions in lower socioeconomic areas influence health behaviors: the UrbASanté study ## Abstract ### Background Mechanisms underlying the associations between changes in the urban environment and changes in health-related outcomes are complex and their study requires specific approaches. We describe the protocol of the interdisciplinary UrbASanté study, which aims to explore how urban interventions can modify environmental exposures (built, social, and food environments; air quality; noise), health-related behaviors, and self-reported health using a natural experiment approach. ### Methods The study is based on a natural experiment design using a before/after protocol with a control group to assess changes in environmental exposures, health-risk behaviors, and self-reported health outcomes of a resident adult population before and after the implementation of a time series of urban interventions in four contiguous neighborhoods in Paris (France). The changes in environmental exposures, health-related behaviors, and self-reported health outcomes of a resident adult population will be concurrently monitored in both intervention and control areas. We will develop a mixed-method framework combining substantial fieldwork with quantitative and qualitative analytical approaches. This study will make use of (i) data relating to exposures and health-related outcomes among all participants and in subsamples and (ii) interviews with residents regarding their perceptions of their neighborhoods and with key stakeholders regarding the urban change processing, and (iii) existing geodatabases and field observations to characterize the built, social, and food environments. The data collected will be analyzed with a focus on interrelationships between environmental exposures and health-related outcomes using appropriate approaches (e.g., interrupted time series, difference–in-differences method). ### Discussion Relying on a natural experiment approach, the research will provide new insights regarding issues such as close collaboration with urban/local stakeholders, recruitment and follow-up of participants, identification of control and intervention areas, timing of the planned urban interventions, and comparison of subjective and objective measurements. Through the collaborative work of a consortium ensuring complementarity between researchers from different disciplines and stakeholders, the UrbASanté study will provide evidence-based guidance for designing future urban planning and public health policies. ### Trial registration This research was registered at the ClinicalTrial.gov (NCT05743257). ## Background It has become widely accepted that overall health and quality of life result from a complex interplay between individual (e.g., age, gender, and socioeconomic position) and contextual built and social characteristics of the environment in which individuals live (e.g., transport infrastructures, land use, food environment, and area-level deprivation) [1, 2]. Thus, urban planning choices can affect the health and well-being of the population and contribute to reducing or, in contrast, enhancement of social health inequalities. The impact of urban redevelopment in a neighborhood can vary considerably depending on the socioeconomic characteristics of populations and environments: in some cases, it may increase social health-related inequalities [3]. Many urban planning and design decisions aimed at improving urban living conditions have been developed in recent years [4]. For example, in the field of air quality, since the year 2000, low-emission zones have been defined in a number of European cities in response to concerns about air pollution and public health. Although such transformations have high political and financial costs, there have been few real-life assessments to date of the well-being and health effects of such environmental policies as well as the role of social health inequalities [5]. Associations between urban changes, environmental exposures, and health-related behaviors generally form a highly complex network involving numerous interactions [6]. More specifically, the mechanisms underlying the associations between changes brought about by urban interventions and changes in environmental risks and health-related behaviors remain understudied. Increased understanding of such mechanisms is needed to clarify the significance and role of specific modifiable determinants. The natural experiment approach is a way to assess the health impacts of urban interventions and represents an opportunity for innovative research. The Medical Research Council (MRC, United Kingdom) defines a natural experiment as an approach “which exploit(s) natural or unplanned variation in exposure, i.e., variation that is not manipulated for the purposes of research” [7, 8]. In a recent “call to action” for the transition to health and sustainable cities, the authors recommended conducting natural experiments to assess the impact of urban interventions on health, social, environmental, and equity outcomes [2]. The natural experiment approach has already been well developed in other research disciplines such as economics. Joshua Angrist, Guido Imbens, and David Card shared the 2021 Nobel prize in economic sciences for their research based on this approach. They showed how causality can be inferred from observational data in real-world natural experiments, specifically regarding labor market issues. In other words, the natural experiment approach takes advantage of the circumstances in which relevant changes occur in a given area and population, e.g., an urban intervention, and then tries to plausibly attribute changes in outcomes of interest to the intervention [9]. Natural experiments are increasingly considered as being able to provide important new input in research and can be very useful for informing policy decisions [10]. In a recent systematic review, 15 studies describing natural experiments using changes in the built environment and investigating health-related consequences were identified. Eight of these reported (favorable) changes in active mobility and dietary intakes when measured on residents after one year post-baseline [4]. However, the authors highlighted that it is difficult to draw overall conclusions, especially because of methodological limitations such as a lack of a comparison group or limited sample sizes. The main objective of the UrbASanté study is to assess how urban interventions can modify environmental exposures (including air quality, noise, food environment, built, and social characteristics), health-related behaviors, and self-reported health using a natural experiment protocol in Paris (France). The specific aims of the project are (i) to develop a combination of methods to assess and monitor environmental exposures, health-risk behaviors, and self-reported health; (ii) to collect relevant data during experimentation at a local level in a real-life setting; and (iii) to analyze the data generated to better understand the effects of urban transformations on environmental exposures, health-risk behaviors, and self-reported health in order to more accurately guide public health and urban planning policies. ## Study design The study is based on a natural experiment approach with a before/after protocol design to assess changes in environmental exposures, health-risk behaviors, and self-reported health outcomes of a resident adult population before (T0) and after (T1) the implementation of a time series of urban interventions. The changes observed in adult residents located in the intervention areas (“exposed population”) will be compared to those in adults living in control areas (“unexposed population”). Control areas are located close to areas with urban intervention sites but are not themselves subject to urban intervention. As described in Fig. 1, the data collection will be implemented in two stages in both the intervention and the control areas: a first stage at baseline (T0, 2022), and a second stage (T1, 2025) after the urban interventions. Fig. 1Overall design of the UrbASanté study. ( adapted from Leatherdale et al., 2019) A particular methodological challenge for our study is the recruitment of participants in a context of (i) low socioeconomic neighborhoods with deprived populations who are generally a hard-to-reach segment of the overall population for research purposes and (ii) a hot topic (urban changes and health) in “hot places” that receive a great deal of political and urban planning attention in the context of ongoing large urban regeneration programs, especially with the perspective of the Olympic and Paralympic Games taking place in Paris in 2024. An additional methodological challenge is the ability to identify intervention and control neighborhoods. Ideally, this involves finding similar intervention and control neighborhoods, with the only difference being that one or several of the neighborhoods will undergo an urban change. In practice, none of the neighborhoods are equally matched in terms of all of the characteristics of interest. In our study, the main objective criterion to identify equally matched neighborhoods is the socioeconomic level, which will be used to define disadvantaged vs. more advantaged areas. This design is appropriate for our study because the intervention and control sites are in the same residential area, with urban intervention taking place in specific neighborhoods. While major transport infrastructures or those related to the Olympic Games will benefit all neighborhoods and more broadly all of Paris and its surrounding areas, specific urban interventions (i.e., housing, street, site) are restricted to specific neighborhoods only. ## Intervention and control neighborhoods The UrbASanté study will examine potential changes in environmental exposures, health-related behaviors, and self-reported health due to urban interventions in an urban area called Porte de la Chapelle in the northern part of the city of Paris (France). This area includes four contiguous neighborhoods located in the northern part of the 18th arrondissement of Paris (Fig. 2). This area exhibits specific characteristics compared to the broader Parisian conurbation: a large part of its population is socioeconomically deprived and it is exposed to high levels of pollutants [11]. This area is also defined as a place of interest in the context of the Olympic and Paralympic games scheduled to take place in Paris in 2024. Urban interventions will occur with different stages of transformation from 2024 (the Paris Olympic and Paralympic games) to 2030 (legacy of the Paris Olympic and Paralympic games) depending on each neighborhood. Two neighborhoods will be referred to as “intervention” neighborhoods, with measures taken before and after a series of urban changes that will start in 2023 (Fig. 2). The main urban changes in the Gare des Mines-Fillettes neighborhood (which includes the two urban sections of Charles Hermite and Valentin Abeille) will comprise the creation or renewal of urban green areas, pedestrian/cycle pathways and urban corridors, housing, redesigning of public spaces, and more specifically the construction of an Olympic sports facility. The urban changes in Porte de la Chapelle Avenue will mainly comprise the development of a new public transport network (bus) as well as the inclusion of cycle and pedestrian pathways and the widening of sidewalks. Two neighborhoods will be referred to as “control” neighborhoods, with measurements at the same time periods as for the “intervention” neighborhoods. Chapelle-international is a new neighborhood built on the site of a former railway wasteland. At the end of 2019, more than 1,500 residents had already moved into this neighborhood. The Chapelle-Evangile neighborhood will not undergo specific urban interventions, except in the northern part, where the renovation and expansion of a green area are currently being finalized. Fig. 2Intervention and control neighborhoods of the study area of the UrbASanté study ## Resident data collection In all studied neighborhoods, the criteria for inclusion will comprise: being an adult (18 years of age or older); residing in the neighborhood; and being able to read, write, or understand French well enough to complete the UrbASanté questionnaire (online or on paper) independently or with assistance from a research assistant. The UrbASanté study is conducted in accordance with the guidelines laid down in the Declaration of Helsinki, and all procedures have been approved by the Institutional Review Board of the French Institute for Health and Medical Research (CEI/IRB INSERM no. IRB00003888, IORG0003254, FWA00005831) and registered with the Commission Nationale de l’Informatique et des Libertés (2220971v0). All participants will provide informed consent. Participants will receive a 15 € voucher for returning a fully completed questionnaire. ## Recruitment plan We will use posters, flyers, social media, and local newspaper advertisements to raise awareness of the survey and increase enrollment, and we will develop a door-to-door recruitment protocol [12]. In a first step, the research team will identify the social housing providers, building superintendents for social housing buildings, and housing organizations for private buildings to be contacted prior to the field data collection. In a second step, we will set up appointments with building superintendents or private housing organizations to place posters announcing the study inside the buildings, and then a few days later, we will conduct the door-to-door surveys. The door-to-door surveys will aim to explain the UrbASanté study, distribute flyers and questionnaires, and collect email addresses for follow-up with inhabitants expressing interest in the research. In addition, we will organize information and survey sessions in the main lobby of the main buildings. We will participate in a variety of community events and use social facilities (e.g., the local library) to conduct drop-in sessions, distribute the study questionnaire, and collect email addresses. ## Mixed-method framework Our analyses will make use of (i) collection of data relating to exposures and health-related outcomes among all participants and in subsamples, (ii) interviews with residents about their perceptions of their neighborhood and interviews with key stakeholders about the urban change processing, and (iii) existing geodatabases and field observations to characterize the built, social, and food environments (Fig. 3). ## Sampling plan According to the 2018 census data from the INSEE (the French National Institute for Statistics and Economic Studies), 13,025 adults (≥ 18 years of age) reside in the entire study area (including the control and intervention neighborhoods). We will aim to recruit 600 adult participants at baseline to form the core data survey groups (300 in the intervention area and 300 in the control area). Although there is little precedent for sample size calculations for natural experiment studies (Kestens et al., 2019), potentially due to the multidimensionality of urban interventions, this sample size would be sufficient to assess a change in health-risk behaviors in the context of urban redevelopment. Indeed, assuming a power of $90\%$ and an alpha of 0.05, 380 participants will be required to detect the effect reported by Pazin et al. [ 2016] for instance, i.e. an increase of 32 ($95\%$ CI: 15–51) min/week of walking after development of a new walking and cycling route [13]. In addition, we will aim to define three subsamples of participants to assess: the nutritional quality of the household food supply through a food supply diary for a subsample among 400 participants (200 in the intervention area and 200 in the control area);the food provisioning practices of individuals through a semi-structured interview for a subsample of 30 participants (15 in the intervention area and 15 in the control area);the mobility-based real-time air pollution and noise exposures through sensor measurements for a subsample of 30 participants (15 in the intervention area and 15 in the control area). Fig. 3Flowchart of the survey in the UrbASanté study ## Questionnaire (core data survey)—all participants A questionnaire-based survey (paper and online versions) will be used to collect self-reported information from individuals regarding socio-demographic characteristics, health outcomes (general health, respiratory health, and weight status), physical activity and sedentary behaviors, perception of neighborhood characteristics, air quality, and noise exposure. The questionnaire includes standardized questions derived from existing validated questionnaires. The questionnaires, the coding procedures, as well as the results of the analyses, will be posted online following an open science strategy using the resources of the Very Large Research Infrastructure called Huma-Num (https://www.huma-num.fr/about-us/), which is supported by the French National Center for Scientific Research (CNRS). ## Health outcomes Self-reported general health will be assessed by the question “How is your health in general? Is it…” (very good/good/fair/bad/very bad). This question is recommended by the WHO as a standard and cost-effective measure in health surveys [14] and is used in national [15] and European longitudinal studies (e.g., the European Health Interview Survey—EHIS). In addition, standardized questionnaires that are widely used in respiratory epidemiological studies in adults will be used (the European Community Respiratory Health Survey—ECRHS [www.ecrhs.org/] and the Epidemiological Study of the Genetics and Environment of Asthma—EGEA [egeanet.vjf.inserm.fr]). To assess weight status, the participants will be asked to self-report their weight and height in order to calculate their body mass index (BMI = weight [kg] divided by height [meters] squared). ## Physical activity, active mobility, and sedentary behaviors Context-specific active mobility, general physical activity, and sedentary behaviors will be evaluated by questions from the Sedentary, Transportation, and Activity Questionnaire (STAQ) [16, 17]. Perception of active modes (walking and cycling) and characteristics of the neighborhood related to active mobility will also be part of the questions asked to the participants. ## Perception and use of neighborhood The perception and use of urban public spaces (such as streets and sidewalks, parks and public-green spaces, and recreational areas) will be assessed by questions from a Daily Life Environment questionnaire (Questionnaire sur l’environnement de vie quotidien, QEVIC) [18]. Specific questions will be also added to include the perception of the food environment (presence of food outlets/restaurants) and details of where, when, and why food outlets were chosen. ## Food supply subsample Participants in the food supply sample will be provided with a short food supply diary to record the details of their household food supplies over a month and asked to save the corresponding grocery store and supermarket receipts, which will be used to accurately assess household food expenditures for food groups. Based on the share of expenditures by food group, the nutritional quality of the household food supplies will be assessed using the revised Healthy Purchase Index (r-HPI) described elsewhere [19]. Briefly, the r-HPI is an index obtained by summing a purchase diversity subscore and a purchase quality subscore, ranging from a minimum score of minus 8 to a maximum score of 17 points, where a higher score reflects a higher quality of the household’s monthly food purchases. ## Subsample of the food provisioning practices of individuals Food provisioning practice will be assessed with semi-structured interviews to improve the understanding of the changes over the period of urban intervention regarding the food provisioning practices of individuals (providing details regarding the choice of food outlets and how they justify their shopping practices). All interviews will be fully recorded and transcribed by a professional transcription company or by a member of the research team. ## Mobility-based real-time air pollution and noise exposure subsample Pollutant concentrations and noise levels will be measured for one week along the daily itineraries of participants with mobile sensors and time-activity diaries. Particulate matter concentrations (PM2.5 and PM10 fractions), the temperature, and the relative humidity will be measured with AirBeam3 sensors (HabitatMap, New York, USA), while the nitrogen dioxide (NO2) and ozone (O3) concentrations will be determined with Cairsens® (ENVEA group, Poissy, France) data loggers that integrate reliable electrochemical gas sensors. Noise levels will be measured using noise dosimeters developed by BruitParif (France). The following noise indicators will be calculated: LAeq, LCeq, LAFmax, and LCFmax data. ## Characteristics of built, social, and retail food environments with GIS Existing geodatabases provided by French public institutions will be used to characterize the social (e.g., population census-INSEE), built (e.g., land-use-IGN), and retail food environments of the study area. For example, the characteristics of the built environment of interest will include the presence of green and blue spaces, street networks, amenities, and pedestrian and cycling infrastructures. Social deprivation can be characterized at the local level based on census variables (housing composition and unemployment rate). The food and built environment characteristics will be completed by field observations (e.g., pedestrian and cycling networks, outlet locations). In addition, we will examine urban and social changes in the study areas and more specifically the possibility of a gentrification effect (e.g., displacement of traditional low-income residents by more affluent households) using census data and field observations. ## Decision-making process of urban interventions Semi-structured interviews will be conducted among key stakeholders to examine the follow-up of the decision-making process related to urban interventions. The process of designing the urban intervention program (e.g., which parties are involved, how, and when) will be analyzed, as well as the content of the urban planning strategy, in particular its relevance with regard to health-related outcomes. For the relevant planners involved in the project, the interviews will focus on the genesis and progression of the urban projects as a part of the assessment of the decision-making process. Existing reports such as urban diagnoses and urban expertise (grey literature) related to the study areas and provided by local practitioners and urban planners will be gathered to obtain a better understanding of the spatial context in terms of populations, economy, and urban history. ## Planned statistical analyses The distributions and correlations among environmental exposures, health-related behaviors, and self-reported health outcomes will be studied at each time point (T0 and T1). The between-period differences will be quantified. Furthermore, we will analyze both cross-sectional (at T0 and T1) and longitudinal dimensions (from T0 to T1) using cluster analyses (e.g., K-means analysis) and model-based methods (e.g., mixture modeling techniques or latent class growth analysis) [20]. The associations between environmental exposures and health-related behaviors will be examined using each indicator individually or using the profiles by multivariate regression and spatially explicit models. For instance, multilevel regression models will be performed to estimate associations between changes in the environment (air quality, noise, built, social, and food) and changes in health-related behaviors (food supply and habitual active mobility). Other approaches will be considered to strengthen the impact attribution of the urban intervention to the observed changes. For instance, interrupted time series regressions will be considered for environmental exposures assessed continuously at evenly spaced intervals over time with one well-defined change point (i.e., urban interventions). The difference-in-difference approach will be used to compare the changes in the outcomes among the participants exposed to the intervention with the change among those who remain unexposed (i.e., the control group). Additionally, time sequences of the observations (“life-segment”) will be generated for participants included in subsamples. The dynamic processes interrelating the exposures (air quality, noise, social, built, and food), the places visited, and the health-risk outcomes will be examined by using space-time disaggregation of the data: case-crossover analyses allowing each person to be compared with themselves will thus be conducted. Possible effect measure modification by socioeconomic status will be explored using interaction terms and stratified models. Particularly, the reasons why people do or do not use local food outlets and how low-income participants access and perceive the food environment compared to more advantaged segments of the population will be examined. ## An interdisciplinary and collaborative work at various levels The UrbASanté study represents an innovative scientific interdisciplinary effort and aims to estimate longitudinal relationships in the interaction between urban characteristics and health-risk behaviors and self-reported health, and more specifically to investigate the processes that may lead to a healthier and more sustainable city. The project will, therefore, be carried out through the collaborative work of a consortium ensuring complementarity between geographers, epidemiologists, experts in atmospheric physics, urban planners, and stakeholders in the field of urban planning and public health. The objectives of UrbASanté will be achieved through close interaction, with a participatory research dynamic, between local stakeholders involved in urban planning and public health and the research community in a type of “action research”. Paris city stakeholders have been involved in the research consortium from the start of the project to provide relevant information regarding research (questions and methods) and local experiences and to facilitate evidence-based decision-making. Overall, the UrbASanté study will contribute to “breaking the silo approach” that tends to exist in urban planning issues involving different urban services, approaches, and disciplines. In addition, the first stage of our fieldwork will be to both identify and meet local stakeholders and practitioners, community and association leaders, and building superintendents to explain the how and why of the UrbASanté research project and that the results that it is expected to yield may be of importance to the residents. We will then participate in the life of the neighborhoods (e.g., by becoming involved in local events and activities) to become known to the residents and to facilitate (future) door-to-door surveys. This process of communication with all local stakeholders also allows for a better understanding of local situations, and additional city and local knowledge users will be included as the research unfolds. This back-and-forth communication between city/local parties and researchers will allow the project to be improved and it may also facilitate dissemination of the results at a later stage. ## Combination of methods We pay particular attention to collect subjective and objective measurements of environmental characteristics of interest [21]. When attempting to explain health-related behaviors in terms of impacts of the environment, it is important to capture both objective and subjective assessments, in other words how the environment is perceived by those who inhabit it [22, 23]. The perceptions of residents are typically obtained through interviews or self-administered questionnaires, as well as objective measures derived from field observations, sensors, and existing geospatial data. ## Limitations and scientific risks The UrbASanté research project involves several risks. The first risk lies with the defining feature of a natural experiment, namely that the implementation of the urban intervention is not under the control of the research consortium. We are aware that the project depends on the schedule of the planned urban changes and, therefore, remains highly dependent on variables external to the project. However, this risk is reduced by involvement of the stakeholders as full partners in our consortium, thereby ensuring visibility of the steps taken to implement urban changes in the field. In addition, the urban interventions used as support of the natural experiment are included in major urban planning projects in the city of Paris. Most of the health-related behaviors and health outcomes are self-reported. This represents a limitation for measurements associated with smoking, physical activity, and sedentary behaviors because of known social desirability bias [24, 25]. However, self-perceived health has received increasing interest in international studies because it correlates strongly with objective measures of health and it is consistent with predictions of future health problems and mortality [26]. Self-perceived health is also a widely used measurement to study trends and inequalities between genders, as well as across population groups [27]. Assessment of the effects of urban changes at the neighborhood level provides a unique opportunity to generate insights regarding the range and the nature of both positive and negative impacts on health-related outcomes. It should help elucidate some of the specific components and mechanisms through which urban changes can influence health-risk behaviors. The results will be of major relevance for research and local public policies (i) in the design of healthy cities and (ii) to maximize the benefits of urban interventions for which quantitative scientific knowledge is largely insufficient compared to the substantial societal expectations. Urban planning and public health authorities need tools to recognize opportunities, support decision-making, provide data to support funding applications, and demonstrate the impact of their initiatives to embed health into urban policy. The results obtained will contribute to filling gaps in major research areas, since the project is fully in line with the objectives of the “Health in All Policies”, developed as a key initiative of the WHO Healthy Cities Network to provide practical strategies for integrating health considerations into all government policies, not only within the health portfolio. ## References 1. Giles-Corti B, Vernez-Moudon A, Reis R, Turrell G, Dannenberg AL, Badland H. City planning and population health: a global challenge. The Lancet déc (2016.0) 388 2912-24. DOI: 10.1016/S0140-6736(16)30066-6 2. Giles-Corti B, Moudon AV, Lowe M, Cerin E, Boeing G, Frumkin H. What next? Expanding our view of city planning and global health, and implementing and monitoring evidence-informed policy. Lancet Glob Health juin (2022.0) 10 e919-26. DOI: 10.1016/S2214-109X(22)00066-3 3. Aschan-Leygonie C, Baudet-Michel S, Mathian H, Sanders L. Gaining a better understanding of respiratory health inequalities among cities: an ecological case study on elderly males in the larger french cities. Int J Health Geogr (2013.0) 12 19. DOI: 10.1186/1476-072X-12-19 4. 4.MacMillan F, George E, Feng X, Merom D, Bennie A, Cook A et al. Do Natural Experiments of Changes in Neighborhood Built Environment Impact Physical Activity and Diet? A Systematic Review. IJERPH. 26 janv 2018;15(2):217. 5. Mudway IS, Dundas I, Wood HE, Marlin N, Jamaludin JB, Bremner SA. Impact of London’s low emission zone on air quality and children’s respiratory health: a sequential annual cross-sectional study. The Lancet Public Health janv (2019.0) 4 e28-40. DOI: 10.1016/S2468-2667(18)30202-0 6. Prins RG, Panter J, Heinen E, Griffin SJ, Ogilvie DB. Causal pathways linking environmental change with health behaviour change: natural experimental study of new transport infrastructure and cycling to work. Prev Med juin (2016.0) 87 175-82. DOI: 10.1016/j.ypmed.2016.02.042 7. 7.Medical Research Council (MRC). Using Natural Experiments to Evaluate Population Health Interventions: Guidance for Producers and Users of Evidence. London:Medical Research Council; 211apr. J.-C. 8. Craig P, Cooper C, Gunnell D, Haw S, Lawson K, Macintyre S. Using natural experiments to evaluate population health interventions: new Medical Research Council guidance. J Epidemiol Community Health déc (2012.0) 66 1182-6. DOI: 10.1136/jech-2011-200375 9. Leatherdale ST. Natural experiment methodology for research: a review of how different methods can support real-world research. Int J Social Res Methodol 2 janv (2019.0) 22 19-35. DOI: 10.1080/13645579.2018.1488449 10. Crane M, Bohn-Goldbaum E, Grunseit A, Bauman A. Using natural experiments to improve public health evidence: a review of context and utility for obesity prevention. Health Res Policy Sys déc (2020.0) 18 48. DOI: 10.1186/s12961-020-00564-2 11. Host S, Laruelle N, Mauclair C, Caudeville J. Cumuls d’expositions environnementales en Ile-de-France, un enjeu de santé publique. Méthode d’identification des secteurs les plus impactés (2022.0) 12. Hillier A, Cannuscio CC, Griffin L, Thomas N, Glanz K. The value of conducting door-to-door surveys. Int J Social Res Methodol 4 mai (2014.0) 17 285-302. DOI: 10.1080/13645579.2012.733173 13. Pazin J, Garcia LMT, Florindo AA, Peres MA, Guimarães AC, de Borgatto A. Effects of a new walking and cycling route on leisure-time physical activity of brazilian adults: a longitudinal quasi-experiment. Health & Place mai (2016.0) 39 18-25. DOI: 10.1016/j.healthplace.2016.02.005 14. de Bruin A, Picavet HS, Nossikov A. Health interview surveys. Towards international harmonization of methods and instruments. WHO Reg Publ Eur Ser (1996.0) 58 i-xiii. PMID: 8857196 15. Singh-Manoux A, Martikainen P, Ferrie J, Zins M, Marmot M, Goldberg M. What does self rated health measure? Results from the british Whitehall II and French Gazel cohort studies. J Epidemiol Community Health avr (2006.0) 60 364-72. DOI: 10.1136/jech.2005.039883 16. Menai M, Charreire H, Feuillet T, Salze P, Weber C, Enaux C. Walking and cycling for commuting, leisure and errands: relations with individual characteristics and leisure-time physical activity in a cross-sectional survey (the ACTI-Cités project). Int J Behav Nutr Phys Act déc (2015.0) 12 150. DOI: 10.1186/s12966-015-0310-5 17. Menai M, Charreire H, Kesse-Guyot E, Andreeva VA, Hercberg S, Galan P. Determining the association between types of sedentary behaviours and cardiometabolic risk factors: a 6-year longitudinal study of french adults. Diabetes Metab avr (2016.0) 42 112-21. DOI: 10.1016/j.diabet.2015.08.004 18. 18.Hess F, Salze P, Weber C, Feuillet T, Charreire H, Menai M, Chen Y et al. éditeur. PLoS ONE. 4 janv 2017;12(1):e0168986. 19. 19.Perignon M, Rollet P, Tharrey M, Recchia D, Drogué S, Caillavet F et al. The revised Healthy Purchase Index (r-HPI): a validated tool for exploring the nutritional quality of household food purchases.Eur J Nutr [Internet].27 août 2022 [cité 17 nov 2022]; Disponible sur: https://link.springer.com/10.1007/s00394-022-02962-4 20. 20.Genolini C, Alacoque X, Sentenac M, Arnaud C. kml and kml3d: R Packages to Cluster Longitudinal Data. J Stat Soft [Internet]. 2015 [cité 1 avr 2021];65(4). Disponible sur: http://www.jstatsoft.org/v65/i04/ 21. Roda C, Charreire H, Feuillet T, Mackenbach JD, Compernolle S, Glonti K. Mismatch between perceived and objectively measured environmental obesogenic features in european neighbourhoods. Obes Rev janv (2016.0) 17 31-41. DOI: 10.1111/obr.12376 22. 22.Chow CK, Lock K, Madhavan M, Corsi DJ, Gilmore AB, Subramanian SV et al. Environmental Profile of a Community’s Health (EPOCH): An Instrument to Measure Environmental Determinants of Cardiovascular Health in Five Countries. Ross JS, éditeur. PLoS ONE. 10 déc 2010;5(12):e14294. 23. Saelens BE, Glanz K, Work Group I. Measures of the Food and Physical Activity Environment. Am J Prev Med avr (2009.0) 36 166-70. DOI: 10.1016/j.amepre.2009.01.006 24. Adams SA, Matthews CE, Ebbeling CB, Moore CG, Cunningham JE, Fulton J. The Effect of Social Desirability and Social approval on self-reports of physical activity. Am J Epidemiol 15 févr (2005.0) 161 389-98. DOI: 10.1093/aje/kwi054 25. Krumpal I. Determinants of social desirability bias in sensitive surveys: a literature review. Qual Quant juin (2013.0) 47 2025-47. DOI: 10.1007/s11135-011-9640-9 26. Jylhä M, Volpato S, Guralnik JM. Self-rated health showed a graded association with frequently used biomarkers in a large population sample. J Clin Epidemiol mai (2006.0) 59 465-71. DOI: 10.1016/j.jclinepi.2005.12.004 27. Jiao J, Drewnowski A, Moudon AV, Aggarwal A, Oppert JM, Charreire H. The impact of area residential property values on self-rated health: a cross-sectional comparative study of Seattle and Paris. Prev Med Rep déc (2016.0) 4 68-74. DOI: 10.1016/j.pmedr.2016.05.008
--- title: Components of the sympathetic nervous system as targets to modulate inflammation – rheumatoid arthritis synovial fibroblasts as neuron-like cells? authors: - Xinkun Cheng - Torsten Lowin - Nadine Honke - Georg Pongratz journal: Journal of Inflammation (London, England) year: 2023 pmcid: PMC10015726 doi: 10.1186/s12950-023-00336-z license: CC BY 4.0 --- # Components of the sympathetic nervous system as targets to modulate inflammation – rheumatoid arthritis synovial fibroblasts as neuron-like cells? ## Abstract ### Background Catecholamines are major neurotransmitters of the sympathetic nervous system (SNS) and they are of pivotal importance in regulating numerous physiological and pathological processes. Rheumatoid arthritis (RA) is influenced by the activity of the SNS and its neurotransmitters norepinephrine (NE) and dopamine (DA) and early sympathectomy alleviates experimental arthritis in mice. In contrast, late sympathectomy aggravates RA, since this procedure eliminates anti-inflammatory, tyrosine hydroxylase (TH) positive cells that appear in the course of RA. While it has been shown that B cells can take up, degrade and synthesize catecholamines it is still unclear whether this also applies to synovial fibroblasts, a mesenchymal cell that is actively engaged in propagating inflammation and cartilage destruction in RA. Therefore, this study aims to present a detailed description of the catecholamine pathway and its influence on human RA synovial fibroblasts (RASFs). ### Results RASFs express all catecholamine-related targets including the synthesizing enzymes TH, DOPA decarboxylase, dopamine beta-hydroxylase, and phenylethanolamine N-methyltransferase. Furthermore, vesicular monoamine transporters $\frac{1}{2}$ (VMAT$\frac{1}{2}$), dopamine transporter (DAT) and norepinephrine transporter (NET) were detected. RASFs are also able to degrade catecholamines as they express monoaminoxidase A and B (MAO-A/MAO-B) and catechol-O-methyltransferase (COMT). TNF upregulated VMAT2, MAO-B and NET levels in RASFs. DA, NE and epinephrine (EPI) were produced by RASFs and extracellular levels were augmented by either MAO, COMT, VMAT or DAT/NET inhibition but also by tumor necrosis factor (TNF) stimulation. While exogenous DA decreased interleukin-6 (IL-6) production and cell viability at the highest concentration (100 μM), NE above 1 μM increased IL-6 levels with a concomitant decrease in cell viability. MAO-A and MAO-B inhibition had differential effects on unstimulated and TNF treated RASFs. The MAO-A inhibitor clorgyline fostered IL-6 production in unstimulated but not TNF stimulated RASFs (10 nM-1 μM) while reducing IL-6 at 100 μM with a dose-dependent decrease in cell viability in both groups. The MAO-B inhibitor lazabemide hydrochloride did only modestly decrease cell viability at 100 μM while enhancing IL-6 production in unstimulated RASFs and decreasing IL-6 in TNF stimulated cells. ### Conclusions RASFs possess a complete and functional catecholamine machinery whose function is altered under inflammatory conditions. Results from this study shed further light on the involvement of sympathetic neurotransmitters in RA pathology and might open therapeutic avenues to counteract inflammation with the MAO enzymes being key candidates. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12950-023-00336-z. ## Introduction The sympathetic nervous system (SNS) is a comprehensive system mediating the ‘fight and flight’ response and is indispensable for body homeostasis during all kinds of activities [1]. It is a constant regulatory machinery transmitting signals from the central nervous system to target tissues thereby regulating their function [2]. The catecholamines norepinephrine (NE), epinephrine (EPI) and dopamine (DA) are the neurotransmitters of the SNS and are of pivotal importance in regulating inflammation and numerous physiological and pathological processes. In the central nervous system, abundant studies have illustrated the constitution and function of the catecholaminergic pathway. It is also clear, that immune cells have the capability to produce and sense catecholamines resulting in modulation of immune function [3]. Rheumatoid arthritis (RA), a chronic autoimmune disorder, is characterized by cartilage and bone damage accompanied by disability causing numerous clinical problems including persistent synovitis and articular dysfunction [4]. The SNS and its neurotransmitters play a dual role in disease onset and severity. In previous studies, it was shown that sympathectomy before the induction of experimental arthritis in mice reduces arthritic score and onset while late sympathectomy aggravated the disease. This suggests a switch from a pro- to an anti-inflammatory effect during the course of RA [5]. The anti-inflammatory effect might be due to newly appearing tyrosine hydroxylase (TH) positive catecholamine-producing cells which have been detected in synovial tissue of RA and osteoarthritis (OA) patients [6]. Mixed synovial cells were assumed to produce and release DA, representing a noncanonical mechanism in the modulation of local joint inflammation by inhibiting tumor necrosis factor (TNF) release [7]. Moreover, it was reported that RA synovial fibroblasts (SFs) possess a dopaminergic system, including dopamine receptors and their activation resulted in a reduction of inflammatory cytokine release by RASFs [8]. However, it is still unclear whether RASFs can produce and store NE or EPI and contribute to catecholamine degradation. Therefore, in this study we investigated the catecholaminergic pathway and its function in RASFs. With the MAO enzymes, we identified a potential therapeutic target, which might help to control cytokine release by RASFs. ## RASFs express all components to synthesize, transport, store, and degrade catecholamines – selective regulation by TNF In a first step, we detected key proteins involved in the synthesis (tyrosine hydroxylase, TH), catecholamine uptake and storage (dopamine transporter, DAT; vesicular monoamine transporters 1 and 2, VMAT $\frac{1}{2}$) and degradation (monoamine oxidases A and B, MAO-A/B) by Immunofluorescence (IF) and Western Blotting (WB) in RASF (Fig. 1a-f; and supplementary fig. 1). As some antibodies were not specific in western blotting, we also detected all target proteins by quantitative polymerase chain reaction (qPCR). In addition, we confirmed the expression of DOPA decarboxylase (DDC), dopamine-beta-hydroxylase (DBH), phenylethanolamine N-methyltransferase (PNMT), Catechol-O-methyltransferase (COMT) by qPCR (Fig. 1g). Since RASF are subjected to pro-inflammatory cytokines in the joint, we also investigated the regulation of these mRNAs at different time points under the influence of TNF. We found DAT to be reduced at 6 h (down 20.68 ± $7.825\%$, $$p \leq 0.0166$$) by TNF and NET (up 179.4 ± $68.20\%$, $$p \leq 0.0170$$) to be increased after 24 h, but the magnitude of regulation was small (Fig. 1g). However, VMAT2 (up 796.3 ± $217.3\%$ at 12 h, $$p \leq 0.0018$$) and MAO-B (up 650.4 ± $229.2\%$ at 24 h, $$p \leq 0.0109$$) were strongly upregulated by TNF treatment (Fig. 1g). Since VMAT and MAO were influenced by TNF on mRNA level, we also confirmed these results by WB (Fig. 2 and supplementary fig. 1). Here, we demonstrated that TNF selectively increases VMAT2 and MAO-B but not VMAT1 or MAO-A (Fig. 2b, d). Interestingly, TNF stimulation induces a specific isoform of MAO-B or introduces a posttranslational modification, as western blotting revealed the appearance of a second band with a slightly higher molecular weight (Fig. 2b).Fig. 1Expression of components of the catecholaminergic pathway in rheumatoid arthritis synovial fibroblasts (RASFs) under basal conditions and after stimulation with tumor necrosis factor (TNF). a-f Immunofluorescence (IF) and western blot (WB) images of dopamine transporter (DAT) (a), tyrosine hydroxylase (TH) (b), monoamine oxidase-A (MAO-A) (c), monoamine oxidase-B (MAO-B) (d), vesicular monoamine transporter 1 (VMAT1) (e) and vesicular monoamine transporter 2 (VMAT2) (f). Staining with respective antibody isotypes (upper left) served as negative control, target proteins are shown in green, cell nuclei (blue) were stained with DAPI (Bar 50 μm). Mouse brain homogenate was used as positive control (Ctl) in WB experiments. g-q Relative mRNA expression of components of the catecholaminergic pathway in RASFs with and without TNF stimulation for 6-, 12- and 24-hours including TH (g), DOPA decarboxylase (DDC) (h), dopamine-beta-hydroxylase (DBH) (i), phenylethanolamine N-methyltransferase (PNMT) (j), norepinephrine transporter (NET) (k), DAT (l), VMAT1 (m), VMAT2 (n), MAO-A (o), MAO-B (p) and Catechol-O-methyltransferase (COMT) (q). ( $$n = 5$$) * $p \leq 0.05$, ** $p \leq 0.01$, comparing with the control group of each stimulation at a given time point by paired Student’s t-test. Black, control conditions; Red: stimulated with TNF (10 ng/mL); 6 h, 12 h, 24 h = stimulated time in hoursFig. 2Relative protein expression of MAO-A, MAO-B, VMAT1 and VMAT2 after TNF stimulation. Synovial fibroblasts were treated with TNF (10 ng/mL) for 72 h or left untreated. Protein levels of MAO-A (a), MAO-B (b), VMAT1 (c) and VMAT2 (d) determined by WB. ( $$n = 5$$) * $p \leq 0.05$, for comparisons between TNF treated and untreated cells by paired t-test. Ctl (black), control group without TNF stimulation; TNF (red), tumor necrosis factor stimulated ## The catecholamine machinery in RASFs is functional Since RASF have all necessary enzymes to synthesize DA, NE and EPI, we assessed whether these catecholamines are actually produced by these cells. As shown in Fig. 3, the concentration of extracellular catecholamines was increased continuously from 2 hours to 24 hours. The strongest increase was found with DA, as levels at baseline were around 49 pg/ml (0.33 nM) but gradually increased over time reaching around 110 pg/mL (0.72 nM) after 24 h (+ 60.54 ± 13.17 pg/mL, $$p \leq 0.0002$$, Fig. 3a). A similar increase was found for NE, whose levels also increased from 246 pg/mL (1.46 nM) at 2 hours to 417 pg/mL after 24 h (2.47 nM; + 171.2 ± 61.88 pg/mL, $$p \leq 0.0077$$, Fig. 3a). Extracellular EPI was also increased from 32 pg/mL (0.17 nM) to 44 pg/mL (0.24 nM) after 24 h (+ 11.93 ± 6.53 pg/mL; $$p \leq 0.0002$$) but overall levels were much lower compared with DA or NE (Fig. 3a). When the conversion from L-DOPA to DA was blocked by the DDC inhibitor benserazide hydrochloride (BZD, 50 μM), both extracellular DA ($$p \leq 0.007$$) and NE ($$p \leq 0.038$$) levels declined significantly (Fig. 3b). A similar trend was observed for EPI (Fig. 3b). The intracellular catecholamine content was not altered by treatment with BZD (Fig. 3c).Fig. 3Catecholamine production by RASFs. a Time-dependent synthesis of dopamine (DA), norepinephrine (NE) and epinephrine (EPI) by RASF. Extracellular levels of catecholamines after 2-, 6-, 12- and 24-hours incubation with complete medium are shown. b, c Extracellular (b) and intracellular (c) catecholamine levels after 72-hour inhibition of DDC with benserazide hydrochloride (BZD, 50 μM). d, e Extracellular (d) and intracellular (e) catecholamine levels after 24-hours inhibition of MAO and COMT with M30 (MAO inhibitor, 10 μM) and OR486 (COMT inhibitor, 10 μM). f, g Extracellular (f) and intracellular (g) catecholamine levels after 24-hours inhibition of VMAT with reserpine (RSP, 10 μM). h, i Extracellular (h) and intracellular (i) catecholamine levels after 24-hours inhibition of DAT and NET with indatraline hydrochloride (IDA, inhibitor of both DAT and NET, 10 μM). ( $$n = 4$$–5) * $p \leq 0.05$, $p \leq 0.01$, * $p \leq 0.001$, for comparisons between inhibitor treated and untreated groups by MANN-Whitney U test, and for comparisons between 2-hour-incubation group and the other three incubation time groups by one-way ANOVA with Tukey’s post-hoc test We also investigated whether inhibition of other components of the catecholaminergic pathway alter levels of DA, NE and EPI. When MAO and COMT were inhibited by the combination of M30 (MAO inhibitor, 10 μM) and OR486 (COMT inhibitor, 10 μM), extracellular NE ($$p \leq 0.028$$) and EPI ($$p \leq 0.028$$) levels were increased (Fig. 3d), while extracellular DA and intracellular levels remained unchanged (Fig. 3d, e). Similarly, when catecholamine storage was targeted by the VMAT inhibitor and releaser reserpine (RSP, 10 μM), extracellular DA ($$p \leq 0.010$$) and intracellular DA ($$p \leq 0.032$$) was increased (Fig. 3f, g), while concomitantly, extracellular NE and EPI showed a decreasing trend (Fig. 3f, g). Targeting re-uptake of catecholamines with indatraline hydrochloride (IDA, 10 μM), an inhibitor of DAT and NET, increased extracellular levels of NE ($$p \leq 0.0317$$, Fig. 3h) while concomitantly extracellular levels of DA and EPI tended to decrease. No difference was found in intracellular levels of DA, NE and EPI (Fig. 3i). ## MAO-A, but not MAO-B inhibition increases extracellular catecholamine levels Since both MAO isoforms are able to degrade catecholamines, we used selective inhibitors to pinpoint the enzyme involved. After 24 hours stimulation, CLG (MAO-A inhibitor) significantly increased extracellular catecholamines ($$p \leq 0.0449$$ for DA and $$p \leq 0.0335$$ for NE, Fig. 4a), except for EPI, while the intracellular catecholamine levels remained unchanged (Fig. 4b). When using the selective MAO-B inhibitor LB, there was no significant modulation of intra- or extracellular catecholamine levels (Fig. 4c, d, $p \leq 0.05$).Fig. 4Extracellular and intracellular catecholamine levels after selective MAO-A or MAO-B inhibition. a, b Extracellular (a) and intracellular (b) catecholamine levels after 24-hour inhibition of MAO-A with clorgyline (CLG, 1 μM). c, d Extracellular (c) and intracellular (d) catecholamine levels after 24-hour inhibition of MAO-B with lazabemide hydrochloride (LB, 10 μM). ( $$n = 4$$–6) * $p \leq 0.05$, for comparisons between inhibitor treated and untreated groups by MANN-Whitney U test ## Exogenous DA and NE modulate interleukin 6 (IL-6) production and cell viability of RASFs Since RASFs are able to produce NE and DA, we were also interested whether these catecholamines regulate cell viability and the production of IL-6, a major cytokine produced by RASFs. After stimulation over 24 hours, we found DA at 100 μM to slightly decrease IL-6 production (Fig. 5a, upper panel), whereas cell viability was reduced already at concentrations above 100 nM (Fig. 5a, lower panel). In contrast, NE increased IL-6 production in concentrations greater than 1 μM reaching a maximum at 100 μM (+ $40\%$, $p \leq 0.0001$, Fig. 5b, upper panel). However, this was also accompanied by a dose-dependent reduction of cell viability starting at 1 μM with a maximal decrease of $24\%$ at 100 μM ($p \leq 0.0001$, Fig. 5b, lower panel).Fig. 5Dopamine and norepinephrine modulate interleukin (IL)-6 production and cell viability of RASFs. IL-6 production (a, upper panel) and cell viability (a, lower panel) after stimulation with dopamine (DA, 10 nM – 100 μM) for 24 hours ($$n = 6$$). IL-6 production (b, upper panel) and cell viability (b, lower panel) after stimulation with norepinephrine (NE, 10 nM – 100 μM) for 24 hours. ( $$n = 3$$). * $p \leq 0.05$, $p \leq 0.01$, * $p \leq 0.001$, for comparisons with the control group (without DA or NE) by one-way ANOVA with Tukey’s post-hoc test ## Modulation of catecholamine synthesis by TNF Since RA is characterized by an inflammatory environment, we also evaluated catecholamine production under the influence of TNF. After 72 hours of stimulation with TNF [10 ng/ml], no obvious difference was found in intracellular catecholamine levels (Fig. 6a-c). However, extracellular DA (+ $18\%$), NE (+ $55\%$) and EPI (+ $208\%$) were significantly increased by TNF treatment ($p \leq 0.05$, Fig. 6d-f). This increase in catecholamines was accompanied by enhanced production of IL-6 (+ $890\%$, $$p \leq 0.0022$$, Fig. 6g).Fig. 6Modulation of intra- and extracellular catecholamine levels and IL-6 production by TNF. a-f Intracellular (a-c) and extracellular (d-f) levels of dopamine (DA) (a, d), norepinephrine (NE) (b, e) and epinephrine (EPI) (c, f) levels under basal conditions and after stimulation with TNF (10 ng/mL for 24 h. ($$n = 4$$–5) g IL-6 production in 24 h after stimulation with TNF vs unstimulated controls ($$n = 3$$). * $p \leq 0.05$, ** $p \leq 0.01$, for comparisons of catecholamine levels versus control by unpaired t-test in a-f, for the comparison of IL-6 production versus control MANN-Whitney U test was employed. Ctl (black), control group; TNF (red), tumor necrosis factor ## Impact of selective MAO-A or MAO-B inhibition on basal and TNF-induced IL-6 production and cell viability We found MAO-B to be highly upregulated by TNF (Figs. 1g and 2b), and, therefore we investigated whether inactivation of this enzyme with the selective MAO-B inhibitor (LB) has any impact on cell viability and IL-6 production by RASF. In addition, we also employed the selective MAO-A inhibitor (CLG) for comparison. When MAO-A was inhibited in RASFs without TNF pretreatment, IL-6 levels were elevated (+ $19\%$ at 1 μM) by low concentrations of CLG (10 nM to 1 μM; $p \leq 0.05$) and declined at 100 μM CLG (− $38\%$; $p \leq 0.0001$, Fig. 7a). In contrast, there was no significant regulation of IL-6 by CLG in RASFs with TNF pretreatment (Fig. 7a). However, cell viability of RASFs was dose-dependently inhibited by CLG regardless of TNF stimulation ($p \leq 0.05$, Fig. 7b). MAO-B inhibition resulted in the opposite: LB significantly suppressed IL-6 production by TNF stimulated RASFs at concentrations above 1 μM with a maximum at 100 μM (− $19\%$, $p \leq 0.0001$, Fig. 7c). Without TNF pretreatment, LB increased IL-6 levels at 100 μM (+ $31\%$, $$p \leq 0.0038$$, Fig. 7c). Cell viability was slightly decreased by 100 μM LB in the control but not in the TNF pre-stimulated group ($p \leq 0.01$, Fig. 7d).Fig. 7Modulation of IL-6 production and cell viability by selective MAO inhibition. IL-6 production and cell viability of RASFs treated with different concentrations of the MAO-A inhibitor clorgyline (CLG, $$n = 6$$) (a, b) or the MAO-B inhibitor lazabemide hydrochloride (LB, $$n = 5$$) (c, d). RASFs were pre-treated with TNF for 72 hours (red) or left untreated (black). MAO-A and MAO-B inhibitors were applied thereafter for 24 hours. * $p \leq 0.05$, $p \leq 0.01$, * $p \leq 0.001$, comparing with the control group without MAOs inhibitors by one-way ANOVA with Tukey’s multiple comparison test ## Discussion To our knowledge, this is the first study to comprehensively characterize the significance of the catecholaminergic machinery in RASF. We demonstrated that RASF not only take up and store catecholamines but are also able to degrade and synthesize DA, NE and EPI. In addition, we have shown that an inflammatory environment mimicked by the addition of TNF influences catecholamine levels but also MAO-B expression which in turn modulated IL-6 production. Over the past decades, the pivotal influence of the SNS on regulating inflammation in RA was increasingly recognized [1]. Early reports already inferred that sympathectomy might be beneficial in the treatment of RA [9] and since then many other studies confirmed extensive crosstalk of the SNS with the immune system [10–13]. In fact, it has been shown that experimental arthritis is attenuated when β2-adrenergic receptor signaling is activated [14, 15]. Similarly, dopamine also influences RA pathology, as inhibition of D1-like or activation of D2-like dopamine receptors ameliorated collagen-induced arthritis in mice [16, 17]. Although the SNS aggravates experimental arthritis at an early stage [18], its effects are reversed during the course of the disease. SNS fibre density decreases in the joint, likely to nerve fibre repulsion initiated by semaphorin 3C production of macrophages and RASF [19, 20]. This decrease in SNS fibres is accompanied by the appearance of anti-inflammatory, TH positive cells in synovial tissue which are able to produce catecholamines [6, 7]. Although RASF have been identified as being TH positive [7], it was unclear until now whether these cells actively synthesize, store or degrade catecholamines. We found that RASF express all enzymes necessary (TH, DDC, DBH, PNMT) to catalyze the steps from tyrosine to epinephrine and, accordingly, we detected DA, NE and EPI both intra- and extracellularly. Although these enzymes were not regulated by TNF, we found increased production of catecholamines in response to this cytokine. TNF increases TH expression in monocytes [21] and our findings confirm results from Miller et al. who showed that increased inflammation is associated with higher TH immunoreactivity and NE production in synovial tissue [22]. In RASF, TNF might enhance catecholamine synthesis not only by increasing TH expression but also by increasing cofactors important for the activity of TH such as tetrahydrobiopterin (BH4) as demonstrated in glioma cells [23, 24]. TNF does increase tetrahydrobiopterin levels by augmenting the expression of BH4-synthesizing enzymes guanosine triphosphate cyclohydrolase I, 6-pyruvoyl tetrahydropterin synthase and sepiapterin reductase [25]. We assessed the effects of exogenous DA and NE on IL-6 production and cell viability of RASFs and found that DA reduced IL-6 levels and cell viability, whereas NE increased IL-6 while concomitantly decreasing cell viability. In a previous study, we found reduced IL-6 and IL-8 production by RASF when challenged with intermediate levels of DA, however culture conditions and incubation time were different compared to this study [8]. The increase of IL-6 by NE has already been briefly described [26] and the involvement of β adrenergic receptors in RA pathology is well established [14, 27, 28]. The reduction of cell viability by the beta-adrenergic agonists isoprenaline and salbutamol but not NE has been shown [27] and intracellular DA is associated with the generation of reactive oxygen species (ROS) and subsequent cell death in neuronal cells [29]. Since relatively high levels of NE and DA are necessary to elicit appreciable effects on cytokine production, it is likely that these catecholamines only engage in autocrine and paracrine signaling. Besides synthesizing enzymes, RASF also express VMAT1 and VMAT2 which regulate catecholamine storage [30]. However, only VMAT2 was strongly upregulated by TNF which might indicate that, under inflammatory conditions, vesicle composition is altered. It has been shown that the affinity towards monoamines is 3-fold higher for VMAT2 which, in contrast to VMAT1, also transports histamine [30]. The increase in VMAT2 might come as a consequence of elevated catecholamine synthesis induced by TNF as it has been shown that high extracellular levels of catecholamines due to stress or drug intake upregulated VMAT2 levels [31, 32]. Interestingly, in western blot analyses we found two bands of VMAT2 with slightly different molecular weights. As VMAT2 localization and function is governed by glycosylation, phosphorylation and nitrosylation [33], TNF might not only regulate overall levels but also VMAT2 activity. Inhibition of VMAT with reserpine disrupts catecholamine storage and we found an increase of extracellular catecholamines. However, reserpine releases catecholamines intracellularly and thereby depletes monoamines as they are degraded by MAO and COMT enzymes [34]. The observed increase in extracellular catecholamines induced by reserpine might be dependent on negative regulation of DAT and NET function [35, 36]. DAT and NET were both found to be expressed by RASF and their ligation with the non-selective reuptake inhibitor indatraline increased extracellular DA, NE and EPI while depleting their levels intracellularly. This is in line with results obtained in healthy volunteers that received intravenous nomifensine, another inhibitor of NET and DAT with similar pharmacologic properties [37]. We recognize that RASF synthesize, store and take up catecholamines, but do they also engage in their degradation? This is clearly the case as we detected both MAO isoforms along with COMT and their combined inhibition elevated extracellular catecholamine levels. This confirms in vivo and in vitro effects of the COMT inhibitor OR-486 or non- selective MAO inhibitors as used in this study [38, 39]. Analogous to VMAT2, MAO-B was strongly upregulated by TNF on mRNA and protein level and this pro-inflammatory cytokine might also induce posttranslational modifications in MAO-B, which was at least indicated by western blot analyses. Although this hasn’t been investigated in previous studies, purification of recombinant MAO-B from yeast revealed extensive acetylation [40] and ubiquitination is required for its insertion into mitochondrial membranes [41]. We further delineated the effects of MAO-A and MAO-B inhibition and investigated IL-6 production and cell viability of RASF under basal and TNF-stimulated conditions. MAO-A inhibition without TNF pre-treatment increased IL-6 production by RASF, whereas TNF stimulation abrogated this effect and cell viability was reduced regardless of TNF stimulation. While there is no data available regarding the influence of MAO-A on RASF function, studies from macrophages suggest that MAO-A breaks down monoamines and by doing so, increases the production of ROS. As a consequence, anti-inflammatory M2 polarization is favored and MAO-A inhibition reversed this effect [42]. In line with this, we also observed an increase in IL-6 upon MAO-A inhibition and the decreased cell viability might be due to the reduced production of ROS at high concentrations of MAO-A inhibitor. Although ROS induces cell death in high concentrations, low to intermediate concentrations actually promote cell survival and proliferation [43, 44]. Consequently, ROS levels need to be strictly controlled as too much or too little might negatively affect cell survival. MAO-B was upregulated by TNF and its inhibition decreased IL-6 production but enhanced it in TNF-naïve RASFs while cell viability was only slightly reduced by high concentration of MAO-B inhibitor in the TNF-naïve group. Although in vitro, both MAO isoforms are able to degrade catecholamines, in vivo, MAO-A was found to be mainly responsible for NE and DA degradation [45, 46], whereas MAO-B is associated with γ-aminobutyric acid (GABA) synthesis and the generation of ROS [46–48]. Accordingly, we also observed an increase in catecholamine levels with MAO-A but not MAO-B inhibition. Our results are in line with those from Won et al. who showed anti-inflammatory effects of MAO-B inhibition on cytokine production and experimental arthritis in mice [48]. Since peripheral GABA is mainly anti-inflammatory [49], the major effect of MAO-B inhibition seems to rely on the reduction of ROS. TNF induces ROS and this is associated with activation of nuclear factor ‘kappa-light-chain-enhancer’ of activated B-cells (NFκB) with pro-inflammatory consequences [50, 51]. Therefore, reduction of ROS by MAO-B inhibition might reduce the activity not only of NFκB signaling but also of other pro-inflammatory ROS-dependent pathways such as Jun activated kinase (JNK) [52] or mitogen-activated kinase [53]. ## Conclusions This is the first comprehensive study of the catecholaminergic system in RASF. We identified all components necessary for the production, storage, reuptake and degradation of catecholamines and an overview is depicted in Fig. 8. Therefore, RASFs, being of mesenchymal origin, resemble pre-synaptic sympathetic neuron-like cells in the joint. Similar to sympathetic neuron-associated macrophages, RASF might take up excess catecholamines from sympathetic nerves in the joint and store them for further use or participate in their degradation [54]. However, in situations where local catecholamine levels drop, e.g., after sympathetic nerve fibre repulsion during the course of arthritis, they might contribute to de novo catecholamine synthesis together with TH-positive lymphocytes and macrophages. Fig. 8Schematic representation of the catecholaminergic pathway in RASFs. 1. Synthesis: RASF are able to synthesize DA, NE and EPI. This needs the participation of the enzymes TH, DDC, DBH and PNMT. 2. Storage: Vesicles are able to store catecholamines by transport through VMAT1 and VMAT2. 3. Release: Catecholamines can be released to the extracellular space by exocytosis and invoke downstream reactions via DRs and ADRs in an autocrine and paracrine fashion. Released catecholamines are able to ligate receptors (ADRs and DRs) and foster or inhibit secretion of IL-6. 4. Reuptake: Exogenous catecholamines are taken up through DAT or NET and repacked into vesicles or are degraded. 5. Degradation: MAO-A, MAO-B and COMT catalyze the degradation of excess catecholamines and thereby produce the metabolites DOPAC and HVA. ( DA, dopamine; NE, norepinephrine; EPI, epinephrine; TH, tyrosine hydroxylase; DDC, DOPA decarboxylase; DBH, dopamine-beta-hydroxylase; PNMT, phenylethanolamine N-methyltransferase; VMAT$\frac{1}{2}$, vesicular monoamine transporter $\frac{1}{2}$; DRs, dopamine receptors; ADRs, adrenergic receptors; DAT, dopamine transporter; NET, norepinephrine transporter; MAO-A/B, monoamine oxidase-A/B; COMT, catechol-O-methyltransferase; DOPAC, dihydroxy-phenyl acetic acid; HVA, homovanillic acid; IL-6, interleukin-6) MAO inhibition might be an attractive therapeutic approach to target excess inflammation in RA. Very early studies already noted an improvement of RA under MAO inhibition [55] and MAO-B inhibition alone provided robust anti-inflammatory effects in experimental arthritis [48]. However, MAO-A also contributes to oxidative stress due to the degradation of catecholamines and it might be necessary to inactivate both isoforms (peripherally) for maximal effect on joint pathology. Therefore, MAO inhibitors might be investigated as adjunct therapy for RA. ## Patients and compounds In this study, 17 patients with long-standing RA fulfilling the American College of Rheumatology revised criteria for RA [56], who underwent elective knee joint replacement surgery, were included. Mean age was 70.12 ± 7.89 years for RA. Mean C-reactive protein (CRP) was 8.41 ± 11.94 mg/L for RA. Rheumatoid factor was 61.24 ± 69.57 IU/mL in RA. In the RA patient group $\frac{4}{17}$ received methotrexate, $\frac{5}{17}$ glucocorticoids and $\frac{5}{17}$ received biologicals or Janus kinase inhibitors. All patients in this study were informed about the purpose and gave written consent before surgery. This study was approved by the Ethics Committees of the University of Düsseldorf (approval number 2018–87-KFogU). The compounds and chemicals with abbreviation, order number, company and concentration used are presented in Table 1.Table 1Compounds and chemicals used in this studyCompoundAbbreviationOrder numberCompanyConcentrationTumor necrosis factorTNF300-01APeroTech10 ng/mlBenserazide hydrochlorideBZDB7283Sigma-Aldrich50 μMClorgylineCLGM3778Sigma-Aldrich1 μMReserpineRSP2742Tocris / Bio-Techne10 μMM30 dihydrochlorideM306067Tocris / Bio-Techne10 μMLazabemide hydrochlorideLB2460Tocris / Bio-Techne10 μMOR-486OR-4860483Tocris / Bio-Techne10 μMIndatraline hydrochlorideIDA1588R&D / Biotechne10 μM ## Synovial tissue preparation and SFs culture The RASF isolation and preparation was performed as described previously for in-vitro experiments [57]. Briefly, synovial tissue samples were immediately collected (up to 9 cm2) upon exposing knee joint capsule. The tissue pieces were carefully cut up into tiny fragments and digested with liberase (Roche Diagnostics, Mannheim, Germany) overnight at 37 °C. The filtration (70 μm) and centrifugation (300 g, 10 min) of the resulting suspension were carried out subsequently. After that, the pellet was obtained, which was then treated with erythrolysis buffer (20.7 g NH4Cl, 1.97 g NH4HCO3, 0.09 g EDTA and 1 L H2O) for 5 minutes. The suspension was centrifuged again for 10 minutes at 300 g. At last, RASFs were resuspended in RPMI-1640 (sigma Aldrich, St. Louis, USA) with $10\%$ FCS. After culturing overnight, cells were treated with fresh medium to wash off dead cells and debris. ## WB Cells were collected for the isolation of protein using RIPA lysis buffer (R 0278; Sigma) with complete protease inhibitor (Roche, Mannheim, Germany). Protein was quantified and subjected to electrophoresis using $12.5\%$ SDS-polyacrylamide gels with the same amount of total protein, running for 60 min at 20 mA (Biorad, Puchheim, Germany). The gels were transferred onto a nitrocellulose membrane (Biorad) at 300 mA for 90 min. Then, membranes were blocked for 1 hour with $5\%$ no-fat milk in TBS-T (Tris-Glycine-SDS Buffer from Sigma containing $0.1\%$ Tween 20) at room temperature. The following primary antibodies shown in Table 2 were applied for overnight incubation at 4 °C. These antibodies are applicable to both human and mouse samples. Subsequently, the membranes were incubated with secondary antibody (goat anti-rabbit IgG HRP, DAKO P0448, 1:2000 in $5\%$ no/fat milk) for 2 hours. Immunoreactive protein bands were visualized by ECL Prime (GE Healthcare, Freiburg, Germany). Membranes were washed three times with TBS-T for 5 min between each step. Proteins of interest were quantified in a V3 Western Workflow (Biorad) and the signals were normalized against that of GAPDH.Table 2Primary antibodies used in this study. WB = western blotting, IF = ImmunofluorescenceCatalogCompanyDilution in WBDilution in IFDAT22,524–1-APProteintech1:10001:250TH25,859–1-APProteintech1:100001:500VMAT1ATM-007Alomone1:2001:100VMAT2ATM-006Alomone1:4001:1000MAOA10,539–1-APProteintech1:30001:300MAOB12,602–1-APProteintech1:40001:50GAPDH2118CellSignalling1:2000– ## Immunofluorescence (IF) IF was performed on cultured RASFs seeded in 96-well plates at approximately 80–$90\%$ confluence. Cells were washed with PBS 5 times and fixed with cold methanol for 20 min at − 20 °C. After rinsing 4 times with PBS, cells were permeabilized with $0.3\%$ (v/v) Triton X-100 diluted in PBS for 5 min at room temperature. Afterwards, cells were blocked for 1 h at room temperature in blocking buffer (PBS with $5\%$ normal swine serum (NSS, Dako, X0901)). Primary antibodies (Table 2) and the same amount of rabbit IgG polyclonal isotype (ab37415, abcam) were separately diluted in blocking buffer. After overnight incubation with primary antibodies or the IgG isotype at 4 °C, cells were washed with $0.3\%$ (v/v) Triton X-100 diluted in PBS 4 times followed by 2 washes with PBS only. Then, secondary antibody (Goat anti-rabbit IgG (Alexa Fluor 488), ab150077, Abcam) labeled with Alexa Fluor 488-FITC was added in blocking buffer for 1 hour at room temperature. At last, cells were rinsed 6 times and covered with the ProLong™ Gold Antifade Mountant with DAPI (P36931, Invitrogen) and coverslips were kept in the dark overnight at 4 °C. Images of each target were captured by using an Axio Observer microscope (Zeiss-Germany) with a digital camera AxioCam (Zeiss-Germany). Zen 2.6 software (Blue edition-Zeiss-Germany) was used to analyze the images. ## Catecholamine enzyme-linked immunosorbent assay (ELISA) For the determination of extracellular catecholamine, cells seeded in 96-well plates were cultured until confluence reached $90\%$. Supernatants were discarded, fresh RPMI with $10\%$ FCS (50 μl/well) with specific inhibitors was added. The supernatants were collected according given protocols (E-EL-0045, E-EL-0046, E-EL-0047, Elabscience). Supernatants were collected after 2, 6, 12 and 24 hours and quantified. To obtain intracellular catecholamines, cell lysates were prepared according to the manufacturer’s instructions. Briefly, cells were seeded in 6-well plates and incubated until at least $90\%$ confluence. After stimulation, cells were collected and centrifuged at 300 g. The cell pellets were subsequently washed with pre-cooled PBS and suspended in distilled water. Cell lysis was achieved by repeated freeze-thaw cycles and by using an ultrasonic cell disrupter. Supernatants were collected after centrifugation and were used in ELISA. ## qPCR Cultured cells with and without TNF treated were harvested after 6, 12 and 24 hours. Total RNA was extracted with the RNA Mini Kit (Qiagen, Hilden, Germany) according to manufacturer’s instructions. By spectrophotometry (260 nm), the content of total RNA was measured. Afterwards, the synthesis of cDNA was performed with iScript™ gDNA clear cDNA Synthesis Kit (BIO-RAD) from the same amount of RNA (1 μg) of different samples. qPCR was performed using qPCRBIO SyGreen Mix Hi-ROX (PCR Biosystems) in a total volume of 20 μl and the StepOnePlus real-time PCR system with the primers shown in Table 3. GAPDH was used as a quantitative control for normalization. The relative expression fold-change was expressed by the values of 2-ΔΔCT [58]. Each qPCR analysis was conducted at least in duplicates. Table 3Primer sequences used in this studyGenesPrimer sequencesForwardReverseTH5′-TGTCCACGCTGTACTGGTTC-3′5′-AGCTCCTGAGCTTGTCCTTG-3′DDC5′-GAACAGACTTAACGGGAGCCTTT-3′5′-AATGCCGGTAGTCAGTGATAAGC-3′DBH5′-GACGCCTGGAGTGACCAGAA-3′5′-CAGTGACCGGAACGGCTC-3′ ## Il-6 ELISA RASFs (10,000 cells/well) were seeded in 96-well plates at least 24 hours before stimulation. Supernatants were collected after indicated time for the quantification of IL-6. Experiments were performed as described in the manufacturer’s protocol (human IL-6, from BD, OptEIA, Heidelberg, Germany). ## Cell viability assay After collecting the supernatants of treated RASFs, cells were incubated with CellTiter-Blue reagent following the instruction of the manufacturer (G8081, Promega). By determining the reduction from resazurin to resorufin, the cell viability was estimated and quantified to reflect the toxic effect of each treatment. ## Statistics All data were presented from at least three independent experiments. GraphPad Prism (GraphPad software Inc., California, USA) was used for data analysis. The statistic tests used are given in the figure legends. When data are presented as line plots, the line represents the mean. When data are presented as bar charts, the top of the bar represents the mean and error bars depict the standard error of the mean (SEM). When data are presented as box plots, the boxes represent the 25th to 75th percentiles, the lines within the boxes represent the median, and the lines outside the boxes represent the 10th and 90th percentiles. The level of significance was $p \leq 0.05.$ ## Supplementary Information Additional file 1: Supplement fig. 1. Original and uncropped images of WB. ## Disclosure The authors have nothing to disclose. ## References 1. Pongratz G, Straub RH. The sympathetic nervous response in inflammation. Arthritis Res Ther (2014.0) 16 1-12. DOI: 10.1186/s13075-014-0504-2 2. Scott-Solomon E, Boehm E, Kuruvilla R. The sympathetic nervous system in development and disease. Nat Rev Neurosci (2021.0) 22 685-702. DOI: 10.1038/s41583-021-00523-y 3. Jenei-Lanzl Z. Anti-inflammatory effects of cell-based therapy with tyrosine hydroxylase-positive catecholaminergic cells in experimental arthritis. Ann Rheum Dis (2015.0) 74 444-451. DOI: 10.1136/annrheumdis-2013-203925 4. Smolen JS, Aletaha D, McInnes IB. Rheumatoid arthritis. Lancet (2016.0) 388 2023-2038. DOI: 10.1016/S0140-6736(16)30173-8 5. Harle P. An opposing time-dependent immune-modulating effect of the sympathetic nervous system conferred by altering the cytokine profile in the local lymph nodes and spleen of mice with type II collagen-induced arthritis. Arthritis Rheum (2005.0) 52 1305-1313. DOI: 10.1002/art.20987 6. Capellino S. First appearance and location of catecholaminergic cells during experimental arthritis and elimination by chemical sympathectomy. Arthritis Rheum (2012.0) 64 1110-1118. DOI: 10.1002/art.33431 7. Capellino S. Catecholamine-producing cells in the synovial tissue during arthritis: modulation of sympathetic neurotransmitters as new therapeutic target. Ann Rheum Dis (2010.0) 69 1853-1860. DOI: 10.1136/ard.2009.119701 8. Capellino S. Increased expression of dopamine receptors in synovial fibroblasts from patients with rheumatoid arthritis: inhibitory effects of dopamine on interleukin-8 and interleukin-6. Arthritis Rheumatol (2014.0) 66 2685-2693. DOI: 10.1002/art.38746 9. Herfort RA. Extended sympathectomy in the treatment of chronic arthritis. J Am Geriatr Soc (1957.0) 5 904-915. DOI: 10.1111/j.1532-5415.1957.tb00490.x 10. 10.Gao D, et al. Neuroimmune Crosstalk in Rheumatoid Arthritis. Int J Mol Sci . 2022;23(15):8158. 10.3390/ijms23158158. 11. 11.Bellinger DL, Lorton D. Sympathetic Nerve Hyperactivity in the Spleen: Causal for Nonpathogenic-Driven Chronic Immune-Mediated Inflammatory Diseases (IMIDs)? Int J Mol Sci . 2018;19(4):1188. 10.3390/ijms19041188. 12. Scanzano A, Cosentino M. Adrenergic regulation of innate immunity: a review. Front Pharmacol (2015.0) 6 171. DOI: 10.3389/fphar.2015.00171 13. Klatt S. Peripheral elimination of the sympathetic nervous system stimulates immunocyte retention in lymph nodes and ameliorates collagen type II arthritis. Brain Behav Immun (2016.0) 54 201-210. DOI: 10.1016/j.bbi.2016.02.006 14. Bellinger DL. Driving beta2- While Suppressing alpha-Adrenergic Receptor Activity Suppresses Joint Pathology in Inflammatory Arthritis. Front Immunol (2021.0) 12 628065. DOI: 10.3389/fimmu.2021.628065 15. Pongratz G, Melzer M, Straub RH. The sympathetic nervous system stimulates anti-inflammatory B cells in collagen-type II-induced arthritis. Ann Rheum Dis (2012.0) 71 432-439. DOI: 10.1136/ard.2011.153056 16. Nakashioya H. Therapeutic effect of D1-like dopamine receptor antagonist on collagen-induced arthritis of mice. Modern rheumatology (2011.0) 21 260-266. DOI: 10.3109/s10165-010-0387-2 17. 17.Lu J-H, et al. Dopamine D2 receptor is involved in alleviation of type II collagen-induced arthritis in mice. Arthritis Rheum. 2008;58(8):2347–55. 10.1002/art.23628. 18. Harle P. An early sympathetic nervous system influence exacerbates collagen-induced arthritis via CD4+CD25+ cells. Arthritis Rheum (2008.0) 58 2347-2355. DOI: 10.1002/art.23628 19. Fassold A. Soluble neuropilin-2, a nerve repellent receptor, is increased in rheumatoid arthritis synovium and aggravates sympathetic fiber repulsion and arthritis. Arthritis Rheum (2009.0) 60 2892-2901. DOI: 10.1002/art.24860 20. Miller LE. Increased prevalence of semaphorin 3C, a repellent of sympathetic nerve fibers, in the synovial tissue of patients with rheumatoid arthritis. Arthritis Rheum (2004.0) 50 1156-1163. DOI: 10.1002/art.20110 21. Gopinath A. TNFalpha increases tyrosine hydroxylase expression in human monocytes. NPJ Parkinsons Dis (2021.0) 7 62. DOI: 10.1038/s41531-021-00201-x 22. Miller LE. Norepinephrine from synovial tyrosine hydroxylase positive cells is a strong indicator of synovial inflammation in rheumatoid arthritis. J Rheum (2002.0) 29 427-435. PMID: 11908553 23. Oka K. Kinetic properties of tyrosine hydroxylase with natural tetrahydrobiopterin as cofactor. Biochimica et Biophysica Acta (BBA)-Enzymology (1981.0) 661 45-53. DOI: 10.1016/0005-2744(81)90082-6 24. Vann LR. Involvement of sphingosine kinase in TNF-alpha-stimulated tetrahydrobiopterin biosynthesis in C6 glioma cells. J Biol Chem (2002.0) 277 12649-12656. DOI: 10.1074/jbc.M109111200 25. Staats Pires A. Kynurenine and Tetrahydrobiopterin Pathways Crosstalk in Pain Hypersensitivity. Front Neurosci (2020.0) 14 620. DOI: 10.3389/fnins.2020.00620 26. Raap T. Neurotransmitter modulation of interleukin 6 (IL-6) and IL-8 secretion of synovial fibroblasts in patients with rheumatoid arthritis compared to osteoarthritis. J Rheumatol (2000.0) 27 2558-2565. PMID: 11093434 27. Wu H. β2-adrenoceptor signaling reduction is involved in the inflammatory response of fibroblast-like synoviocytes from adjuvant-induced arthritic rats. Inflammopharmacology (2019.0) 27 271-279. DOI: 10.1007/s10787-018-0477-x 28. Honke N, Wiest CJ, Pongratz G. beta2-Adrenergic Receptor Expression and Intracellular Signaling in B Cells Are Highly Dynamic during Collagen-Induced Arthritis. Biomedicines (2022.0) 10 1950. DOI: 10.3390/biomedicines10081950 29. Sango J. USP10 inhibits the dopamine-induced reactive oxygen species-dependent apoptosis of neuronal cells by stimulating the antioxidant Nrf2 activity. J Biol Chem (2022.0) 298 101448. DOI: 10.1016/j.jbc.2021.101448 30. Erickson JD. Distinct pharmacological properties and distribution in neurons and endocrine cells of two isoforms of the human vesicular monoamine transporter. Proc Natl Acad Sci (1996.0) 93 5166-5171. DOI: 10.1073/pnas.93.10.5166 31. Tillinger A. Vesicular monoamine transporters (VMATs) in adrenal chromaffin cells: stress-triggered induction of VMAT2 and expression in epinephrine synthesizing cells. Cell Mol Neurobiol (2010.0) 30 1459-1465. DOI: 10.1007/s10571-010-9575-z 32. Schwartz K. Cocaine, but not amphetamine, short term treatment elevates the density of rat brain vesicular monoamine transporter 2. J Neural Transm (2007.0) 114 427-430. DOI: 10.1007/s00702-006-0549-8 33. German CL. Regulation of the Dopamine and Vesicular Monoamine Transporters: Pharmacological Targets and Implications for Disease. Pharmacol Rev (2015.0) 67 1005-1024. DOI: 10.1124/pr.114.010397 34. 34.Cheung M, Parmar M. Reserpine. in StatPearls. Front Aging Neurosci. 2017;27(9):78. 10.3389/fnagi.2017.00078. 35. Metzger RR. Inhibitory effect of reserpine on dopamine transporter function. Eur J Pharmacol (2002.0) 456 39-43. DOI: 10.1016/S0014-2999(02)02647-X 36. Mandela P. Reserpine-induced reduction in norepinephrine transporter function requires catecholamine storage vesicles. Neurochem Int (2010.0) 56 760-767. DOI: 10.1016/j.neuint.2010.02.011 37. Scheinin M. Noradrenergic and dopaminergic effects of nomifensine in healthy volunteers. Clin Pharmacol Ther (1987.0) 41 88-96. DOI: 10.1038/clpt.1987.15 38. Su Y. Membrane bound catechol-O-methytransferase is the dominant isoform for dopamine metabolism in PC12 cells and rat brain. Eur J Pharmacol (2021.0) 896 173909. DOI: 10.1016/j.ejphar.2021.173909 39. Aubin N. SL25. 1131 [3 (S), 3a (S)-3-Methoxymethyl-7-[4, 4, 4-trifluorobutoxy]-3, 3a, 4, 5-tetrahydro-1, 3-oxazolo [3, 4-a] quinolin-1-one], a new, reversible, and mixed inhibitor of monoamine oxidase-A and monoamine oxidase-B: biochemical and behavioral profile. J Pharmacol Exp Ther (2004.0) 310 1171-1182. DOI: 10.1124/jpet.103.064782 40. Newton-Vinson P, Hubalek F, Edmondson DE. High-level expression of human liver monoamine oxidase B in Pichia pastoris. Protein Expr Purif (2000.0) 20 334-345. DOI: 10.1006/prep.2000.1309 41. Zhaung Z, McCauley R. Ubiquitin is involved in the in vitro insertion of monoamine oxidase B into mitochondrial outer membranes. J Biol Chem (1989.0) 264 14594-14596. DOI: 10.1016/S0021-9258(18)63734-2 42. Wang YC. Targeting monoamine oxidase A-regulated tumor-associated macrophage polarization for cancer immunotherapy. Nat Commun (2021.0) 12 3530. DOI: 10.1038/s41467-021-23164-2 43. Graceffa V. Therapeutic Potential of Reactive Oxygen Species: State of the Art and Recent Advances. SLAS Technol (2021.0) 26 140-158. DOI: 10.1177/2472630320977450 44. Kim BY, Han MJ, Chung AS. Effects of reactive oxygen species on proliferation of Chinese hamster lung fibroblast (V79) cells. Free Radic Biol Med (2001.0) 30 686-698. DOI: 10.1016/S0891-5849(00)00514-1 45. Sturza A. Monoamine Oxidase-Related Vascular Oxidative Stress in Diseases Associated with Inflammatory Burden. Oxid Med Cell Longev (2019.0) 2019 8954201. DOI: 10.1155/2019/8954201 46. Cho HU. Redefining differential roles of MAO-A in dopamine degradation and MAO-B in tonic GABA synthesis. Exp Mol Med (2021.0) 53 1148-1158. DOI: 10.1038/s12276-021-00646-3 47. Yoon BE. Glial GABA, synthesized by monoamine oxidase B, mediates tonic inhibition. J Physiol (2014.0) 592 4951-4968. DOI: 10.1113/jphysiol.2014.278754 48. Won W. Inhibiting peripheral and central MAO-B ameliorates joint inflammation and cognitive impairment in rheumatoid arthritis. Exp Mol Med (2022.0) 54 1188-1200. DOI: 10.1038/s12276-022-00830-z 49. Bhandage AK, Barragan A. GABAergic signaling by cells of the immune system: more the rule than the exception. Cell Mol Life Sci (2021.0) 78 5667-5679. DOI: 10.1007/s00018-021-03881-z 50. Kastl L. TNF-α mediates mitochondrial uncoupling and enhances ROS-dependent cell migration via NF-κB activation in liver cells. FEBS letters (2014.0) 588 175-183. DOI: 10.1016/j.febslet.2013.11.033 51. Blaser H. TNF and ROS crosstalk in inflammation. Trends Cell Biol (2016.0) 26 249-261. DOI: 10.1016/j.tcb.2015.12.002 52. Shen H-M, Liu Z-G. JNK signaling pathway is a key modulator in cell death mediated by reactive oxygen and nitrogen species. Free Radic Biol Med (2006.0) 40 928-939. DOI: 10.1016/j.freeradbiomed.2005.10.056 53. Schieber M, Chandel NS. ROS function in redox signaling and oxidative stress. Curr Biol (2014.0) 24 R453-R462. DOI: 10.1016/j.cub.2014.03.034 54. Pirzgalska RM. Sympathetic neuron-associated macrophages contribute to obesity by importing and metabolizing norepinephrine. Nat Med (2017.0) 23 1309-1318. DOI: 10.1038/nm.4422 55. Scherbel AL. The effect of isoniazid and of iproniazid in patients with rheumatoid arthritis. Cleve Clin Q (1957.0) 24 90-97. DOI: 10.3949/ccjm.24.2.90 56. Aletaha D. 2010 Rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis Rheum (2010.0) 62 2569-2581. DOI: 10.1002/art.27584 57. Lowin T. Selective killing of proinflammatory synovial fibroblasts via activation of transient receptor potential ankyrin (TRPA1). Biochem Pharmacol (2018.0) 154 293-302. DOI: 10.1016/j.bcp.2018.05.015 58. Pfaffl MW. A new mathematical model for relative quantification in real-time RT–PCR. Nucleic Acids Res (2001.0) 29 e45. DOI: 10.1093/nar/29.9.e45
--- title: Temporal relationship between sleep duration and obesity among Chinese Han people and ethnic minorities authors: - Zhengxing Xu - Min Chen - Yuntong Yao - Lisha Yu - Peijing Yan - Huijie Cui - Ping Li - Jiaqiang Liao - Ben Zhang - Yuqin Yao - Zhenmi Liu - Xia Jiang - Tao Liu - Chenghan Xiao journal: BMC Public Health year: 2023 pmcid: PMC10015728 doi: 10.1186/s12889-023-15413-4 license: CC BY 4.0 --- # Temporal relationship between sleep duration and obesity among Chinese Han people and ethnic minorities ## Abstract ### Background No studies have assessed the association between sleep duration and obesity in Chinese ethnic minorities. Whether the relationship between sleep duration and obesity is different between Chinese Han people and Chinese ethnic minorities remains unclear. The study aimed to explore the relationship between sleep duration and obesity among Chinese Han people and Chinese ethnic minorities. ### Methods We applied data from the Guizhou Population Health Cohort Study (GPHCS), which 9,280 participants were recruited in the baseline survey from 2010 to 2012, and 8,163 completed the follow-up survey from 2016 to 2020. A total of 5,096 participants (3,188 Han Chinese and 1,908 ethnic minorities) were included in the ultimate analysis. Information on sleep duration (total 24-hour sleep time), body mass index (BMI), and waist circumference (WC) was collected at the baseline and follow-up survey, respectively. Cross-lagged panel analyses were conducted to explore the temporal relationship between sleep duration and obesity for Han people and ethnic minorities. ### Results For Han people, the results from cross-lagged panel analyses indicated that baseline sleep duration was significantly associated with follow-up BMI (βBMI = -0.041, $95\%$ CIBMI: -0.072 ~ -0.009) and follow-up WC (βWC = -0.070, $95\%$CIWC: -0.103 ~ -0.038), but baseline BMI (βBMI = -0.016, $95\%$ CIBMI: -0.050 ~ 0.018) and baseline WC (βWC = -0.019, $95\%$ CIWC: -0.053 ~ 0.016) were not associated with follow-up sleep duration. In addition, the relationship between baseline sleep duration and follow-up BMI was gender-specific and significant only in the Han people female (βBMI = -0.047, $95\%$ CIBMI: -0.090 ~ -0.003) but not in the Han people male (βBMI = -0.029, $95\%$ CIBMI: -0.075 ~ 0.016). For ethnic minorities, the results indicated that there was no relationship between sleep duration and obesity at all, either from sleep duration to obesity (βBMI = 0.028, $95\%$CIBMI: -0.012 ~ 0.068; βWC = 0.020, $95\%$CIWC: -0.022 ~ 0.062), or from obesity to sleep duration (βBMI = -0.022, $95\%$CIBMI: -0.067 ~ 0.022; βWC = -0.042, $95\%$CIWC: -0.087 ~ 0.003). ### Conclusion The relationship pattern between sleep duration and obesity across Han people and ethnic minorities is different. Future sleep-aimed overweight and obesity intervention should be conducted according to population characteristics. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12889-023-15413-4. ## Background Obesity has become a global epidemic and a significant health challenge worldwide [1]. There were more than 650 million adults suffering from obesity, according to the world health organization report [2]. Thus, a growing body of studies has been done to identify modifiable risk factors for obesity. In the recent two decades, sleep duration, a critical measurement reflecting sleep quality, has been proposed as a potential factor contributing to obesity [3]. In general, most existing evidence consented to the association between short sleep duration and obesity in adults [4–8]. However, some evidence indicated that the effect size and direction of this association might vary across populations and countries [9–12]. For findings about effect size, one study suggested that African Americans with short sleep duration are more susceptible to obesity than Caucasians [9]. Furthermore, for findings about the association direction, though most studies indicated a unidirectional inversely effect from sleep duration to obesity [4–6], some studies from western developed countries indicated different association patterns [10–12]. For instance, studies from the United States and Britain consented to a unidirectional inversely effect from obesity to sleep duration [10, 11], while a study from the Netherlands indicated a bidirectional inversely relationship between obesity and sleep duration [12]. One explanation for the inconsistent findings is that some sociodemographic factors may moderate the effect size and direction of the association between sleep duration and obesity. Unfortunately, existing evidence mainly comes from western developed countries. We know little if the relationship between sleep duration and obesity will vary among people in developing countries. With regard to developing countries, China has been suffering a surge of obesity in the past decades [13]. About 85 million Chinese adults with body mass index (BMI) ≥ 28.0 kg/m² in 2018, the figure was three times compared with 2004 [14]. Furthermore, China is a unified multi-ethnic country consisting of Han people and 55 ethnic minorities, of which the minority population exceeds 125 million [15]. Sleep duration and obesity are significantly different among Han and ethnic minorities in China due to variations in sociodemographic factors [16, 17]. However, it remains unclear whether the relationship between sleep duration and obesity is different between Han people and ethnic minorities. When exploring the association between sleep duration and obesity in China, most available studies only focused on Chinese Han people [18, 19], leading to a poor understanding of the association for Chinese ethnic minorities. For the paucity of previous ethnic minorities studies, one possible reason is the inadequacy of minority samples. Although ethnic minorities account for $8.89\%$ of China’s total population [15], most sample surveys fail to obtain sufficient representative samples for valid statistical inference due to widespread distribution across the country [20]. Guizhou province is located in southwestern China, it is one of the primary concentrations of ethnic minorities in China, with more than $36.44\%$ of the whole province’s population being ethnic minorities, including the Miao, Buyi, Dong and so forth [21]. The extensive minority population in Guizhou province offers the possibility to explore the relationship between sleep duration and obesity among ethnic minorities. Therefore, leveraging a longitudinal data from Guizhou containing a significant proportion of Chinese ethnic minorities ($37.44\%$), this study attempted to examine the relationship between sleep duration and obesity across Chinese Han people and ethnic minorities. Given the evidence from relevant studies in western developed countries [10–12] and also considering that Chinese Han people and ethnic minorities have different demographic characteristics [16, 17], this study hypothesized that the relationship between sleep duration and obesity is different among Chinese Han people and ethnic minorities. This study can contribute to the knowledge about the association between sleep duration and obesity, and the moderating effect of ethnicity on the relationship among Chinese people. ## Study population and sample We used data from two stages of the Guizhou Population Health Cohort Study (GPHCS) to accomplish the analyses. The GPHCS conducted a multistage stratified cluster random sampling method to recruit participants in Guizhou province, China. Detailed information related to study design and sampling strategy has been reported elsewhere [22]. Briefly, a total of 9,280 individuals aged 18 years and older from 48 townships of 12 districts in Guizhou province were recruited from November 2010 to December 2012, and 8,163 individuals completed the follow-up survey from December 2016 to June 2020. This study was approved by the Institutional Review Board of Guizhou Province Centre for Disease Control and Prevention (No. S2017-02) [22]. All participants signed informed consent before the data collection. To explore the relationship between sleep duration and obesity, we excluded 3,067 individuals with missing or invalid information for sleep duration, height, weight, waist circumference, or other covariates (e.g., drinking, energy intake, or physical activity). At last, we included a total of 5,096 participants in the subsequent analysis, with an average follow-up period of 7.12 years (standard deviation = 1.13 years) (Fig. 1). Fig. 1Flow chart of participants ## Assessment of sleep duration Self-reported sleep duration was obtained through a questionnaire by asking “how long do you sleep on a typical day?“. It’s important to note that the sleep duration assessed in this study refers to the 24-hour total sleep time. Each participant’s answer was converted to hours to represent the total sleep duration per day. ## Anthropometric measurements Anthropometric measurements, including height, weight, and waist circumference (WC), were obtained by a standardized physical examination. Height and weight were measured using unified height meters (accuracy is 0.1 cm) and electronic weight scales (accuracy is 0.1 kg). Body mass index (BMI) was calculated as weight in kilograms (kg) divided by height in meters squared (m2). WC (cm) was measured using a waist ruler (accuracy is 0.1 cm) at the midpoint between the lower rib cage and the iliac crest. ## Measurement of covariates Information on several covariates, including age, gender, place of residence, education levels, marital status, smoking status, alcohol consumption, dietary energy intake, physical activity, and sedentary behavior, were collected to adjust for confounding factors. Place of residence was categorized as 1 = rural and 2 = urban. Education was classified into four categories: 1 = illiterate, 2 = primary school, 3 = secondary school, and 4 = college and above. Marital status was classed as 1 = married and 2 = other (separation, divorce, widowed, spinsterhood, or cohabit). Smoking status was categorized as 1 = non-smoker and 2 = smoker. Drinking status was categorized as 1 = non-drinker and 2 = drinker. Energy intake (kcal/day) was assessed by daily dietary intake. Specifically, habitual diets regarding the previous 12 months were assessed using a semi-quantitative food frequency questionnaire (FFQ) with 14 food groups (cereals, tubers, pork, livestock, poultry, aquatic products, vegetables, fruits, juice and beverage, eggs, dairy products, bean products, and fried products) [23]. For each food group, participants were required to report the quantity and frequency. According to the information from FFQ, we estimated the total daily energy intake based on the Chinese Food Composition Tables published in 2009 [24]. Physical activity and sedentary behavior were assessed using the Global Physical Activity Questionnaire [25], and physical activity intensity level was classified as 1 = low, 2 = moderate, and 3 = high [26]. ## Statistical analyses Analyses were performed by using R version 4.0.5. All tests were conducted on two-sided, and P value less than or equal to 0.05 was considered statistically significant. Cross-lagged panel analyses were performed to examine the longitudinal relationship of sleep duration with BMI and WC across Han people and ethnic minorities. The cross-lagged panel analysis is a form of path analysis that simultaneously examines reciprocal, longitudinal relationships among a set of intercorrelated variables [27], which has been widely used in epidemiological studies [11, 28, 29]. A parsimonious model version is depicted in Fig. 2. A significant path coefficient (β1 or β2) suggests the directionality between the two variables measured over time. The cross-lagged path models were estimated based on the correlation matrix using the maximum likelihood method by the R package “Lavaan“ [30]. The validity of model fitting was assessed by root mean square residual (RMR) and comparative fit index (CFI) [31]. RMR < 0.05 and CFI > 0.90 indicate a relatively good fit for the observed data [28, 29]. Fig. 2Cross-lagged path analysis of sleep duration with BMI and WC in the Han people (A) and ethnic minorities (B), adjusted for age, gender, place of residence, education levels, marital status, smoking status, alcohol consumption, dietary energy intake, physical activity, sedentary behavior, and follow-up years; β1 represents cross-lagged path coefficients from baseline sleep duration to follow-up BMI or WC; β2 represents from baseline BMI or WC to follow-up sleep duration; r1 and r2 represent tracking correlations; r3 represent synchronous correlations; R2 represents variance explained. Goodness-of-fit (A, Han people, BMI): CFI = 1, RMR = 0.004; Goodness-of-fit (A, Han people, WC): CFI = 0.994, RMR = 0.010; Goodness-of-fit (B, Ethnic minorities, BMI): CFI = 1, RMR = 0.007; Goodness-of-fit (B, Ethnic minorities, WC): CFI = 0.996, RMR = 0.010. The cross-lagged path coefficients are presented as β (lower $95\%$ CI, upper $95\%$ CI). * P ≤ 0.05, P ≤ 0.01, P ≤ 0.001 Before cross-lagged path analysis, the baseline and follow-up of sleep duration, BMI and WC were adjusted for all covariates mentioned above by regression residual analyses. Then, the residual was standardized with Z-transformation (mean = 0; standard deviation = 1) [28, 29] in Han people and ethnic minorities, respectively. At last, the standard Z score of sleep duration, BMI and WC were applied in the cross-lagged path analyses. In addition, two sensitivity analyses were carried out in this study. First, the bootstrap simulation with 1000 replicates was performed to obtain $95\%$ confidence intervals (CIs) to evaluate sensitivity to the distributions of cross-lagged path coefficients. The second analysis was executed by gender subgroups in Han people and ethnic minorities to test whether gender influences the temporal relationship of sleep duration with BMI and WC. ## Characteristics of study participants As shown in Tables 1, a total of 3,188 Han people participants and 1,908 ethnic minorities participants were included in the analyses. There was a significant difference between Han people and ethnic minorities in almost all characteristics, except for age and gender. In addition, from baseline to follow-up, compared with Han people, ethnic minorities experienced a more significant reduction in sleep duration (Han people: 0.21 h, Ethnic minority: 0.31 h, $$P \leq 0.049$$) but a lesser increase in BMI (Han people: 0.69 kg/m2, Ethnic minority: 0.57 kg/m2, $$P \leq 0.207$$) and WC (Han people: 5.96 cm, Ethnic minority: 5.23 cm, $$P \leq 0.013$$). Table 1Characteristics at baseline and follow-up by Han people and Ethnic minoritiesCharacteristicsTotal ($$n = 5$$,096)Han people($$n = 3$$,188)Ethnic minorities ($$n = 1$$,908)P-value Baseline Age, years44.02 ± 14.9544.33 ± 15.0843.49 ± 14.720.053Male2,425 (47.57)1,492 (46.80)933 (48.90)0.147Urban1,796 (35.24)1,607 (50.41)189 (9.91)< 0.001Education< 0.001Illiteracy1,036 (20.33)602 (18.89)434 (22.75)Primary1,734 (34.03)1,013 (31.78)721 (37.79)Secondary2,050 (40.23)1,366 (42.85)684 (35.85)Collage and above276 (5.42)207 (6.49)69 (3.62)Married4,099 (80.44)2,608 (81.81)1,491 (78.15)0.001Smoking1,480 (29.04)1,002 (31.43)478 (25.05)< 0.001Drinking1,694 (33.24)980 (30.74)714 (37.42)< 0.001Energy intake, kcal/d2,112.09 ± 855.572,030.76 ± 826.382,247.98 ± 885.94< 0.001Physical activity< 0.001Low1,289 (25.29)743 (23.31)546 (28.62)Moderate1,126 (22.10)790 (24.78)336 (17.61)High2,681 (52.61)1,655 (51.91)1,026 (53.77)Sedentary duration, h4.09 ± 2.254.28 ± 2.313.77 ± 2.10< 0.001Sleep duration, h7.88 ± 1.147.79 ± 1.178.05 ± 1.07< 0.001BMI, kg/m222.86 ± 3.1022.99 ± 3.1122.64 ± 3.06< 0.001WC, cm76.74 ± 9.2777.33 ± 9.3375.77 ± 9.08< 0.001 Follow up Sleep duration, h7.64 ± 1.487.58 ± 1.557.74 ± 1.34< 0.001BMI, kg/m223.51 ± 3.2523.68 ± 3.2823.21 ± 3.18< 0.001WC, cm82.43 ± 9.3183.29 ± 9.3381.00 ± 9.10< 0.001 Change from baseline to follow up Δ Sleep duration, h-0.25 ± 1.75-0.21 ± 1.18-0.31 ± 1.630.049Δ BMI, kg/m20.64 ± 3.370.69 ± 3.450.57 ± 3.240.207Δ WC, cm5.69 ± 10.335.96 ± 10.455.23 ± 10.130.013Abbreviation: BMI = body mass index, WC = waist circumferenceΔ represents the mean change from baseline to follow-up measurementsData are frequency (%) for categorical variables and mean ± standard deviation for continuous variables ## Cross-lagged panel analyses between sleep duration and obesity As shown in Fig. 2, results from Cross-lagged panel analyses indicated that the relationship pattern between sleep duration and obesity across Han people and ethnic minorities could be distinct. For Han people (Fig. 2A), results indicated that baseline sleep duration was significantly associated with follow-up BMI (βBMI = -0.041, $95\%$CIBMI: -0.072 ~ -0.009), and WC (βWC = -0.070, $95\%$CIWC: -0.103 ~ -0.038). However, our results rejected the inverse effect from either baseline BMI (βBMI = -0.016, $95\%$CIBMI: -0.050 ~ 0.018) nor baseline WC (βWC = -0.019, $95\%$CIWC: -0.053 ~ 0.016) to follow-up sleep duration. Model fitting parameters (RMR = 0.004 and CFI = 1 in the Sleep duration-BMI model and RMR = 0.010 and CFI = 0.994 in the Sleep duration-WC model) indicated an acceptable model fitness. Distinct from Han people, our results indicated that there was no relationship between sleep duration and obesity for ethnic minorities at all (Fig. 2B). The significant relationship between baseline sleep duration and follow-up obesity index disappeared for this group (βBMI = 0.028, $95\%$CIBMI: -0.012 ~ 0.068; βWC = 0.020, $95\%$CIWC: -0.022 ~ 0.062). Model fitting parameters were RMR = 0.007 and CFI = 1 in the Sleep duration-BMI model and RMR = 0.010 and CFI = 0.996 in the Sleep duration-WC model, indicating an acceptable model fitness. ## Sensitivity analyses We performed the first sensitivity analysis using bootstrap simulation and obtained $95\%$ confidence intervals for cross-lagged path coefficients (Supplement Fig. 1). Though the confidence interval changed slightly, the results corroborated with the conclusion from cross-lagged path analyses. We performed the second sensitivity analysis by estimating the cross-lagged path coefficients for males and females separately. For Han people, the results showed a subtle gender difference in the relationship between sleep duration and follow up BMI, but not follow up WC. Precisely, the path coefficient from baseline sleep duration to follow-up BMI was significant in the Han people female (βBMI = -0.047, $95\%$CIBMI: -0.090 ~ -0.003) but not significant in the Han people male (βBMI= -0.029, $95\%$CIBMI: -0.075 ~ 0.016) (Fig. 3A). For ethnic minorities, the results showed no significant relationship of sleep duration with BMI and WC, and this non-significant relationship did not differ by gender (Fig. 3B). Fig. 3Cross-lagged path analysis of sleep duration with BMI and WC in the Han people (A) and Ethnic minorities (B) by sex groups, adjusted for age, place of residence, education levels, marital status, smoking status, alcohol consumption, dietary energy intake, physical activity, sedentary behavior, and follow-up years; β1 represents cross-lagged path coefficients from baseline sleep duration to follow-up BMI or WC; β2 represents from baseline BMI or WC to follow-up sleep duration; r1 and r2 represent tracking correlations; r3 represent synchronous correlations; R2 represents variance explained. Goodness-of-fit (A, Han people, BMI, male): CFI = 1, RMR = 0.004; Goodness-of-fit (A, Han people, BMI, female): CFI = 1, RMR = 0.003; Goodness-of-fit (A, Han people, WC, male): CFI = 0.989, RMR = 0.014; Goodness-of-fit (A, Han people, WC, female): CFI = 1, RMR = 0.006; Goodness-of-fit (B, Ethnic minorities, BMI, male): CFI = 1, RMR = 0.009; Goodness-of-fit (B, Ethnic minorities, BMI, female): CFI = 1, RMR = 0.003; Goodness-of-fit (B, Ethnic minorities, WC, male): CFI = 1, RMR = 0.007; Goodness-of-fit (B, Ethnic minorities, WC, female): CFI = 0.998, RMR = 0.011. The cross-lagged path coefficients are presented as β (lower $95\%$ CI, upper $95\%$ CI). P ≤ 0.05, P ≤ 0.01, P ≤ 0.001 ## Discussion To the best of our knowledge, this study is the first to investigate the temporal relationship between sleep duration and obesity among Chinese Han people and Chinese ethnic minorities. The results showed that the relationship pattern between sleep duration and obesity across Han people and ethnic minorities was different. For Han people, our results showed that sleep duration has an impact on subsequent obesity, but not vice versa. For the inverse effect of sleep duration on obesity, our results are consistent with previous findings in the general population [7, 8]. Previous studies have shown that sleep deprivation may decrease leptin and increase gastric hunger hormone levels, leading to increased appetite and increased food intake [32]. In parallel, sleep deprivation, as a metabolic stressor, may activate the hypothalamic-pituitary-adrenal (HPA) axis and increase cortisol production, thereby increasing food intake and leading to visceral fat accumulation [33, 34]. In addition, people who sleep less may be more fatigued, which may reduce physical activity and increase sedentary time, thereby leading to obesity [33]. These potential mechanisms explain, to some extent, the inverse association between sleep duration and subsequent obesity. However, for the insignificant effect of BMI and WC on sleep duration, our findings are not consistent with several prior studies [10, 11]. This difference in the inverse association may be due to differences in sleep duration measurements, adjustment for confounders, and characteristics of the participants. Besides, For Han people, this study also found gender differences in the effect of sleep duration on BMI. Specifically, the negative relationship between baseline sleep duration and follow-up BMI was significant only in female, but not in male, which is consistent with previous findings [35, 36]. However, the exact biological mechanism for this gender difference is not clear. One possible explanation is that it is related to sexual hormones [37]. Previous studies have found that changes in sexual hormones not only lead to a decrease in sleep quality [38], but also to an increase in body weight [39]. Compared to males, females are more likely to suffer from changes in sexual hormones (such as menopause) and experience a decrease in sleep duration leading to obesity. For ethnic minorities, our findings showed that there was no relationship between sleep duration and obesity at all. This is consistent with a few studies of ethnic minorities in the United States failing to find a link between sleep duration and obesity [40]. For this non-significant association in ethnic minorities, we argued that the relatively long sleep duration of ethnic minorities may play an important role. Previous studies have shown that there is no significant relationship between sleep duration and obesity for people with a long sleep duration [4]. From this prospective, our results did show that ethnic minorities had a longer sleep duration than Han people, thus the significant association disappearing. As to the reason why the ethnic minorities having a longer sleep duration, the socioeconomic status (SES) may be one of the key factors. Compared to Han people, most ethnic minorities live in remote and poor areas and have lower SES [20]. Previous studies indicated that lower SES is associated with longer sleep duration [41]. For instance, some studies found that living in a rural area and having a lower level of educational attainment are protective factors for long sleep duration [42, 43]. Thus, among ethnic minorities with lower SES can have a longer sleep duration, attributing to a non-significant association between sleep duration and obesity. Nevertheless, due to the lack of relevant studies on the relationship between sleep duration and obesity among Chinese ethnic minorities, further studies with large samples are still needed to confirm our findings. There are several limitations to this study. First, participants self-reported sleep duration might not reflect the actual sleep duration. Objective measurements of sleep duration are worth considering in future studies. Moreover, information for other sleep-related characteristics, such as sleep quality, sleep patterns and sleep disorder, was missed due to the limitations of the dataset. The lack of considering these potential confounders may bring about a spurious link between sleep duration and obesity. In addition, the results of this study are based on data from Guizhou province, China. The* generalization of the findings to the whole Chinese or global population may be limited. ## Conclusion Our results suggested an ethnic difference in the relationship between sleep and obesity for Chinese populations. Given the backdrop of the obesity epidemic in China, future sleep-aimed overweight and obesity interventions should be conducted according to population characteristics. ## Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 ## References 1. Bluher M. Obesity: global epidemiology and pathogenesis. Nat Rev Endocrinol (2019.0) 15 288-98. DOI: 10.1038/s41574-019-0176-8 2. Liu J, He Z, Ma N, Chen ZY. Beneficial Effects of Dietary Polyphenols on High-Fat Diet-Induced obesity linking with modulation of gut microbiota. J Agric Food Chem (2020.0) 68 33-47. DOI: 10.1021/acs.jafc.9b06817 3. 3.Garfield V. The Association between Body Mass Index (BMI) and sleep duration: where are we after nearly two decades of Epidemiological Research? Int J Environ Res Public Health. 2019;16(22). 10.3390/ijerph16224327. 4. Zhou Q, Zhang M, Hu D. Dose-response association between sleep duration and obesity risk: a systematic review and meta-analysis of prospective cohort studies. Sleep Breath (2019.0) 23 1035-45. DOI: 10.1007/s11325-019-01824-4 5. Bacaro V, Ballesio A, Cerolini S, Vacca M, Poggiogalle E, Donini LM. Sleep duration and obesity in adulthood: an updated systematic review and meta-analysis. Obes Res Clin Pract (2020.0) 14 301-9. DOI: 10.1016/j.orcp.2020.03.004 6. Guimaraes KC, Silva CM, Latorraca COC, Oliveira RA, Crispim CA. Is self-reported short sleep duration associated with obesity? A systematic review and meta-analysis of cohort studies. Nutr Rev (2022.0) 80 983-1000. DOI: 10.1093/nutrit/nuab064 7. Cho KH, Cho EH, Hur J, Shin D. Association of Sleep duration and obesity according to gender and age in korean adults: results from the Korea National Health and Nutrition Examination Survey 2007–2015. J Korean Med Sci (2018.0) 33 e345. DOI: 10.3346/jkms.2018.33.e345 8. Jefferson T, Addison C, Sharma M, Payton M, Jenkins BC. Association between sleep and obesity in African Americans in the Jackson Heart Study. J Am Osteopath Assoc (2019.0) 119 656-66. DOI: 10.7556/jaoa.2019.113 9. Singh M, Drake CL, Roehrs T, Hudgel DW, Roth T. The association between obesity and short sleep duration: a population-based study. J Clin Sleep Med (2005.0) 1 357-63. DOI: 10.5664/jcsm.26361 10. Garfield V, Llewellyn CH, Steptoe A, Kumari M. Investigating the bidirectional Associations of Adiposity with Sleep Duration in older adults: the English Longitudinal Study of Ageing (ELSA). Sci Rep (2017.0) 7 40250. DOI: 10.1038/srep40250 11. Sokol RL, Grummon AH, Lytle LA. Sleep duration and body mass: direction of the associations from adolescence to young adulthood. Int J Obes (Lond) (2020.0) 44 852-6. DOI: 10.1038/s41366-019-0462-5 12. Koolhaas CM, Kocevska D, Te Lindert BHW, Erler NS, Franco OH, Luik AI. Objectively measured sleep and body mass index: a prospective bidirectional study in middle-aged and older adults. Sleep Med (2019.0) 57 43-50. DOI: 10.1016/j.sleep.2019.01.034 13. Pan XF, Wang L, Pan A. Epidemiology and determinants of obesity in China. Lancet Diabetes Endocrinol (2021.0) 9 373-92. DOI: 10.1016/S2213-8587(21)00045-0 14. Wang L, Zhou B, Zhao Z, Yang L, Zhang M, Jiang Y. Body-mass index and obesity in urban and rural China: findings from consecutive nationally representative surveys during 2004-18. Lancet (2021.0) 398 53-63. DOI: 10.1016/S0140-6736(21)00798-4 15. Jin Y, Luo Y, He P. Hypertension, socioeconomic status and depressive symptoms in chinese middle-aged and older adults: findings from the China health and retirement longitudinal study. J Affect Disord (2019.0) 252 237-44. DOI: 10.1016/j.jad.2019.04.002 16. Tong X, Wang X, Wang D, Chen D, Qi D, Zhang H. Prevalence and ethnic pattern of overweight and obesity among middle-aged and elderly adults in China. Eur J Prev Cardiol (2019.0) 26 1785-9. DOI: 10.1177/2047487319845129 17. Lu WH, Zhang WQ, Sun F, Gao YT, Zhao YJ, Liu JW. Correlation between occupational stress and Coronary Heart Disease in Northwestern China: a case study of Xinjiang. Biomed Res Int (2021.0) 2021 8127873. DOI: 10.1155/2021/8127873 18. Ning X, Lv J, Guo Y, Bian Z, Tan Y, Pei P. Association of Sleep Duration with Weight Gain and General and central obesity risk in chinese adults: a prospective study. Obes (Silver Spring) (2020.0) 28 468-74. DOI: 10.1002/oby.22713 19. Zhou Q, Wu X, Zhang D, Liu L, Wang J, Cheng R. Age and sex differences in the association between sleep duration and general and abdominal obesity at 6-year follow-up: the rural chinese cohort study. Sleep Med (2020.0) 69 71-7. DOI: 10.1016/j.sleep.2019.12.025 20. Wang YJ, Chen XP, Chen WJ, Zhang ZL, Zhou YP, Jia Z. Ethnicity and health inequalities: an empirical study based on the 2010 China survey of social change (CSSC) in Western China. BMC Public Health (2020.0) 20 637. DOI: 10.1186/s12889-020-08579-8 21. Huang CQ, Dong BR, Lu ZC, Yue JR, Liu QX. Chronic diseases and risk for depression in old age: a meta-analysis of published literature. Ageing Res Rev (2010.0) 9 131-41. DOI: 10.1016/j.arr.2009.05.005 22. 22.Chen Y, Wang Y, Xu K, Zhou J, Yu L, Wang N, et al. Adiposity and long-term Adiposity Change are Associated with Incident Diabetes: a prospective cohort study in Southwest China. Int J Environ Res Public Health. 2021;18(21). 10.3390/ijerph182111481. 23. 23.Zhang Y, Wang Y, Chen Y, Zhou J, Xu L, Xu K, et al. Associations of dietary patterns and risk of hypertension in Southwest China: a prospective cohort study. Int J Environ Res Public Health. 2021;18(23). 10.3390/ijerph182312378. 24. 24.Liu MJ, Li HT, Yu LX, Xu GS, Ge H, Wang LL, et al. A correlation study of DHA Dietary Intake and plasma, erythrocyte and breast milk DHA concentrations in Lactating Women from Coastland, Lakeland, and Inland Areas of China. Nutrients. 2016;8(5). 10.3390/nu8050312. 25. Bull FC, Maslin TS, Armstrong T. Global physical activity questionnaire (GPAQ): nine country reliability and validity study. J Phys Act Health (2009.0) 6 790-804. DOI: 10.1123/jpah.6.6.790 26. Hamrik Z, Sigmundova D, Kalman M, Pavelka J, Sigmund E. Physical activity and sedentary behaviour in Czech adults: results from the GPAQ study. Eur J Sport Sci (2014.0) 14 193-8. DOI: 10.1080/17461391.2013.822565 27. Kivimaki M, Feldt T, Vahtera J, Nurmi JE. Sense of coherence and health: evidence from two cross-lagged longitudinal samples. Soc Sci Med (2000.0) 50 583-97. DOI: 10.1016/s0277-9536(99)00326-3 28. Zhang T, Zhang H, Li Y, Sun D, Li S, Fernandez C. Temporal relationship between Childhood Body Mass Index and insulin and its impact on adult hypertension: the Bogalusa Heart Study. Hypertension (2016.0) 68 818-23. DOI: 10.1161/HYPERTENSIONAHA.116.07991 29. Han T, Lan L, Qu R, Xu Q, Jiang R, Na L. Temporal relationship between hyperuricemia and insulin resistance and its impact on future risk of hypertension. Hypertension (2017.0) 70 703-11. DOI: 10.1161/HYPERTENSIONAHA.117.09508 30. Rosseel Y. lavaan: an R package for structural equation modeling. J Stat Softw (2012.0) 48 1-36. DOI: 10.18637/jss.v048.i02 31. Joreskog KG. Modeling development: using covariance structure models in longitudinal research. Eur Child Adolesc Psychiatry (1996.0) 5 8-10. DOI: 10.1007/bf00538536 32. Taheri S, Lin L, Austin D, Young T, Mignot E. Short sleep duration is associated with reduced leptin, elevated ghrelin, and increased body mass index. PLoS Med (2004.0) 1 e62. DOI: 10.1371/journal.pmed.0010062 33. Kyrou I, Tsigos C. Stress hormones: physiological stress and regulation of metabolism. Curr Opin Pharmacol (2009.0) 9 787-93. DOI: 10.1016/j.coph.2009.08.007 34. 34.Magee CA, Huang XF, Iverson DC, Caputi P. Examining the pathways linking chronic sleep restriction to obesity. J Obes 2010, 2010. 10.1155/2010/821710. 35. Cournot M, Ruidavets JB, Marquie JC, Esquirol Y, Baracat B, Ferrieres J. Environmental factors associated with body mass index in a population of Southern France. Eur J Cardiovasc Prev Rehabil (2004.0) 11 291-7. DOI: 10.1097/01.hjr.0000129738.22970.62 36. Ogilvie RP, Bazzano LA, Gustat J, Harville EW, Chen W, Patel SR. Sex and race differences in the association between sleep duration and adiposity: the Bogalusa Heart Study. Sleep Health (2019.0) 5 84-90. DOI: 10.1016/j.sleh.2018.10.010 37. Meyer KA, Wall MM, Larson NI, Laska MN, Neumark-Sztainer D. Sleep duration and BMI in a sample of young adults. Obes (Silver Spring) (2012.0) 20 1279-87. DOI: 10.1038/oby.2011.381 38. Sowers MF, Zheng H, Kravitz HM, Matthews K, Bromberger JT, Gold EB. Sex steroid hormone profiles are related to sleep measures from polysomnography and the Pittsburgh Sleep Quality Index. Sleep (2008.0) 31 1339-49. PMID: 18853931 39. Lovejoy JC. The influence of sex hormones on obesity across the female life span. J Womens Health (1998.0) 7 1247-56. DOI: 10.1089/jwh.1998.7.1247 40. Sun X, Gustat J, Bertisch SM, Redline S, Bazzano L. The association between sleep chronotype and obesity among black and white participants of the Bogalusa Heart Study. Chronobiol Int (2020.0) 37 123-34. DOI: 10.1080/07420528.2019.1689398 41. Patel SR, Malhotra A, Gottlieb DJ, White DP, Hu FB. Correlates of long sleep duration. Sleep (2006.0) 29 881-9. DOI: 10.1093/sleep/29.7.881 42. Ren Y, Liu Y, Meng T, Liu W, Qiao Y, Gu Y. Social-biological influences on sleep duration among adult residents of northeastern China. Health Qual Life Outcomes (2019.0) 17 47. DOI: 10.1186/s12955-019-1111-3 43. Wang S, Li B, Wu Y, Ungvari GS, Ng CH, Fu Y. Relationship of Sleep Duration with Sociodemographic characteristics, Lifestyle, Mental Health, and chronic Diseases in a large Chinese Adult Population. J Clin Sleep Med (2017.0) 13 377-84. DOI: 10.5664/jcsm.6484
--- title: 'Provider implicit and explicit bias in person-centered maternity care: a cross-sectional study with maternity providers in Northern Ghana' authors: - Patience A. Afulani - Jaffer Okiring - Raymond A. Aborigo - Jerry John Nutor - Irene Kuwolamo - John Baptist K. Dorzie - Sierra Semko - Jason A. Okonofua - Wendy Berry Mendes journal: BMC Health Services Research year: 2023 pmcid: PMC10015736 doi: 10.1186/s12913-023-09261-6 license: CC BY 4.0 --- # Provider implicit and explicit bias in person-centered maternity care: a cross-sectional study with maternity providers in Northern Ghana ## Abstract ### Background Person-centered maternity care (PCMC) has become a priority in the global health discourse on quality of care due to the high prevalence of disrespectful and lack of responsive care during facility-based childbirth. Although PCMC is generally sub-optimal, there are significant disparities. On average, women of low socioeconomic status (SES) tend to receive poorer PCMC than women of higher SES. Yet few studies have explored factors underlying these inequities. In this study, we examined provider implicit and explicit biases that could lead to inequitable PCMC based on SES. ### Methods Data are from a cross-sectional survey with 150 providers recruited from 19 health facilities in the Upper East region of Ghana from October 2020 to January 2021. Explicit SES bias was assessed using situationally-specific vignettes (low SES and high SES characteristics) on providers’ perceptions of women’s expectations, attitudes, and behaviors. Implicit SES bias was assessed using an Implicit Association Test (IAT) that measures associations between women’s SES characteristics and providers’ perceptions of women as ‘difficult’ or ‘good’. Analysis included descriptive statistics, mixed-model ANOVA, and bivariate and multivariate linear regression. ### Results The average explicit bias score was 18.1 out of 28 (SD = 3.60) for the low SES woman vignette and 16.9 out of 28 (SD = 3.15) for the high SES woman vignette ($p \leq 0.001$), suggesting stronger negative explicit bias towards the lower SES woman. These biases manifested in higher agreement to statements such as the low SES woman in the vignette is not likely to expect providers to introduce themselves and is not likely to understand explanations. The average IAT score was 0.71 (SD = 0.43), indicating a significant bias in associating positive characteristics with high SES women and negative characteristics with low SES women. Providers with higher education had significantly lower explicit bias scores on the low SES vignette than those with less education. Providers in private facilities had higher IAT scores than those in government hospitals. ### Conclusions The findings provide evidence of both implicit and explicit SES bias among maternity providers. These biases need to be addressed in interventions to achieve equity in PCMC and to improve PCMC for all women. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12913-023-09261-6. ## Background Person-centered maternity care (PCMC) refers to care during childbirth that is respectful and responsive to women’s preferences, needs, and values [1, 2]. PCMC emphasizes the continuum of people’s experience of care during childbirth including responsive and supportive care, dignified and respectful care, effective communication, and respect for people’s autonomy. It is a measure of respect for people’s human rights as well as a measure of the quality of care [1, 3–5]. PCMC is a more inclusive term that captures other terminologies used in maternal health such as respectful maternity care and compassionate care. Other terms such as mistreatment, disrespect and abuse, and dehumanized care represent poor PCMC. Several studies globally have documented a high prevalence of disrespectful, abusive, and neglectful treatment of women during facility-based childbirth [6–9]. Such poor PCMC leads to lack of, delayed, inadequate, unnecessary, or harmful care [10]. Mistreatment deters women from giving birth in health facilities; the experience of poor PCMC, even by a few women, leads to negative community perceptions of quality of care, which discourages other women from giving birth in health facilities [11–14]. On the other hand, positive healthcare experiences can improve health outcomes through pathways such as patient engagement, safety, trust, higher patient and provider satisfaction, and improved psychosocial health,[15, 16]. Further, core components of PCMC such as birth companionship is associated with improved birth outcomes such as shorter duration of labor, decreased caesarean and instrumental vaginal birth, and higher five-minute Apgar scores, which decreases need for neonatal resuscitation [17, 18]. Recent studies have also linked PCMC to improved postpartum outcomes such as lower risk of reporting maternal complications, screening positive for post-partum depression, and reporting newborn complications [19, 20]. Poor PCMC therefore undermines health gains for mothers and babies [10]. Studies in sub-Saharan Africa (SSA) have shown significant gaps in PCMC as evidenced by disrespect and abuse, poor communication, lack of respect for women’s autonomy, and lack of supportive care during prenatal care and childbirth [8, 12, 21, 22]. One study in Ghana, Guinea, Myanmar, and Nigeria found that $42\%$ ($$n = 2016$$) of women observed were physically or verbally abused, and $35\%$ ($$n = 2672$$) of women interviewed reported mistreatment such as stigma and discrimination during childbirth in healthcare facilities [9]. Another study using a continuous measure of PCMC with scores ranging from 0–100 (higher scores indicative of higher PCMC) found average PCMC scores below 70 among women surveyed in Ghana, Kenya, and India [8]. The lowest scores were in the communication and autonomy domains, with $60\%$ of women in the Ghana sample reporting providers never explained the purpose of examinations or procedures and $44\%$ reporting providers never asked for their consent before exams and procedures. Over $80\%$ of women in the Ghana sample also reported providers never introduced themselves before attending to them. Prior research also shows disparities in PCMC, especially by socioeconomic status (SES). Across several quantitative studies, women of low SES (measured variously by wealth quintiles, education, literacy, age, and empowerment measures) tend to report poorer PCMC than women of higher SES [8, 9, 21, 23–25]. These disparities are also highlighted in qualitative studies where women reported being discriminated against based on their status, which influences their decision-making on where to give birth [7, 12, 26–28]. We hypothesized that provider biases may be reinforcing these patterns of abuse against low SES women [29, 30]. Bias can be explicit/conscious or implicit/unconscious [31]. Explicit bias refers to conscious attitudes, beliefs, and perceptions about a group, often manifesting as discrimination—the act of treating people differently according to their perceived group [32]. Implicit bias on the other hand operates at an unintentional level and does not require a person to endorse or devote attention to its expression. Instead, it can be activated quickly and unknowingly by situational cues such as a person’s skin color, accent, clothes, or other outwardly appearance [31, 33]. Implicit bias is prevalent in every society, although the content (e.g., stereotypic associations) of biases may differ across contexts [34, 35]. While people may have similar explicit and implicit stereotypes, there is often little correlation between measures of explicit and implicit bias as reporting on explicit bias is prone to social desirability bias [36–39]. Most studies on implicit and explicit bias in health care settings have been conducted in the United States (US), with racial bias commonly studied. For example, several studies have documented anti-Black bias among physicians contributing to lower likelihood of evidence-based prescribing and lower quality interpersonal care for Black compared to White patients [40–42]. SES bias is, however, likely a key bias in context where the predominant bias is not racial. Within healthcare in the US, SES bias is prevalent and influential [43]. Research with physicians supports this claim, finding that low-SES patients are perceived to be less intelligent, less compliant, and less interested in promoting their own health relative to higher-SES patients [44, 45]. Such differential attitudes by patient SES translate into physician behavior with low-income patients receiving shorter consultations and fewer medical tests than patients with higher income [46]. From the patient perspective, SES bias is felt. Patients report that quality of treatment provided, access to care, and patient-provider interactions is affected by their status, offering evidence of the role of implicit bias [47]. Patients also report experiencing discrimination due to their status, suggesting the persistence of explicit or implicit bias in healthcare [48]. SES bias experienced in interactions with healthcare professionals contributes to distrust and lack of treatment adherence [49], thereby contributing to poorer health over time [50]. Though provider bias based on patient’s SES has been less thoroughly studied in other settings [51], disparities in person-centered care in the healthcare system based on patient SES have been observed worldwide. For example, a study of several countries in Europe found that compared to low SES patients (measured by patient education), high SES patients in Spain, Italy, and France experienced shorter waiting time for specialist consultations [52]. In low and middle-income settings, several qualitative studies have documented preferential treatment of higher SES patients compared to low SES patients, suggesting a role of provider bias [12, 53–56] Studies on bias in PCMC in Africa are limited. To our knowledge only one prior study empirically examined the role of both provider implicit and explicit bias on PCMC in SSA. This work in Kenya provided initial empirical evidence for the role of both implicit and explicit bias in PCMC disparities by socioeconomic status in SSA [21, 39, 55, 57]. This manifested as providers’ reactions to women's appearances, assumptions about who is more likely to understand or be cooperative, and perceptions of women's expectations and attitudes. These factors, including women's ability to advocate for themselves or hold providers accountable interact to produce PCMC disparities [39]. There is also limited research on factors that may be associated with provider bias. Prior research however suggests that while implicit bias may be similar among providers in the same contexts, some provider factors such as education may be associated with explicit bias [36, 39]. In this study, we sought to extend the evidence base for the role of both explicit and implicit bias in PCMC disparities by socioeconomic status using data from another setting in SSA. This study included a larger sample of maternity providers in Ghana, which provides more statistical power to assess various associations. The primary aim of the study was to assess the extent of provider implicit and explicit SES bias that may contribute to disparities in PCMC in Ghana. A secondary aim was to identify provider and facility-level factors associated with these implicit and explicit SES biases. ## Design, participants, setting, and data collection The data are from a cross-sectional study with healthcare providers who work in maternity units in the Upper East region (UER) of Ghana. The setting and data collection procedures have been previously described [58] and are briefly summarized here. The UER is one of poorest regions in Ghana. The literacy rate (proportion of people from age 6 who can read and write) for the region is about $48\%$ compared to the national average of about $70\%$. About $37\%$ of the population aged 3 and older have never been to school, and of those over 15 years who have some schooling, less than $20\%$ have more than a secondary school education [59]. The region is divided into 15 administrative municipalities/districts, of which 10 have district hospitals. The doctor-patient ratio for the region is about 1:27,652 and the nurse-patient ratio is about 1:500 [60]. We recruited a total of 150 Providers from the 19 highest volume delivery health facilities (most with an average of 75 births per month or more in the prior year) across the 15 districts in the region from October 2020 to January 2021. There are about 94 high volume delivery facilities across the region (hospitals and health centers that conduct at least 100 deliveries per year). All providers who worked in maternity units in the selected facilities for a minimum of six months at the time of the survey, inclusive of doctors, medical assistants, midwives, nurses, and support staff, were eligible to participate. Two trained research officers (one male and one female) conducted the interviews at private locations at the health facilities or elsewhere based on the provider preference. With approval from the Regional Director of Health Services and permission from leadership of the various health facilities, providers designated to the maternity unit who were available at the time of the visit were invited to participate in the study. The response rate was $80\%$. Ethics approval was obtained from the Navrongo Health Research Center and the University of California, San Francisco Institutional Review Boards, with additional approval from the UER Director of Health Services. All participants provided written informed consent following receipt of information about the study. All the interviews were conducted in English using a structured questionnaire in the REDCap mobile application [61], and lasted about one hour. The questionnaire included several questions related to explicit bias and provider and facility characteristics. Following the interview, each respondent took a computer-based implicit bias test described below. ## Measures The measures of implicit and explicit bias used have been previously described [39] and briefly summarized below: Explicit bias was assessed using providers’ perceptions of women’s PCMC expectations and behaviors based on SES, preference for low and high SES women, and a feeling of connection to low and high SES women. Two vignettes (Table 1) were read in counter-balanced order to each provider, followed by ten questions. The first eight questions assessed providers’ perceptions of the woman in the vignette’s expectations for introductions, consenting, and companionship; potential to cooperate, understand explanations, exaggerate pain, and to litigate; as well as provider behavior needed to convey seriousness and gain cooperation. Response options ranged from strongly disagree to strongly agree on a 4-point scale (Table 2). Two final questions asked providers to what extent they would want to be a provider for the woman in the vignette and how connected they felt to her on a scale of 1 to 10. All participants responded to these questions about both the low-SES and high-SES woman. The development of this measure was informed by measurement of explicit bias in prior literature [37, 40] and prior research in Kenya [55]. It was piloted with five providers in Ghana prior to the study. Table 1Vignettes to assess explicit biasScenario 1: Woman with markers of low SES: A 30-year-old poor farmer from one of the villages in the county is admitted to the ward. She dropped out of school in primary two and cannot read or write. She is not covered by insurance and attended ANC only once. She looks very unkempt and did not bring anything with her to be used for the delivery. She presented in labor with her mother-in-law and is complaining of severe abdominal pain. Thinking about this patient: How strongly do you agree/disagree with these statements?Scenario 2: Woman with markers of high SES: A 30-year-old woman who is the wife of a doctor in the hospital is admitted to your ward. She also works at the local bank and is covered by private health insurance. She received ANC 6 times during her pregnancy. She is very well dressed and has come with all the required items for her labor. She presented in labor with her mother-in-law and is complaining of severe abdominal pain. Thinking about this patient: How strongly do you agree/disagree with these statements?Statements 1. She is not likely to expect providers to introduce themselves to her 2. She is not likely to understand any explanations 3. Since she has come to the facility, it means she has consented to all examinations and treatments 4. She is likely exaggerating her pain 5. She will not need a companion to stay with her 6. The provider needs to be stern for her to understand the seriousness of the situation 7. She is likely going to be uncooperative when it is time to push and need to be physically restrained 8. She is likely to sue you if something goes wrong 9. On a scale of 1 to 10, where 1 represents lack of connection or warm feelings towards the patient and 10 represents strong connection or strong feelings of warmth towards the patient how connected or warm are you likely to feel towards this patient? 10. To what extent do think you will want to be a provider for patients like her?Table 2Distribution of responses to the individual questions in the vignettesStatement ResponseVignette: N (%)Low SES ($$n = 148$$)High SES ($$n = 148$$)p-valueNot likely to expect providers to introduce themselves to herStrongly disagree21 (14.2)25 (16.9) < 0.001Disagree58 (39.2)92 (62.2)Agree57 (38.5)21 (14.2)Strongly agree12 (8.1)10 (6.7)Not likely to understand any explanationsStrongly disagree31 (21.0)48 (32.4) < 0.001Disagree59 (39.9)70 (47.3)Agree35 (23.7)25 (16.9)Strongly agree23 (15.5)5 (3.4)Has come to the facility, it means she has consented to all examinationsStrongly disagree21 (14.2)17 (11.5)0.772Disagree57 (38.5)56 (37.8)Agree44 (29.7)44 (29.7)Strongly agree26 (17.6)31 (21.0)Likely to exaggerate her painStrongly disagree31 (21.0)24 (16.2)0.004Disagree83 (56.1)62 (41.9)Agree23 (15.5)49 (33.1)Strongly agree11 (7.4)13 (8.8)Will not need a companion to stay with herStrongly disagree43 (29.1)46 (31.1)0.251Disagree94 (63.5)96 (64.9)Agree7 (4.7)6 (4.0)Strongly agree4 (2.7)0 (0.0)Provider needs to be stern for her to understand the seriousness of the situationStrongly disagree24 (16.2)21 (14.2)0.818Disagree57 (38.5)60 (40.5)Agree51 (34.5)47 (31.8)Strongly agree16 (10.8)20 (13.5)Likely going to be uncooperative when it is time to push and need to be restrainedStrongly disagree27 (18.2)25 (16.9)0.277Disagree84 (56.8)72 (48.6)Agree22 (14.9)34 (23.0)Strongly agree15 (10.1)17 (11.5)Likely to sue you if something goes wrongStrongly disagree14 (9.5)1 (0.7) < 0.001Disagree50 (33.8)4 (2.7)Agree58 (39.2)55 (37.2)Strongly agree26 (17.6)88 (59.5)Would like to be a provider for this patientNot at all3 (2.0)2 (1.4)0.046A little27 (18.2)47 (31.8)Very much118 (79.7)98 (66.2)1 (0.7Feeling connected to patient on a scale of 0 to 10. $$n = 149$$Mean (SD)7.7 (1.9)7.5 (1.7)0.246 Implicit bias was measured using an Implicit Association Test (IAT) implemented in Inquisit Lab version 5 [62], which was first developed in a study in Kenya [39]. The IAT is a cognitive-behavioral test that measures the strength of automatic associations between concepts in people’s minds based on a sorting task [63]. It has been shown to be a valid and reliable way of measuring implicit bias based on various factors such as race, gender, SES, religion, etc., [ 36, 64]. The IAT used for this study assessed associations between women’s SES characteristics and providers’ perceptions of women as ‘difficult’ or ‘good’ [55, 65]. Attributes of ‘good’ patients used were likable, cooperative, respectful, intelligent, and responsible; whereas attributes of ‘difficult’ patients were irresponsible, uncooperative, rude, annoying, and stupid. High SES descriptors were wealthy, well-educated, well-dressed, and a banker; low SES descriptors included poor, uneducated, old/torn clothes, and a cleaner. An individual’s IAT score represents the difference in the average length of time they took to sort words during various sections of the test. It is assumed that people will more quickly sort words they associate together than those they do not. IAT scores vary between -2 and + 2. In this study, a positive score indicates a stronger association between high status and good patient and between low status and difficult patient. Increasing positive scores can thus be interpreted as stronger implicit bias in favor of high SES patients. A negative score indicates a stronger association between high status and difficult patient and low status and good patient—implying implicit bias in favor of low SES patients. The IAT has been used in prior studies to assess implicit bias in healthcare settings [40, 42]. ## Statistical analysis Initial analysis included descriptive statistics to characterize the sample and measures and factor analysis to assess the psychometric properties of the composite measures. We then generated explicit bias scores by summing responses to the questions for each vignette. We used mixed-model ANOVA to assess if responses based on the two SES vignettes differed. For implicit bias, we used dependent samples t-test to test whether the average IAT score differed significantly from zero—zero indicating no bias. We then analyzed the associations between bias measures and provider and facility characteristics using cross-tabulations and bivariate linear regressions with robust standard errors; and multivariate associations using multilevel linear regressions. In model building, we included all variables with p-values of 0.2 and below from the bivariate analysis to minimize negative confounding, and those with known relationship with the outcome of interest. We then systematically removed non-significant variables from the model until the best fit was attained using the Akaike information criteria. We used STATA version 14.1 for all analysis (College Station, TX). ## Demographics Of the 150 providers who participated in the interviews, most were female ($97.3\%$), married ($74.0\%$), between 30 and 52 years of age ($71.3\%$), and were nurses or midwives ($97\%$). No doctor participated. About two thirds worked in government hospitals ($61.4\%$), with $21.3\%$ working in government health centers, and $17.3\%$ in Mission/private facilities. Close to half ($46.0\%$) had been in their positions for five years or less (Table 3).Table 3Participant characteristicsCharacteristicCategorySurvey ($$n = 150$$)No. (%) Facility typeGovt hospital92 (61.4)Govt health center/Dispensary32 (21.3)Mission/private26 (17.3)PositionNurse/Midwife145 (96.7)Support5 (3.3)GenderMale4 (2.7)Female146 (97.3)Age23–29 years43 (28.7)30–39 years84 (56.0)40–52 years23 (15.3)Marital statusMarried111 (74.0)Single39 (26.0)Number of childrenNo children35 (23.3)1 to 2 children84 (56.0)3 or more children31 (20.7)Educational levelTraining college and below128 (85.3)University and above22 (14.7)Monthly salaryLess than 2000 GHS113 (75.3)2000–3000 GHS37 (24.7)Years as provider0–5 years69 (46.0)6–10 years43 (28.7)More than 10 years38 (25.3)Perceived social status of family growing upBottom half104 (69.3)Upper half46 (30.7)Perceived social status of self nowBottom half59 (39.3)Upper half91 (60.7)Social mobilityUpward mobility102 (68.0)No change35 (23.3)Downward mobility13 (8.7)ReligionCatholic121 (80.7)Methodist/Presby/Anglican29 (19.3)Training on interpersonal interactionsNo60 (40.0)Yes90 (60.0) All providers completed the questions from the two vignettes, but two respondents were excluded because of incomplete data ($$n = 148$$). The questions for which there were significant differences by SES were introductions, understanding, exaggerating pain, and litigation (Table 2). Close to half ($47\%$) of providers agreed (agree or strongly agree) that the low SES woman was not likely to expect providers to introduce themselves compared to $21\%$ for the high SES woman; and $39\%$ agreed that the low SES woman was not likely to understand explanations compared to $21\%$ for the high SES woman. On the other hand, providers were more likely to agree that the high SES woman was likely exaggerating her pain ($23\%$ for low SES and $42\%$ for high SES) and was more likely to sue them if something goes wrong compared to the low SES woman ($57\%$ for low SES and $97\%$ for high SES). For the other items, there were no statistically significant differences by SES, although the direction of association and magnitude of some of the differences are worth noting. For example, in both vignettes, close to half of providers agreed that since the woman came to the facility, it means she has consented to all examinations and treatment ($47\%$ for low SES and $51\%$ for high SES). Also, close to half agreed that the provider needs to be stern for the woman to understand the seriousness of the situation ($45\%$ for low SES and $46\%$ for high SES), and about one-third agreed that the woman was likely to be uncooperative when it was time to push and would need to be physically restrained ($25\%$ for low SES and $34\%$ for high SES). Very few providers agreed that the woman would not need a companion ($7\%$ for low SES and $5\%$ for high SES). More providers stated they would very much want to be a provider for the woman with the lower SES ($78\%$) than the one with the higher SES ($66\%$), but there were no differences in the extent to which they felt connected with the two patients (average feelings of connectedness of about 8 out of 10 for both: Table 2). Exploratory factor analysis of the eight PCMC perceptions yielded one factor with eigenvalue > 1 for both vignettes (Table S1). The question on litigation had low loadings on the first factor for both vignettes and was dropped. For the response to the low SES vignette, all other items had factor loadings of > 0.3 on the first factor. Three items (introductions, consent, and companion) had loadings between 0.19 and 0.27 on the high SES vignette but were retained based on their conceptual relevance. Cronbach’s alpha for the seven items was 0.65 for the low SES Vignette and 0.61 for the high SES vignette. The hypothetical range of scores on the composite measure from the seven items is from 7 to 28, with higher scores indicating stronger explicit bias. Hypothetically a score of 7 represents no explicit bias and 28 represents the strongest explicit bias. The average explicit bias score was 18.1 (SD = 3.60; range 9–28) for the low SES woman vignette and 16.9 (SD = 3.15; range 8–27) for the high SES woman vignette. Scores did not differ significantly by order of vignette presentation. Mixed-model ANOVA showed a significant difference between the two composite scores ($p \leq 0.001$), suggesting a significant difference in associating negative perceptions towards the lower SES woman than the higher SES woman. This implies that on average the providers in the sample have stronger negative explicit bias towards the low SES patient than the high SES patient. ## Implicit SES bias All providers ($$n = 150$$) took the IAT. IAT scores ranged from -0.47 to 1.43, with a mean of 0.71 (SD = 0.43; $95\%$CI = 0.64 to 0.78). Most providers ($90.7\%$) had an IAT score greater than zero. Thus, on average providers in this sample had stronger implicit bias in favor of high SES patients—i.e., bias towards of associating positive characteristics with high SES women and negative characteristics with low SES women. ## Factors associated with explicit and implicit bias There was a strong correlation between the explicit bias scores from the two vignettes ($r = 0.60$, $p \leq 0.001$), but as expected, little correlation between the explicit bias scores and the IAT score ($p \leq 0.6$) (Table 4). The means bias scores by provider and facility characteristics are presented in Table 4. In the bivariate analysis, only educational status and parity were significantly associated explicit bias. On average providers with lower education had higher scores on the low SES vignette than those with higher education—indicating stronger negative explicit bias towards the low SES woman among providers with lower education than among those with higher education. Providers with higher parity had, on average, stronger negative explicit bias towards the high SES woman than providers with lower parity. For the IAT scores, only facility type was significant in bivariate analysis. On average providers working in government hospitals had lower IAT scores—indicating weaker implicit bias in favor of high SES patients among these providers than those working in government health centers and private facilities (Table 4).Table 4Bivariate analysis—correlation between provider bias measures and cross tabulations of mean bias scores by provider and facility characteristicsCharacteristicCategoryScore on low SES woman vignette, ($$n = 148$$)Score on high SES woman vignette, ($$n = 148$$)IAT score ($$n = 150$$)CorrelationsNrP-valueNrP-valueNrP-valueScore on low SES woman vignette1460.60 < 0.0011480.010.946Score on high SES woman vignette1460.60 < 0.001148-0.030.719IAT score1500.010.946148-0.030.719CrosstabulationsNMeanSDP-valueNMeanSDP-valueNMeanSDP-valueTotal14818.073.6014816.883.151500.710.43Facility typeGovt hospital9117.83.370.5119116.92.780.287920.640.430.041Govt health center3218.53.923217.43.95320.830.37Mission/private2518.64.032516.13.26260.800.44PositionNurse12917.93.690.14613116.83.240.1891310.700.430.265Midwife1118.12.551016.91.97110.900.35Ward aid/assistant/support820.52.62719.02.2480.620.44GenderMale4181.150.967416.82.500.93541.050.550.108Female14418.13.6514416.93.181460.700.42Age23–29 years4318.33.240.2234316.72.710.066430.670.450.61230–39 years8317.73.738316.63.45840.710.4240–52 years2219.13.682218.32.40230.780.40Marital statusMarried10918.13.730.80111017.03.350.3611110.680.440.110Single3917.93.263816.52.50390.800.37Number of childrenNo children3517.93.470.2413416.02.530.015350.790.430.3181 to 2 children8317.83.678416.83.24840.660.433 or more children3019.13.513018.23.18310.740.41Educational levelTraining College and below12618.43.370.00412617.13.070.0751280.700.440.705University and above2216.04.272215.83.46220.740.39Monthly salaryLess than 2000 GHS11118.23.570.57311116.83.170.7421130.700.440.0532000–3000 GHS3717.83.743717.03.14370.830.38Years as provider0–5 years6818.43.220.6676816.83.030.443690.690.430.2326–10 years4217.73.284317.32.79430.650.43More than 10 years3817.94.543716.53.74380.810.41Perceived social status of family growing upBottom half10217.93.570.38810216.93.290.8051040.720.400.672Upper half4618.53.694616.82.85460.690.48Perceived social status of self nowBottom half5717.83.770.4485816.63.230.340590.670.440.387Upper half9118.33.509017.13.10910.730.42Social mobilityUpward mobility10117.93.510.15410016.93.170.8081020.730.430.126No change3419.13.573517.13.21350.720.36Downward mobility1317.24.161316.43.04130.480.57ReligionCatholic12018.03.670.49012016.93.670.8731210.710.420.899Methodist/Presby/Anglican2818.53.322817.02.50290.700.47Training on interpersonal interactionsNo6018.53.530.2195917.22.710.335600.710.440.924Yes8817.83.648916.73.42900.710.42Higher mean scores on the low SES vignette indicate stronger negative explicit bias towards the low SES woman, while higher mean scores on the high SES vignette indicate stronger negative explicit bias towards the high SES woman. Higher IAT scores indicate stronger bias in favor of high SES patients In multivariate analysis (Table 5), education was again significantly associated with low SES bias scores, with those with higher education having a lower score on the low SES vignette than those with lower education—indicating providers with higher education had weaker negative explicit bias towards the low SES woman. Also, providers who reported no change in social mobility had stronger negative explicit bias towards the low SES woman than those who had experienced upward mobility. On the high SES vignette, providers in private facilities had lower bias scores than those in government hospitals, indicating providers in private hospitals had weaker negative explicit bias towards the high SES woman than those in government hospitals. Providers with more than 10 years of experience also had weaker explicit bias toward the high SES woman than those with fewer years of experience. Older age and higher parity were associated with stronger negative explicit bias towards the high SES woman. For implicit bias, providers in private facilities and government health centers had higher IAT scores than those in government hospitals, indicating providers in private facilities and lower level government facilities have stronger implicit bias in favor of high SES women—i.e., associating positive characteristics with high SES women and negative characteristics with low SES women. Also, providers with higher income had stronger implicit bias in favor of high SES women than those with lower income. Compared those who had achieved upward mobility, those who reported downward change in social mobility had stronger implicit bias in favor of low SES women. Table 5Multivariate analysis of factors associated with provider explicit and implicit biasCharacteristicCategoryScore on low SES woman vignette, ($$n = 148$$)Score on high SES woman vignette, ($$n = 148$$)IAT score, ($$n = 150$$)Coeff ($95\%$ CI)p-valueCoeff ($95\%$ CI)p-valueCoeff ($95\%$ CI)p-valueFacility typeGovt hospitalReference-Reference-Govt health center/Dispensary0.90 (-0.41,2.21)0.1800.18 (0.04,0.32)0.013Mission/private-0.89 (-1.64, -0.15)0.0190.18 (0.07,0.29)0.002Age23–29 yearsReference-30–39 years0.10 (-1.40,1.59)0.90040–52 years2.53 (0.51,4.54)0.014Marital statusMarriedReference-Single0.15 (-0.02,0.32)0.091Number of childrenNo childrenReference-1 to 2 children0.92 (-0.49,2.32)0.2003 or more children2.40 (0.55,4.24)0.011Educational levelTraining college and belowReference-University and above-2.17 (-3.74, -0.60)0.007Monthly salaryLess than 2000 GHSReference-2000–3000 GHS0.12 (0.01,0.23)0.041Years as provider0–5 yearsReference-6–10 years-0.35 (-2.07,1.38)0.694More than 10 years-2.87 (-4.87, -0.88)0.005Social mobilityUpward mobilityReference-Reference-No change1.11 (0.04,2.19)0.043-0.04 (-0.17, 0.08)0.493Downward mobility-0.51 (-1.99,0.97)0.500-0.24 (-0.43, -0.05)0.013Training on interpersonal interactionsNoReference-Yes-0.57 (-1.41,0.28)0.189Higher mean scores on the low SES vignette indicate stronger negative explicit bias towards the low SES woman, while higher mean scores on the high SES vignette indicate stronger negative explicit bias towards the high SES woman. Higher IAT scores indicate stronger implicit bias in favor of high SES patients ## Discussion This study provides further evidence of explicit and implicit biases among maternity providers that could lead to disparities in PCMC based on SES. When presented with vignettes representing a woman of low SES and one of high SES, overall, providers had more negative perceptions about the low SES woman manifested in their higher agreement to statements such as the low SES woman is not likely to expect providers to introduce themselves and not likely to understand explanations when compared to responses about the high SES woman. On the other hand, they were more likely to agree that the high SES woman was likely exaggerating her pain and was more likely to sue them if something went wrong, compared to responses for the low SES woman. Further, providers overall, showed implicit bias between women’s SES and their perceptions of those women as good or difficult patients. Specifically, providers were more likely to associate higher SES characteristics with good attributes and lower SES with negative attributes than the reverse. Such perceptions likely contribute to the poorer PCMC experiences among women of lower SES. To our knowledge, this is the second study to examine provider implicit biases in PCMC in SSA and the first in Ghana. The current study and the previous one in Kenya provide consistent evidence on the role of implicit bias in a context where the predominant bias is not racial bias. As has been previously noted, bias (implicit and explicit) is prevalent in every society, although the content of biases may differ across different contexts [35]. For instance, in contexts like the US, racial bias is a key contributor to health disparities. SES bias likely plays a greater role in other settings where there is less racial diversity (as socially constructed), and social class is an important determinant of how people are treated in societies. The significant bias in favor of associating positive characteristics with high SES women and negative characteristics with low SES women likely influences how providers interact with each group. Prior research in Kenya, showed that providers sometimes unconsciously treated higher SES women better based on their attractions to their physical appearance, although they did not necessarily prefer higher SES women as patients [39]. This is consistent with the findings here, where there were no differences in the extent to which they reported feeling connected with the two patients in the vignettes, despite more negative implicit bias towards the lower SES woman. This study provides stronger evidence on the role of explicit SES bias in PCMC. Unlike the Kenya study, where we did not find statistically significant overall differences in the composite explicit bias scores for the two vignettes, we did observe statistically significant differences supporting more negative biases towards low SES women in this study. This is likely because of the larger sample size and bigger differences in the magnitude of the associations for the individual items. In both studies, providers were more likely to agree that the low SES woman is not likely to expect providers to introduce themselves and is not likely to understand explanations, compared to the high SES woman. These likely explain SES differences in PCMC. If providers perceive less expectation of self-introduction and lower capacity for understanding explanations in low-SES patients, less information-giving and relationship-building is likely to follow. This has been previously documented in studies with both patients and providers in Kenya [12, 39, 57]. In other analysis, only $21\%$ of providers in this sample reported always introducing themselves and $49\%$ reported always explaining the purpose of examinations and procedures to their patients, which is informative given they serve predominantly low SES women [58]. Similarly, in both studies, providers were more likely to agree that the high SES woman is likely exaggerating her pain and is more likely to sue them if something goes wrong compared to the low SES woman. Such perceptions may lead to pain medication being withheld from high SES women who genuinely need it. Higher SES women may however still obtain adequate pain medication and experience more positive person-centered care because they are able to demand and advocate for their needs and are perceived to have the means to pursue legal redress [57, 66]. On the other hand, poorer women may be treated negligently because of the perception that they will be unable to seek legal redress. Women’s ability to advocate for themselves or hold providers accountable is a key factor in how they are treated [55, 57, 66]. In both the current and the prior Kenya study, close to half of providers agreed that since the woman came to the facility, it meant she had consented to all examinations and treatment and that the provider needed to be stern for women to understand the seriousness of the situation. Further, about one-third of providers in both the Ghana and Kenya samples agreed that women were likely to be uncooperative when it was time to push and would need to be physically restrained. Such perceptions likely contribute to findings from previous studies in this setting where women reported experiences of providers not asking for consent before doing examinations and procedures on them and providers being rude and physically abusive [8, 12, 67]. Providers have also reported using physical and verbal abuse as a means of gaining compliance during difficult situations [55, 57, 68]. These findings, though not necessarily indicative of bias, are likely a reflection of the common training, experiences, and health system culture, and need to be addressed in interventions to improve PCMC. Intervention strategies should include training providers on the importance of consenting and patient autonomy, how to communicate complications to patients in a respectful and supportive manner and how to handle difficult situations where patients may not be as compliant. Training should be accompanied by strategies to motivate and support providers, reinforce positive behaviors, as well as well as strategies to hold them accountable for negative behaviors. Interestingly, more providers stated they would want to be a provider for the woman with lower SES than the one with the higher SES. This is consistent with qualitative data, which illuminates the contradictory ways various factors influence provider behavior. Providers preferred to care for lower SES patients because they often “did what they were told,” but ended up providing poorer care to them because they were perceived to be less likely to understand what they were told, had lower expectations, were less likely to advocate for themselves, and were less likely to hold them accountable [39, 55, 57]. Interventions that educate low SES patients on their rights and empower them to communicate their expectations and hold providers accountable may enable them to advocate for better care. However, such interventions place the onus of receiving good care on the patient and not the provider. Provider and system level interventions are thus required, as discussed subsequently. We found associations between some provider socio-demographic factors and explicit bias scores in this and the prior Kenya study. This is not surprising given that reports on PCMC perceptions are influenced by knowledge, which in turn are influenced by socio-demographic factors like education [36, 37]. But unlike in the Kenya study, where no provider characteristics were associated with implicit bias, we found some associations between implicit bias and facility type as well as some provider socio-demographic factors. The role of facility type is especially important given prior research suggests women who give birth in private facilities receive better PCMC [8, 21, 25]. Given people who seek care in these private facilities are more likely to be of higher SES, higher care in these facilities may be due to a combination of factors including bias in favor of higher SES patients. This means that low SES patients seeking care in these private facilities may still receive poorer care if other factors such as higher accountability at the institutional level are not enforced. Such accountability measures include creating mechanisms for all patients to provide feedback on their care experience and providing opportunities for redress. Studies on SES bias in high income countries such as the United States also support the findings presented here. Health visits with lower-SES patients are found to have less time spent on patient questions and assessment of patient’s health knowledge, as well as less socioemotional support and partnership-building conversations [69]. Furthermore, implicit bias towards low-SES patients (and resultant implicit preference towards high-SES patients) has been documented in studies of healthcare professionals in these settings [37, 70, 71]. Research indicates that implicit bias towards low-SES patients may translate into poorer person-centered care, with low-SES patients experiencing less involvement in treatment decisions and lower control over communication [72]. Across studies, lower-SES patients report that their providers communicate poorly, thereby failing to exhibit a core tenet of person-centered care [73, 74]. Our findings imply a need for multilevel interventions to address both implicit and explicit provider biases to reduce the disparities in PCMC. Prior research has used strategies such providing lists of questions for patients to ask doctors during health visits [75] and using coaching to teach communication skills to patients [76], as a way of empowering low SES patients to communicate their expectations and advocate for themselves. As noted however, such interventions place the onus of receiving good care on the patient and not the provider, which should not be the case. Further, requiring time from low-SES patients over and above that spent seeking care to increase their chances of receiving good care is certain to present additional challenges to patients who may already be dealing with several challenges. An alternative and perhaps more effective avenue for designing interventions to improve outcomes for low-SES patients is to target healthcare providers. Training providers to recognize their biases, to be concerned about the effects of bias, to be motivated to identify and learn to replace biased response with responses more consistent with their goals, have been shown to be effective in reducing racial bias [77, 78]. Further, emerging research suggests that it is possible to reduce the effects of people’s bias through activities that elevate the alternative selves and goals that people endorse, without actually removing their deep-seated biases—referred to as sidelining bias [79]. For example, when probation officers adopt a mindset focused on reaching their professional goals to help people get back on their feet—especially people who may not receive that support elsewhere due to previous incarceration—biases against those stigmatized people are rendered dysfunctional to those officers’ reaching their goals; and in turn, mitigate disparities in life outcomes (e.g., recidivism to jail) for previously incarcerated people the officers supervise [80]. Likewise, healthcare providers can be strategically reminded of their professional goals to help people—especially those most in need and unable to otherwise get support—in a way that would render bias and its consequences dysfunctional. Interventions can thus shape healthcare provider mindsets towards empathy. Such interventions have also sidelined consequences of teachers’ biases against students from stigmatized groups and mitigated disparate outcomes in discipline that remove students from the learning environment [80–83]. Evaluations in health care settings are however needed. Beyond these, there is a need to create structures to minimize the effects of people’s individual biases [34]. These can include institutional policies around introductions, communicating procedures, consenting, pain management, among others, and institutional structures for accountability. ## Limitations and strengths First, bias is not a socially desirable attitude, provider’s responses to the questions on explicit bias is thus likely influenced by their perceptions of what they think is the right answer leading to social desirability bias. However, the variation in responses including evidence of bias suggest that providers are willing to explicitly note their bias based on SES characteristics. The relatively low Cronbach alpha for the explicit bias measures is also a limitation. Second, the predictive validity of the IAT in terms of predicting behavior remains disputed, with lack of clarity on whether implicit bias would translate to behavioral differences towards patients among health care professionals [84–86]. Studies examining such relationships are needed. Finally, our sample is drawn from providers (mostly nurses and midwives) working in high-volume maternity units in one region in Ghana and thus may not be generalizable to all providers. Nonetheless, this study makes an important contribution to the literature to achieve equity in PCMC. It is one of the few studies on sources of PCMC disparities in a low-resource setting, and only the second to examine implicit bias in SSA. ## Conclusions The findings from this study strengthens the evidence on the presence of both implicit and explicit SES bias among maternity providers in SSA. This study is important given the dearth of research on how to improve PCMC for low SES patients. The findings provide insights on alternative interventions to achieve equity in PCMC. Such interventions can be approached at different levels including increasing low SES women’s ability to advocate for themselves and interventions that target providers attitudes, mindset, and behavior. Lasting change will however likely come from health system interventions that both motivate and hold providers accountable for equity in PCMC, as well as strengthen the overall health system. Research to develop and test such interventions are urgently needed to reduce disparities in PCMC and to improve PCMC for all women as part of efforts to achieve the fundamental human right of dignity and respect and to achieve the global goals of reducing inequities in maternal mortality and morbidity. ## Supplementary Information Additional file 1.Additional file 2. ## References 1. 1.Institute of Medicine (US) Committee on Quality of Health Care in America. Crossing the Quality Chasm: A New Health System for the 21st Century. Washington (DC): National Academies Press (US); 2001. 2. 2.Afulani PA, Diamond-Smith N, Golub G, Sudhinaraset M. Development of a tool to measure person-centered maternity care in developing settings: validation in a rural and urban Kenyan population. Reprod Health. 2017;14:118 3. 3.White Ribbon Alliance. Respectful Maternity Care: The Universal Rights of Childbearing Women (Full charter). 2011. http://www.healthpolicyproject.com/index.cfm?ID=publications&get=pubID&pubID=46. Accessed 28 Mar 2017. 4. 4.WHO. Prevention and elimination of disrespect and abuse during childbirth. WHO. 2014. http://www.who.int/reproductivehealth/topics/maternal_perinatal/statement-childbirth/en/. Accessed 4 Apr 2017. 5. Tunçalp Ӧ, Were W, MacLennan C, Oladapo O, Gülmezoglu A, Bahl R. Quality of care for pregnant women and newborns—the WHO vision. BJOG Int J Obstet Gynaecol (2015.0) 122 1045-1049. DOI: 10.1111/1471-0528.13451 6. 6.Bowser D, Hill K. Exploring Evidence for Disrespect and Abuse in Facility-Based Childbirth: Report of a Landscape Analysis \| Traction Project. 2010. http://www.tractionproject.org/resources/access-skilled-care-respectful-maternal-care/exploring-evidence-disrespect-and-abuse. Accessed 31 Aug 2015. 7. 7.Bohren MA, Vogel JP, Hunter EC, Lutsiv O, Makh SK, Souza JP, et al. The Mistreatment of Women during Childbirth in Health Facilities Globally: A Mixed-Methods Systematic Review. PLoS Med. 2015;12:e1001847. 8. Afulani PA, Phillips B, Aborigo RA, Moyer CA. Person-centred maternity care in low-income and middle-income countries: analysis of data from Kenya, Ghana, and India. Lancet Glob Health (2019.0) 7 e96-109. DOI: 10.1016/S2214-109X(18)30403-0 9. Bohren MA, Mehrtash H, Fawole B, Maung TM, Balde MD, Maya E. How women are treated during facility-based childbirth in four countries: a cross-sectional study with labour observations and community-based surveys. Lancet (2019.0) 394 1750-1763. DOI: 10.1016/S0140-6736(19)31992-0 10. Miller S, Abalos E, Chamillard M, Ciapponi A, Colaci D, Comandé D. Beyond too little, too late and too much, too soon: a pathway towards evidence-based, respectful maternity care worldwide. Lancet (2016.0) 388 2176-2192. DOI: 10.1016/S0140-6736(16)31472-6 11. Bohren MA, Hunter EC, Munthe-Kaas HM, Souza JP, Vogel JP, Gülmezoglu AM. Facilitators and barriers to facility-based delivery in low- and middle-income countries: a qualitative evidence synthesis. Reprod Health (2014.0) 11 71. DOI: 10.1186/1742-4755-11-71 12. Afulani PA, Kirumbi L, Lyndon A. What makes or mars the facility-based childbirth experience: thematic analysis of women’s childbirth experiences in western Kenya. Reprod Health (2017.0) 14 180. DOI: 10.1186/s12978-017-0446-7 13. Odiase O, Akinyi B, Kinyua J, Afulani P. Community Perceptions of Person-Centered Maternity Care in Migori County. Kenya Front Glob Womens Health (2021.0) 2 71 14. Moyer CA, Mustafa A. Drivers and deterrents of facility delivery in sub-Saharan Africa: a systematic review. Reprod Health (2013.0) 10 40. DOI: 10.1186/1742-4755-10-40 15. 15.Doyle C, Lennox L, Bell D. A systematic review of evidence on the links between patient experience and clinical safety and effectiveness. BMJ Open. 2013;3:e001570 16. Oliveira VC, Refshauge KM, Ferreira ML, Pinto RZ, Beckenkamp PR, Negrao Filho RF. Communication that values patient autonomy is associated with satisfaction with care: a systematic review. J Physiother (2012.0) 58 215-229. DOI: 10.1016/S1836-9553(12)70123-6 17. 17.Hodnett ED, Gates S, Hofmeyr GJ, Sakala C. Continuous support for women during childbirth. In: The Cochrane Collaboration, Hodnett ED, editors. Cochrane Database of Systematic Reviews. Chichester, UK: John Wiley & Sons, Ltd; 2013. 18. 18.Bohren MA, Hofmeyr GJ, Sakala C, Fukuzawa RK, Cuthbert A. Continuous support for women during childbirth. Cochrane Database Syst Rev. 2017;7:CD003766. 19. 19.Sudhinaraset M, Landrian A, Afulani PA, Diamond-Smith N, Golub G. Association between person-centered maternity care and newborn complications in Kenya. Int J Gynaecol Obstet Off Organ Int Fed Gynaecol Obstet. 2019. 10.1002/ijgo.12978. 20. 20.Sudhinaraset M, Landrian A, Golub GM, Cotter SY, Afulani PA. Person-centered maternity care and postnatal health: associations with maternal and newborn health outcomes. AJOG Glob Rep. 2021;1:100005. 21. Afulani PA, Sayi TS, Montagu D. Predictors of person-centered maternity care: the role of socioeconomic status, empowerment, and facility type. BMC Health Serv Res (2018.0) 18 360. DOI: 10.1186/s12913-018-3183-x 22. 22.Afulani PA, Aborigo RA, Walker D, Moyer CA, Cohen S, Williams J. Can an integrated obstetric emergency simulation training improve respectful maternity care? Results from a pilot study in Ghana. Birth Berkeley Calif. 2019. 10.1111/birt.12418. 23. Diamond-Smith N, Treleaven E, Murthy N, Sudhinaraset M. Women’s empowerment and experiences of mistreatment during childbirth in facilities in Lucknow, India: results from a cross-sectional study. BMC Pregnancy Childbirth (2017.0) 17 335. DOI: 10.1186/s12884-017-1501-7 24. Afulani PA, Buback L, Essandoh F, Kinyua J, Kirumbi L, Cohen CR. Quality of antenatal care and associated factors in a rural county in Kenya: an assessment of service provision and experience dimensions. BMC Health Serv Res (2019.0) 19 1-16. DOI: 10.1186/s12913-019-4476-4 25. 25.Montagu D, Ladrian A, Kumar V, Phillips BS, Singah S, Mishra S, et al. Patient-experience during delivery in public health facilities in Uttar Pradesh, India. Health Policy Plan.2019. 10.1093/heapol/czz067. 26. Sudhinaraset M, Beyeler N, Barge S, Diamond-Smith N. Decision-making for delivery location and quality of care among slum-dwellers: a qualitative study in Uttar Pradesh. India BMC Pregnancy Childbirth (2016.0) 16 148. DOI: 10.1186/s12884-016-0942-8 27. Moyer CA, Adongo PB, Aborigo RA, Hodgson A, Engmann CM. “They treat you like you are not a human being”: maltreatment during labour and delivery in rural northern Ghana. Midwifery (2014.0) 30 262-268. DOI: 10.1016/j.midw.2013.05.006 28. Oluoch-Aridi J, Afulani PA, Guzman DB, Makanga C, Miller-Graff L. Exploring women’s childbirth experiences and perceptions of delivery care in peri-urban settings in Nairobi. Kenya Reprod Health (2021.0) 18 83. DOI: 10.1186/s12978-021-01129-4 29. Leape LL, Shore MF, Dienstag JL, Mayer RJ, Edgman-Levitan S, Meyer GS. Perspective: a culture of respect, part 1: the nature and causes of disrespectful behavior by physicians. Acad Med J Assoc Am Med Coll (2012.0) 87 845-852. DOI: 10.1097/ACM.0b013e318258338d 30. Leape LL, Shore MF, Dienstag JL, Mayer RJ, Edgman-Levitan S, Meyer GS. Perspective: a culture of respect, part 2: creating a culture of respect. Acad Med J Assoc Am Med Coll (2012.0) 87 853-858. DOI: 10.1097/ACM.0b013e3182583536 31. Blair IV, Steiner JF, Havranek EP. Unconscious (Implicit) Bias and Health Disparities: Where Do We Go from Here?. Perm J (2011.0) 15 71-78. DOI: 10.7812/TPP/11.979 32. Daumeyer NM, Onyeador IN, Brown X, Richeson JA. Consequences of attributing discrimination to implicit vs. explicit bias. J Exp Soc Psychol (2019.0) 84 103812. DOI: 10.1016/j.jesp.2019.04.010 33. Mendes WB, Koslov K. Brittle smiles: positive biases toward stigmatized and outgroup targets. J Exp Psychol Gen (2013.0) 142 923-933. DOI: 10.1037/a0029663 34. 34.UNC Executive Development. The Real Effects of Unconscious Bias in the Workplace. 2015. http://execdev.kenan-flagler.unc.edu/blog/the-real-effects-of-unconscious-bias-in-the-workplace-0. Accessed 26 Oct 2016. 35. Nosek BA, Ranganath KA, Smith CT, Chugh D, Olson KR, Lindner NM. Pervasiveness and Correlates of Implicit Attitudes and Stereotypes (2007.0) 36. Nosek BA, Smyth FL, Hansen JJ, Devos T, Lindner NM, Ranganath KA. Pervasiveness and correlates of implicit attitudes and stereotypes. Eur Rev Soc Psychol (2007.0) 18 36-88. DOI: 10.1080/10463280701489053 37. Haider AH, Schneider EB, Sriram N, Scott VK, Swoboda SM, Zogg CK. Unconscious Race and Class Biases among Registered Nurses: Vignette-Based Study Using Implicit Association Testing. J Am Coll Surg (2015.0) 220 1077-1086.e3. DOI: 10.1016/j.jamcollsurg.2015.01.065 38. Forscher PS, Lai CK, Axt JR, Ebersole CR, Herman M, Devine PG. A meta-analysis of procedures to change implicit measures. J Pers Soc Psychol (2019.0) 117 522-559. DOI: 10.1037/pspa0000160 39. 39.Afulani PA, Ogolla BA, Oboke EN, Ongeri L, Weiss SJ, Lyndon A, et al. Understanding disparities in person-centred maternity care: the potential role of provider implicit and explicit bias. Health Policy Plan. 2021. 10.1093/heapol/czaa190. 40. Green AR, Carney DR, Pallin DJ, Ngo LH, Raymond KL, Iezzoni LI. Implicit bias among physicians and its prediction of thrombolysis decisions for black and white patients. J Gen Intern Med (2007.0) 22 1231-1238. DOI: 10.1007/s11606-007-0258-5 41. Cooper LA, Roter DL, Carson KA, Beach MC, Sabin JA, Greenwald AG. The Associations of Clinicians’ Implicit Attitudes About Race With Medical Visit Communication and Patient Ratings of Interpersonal Care. Am J Public Health (2012.0) 102 979-987. DOI: 10.2105/AJPH.2011.300558 42. Sabin JA, Greenwald AG. The influence of implicit bias on treatment recommendations for 4 common pediatric conditions: pain, urinary tract infection, attention deficit hyperactivity disorder, and asthma. Am J Public Health (2012.0) 102 988-995. DOI: 10.2105/AJPH.2011.300621 43. Fiscella K. Socioeconomic status disparities in healthcare outcomes: selection bias or biased treatment?. Med Care (2004.0) 42 939-942. DOI: 10.1097/00005650-200410000-00001 44. van Ryn M, Burke J. The effect of patient race and socio-economic status on physicians’ perceptions of patients. Soc Sci Med (2000.0) 50 813-828. DOI: 10.1016/S0277-9536(99)00338-X 45. Bernheim SM, Ross JS, Krumholz HM, Bradley EH. Influence of Patients’ Socioeconomic Status on Clinical Management Decisions: A Qualitative Study. Ann Fam Med (2008.0) 6 53-59. DOI: 10.1370/afm.749 46. Brekke KR, Holmås TH, Monstad K, Straume OR. Socio-economic status and physicians’ treatment decisions. Health Econ (2018.0) 27 e77-89. DOI: 10.1002/hec.3621 47. Arpey NC, Gaglioti AH, Rosenbaum ME. How Socioeconomic Status Affects Patient Perceptions of Health Care: A Qualitative Study. J Prim Care Community Health (2017.0) 8 169-175. DOI: 10.1177/2150131917697439 48. Piette JD, Bibbins-Domingo K, Schillinger D. Health care discrimination, processes of care, and diabetes patients’ health status. Patient Educ Couns (2006.0) 60 41-48. DOI: 10.1016/j.pec.2004.12.001 49. Moore PJ, Sickel AE, Malat J, Williams D, Jackson J, Adler NE. Psychosocial Factors in Medical and Psychological Treatment Avoidance: The Role of the Doctor-Patient Relationship. J Health Psychol (2004.0) 9 421-433. DOI: 10.1177/1359105304042351 50. Lazar M, Davenport L. Barriers to Health Care Access for Low Income Families: A Review of Literature. J Community Health Nurs (2018.0) 35 28-37. DOI: 10.1080/07370016.2018.1404832 51. Job C, Adenipekun B, Cleves A, Samuriwo R. Health professional’s implicit bias of adult patients with low socioeconomic status (SES) and its effects on clinical decision-making: a scoping review protocol. BMJ Open (2022.0) 12 e059837. DOI: 10.1136/bmjopen-2021-059837 52. Siciliani L, Verzulli R. Waiting times and socioeconomic status among elderly Europeans: evidence from SHARE. Health Econ (2009.0) 18 1295-1306. DOI: 10.1002/hec.1429 53. Andersen HM. “Villagers”: Differential treatment in a Ghanaian hospital. Soc Sci Med (2004.0) 59 2003-2012. DOI: 10.1016/j.socscimed.2004.03.005 54. Sudhinaraset M, Treleaven E, Melo J, Singh K, Diamond-Smith N. Women’s status and experiences of mistreatment during childbirth in Uttar Pradesh: a mixed methods study using cultural health capital theory. BMC Pregnancy Childbirth (2016.0) 16 332. DOI: 10.1186/s12884-016-1124-4 55. 55.Afulani PA, Kelly AM, Buback L, Asunka J, Kirumbi L, Lyndon A. Providers’ perceptions of disrespect and abuse during childbirth: a mixed-methods study in Kenya. Health Policy Plan. 2020. 10.1093/heapol/czaa009. 56. Vargas B, Louzado-Feliciano P, Santos N, Fuller S, Jimsheleishvili S, Quiñones Á. An exploration of patient-provider dynamics and childbirth experiences in rural and urban Peru: a qualitative study. BMC Pregnancy Childbirth (2021.0) 21 135. DOI: 10.1186/s12884-021-03586-y 57. 57.Afulani PA, Buback L, Kelly AM, Kirumbi L, Cohen CR, Lyndon A. Providers’ perceptions of communication and women’s autonomy during childbirth: a mixed methods study in Kenya. Reprod Health. 2020;17:85. 58. Afulani PA, Aborigo RA, Nutor JJ, Okiring J, Kuwolamo I, Ogolla BA. Self-reported provision of person-centred maternity care among providers in Kenya and Ghana: scale validation and examination of associated factors. BMJ Glob Health (2021.0) 6 e007415. DOI: 10.1136/bmjgh-2021-007415 59. 59.Ghana Statistical Service2021 Population and Housing Census2022. 2021 Population and Housing Census (2022.0) 60. 60.Ministry of Health GhanaHolistic assessment of 2017 health sector programme of work2018. Holistic assessment of 2017 health sector programme of work (2018.0) 61. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research Electronic Data Capture (REDCap) - A metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform (2009.0) 42 377-381. DOI: 10.1016/j.jbi.2008.08.010 62. 62.Inquisit Lab Overview. https://www.millisecond.com/products/lab. Accessed 3 Nov 2022. 63. Greenwald AG, McGhee DE, Schwartz JLK. Measuring individual differences in implicit cognition: The implicit association test. J Pers Soc Psychol (1998.0) 74 1464-1480. DOI: 10.1037/0022-3514.74.6.1464 64. Nosek BA, Greenwald AG, Banaji MR, Bargh JA. The Implicit Association Test at Age 7: A Methodological and Conceptual Review. Social psychology and the unconscious: The automaticity of higher mental processes (2007.0) 265-292 65. Adams J, Murray R. The general approach to the difficult patient. Emerg Med Clin North Am (1998.0) 16 v. PMID: 9889735 66. Buback L, Kinyua J, Akinyi B, Walker D, Afulani PA. Provider perceptions of lack of supportive care during childbirth: A mixed methods study in Kenya. Health Care Women Int. (2021.0) 0 1-22 67. 67.Abuya T, Warren CE, Miller N, Njuki R, Ndwiga C, Maranga A, et al. Exploring the Prevalence of Disrespect and Abuse during Childbirth in Kenya. PLoS One. 2015;10:e0123606. 68. Bohren MA, Vogel JP, Tunçalp Ö, Fawole B, Titiloye MA, Olutayo AO. “By slapping their laps, the patient will know that you truly care for her”: A qualitative study on social norms and acceptability of the mistreatment of women during childbirth in Abuja, Nigeria. SSM - Popul Health (2016.0) 2 640-655. DOI: 10.1016/j.ssmph.2016.07.003 69. Fiscella K, Goodwin MA, Stange KC. Does patient educational level affect office visits to family physicians?. J Natl Med Assoc (2002.0) 94 157-165. PMID: 11918385 70. Haider AH, Schneider EB, Sriram N, Dossick DS, Scott VK, Swoboda SM. Unconscious race and social class bias among acute care surgical clinicians and clinical treatment decisions. JAMA Surg (2015.0) 150 457-464. DOI: 10.1001/jamasurg.2014.4038 71. Willems S, De Maesschalck S, Deveugele M, Derese A, De Maeseneer J. Socio-economic status of the patient and doctor–patient communication: does it make a difference?. Patient Educ Couns (2005.0) 56 139-146. DOI: 10.1016/j.pec.2004.02.011 72. Verlinde E, De Laender N, De Maesschalck S, Deveugele M, Willems S. The social gradient in doctor-patient communication. Int J Equity Health (2012.0) 11 12. DOI: 10.1186/1475-9276-11-12 73. DeVoe JE, Wallace LS, Fryer GE. Measuring patients’ perceptions of communication with healthcare providers: Do differences in demographic and socioeconomic characteristics matter?. Health Expect (2009.0) 12 70-80. DOI: 10.1111/j.1369-7625.2008.00516.x 74. Kangovi S, Barg FK, Carter T, Levy K, Sellman J, Long JA. Challenges Faced by Patients with Low Socioeconomic Status During the Post-Hospital Transition. J Gen Intern Med (2014.0) 29 283-289. DOI: 10.1007/s11606-013-2571-5 75. Eggly S, Hamel LM, Foster TS, Albrecht TL, Chapman R, Harper FWK. Randomized trial of a question prompt list to increase patient active participation during interactions with black patients and their oncologists. Patient Educ Couns (2017.0) 100 818-826. DOI: 10.1016/j.pec.2016.12.026 76. Street RL, Slee C, Kalauokalani DK, Dean DE, Tancredi DJ, Kravitz RL. Improving physician–patient communication about cancer pain with a tailored education-coaching intervention. Patient Educ Couns (2010.0) 80 42-47. DOI: 10.1016/j.pec.2009.10.009 77. Devine PG, Forscher PS, Austin AJ, Cox WTL. Long-term reduction in implicit race bias: A prejudice habit-breaking intervention. J Exp Soc Psychol (2012.0) 48 1267-1278. DOI: 10.1016/j.jesp.2012.06.003 78. Forscher PS, Mitamura C, Dix EL, Cox WTL, Devine PG. Breaking the prejudice habit: Mechanisms, timecourse, and longevity. J Exp Soc Psychol (2017.0) 72 133-146. DOI: 10.1016/j.jesp.2017.04.009 79. 79.Okonofua JA, Harris LT, Walton GM. Sidelining Bias: A Situationist Approach to Reduce the Consequences of Bias in Real-World Contexts. Curr Dir Psychol Sci. 2022;31:395–404. 80. Okonofua JA, Saadatian K, Ocampo J, Ruiz M, Oxholm PD. A scalable empathic supervision intervention to mitigate recidivism from probation and parole. Proc Natl Acad Sci U S A (2021.0) 118 e2018036118. DOI: 10.1073/pnas.2018036118 81. Okonofua JA, Paunesku D, Walton GM. Brief intervention to encourage empathic discipline cuts suspension rates in half among adolescents. PNAS Proc Natl Acad Sci U S Am (2016.0) 113 5221-5226. DOI: 10.1073/pnas.1523698113 82. Okonofua JA, Perez AD, Darling-Hammond S. When policy and psychology meet: Mitigating the consequences of bias in schools. Sci Adv (2020.0) 6 eaba9479. DOI: 10.1126/sciadv.aba9479 83. Okonofua JA, Goyer JP, Lindsay CA, Haugabrook J, Walton GM. A scalable empathic-mindset intervention reduces group disparities in school suspensions. Sci Adv (2021.0) 8 eabj0691. DOI: 10.1126/sciadv.abj0691 84. Greenwald AG, Poehlman TA, Uhlmann EL, Banaji MR. Understanding and using the Implicit Association Test: III. Meta-analysis of predictive validity. J Pers Soc Psychol (2009.0) 97 17-41. DOI: 10.1037/a0015575 85. Blanton H, Jaccard J, Klick J, Mellers B, Mitchell G, Tetlock PE. Strong claims and weak evidence: reassessing the predictive validity of the IAT. J Appl Psychol (2009.0) 94 567-82. DOI: 10.1037/a0014665 86. Oswald FL, Mitchell G, Blanton H, Jaccard J, Tetlock PE. Predicting ethnic and racial discrimination: A meta-analysis of IAT criterion studies. J Pers Soc Psychol (2013.0) 105 171-192. DOI: 10.1037/a0032734
--- title: 'Significant risk of repeat adverse outcomes in recurrent gestational diabetes pregnancy: a retrospective cohort study' authors: - Sue Lynn Lau - Alex Chung - Joanna Kao - Susan Hendon - Wendy Hawke - Sue Mei Lau journal: Clinical Diabetes and Endocrinology year: 2023 pmcid: PMC10015739 doi: 10.1186/s40842-023-00149-2 license: CC BY 4.0 --- # Significant risk of repeat adverse outcomes in recurrent gestational diabetes pregnancy: a retrospective cohort study ## Abstract ### Background The risk of adverse outcomes in recurrent GDM pregnancy has not been well documented, particularly in women who have already had an adverse outcome. The aim of this study was to compare the risk of recurrent adverse delivery outcome (ADO) or adverse neonatal outcome (ANO) between consecutive gestational diabetes (GDM) pregnancies. ### Methods In this retrospective study of 424 pairs of consecutive (“index” and “subsequent”) GDM pregnancies, we compared the risk of ADO (instrumental delivery, emergency Caesarean section) and ANO (large for gestational age (LGA and small for gestational age (SGA)) in women with and without a history of adverse outcome in their index pregnancy. ### Results Subsequent pregnancies had higher rates of elective Caesarean ($30.4\%$ vs $17.0\%$, $p \leq 0.001$) and lower rates of instrumental delivery ($5\%$ vs $13.9\%$, $p \leq 0.001$), emergency Caesarean ($7.1\%$ vs $16.3\%$, $p \leq 0.001$) and vaginal delivery ($62.3\%$ vs $66.3\%$, $$p \leq 0.01$$). Index pregnancy adverse outcome was associated with a higher risk of repeat outcome: RR 3.09 ($95\%$CI:1.30,7.34) for instrumental delivery, RR 2.20 ($95\%$CI:1.06,4.61) for emergency Caesarean, RR 4.55 ($95\%$CI:3.03,6.82) for LGA, and RR 5.01 ($95\%$CI:2.73,9.22) for SGA). The greatest risk factor for subsequent LGA (RR 3.13 ($95\%$CI:2.20,4.47)) or SGA (RR 4.71 ($95\%$CI:2.66,8.36)) was having that outcome in the index pregnancy. ### Conclusion A history of an adverse outcome is a powerful predictor of the same outcome in the subsequent GDM pregnancy. These high-risk women may warrant more directed management over routine GDM care such as altered glucose targets or increased frequency of ultrasound assessment. ## Introduction Gestational diabetes (GDM) is defined as glucose intolerance first diagnosed in pregnancy [1]. Risk factors include age, ethnicity, family history of diabetes and obesity. Notably, a history of prior GDM confers an estimated 30–$60\%$ risk of recurrent GDM [2–7]. GDM is associated with adverse outcomes for both mother and fetus, the rates of which have been well documented [8, 9]. However, the risk of complications in recurrent GDM has not been as clearly defined. There are currently no evidence-based guidelines for managing recurrent GDM or data on which women are at highest risk of adverse outcomes. Only four studies have examined the risk of adverse outcomes in recurrent GDM [10–12]. One retrospective study of 389 women observed higher fasting glucose levels and pre-pregnancy BMI in the second GDM pregnancy compared to the first, with no increase in LGA or adverse neonatal outcomes [10]. Another study found similar LGA rates in first-time and recurrent GDM [12]. Both studies did not examine individual level data to determine the rate of repeat adverse outcome. In contrast, another study found a higher rate of LGA in the subsequent pregnancy versus the index ($22.4\%$ vs $13.8\%$) [11] and another found a decreased rate of macrosomia and increased rate of SGA in the subsequent GDM pregnancy [13]. The aims of this study were to quantitate the risk of adverse delivery outcome (ADO) and adverse neonatal outcome (ANO) in consecutive GDM pregnancies. We assessed the predictive value of adverse outcome in the index GDM pregnancy on the next GDM pregnancy, in the context of other risk factors. ## Cohort This is a retrospective longitudinal study of 424 GDM pregnancy-pairs, conducted in two centres: the Royal Hospital for Women (RHW), a tertiary maternity hospital in Eastern Sydney, and Blacktown-Mount Druitt Hospital (BMDH), a hospital in Western Sydney with the highest annual number of births statewide. Women who attended GDM clinics from 2003–2015 with more than one GDM pregnancy were identified. Each pregnancy-pair comprised two consecutive singleton GDM pregnancies (“index “ and “subsequent” pregnancies). In women with > 2 GDM pregnancies, each set of consecutive GDM pregnancies was considered a pregnancy-pair- e.g. in a woman with three GDM pregnancies, the first and second pregnancy and the second and third pregnancy were each considered as pregnancy-pairs. Both centres used the Australasian Diabetes in Pregnancy Society diagnostic criteria at the time of a fasting plasma glucose ≥ 5.5 mmol/L and/or a 2-h glucose ≥ 8.0 mmol/L on the 2-h 75 g oral glucose tolerance test (GTT), which was performed in women with a 1-h plasma glucose of ≥ 7.8 mmol/L after a 50 g glucose challenge at 24–28 weeks gestation. Early screening for GDM was performed in the early second trimester in women with a history of GDM in a prior pregnancy, polycystic ovarian syndrome, BMI ≥ 35 kg/m2, maternal age ≥ 40 years or a first-degree relative with type 2 diabetes. Glucose targets were ≤ 5.0 mmol/L fasting and ≤ 7.0 mmol/L two hours after a meal at the RHW, and ≤ 5.5 mmol/L and ≤ 7.0 mmol/L respectively at BMDH. Women were referred to the diabetes educator, instructed on home blood glucose monitoring and a low glycemic index diet and encouraged to do 30 min of exercise per day. They attended one- to four- weekly doctor appointments at the GDM clinic. Insulin was commenced in women who did not regularly meet their blood glucose targets. Diagnostic criteria, glucose targets and guidelines for GDM management remained consistent during the study period. ## Demographic and outcome data Data were obtained from in-house databases, medical files and the Obstetrix Clinical Database System (http://www.meridianhi.com/index.php/obstetrix), a statewide database that accesses data from the New South Wales Perinatal Data Collection, a population-based surveillance system covering all births in the state. Maternal data collected included age at estimated date of confinement, ethnicity, height, weight at booking-in, week of booking-in, week of diagnosis of GDM, results of the GTT, requirement for and starting date of insulin and/or metformin, mode of delivery and need for instrumental delivery. Preterm birth was defined as delivery before 37 weeks gestation. Early GDM was defined as GDM diagnosed before 22 weeks gestation. Instrumental delivery or emergency Caesarean section were considered adverse delivery outcomes (ADO). Neonatal data included gestational age at delivery, sex, birth weight, shoulder dystocia and fetal or neonatal death. Birth centiles were calculated using the Perinatal Institute’s customised centile calculator (https://www.gestation.net/birthweightcentiles/birthweightcentiles.htm) which accounts for maternal height, weight, ethnicity, parity, sex of the child and gestational age at birth, for an Australian population. LGA was defined as a birth weight centile ≥ $90\%$, and SGA was defined as a birth weight centile ≤ $10\%$. At the RHW, neonatal hypoglycemia was defined as capillary blood glucose < 2.2 mmol/L. At BMDH, neonatal hypoglycemia was recorded if this diagnosis had been entered into the Obstetrix database. The primary adverse neonatal outcomes (ANO) studied were LGA and SGA. In addition, a composite ANO was defined as the presence of at least one of the following: shoulder dystocia, perinatal death, LGA or SGA. ## Ethics This study was approved by the South Eastern Sydney Local Health District-Northern Network and the Western Sydney Local Health District Human Research Ethics Committees. ## Statistical analysis Statistical analysis was performed using SPSS 26.0 and SAS 9.4 software. Index and subsequent pregnancies were compared using paired t-tests, Wilcoxon signed-rank tests and McNemar’s test. Chi-squared tests were used to calculate the relative risk (RR) of adverse outcomes in subsequent pregnancies. Subsequent pregnancies with and without LGA, and with and without SGA, were compared using independent t-tests, Mann–Whitney U tests and chi-squared tests. Binomial regression analysis was performed to estimate the RR of recurrent SGA and LGA. Potential factors identified on univariate analysis were included in the model and backward stepwise removal was performed in order to identify independent predictors of each outcome of interest and their adjusted RR. Results are expressed as mean ± standard deviation for parametric data and median and interquartile range for non-parametric data, unless otherwise stated. Critical significance is taken at $5\%$. ## Maternal characteristics 424 pregnancy-pairs were analysed: 170 pairs from RHW (centre 1) and 254 pregnancy-pairs from BMDH (centre 2). There were 804 pregnancies in 380 women, with 32 women having three GDM pregnancies and six having four GDM pregnancies. Maternal characteristics in index and subsequent GDM pregnancies are shown in Table 1. The mean age was 30.6 ± 4.9 years in the index pregnancy and 33.5 ± 4.9 years in the subsequent pregnancy. Booking-in weight and BMI were higher in subsequent pregnancies ($p \leq 0.001$). GDM was diagnosed three weeks earlier ($p \leq 0.001$). The rate of medication use (insulin, metformin or both) was higher in subsequent pregnancies ($63.7\%$ vs $54.0\%$, $p \leq 0.001$) and medication was started three weeks earlier ($p \leq 0.001$). There were no differences in GTT results. Table 1Characteristics of index and subsequent GDM pregnancies. Data expressed as mean (SD), median (IQR) or n (%)GDM pregnancy-pairs ($$n = 424$$)IndexSubsequentp valueAge (years)30.6 (4.9)33.5 (4.9) < 0.001Booking-in weight (kg)67 (58.0,82.0)70 (59.0,87.0) < 0.001Booking-in BMI (kg/m2)26.2 (22.6, 31.6)27.1 (23.4,32.5) < 0.001Week of booking12 (6.0)13 (5.7)0.42Ethnicity (n,%) -Europid178 (42.0) -East Asia89 (21.0) -South Asia68 (16.0) -Middle East57 (13.4) -Other32 (7.6)Parity (n,%) 0238 (56.1)- < 0.001 1104 (24.5)223 (52.7) 245 (10.6)110 (26.0) > 237 (8.7)90 (21.3)Week of GDM diagnosis27 [25,29]24 [16,27] < 0.001Early vs late GDM (n,%) -Early41 (9.7)180 (42.5) < 0.001 -Late383 (90.3)244 (57.5)GTT- fasting glucose (mmol/L)4.9 (0.9)5.0 (1.0)0.22GTT- 2 h glucose (mmol/L)9.0 (1.4)8.9 (1.7)0.51Medication required (%) -Yes229 (54.0)270 (63.7) < 0.001 -No195 (46.0)154 (36.3)Week medication started30 [28,33]27 [21,30] < 0.001Mode of delivery (%) -Vaginal birth281 (66.3)264 (62.3)0.01 -Caesarean section143 (33.7)160 (37.7)Induction of labour (n, %) -Yes200 (47.2)143 (33.7) < 0.001 -No224 (52.8)281 (66.3)Instrumental delivery (n, %) -Yes59 (13.9)21 (5.0) < 0.001 -No365 (86.1)403 (95.0)Emergency Caesarean (n, %) -Yes69 (16.3)30 (7.1) < 0.001 -No355 (83.7)394 (92.9)Elective Caesarean (n, %) -Yes72 (17.0)129 (30.4) < 0.001 -No352 (83.0)295 (69.6)Gestation at delivery (weeks)38.7 (1.6)38.5 (1.4)0.02Delivery < 37 wks -Yes30 (7.1)40 (9.4)0.20 -No394 (92.9)384 (90.6)Birth weight (g)3315 [554]3392 [587]0.005Birth weight centile (%)50 (27.77)54 [28,80]0.23SGA (%) -Yes33 (7.8)37 (8.7)0.67 -No391 (92.2)387 (91.3)LGA (%) -Yes71 (16.7)67 (15.8)0.73 -No353 (83.3)357 (84.2)Birth length (cm)50.1 (3.0)50.3 (2.6)0.12Fetal/neonatal death (n, %) -Yes3 (0.7)1 (0.2)0.63 -No421 (99.3)423 (99.8)Dystocia (n, %) -Yes17 (4.0)8 (1.9)0.09 -No407 (96.0)416 (98.1)Neonatal hypoglycemia (n, %) -Yes57 (13.4)69 (16.3)0.25 -No367 (86.6)355 (83.7)Composite neonatal outcome (death/dystocia/LGA/ SGA) (n, %) -Yes114 (26.9)108 (25.5)0.66 -No310 (73.1)316 (74.5) While there were minor differences in maternal characteristics between the centres, interpregnancy changes were generally comparable. The mean increment in age between index and subsequent pregnancies was slightly smaller in centre 1 versus centre 2 (2.6 ± 1.3 vs 3.1 ± 1.7 years, $$p \leq 0.001$$). Similarly, the mean increment in body weight at booking-in (the first antenatal visit) was smaller in centre 1 (1.8 ± 5.7 vs 3.1 ± 6.1 kg, $$p \leq 0.03$$). There were no differences in interval changes between centres for other maternal parameters including BMI at booking-in and glucose levels on the GTT. ## Adverse delivery outcomes (ADO) Instrumental delivery ($5\%$ vs $13.9\%$, $p \leq 0.001$) and emergency Caesarean Sect. ( $7.1\%$ vs $16.3\%$, $p \leq 0.001$) were decreased in the subsequent pregnancy. However, there was a higher rate of elective Caesarean Sect. ( $30.4\%$ vs $17.0\%$, $p \leq 0.001$) and a lower rate of vaginal delivery ($62.3\%$ vs $66.3\%$, $$p \leq 0.01$$) and induction of labour ($33.7\%$ vs $47.2\%$, $p \leq 0.001$) (Table 1). Only $2.8\%$ of women with a Caesarean section in the index pregnancy went on to have a vaginal delivery in the next pregnancy, with $12.2\%$ having an emergency Caesarean section and $79.7\%$ having an elective Caesarean; this was different to women who delivered via vaginal birth in the index pregnancy, of whom $90.1\%$ delivered via vaginal birth, $4.6\%$ via emergency Caesarean section and $5.3\%$ via elective Caesarean section ($p \leq 0.001$). Mode of delivery in those with a composite neonatal adverse outcome in the index pregnancy (vaginal delivery $55.5\%$, emergency Caesarean $9.2\%$, elective Caesarean $35.3\%$) was similar to those without adverse outcome (vaginal delivery $64.5\%$, emergency Caesarean $6.4\%$, elective Caesarean $29.1\%$) ($$p \leq 0.20$$). ## Adverse neonatal outcomes (ANO) There were no differences in the rates of SGA, LGA, fetal/neonatal death, neonatal hypoglycemia, or the composite ANO (death/dystocia/LGA/SGA) when index and subsequent GDM pregnancies were compared as a group (Table 1). Babies from subsequent pregnancies were delivered slightly earlier (38.5 ± 1.4 vs 38.7 ± 1.6 weeks gestation, $$p \leq 0.02$$) than those from index pregnancies. While birth weight was increased in the subsequent pregnancy (3392 ± 587 g vs 3315 ± 554 g, $$p \leq 0.005$$), customised birth centiles were similar (54 (28–80) % vs 50 (27–77) %, $$p \leq 0.23$$) (Table 1). ## Risk of recurrent adverse delivery outcomes The risk of ADO in the subsequent pregnancy was greatly increased in those women who had ADO in their index pregnancies, with a threefold risk of instrumental delivery in those women who required it in their index pregnancy, and a 2.2-fold risk of emergency Caesarean section compared to women who did not. Similarly, women who had early GDM, requirement for medication, or preterm birth in their index pregnancies were at much higher risk of developing the same outcomes in their subsequent pregnancy (Table 2).Table 2Rate and relative risk of recurrent maternal and neonatal outcomes in the subsequent GDM pregnancyOutcomeRate a (%)$$n = 424$$Relative risk b ($95\%$ CI) in subsequent pregnancyp valueEarly GDM67.51.65 (1.29, 2.11) < 0.001GDM requiring medication87.32.40 (1.98, 2.91) < 0.001Instrumental delivery11.93.09 (1.30, 7.34)0.011Emergency Caesarean section13.02.20 (1.06, 4.61)0.036SGA33.35.01 (2.73, 9.22) < 0.001LGA45.14.55 (3.03, 6.82) < 0.001Preterm birth33.34.38 (2.37, 8.07) < 0.001Composite neonatal outcome (death/dystocia/LGA/ SGA)41.22.10 (1.53, 2.87) < 0.001ain women with this outcome in the index pregnancybcompared to women without this outcome in their index pregnancy ## Risk of recurrent adverse neonatal outcomes While the rate of LGA or SGA was not different in index versus subsequent pregnancies as a group, the risk of these complications in their subsequent pregnancy was greatly increased in those women who had LGA or SGA in their index pregnancy compared to those who did not (Fig. 1). For example, while the rate of LGA was similar in index and subsequent pregnancies ($16.7\%$ vs $15.8\%$, p = NS), the rate of LGA in the subsequent pregnancy was $45.1\%$ in those women who had LGA in their index pregnancy, with a RR of 4.55 compared to women who did not. Likewise, while the rate of SGA was similar in index and subsequent pregnancies ($7.8\%$ vs $8.7\%$, p = NS), the rate of SGA in the subsequent pregnancy was $33.3\%$ in women who had SGA in their index pregnancy, with a RR of 5.01 compared to women who did not (Fig. 1). This greatly increased risk was also the case for the composite ANO (death/dystocia/LGA/SGA) (Table 2).Fig. 1Small for gestational age (SGA) and large for gestational age (LGA) outcomes in index and subsequent pregnancies. Data expressed as n (%) Conversely, having an SGA baby in the index GDM pregnancy was associated with a below average rate of LGA ($6.1\%$, $$n = 2$$/33), and having prior LGA was associated with an SGA rate of only $1.4\%$ ($$n = 1$$/71) in the subsequent GDM pregnancy. Women with no LGA or SGA history had a $7.8\%$ rate of SGA and $10.3\%$ rate of LGA in their subsequent GDM pregnancy (Table 2). ## Factors associated with ANO in the subsequent GDM pregnancy In women with LGA in the index pregnancy, those who went on to have another LGA pregnancy had a higher booking-in BMI in that index pregnancy (30.5 (26.6–40.7) vs 25.7 (23.1–30.6) kg/m2, $$p \leq 0.008$$) as well as the subsequent pregnancy (32.2 (28.3–41.9) vs 27.3 (24.0–30.4) kg/m2, $$p \leq 0.001$$) compared to women who did not have another LGA pregnancy. They also had a higher parity ($50.0\%$ vs $9.0\%$ > 2, $$p \leq 0.02$$) in the index pregnancy and a higher 2 h glucose level on the GTT (10.6 ± 2.5 vs 8.8 ± 2.6 mmol/L, $$p \leq 0.02$$) compared to women who did not have a repeat LGA pregnancy. There were no differences in age or interval between pregnancies. On univariate analysis, women with LGA in their subsequent GDM pregnancy were slightly younger with higher parity compared to those without LGA. They had a 17.5 kg greater median booking-in weight (84.5 (69.0–105.0) vs 67.0 (58.0–82.0) kg, $p \leq 0.001$), higher booking-in BMI (31.3 (26.9–37.4)) vs 26.5 (23.1–32.0) kg/m2, $p \leq 0.001$) and a 2.5 kg greater interpregnancy weight gain than women without LGA (4.7 ± 8.4 vs 2.2 ± 5.4 kg, $$p \leq 0.002$$), despite a similar interpregnancy interval. They had a higher fasting and two-hour glucose on the diagnostic GTT. $47.8\%$ had LGA in their index pregnancy, whereas only $10.9\%$ of women without LGA in the subsequent pregnancy had LGA in the index pregnancy ($p \leq 0.001$) (Table 3).Table 3Characteristics of subsequent GDM pregnancies with and without LGA and SGA. Data expressed as mean (SD), median (IQR) or n (%)No LGA($$n = 67$$)LGA($$n = 357$$)p valueNo SGA ($$n = 387$$)SGA($$n = 387$$)p valueAge (years)33.7 (4.7)32.4 (5.6)0.0233.5 (5.0)33.6 (4.0)0.94Interval between pregnancies (years)3.0 (1.5)2.7 (1.7)0.282.8 (1.5)3.9 (2.1) < 0.001Booking-in weight (kg)67.0(58.0,82.0)84.5[69,105] < 0.00171.0(59.3,88.0)65.0(56.0,77.5)0.07Weight change between pregnancies (kg)2.2 (5.4)4.7 (8.4)0.0022.6 (6.1)2.4 (5.5)0.87Booking-in BMI (kg/m2)26.5(23.1,32.0)31.3(26.9, 37.4) < 0.00127.1(23,4,32.8)23.2(25.8,31.6)0.46Ethnicity (n,%)0.080.11 -Europid38.450.741.627.0 -Non-Europid61.649.358.473.0Parity (n,%)0.0020.87 155.935.852.356.8 225.628.426.224.3 > 218.935.821.518.9Week of GDM diagnosis24.0(17.0,27.0)23.5(14.8,28.0)0.8924.0(16.0,27.0)20.0(15.5,26.0)0.10Early vs late GDM (n,%)0.420.22 -Early42.748.542.654.1 -Late57.351.557.445.9GTT- fasting glucose (mmol/L)4.9 (0.8)5.6 (1.5) < 0.0015.0 (1.0)4.7 (0.7)0.27GTT- 2 h glucose (mmol/L)8.8 (1.6)9.7 (2.2)0.0029.0 (1.7)8.8 (1.5)0.71Medication required (%)0.170.72 -Yes62.571.663.367.6 -No37.528.436.432.4Same outcome in previous GDM preg < 0.001 < 0.001 -Yes10.947.85.770.3 -No89.152.294.329.7 In women with SGA in the index pregnancy, those who went on to have another SGA pregnancy had a longer interval between pregnancies (4.8 ± 2.5 vs 2.8 ± 1.3 years, $$p \leq 0.04$$) compared to those who did not have another SGA pregnancy. There were no differences in age, parity, booking-in BMI or GTT results. On univariate analysis, women with SGA in their subsequent GDM pregnancy had a longer interpregnancy interval (3.9 ± 2.1 vs 2.8 ± 1.5 years, $p \leq 0.001$) compared to women without SGA. $70.3\%$ had SGA in their index pregnancy, versus $5.7\%$ of women without SGA in their subsequent pregnancy ($p \leq 0.001$). There was a trend to lower booking-in weight (65.0 (56.0–77.5) vs 71.0 (59.3–88.0) kg, $$p \leq 0.07$$) but no differences in booking-in BMI or interpregnancy weight change (Table 3). Based on results of univariate analysis, potential predictors of LGA in the subsequent pregnancy were included in a binomial regression model (prior LGA, BMI at booking-in, interpregnancy weight gain, and fasting glucose at diagnostic OGTT). After backward stepwise removal, LGA in the index pregnancy remained the strongest predictor of subsequent LGA, with a RR of 3.13 ($95\%$CI:2.20, 4.47, $p \leq 0.001$) compared to women without prior LGA. Booking-in BMI showed a modest association with LGA outcome- RR 1.04 ($95\%$CI:1.02, 1.07, $p \leq 0.001$). For the outcome of SGA in the subsequent pregnancy, prior SGA, interpregnancy interval and booking-in weight were included in the model. After adjustment, the RR of SGA in women with SGA in the index pregnancy was 4.71 ($95\%$CI:2.66, 8.36, $p \leq 0.001$). For every one-year increase in the interpregnancy interval, the RR of SGA was 1.51 ($95\%$CI:1.19, 1.91, $p \leq 0.001.$ ## Discussion In this study of 424 pairs of consecutive GDM pregnancies, an ADO or ANO in the index GDM pregnancy conveyed a greatly increased risk of the same outcome in the subsequent GDM pregnancy. While these risks have been described in the general antenatal population, they have not been previously quantitated in GDM. Compared to index GDM pregnancies, the rates of instrumental delivery and emergency Caesarean section were more than halved in subsequent pregnancies, with correspondingly increased rates of elective Caesarean section, lower rates of vaginal delivery and induction of labour. This could be explained by a greater consideration of elective Caesarean in women who had a Caesarean section in the index pregnancy, with only $2.8\%$ of these women delivering vaginally in the subsequent pregnancy, and nearly $80\%$ delivering via elective Caesarean. At our centre, all women with a prior Caesarean section are counseled about the small but significant risk of uterine rupture in a subsequent labour and the majority choose to have an elective Caesarean. While ADOs were improved in subsequent pregnancies, the risk of having an ADO was still far greater in women with a history of the same ADO in the index pregnancy, with a RR of 3.09 for instrumental delivery and 2.20 for emergency Caesarean. These risks may justify a lower threshold for elective Caesarean in women with recurrent GDM and history of instrumental delivery or emergency Caesarean. While delivery outcomes were improved in subsequent GDM pregnancies, ANO rates were unchanged in index versus subsequent pregnancies, with LGA rates of ~ $16\%$, SGA rates of ~ $8\%$ and overall composite ANO rates of ~ $26\%$. Given that women with subsequent GDM pregnancies were older, had a higher BMI and were more likely to require medication, it could be hypothesised that they should have had a higher rate of adverse outcomes, which was not the case. One explanation could be that women were diagnosed earlier due to earlier screening which may have affected ANO rates. Four retrospective studies have examined the comparative rates of LGA in first and second GDM pregnancies [10–13], although none have analyzed detailed individual-level data across consecutive pregnancies. Two studies [10, 12] also found that LGA and SGA rates were not significantly different between pregnancies with first time recurrent GDM. In a study of GDM pregnancy-pairs, [11]. LGA rate increased in the second GDM pregnancy ($22.4\%$ vs $13.8\%$, $p \leq 0.05$). The risk of recurrent LGA was $55.7\%$, comparable to our rate of $45.1\%$. It did not examine other ANOs such as SGA and fetal or neonatal death, and clinical information such as timing of diagnosis of GDM, results of the GTT, maternal BMI and interpregnancy interval were not included in the analyses. Thus, they were not able to evaluate for effects of other factors associated with increased risk of recurrent LGA. A fourth study [13] found a decreased rate of SGA in 56 Chinese women wtih recurrent versus index GDM pregnancies. However, this study was confined to diet-controlled GDM. Our study lends a new perspective by tracking the incidence of ANOs in individual women over consecutive GDM pregnancies. The RR of repeat outcome for women with LGA, SGA or any ANO in their GDM pregnancy was 4.5 fold, 5.0 and 2.1 respectively. Put another way, nearly half of women with LGA, $70\%$ of women with SGA and $44\%$ of women with the composite ANO in their subsequent pregnancy had the same outcome in their index pregnancy. Thus while ANO rates were similar in index and subsequent pregnancies as a group, a substantial proportion of adverse outcomes were occurring in the same women. Of additional interest is the very low risk of SGA in women with prior LGA, and the low risk of LGA in women with prior SGA. Multivariate analysis of ANO in the subsequent pregnancy showed that having the same outcome in the index pregnancy was by far the strongest risk factor, with a 3.1-fold risk of LGA and 4.7-fold risk of SGA. In LGA, this risk far outweighed that of maternal BMI, a well-established risk factor for LGA [14, 15]. In SGA, this risk far outweighed that associated with increasing interpregnancy interval. The risks of recurrent LGA or SGA have not been previously described in GDM. Our calculated risks are similar in magnitude to the five-fold risk observed in general obstetric cohorts for LGA [16, 17] as well as for SGA [18, 19]. Moreover, severity of GDM based on the GTT results, need for medication and timing of diagnosis were not associated with subsequent pregnancy outcomes. While on the surface these data may suggest that GDM does not impact greatly on risk of recurrent ANOs, we were not able to include the modifiable factors of gestational weight gain and a measure of glycaemia such as HbA1c in our model, and cannot discount the importance of weight and glycemic management. Limitations of this study include unavailable data on maternal smoking status, gestational weight gain, hypertensive disorders as well as weight gain and glycemic control during pregnancy. There were slight differences between centres in glucose targets and population demographics although reassuringly, interpregnancy changes did not differ between centres. We relied on medication requirement as a surrogate marker of glycaemia. It is possible some of the participants may have had undiagnosed type 2 diabetes as the ADIPS criteria for GDM does not specifically exclude women who may have had undiagnosed type 2 diabetes prior to pregnancy. Strengths of this study included the inclusion of consecutive GDM pregnancy-pairs and the analysis of longitudinal data in individual patients that allowed us to assess risk of adverse outcomes in the context of previous complications. We were able to adjust for relevant clinical covariates including interpregnancy duration and interpregnancy weight gain, both pertinent to recurrent GDM. The definitions of LGA and SGA were based on customised centiles for an Australian population. According to current standards, diagnostic criteria, glucose targets and weight gain targets are applicable to all women with GDM, irrespective of their history of ADO/ANO. There are currently no evidence-based guidelines for managing GDM. Our data support a more individualised management of GDM in those with previous ADO and ANO. ## Conclusions While rates of ANO were similar in index and subsequent GDM pregnancies, the risk was greatly increased in women who had ANO in the index pregnancy. This is despite decreased rates of ADO. Our study identifies a group of women with recurrent GDM and previous LGA who may stand to gain the most from intensive management of their glucose levels and weight. This may include tighter glucose targets and/or more frequent ultrasound assessment. Our study also identifies a group of women with recurrent GDM and previous SGA in whom intensive or early therapy might potentially be unwarranted, given the high risk of recurrent SGA and the low risk of LGA. ## Statements and declarations The authors have no financial or non-financial interests to disclose that are directly or indirectly related to this work. There has been no funding received for this work. ## References 1. Metzger BE. Summary and recommendations of the Third International Workshop-Conference on Gestational Diabetes Mellitus. Diabetes (1991) 40 197-201. DOI: 10.2337/diab.40.2.S197 2. Kim C, Berger DK, Chamany S. Recurrence of gestational diabetes mellitus: a systematic review. Diabetes Care (2007) 30 1314-1319. DOI: 10.2337/dc06-2517 3. MacNeill S, Dodds L, Hamilton DC, Armson BA, VandenHof M. Rates and risk factors for recurrence of gestational diabetes. Diabetes Care (2001) 24 659-662. DOI: 10.2337/diacare.24.4.659 4. Bernstein JA, Quinn E, Ameli O, Craig M, Heeren T, Lee-Parritz A. Follow-up after gestational diabetes: a fixable gap in women's preventive healthcare. BMJ Open Diabetes Res Care (2017) 5 e000445. DOI: 10.1136/bmjdrc-2017-000445 5. Gaudier FL, Hauth JC, Poist M, Corbett D, Cliver SP. Recurrence of gestational diabetes mellitus. Obstet Gynecol (1992) 80 755-758. PMID: 1407910 6. Moses RG. The recurrence rate of gestational diabetes in subsequent pregnancies. Diabetes Care (1996) 19 1348-1350. DOI: 10.2337/diacare.19.12.1348 7. Major CA, deVeciana M, Weeks J, Morgan MA. Recurrence of gestational diabetes: who is at risk?. Am J Obstet Gynecol (1998) 179 1038-1042. DOI: 10.1016/S0002-9378(98)70211-X 8. Adams KM, Li H, Nelson RL, Ogburn PL, Danilenko-Dixon DR. Sequelae of unrecognized gestational diabetes. Am J Obstet Gynecol (1998) 178 1321-1332. DOI: 10.1016/S0002-9378(98)70339-4 9. Xiong X, Saunders LD, Wang FL, Demianczuk NN. Gestational diabetes mellitus: prevalence, risk factors, maternal and infant outcomes. Int J Gynaecol Obstet (2001) 75 221-228. DOI: 10.1016/S0020-7292(01)00496-9 10. Yogev Y, Langer O. Recurrence of gestational diabetes: pregnancy outcome and birth weight diversity. J Matern Fetal Neonatal Med (2004) 15 56-60. DOI: 10.1080/14767050310001650734 11. Kim SY, Kotelchuck M, Wilson HG, Diop H, Shapiro-Mendoza CK, England LJ. Prevalence of adverse pregnancy outcomes, by maternal diabetes status at first and second deliveries, massachusetts, 1998–2007. Prev Chronic Dis (2015) 12 E218. DOI: 10.5888/pcd12.150362 12. Boghossian NS, Yeung E, Albert PS, Mendola P, Laughon SK, Hinkle SN. Changes in diabetes status between pregnancies and impact on subsequent newborn outcomes. Am J Obstet Gynecol (2014) 210 431 e1-14. DOI: 10.1016/j.ajog.2013.12.026 13. Wang N, Lu W, Xu Y, Mao S, He M, Lin X. Recurrence of diet-treated gestational diabetes in primiparous women in northern Zhejiang, China: Epidemiology, risk factors and implications. J Obstet Gynaecol Res (2018) 44 1391-1396. DOI: 10.1111/jog.13688 14. Catalano PM, McIntyre HD, Cruickshank JK, McCance DR, Dyer AR, Metzger BE. The hyperglycemia and adverse pregnancy outcome study: associations of GDM and obesity with pregnancy outcomes. Diabetes Care (2012) 35 780-786. DOI: 10.2337/dc11-1790 15. Barbour LA. Metabolic culprits in obese pregnancies and gestational diabetes mellitus: big babies, big twists, big picture : the 2018 norbert freinkel award lecture. Diabetes Care (2019) 42 718-726. DOI: 10.2337/dci18-0048 16. Hiersch L, Shinar S, Melamed N, Aviram A, Hadar E, Yogev Y. Birthweight and large for gestational age trends in non-diabetic women with three consecutive term deliveries. Arch Gynecol Obstet (2018) 298 725-730. DOI: 10.1007/s00404-018-4872-8 17. Wallace JM, Bhattacharya S, Campbell DM, Horgan GW. Inter-pregnancy weight change and the risk of recurrent pregnancy complications. PLoS One (2016) 11 e0154812. DOI: 10.1371/journal.pone.0154812 18. Hinkle SN, Albert PS, Mendola P, Sjaarda LA, Boghossian NS, Yeung E. Differences in risk factors for incident and recurrent small-for-gestational-age birthweight: a hospital-based cohort study. BJOG. (2014) 121 1080-8. DOI: 10.1111/1471-0528.12628 19. Manzanares S, Maroto-Martin MT, Naveiro M, Sanchez-Gila M, Lopez-Criado S, Puertas A. Risk of recurrence of small-for-gestational-age foetus after first pregnancy. J Obstet Gynaecol (2017) 37 723-726. DOI: 10.1080/01443615.2017.1290057
--- title: 'Relationship Between Vitamin D Deficiency and Non-alcoholic Fatty Liver Disease: A Cross-Sectional Study From a Tertiary Care Center in Northern India' journal: Cureus year: 2023 pmcid: PMC10015758 doi: 10.7759/cureus.34921 license: CC BY 3.0 --- # Relationship Between Vitamin D Deficiency and Non-alcoholic Fatty Liver Disease: A Cross-Sectional Study From a Tertiary Care Center in Northern India ## Abstract Background Vitamin D levels are strongly associated with myocardial infarction, coronary artery disease, heart dysfunction, and even mortality. Non-alcoholic fatty liver disease (NAFLD) is a prevalent hepatic illness whose incidence has grown dramatically over the past several decades. Methodology This observational, cross-sectional study was conducted over 1.5 years (January 2019 to June 2020) at the Department of General Medicine of a tertiary care hospital in northern India on 100 adult patients with NAFLD admitted to the emergency ward, intensive care unit, and medical ward. Results In our study, of the 100 patients, $45.0\%$, $16.0\%$, and $39.0\%$ of patients exhibited vitamin D deficiency, insufficiency, and sufficiency, respectively. Vitamin D deficiency was the highest among those aged 41-50 ($54.2\%$) and lowest among those aged 30-40 ($29.0\%$). We observed that vitamin D deficiency was less prevalent in people with a normal body mass index ($39.1\%$) than in those who were overweight ($91.7\%$). There was a significant ($p \leq 0.05$) association between the severity of vitamin D deficiency and the presence of hepatomegaly, splenomegaly, and ascites. Overall, the incidence of fatty liver was $49\%$ among patients. There was a significant ($$p \leq 0.0001$$) correlation between fatty liver and serum vitamin D levels. The association between the proportion of patients with fatty liver and the degree of vitamin D deficiency was found to be significant ($$p \leq 0.04$$). The relationship between the distribution of patients according to insulin resistance and the degree of vitamin D deficiency was also statistically significant ($p \leq 0.001$). Conclusions Vitamin D deficiency is associated with an increased risk of NAFLD, as well as with the severity of NAFLD. ## Introduction The primary function of vitamin D is to control bone metabolism; nevertheless, its deficiency is associated with many other organ systems. Diabetes, hypertension, hyperlipidemia, and peripheral vascular disease are more prevalent in those with vitamin D deficiency. Vitamin D levels are also strongly linked to coronary artery disease (CAD), myocardial infarction, heart failure, stroke, and incident mortality [1]. Vitamin D deficiency, defined by low serum 25(OH)D levels, may develop from inadequate sun exposure, poor vitamin D consumption, or malabsorption. Vitamin D insufficiency is common among adults and ranges between $24\%$ and $49\%$ of the global population [2]. It has been shown to be associated with a variety of health issues. It is commonly recognized that vitamin D deficiency leads to osteoporosis, osteomalacia, and increased fracture risk. Vitamin D deficiency has been previously linked to an increase in the prevalence of hypertension, diabetes, peripheral vascular disease, and hyperlipidemia [3]. Some studies also indicate that vitamin D levels are associated with obesity, inflammation, and insulin resistance [4]. In addition, it has been demonstrated that vitamin D supplementation lowers free fatty acid (FFA)-induced insulin resistance in animal models [5]. Non-alcoholic fatty liver disease (NAFLD) is perhaps the most common form of liver disease in adults with a worldwide prevalence of $32.4\%$ [6]. Its prevalence has increased rapidly over the past few decades. NAFLD is an umbrella term that encompasses benign adipose tissue accumulation in the liver, progressive steatosis with hepatitis, fibrosis, cirrhosis, and, in rare instances, hepatocellular carcinoma (HCC). NAFLD comprises the following two conditions: non-alcoholic fatty liver (NAFL) and non-alcoholic steatohepatitis (NASH). NAFL is defined by liver steatosis involving more than $5\%$ of the parenchyma without any signs of hepatocyte damage. Histologic criteria identify NASH as a necro-inflammatory process in which steatosis causes liver cells to become destroyed. Despite the fact that the natural history of NAFLD is still being investigated, there is a risk of progression to cirrhosis and HCC, according to the existing evidence. It is currently one of the most prevalent chronic liver diseases worldwide. According to its name, the primary hallmark of NAFLD is an abundance of fat deposits in liver cells. NASH is an aggressive form of fatty liver disease marked by liver inflammation that, in some NAFLD patients, can progress to cirrhosis and ultimately liver failure. This damage is comparable to that caused by excessive alcohol intake. NAFLD is described as the accumulation of fat in the liver in the absence of other factors such as viral hepatitis, alcohol abuse, and others. NAFLD is presently recognized as a significant component of metabolic syndrome [7] and has emerged as a developing clinical entity. Numerous related pathways contribute to the development of NAFLD, and several potential risk factors, including metabolic syndrome, insulin resistance, and obesity, have been discovered. Several studies have shown that vitamin D levels in adults are inversely linked with NAFLD [8,9]. Vitamin D deficiency has long been considered a risk factor for the development of NAFLD. However, the association between vitamin D deficiency and NAFLD has only been investigated and assessed in a small number of studies. ## Materials and methods Study setting and oversight This observational, cross-sectional study was conducted in the Department of General Medicine at a tertiary care hospital in northern India on 100 adult patients with NAFLD admitted to the emergency ward, intensive care unit (ICU), and medical ward over the course of 1.5 years (January 2019 to June 2020). Patients who gave informed consent and fulfilled the inclusion criteria were enrolled in this research. The Institutional Ethics Committee of the Uttar Pradesh University of Medical Sciences granted ethical approval (reference number: $\frac{139}{2018}$). Patients All adult (over 18 years of age) patients with newly diagnosed NAFLD were eligible for inclusion. Key exclusion criteria included patients with significant alcohol consumption or alcohol intake >20 g/day; patients with chronic liver disease, including those seropositive for hepatitis B and C viruses; patients exposed to metals such as antimony, barium salts, chromates, phosphorus, thallium compounds, and uranium compounds; patients who had undergone surgical procedures such as biliopancreatic diversion, extensive small bowel resection, gastric bypass and jejunoileal bypass, abnormal thyroid function test; and patients who refused to give consent. Methodology Patients who fulfilled the study’s inclusion criteria were identified. Patients were informed of the purpose of the research, and those who agreed to participate were enrolled in the study. After obtaining informed consent, each patient’s complete medical history was collected and recorded using a standardized questionnaire that covered demographics, leisure activities, past medical history, family medical history, medication history, nicotine and alcohol use, and dietary patterns. All measurements, including body height and weight, hip, and waist-to-hip ratio (WHR), were measured in accordance with the World Health Organization (WHO) recommendations. After eliminating alcohol consumption and viral or other liver illnesses, NAFLD was diagnosed using abdominal ultrasound findings and graded according to severity. A structured form was utilized to record the results of the examination. During the examination, the liver, gallbladder, kidneys, and spleen were among the organs evaluated. The liver size, presence of localized lesions, and hepatic steatosis were assessed. According to previous research [10-12], hepatic steatosis was diagnosed based on predetermined criteria. The presence of comorbidities such as hepatomegaly, splenomegaly, and ascites was also noted. Complete blood count (CBC), fasting lipid profile, serum total cholesterol (TC) (cholesterol esterase method), high-density lipoprotein (HDL-C) (cholesterol esterase method after precipitation with phosphotungstate method), low-density lipoprotein (LDL-C) (Fried Wald’s Formula: LDL - C = TC - HDL - C + TG/5), viral markers, fasting blood sugar, fasting serum insulin, liver function test, and kidney function test estimates were performed using the Selectra Pro XL fully automated clinical chemistry analyzer with reagent (ELITechGroup, Puteaux, France). Viral markers were determined using chemiluminescent immunoassays. The CBC was calculated using an automated hematological analyzer (Sysmex XS-1000i, Sysmex Corporation, Kobe, Japan). Serum 25(OH)D levels were measured using an Achitect 25-OH vitamin D test (Abbott Diagnostics, Lake Forest, IL, USA). The assay has a minimum limit of detection of 2.8 ng/mL and a maximum limit of 147.8 ng/mL. 25(OH)D levels were categorized as deficient (20 ng/mL), insufficient (20-29 ng/mL), and sufficient (30 ng/mL). Statistical analysis Microsoft Office Excel version 2010 was used in creating the database and producing graphs, while the data were analyzed using SPSS version 26 for Windows (IBM Corp., Armonk, NY, USA). Various tests of significance were employed for the analysis of data, as mentioned in the Results section. Categorical data were evaluated using the chi-square test. A p-value <0.05 was used as the level of significance. ## Results In our study, of the total 100 patients, $45.0\%$, $16.0\%$, and $39.0\%$ of patients were vitamin D deficient, insufficient, and sufficient, respectively (Figure 1). Overall, $34\%$ of patients were between 30 and 40 years old, followed by $32\%$ aged >50, $24\%$ aged 41-50, and $10\%$ aged 30 years. Vitamin D deficiency was the highest among those aged 41-50 ($54.2\%$) and lowest among those aged 30-40 ($29.0\%$) (Figure 2). However, there was no significant correlation between the severity of vitamin D insufficiency and age ($p \leq 0.05$). Further, $64\%$ of the patients were female. Vitamin D deficiency was greater in females ($46.9\%$) than in males ($41.7\%$), although there was no significant correlation ($p \leq 0.05$) between the degree of vitamin D deficiency and gender. Figure 1: Serum vitamin D status in NAFLD patients.NAFLD = non-alcoholic fatty liver disease Figure 2: Distribution of patients based on age and serum vitamin D status. The clinical history of patients is presented in Table 1, with abdominal discomfort being the most prevalent ($18.0\%$) symptom. Not a single patient was an alcoholic. About $35\%$ of patients were smokers. Vitamin D insufficiency was lower among smokers ($42.9\%$) than among non-smokers ($46.2\%$). The percentage of overweight (body mass index (BMI) ≥25) patients was $12\%$. Vitamin D deficiency was lower in normal people ($39.1\%$) than in obese patients ($91.7\%$). Overall, $9\%$ and $21\%$ of patients, respectively, exhibited splenomegaly and hepatomegaly. Vitamin D deficiency was greater among those with such comorbidities than among those without, with a significant ($p \leq 0.05$) correlation between severe vitamin D deficiency and the presence of comorbidity (Table 2). Table 3 depicts the relationship between several blood tests and the degree of vitamin D deficiency. All patients were negative for viral indicators. According to the degree of vitamin D deficiency, the results of the liver function test, lipid profile, uric acid, and platelet count were determined and investigated. The primary incidence of fatty liver in patients was $49\%$. The association between fatty liver and vitamin D deficiency was statistically significant ($$p \leq 0.0001$$) The correlation between vitamin D deficiency and the proportion of patients with fatty liver was found to be statistically significant ($$p \leq 0.04$$). The correlation between the distribution of patients by insulin resistance and the degree of vitamin D deficiency was also statistically significant ($p \leq 0.001$) (Table 4). ## Discussion Vitamin D deficiency was linked to an increased risk of NAFLD in this cross-sectional research. Although the exact relationship between vitamin D and NAFLD is unknown, several studies have suggested that vitamin D deficiency is linked to the prevalence and severity of NAFLD [13-15]. In adults with normal serum liver enzymes, low vitamin D levels are linked to the development of NAFLD independent of metabolic syndrome. The proportion of female patients was greater than male patients among the 100 patients investigated, with a male-to-female ratio of 36:64. In their research, Lonardo and Suzuki also noted a gender difference [16]. According to a previous study, NAFLD was highly prevalent among women. The gender disparity in the prevalence of NAFLD might be related to differences in fat distribution between men and women. The bulk of NAFLD cases were in their fourth and fifth decades, although there was no significant relationship, according to our data. Vitamin D deficiency was the highest in those aged 41-50 years ($54.2\%$) and lowest in those aged 30-40 years ($29.4\%$). Clinical characteristics were investigated by Basaranoglu and Neuschwander-Tetri, and most occurrences were observed in people in their 40s or 50s, though the entire spectrum has also been reported in children. Obesity and hypertension are common in NAFLD patients. Although many individuals have no symptoms, tiredness and dullness are the most common complaints. Mild-to-moderate hepatomegaly is one of the most prevalent physical examination findings. Patients with NAFLD may have hyperlipidemia, hyperglycemia, hyperinsulinemia, and decreased insulin sensitivity on a biochemical level [17]. Hadizadeh et al. conducted a study and reported analogous findings [18]. The analysis of variance test revealed a significant ($$p \leq 0.02$$) difference in systolic blood pressure between the severity of vitamin D deficiency in our research. The functional crosstalk between NAFLD and hypertension has been reported and discussed by some studies; moreover, in non-hypertensive people, no study has described the salient features of NAFLD. Our findings show that blood pressure is linked to NAFLD in non-hypertensive individuals; both systolic and diastolic blood pressure have been identified as independent risk factors for NAFLD [19]. Patients with NAFLD were shown to have higher body weight, fasting blood glucose, and blood pressure concentration than those without NAFLD ($p \leq 0.05$), according to another study. Insulin resistance, which is important in the development of both illnesses, is linked to NAFLD and hypertension. Elevated blood pressure has also been demonstrated as a sign of NAFLD [20]. We also observed that in NAFLD patients, vitamin D deficiency was significantly associated with insulin resistance and higher homeostasis model assessment-estimated insulin resistance values compared to those without insulin resistance. Multiple studies have demonstrated the association between obesity and vitamin D deficiency, with both working synergistically to influence the risk of insulin resistance [21]. Low levels of blood 25(OH)D are inversely linked with markers of obesity, including BMI (30 kg/m2), fat mass, and waist circumference [22]. In bidirectional genetic studies, high BMI was associated with decreased 25(OH)D; each unit increase in BMI was associated with a $1.15\%$ decrease in blood 25(OH)D concentration [23,24]. According to our findings, which are analogous to those of Glass et al. [ 25], vitamin D deficiency is significantly associated with the presence of hepatomegaly, splenomegaly, and ascites in NAFLD. In our investigation, vitamin D deficiency was also shown to be associated with elevated alanine transaminase and aspartate aminotransferase levels in NAFLD patients. The subsequent results may explain the potential processes underlying the link between vitamin D and NAFLD. Vitamin D influences extraskeletal metabolic organs, which can indirectly influence hepatic lipid metabolism and lower blood TC, TG, and LDL levels, leading to additional liver fat buildup [26]. Vitamin D modulates the immune system and has anti-inflammatory properties. Vitamin D deficiency results in dysfunctional adipose tissue and, consequently, chronic inflammation, which may contribute to the development of NAFLD [27,28]. According to Chatrath et al., NAFLD patients demonstrated atypical lipid metabolism and metabolic syndrome characteristics. NAFLD is characterized by elevated triglycerides, elevated LDL (non-type A), and decreased HDL. NAFLD is caused by the excessive synthesis of very low-density lipoprotein by the liver and the decreased clearance of lipids by the liver, which results in abnormal lipid metabolism. The atherogenic lipid profile of NAFLD is driven by insulin resistance, which may be aggravated by vitamin D deficiency [29]. The distribution of patients according to the grade of fatty liver and its relationship with the degree of vitamin D insufficiency was also discussed in our study. Fatty liver grade, as assessed by ultrasonography, was shown to have a significant ($$p \leq 0.0001$$) relationship with the degree of vitamin D deficiency. In 2014, Park et al. performed a cross-sectional study to investigate the relationship between vitamin D levels and NAFLD and reported a substantial link between vitamin D insufficiency and NAFLD as well [30]. Our study has numerous limitations, including the fact that it was a single-center study based on a small sample, prohibiting generalizations regarding the prevalence of fatty liver in the Indian community. In cross-sectional studies, it might not always be feasible to eliminate selection bias. If patients with complications die prematurely, it can lead to the selection of survivors. This is an imminent problem with older age groups. To establish a stronger link, more research with a larger study population is required. ## Conclusions Vitamin D deficiency is linked to an increased incidence of NAFLD as well as the severity of NAFLD grade. Estimating vitamin D levels can assist in minimizing the risk of NAFLD. In NAFLD cases, vitamin D supplementation can help slow down the progression of the disease. In this study, vitamin D deficiency was also linked to deranged liver enzymes and dyslipidemia in NAFLD patients. ## References 1. Anderson JL, May HT, Horne BD. Relation of vitamin D deficiency to cardiovascular risk factors, disease status, and incident events in a general healthcare population. Am J Cardiol (2010) 106 963-968. PMID: 20854958 2. Cashman KD. Global differences in vitamin D status and dietary intake: a review of the data. Endocr Connect (2022) 11 0 3. Papandreou D, Hamid ZT. The role of vitamin D in diabetes and cardiovascular disease: an updated review of the literature. Dis Markers (2015) 2015 580474. PMID: 26576069 4. Chagas CE, Borges MC, Martini LA, Rogero MM. Focus on vitamin D, inflammation and type 2 diabetes. Nutrients (2012) 4 52-67. PMID: 22347618 5. Zhou QG, Hou FF, Guo ZJ, Liang M, Wang GB, Zhang X. 1,25-Dihydroxyvitamin D improved the free fatty-acid-induced insulin resistance in cultured C2C12 cells. Diabetes Metab Res Rev (2008) 24 459-464. PMID: 18551686 6. Riazi K, Azhari H, Charette JH. The prevalence and incidence of NAFLD worldwide: a systematic review and meta-analysis. Lancet Gastroenterol Hepatol (2022) 7 851-861. PMID: 35798021 7. Bhatia LS, Curzen NP, Calder PC, Byrne CD. Non-alcoholic fatty liver disease: a new and important cardiovascular risk factor?. Eur Heart J (2012) 33 1190-1200. PMID: 22408036 8. Gad AI, Elmedames MR, Abdelhai AR, Marei AM. The association between vitamin D status and non-alcoholic fatty liver disease in adults: a hospital-based study. Egypt Liver J (2020) 10 9. Hariri M, Zohdi S. Effect of vitamin D on non-alcoholic fatty liver disease: a systematic review of randomized controlled clinical trials. Int J Prev Med (2019) 10 14. PMID: 30774848 10. Saverymuttu SH, Joseph AE, Maxwell JD. Ultrasound scanning in the detection of hepatic fibrosis and steatosis. Br Med J (Clin Res Ed) (1986) 292 13-15 11. Hamaguchi M, Kojima T, Itoh Y. The severity of ultrasonographic findings in nonalcoholic fatty liver disease reflects the metabolic syndrome and visceral fat accumulation. Am J Gastroenterol (2007) 102 2708-2715. PMID: 17894848 12. Charatcharoenwitthaya P, Lindor KD. Role of radiologic modalities in the management of non-alcoholic steatohepatitis. Clin Liver Dis (2007) 11 37-0. PMID: 17544971 13. Chung GE, Kim D, Kwak MS, Yang JI, Yim JY, Lim SH, Itani M. The serum vitamin D level is inversely correlated with nonalcoholic fatty liver disease. Clin Mol Hepatol (2016) 22 146-151. PMID: 27044765 14. Barchetta I, Angelico F, Del Ben M, Baroni MG, Pozzilli P, Morini S, Cavallo MG. Strong association between non alcoholic fatty liver disease (NAFLD) and low 25(OH) vitamin D levels in an adult population with normal serum liver enzymes. BMC Med (2011) 9 85. PMID: 21749681 15. Wang D, Lin H, Xia M. Vitamin D levels are inversely associated with liver fat content and risk of non-alcoholic fatty liver disease in a Chinese middle-aged and elderly population: the Shanghai Changfeng study. PLoS One (2016) 11 0 16. Lonardo A, Suzuki A. Nonalcoholic fatty liver disease: does sex matter?. Hepatobiliary Surg Nutr (2019) 8 164-166. PMID: 31098369 17. Basaranoglu M, Neuschwander-Tetri BA. Nonalcoholic fatty liver disease: clinical features and pathogenesis. Gastroenterol Hepatol (N Y) (2006) 2 282-291. PMID: 28286458 18. Hadizadeh F, Faghihimani E, Adibi P. Nonalcoholic fatty liver disease: diagnostic biomarkers. World J Gastrointest Pathophysiol (2017) 8 11-26. PMID: 28573064 19. Qian LY, Tu JF, Ding YH, Pang J, Che XD, Zou H, Huang DS. Association of blood pressure level with nonalcoholic fatty liver disease in nonhypertensive population: normal is not the new normal. Medicine (Baltimore) (2016) 95 0 20. Lv WS, Sun RX, Gao YY. Nonalcoholic fatty liver disease and microvascular complications in type 2 diabetes. World J Gastroenterol (2013) 19 3134-3142. PMID: 23716995 21. Tzotzas T, Papadopoulou FG, Tziomalos K, Karras S, Gastaris K, Perros P, Krassas GE. Rising serum 25-hydroxy-vitamin D levels after weight loss in obese women correlate with improvement in insulin resistance. J Clin Endocrinol Metab (2010) 95 4251-4257. PMID: 20534751 22. Rajakumar K, de las Heras J, Chen TC, Lee S, Holick MF, Arslanian SA. Vitamin D status, adiposity, and lipids in black American and Caucasian children. J Clin Endocrinol Metab (2011) 96 1560-1567. PMID: 21367931 23. Jorde R, Sneve M, Emaus N, Figenschau Y, Grimnes G. Cross-sectional and longitudinal relation between serum 25-hydroxyvitamin D and body mass index: the Tromsø study. Eur J Nutr (2010) 49 401-407. PMID: 20204652 24. Vimaleswaran KS, Berry DJ, Lu C. Causal relationship between obesity and vitamin D status: bi-directional Mendelian randomization analysis of multiple cohorts. PLoS Med (2013) 10 0 25. Glass LM, Hunt CM, Fuchs M, Su GL. Comorbidities and nonalcoholic fatty liver disease: the chicken, the egg, or both?. Fed Pract (2019) 36 64-71. PMID: 30867626 26. Jafari T, Fallah AA, Barani A. Effects of vitamin D on serum lipid profile in patients with type 2 diabetes: a meta-analysis of randomized controlled trials. Clin Nutr (2016) 35 1259-1268. PMID: 27020528 27. Cimini FA, Barchetta I, Carotti S. Relationship between adipose tissue dysfunction, vitamin D deficiency and the pathogenesis of non-alcoholic fatty liver disease. World J Gastroenterol (2017) 23 3407-3417. PMID: 28596677 28. Bril F, Maximos M, Portillo-Sanchez P. Relationship of vitamin D with insulin resistance and disease severity in non-alcoholic steatohepatitis. J Hepatol (2015) 62 405-411. PMID: 25195551 29. Chatrath H, Vuppalanchi R, Chalasani N. Dyslipidemia in patients with nonalcoholic fatty liver disease. Semin Liver Dis (2012) 32 22-29. PMID: 22418885 30. Park D, Kwon H, Oh SW. Is vitamin D an independent risk factor of nonalcoholic fatty liver disease?: a cross-sectional study of the healthy population. J Korean Med Sci (2017) 32 95-101. PMID: 27914137
--- title: 'High prevalence of cardiometabolic risk factors amongst young adults in the United Arab Emirates: the UAE Healthy Future Study' authors: - Fatima Mezhal - Abderrahim Oulhaj - Abdishakur Abdulle - Abdulla AlJunaibi - Abdulla Alnaeemi - Amar Ahmad - Andrea Leinberger-Jabari - Ayesha S. Al Dhaheri - Eiman AlZaabi - Fatma Al-Maskari - Fatme Alanouti - Fayza Alameri - Habiba Alsafar - Hamad Alblooshi - Juma Alkaabi - Laila Abdel Wareth - Mai Aljaber - Marina Kazim - Michael Weitzman - Mohammad Al-Houqani - Mohammad Hag Ali - E. Murat Tuzcu - Naima Oumeziane - Omar El-Shahawy - Rami H. Al-Rifai - Scott Sherman - Syed M. Shah - Thekra Alzaabi - Tom Loney - Wael Almahmeed - Youssef Idaghdour - Luai A. Ahmed - Raghib Ali journal: BMC Cardiovascular Disorders year: 2023 pmcid: PMC10015775 doi: 10.1186/s12872-023-03165-3 license: CC BY 4.0 --- # High prevalence of cardiometabolic risk factors amongst young adults in the United Arab Emirates: the UAE Healthy Future Study ## Abstract ### Background Cardiovascular disease (CVD) is the leading cause of death in the world. In the United Arab Emirates (UAE), it accounts for $40\%$ of mortality. CVD is caused by multiple cardiometabolic risk factors (CRFs) including obesity, dysglycemia, dyslipidemia, hypertension and central obesity. However, there are limited studies focusing on the CVD risk burden among young Emirati adults. This study investigates the burden of CRFs in a sample of young Emiratis, and estimates the distribution in relation to sociodemographic and behavioral determinants. ### Methods Data was used from the baseline data of the UAE Healthy Future Study volunteers. The study participants were aged 18 to 40 years. The study analysis was based on self-reported questionnaires, anthropometric and blood pressure measurements, as well as blood analysis. ### Results A total of 5167 participants were included in the analysis; $62\%$ were males and the mean age of the sample was 25.7 years. The age-adjusted prevalence was $26.5\%$ for obesity, $11.7\%$ for dysglycemia, $62.7\%$ for dyslipidemia, $22.4\%$ for hypertension and $22.5\%$ for central obesity. The CRFs were distributed differently when compared within social and behavioral groups. For example, obesity, dyslipidemia and central obesity in men were found higher among smokers than non-smokers ($p \leq 0.05$). And among women with lower education, all CRFs were reported significantly higher than those with higher education, except for hypertension. Most CRFs were significantly higher among men and women with positive family history of common non-communicable diseases. ### Conclusions CRFs are highly prevalent in the young Emirati adults of the UAE Healthy Future Study. The difference in CRF distribution among social and behavioral groups can be taken into account to target group-specific prevention measures. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12872-023-03165-3. ## Background Cardiovascular diseases (CVD) are the most common non-communicable diseases (NCDs) globally, and constitute the leading cause of global mortality, as well as a major contributor to reduced quality of life [1]. CVD death rates have increased steadily from 12.1 million, in 1990, to 18.6 million in 2019 [2]. In 2017, it was responsible for 17.8 million deaths worldwide and corresponded to 35.6 million years lived with related disability [1, 2]. Approximately $80\%$ of CVD-related deaths are caused by coronary heart disease (CHD), and strokes. The World Health Organization (WHO) reports that NCDs account for $77\%$ of all deaths in the United Arab Emirates (UAE); CVDs account for $40\%$ of the causes [3]. The UAE Ministry of Health and Prevention report [2019] revealed that $22\%$ of CVD-related deaths were attributable to acute myocardial infarction, followed by cerebrovascular disease, ischemic heart disease, and hypertension [4]. Individuals at risk of CVD may have a cluster of risk factors including obesity, raised blood pressure, high blood glucose, abnormal lipids as well as abdominal obesity. These are the most common cardiovascular risk factors, also referred to as cardiometabolic risk factors (CRFs). The INTERHEART study, which included data from 52 countries across the world, showed that smoking, hypertension, high low-density lipoprotein level, and diabetes accounted for $76\%$ of the risk of myocardial infarction [5]. Another study that compiled data from 14 clinical trials, involving 122,458 patients, similarly concluded that smoking, diabetes, hyperlipidemia, and hypertension are affected by these same main risk factors [6]. In 2012, it was reported that there was a higher prevalence of CRFs in the UAE, as opposed to other developed countries, and the related deaths were above the global average [7]. Although clinical signs of CVD usually present in adulthood, early atherosclerotic changes occur during adolescence. The Framingham Offspring Study showed that risk factor exposure during early adulthood (ages 20–39 years) was associated with coronary heart events after the age of 40 years [8]. The study showed that high blood pressure, and abnormal lipid levels, were associated with an 8 to 30-fold increase in cardiac events. The most important behavioral risk factors for CVD that are comprehensively reported in the literature, include: tobacco use, physical inactivity, and poor diet [9–12]. Additionally, there are a number of underlying determinants of CVD, such as socioeconomic status (SES) and hereditary factors. Examples of SES indicators, on the individual level, include: education, occupation and marital status. With regards to education, studies have shown that the higher the education level of the individual is, the greater the possibility of adequate life choices, which in turn leads to a reduced prevalence of hypertension, diabetes and obesity [13–15]. With regards to employment, although evidence that having a financial income can, for example, increase health quality by being able to have better access to healthy food options, however the stress and demands of a job can increase CVD risk by $50\%$ [16]. A meta-analysis on marital status, as a social factor affecting CVD risk, concluded that, being unmarried increased the odds of CVD by $42\%$ and CHD by $16\%$ compared to married individuals [17]. Looking at hereditary factors, the World Heart Federation (WHF) states that if a first-degree relative suffered from a heart attack before the age of 55 for men, or 65 for women, the subject is at greater risk of developing the disease [18]. There are limited studies in the UAE that focus primarily on young adults in the context of CVD and associated risk factors. The objective of this study was to investigate the burden of CRFs in a sample of young Emiratis, and to estimate the prevalence of CRFs within social and behavioral determinants. ## Study population The study sample includes participants from the UAE Healthy Future Study (UAEHFS) [19]. The UAEHFS is a population-based prospective cohort study recruiting 20,000 adult Emiratis to explore the risk factors for NCDs in the UAE. Participants are opportunistically recruited at multiple sites including health centers, universities and companies [19]. The study was based on the cross-sectional analysis of baseline data from the UAEHFS cohort, recruited between February 2016 and December 2018. Subjects were Emirati nationals aged 18 to 40 years. All participants were required to provide informed consent. Participants who reported any acute infection at the time of recruitment and pregnant women were excluded from the study. This study was approved by the Abu Dhabi Health Research and Technology Committee (ref. DOH/HQD/$\frac{2020}{516}$). Additional information on the UAEHFS methodology is published elsewhere [19]. ## Data collection protocol for UAEHFS Participants answered a self-completed questionnaire, underwent physical measurements, and gave a blood sample. The questionnaire collected information on risk factors that pertain to NCD development. The questions explored socio-demographic factors, general health, and early life exposures. The family history of NCDs was also considered to see whether a parent may have had a heart disease, stroke, or a combination of a known CVD risk factor such as high cholesterol, hypertension, diabetes and obesity. Physical activity was also assessed using the WHO’s physical activity tool; the Global Physical Activity Questionnaire (GPAQ) [20]. The tool quantifies the physical activity levels and time spent into metabolic equivalents that can be calculated and categorized into low, moderate and high physical activity levels. The self-completed questionnaire also addressed tobacco smoking status and types (cigarette, midwakh, or water-pipe “shisha”). The subsequent steps of physical assessments included measuring brachial blood pressure and anthropometric measures (BMI, waist and hip circumferences) were performed by trained nurses that followed a standardized protocol. Finally, a sample of random venous blood was collected and used for analyzing blood lipids and HbA1c. Only fasting blood samples were used to analyze plasma glucose. ## Cardiometabolic risk factors (CRFs) Body mass index (BMI) was categorized according to the WHO definitions; a BMI less than 25.0 kg/m2 was considered normal, a BMI that lies between 25.0 and 29.9 kg/m2 was considered overweight, and a BMI above 30.0 Kg/m2 was classified as obese. Dysglycemia, or abnormal glycemic status, was defined as having one or more of the following; HbA1c ≥ $5.7\%$, self-reporting diabetes or taking antidiabetic medication in the questionnaire, or fasting blood glucose (FBG) ≥ 100 mg/dL (this was only done for a subset of 1080 participants). Dyslipidemia was defined as either self-reported diagnosis of abnormal cholesterol level, or taking a lipid-controlling medication in the questionnaire, or having an abnormal test level of any of the following; low-density lipoprotein (LDL) cholesterol level of ≥130 mg/dL, high-density lipoprotein (HDL) cholesterol level of ≤40 mg/dL for men or ≤ 50 mg/dL for women, total cholesterol ≥200 mg/dL, or triglycerides ≥150 mg/dL for fasting samples and ≥ 175 mg/dL for non-fasting samples [21, 22]. Elevated blood pressure, or hypertension, was defined as having 2 or 3 blood pressure readings of ≥140 mmHg systolic and/or ≥ 90 mmHg diastolic according to the American Heart Association guidelines [23]. Hypertension was also defined as having self-reported “hypertensive” on the questionnaire and/or whether they are taking blood pressure-controlling medication. Central obesity was indicated if the waist-to-hip ratio was ≥0.85 for women and ≥ 0.90 for men [24]. ## Statistical analyses Categorical data was presented as frequencies and percentages. Continuous variables were presented as means ± standard deviation. The frequencies and percentages were tested for significance of any differences in distribution between two or more groups using Chi-square and Fisher’s exact tests, as appropriate. For continuous variables, differences in means were measured by t-tests and one-way ANOVA tests. Logistic regression models were used to estimate the overall age-adjusted prevalence for each CRF, and to estimate age-adjusted prevalence of every CRF within each social and behavioral factor. Estimates were reported with the $95\%$ confidence interval ($95\%$ CI). Two-sided tests with $P \leq 0.05$ were considered statistically significant. The analyses were performed using Stata statistical software version 15 [25]. ## Results Overall, a total of 5167 subjects aged between 18 and 40 years were recruited between February 2016 and December 2018. Participants were from different cities; with the majority (around $70\%$) from Abu Dhabi emirate. Complete self-reported data was available for up to $85\%$ of the participants, depending on the data point concerned. Complete body measurements including anthropometrics and blood pressure was available for $94\%$ of the sample. Finally, complete blood sample testing was available for $98\%$ of the sample; where $79.1\%$ were non-fasting and $20.9\%$ were fasting samples. Table 1 summarizes the study sample’s social and behavioral characteristics. It included $38\%$ females and $62\%$ males. The mean age for the sample was 25.7 (±6.2) years with a median age of 24 years. The age distribution was significantly different between women and men, where women were generally younger in the sample. Most of the participants were single ($63.6\%$) and employed ($53.9\%$). About half of the participants had college or post-graduate degree ($46\%$) while the other half had a high-school diploma or below ($54\%$). Among females, the majority were students ($44.3\%$), while most men were employed ($68.5\%$) ($P \leq 0.001$). Family history of NCDs was reported by $56\%$ of the overall study population. Table 1Social and behavioral characteristics of the UAE Healthy Future Study participantsOVERALLMENWOMEN P-value $$n = 51673202$$ ($62\%$)1965 ($38\%$) Social factors Age (years), mean (SD)25.7 (6.2)26.4 (5.9)24.5 (6.3)< 0.001 Age groups 18–19872 ($16.9\%$)374 ($11.7\%$)498 ($25.3\%$) 20–241824 ($35.3\%$)1091 ($34.1\%$)733 ($37.3\%$) 25–291068 ($20.7\%$)778 ($24.3\%$)290 ($14.8\%$) 30–34784 ($15.2\%$)563 ($17.6\%$)221 ($11.3\%$) 35–40619 ($12.0\%$)396 ($12.4\%$)223 ($11.4\%$) Missing a 0 0 0 Marital status < 0.001 Single2804 ($63.6\%$)1522 ($56.2\%$)1282 ($75.4\%$) Married1497 ($34\%$)1144 ($42.2\%$)353 ($20.8\%$) Divorced/widowed108 ($2.5\%$)43 ($1.6\%$)65 ($3.8\%$) Missing a 493 ($15.4\%$) 265 ($13.5\%$) 758 ($14.7\%$) Employment < 0.001 Employed1978 ($53.9\%$)1535 ($68.5\%$)443 ($31\%$) Students1032 ($28.1\%$)399 ($17.8\%$)633 ($44.3\%$) Unemployed658 ($17.9\%$)306 ($13.7\%$)352 ($24.7\%$) Missing a 962 ($30.0\%$) 537 ($27.3\%$) 1499 ($29.0\%$) Education level < 0.001 High school& below2326 ($54.0\%$)1468 ($55.3\%$)858 ($51.9\%$) College & above1984 ($46.0\%$)1189 ($44.7\%$)795 (48.1) Missing 545 ($17.0\%$) 312 ($15.9\%$) 857 ($16.6\%$) Family history of NCD < 0.001 No2228 ($44.0\%$)1459 ($46.5\%$)769 ($40.0\%$) Yes2830 ($56.0\%$)1677 ($53.5\%$)1153 ($60.0\%$) Missing a 66 ($2.1\%$) 43 ($2.2\%$) 1.9 ($2.1\%$) Behavioral factors Smoking Non-smoker:2628 ($66.9\%$)1170 ($49.0\%$)1458 ($94.8\%$)< 0.001 Current smoker:1299 ($33.1\%$)1219 ($51\%$)80 ($5.2\%$) Cigarette666 ($17.7\%$)643 ($28.5\%$)23 ($1.5\%$) Midwakh802 ($21.2\%$)779 ($34.3\%$)23 ($1.5\%$) Shisha791 ($21.2\%$)722 ($32.3\%$)69 ($4.6\%$) Missing a 813 ($25.4\%$) 427 ($21.7\%$) 1240 ($24.0\%$) Physical activity Metabolic Equivalent (minutes/week), mean (SD)5514.3 (6306.8)6456.4 (7003.2)4129.9 (4792.4)< 0.001 Low1894 ($80.9\%$)1119 ($80.3\%$)775 ($81.8\%$)< 0.001 Moderate189 ($8.1\%$)95 ($6.8\%$)94 ($9.9\%$) High258 ($11\%$)179 ($12.9\%$)79 ($8.3\%$) Missing a 1809 ($56.5\%$) 1017 ($51.8\%$) 2826 ($54.7\%$) Data is presented as mean values with standard deviation (SD) or frequency numbers (N) and percentages (%). P-values are derived from t-tests for continuous measures and chi-square tests for categorical variables aMissing numbers were not included in the column percentages Smoking was self-reported in $33.1\%$ of the study sample, including three different types of tobacco smoking; cigarette, shisha, and midwakh. There were more male smokers ($51\%$) than female smokers ($5.2\%$) ($P \leq 0.001$). In men, the prevalence of tobacco use was similar amongst the three types of tobacco smoking. However, smoking shisha was more common than smoking other types of tobacco amongst women. For physical activity, approximately $81\%$ were categorized as performing low-physical activity and $19\%$ as moderate-to-high physical activity. Men reported higher number of active minutes per week compared to females ($P \leq 0.001$) (Table 1). The mean values of cardiometabolic markers of the study sample are presented in Table 2. All biomarkers were significantly higher in men than women ($P \leq 0.001$) with an inverse in high-density lipoprotein (HDL). In the overall sample, obesity was estimated as $27.2\%$ (25.9–28.4), the age-adjusted prevalence was $26.5\%$ (25.2–27.7). Based on HbA1c analysis, $6.5\%$ of the sample had prediabetes and $1.9\%$ had diabetes. Fasting serum glucose yielded a prevalence of $17.8\%$ for prediabetes and $2.7\%$ for diabetes. Together, glycated hemoglobin, fasting blood glucose and self-reported diagnosis or medication identified prediabetes prevalence as $8.2\%$ and diabetes as $3.5\%$. The overall prevalence of dysglycemia (prediabetes and diabetes) was $12.5\%$ (11.7–13.25); and the age-adjusted prevalence was $11.7\%$ (10.8–12.7), as presented in Table 3.Table 2Mean values of cardiometabolic markers of the UAE Healthy Future Study participantsCardiometabolic markersOverall, Mean (SD)Men, Mean (SD)Women, Mean (SD) P-valueBody Mass Index (BMI), kg/m2 26.9 (6.3)27.7 (6.0)25.8 (6.6)< 0.001Waist – hip ratio0.82 (0.09)0.87 (0.07)0.77 (0.08)< 0.001Systolic blood pressure, mmHg126.0 (14.1)131.2 (13.1)117.8 (11.6)< 0.001Diastolic blood pressure, mmHg78.0 (77.7)80.3 (10.2)74.3 (8.6)< 0.001Low-density lipoprotein (LDL), mg/dL115.9 (34.0)122.0 (35.8)105.9 (28.1)< 0.001High-density lipoprotein (HDL), mg/dL48.4 (12.9)43.9 (10.6)55.7 (13.0)< 0.001Total cholesterol, mg/dL182.5 (36.0)185.8 (38.6)177.0 (30.3)< 0.001Triglycerides (TG), mg/dL103.9 (77.6)118.8 (86.0)79.3 (52.8)< 0.001Glycated hemoglobin (HbA1c), %5.26 (0.66)5.29 (0.71)5.21 (0.58)< 0.001Fasting blood glucose (FBG), mg/dL94.3 [25]96.4 (25.7)88.8 (22.3)< 0.001Data is presented as means (standard deviation). P-values are derived from t-testsTable 3Age-adjusted prevalence of cardiometabolic risk factors of the UAE Healthy Future Study participantsCRFsOverallMenWomen P value Obesity 26.5 (25.2–27.7)29.7 (28–31.4)21.6 (19.7–23.5)< 0.001 Dysglycemia 11.7 (10.8–12.7)14.0 (12.7–15.2)8.3 (7.0–9.6)< 0.001 Dyslipidemia 62.7 (61.3–64.0)68.0 (66.3–69.7)54.2 (52.0–56.5)< 0.001 Hypertension 22.4 (21.2–23.6)30.9 (29.2–32.6)9.2 (7.8–10.5)< 0.001 Central obesity 22.5 (21.3–23.8)29.6 (27.9–31.3)12.5 (10.9–14.0)< 0.001Data is presented as percentage ($95\%$ CI) Dyslipidemia was reported in $62.7\%$ (61.3–64.0) of the study sample, and it was higher in men ($68.0\%$ (66.3–69.7)) than women (($54.2\%$ (52.0–56.5)) ($P \leq 0.001$). The mean systolic and diastolic blood pressures were significantly different between men and women ($P \leq 0.001$). The overall age-adjusted prevalence of hypertension was observed in $22.4\%$ (21.2–23.6) of the sample; $30.9\%$ (29.2–32.6) in men and $9.2\%$ (7.8–10.5) in women. Finally, age-adjusted central obesity was $22.5\%$ (21.3–23.8). Nearly a third of men had an increased waist-to-hip ratio ($29.6\%$ (27.9–31.3)), while only $12.5\%$ (10.9–14.0) of women had central obesity. Table 3 summarizes the age-adjusted prevalence for the CRFs. The prevalence for each of the five CRFs are significantly different across age groups in men and women, as visualized in Supplementary Fig. 1. The age-adjusted distribution of the CRFs was assessed within the social and behavioral determinants in men and women; presented in Tables 4 and 5. In men, smokers had higher prevalence of obesity, dyslipidemia, and central obesity than non-smokers ($p \leq 0.05$). Men in the lower education group had higher obesity and hypertension cases than men within higher education groups. Whereas, unemployed men had higher dysglycemia than students. Additionally, single or divorced men tended to be more hypertensive than married men. Table 4Age-adjusted prevalence of CRFs by social and behavioral determinants in menMENObesityDysglycemiaDyslipidemiaHypertensionCentral obesity Marital Status Single/divorced29.1 (26.4–31.8)14.1 (12.1–16.1)68.3 (65.5–71.2)33.8 (31–36.6)a 27.8 (25.1–30.5) Married29.7 (26.4–32.9)15.8 (13.2–18.3)67.5 (64–70.9)28.6 (25.4–31.8)a 30.8 (27.4–34.1) Employment Status Unemployed29.1 (23.5–34.6)17.8 (13.1–17.7)a 68.2 (62.8–73.6)30.8 (25.5–36.1)27.0 (22.0–32.1) Employed29.8 (27.3–32.4)15.7 (13.7–17.7)68.2 (65.7–70.7)33.6 (31.1–36.0)36 (33.6–38.5) Student31.8 (26.3–37.3)9.4 (5.9–12.8)a 69.8 (64.8–74.8)26.2 (21.8–30.5)18.6 (14.7–22.4) Education level High School & below32.7 (30.2–35.2)a 15.7 (13.8–17.6)68.6 (66–71.2)33.7 (31.2–36.3)a 30.6 (28–33.1) College & above25.1 (22.6–27.7)a 13.7 (11.7–15.6)67.1 (64.3–70)28.6 (25.9–31.2)a 27.8 (25.1–30.5) Family history of NCDs No27.1 (24.7–29.6)a 12.1 (10.5–13.8)a 65.7 (63.2–68.3)a 25.7 (23.3–28.1)a 29.3 (26.7–31.9) Yes31.8 (29.5–34.1)a 15.6 (13.9–17.4)a 69.7 (67.4–72)a 35.1 (32.8–37.4)a 29.8 (27.5–32.1) Smoking Non-smoking27.4 (24.7–30)a 15.1 (13–17.2)63.6 (60.7–66.5)a 32.7 (30–35.5)26.4 (23.7–29.1)a Smoking31.2 (28.5–33.9)a 14.4 (12.4–16.4)71.1 (68.5–73.8)a 31.0 (28.3–33.6)31.9 (29.2–34.7)a Physical Activity Moderate/High25.2 (20–30.5)11 (7.3–14.6)66.8 (61–72.6)33.1 (27.5–38.8)22.9 (17.7–28) Low PA27.9 (25.2–30.6)12 (10.1–14)66.3 (63.4–69.2)30.8 (28.1–33.6)28.3 (25.5–31.1)Data is presented as percentages ($95\%$ CI). Percentages are derived from logistic regression analyses adjusting for age ahave significant difference in proportions of CRFs ($P \leq 0.05$)Table 5Age-adjusted prevalence of cardiometabolic risk factors by social and behavioral determinants in womenWOMENObesityDysglycemiaDyslipidemiaHypertensionCentral obesity Marital Status Single/divorced20.7 (18.5–23.0)7.6 (6.1–9.0)53.4 (50.6–56.2)9.8 (8.1–11.4)12.1 (10.2–13.9) Married22.3 (17.5–27.0)10.4 (7.1–13.8)60.6 (54.6–66.6)9.0 (5.9–12.1)12.6 (9.0–16.1) Employment Status Unemployed24.5 (19.8–29.1)a 11.3 (8–14.7)a 61.6 (56.4–66.8)11.5 (8.1–15.0)14.2 (10.5–17.9) Employed23.9 (19.2–28.5)10.1 (6.9–13.3)51.9 (46.3–57.6)9.8 (6.6–12.9)13.2 (9.6–16.8) Student17.7 (14.0–21.3)a 4.8 (2.8–6.7)a 53.6 (48.9–58.3)10.0 (7.0–12.9)10.6 (7.5–13.6) Education level High school & below25.9 (22.7–29.0)a 9.4 (7.4–11.5)a 59.0 (55.6–62.4)a 10.5 (8.4–12.7)14.3 (11.8–16.8)a College & above15.8 (13.1–18.4)a 6.7 (4.9–8.4)a 50.0 (46.4–53.7)a 8.8 (6.8–10.9)10.2 (8.1–12.3)a Family history of NCDs No18.2 (15.34–21.0)a 6.6 (4.8–8.3)a 50.3 (46.7–53.9)a 7.0 (5.1–8.8)a 11.3 (9.0–13.7) Yes24.2 (21.6–26.7)a 9.4 (7.7–11.1)a 57.2 (54.3–60.1)a 10.9 (9.0–12.8)a 13.2 (11.1–15.2) Smoking Non-smoking20.2 (18.1–22.4)7.6 (6.2–9.0)54.2 (51.6–56.8)9.5 (7.9–11.1)11.3 (9.6–13) Smoking28.9 (18.9–38.9)10.7 (4.3–17.2)57.3 (46.3–68.3)14.0 (6.5–21.5)13.9 (6.6–21.2) Physical Activity Moderate/High21.9 (15.5–28.3)9.0 (4.7–13.4)58.4 (50.9–65.8)10.7 (5.9–15.5)7.7 (3.7–11.8) Low PA19.3 (16.5–22.2)6.5 (4.7–8.3)52.4 (48.8–55.9)9.5 (7.4–11.7)10.2 (7.9–12.4)Data is presented as percentages ($95\%$ CI). Percentages are derived from logistic regression analyses adjusting for age ahave significant difference in proportions of CRFs ($P \leq 0.05$) In women, obesity and dysglycemia were significantly higher in the unemployed group compared to students ($p \leq 0.05$). Women in the lower education group reported significantly higher prevalence of all CRFs, except for hypertension, than those in the higher education one ($p \leq 0.05$). Interestingly, obesity, dysglycemia, dyslipidemia and hypertension were higher in the more the physically active group, however these findings were not statistically significant. All CRFs, with the exception of central obesity, across both sexes were significantly higher among participants with family history of NCDs ($P \leq 0.05$). ## Discussion This study presents the first comprehensive epidemiological assessment of the major CRFs in a large sample of young Emirati adults, including obesity, dysglycemia, dyslipidemia, hypertension and central obesity. All CRFs were highly prevalent across the whole sample, but significantly higher in men compared to women. This study investigated, for the first time, how CRFs prevalence differs among different social and behavioral determinants. Obesity was present in $26.5\%$ of our population. This estimate was similar to the prevalence reported in earlier work, where the prevalence estimates of obesity ranged from 25 to $35.4\%$ in similar age groups [4, 26–28]. In this study, obesity was higher in men than in women. This trend was similar to another nation-wide study published in 2012 [26]. A review led by Azizi et al. on the metabolic health status in the Middle East and North Africa (MENA) region projected a further increase in high BMI prevalence in 2025 to $36.3\%$ (25.0–48.5) in men and $47.8\%$ (37.1–58.9) in women [29]. The age-adjusted prevalence for pre-diabetes was $8.2\%$ and for diabetes was $3.5\%$ in the whole study population. These prevalence estimates were similar to the estimates reported by the UAE national survey for the age group 18–44 years; where diabetes had a prevalence of $3.3\%$ and prediabetes was $6.5\%$ among Emiratis [4]. The study findings showed that the age-adjusted prevalence of dysglycemia in this population was $11.7\%$ and it was higher in males ($14.0\%$) than in females ($8.3\%$). In this analysis, the prevalence of dysglycemia doubled from the youngest age group (below 20 years) to the oldest age group (35 to 40 years) ($p \leq 0.01$). It was found that $7.6\%$ of participants aged 18 and 19 years, and $8.3\%$ of participants between 20 and 24 years had abnormal glycemic status. This supports the international connotation that prediabetes and diabetes are rapidly rising in the adolescents and young adults as reported by the Centre for Disease Control and Prevention (CDC) [30]. According to the International Diabetes Federation (IDF), the age-adjusted prevalence of diabetes was $16.3\%$ in UAE, while it is $12.2\%$ in the Middle East and North Africa (MENA) region in 2019 [31]. The MENA region had the highest prevalence compared to other parts of the world. A recent analysis on 33,000 men in the UAE revealed a relatively higher prediabetes prevalence of $33\%$ in the 18–19-year-old age group, and $40.2\%$ in the 20–24-year-old age group based on fasting blood glucose measurements [27]. Projected estimates of 2025 state that diabetes will increase to $19.9\%$ (8.0–41.1) in men and women [29]. With the broad definition of dyslipidemia applied in this study, the results revealed that $62.7\%$ of the whole sample had abnormal lipid profiles. This high proportion of dyslipidemia might not be comparable to other local studies due to the difference in the definition criteria and methods of blood sampling; fasting or random [26, 27]. The global prevalence of dyslipidemia among adults was reported as high as $39\%$ in 2008 [32]. They showed that the prevalence of dyslipidemia was positively associated with the income of the country and estimates were double in high-income countries compared to low-income countries. Elevated blood pressure was identified in $22.4\%$ of the sample. Hypertension in men was 3-folds higher than in women, $30.9\%$ versus $9.2\%$, respectively. In men, hypertension was highest ($30.0\%$) in the 20–24 age group, whereas in women, the prevalence was highest ($25\%$) in the oldest age group; 35–40 years. In line with other reports, men consistently had a higher prevalence for hypertension than women [4, 26]. A global prevalence of $26.4\%$ was estimated among adults in year 2000 [33]. In the age and gender breakdown, hypertension was reported in $12.7\%$ among the 20–29-year-old age group and $18.4\%$ in the 30–39-year-old age group in men. Men had double the rates reported for women in all age groups. NCD trends in UAE show that hypertension prevalence decreases when compared to data from 1975 to 2015 [29]. A possible explanation to the reduction, despite unfavorable trends in sodium intake, obesity and physical inactivity, maybe be due to the use of antihypertensive drugs among other unknown factors. The prevalence of abdominal obesity in this study population was estimated as $24.3\%$, and males had a higher prevalence than women; 29.6 and $12.5\%$, respectively. However, these findings were lower than that of the Weqaya study, where the prevalence for abdominal obesity was $46.5\%$ in men and $36.4\%$ in women, aged 18–39 years [26]. In a smaller local study on young women aged 18–25 years, high waist circumference was detected in $18.2\%$ of the sample [34]. In the US, the National Health and Nutrition Examination Survey (NHANES) report of 2007–2010 estimated abdominal obesity in 18–39 year old age group as $38.7\%$ [35]. Global patterns of abdominal obesity show that women generally have higher prevalence than men [36]. The Weqaya study showed that, by stratifying by age and gender, women in the younger age-groups had lower rates of central obesity than men. However, in the sixty age-group, women shifted to have higher rates of central obesity than men [26]. This could be explained by the effect of menopause. The literature shows that central obesity is also associated with low levels of testosterone; a hormone that promotes fat metabolism and decreases central obesity [37, 38]. However, this pattern was not detected in our young sample below the age of 40 years. To exclude collinearity between obesity and central obesity, correlation tests between BMI and waist-to-hip ratio were carried out; the estimated correlation was 0.42. We found that among individuals that did not have central obesity, $18.5\%$ were BMI-obese. As for those that did have central obesity, only $54\%$ had BMI-obesity. In this study, we also investigated the CRFs distribution within different social characteristics, such as marital status, employment and educational attainment, as well as behavioral determinants, such as smoking and physical activity. A positive family history of NCDs was also investigated. In men, the prevalence of obesity, dyslipidemia and central obesity were higher among smokers. Although obesity is usually lower among smokers than non-smokers, this was not the case in our sample. Obesity was significantly higher in smokers $31.2\%$ (28.5–33.9) versus $27.4\%$ (24.7–30.0) in non-smokers. This finding is in accordance with Sulaiman et al. ’s report where smokers had higher BMI than non-smokers; $28.7\%$ vs. $20.7\%$ respectively [39]. Similarly, the Northern Finland Birth Cohort 1966 study sample showed that smokers had a higher BMI, waist circumference, dyslipidemia and hypertension when compared to non-smokers [40]. Furthermore, lower education attainment in both men and women showed significantly higher CRFs; 2 out of 5 the CRFs in men and 4 out of the 5 CRFs in women. Similarly, in a study that investigated the relationship between education and CVD incidence, those with higher education (of a university degree) had a smaller percentage of people with hypertension, BMI and diabetes compared to people with lower education ($P \leq 0.001$) [41]. Unemployed men and women had a higher prevalence of dysglycemia than in students. This finding can be supported by Rautio et al. ’s [42] conclusion that unemployment was related to prediabetes and diabetes. Both men and women with a positive family history of NCDs had significantly higher prevalence of CRFs compared to those with no family history ($P \leq 0.05$). It is well established that family history of disease and metabolic abnormality play a big role on offspring, due to the combination of both genetic and environmental factors [43]. There was no significant difference in the distribution of risk factors according to physical activity levels. This could be attributable to the fact that $81\%$ of the sample were classified as low physically active. This study used a broad definition for dyslipidemia, which was based on four lipid markers, self-report and the use of lipid-lowering medication. This definition was recommended by the ATP3 guidelines for persons above 20 years old [44]. We also used random non-fasting samples for the analysis, which is unconventional to normal practice. Traditionally, blood collection for lipid testing purposes is required to be fasting samples. However, recent reports show that random blood samples are acceptable. Observational studies demonstrate that in comparison to fasting level, measurements only altered minimally, by 8 mg/dL or 0.2 mmol/L, when compared to fasting lipid levels [45]. So far, there is no robust -scientific evidence to why fasting samples are better than random samples when evaluating lipid profile for cardiovascular risk prediction. In fact, most studies now recommend non-fasting samples as they are easier to collect during the day and represent the normal postprandial state of individuals. Many countries are now changing their guidelines towards a consensus on measuring lipid profiles for cardiovascular risk prediction in the non-fasting state to simplify blood sampling for patients, laboratories, and clinicians worldwide [46]. The principal strengths of this study include the large sample size of young Emiratis, and the extensive information collected. This study mainly focused on recruiting young adults, who are often underrepresented in other non-communicable disease studies. Another strength is the thorough process, the use of objective tools and the various data points, from sociodemographic, to lifestyle behaviors, health and family history. All blood samples and physical measurements were collected in a standardized procedure to ensure consistent quality and reduce the risk of information bias. All of these data points allowed us to employ detailed and specific disease-identification criteria. The main weakness of this study is that it is based on opportunistic recruitment of study volunteers. This might introduce the risk of having selection bias and potentially affect the representativeness of the study sample. Another limitation observed is that more males ($62\%$) were recruited than females ($38\%$), and that they were recruited from different centers. The analysis of the results therefore varied and were described separately for each gender. Moreover, it is essential to address a major limitation related to cross-sectional studies, which is the inability to identify a causal relationship between the potential risk factor and outcome. Therefore, the results of this study must be interpreted cautiously and inferred to the local populations of similar age and sociodemographic characteristics. Similar to other observational studies, this study is prone to measurement and recall bias. Moreover, the high number of missing data for physical activity have possibly affected the capability to capture a relationship to CRFs in the model. Although validation studies conclude that generally GPAQ is an acceptable measure of physical activity; results ranged between fair-to-moderate validity [47, 48]. However, it does not adequately assess sedentary behavior. Sedentary behavior is not synonymous with physical inactivity. An individual can be physically active, and have long hours of sedentary behavior [49]. Therefore, it is important to address sedentary behavior independently from physical activity. Finally, the lack of dietary data, which is another important behavioral risk factor that is known to affect cardiometabolic health, was another limiting factor to the study. ## Conclusion This study on cardiometabolic risk factors provided thorough information about the cardiovascular risk in young adults of the United Arab Emirates, which represent the majority age demographic of the country, where $95\%$ of the UAE population is younger than 40 years. This study suggests that the prevalence of obesity, dysglycemia, dyslipidemia, hypertension and central obesity are high. The study showed variation in the distribution of CRFs by social and behavioral characteristics. Understanding that some social groups are more prone for developing a metabolic abnormality can help design specific prevention measures towards them. ## Supplementary Information Additional file 1: Supplementary Fig. 1. Cardiometabolic risk factors across age groups of the UAE Healthy Future Study Participants. ## References 1. Kyu HH. Global, regional, and national disability-adjusted life-years (DALYs) for 359 diseases and injuries and healthy life expectancy (HALE) for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet (2018.0) 392 1859-1922. DOI: 10.1016/S0140-6736(18)32335-3 2. Roth GA. Global Burden of Cardiovascular Diseases and Risk Factors, 1990–2019: Update From the GBD 2019 Study. J Am Coll Cardiol (2020.0) 76 2982-3021. DOI: 10.1016/j.jacc.2020.11.010 3. 3.World Health Organization. (2018). Noncommunicable diseases country profiles 2018. World Health Organization. https://apps.who.int/iris/handle/10665/274512. License: CC BY-NC-SA 3.0 IGO 4. 4.Ministry of Health and Prevention, UAE National Health Survey Report 2017–2018. Statistics & research center (SARC). (UAE 2019). 5. Yusuf S. Effect of potentially modifiable risk factors associated with myocardial infarction in 52 countries (the INTERHEART study): case-control study. Lancet (London, England) (2004.0) 364 937-952. DOI: 10.1016/s0140-6736(04)17018-9 6. Khot UN. Prevalence of conventional risk factors in patients with coronary heart disease. JAMA (2003.0) 290 898-904. DOI: 10.1001/jama.290.7.898 7. Yusufali A. Opportunistic screening for CVD risk factors: the Dubai shopping for cardiovascular risk study (DISCOVERY). Glob Heart (2015.0) 10 265-272. DOI: 10.1016/j.gheart.2015.04.008 8. Pletcher MJ, Vittinghoff E, Thanataveerat A, Bibbins-Domingo K, Moran AE. Young adult exposure to cardiovascular risk factors and risk of events later in life: the Framingham offspring study. PLoS One (2016.0) 11 e0154288. DOI: 10.1371/journal.pone.0154288 9. Lakier JB. Smoking and cardiovascular disease. Am J Med (1992.0) 93 8s-12s. DOI: 10.1016/0002-9343(92)90620-q 10. Banks E. Tobacco smoking and risk of 36 cardiovascular disease subtypes: fatal and non-fatal outcomes in a large prospective Australian study. BMC Med (2019.0) 17 128. DOI: 10.1186/s12916-019-1351-4 11. Lavie CJ, Ozemek C, Carbone S, Katzmarzyk PT, Blair SN. Sedentary Behavior, Exercise, and Cardiovascular Health. Circulation Res (2019.0) 124 799-815. DOI: 10.1161/CIRCRESAHA.118.312669 12. 12.Casas R, Castro-Barquero S, Estruch R, Sacanella E. Nutrition and Cardiovascular Health. Int J Mol Sci. 2018;19. 10.3390/ijms19123988. 13. Isomaa B. Cardiovascular morbidity and mortality associated with the metabolic syndrome. Diabetes Care (2001.0) 24 683-689. DOI: 10.2337/diacare.24.4.683 14. Le F, Ahern J, Galea S. Neighborhood education inequality and drinking behavior. Drug Alcohol Depend (2010.0) 112 18-26. DOI: 10.1016/j.drugalcdep.2010.05.005 15. de Walque D. Does education affect smoking behaviors? Evidence using the Vietnam draft as an instrument for college education. J Health Econ (2007.0) 26 877-895. DOI: 10.1016/j.jhealeco.2006.12.005 16. Kivimaki M. Work stress in the etiology of coronary heart disease--a meta-analysis. Scand J Work Environ Health (2006.0) 32 431-442. DOI: 10.5271/sjweh.1049 17. Wong CW. Marital status and risk of cardiovascular diseases: a systematic review and meta-analysis. Heart (British Cardiac Society) (2018.0) 104 1937-1948. DOI: 10.1136/heartjnl-2018-313005 18. 18.World Health FederationRisk Factors Fact Sheet2017. Risk Factors Fact Sheet (2017.0) 19. Abdulle A. The UAE healthy future study: a pilot for a prospective cohort study of 20,000 United Arab Emirates nationals. BMC Public Health (2018.0) 18 101. DOI: 10.1186/s12889-017-5012-2 20. 20.WHO. Global Physical Activity Questionnaire (GPAQ): WHO STEPwise approach to NCD risk factor surveillance. Geneva: World Health Organization. https://www.who.int/publications/m/item/global-physical-activity-questionnaire. 21. Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III) Final Report. Circulation (2002.0) 106 3143. DOI: 10.1161/circ.106.25.3143 22. Lipid Modification: Cardiovascular Risk Assessment and the Modification of Blood Lipids for the Primary and Secondary Prevention of Cardiovascular Disease. Royal College of General Practitioners (UK) Royal College of General Practitioners (2008.0) 23. Whelton PK. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA Guideline for the Prevention, Detection, Evaluation, and Management of High Blood Pressure in Adults: Executive Summary: A Report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Hypertension (Dallas, Tex 1979) (2018.0) 71 1269-1324. DOI: 10.1161/hyp.0000000000000066 24. 24.World Health OrganizationWaist Cicumference and waist-hip ratio2008GenevaWorld Health Organization. Waist Cicumference and waist-hip ratio (2008.0) 25. 25.Stata Statistical SoftwareRelease 152017College StattionStataCorp. Release 15 (2017.0) 26. Hajat C, Harrison O, Al Siksek Z. Weqaya: a population-wide cardiovascular screening program in Abu Dhabi, United Arab Emirates. Am J Public Health (2012.0) 102 909-914. DOI: 10.2105/ajph.2011.300290 27. Alzaabi A, Al-Kaabi J, Al-Maskari F, Farhood AF, Ahmed LA. Prevalence of diabetes and cardio-metabolic risk factors in young men in the United Arab Emirates: a cross-sectional national survey. Endocrinol Diabetes Metab (2019.0) 2 e00081. DOI: 10.1002/edm2.81 28. 28.Radaideh G, et al. Cardiovascular Risk Factor Burden in the United Arab Emirates (UAE): The Africa Middle East (AfME) Cardiovascular Epidemiological (ACE) Study Sub-analysis. Int Cardiovasc Forum J. 2017;11. https://doi.org/10.17987/icfj.v11i0.414%J. 29. Azizi F. Metabolic health in the Middle East and north Africa. Lancet Diabetes Endocrinol (2019.0) 7 866-879. DOI: 10.1016/S2213-8587(19)30179-2 30. 30.CDC. Prediabetes: An emerging health threat can lead to type 2 diabetes (ed CDC). USA: Centers for Disease Control and Prevention; 2019. 31. 31.International Diabetes FederationInternational Diabetes Federation2015. International Diabetes Federation (2015.0) 32. 32.WHOGlobal Health Observatory (GHO) data2014. Global Health Observatory (GHO) data (2014.0) 33. Kearney PM. Global burden of hypertension: analysis of worldwide data. Lancet (London, England) (2005.0) 365 217-223. DOI: 10.1016/s0140-6736(05)17741-1 34. Al Dhaheri AS. A cross-sectional study of the prevalence of metabolic syndrome among young female Emirati adults. PLoS One (2016.0) 11 e0159378. DOI: 10.1371/journal.pone.0159378 35. Ostchega Y, Hughes JP, Terry A, Fakhouri THI, Miller I. Abdominal Obesity, Body Mass Index, and Hypertension in US Adults: NHANES 2007–2010. Am J Hypertens (2012.0) 25 1271-1278. DOI: 10.1038/ajh.2012.120%J 36. Wong MCS. Global, regional and time-trend prevalence of central obesity: a systematic review and meta-analysis of 13.2 million subjects. Eur J Epidemiol (2020.0) 35 673-683. DOI: 10.1007/s10654-020-00650-3 37. Laaksonen DE. Sex hormones, inflammation and the metabolic syndrome: a population-based study. Eur J Endocrinol (2003.0) 149 601-608. DOI: 10.1530/eje.0.1490601 38. Schunkert H, Hense HW, Andus T, Riegger GA, Straub RH. Relation between dehydroepiandrosterone sulfate and blood pressure levels in a population-based sample. Am J Hypertens (1999.0) 12 1140-1143. DOI: 10.1016/s0895-7061(99)00128-4 39. Sulaiman N. Prevalence of overweight and obesity in United Arab Emirates expatriates: the UAE National Diabetes and lifestyle study. Diabetol Metab Syndr (2017.0) 9 88. DOI: 10.1186/s13098-017-0287-0 40. Keto J. Cardiovascular disease risk factors in relation to smoking behaviour and history: a population-based cohort study. Open Heart (2016.0) 3 e000358. DOI: 10.1136/openhrt-2015-000358 41. Dégano IR. The association between education and cardiovascular disease incidence is mediated by hypertension, diabetes, and body mass index. Sci Rep (2017.0) 7 12370. DOI: 10.1038/s41598-017-10775-3 42. Rautio N. Accumulated exposure to unemployment is related to impaired glucose metabolism in middle-aged men: a follow-up of the northern Finland birth cohort 1966. Prim Care Diabetes (2017.0) 11 365-372. DOI: 10.1016/j.pcd.2017.03.010 43. Harrison TA. Family history of diabetes as a potential public health tool. Am J Prev Med (2003.0) 24 152-159. DOI: 10.1016/s0749-3797(02)00588-3 44. 44.Chou R, D. T, Blazina I. Screening for Dyslipidemia in Younger Adults: A Systematic Review to Update the 2008 U. S, https://www.ncbi.nlm.nih.gov/books/NBK396239/?report=classic (2016). 45. Nordestgaard BG. Fasting is not routinely required for determination of a lipid profile: clinical and laboratory implications including flagging at desirable concentration cut-points—a joint consensus statement from the European Atherosclerosis Society and European Federation of Clinical Chemistry and Laboratory Medicine. Eur Heart J (2016.0) 37 1944-1958. DOI: 10.1093/eurheartj/ehw152%J 46. Langsted A, Nordestgaard BG. Nonfasting versus fasting lipid profile for cardiovascular risk prediction. Pathology (2019.0) 51 131-141. DOI: 10.1016/j.pathol.2018.09.062 47. Wanner M. Validation of the global physical activity questionnaire for self-administration in a European context. BMJ Open Sport Exerc Med (2017.0) 3 e000206. DOI: 10.1136/bmjsem-2016-000206 48. Doyle C, Khan A, Burton N. Reliability and validity of a self-administered Arabic version of the global physical activity questionnaire (GPAQ-A). J Sports Med Phys Fitness (2019.0) 59 1221-1228. DOI: 10.23736/s0022-4707.18.09186-7 49. 49.Winzer EB, Woitek F, Linke A. Physical Activity in the Prevention and Treatment of Coronary Artery Disease. J Am Heart Assoc. 2018;7. 10.1161/jaha.117.007725.
--- title: 'NEMoE: a nutrition aware regularized mixture of experts model to identify heterogeneous diet-microbiome-host health interactions' authors: - Xiangnan Xu - Michal Lubomski - Andrew J. Holmes - Carolyn M. Sue - Ryan L. Davis - Samuel Muller - Jean Y. H. Yang journal: Microbiome year: 2023 pmcid: PMC10015776 doi: 10.1186/s40168-023-01475-4 license: CC BY 4.0 --- # NEMoE: a nutrition aware regularized mixture of experts model to identify heterogeneous diet-microbiome-host health interactions ## Abstract ### Background Unrevealing the interplay between diet, the microbiome, and the health state could enable the design of personalized intervention strategies and improve the health and well-being of individuals. A common approach to this is to divide the study population into smaller cohorts based on dietary preferences in the hope of identifying specific microbial signatures. However, classification of patients based solely on diet is unlikely to reflect the microbiome-host health relationship or the taxonomic microbiome makeup. ### Results We present a novel approach, the Nutrition-Ecotype Mixture of Experts (NEMoE) model, for establishing associations between gut microbiota and health state that accounts for diet-specific cohort variability using a regularized mixture of experts model framework with an integrated parameter sharing strategy to ensure data-driven diet-cohort identification consistency across taxonomic levels. The success of our approach was demonstrated through a series of simulation studies, in which NEMoE showed robustness with regard to parameter selection and varying degrees of data heterogeneity. Further application to real-world microbiome data from a Parkinson’s disease cohort revealed that NEMoE is capable of not only improving predictive performance for Parkinson’s Disease but also for identifying diet-specific microbial signatures of disease. ### Conclusion In summary, NEMoE can be used to uncover diet-specific relationships between nutritional-ecotype and patient health and to contextualize precision nutrition for different diseases. Video Abstract ### Supplementary Information The online version contains supplementary material available at 10.1186/s40168-023-01475-4. ## Background The human body is home to complex microbial communities, collectively known as the microbiome, which is mostly made up of prokaryotes (bacteria) and archaea [1]. Considerable evidence has emerged indicating that the microbiome is an important contributor to an individual’s health [2]. This has been illustrated by links between the gut microbiome and numerous diseases, including irritable bowel syndrome [3], Crohn’s disease [4], type 2 diabetes [5], cardiovascular disease [6], and Parkinson’s disease (PD) [7]. The gut microbiome is known to change throughout our lives as a result of various environmental influences. Diet, being one of these factors, has the greatest known long-term interaction with the gut microbiome [8]. Thus, a deep understanding of the relationship between diet and the gut microbiome and the consequential impact on disease processes holds promise for developing personalized dietary intervention strategies to modulate and maintain a healthy microbiome population [9, 10]. Diet has a direct impact on the microbial community in the gut, which governs the activity of the intestinal ecosystem and can have considerable implications for an individual’s health [11, 12]. This is conceptualized in Fig. 1 where, for illustration purposes, the macronutrient intake is separated into three perfectly distinct subcohorts with different associations between microbiome composition and PD. In practice, several studies have demonstrated that variations in nutrient intake, such as different ratios of protein, carbohydrate [13], or dietary fiber [14] intake, can influence the host-microbiome association. These discoveries are generally based on an elaborate experimental design using model organisms [13] or dietary interventions [15–17]. Recent observational studies suggest that long-term diets could be associated with the microbiome [18], and this can further affect overall health. In a similar context, our recent study of the gut microbiome in PD showed that when partitioning individuals based on carbohydrate intake, the predictive performance of the microbiota profile to indicate PD was increased [19, 20]. Together, these studies suggest that dietary differences can impact relationships between microbiome composition and host health/disease status. Fig. 1Illustration of NEMoE: a, b The input matrix of NEMoE: n samples with q nutrient features and p microbial features. c A conceptual workflow of NEMoE, where the joint optimization is achieved by EM algorithm to maximize the regularized likelihood function. d A toy example showing a nutritional-ecotype in the microbiome PD study. The nutrient intake is clustered into K latent classes. e In each latent class, the microbial signatures of PD are different, which is reflected by the coefficients in the experts network To uncover complex heterogeneous relationship structure between diet, microbiome, and host health, it is important to identify homogeneous subcohort or latent structure in data that can be explained by a set of features. This is similar to the concept of “ecotype”, which is commonly used to refer to a variant which has observable phenotypically difference in a local environment [21]. Hence, using a data-driven approach, it is able to divide a population into multiple subcohorts with distinct microbiological signatures for health that can be best described by nutrient combinations, resulting in what we term “nutritional-ecotypes.” These subcohorts can be thought of as diet-based latent classes where they capture interaction between the constraints imposed by nutrient intake of individuals on the community dynamics of their microbiomes [22, 23]. Methods to discover such diet-based latent classes could be hypothesis-driven based on prior knowledge [24, 25] or guided by an unsupervised statistical learning method, such as clustering [26], followed by latent class analysis [27]. Although these methods identify nutrient-classes with an altered overall nutritional profile, one limitation is that the defined cohorts may not reflect the heterogeneous microbiome-host health relationship: the drivers of “diet x microbiome” outcomes, “diet x host” outcomes, and “host x microbiome outcomes” are overlapping, but not perfectly congruent. Consequently, classification models built within a subcohort defined just by diet (or microbiome) will not necessarily improve prediction of the health/disease state [28]. Similar concepts of identifying cohort heterogeneity to improve prediction performance have been developed in other omics settings and for other diseases [29, 30]. However, simple adaptations of methodologies developed for other omics platforms remain challenging as these do not account for the hierarchical taxonomic structure observed in the study of the diet-microbiome-host interaction. That is, each individual should be in the same diet-specific cohort across all taxonomic levels to keep hierarchical fidelity of the microbial community, i.e., a consistent nutrition class across Phylum, Class, Family, Genus, etc. To this end, we propose a novel Nutritional-Ecotype Mixture of Experts (NEMoE) approach for uncovering associations between the gut microbiome profile and the health state of an individual that takes into account diet-specific cohort heterogeneity (Fig. 1 and Supplementary Fig. 1 and 2). This is achieved by using a regularized mixture of experts model to simultaneously optimize the separations between nutritional-ecotypes, classification performance of microbiota, and the health state. The mixture of experts models has been widely used in integrating different types of data. Kim and colleagues [31] have used it for combining clinical data and genomics data. However, this work does not use sparse regularization and lacks interpretability, i.e., unable to identify unique markers in each experts network. NEMoE also integrates a model parameter sharing strategy to account for the taxonomic information contained in microbiome data, ensuring coherent nutritional classification is maintained across all taxonomic levels. We show through empirical computational simulation research that NEMoE is robust to parameter changes. We also apply NEMoE to real microbiome data from a PD cohort and show that the model outperforms existing approaches of predictive performance and is able to uncover candidate diet-specific microbiome markers of complex disease. ## NEMoE, a novel method for jointly identifying nutritional-ecotype and for modeling the relationship between microbiota and health state NEMoE identifies nutritional-ecotypes that represent differential dietary intake as well as the relationship between microbiome structure and host health (Fig. 1 and Supplementary Fig. 1). This approach has two distinct components: first, a gating network aimed at estimating latent classes shaped by nutritional intake, and second, an experts network aimed at modeling the relationship between the microbiota composition and the health state within each latent class [31, 32]. The input of the gating network is a nutrition matrix, with each variable being the nutrients intake of the individual and the corresponding microbiome measurements are used as input of the experts network. Similar to non-regularized mixture of experts (MoE) models, fitting NEMoE involves estimating the parameters via maximum likelihood estimation to simultaneously optimize the separations among nutritional-ecotypes, microbiome classification performance, and the health state (Supplementary Fig. 2). The optimization procedure is usually achieved by an expectation maximization (EM) algorithm. However, the MoE model does not extend to a large number of feature variables (p) and small sample size (n) framework, which often occurs in diet and microbiome data where there are many more features than observations. Instead, NEMoE adopts a regularization component to the MoE (RMoE [33]) by adding elastic net penalties [34] on both the gating function and the experts network (details in the Methods section). Next, NEMoE employs a parameter sharing strategy that involves a shared gating network for the microbiome relative abundance matrices across taxonomic levels, to ensure coherent latent classes across all taxonomic levels. Compared with a latent class using purely nutritional intake, our nutritional-ecotype has two advantages: (i) it takes the relationship between microbiome and health outcome into account and is beneficial for identifying diet-specific microbial signatures (Supplementary Fig. 1). ( ii) It incorporates the taxonomic structure in the latent class and keeps hierarchical fidelity of the microbial community. ## NEMoE is able to accurately identify nutritional latent classes shared across different taxonomic levels We evaluated the efficiency of NEMoE in determining nutritional-ecotypes based on microbiota across different taxonomic levels using both simulated and experimental data. In our simulation study (see Supplementary Notes), we created a four-level dataset of 500 samples with shared latent structure, where each individual belonged to a nutritional-ecotype and the relationship between microbiota and health status differed between two simulated nutritional-ecotypes. The adjusted rand index (ARI), a cluster comparison statistic, was used to compare the estimated nutritional-ecotypes and the underlying simulated latent classes (Fig. 2a). We discovered that by incorporating hierarchical taxonomy information in our NEMoE approach, the estimated nutritional class was cohesive and performed better (higher ARI = 0.80) than nutritional-ecotypes estimated from a single taxonomy level (ARI = 0.75). NEMoE achieved this by sharing information across taxonomic levels and the estimated latent class incorporated information from all levels. Fig. 2Identification of nutritional-ecotype by NEMoE. a Boxplot comparing NEMoE and single-level NEMoE in estimating shared latent classes. The ARI (x-axis) is calculated by comparing the estimated latent class and the true latent class from the data-generating model. In all settings, NEMoE using multiple-level information performs better. b PCA plot of scaled nutrient intake for subjects colored by the two nutrition classes as estimated by NEMoE. Estimated coefficients of the gating network showed high coefficients for sugar, protein:carbohydrate, and moisture. We denote the two nutrition classes as prot-CARB and PROT-carb with low protein-high carbohydrate intake and vice versa. c Scatter plot of genera Fusicatenibacter and Anaerostipes. Left panel shows that Parkinson’s disease and healthy controls in the prot-CARB subcohort roughly separate but there is no such separation in the PROT-carb right panel. d Scatter plot of genera Erysipelotrichaceae UCG-003 and [Ruminococcus] torques group showed a different relationship between Parkinson’s Disease and Healthy Controls in two nutritional-ecotypes Next, we applied NEMoE to our in-house data from a gut microbiome PD study. A scatter plot from the first two components of a principal component analysis (PCA) of scaled nutrient intake (see Methods section) from all individuals is shown in Fig. 2b, with the two nutritional-ecotypes best described as “high protein”–“low carbohydrate” (PROT-carb; shown in red) and “low protein”–“high carbohydrate” (prot-CARB; blue). The corresponding loadings show that these two ecotypes have very different ratios of protein to carbohydrate intake: Sugars and %EC (percentage of energy intake as carbohydrate) showed negative coefficient (γ < 0); P:C, Moisture and %EP (percentage of energy intake as protein) showed a positive coefficient of the gating network (γ > 0). Based on the meaning of these variables, we described the groups as “PROT-carb” and “prot-CARB,” with capital letters indicating the variable with a positive coefficient. Figure 2c and d illustrate that the relationships between gut microbiota and PD status are different between these two nutrition-ecotypes, PROT-carb, and prot-CARB. It is important to note that two identified subcohorts are significantly different to clusters identified by unsupervised clustering, such as subcohorts estimated by the k-means algorithm (ARI ~ 0, Supplementary Fig. 3). We further established the generalizability of NEMoE by examining its impact when applied to data with different levels of heterogeneity. Here, we created synthetic datasets with four different degrees of separation (Fig. 3a, b and Supplementary Notes) and demonstrated that NEMoE performs better than other existing approaches in detecting latent classes and this difference was more evident in challenging situations where the true separation between latent classes was small (Supplementary Fig. 4). This implies that NEMoE has potential to perform well in many observational studies where nutrient intake patterns are mixed or difficult to separate, and hence the NEMoE approach can be applied broadly to human disease datasets with diverse dietary intake. Fig. 3Comparison of NEMoE on simulation dataset and real dataset. a An illustration of a non-separable case where nutrition intake does not show a difference between two nutritional-ecotypes, but each subcohort shows a different relationship between microbiome taxa and health state. b An illustration of a separable case where nutrition intake is significantly different between two nutritional-ecotypes and relationships in each model are similar to the illustration in a. Simulation studies showed that NEMoE can identify both case a and case b. c Receiver operating characteristics curve of different methods (see Table 1) in predicting Parkinson’s disease using LOOCV. NEMoE showed the best LOOCV-AUC (AUC = 0.78). d ROC plot of NEMoE at different taxonomic levels using LOOCV. Genus level showed the best predictive performance (AUC = 0.78) ## NEMoE outperforms existing supervised methods in predicting Parkinson’s disease state We evaluated the predictive performance of NEMoE using both simulation and real data based on leave-one-out cross-validation (LOOCV; see Supplementary Notes) to the area under the receiver operating characteristics curve (AUC) for the various models described in Table 1. In simulation studies, we showed that under all comparison settings, NEMoE was able to achieve higher prediction accuracy (Supplementary Fig. 4), which implies NEMoE is robust to different parameter settings, such as n and p. Figure 3c highlights that when NEMoE was applied to our in-house dataset from a gut microbiome PD study [20] with 2 latent classes (AUC = 0.78), it outperformed all other approaches, with the next best being random forest (AUC = 0.71). Supplementary Fig. 6 further highlights that increasing the number of latent classes for this data did not improve the overall AUC.Table 1Summary of methods for comparisonMethodInput data of identified subcohortInput data of modeling within each subcohortModelsLRMicrobiomeSparse logistic regressionSVMMicrobiomeSupport vector machineRFMicrobiomeRandom forestsLR KaNutritionMicrobiomeTwo-stage sLR with K latent classSVM IINutritionMicrobiomeTwo-stage SVMRF IINutritionMicrobiomeTwo-stage RFNEMoE KbNutritionMicrobiomeNEMoE with K latent classMMMoEcMicrobiomeMicrobiomeRMoENNMoENutritionNutritionRMoEMNMoEMicrobiomeNutritionRMoEComb-MoEMicrobiome+nutritionMicrobiome+nutritionRMoEaTwo stage sparse logistic regression fitted with two, three four latent classes were denoted as sLR II, sLR III, and sLR IVbNEMoE fitted with two, three four latent classes were denoted as NEMoE II, NEMoE III, and NEMoE IV. When not explicitly including the number of latent classes, we refer to NEMoE IIcOur NEMoE is easy to extend to partition the population with different types of data. We also investigate the different types of data as input of the NEMoE model. Results showed using nutrition to split the population obtained the best performance in our dataset NEMoE’s ability to detect meaningful subcohorts via its joint optimization approach is a key driver of this increase in accuracy. For example, when comparing to a naive two-stage model that uses unsupervised clustering to identify latent classes before fitting two independent models, the performance of NEMoE is considerably better, as indicated by the large difference in AUC (NEMoE = 0.78, sLR II= 0.6). We further assessed NEMoE’s capabilities on enterotype-separated subcohorts [35] within our PD dataset. Enterotype, a widely used concept in microbiome research, refers to the categorization of an individual’s microbiomes by the variance in composition [2, 36]. It is widely accepted that enterotype captures stable compositional features of individuals and differences in community-type prevalence across populations with different long-term diets. In this study, we classify 87 samples as Enterotype B, 81 samples as Enterotype F, and no samples as Enterotype P. The cluster memberships between the subcohorts determined by NEMoE and by enterotype had no more overlap than pure chance (ARI = 0). Furthermore, building a different classifier for each of the two enterotypes had a much lower (LOOCV-AUC = 0.65) predictive ability than NEMoE (LOOCV-AUC = 0.78). This suggests that NEMoE allows the model to focus more on each latent class and increases prediction performance by more precisely identifying subcohorts with differential microbiome-PD relationships. ## Identification of informative taxonomic levels and consensus candidate microbial PD signatures in multiple independent cohorts In our in-house gut microbiome PD investigation, NEMoE provided a natural criterion to examine which of the taxonomic levels (Phylum, Order, Family, Genus, and ASV) was most informative with respect to different nutrient intakes. We achieved this by evaluating predictive performance for PD at each taxonomic level to determine the most informative. Figure 3c shows that genus was most predictive compared to the other taxonomic levels, with an LOOCV-AUC of 0.78. Next, our NEMoE model determined a separate set of PD microbial signatures for each nutritional-ecotype. The derived coefficients represent the level of association between microbiota and health/disease state in each nutritional-ecotype (Fig. 4a and b) and results for all taxa are given in Supplementary Data 1. We can broadly group the microbiota taxa into five categories based on their coefficient estimates: (i) significant in both classes with different directions; (ii) significant in both classes with the same direction; (iii) significant in prot-CARB only, (iv) significant in PROT-carb only and (v) not-significant in both classes. The first category “significant in both classes with different directions” represents consistent abundance changes in both nutritional-ecotypes (Fig. 4b). It was noted that the genera Fusicatenibacter and Blautia showed consistent negative coefficients in both PROT-carb and prot-CARB nutritional-ecotypes. Such genera may be considered stable PD microbial signatures, with several studies showing their underrepresentation in PD. [ 19, 20, 38–42]Fig. 4Results of NEMoE on gut microbiome-PD study. a Coefficients of experts network in NEMoE at different taxonomic levels. The two latent classes showed distinctly different microbiome patterns. b Identification of diet-specific microbial signatures of PD. The “Same direction” class showed consistent function in different dietary patterns. The “PROT-carb only” and “prot-CARB only” classes tended to be important only with specific dietary intake. The “Different direction” class changed their coefficients in different dietary patterns. c Validation of differential relative abundance of genus Fusicatenibacter in 11 different datasets. With the exception of one dataset (Jin et al. [ 37]) all other datasets showed decreasing Fusicatenibacter in PD. d Forest plot of $95\%$ confidence interval of selected taxa showed NEMoE is able to identify the species that are differentially represented in specific nutritional-ecotypes The underrepresentation of Fusicatenibacter and Blautia was further validated using data from eight independent PD microbiome studies (Table 2). We processed the publicly available datasets using the dada2 pipeline [49](v1.16) and taxonomy reference “silva 138” [48, 50]. The relative abundance changes of the genus Fusicatenibacter were examined across all datasets, as shown in Fig. 4c. In all but one dataset [37], Fusicatenibacter had significantly lower relative abundance among PD individuals. Similar results were observed for Blautia (Supplementary Fig. 5), verifying NEMoE’s ability to identify consensus microbial signatures of PD in multiple independent cohorts. Table 2Summary of eight publicly available Parkinson’s disease microbiome studies used for validation of the NEMoE modelStudyDesignCountrySample sizeSamplingDNA extraction16S regionENA Accession NumberLubomski_0 [19, 39]Lubomski_6Lubomski_12LongitudinalAustralia74PD, 74HCHome collection, stored at −80 °CMP Biomedicals FastDNATM SPIN KitV3-V4PRJNA808166Wallen_1 [36]Cross-sectionalUSA323PD, 184HCHome collection, swabs, stored at −20 °CMoBio PowerSoil DNA Isolation KitV4PRJNA601994Wallen_2 [36, 43]Cross-sectionalUSA197PD, 130HCSwabs, delivered at RTMoBio PowerMag Soil kitV4PRJNA601994Aho (baseline) [44]Aho (follow-up)LongitudinalFinland64PD, 64HCHome collection, DNA stabilizer, stored in fridgePSP-Spin Stool KitV3-V4PRJEB27564Weis [45]Cross-sectionalGermany34PD, 25HCMED AUXIL fecal collector setFastDNA Spin KitV4-V5PRJEB30615Pietrucci [46]Cross-sectionalItaly80PD, 72HCHome collection, DNA stabilizerPSP-Spin Stool KitV3-V4PRJNA510730Scheperjans [47]Cross-sectionalFinland72PD, 72HCHome collection, DNA stabilizer, stored in fridgePSP-Spin Stool KitV1-V3PRJEB4927Jin [48]Cross-sectionalChina72PD, 68HCNANAV3-V4PRJEB588834Studies Lubomski_0, Lubomski_6, and Lubomski_12 were part of the same longitudinal data set by Lubomski and colleagues [2] and they represent samples that were measured at 0, 6, and 12 months, respectivelyStudies Aho (baseline) and Aho (follow-up) were part of the same longitudinal data set by Aho and colleagues [44]. The same subjects were measured twice, at baseline and then later at follow-up, which was on average 2.25 years apartStudies Wallen_1 and Wallen_2 were part of two large cohort studies set by Wallen and colleagues [38] ## Identification of the microbiome that are differentially represented in specific nutritional classes We note that taxa categories (i)–(iii) represent differential abundance changes that are unique in the two nutritional-ecotypes prot-CARB and PROT-carb, which indicate some microbial signatures of PD are diet-specific (Fig. 4c). We discovered that the genus Escherichia-Shigella was significantly underrepresented in the prot-CARB nutritional-ecotype but not in the PROT-carb ecotype. This genus belongs to the family Enterobacteriaceae (including E. coli, Shigella, Salmonella, and Klebsiella), which are facultative anaerobes and known for utilizing soluble sugars as a carbon source. When an individual’s diet has a higher intake of sugars (or simple starch) it can be expected that the relative abundance of these microbiota will likely increase. Recent studies found that Escherichia-Shigella is a pathogenic bacteria that potentially reduces short-chain fatty acid production and produces endotoxins and neurotoxins [51, 52]. We also found a significant increase in the relative abundance of the genus Akkermansia, but only in the PROT-carb class (Fig. 4d). These bacteria are known to impact immune response and constipation, with many studies reporting an overrepresentation in PD [39, 40, 42, 53]. Akkermansia breaks down mucins and turns them into short-chain fatty acids; further, their relative abundance is thought to increase when “diet-specialize bacteria” decline as a direct impact of changes in microbially accessible carbohydrates (MAC). Generally, a low carbohydrate diet will lower MAC, thus lowering the number of diet-specialist microbes and allowing Akkermansia to become overrepresented, consistent with our discovery. Most importantly, neither of these two genera (Escherichia-Shigella, Akkermansia) was discovered in our previous analysis using the ALDE model [54], where both classes were combined for microbiome biomarker identification (Escherichia-Shigella: p-value 0.14, Akkermansia: p-value 0.55) [20]. This highlights the relevance and importance of nutritional-ecotypes identification in microbiome marker discovery. ## Discussion The aim of this study is to investigate and unravel the complex interaction between diet, the microbiome and an individual’s health. We achieve this by exploring the effects of dietary pattern (or composition) on the relationship between the microbiome and host health and by developing a method called NEMoE that detects such heterogeneity. Through a series of simulation studies, NEMoE shows strong prediction performance when the underlying data show heterogeneity explained by different nutrient intake. Furthermore, we illustrate the practical performance of NEMoE on a gut microbiome PD study in which nutritional-ecotypes and microbial signatures of disease are found. We show that NEMoE outperforms the predictive accuracy of previous models (higher AUC) and identifies multiple known PD microbiome markers. Two different nutritional-ecotypes are also identified within our data with distinct protein-to-carbohydrate intake ratios and novel candidate signatures that were indicative of a diet-specific cohort. While we focus on the discovery of microbial signatures of PD by splitting the population based on dietary profile, the architecture of NEMoE means its flexible algorithm can take different types of data for subcohort detection (data used for gating networks) or biomarker identification (data used for expert networks). Therefore, an alternate research question could be to identify nutrients as disease markers for diverse microbiome profiles, and the NEMoE system can readily adapt to this new problem by changing the input of the gating network and experts network. Often, clinical knowledge or interest guides the decision on question formulation. However, if we consider both the dietary and microbiome profiles to be equivalent proxies for one’s nutrition system, then performing NEMoE in two different ways allows us to empirically compare the effectiveness of nutritional signatures versus microbial signatures and provides us with insight into the natural heterogeneity in the microbiome and in nutritional intake. NEMoE is designed to partition samples based on their associated nutrient intake and can be viewed as a data-driven strategy for subcohort or latent class identification. An alternative option is to investigate a knowledge-driven strategy to achieve the same goal and one example is the use of “enterotype.” Similar to unsupervised learning, stratifying samples based on “enterotype” while providing an alternative way to stratify samples, does not explicitly take disease prediction performance into account. As a result, the aggregate predictive ability of the three separate enterotypes is lower than the nutritional-ecotypes division discovered by the NEMoE approach. The proposed NEMoE method is based on diet-microbiome-host health interaction. However, it is not restricted to diet and microbiome data. Our method can be expanded to other multi-omics studies to identify subcohorts determined by the heterogeneity in relationships between covariates and response. One potential application is in the clinical heterogeneity of the relationship between multi-omics and host health. In such scenarios, the subcohorts are determined by their clinical index while the omics data are used to model the relationship between host health and information from a specific molecular platform. In summary, we present NEMoE, a novel statistical method to model heterogeneity of diet and the gut microbiome in disease. NEMoE identifies nutritional-ecotypes based on a maximum likelihood framework and using an Expectation-Maximization step to estimate the model parameters. Our proposed framework also enables identification and then accounts for multiple levels of structure in the feature set, a unique characteristic in microbiome data, where we are able to estimate a shared latent class for each individual at different taxonomic levels. Effectiveness of NEMoE is validated at three levels. First, we demonstrate through a series of extensive simulation studies the model’s ability to accurately identify latent classes and to increase microbiome predictability. Second, we validate the performance of NEMoE on a real disease dataset and show that this method outperforms existing two-stage methods. Finally, the downstream impact and practical importance of NEMoE is further demonstrated by the discovery of diet-specific PD microbiome markers, such as Escherichia-Shigella and Akkermansia, which are not identified by the ALDE model [54]. ## In-house studies Our in-house gut microbiome PD data collection includes stool samples from 101 PD patients and 83 healthy controls across three timepoints (0-, 6-, and 12-month time points). The samples were collected and 16S rRNA V3–V4 amplicon sequencing was performed on an Illumina MiSeq platform. Details of the experimental setting can be found in Lubomski et al. [ 19, 20]. We denoted data corresponding to each timepoints as Lubomski_0, Lubomski_6, and Lubomski_12, respectively. ## PD-diet Dietary information was collected by a comprehensive Food Frequency Questionnaire and resulted in a table of nutrient intake with 23 macronutrients, presented earlier [43]. Details of the sample information and sequence processing can be found in Lubomski et al. [ 19, 20]. ## Public validation (PV) studies We curated a series of datasets from eight different publicly-available microbiome studies [37, 38, 44–46, 51] to further validate results from NEMoE. All the datasets were processed using the dada2 pipeline [49] (v1.16) and microbiome taxa were annotated using taxonomy reference “silva 138” [48, 50]. Samples with low sequence reads (<1000) were excluded from the analysis. More information on these datasets can be found in Table 2. For the longitudinal datasets Aho [44], the data for baseline and follow-up, which were collected after 2.5 years, are denoted as Aho (baseline) and Aho (follow-up) respectively. ## PD-microbiome data processing We excluded 7 samples with extremely large energy intake (>20,000 kJ per day), one subject with low microbial read counts (total counts < 10,000), and two samples with missing nutrition measurements, resulting in 175 samples (75 HC individuals and 100 PD individuals). Raw counts from microbiome data were first normalized by total sum scaling, i.e., the counts (totals) were normalized into a composition proportion. Then core microbial features were kept and further transformed: Features that had more than $30\%$ zeros in the n samples and features which had sample variance smaller than 10−5 were filtered out at each taxonomic rank resulting in the core microbial features of 7 Phylum, 19 Order, 27 Family, 41 Genus, and 101 ASVs, and 3,152,746 total reads were kept from 6,024,011 reads; variance stability transformation, i.e. an arcsin square root transformation, was performed on taxa proportion [47, 55]; the arcsin transformed data were further standardized to have mean zero and unit variance (z-score). We also performed z-score and central log transformation and the corresponding result are shown in Fig. S7. ## PD-diet features construction In addition to the nutrients intake values, we calculated the percentage of energy intake as protein (%EP), percentage of energy intake as fat (%EF), percentage of energy intake as carbohydrate (%EC), and protein intake and carbohydrate intake ratio (P:C) as additional variables. These transformations of nutritional features are widely used in nutri-omics studies [56, 57]. All of the 27 nutritional features were z-scored. ## Nutrition-ecotype mixture of expert (NEMoE) model The development of NEMoE was inspired by a mixture of experts approach to model heterogeneous data as shown in Supplementary Fig. 2a. In machine learning, the concept of “gate” [58] can be thought of as a decision-making component given some input. Our approach consists of two key components, a “gating network” that is set up to determine which nutritional-class the sample belongs to and a “k-experts network” of size k to build classifiers for each nutritional-class. NEMoE uses a regularized MoE (RMoE) model, which adds elastic-net penalties to both the gating network and the experts network. Regularization is needed here because a non-regularized MoE does not extend to a large p small n framework [59] where the number of features (p) is much larger than the number of samples (n). This data characteristic often occurs in diet and microbiome data where there are many more microbial features (p) than individual samples (n). NEMoE further incorporates the taxonomic information into RMoE by jointly optimizing RMoE models from all taxonomic levels with the added constraint that all RMoE share the same gating network (Supplementary Fig. 2b). ## Mathematical formulation of NEMoE For a transformed microbiome data at taxonomic level l, we use the matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X_{n\times {p}_l}}^{(l)}$$\end{document}Xn×pl(l)to denote the relative abundance in n samples of pl taxa. The corresponding diet information, measured as a nutrients intake matrix, is denoted as Wn × q, where the q columns are the nutrient metrics for the same n samples Let Yn denote the binary response of the health outcome, with $Y = 1$ and $Y = 0$ representing individuals with and without disease, respectively. NEMoE models the heterogeneous relationship between the microbiome and the health outcome by a mixture distribution, i.e.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_l\left($Y = 1$\|{X}^{(l)},W\right)={\sum}_{$k = 1$}^K{\pi}_k\frac{\mathit{\exp}\left({X}^{(l)}{\beta_k}^{(l)}\right)}{1+\mathit{\exp}\left({X}^{(l)}{\beta_k}^{(l)}\right)},$$\end{document}PlY=1\|X(l),W=∑$k = 1$KπkexpX(l)βk(l)1+expX(l)βk(l),where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\pi}_k=\frac{\mathit{\exp}\left(W{\gamma}_k\right)}{\sum_{$i = 1$}^K\mathit{\exp}\left(W{\gamma}_i\right)}$$\end{document}πk=expWγk∑$i = 1$KexpWγi is the nutrition class mixing weight of shared components determined by nutrients intake, and where γk and βk are the corresponding effect size for the gating network and the experts network, respectively, and K denotes the predetermined number of nutrition classes. NEMoE estimates the regularized sum of all levels of the log-likelihood function in Equation [1], where the regularization term consists of elastic net penalties for both the gating network and the experts network:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$rLL={\sum}_{$l = 1$}^L{\sum}_{$k = 1$}^K\left\{{\sum}_{$i = 1$}^n\mathit{\log}\left[P\left({Y}_i\|{X_i}^{(l)},{W}_i\right)\right]-\phi \left({\lambda_{1k}}^{(l)},{\alpha_{1k}}^{(l)},{\beta_k}^{(l)}\right)\right\}-\phi \left({\lambda}_2,{\alpha}_2,\gamma \right),$$\end{document}rLL=∑$l = 1$L∑$k = 1$K∑$i = 1$nlogPYi\|Xi(l),Wi-ϕλ1k(l),α1k(l),βk(l)-ϕλ2,α2,γ,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi \left(\lambda, \alpha, \beta \right)=\lambda \Big[\alpha \left\|\beta \right\|+\frac{1}{2}\left(1-\alpha \right){\left\Vert \beta \right\Vert}_2^2$$\end{document}ϕλ,α,β=λ[αβ+121-αβ22] is the elastic net penalty function and λ1k(l), α1k(l), λ2, α2 are the corresponding parameters for penalties in the experts network and in the gating function. The regularized LL can be maximized through a proximal Newton Expectation Maximization algorithm [59]. Details of the optimization procedure can be found in the reference manual of the NEMoE package https://sydneybiox.github.io/NEMoE. ## Comparison methods Table 1 contains a summary of all methods used in the comparison study. We included the most commonly used methods in microbiome analysis as well as a naive two-stage approach. All of the comparisons were performed on simulation datasets and on in-house data on the Genus level. ## Naive two-stage approach The approach first clustered the nutrition data using unsupervised learning methods such as k-means. Then, based on the clustering result, samples in each cluster were used to build a classification model of microbiome and health state. The choice of classification models we used in our simulation includes sparse logistic regression (glmnet v4.1-2), support vector machine (e1071 v1.7-11), and random forest (randomForest v4.6-14). ## Differential abundance We compared differential relative abundance between PD and HC in all datasets. The comparison was based on a non-parametric bootstrapping procedure. We resampled the data with replacement, then calculated the difference of the average relative abundance between PD and HC. This procedure was repeated 10,000 times for each taxon and the $95\%$ confidence interval of the differential relative abundance was calculated. ## Simulation framework Our simulation first generated independent data of 2n samples from the procedure described above, then the first n samples were used for training and another n samples were used to calculate the predicted accuracy. The details of parameter settings in each simulation are described in Table 3. Implementation Table 3Summary of Simulation settingsSimulation Description n p c q η c e c g K ρ Evaluate the effect of n[100, 200, 500, 1000]50300.12220Evaluate the effect of η2005030(0, 0.1, 0.3, 0.5)2220Evaluate the effect of p200[30, 50, 80, 100]300.12220Evaluate the effect of q20050[30, 50, 80, 100]0.12220Evaluate the effect of ρ20050300.1222(0, 0.1, 0.3, 0.5)Evaluate the effect of Ka[100, 200, 500, 1000]5030(0, 0.1, 0.3, 0.5)2230Evaluate the multi-level datab500100300.12220aFor the evaluation of the effect of K, the underlying simulation data is generated based on $K = 3$, while the fitted NEMoE is based using K ranging from 2 to 4bFor the evaluation of the multi-level, we compare the adjusted rand index between NEMoE using all 5 levels of data (Phylum, Order, Family, Genus, and ASV) with NEMoE using only one level datacExcept the evaluation of multi-level, all evaluations were performed based on single-level data. For the multi-level data, the number of variables for Phylum, Order, Family, Genus, and ASV levels are 30, 50, 80, and 100, respectively ## Supplementary Information Additional file 1: Supplementary notes. Supplementary Fig. 1. Illustration of NEMoE and two-stage model. Supplementary Fig. 2. Graphical model representation of NEMoE. Supplementary Fig. 3. Nutrition classes determined by k-means do not show an informative relationship between microbiome and PD. Supplementary Fig 4. Simulation results of NEMoE and other methods under different settings. Supplementary Fig 5. External validation of consensus taxa Faecalibacterium and Blautia. Supplementary Fig 6. Prediction performance of different types of input for NEMoE. Supplementary Fig 7. ROC curves for different standardization methods of microbiome composition data analysis. ## Code availability NEMoE is implemented using Rcpp and available at https://github.com/SydneyBioX/NEMoE and in the process of submission to the BioConductor repository. All code used in this paper is freely available from our GitHub repository https://github.com/SydneyBioX/NEMoE_MS. ## Conflict of interest Not industry sponsored. All authors report no relevant disclosures. ## References 1. Li H. Microbiome, Metagenomics, and high-dimensional compositional data analysis. Annu Rev Stat Appl Annual Reviews (2015.0) 2 73-94. DOI: 10.1146/annurev-statistics-010814-020351 2. Wu GD, Chen J, Hoffmann C, Bittinger K, Chen Y-Y, Keilbaugh SA. Linking long-term dietary patterns with gut microbial enterotypes. Science. (2011.0) 334 105-108. DOI: 10.1126/science.1208344 3. Cho JH, Abraham C. Inflammatory bowel disease genetics: Nod2. Annu Rev Med (2007.0) 58 401-416. DOI: 10.1146/annurev.med.58.061705.145024 4. Pascal V, Pozuelo M, Borruel N, Casellas F, Campos D, Santiago A. A microbial signature for Crohn’s disease. Gut. (2017.0) 66 813-822. DOI: 10.1136/gutjnl-2016-313235 5. Qin J, Li Y, Cai Z, Li S, Zhu J, Zhang F. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature. (2012.0) 490 55-60. DOI: 10.1038/nature11450 6. Koeth RA, Wang Z, Levison BS, Buffa JA, Org E, Sheehy BT. Intestinal microbiota metabolism of L-carnitine, a nutrient in red meat, promotes atherosclerosis. Nat Med (2013.0) 19 576-585. DOI: 10.1038/nm.3145 7. Lubomski M, Tan AH, Lim S-Y, Holmes AJ, Davis RL, Sue CM. Parkinson’s disease and the gastrointestinal microbiome. J Neurol (2020.0) 267 2507-2523. DOI: 10.1007/s00415-019-09320-1 8. Yu D, Nguyen SM, Yang Y, Xu W, Cai H, Wu J. Long-term diet quality is associated with gut microbiome diversity and composition among urban Chinese adults. Am J Clin Nutr (2021.0) 113 684-694. DOI: 10.1093/ajcn/nqaa350 9. Xu Z, Knight R. Dietary effects on human gut microbiome diversity. Br J Nutr (2015.0) 113 Suppl S1-S5. DOI: 10.1017/S0007114514004127 10. McBurney MI, Davis C, Fraser CM, Schneeman BO, Huttenhower C, Verbeke K. Establishing what constitutes a healthy human gut microbiome: state of the science, regulatory considerations, and future directions. J Nutr (2019.0) 149 1882-1895. DOI: 10.1093/jn/nxz154 11. De Filippis F, Pellegrini N, Vannini L, Jeffery IB, La Storia A, Laghi L. High-level adherence to a Mediterranean diet beneficially impacts the gut microbiota and associated metabolome. Gut. (2016.0) 65 1812-1821. DOI: 10.1136/gutjnl-2015-309957 12. Read MN, Holmes AJ. Towards an integrative understanding of diet–host–gut microbiome interactions. Front Immunol (2017.0) 8 538. DOI: 10.3389/fimmu.2017.00538 13. Holmes AJ, Chew YV, Colakoglu F, Cliff JB, Klaassens E, Read MN. Diet-microbiome interactions in health are controlled by intestinal nitrogen source constraints. Cell Metab (2017.0) 25 140-151. DOI: 10.1016/j.cmet.2016.10.021 14. 14.Cronin P, Joyce SA, O’Toole PW, O’Connor EM. Dietary fibre modulates the gut microbiota. Nutrients. 2021:13. 10.3390/nu13051655. 15. David LA, Maurice CF, Carmody RN, Gootenberg DB, Button JE, Wolfe BE. Diet rapidly and reproducibly alters the human gut microbiome. Nature. (2014.0) 505 559-563. DOI: 10.1038/nature12820 16. 16.Hegelmaier T, Lebbing M, Duscha A, Tomaske L, Tönges L, Holm JB, et al. Interventional influence of the intestinal microbiome through dietary intervention and bowel cleansing might improve motor symptoms in Parkinson’s disease. Cells. 2020;9. 10.3390/cells9020376. 17. Zeevi D, Korem T, Zmora N, Israeli D, Rothschild D, Weinberger A. Personalized nutrition by prediction of glycemic responses. Cell. (2015.0) 163 1079-1094. DOI: 10.1016/j.cell.2015.11.001 18. Asnicar F, Berry SE, Valdes AM, Nguyen LH, Piccinno G, Drew DA. Microbiome connections with host metabolism and habitual diet from 1,098 deeply phenotyped individuals. Nat Med (2021.0) 27 321-332. DOI: 10.1038/s41591-020-01183-8 19. 19.Lubomski M, Xu X, Holmes A, Muller S, Yang JYH, Davis RL, et al. Nutritional intake and gut microbiome composition predict Parkinson’s disease. Front Aging Neurosci. 10.3389/fnagi.2022.881872. 20. 20.Lubomski M, Xu X, Holmes A, Muller S, Yang JYH, Davis RL, et al. The gut microbiome in Parkinson’s disease: a longitudinal study of the impacts on disease progression and the use of device-assisted therapies. Front Aging Neurosci. 10.3389/fnagi.2022.875261. 21. 21.Liang D, Leung RK-K, Guan W, Au WW. Involvement of gut microbiome in human health and disease: brief overview, knowledge gaps and research opportunities. Gut Pathogens. 2018. 10.1186/s13099-018-0230-4. 22. 22.Schulz C-A, Oluwagbemigun K, Nöthlings U. Advances in dietary pattern analysis in nutritional epidemiology. Eur J Nutr. 2021. 10.1007/s00394-021-02545-9. 23. Tebani A, Bekri S. Paving the way to precision nutrition through metabolomics. Front Nutr (2019.0) 6 41. DOI: 10.3389/fnut.2019.00041 24. Jannasch F, Riordan F, Andersen LF, Schulze MB. Exploratory dietary patterns: a systematic review of methods applied in pan-European studies and of validation studies. Br J Nutr (2018.0) 120 601-611. DOI: 10.1017/S0007114518001800 25. Schulze MB, Martínez-González MA, Fung TT, Lichtenstein AH, Forouhi NG. Food based dietary patterns and chronic disease prevention. BMJ. (2018.0) 361 k2396. DOI: 10.1136/bmj.k2396 26. Hughes RL, Kable ME, Marco M, Keim NL. The role of the gut microbiome in predicting response to diet and the development of precision nutrition models. Part II: results. Adv Nutr (2019.0) 10 979-998. DOI: 10.1093/advances/nmz049 27. Hose AJ, Pagani G, Karvonen AM, Kirjavainen PV, Roduit C, Genuneit J. Excessive unbalanced meat consumption in the first year of life increases asthma risk in the PASTURE and LUKAS2 birth cohorts. Front Immunol (2021.0) 12 651709. DOI: 10.3389/fimmu.2021.651709 28. Tap J, Störsrud S, Le Nevé B, Cotillard A, Pons N, Doré J. Diet and gut microbiome interactions of relevance for symptoms in irritable bowel syndrome. Microbiome. (2021.0) 9 74. DOI: 10.1186/s40168-021-01018-9 29. 29.Patrick E, Schramm S-J, Ormerod JT, Scolyer RA, Mann GJ, Mueller S, et al. A multi-step classifier addressing cohort heterogeneity improves performance of prognostic biomarkers in three cancer types. Oncotarget. 2017:2807–15. 30. Tan AH, Chong CW, Lim S-Y, Yap IKS, Teh CSJ, Loke MF. Gut microbial ecosystem in Parkinson disease: new Clinicobiological insights from multi-Omics. Ann Neurol (2021.0) 89 546-559. DOI: 10.1002/ana.25982 31. Lê Cao K-A, Meugnier E, McLachlan GJ. Integrative mixture of experts to combine clinical factors and gene markers. Bioinformatics. (2010.0) 26 1192-1198. DOI: 10.1093/bioinformatics/btq107 32. Yuksel SE, Wilson JN, Gader PD. Twenty years of mixture of experts. IEEE Trans Neural Netw Learn Syst (2012.0) 23 1177-1193. DOI: 10.1109/TNNLS.2012.2200299 33. Huynh BT, Chamroukhi F. Estimation and feature selection in mixtures of generalized linear experts models (2019.0) 34. 34.Zou H, Hastie T. Regularization and variable selection via the elastic net. J Royal Stat Soc: Ser B (Statistical Methodology). 2005:301–20. 35. Costea PI, Hildebrand F, Arumugam M, Bäckhed F, Blaser MJ, Bushman FD. Enterotypes in the landscape of gut microbial community composition. Nat Microbiol (2018.0) 3 8-16. DOI: 10.1038/s41564-017-0072-8 36. Arumugam M, Raes J, Pelletier E, Le Paslier D, Yamada T, Mende DR. Enterotypes of the human gut microbiome. Nature (2011.0) 473 174-180. DOI: 10.1038/nature09944 37. Jin M, Li J, Liu F, Lyu N, Wang K, Wang L. Analysis of the gut microflora in patients with Parkinson’s disease. Front Neurosci (2019.0) 13 1184. DOI: 10.3389/fnins.2019.01184 38. Wallen ZD, Appah M, Dean MN, Sesler CL, Factor SA, Molho E. Characterizing dysbiosis of gut microbiome in PD: evidence for overabundance of opportunistic pathogens. NPJ Parkinsons Dis (2020.0) 6 11. DOI: 10.1038/s41531-020-0112-6 39. 39.Gerhardt S, Mohajeri M. Changes of colonic bacterial composition in Parkinson’s disease and other neurodegenerative diseases. Nutrients. 2018:708. 40. 40.Lubomski M, Xu X, Holmes AJ, Yang JYH, Sue CM, Davis RL. The impact of device-assisted therapies on the gut microbiome in Parkinson’s disease. J Neurol. 2021. 10.1007/s00415-021-10657-9. 41. Keshavarzian A, Green SJ, Engen PA, Voigt RM, Naqib A, Forsyth CB. Colonic bacterial composition in Parkinson’s disease. Mov Disord (2015.0) 30 1351-1360. DOI: 10.1002/mds.26307 42. Romano S, Savva GM, Bedarf JR, Charles IG, Hildebrand F, Narbad A. Meta-analysis of the Parkinson’s disease gut microbiome suggests alterations linked to intestinal inflammation. NPJ Parkinson’s Dis (2021.0) 7 1-13. PMID: 33397996 43. Palavra NC, Lubomski M, Flood VM, Davis RL, Sue CM. Increased added sugar consumption is common in Parkinson’s disease. Front Nutr (2021.0) 8 628845. DOI: 10.3389/fnut.2021.628845 44. Hill-Burns EM, Debelius JW, Morton JT, Wissemann WT, Lewis MR, Wallen ZD. Parkinson’s disease and Parkinson's disease medications have distinct signatures of the gut microbiome. Mov Disord (2017.0) 32 739-749. DOI: 10.1002/mds.26942 45. Scheperjans F, Aho V, Pereira PAB, Koskinen K, Paulin L, Pekkonen E. Gut microbiota are related to Parkinson’s disease and clinical phenotype. Mov Disord (2015.0) 30 350-358. DOI: 10.1002/mds.26069 46. 46.Weis S, Schwiertz A, Unger MM, Becker A, Faßbender K, Ratering S, et al. Effect of Parkinson’s disease and related medications on the composition of the fecal bacterial microbiota. NPJ Parkinson’s Dis. 2019. 47. Dong M, Li L, Chen M, Kusalik A, Xu W. Predictive analysis methods for human microbiome data with application to Parkinson’s disease. PLoS One (2020.0) 15 e0237779. DOI: 10.1371/journal.pone.0237779 48. Quast C, Pruesse E, Yilmaz P, Gerken J, Schweer T, Yarza P. The SILVA ribosomal RNA gene database project: improved data processing and web-based tools. Nucleic Acids Res (2013.0) 41 D590-D596. DOI: 10.1093/nar/gks1219 49. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJA, Holmes SP. DADA2: high-resolution sample inference from Illumina amplicon data. Nat Methods (2016.0) 13 581-583. DOI: 10.1038/nmeth.3869 50. 50.Yilmaz P, Parfrey LW, Yarza P, Gerken J, Pruesse E, Quast C, et al. The SILVA and “all-species living tree project (LTP)” taxonomic frameworks. Nucleic Acids Res. 2014:D643–8. 51. Aho VTE, Houser MC, Pereira PAB, Chang J, Rudi K, Paulin L. Relationships of gut microbiota, short-chain fatty acids, inflammation, and the gut barrier in Parkinson’s disease. Mol Neurodegener (2021.0) 16 6. DOI: 10.1186/s13024-021-00427-6 52. 52.Kang Y, Kang X, Zhang H, Liu Q, Yang H, Fan W. Gut microbiota and Parkinson’s disease: implications for Faecal microbiota transplantation therapy. ASN Neuro. 2021:175909142110162. 53. Bullich C, Keshavarzian A, Garssen J, Kraneveld A, Perez-Pardo P. Gut vibes in Parkinson’s disease: the microbiota-gut-brain Axis. Mov Disord Clin Pract (2019.0) 6 639-651. DOI: 10.1002/mdc3.12840 54. Fernandes AD, Reid JNS, Macklaim JM, McMurrough TA, Edgell DR, Gloor GB. Unifying the analysis of high-throughput sequencing datasets: characterizing RNA-seq, 16S rRNA gene sequencing and selective growth experiments by compositional data analysis. Microbiome (2014.0) 2 15. DOI: 10.1186/2049-2618-2-15 55. 55.Ma S, Shungin D, Mallick H, Schirmer M, Nguyen LH, Kolde R, et al. Population structure discovery in meta-analyzed microbial communities and inflammatory bowel disease. bioRxiv. 2020:2020.08.31.261214. 56. Simpson SJ, Le Couteur DG, James DE, George J, Gunton JE, Solon-Biet SM. The geometric framework for nutrition as a tool in precision medicine. Nutr Healthy Aging (2017.0) 4 217-226. DOI: 10.3233/NHA-170027 57. 57.Raubenheimer D, Simpson SJ. Nutritional ecology and human health. Annu Rev Nutr. 2016:603–26. 58. Makkuva A, Oh S, Kannan S, Viswanath P, Chiappa S, Calandra R. Learning in gated neural networks. Proceedings of the twenty third international conference on artificial intelligence and statistics (2020.0) 3338-3348 59. Fruhwirth-Schnatter S, Celeux G, Robert CP. Handbook of mixture analysis (2019.0)
--- title: LncRNA GAS5 suppresses TGF-β1-induced transformation of pulmonary pericytes into myofibroblasts by recruiting KDM5B and promoting H3K4me2/3 demethylation of the PDGFRα/β promoter authors: - Yichun Wang - Diyu Chen - Han Xie - Shuhua Zhou - Mingwang Jia - Xiaobo He - Feifei Guo - Yihuan Lai - Xiao Xiao Tang journal: Molecular Medicine year: 2023 pmcid: PMC10015786 doi: 10.1186/s10020-023-00620-x license: CC BY 4.0 --- # LncRNA GAS5 suppresses TGF-β1-induced transformation of pulmonary pericytes into myofibroblasts by recruiting KDM5B and promoting H3K4me2/3 demethylation of the PDGFRα/β promoter ## Abstract ### Background Idiopathic pulmonary fibrosis (IPF) is a condition that may cause persistent pulmonary damage. The transformation of pericytes into myofibroblasts has been recognized as a key player during IPF progression. This study aimed to investigate the functions of lncRNA growth arrest-specific transcript 5 (GAS5) in myofibroblast transformation during IPF progression. ### Methods We created a mouse model of pulmonary fibrosis (PF) via intratracheal administration of bleomycin. Pericytes were challenged with exogenous transforming growth factor-β1 (TGF-β1). To determine the expression of target molecules, we employed quantitative reverse transcription-polymerase chain reaction, Western blotting, and immunohistochemical and immunofluorescence staining. The pathological changes in the lungs were evaluated via H&E and Masson staining. Furthermore, the subcellular distribution of GAS5 was examined using FISH. Dual-luciferase reporter assay, ChIP, RNA pull-down, and RIP experiments were conducted to determine the molecular interaction. ### Results GAS5 expression decreased whereas PDGFRα/β expression increased in the lungs of IPF patients and mice with bleomycin-induced PF. The in vitro overexpression of GAS5 or silencing of PDGFRα/β inhibited the TGF-β1-induced differentiation of pericytes to myofibroblasts, as evidenced by the upregulation of pericyte markers NG2 and desmin as well as downregulation of myofibroblast markers α-SMA and collagen I. Further mechanistic analysis revealed that GAS5 recruited KDM5B to promote H3K4me$\frac{2}{3}$ demethylation, thereby suppressing PDGFRα/β expression. In addition, KDM5B overexpression inhibited pericyte–myofibroblast transformation and counteracted the promotional effect of GAS5 knockdown on pericyte–myofibroblast transformation. Lung fibrosis in mice was attenuated by GAS5 overexpression but promoted by GAS5 deficiency. ### Conclusion GAS5 represses pericyte–myofibroblast transformation by inhibiting PDGFRα/β expression via KDM5B-mediated H3K4me$\frac{2}{3}$ demethylation in IPF, identifying GAS5 as an intervention target for IPF. ### Supplementary Information The online version contains supplementary material available at 10.1186/s10020-023-00620-x. ## Introduction Idiopathic pulmonary fibrosis (IPF) is an irreversible and fatal lung disease featured by progressive scarring of the pulmonary parenchyma with a median survival of 3 years. It currently affects 14–43 per 100,000 individuals, and its incidence increases with age (Maher et al. 2021; Raghu et al. 2006; Yu and Tang 2022). According to current pathogenic theories, fibroblast activation and differentiation play important roles in IPF (Yamaguchi et al. 2020). The exposure of quiescent fibroblasts to a variety of profibrotic mediators, such as transforming growth factor-beta 1 (TGF-β1) and platelet-derived growth factor (PDGF), results in phenotypic differentiation into myofibroblasts, which causes excess deposition of extracellular matrix (ECM) during IPF progression (Yin et al. 2018; Xie et al. 2020). Over the past several years, researchers have found that pericytes are a substantial source of myofibroblasts and that atypical pericyte activation may be a significant cause of IPF (Chou et al. 2020). The biological functions of PDGF are mediated by binding to specific receptors (PDGFRα/β) (Chou et al. 2020; Djudjaj and Boor 2019). Blockage of the PDGFRα/β signaling pathway activation can delay IPF progression (Kishi et al. 2018; Vuorinen et al. 2007; Wang et al. 2021). Thus, elucidation of the regulatory mechanisms of the PDGFRα/β signaling pathway is crucial for IPF prevention and treatment. Lysine-specific demethylase 5B (KDM5B), a demethylase at the K4 position of H3, acts as a transcriptional inhibitor in various biological processes. For instance, KDM5B contributed to colorectal cancer proliferation via CDX2 inhibition by H3K4me3 demethylation (Huang et al. 2020). It has been suggested that HIF-1α can trigger tissue fibrosis through KDM5B-mediated transcriptional repression via H3K4me$\frac{2}{3}$ demethylation (Salminen et al. 2016). To date, it remains unclear whether KDM5B is involved in the pathogenesis of IPF. As predicted by the AnimalTFDB and RPISeq databases, KDM5B could bind to the PDGFRα/β promoter. Thus, we speculated that KDM5B transcriptionally represses PDGFRα/β via H3K4me$\frac{2}{3}$ demethylation to delay IPF progression. Long noncoding RNAs (lncRNAs) are specialized RNAs that have been proposed as therapeutic targets for a variety of diseases, including IPF. Chen et al. reported that lncRNA CTD-2528L19.6 attenuated IPF via fibroblast activation repression (Chen et al. 2021). It has been recognized that lncRNA growth arrest-specific 5 (GAS5) participates in multiple lung disorders, such as asthma and lung cancer (Poulet et al. 2020). However, the involvement of GAS5 in the pathological development of IPF remains unclear. In our preliminary experiments, we found that GAS5 was downregulated in patients and mice with IPF. Furthermore, bioinformatics analysis revealed that GAS5 could directly bind to KDM5B. In this context, we speculated that GAS5 recruited KDM5B to restrain PDGFRα/β transcription by promoting H3K4me$\frac{2}{3}$ demethylation, thereby inhibiting the transformation of lung pericytes to myofibroblasts. This study aimed to validate the aforementioned hypothesis in TGF-β1-exposed lung pericytes and bleomycin-treated mice. Our findings clarify the regulatory role of GAS5 in IPF progression, identifying GAS5 as a novel intervention target for IPF therapy. ## Clinical samples Lung tissues were obtained from 33 IPF patients (male, 25; female, 8; mean age, 40.14 ± 13.63 years) who underwent surgical resection at the third Affiliated Hospital of Guangzhou Medical University. Prior to the procedure, none of the patients had received any therapy. In addition, normal lung tissues (tumor-adjacent tissues) were obtained and used as controls. Informed consent was obtained from all participants. ## Isolation and culture of primary pericytes All experiments have been approved by the ethics committee of the Third Affiliated Hospital of Guangzhou Medical University, and pericytes were purified as previously described (Hung et al. 2017). Fresh lung tissues were collected from mice and cut into small pieces. After digestion and filtration through nylon meshes, single cells were resuspended in culture media. Pericytes with negative CD45, CD31, and CD326 expressions and positive PDGFRβ expression were selected using MACS (Miltenyi Biotec, USA). The isolated pericytes were seeded into gelatin-coated culture plates and maintained using DMEM/F-12 (11330032, Thermo Fisher Scientific Inc., Waltham, MA, USA) supplemented with $10\%$ FBS (10100, Thermo Fisher). ## Cell transfection For the induction of fibrogenesis, mouse pulmonary microvascular pericytes were treated with TGF-β1 (HY-P7117, MCE, NJ, USA) for 30 min at 37 °C. The shRNA targeting GAS5 (shGAS5), shPDGFRα, shPDGFRβ, shKDM5B, and negative control (shNC) were synthesized by GenePharma (Shanghai, China). The specific sequences for shRNAs are presented in Table 1. For the overexpression of GAS5 and PDGFRα/β, GAS5 and PDGFRα/β sequences amplified by RT-qPCR were ligated with the pcDNA3.0 plasmids (#13031, Addgene, Watertown, MA, USA) to establish the pcDNA3.0-GAS5 and pcDNA3.0-PDGFRα/β recombinant plasmids, respectively. The primers used for establishing the overexpression constructs are presented in Table 2. The aforementioned plasmids and sequences were transfected into cells using Lipofectamine 3000 (L3000001, Thermo Fisher Scientific) following the manufacturer’s protocols. Table 1Sequences for shRNAsshGAS5#1 targeting sequence5′-GGATCTCACAGCCAGTTCTGT-3′shGAS5#2 targeting sequence5′-GGATCTCACAGCCAGTTCTGT-3′shGAS5#3 targeting sequence5′-GCAATGTGCTAGAATAGAAGA-3′shGAS5#4 targeting sequence5′-GAGGCTGGATAGACAGTTTGA-3′shKDM5B#1 targeting sequence5′-CGGTGCTATTTCTATTCCTTA-3′shKDM5B#2 targeting sequence5′-GCCTACATCATGTGAAAGAAT-3′shKDM5B#3 targeting sequence5′-CCTGAAATTCAGGAGCTTTAT-3′shKDM5B#4 targeting sequence5′-ATCGCTTGCTGCACCGTTATT-3′shPDGFRα#1 targeting sequence5′-GCCAGCTCTTATTACCCTCTA-3′shPDGFRα#2 targeting sequence5′-GCCACATTTGAACATTGTGAA-3′shPDGFRα#3 targeting sequence5′-CCTGGAGAAGTGAGAAACAAA-3′shPDGFRα#4 targeting sequence5′-GATGATCTGCAAGCATATTAA-3′shPDGFRβ#1 targeting sequence5′-CCATGCTTAATGAATGCTGTT-3′shPDGFRβ#2 targeting sequence5′-CCTTGAATGAAGTCAACACTT-3′shPDGFRβ#3 targeting sequence5′-GAAGCGGGCTACTATACTATG-3′shPDGFRβ#4 targeting sequence5′-GATGTCACTGAGACGACAATT-3′Table 2The primers used for establishing the overexpression constructsGAS5 (gene ID: 14455)F: 5′-CTTTCGGAGCTGTGCGGCATTCTGA-3′R: 5′-TTCATGTTATAATACACTTTAATGG-3′PDGFRα (gene ID: 18595)F: 5′-ATGAGGACCTGGGCTTGCCTGCTGC-3′R: 5′-TTAGGTGGGTTTTAACCTTTTCCTT-3′PDGFRβ (gene ID: 18596)F: 5′-ATGCTGAGCGACCACTCCATCCGCT-3′R: 5′-CTAGGCTCCGAGGGTCTCCTTCAGG-3′KDM5B (gene ID: 75605)F: 5′-ATGGAGCCGGCCACCACGCTGCCCC-3′R: 5′-TTACTTTCGGCTTGGTGCGTCCTTC-3′ ## Mouse model of pulmonary fibrosis (PF) As previously described (Pan et al. 2020), 7–8-week-old male C57BL/6 mice weighing 20–22 g were obtained from Hunan Slac Jingda Laboratory Animal Co., Ltd. (Changsha, China). The mice were randomly assigned into six groups ($$n = 7$$ per group): control, bleomycin, bleomycin + NC, bleomycin + GAS5, bleomycin + shNC, and bleomycin + shGAS5. Intraperitoneal injection with 30-mg/kg sodium pentobarbital was adopted to anesthetize the mice. To inducePF, 2.5-mg/kg bleomycin (HY-108345, MCE) was intratracheally administered to the mice. The control group was administered with the same volume of PBS (10010001, Thermo Fisher Scientific). Five days before bleomycin administration, the mice were intratracheally injected with lentiviruses containing overexpression of GAS5, shGAS5, or NC (1 × 107 TU, GenePharma). Then, 2 weeks after bleomycin administration, the mice were sacrificed, and lung tissues were collected for further analysis. All animal protocols were approved by the Animal Care and Use Committee of Guangzhou Medical University. ## Hematoxylin and eosin staining Before being paraffin-embedded, the lung samples of mice were fixed in $4\%$ paraformaldehyde (P1110, Solarbio, Beijing, China). Subsequently, the lungs were cut into 4-µm slices and stained with the H&E staining kit (G1120, Solarbio) and examined under a microscope (Olympus, Japan) for pathological and fibrosis assessment. ## Masson staining Collagen deposition was evaluated via Masson staining on lung slices. In brief, the 4-μm slices of lungs were stained using the Masson trichrome staining kit (G1340, Solarbio). The images were photographed under a microscope. ## Immunohistochemical staining The 4-µm paraffin-embedded lung slices were dewaxed and rehydrated in ethanol gradient. After blocking endogenous peroxidases with $0.3\%$ hydrogen peroxide (7722-84-1, Sigma–Aldrich, Saint Louis, MO, USA), the slices were probed overnight at 4 °C with primary antibodies against α-SMA (ab124964, 1:50, Abcam, Cambridge, MA, UK), collagen I (ab88147, 1:50, Abcam), and desmin (ab227651, 1:2000, Abcam). Then, the slices were rinsed with PBS and reacted with the secondary antibody. Subsequently, 3,3′-diaminobenzidine was used to visualize the slices. A light microscope was used to obtain the photographs. ## Real-time quantitative polymerase chain reaction (RT-qPCR) Using the TRIzol reagent (T9424, Sigma–Aldrich), total RNA was isolated from the pericytes and tissues, followed by reverse transcription into cDNA using the PrimeScript RT reagent kit (RR047AA, Takara, Tokyo, Japan). Then, qPCR was performed using the SYBR Premix Ex Taq II kit (RR820A, Takara). GAPDH was used as the housekeeping gene to calculate the relative gene expression using the 2−ΔΔCT method as previously described (Zuo et al. 2022). The primers used in RT-qPCR are listed in Table 3.Table 3Primers used for RT-qPCR analysisGenesPrimer sequences (5′–3′)Human GAS5F: 5′-GTGAGGTATGGTGCTGGGTG-3′R: 5′-GCCAATGGCTTGAGTTAGGC-3′Human PDGFRAF: 5′-GAAGCTGTCAACCTGCATGA-3′R: 5′-CTTCCTTAGCACGGATCAGC-3′Human PDGFRBF: 5′-CCCAATGAGGGTGACAACGA-3′R: 5′-AAGCTATCCTCTGCTTCCGC-3′Human GAPDHF: 5′-CCAGGTGGTCTCCTCTGA-3′R: 5′-GCTGTAGCCAAATCGTTGT-3′Mouse Gas5F: 5′-GAATGGCAGTGTGGACCTCT-3′R: 5′-CAGCCTCAAACTCCACCATT-3′Mouse PdgfraF: 5′-TGGCATGATGGTCGATTCTA-3′R: 5′-CGCTGAGGTGGTAGAAGGAG-3′Mouse PdgfrbF: 5′-CCTTCTCCAGTGTGCTGACA-3′R: 5′-TCATGTAGCGTCACCTCCAG-3′Mouse GapdhF: 5′-AGCCCAAGATGCCCTTCAGT-3′R: 5′-CCGTGTTCCTACCCCCAATG-3′ ## Western blotting The cells or tissues were lysed in RIPA solution (P0013B, Beyotime, Haimen, China) and quantified using BCA protein assay (P0012, Beyotime). The protein (50 µg) was separated on SDS-PAGE and then transferred to PVDF membranes (3010040001, Roche, Basel, Switzerland). After blocking with $5\%$ skimmed milk for 1 h, the membranes were incubated with primary antibodies overnight at 4 °C. The following primary antibodies were used: PDGFR-α (ab203491, 1:1000, Abcam), PDGFR-β (ab69506, 1:1000, Abcam), α-SMA (ab124964, 1:2000, Abcam), NG-2 (ab275024, 1:1000, Abcam), collagen I (ab260043, 1:1000, Abcam), desmin (ab227651, 1:5000, Abcam), KDM5B (ab181089, 1:1000, Abcam), H3K4me2 (#9725, 1:1000, CST, USA), H3K4me3 (#9751, 1:1000, CST, Danvers, MA, USA), and β-actin (ab8226, 1:1000, Abcam). After incubation with a secondary antibody, the membranes were treated with the ECL-chemiluminescent kit (34580, Thermo Fisher Scientific) to visualize the protein bands. ## Immunofluorescence staining Primary pericytes were fixed in $4\%$ paraformaldehyde, permeabilized with $0.1\%$ Triton X-100 (T8787, Sigma–Aldrich), and blocked with $1\%$ BSA (A1595, Sigma–Aldrich). The antibodies, including α-SMA (ab124964, 1:200, Abcam) and desmin (ab227651, 1:500, Abcam), were applied overnight at 4 °C. The pericytes were probed with goat anti-rabbit IgG (H + L) secondary antibody (ab150077, 1:200, Abcam) for 1 h. A fluorescence microscope (Zeiss, Germany) was used to observe and photograph the outcomes of the immunofluorescence assays. ## Fluorescence in situ hybridization (FISH) FISH was employed to determine GAS5 localization in pericytes. Briefly, the pericytes were seeded in a 24-well plate. After fixation and permeabilization, the cells were probed with GAS5 probe. Subsequently, the cells were rinsed in a hybridization solution and stained with DAPI (MBD0015, Sigma–Aldrich) in the dark. Images were obtained using a laser scanning confocal microscope (Leica, Germany). ## Chromatin immunoprecipitation (ChIP) assay ChIP assay was conducted using the Magna ChIP and EZ-Magna ChIP kit (17-10461, Millipore, Billerica, MA, USA). Sequences for the PDGFRα/β promoter containing the wild-type (WT) or mutant (MUT) binding sites in KDM5B are presented in Table 4. Generally, pericytes were extracted and cross-linked with $1\%$ formaldehyde before being treated with 125-mM glycine (67419, Sigma–Aldrich). The cell suspension was then sonicated and precipitated with anti-KDM5B (#15327, CST), anti-H3K4me2 (#9725, CST), anti-H3K4me3 (#9751, CST), or anti-control rabbit IgG (ab37415, Abcam). PDGFRα/β enrichment in the immunoprecipitated complex was assessed via quantitative PCR.Table 4Sequences for PDGFRα/β promoter containing the wild type (MT) or mutant (MUT) binding sites in KDM5BWT-PDGFRα-BS1(-582/-558)5′-GGGAGAGAAACAAACGGAGGAGCTG-3′WT-PDGFRα-BS2(-487/-463)5′-GAAGGAGAAGGTAAGGGAGAGGAAA-3′WT-PDGFRα-BS3(-487/-463)5′-AAAAAAAAAGAAAAGAAAAAGAAAA-3′MUT-PDGFRα-BS1(-582/-558)5′- CCCTCTCTTTGTTTGCCTCCTCGAC-3′MUT-PDGFRα-BS2(-487/-463)5′-CTTCCTCTTCCATTCCCTCTCCTTT-3′MUT-PDGFRα-BS3(-487/-463)5′-TTTTTTTTCTTTTCTTTTTCTTTT-3′WT-PDGFRβ-BS1(-1432/-1417)5′-CTCTTCCTGTTTCCTC-3′WT-PDGFRβ-BS2(-196/-172)5′-GGAAAGGAGGAAGAAAAACAAGAAA-3′WT-PDGFRβ-BS3(-165/-141)5′-GGAAAAGAAAGAGAGGAAAAAAAA-3′MUT-PDGFRβ-BS1(-1432/-1417)5′-GAGAAGGACAAAGGAG-3′MUT-PDGFRβ-BS2(-196/-172)5′-CCTTTCCTCCTTCTTTTTGTTCTTT-3′MUT-PDGFRβ-BS3(-165/-141)5′-CCTTTTCTTTCTCTCCTTTTTTTT-3′ ## Dual-luciferase reporter assay A luciferase reporter assay was conducted to confirm the target relationship between KDM5B and PDGFRα/β. The sequences of the PDGFRα/β promoter containing the WT or MUT binding sites in KDM5B were synthesized and introduced into the pmirGLO vector (E1330, Promega, Madison, WI, USA). In 96-well plates, 293 T cells were co-transfected with the aforementioned reporter vectors and shKDM5B or shNC using Lipofectamine 3000 (L3000001, Thermo Fisher Scientific). After 48 h, the luciferase activity was measured using a dual-luciferase reporter system (E1960, Promega). ## RNA pull-down assay The biotin-labeled GAS5 probe was provided by RiBobio (Guangzhou, China) and transfected into pericytes. Then, cell lysate was prepared using lysis buffer and incubated with streptavidin–agarose beads (16-126, Millipore) for 3 h at 4 °C. The RNA-binding protein complexes in agarose beads were washed and boiled at 95 °C–100 °C. Finally, the eluted protein was assessed via Western blotting. ## RNA immunoprecipitation assay Pericytes were lysed using the lysis buffer from the Magna RIP kit (17-700, Millipore). The cell lysate was treated overnight at 4 °C with RIP buffer containing magnetic beads pre-coated with anti-IgG (#3900, CST) or anti-KDM5B (#15327, CST). To remove the protein, proteinase K (70663, Sigma–Aldrich) was added to the eluted samples at 55 °C for 30 min. Subsequently, the immunoprecipitated RNA was isolated, and GAS5 enrichment was determined via RT-qPCR. ## Statistical analysis The data are expressed as mean ± standard deviation (SD). Statistical analysis was conducted using GraphPad Prism. Student’s t-test was employed to determine the difference between two groups. One-way analysis of variance was employed to compare multiple groups, followed by Tukey’s post hoc test. $P \leq 0.05$ was considered to indicate statistical significance. ## Downregulation of GAS5 and upregulation of PDGFR α/β in the lung of IPF patients and mice with bleomycin-induced pulmonary fibrosis First, the dysregulation of GAS5 and PDGFR α/β in IPF was evaluated via RT-qPCR. We found that the GAS5 expression decreased whereas the PDGFR α/β expression increased in the IPF lung tissues (Fig. 1A). Furthermore, the protein levels of PDGFR α/β, collagen I, and α-SMA were enhanced, but the levels of pericyte markers desmin and NG2 decreased in the IPF specimens (Fig. 1B). To further confirm the above findings, a bleomycin-induced mouse model of PF was created. The fibrosis and collagen deposition in the lung tissues of bleomycin-challenged mice were observed via H&E and Masson staining (Fig. 1C, D). Immunohistochemical staining further confirmed the enhanced expressions of α-SMA and collagen I in PF mice (Fig. 1E). Similarly, GAS5, desmin, and NG2 were downregulated whereas PDGFR α/β, collagen I, and α-SMA were upregulated in the bleomycin-induced PF mouse model (Fig. 1F, G). Taken together, GAS5 was aberrantly downregulated and PDGFR α/β was upregulated in patients and mice with IPF.Fig. 1A low GAS5 expression and a high PDGFR α/β expression were observed in the lungs of patients and mice with IPF. A The expressions of GAS5 and PDGFR α/β in the IPF specimens were examined via RT-qPCR. B Western blotting was employed to explore the expressions of desmin, NG2, α-SMA, collagen I, and PDGFR α/β in IPF patients. C The pathological changes in the lung tissues of bleomycin-induced mice were evaluated via H&E staining. D Masson staining was used to determine collagen formation in the lung tissues. E Immunohistochemical staining was employed to detect the expressions of α-SMA and collagen I. F RT-qPCR analysis of the expressions of GAS5 and PDGFR α/β in IPF mice. G Western blotting was used to examine the expressions of desmin, NG2, α-SMA, collagen I, and PDGFR α/β in the lungs. For A and B, $$n = 33$$; for C–F, $$n = 6$.$ * $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$ ## TGF-β1 exposure-induced phenotypic differentiation of pericytes into myofibroblasts in a dose- and time-dependent manner We tested the effect of TGF-β1 on the transformation of pericytes into myofibroblasts. The results of Western blotting indicated that the levels of myofibroblasts (collagen I and α-SMA) were dose-dependently elevated whereas the levels of pericyte markers (desmin and NG2) were declined by TGF-β1 stimulation (Fig. 2A). Furthermore, TGF-β1 exposure reduced GAS5 expression and increased PDGFR α/β expression dose-dependently (Fig. 2B, C). It was also found that with the extension of time, GAS5 was downregulated and PDGFR α/β were upregulated in TGF-β1-exposed pericytes (Fig. 2D, E). These data suggested that TGF-β1 induced the phenotypic transformation of pericytes into myofibroblasts in a dose- and time-dependent manner. Fig. 2TGF-β1 induced pericyte–myofibroblast transformation in a dose- or time-dependent manner. Pericytes were exposed to different TGF-β1 concentrations (0, 1, 5, 10, and 15 ng/mL) for 12 h. A Western blotting analysis of the levels of pericyte markers (desmin, NG-2) and myofibroblast markers (α-SMA, collagen I). B RT-qPCR analysis of the expressions of GAS5 and PDGFR α/β in pericytes. C The expression of PDGFR α/β was evaluated via Western blotting. Pericytes were treated with 10 ng/mL of TGF-β1 at different time periods (1, 3, 6, 12, 24, and 48 h). D The expressions of GAS5 and PDGFR α/β were examined via RT-qPCR. E Western blotting analysis of the expression of PDGFR α/β. For A–E, $$n = 3$.$ $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$ ## GAS5 repressed the TGF-β1-induced transformation of pericytes into myofibroblasts Given the differential expression of GAS5 in IPF, we further investigated the impact of GAS5 on TGF-β1-induced pericyte–myofibroblast transformation. The overexpression or silencing efficiency of GAS5 in pericytes was confirmed by RT-qPCR (Fig. 3A, Additional file 1: Fig. S1A, B). The result indicated that shGAS5-1# was adopted in the following experiments. As shown in Fig. 3B, the declined expression of GAS5 in TGF-β1-challenged pericytes was reversed by GAS5 overexpression. Contrarily, GAS5 depletion further reduced the GAS5 level in TGF-β1-stimulated pericytes. In addition, immunofluorescence staining demonstrated that the TGF-β1-induced reduction in desmin expression and increase in α-SMA expression could be counteracted by GAS5 overexpression but reinforced by GAS5 knockdown (Fig. 3C). Furthermore, enforced GAS5 expression reversed the downregulation of desmin and NG2 and upregulation of α-SMA, collagen I, and PDGFR α/β mediated by TGF-β1 exposure, GAS5 depletion yielded opposite results (Fig. 3D, E). Therefore, GAS5 could repress the TGF-β1-induced transformation of pericytes into myofibroblasts. Fig. 3GAS5 repressed the TGF-β1-induced pericyte–myofibroblast transformation. Pericytes were transfected with pcDNA3.0 containing GAS5 (GAS5) or shGAS5 and then exposed to TGF-β1 (10 ng/mL) for 12 h. They were divided into six groups: control, TGF-β1, TGF-β1 + vector, TGF-β1 + GAS5, TGF-β1 + shNC, and TGF-β1 + shGAS5. A, B RT-qPCR examination of the GAS5 level. C Immunofluorescence staining detection of desmin and α-SMA expressions. D Western blotting analysis of the protein levels of desmin, NG-2, α-SMA, and collagen I. E Western blotting was used to detect the expression of PDGFR α/β. For A–E, $$n = 3$.$ $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$ ## PDGFR α/β facilitated pericyte–myofibroblast transformation induced by TGF-β1 To explore the function of PDGFR α/β in TGF-β1-induced pericyte–myofibroblast transformation, pericytes were transfected with pcDNA3.0 containing PDGFR α/β or shPDGFR α/β. As confirmed by RT-qPCR and Western blotting, PDGFR α/β was overexpressed or silenced in pericytes following transfection (Fig. 4A, Additional file 1: Fig. S1C–F). shPDGFRα-4# and shPDGFRβ-4# exhibited the highest silencing efficiency, which were selected for future studies (Fig. S1C–F). Furthermore, the TGF-β1-induced increase in the PDGFR α/β levels was enhanced by PDGFR α/β overexpression but abolished by PDGFR α/β knockdown (Fig. 4B). Immunofluorescence examination revealed that the decreased desmin expression and increased α-SMA expression in TGF-β1-exposed pericytes could be strengthened by PDGFR α/β overexpression; however, opposite changes were observed after PDGFR α/β silencing (Fig. 4C). Consistently, Western blotting confirmed that the TGF-β1-induced downregulation of desmin and NG2 and the upregulation of α-SMA, collagen I, and PDGFR α/β. PDGFR α/β were reinforced by PDGFR α/β overexpression but abrogated by PDGFR α/β depletion (Fig. 4D, E). These findings indicated that the TGF-β1-induced transformation of pericytes into myofibroblasts was promoted by PDGFR α/β. Fig. 4PDGFR α/β facilitated pericyte–myofibroblast transformation induced by TGF-β1. Pericytes were transfected with pcDNA3.0 containing PDGFR α/β (PDGFR α/β) or shPDGFR α/β and then exposed to TGF-β1 (10 ng/mL) for 12 h. They were divided into the following groups: control, TGF-β1, TGF-β1 + vector, TGF-β1 + PDGFR α/β, TGF-β1 + shNC, TGF-β1 + shPDGFR α/β. A, B The mRNA expression of PDGFR α/β was examined via RT-qPCR. C Immunofluorescence staining was employed to evaluate desmin and α-SMA expressions. D, E Western blotting analysis of the protein levels of desmin, NG-2, α-SMA, collagen I, and PDGFR α/β. For A–E, $$n = 3$.$ $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$ ## GAS5 interacted with KDM5B to regulate PDGFR α/β expression We explored the underlying mechanism of GAS5-mediated modulation of PDGFR α/β expression during IPF progression. The FISH assay revealed that GAS5 was distributed in the cytoplasm and nucleus of pericytes (Fig. 5A). Furthermore, GAS5 was shown to be a little more located in the cytoplasm than in the nucleus (Fig. 5B). The silencing efficiency of shKDM5B-1-4# was confirmed by Western blotting, and shKDM5B-3#, which has the highest silencing efficiency, was selected (Additional file 1: Fig. S1G). Notably, we found that the PDGFR α/β levels were significantly elevated in KDM5B-depleted pericytes (Fig. 5C). The AnimalTFDB database predicted two binding sites (BS1-2) of KDM5B in the PDGFR α/β promoter (Fig. 5D, G). The ChIP assay validated that KDM5B directly bound to BS2 and BS3 in the PDGFR α/β promoter (Fig. 5E, H). In addition, the luciferase activity of WT PDGFR α/β was evidently increased after KDM5B silencing (Fig. 5F, I). As presented in Additional file 2: Fig. S2A, the binding of GAS5 to KDM5B was predicted using the RNA–Protein Interaction Prediction (RPISeq) database (http://pridb.gdcb.iastate.edu/RPISeq/index.html). The possibility of the binding between GAS5 and KDM5B was relatively high (RF = 0.85, SVM = 0.99). The secondary structure of GAS5 is presented in Additional file 2: Fig. S2B. The interaction between GAS5 and KDM5B was also confirmed by RNA pull-down and RIP assays (Fig. 5J, K). Furthermore, GAS5 depletion weakened the binding of KDM5B to the PDGFR α/β promoter (Fig. 5L). Also, the binding between GAS5 and KDM5B was enhanced by GAS5 overexpression but restrained by GAS5 deficiency (Fig. 5M). As presented in Additional file 2: Fig. S2C, RNA pull-down assay detected the binding of a series of GAS5 mutants (GAS5△1, GAS5△2, GAS5△3, and GAS5△4) to KDM5B. The results indicated that KDM5B could bind to GAS5△3 (Additional file 2: Fig. S2C). Meanwhile, the domain patterns of KDM5B binding to GAS5 were predicted using catRAPID (http://service.tartaglialab.com/update_submission/508192/f1d6a86b9b). We found that KDM5B domains JmJN(32-73AA), ARID(97-187AA), and JmjC(453-619AA) might bind to GAS5 (Fig. S2D). Further RNA pull-down assay determined the binding of various KDM5B splice variants (KDM5B△1 [loss of JmJN], KDM5B△2 [loss of ARID], KDM5B△3 [loss of JmjC]) to GAS5. The results indicated that the KDM5B JmjC (453-619AA) domain could directly interact with GAS5 (Additional file 2: Fig. S2D). Overall, PDGFR α/β transcription and expression were modulated by GAS5 through interaction with KDM5B.Fig. 5GAS5 recruited KDM5B to repress PDGFR α/β expression. A FISH assay was employed to observe the distribution of GAS5 in pericytes. B The nuclear and cytoplasmic expression of GAS5 in pericytes was detected via RT-qPCR. C Western blotting analysis of the protein levels of KDM5B and PDGFR α/β after KDM5B silencing. D/G The AnimalTFDB database predicted the binding sites of KDM5B in the promoter of PDGFR α/β. E/H ChIP assay was employed to examine the binding between KDM5B and the PDGFR α/β promoter. F/I Dual-luciferase assay confirmed the interaction between DKM5B and PDGFR α/β. J The interaction between GAS5 and KDM5B was assessed via RNA pull-down assay. K The direct binding between GAS5 and KDM5B was confirmed by RIP assay. L The binding between KDM5B and the PDGFR α/β promoter after GAS5 knockdown was detected via ChIP assay. M The interplay between GAS5 and KDM5B after GAS5 silencing or overexpression was examined via RNA pull-down assay. For A–M $$n = 3$.$ $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$ ## GAS5 promoted H3K4me2/3 demethylation in PDGFRα/β promoter by recruiting KDM5B To further elucidate the downstream mechanism, H3K4me$\frac{2}{3}$ attracted our attention. As presented in Fig. 6A, PDGFRα/β enrichment was significantly enhanced after precipitation with the H3K4me2 or H3K4me3 antibody. As presented in Additional file 1: Fig. S1H, the overexpression efficiency of KDM5B was confirmed by Western blotting. Furthermore, KDM5B overexpression upregulated H3Kme$\frac{2}{3}$ in pericytes, whereas KDM5B deficiency downregulated H3K4me$\frac{2}{3}$ (Fig. 6B). We also found that the interplay between PDGFRα/β and H3K4me$\frac{2}{3}$ was repressed by KDM5B silencing but promoted by KDM5B overexpression (Fig. 6C). Similar results indicated that the direct binding between PDGFRα/β and H3K4me$\frac{2}{3}$ was suppressed in GAS5-overexpressed cells but enhanced in GAS5-depleted cells (Fig. 6D). Collectively, GAS recruited KDM5B to modulate H3K4me$\frac{2}{3}$ demethylation in the PDGFR α/β promoter. Fig. 6GAS5 facilitated H3K4me$\frac{2}{3}$ demethylation to downregulate PDGFRα/β by recruiting KDM5B. A The binding between H3K4me$\frac{2}{3}$ and the PDGFR α/β promoter was confirmed by ChIP assay. B The protein levels of KDM5B and H3K4me$\frac{2}{3}$ after KDM5B knockdown or overexpression were detected via Western blotting. C ChIP assay analysis of the binding between H3K4me$\frac{2}{3}$ and the PDGFR α/β promoter in KDM5B-silenced or KDM5B-overexpressed pericytes. D The interaction between H3K4me$\frac{2}{3}$ and the PDGFR α/β promoter after GAS5 silencing or overexpression was determined via ChIP assay. For A–D, $$n = 3$.$ $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$ ## GAS5 repressed TGF-β1-induced pericyte–myofibroblast transformation via the KDM5B/PDGFR α/β signaling pathway To determine whether the KDM5B/PDGFRα/β signaling pathway was involved in the GAS5-mediated transformation of pericytes to myofibroblasts, pericytes were transfected with shGAS5, pcDNA3.0-KDM5B, or a combination of them. KDM5B was overexpressed in pericytes transfected with pcDNA3.0-KDM5B (Fig. 7A). As presented in Fig. 7B, the increased PDGFRα/β expression and reduced KDM5B expression in TGF-β1-exposed pericytes were augmented by GAS5 knockdown. However, KDM5B overexpression yielded opposite results and counteracted above shGAS5-mediated changes (Fig. 7B). Furthermore, KDM5B overexpression increased desmin and NG2 expressions and reduced α-SMA and collagen I expressions in TGF-β1-stimulated pericytes and abolished GAS5 depletion-induced reduction in desmin and NG2 expressions and elevation in α-SMA and collagen I expressions (Fig. 7C, D). These data indicated that the KDM5B/PDGFR α/β axis participated in the GAS5-mediated inhibition of pericyte–myofibroblast transformation. Fig. 7GAS5 repressed TGF-β1-induced pericyte–myofibroblast transformation by regulating the KDM5B/PDGFR α/β signaling pathway. Pericytes were transfected with pcDNA3.0 containing KDM5B (KDM5B) and/or shGAS5 and then exposed to TGF-β1 (10 ng/mL) for 12 h. They were divided into the following groups: control, TGF-β1, TGF-β1 + NC, TGF-β1 + shGAS5, TGF-β1 + KDM5B, and TGF-β1 + shGAS5 + KDM5B. A, B Western blotting analysis of the expressions of KDM5B and PDGFR α/β after KDM5B overexpression. C Immunofluorescence staining to evaluate the expressions of desmin and α-SMA. D The protein levels of desmin, NG-2, α-SMA, and collagen I were assessed via Western blotting. For A–D, $$n = 3$.$ $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$ ## GAS5 inhibited PF progression in bleomycin-induced mice in vivo Finally, we validated the function of GAS5 in the mouse model of PF in vivo. Histopathological examination revealed that the pathological changes and PF that occurred in the bleomycin group were attenuated by GAS5 overexpression but aggravated by GAS5 knockdown (Fig. 8A). Furthermore, GAS5 overexpression attenuated collagen deposition in the lungs of PF mice, whereas GAS5 deficiency yielded opposite results (Fig. 8B). Immunohistochemical staining showed that the bleomycin-induced reduction in desmin expression and increase in α-SMA and collagen I expressions were attenuated in the GAS5 overexpression group but aggravated in the shGAS5 group (Fig. 8C). RT-qPCR confirmed the downregulation of GAS5 in the bleomycin group and overexpression or silencing of GAS5 in the PF mouse model (Fig. 8D). Accordingly, GAS5 overexpression increased the low levels of desmin, NG-2, and KDM5B but decreased the high levels of α-SMA, collagen I, and PDGFRα/β in the lung tissues of PF mice. Conversely, the reverse effect of GAS5 knockdown was observed (Fig. 8E). The above results indicated that GAS5 attenuated lung fibrosis in bleomycin-induced mice in vivo by regulating the KDM5B/PDGFR α/β signaling pathway. Fig. 8GAS5 inhibited pulmonary fibrosis in the mouse model of pulmonary fibrosis in vivo. There were six experimental groups: control, bleomycin, bleomycin + NC, bleomycin + GAS5, bleomycin + shNC, and bleomycin + shGAS5. A, B The degree of pulmonary fibrosis and collagen deposition in the lungs of mice was assessed via H&E and Masson staining. C The expressions of desmin, α-SMA, and collagen I in the lung tissues were determined via immunohistochemical staining. D RT-qPCR analysis of GAS5 expression in different groups. E The protein levels of desmin, NG-2, α-SMA, collagen I, KDM5B, and PDGFR α/β in the lungs were examined via Western blotting. For A–E, $$n = 6$.$ $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$ ## Discussion To date, IPF remains a progressive and fatal disorder, with increasing incidence and prevalence rates (Saito et al. 2019). The median survival of IPF is 2–3 years after diagnosis (Sharif 2017). It has a high mortality rate, and the available treatments for this disease are few; thus, safe and effective therapeutic agents need to be developed. Growing evidence has indicated that pericytes are the origin of myofibroblasts that contribute to the incidence of IPF (Yamaguchi et al. 2020); thus, elucidation of the mechanism underlying pericyte–myofibroblast transformation is crucial for the development of novel treatment for IPF. This study found that GAS5 was downregulated whereas PDGFR α/β was upregulated in the lungs of patients and mice with IPF. Moreover, GAS5 overexpression or PDGFR α/β knockdown inhibited TGF-β1-induced pericyte–myofibroblast differentiation. GAS5 recruited KDM5B to repress PDGFR α/β expression via H3K4me$\frac{2}{3}$ demethylation. Finally, GAS5 overexpression attenuated lung fibrosis in bleomycin-induced mice by regulating the KDM5B/PDGFR α/β signaling pathway. Our observations indicated that GAS5 is a novel target for the treatment of IPF. LncRNAs have been found to participate in IPF progression. For example, Yang et al. found that lncRNA ZFAS1 triggered ferroptosis to promote IPF progression via the miR-150-5p/SLC38A1 signaling pathway (Yang et al. 2020). The influence of GAS5 on fibrosis has been widely documented in multiple tissues. For instance, GAS5 was reported to attenuate renal interstitial fibrosis in diabetic rats by regulating the EZH2/MMP9 signaling pathway (Zhang et al. 2020). A previous study demonstrated that GAS5 attenuated renal fibrosis by modulating the Smad3/miRNA-142-5p axis (Zhang et al. 2021). Tao et al. showed that GAS5 suppression by MeCP2 led to cardiac fibrosis (Tao et al. 2020). To date, the function of GAS5 in IPF has not yet been elucidated. The mouse model of intratracheal bleomycin administration is the most well-characterized animal model currently available for preclinical PF research (Jenkins et al. 2017). Thus, we used bleomycin to create an in vivo PF mouse model in this study. Our data indicated that GAS5 was downregulated in the lungs of patients and mice with IPF. Furthermore, we observed an enhanced expression of PDGFRα/β in the lungs of these patients. Excessive activation of the PDGFR α/β signaling pathway may result in tissue fibrosis (Ieronimakis et al. 2016). It has also been suggested that inactivation of the PDGFRα/β signaling pathway is essential for delaying IPF progression (Kishi et al. 2018; Vuorinen et al. 2007). Fibrosis progression caused by myofibroblast activation may destroy the normal structure and function of the lung tissue. Inhibiting myofibroblast activation is an effective strategy for treating fibrosis (Morelli et al. 2019; Feng et al. 2022). Wang et al. demonstrated that inhibition of exosomal microRNA-92a restrained the activation of cardiac fibroblasts, thereby attenuating cardiac fibrosis (Wang et al. 2020). TGF-β1, a critical inducer of tissue fibrosis (including IPF), can activate myofibroblasts and induce excessive ECM deposition (Hu et al. 2018; Zainal Abidin et al. 2021; Jordan et al. 2021). Inactivation of the TGFβ signaling pathway was proven to reduce the level of fibrosis-related proteins in myofibroblasts (Yousefi et al. 2021). α-SMA is a common marker for myofibroblasts, and ECM deposition is predominantly indicated by increased collagen I expression (Li et al. 2019). Herein, our findings indicated that GAS5 overexpression or PDGFRα/β deficiency effectively inhibited TGF-β1-induced pericyte–myofibroblast transformation by repressing α-SMA and collagen I expressions and elevating pericyte markers desmin and NG2 expression. Conversely, GAS5 silencing or PDGFRα/β overexpression yielded opposite results. Collectively, loss of GAS5 or PDGFRα/β overexpression promoted IPF progression by inducing pericyte–myofibroblast transformation. In view of the important function of GAS5 and PDGFRα/β in IPF, we further elucidated the underlying regulatory mechanism. LncRNAs have been demonstrated to act as scaffolds, decoys, or miRNA sequesters, thereby exerting their molecular functions (Wang and Chang 2011). GAS5 has been shown to affect various signaling cascades by recruiting protein for the formation of RNA–protein complexes. For instance, it has been suggested that GAS5 inactivates the TGF-β/Smad signaling pathway during skin fibrosis by facilitating PPM1A-mediated Smad3 dephosphorylation (Tang et al. 2020). Interestingly, bioinformatics analysis revealed that both GAS5 and PDGFRα/β bind to KDM5B. KDM5B, as a histone demethylase, plays pivotal roles in H3K4me$\frac{2}{3}$ demethylation (Zhang et al. 2014). H3K4me$\frac{2}{3}$ residue has been recognized as a gene transcription initiation site, and H3K4me$\frac{2}{3}$ demethylation indicates the inhibition of transcription, suggesting its active role in suppressing gene expression (Zheng et al. 2019). A previous study demonstrated that KDM5B participated in osteoblast differentiation regulation by epigenetic modulation of Runx2* gene via H3K4me3 (Rojas et al. 2015). Here, we found that GAS5 acted as a scaffold that recruited KDM5B to the PDGFRα/β promoter and consequently repressed PDGFRα/β expression via H3K4me$\frac{2}{3}$ demethylation. Therefore, GAS5 inhibited pericyte–myofibroblast transformation by recruiting KDM5B to repress PDGFRα/β expression. ## Conclusion In conclusion, the current study demonstrated that GAS5 overexpression restrained TGF-β1-induced pericyte–myofibroblast transformation and bleomycin-induced PF by reducing PDGFR α/β expression through interaction with KDM5B to promote H3K4me$\frac{2}{3}$ demethylation, suggesting a negative regulatory role of GAS5 in the etiology of IPF (Fig. 9). Our findings identified GAS5 as a potential therapeutic target for IPF.Fig. 9A schematic map illustrating the role and mechanism of lncRNA GAS5 in pulmonary fibrosis. GAS5 recruited KDM5B to facilitate H3K4me$\frac{2}{3}$ demethylation that reduced PDGFR α/β transcription and expression, thus inhibiting pulmonary fibrosis ## Supplementary Information Additional file 1: Figure S1. Confirmation of the silencing or overexpression efficiency. ( A) RT-qPCR analysis of the expression of GAS5 in pericytes after transfection with shGAS5-1–4#. ( B) RT-qPCR analysis of the level of GAS5 in pericytes after transfection with pcDNA3.0-GAS5. ( C and D) Western blotting analysis of the levels of PDGFRα/β in pericytes transfected with shPDGFRα/β-1-4#. ( E and F) The protein levels of PDGFRα/β in pericytes after transfection with pcDNA3.0-PDGFR α/β were evaluated via Western blotting. ( G) Western blotting analysis of the level of KDM5B in pericytes transfected with shKDM5B-1-4#. ( H) The protein level of KDM5B in pericytes after transfection with pcDNA3.0-KDM5B was detected via Western blotting. * $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001.$Additional file 2: Fig. S2. Direct interaction between GAS5 and KDM5B. (A) The RNA–Protein Interaction Prediction (RPISeq) database predicted the binding sites between GAS5 and KDM5B. (B) Secondary structure of GAS5. ( C) The binding of a series of GAS5 mutants (GAS5△1, GAS5△2, GAS5△3, and GAS5△4) to KDM5B was validated via RNA pull-down assay. ( D) RNA pull-down assay analysis of the binding of various KDM5B splice variants (KDM5B△1(loss of JmJN), KDM5B△2(loss of ARID), and KDM5B△3(loss of JmjC)) to GAS5. ## References 1. Chen T. LncRNA CTD-2528L19.6 prevents the progression of IPF by alleviating fibroblast activation. Cell Death Dis* (2021) 12 600. DOI: 10.1038/s41419-021-03884-5 2. Chou YH. Methylation in pericytes after acute injury promotes chronic kidney disease. J Clin Invest (2020) 130 4845-4857. DOI: 10.1172/JCI135773 3. Djudjaj S, Boor P. Cellular and molecular mechanisms of kidney fibrosis. Mol Aspects Med (2019) 65 16-36. DOI: 10.1016/j.mam.2018.06.002 4. Feng Y. MicroRNA-130a attenuates cardiac fibrosis after myocardial infarction through TGF-beta/Smad signaling by directly targeting TGF-beta receptor 1. Bioengineered (2022) 13 5779-5791. DOI: 10.1080/21655979.2022.2033380 5. Hu HH. New insights into TGF-beta/Smad signaling in tissue fibrosis. Chem Biol Interact (2018) 292 76-83. DOI: 10.1016/j.cbi.2018.07.008 6. Huang D. JARID1B promotes colorectal cancer proliferation and Wnt/beta-catenin signaling via decreasing CDX2 level. Cell Commun Signal (2020) 18 169. DOI: 10.1186/s12964-020-00660-4 7. Hung CF. Lung pericyte-like cells are functional interstitial immune sentinel cells. Am J Physiol Lung Cell Mol Physiol (2017) 312 L556-L567. DOI: 10.1152/ajplung.00349.2016 8. Ieronimakis N. PDGFRalpha signalling promotes fibrogenic responses in collagen-producing cells in Duchenne muscular dystrophy. J Pathol (2016) 240 410-424. DOI: 10.1002/path.4801 9. Jenkins RG. An official american thoracic society workshop report: use of animal models for the preclinical assessment of potential therapies for pulmonary fibrosis. Am J Respir Cell Mol Biol (2017) 56 667-679. DOI: 10.1165/rcmb.2017-0096ST 10. Jordan NP. MiR-126-3p is dynamically regulated in endothelial-to-mesenchymal transition during fibrosis. Int J Mol Sci (2021) 22 8629. DOI: 10.3390/ijms22168629 11. Kishi M. Blockade of platelet-derived growth factor receptor-beta, not receptor-alpha ameliorates bleomycin-induced pulmonary fibrosis in mice. PLoS ONE (2018) 13 e0209786. DOI: 10.1371/journal.pone.0209786 12. Li X. Antifibrotic mechanism of cinobufagin in bleomycin-induced pulmonary fibrosis in mice. Front Pharmacol (2019) 10 1021. DOI: 10.3389/fphar.2019.01021 13. Maher TM. Global incidence and prevalence of idiopathic pulmonary fibrosis. Respir Res (2021) 22 197. DOI: 10.1186/s12931-021-01791-z 14. Morelli MB, Shu J, Sardu C, Matarese A, Santulli G. Cardiosomal microRNAs are essential in post-infarction myofibroblast phenoconversion. Int J Mol Sci (2019) 21 201. DOI: 10.3390/ijms21010201 15. Pan X. Lysine-specific demethylase-1 regulates fibroblast activation in pulmonary fibrosis via TGF-beta1/Smad3 pathway. Pharmacol Res (2020) 152 104592. DOI: 10.1016/j.phrs.2019.104592 16. Poulet C. Exosomal long Non-coding RNAs in lung diseases. Int J Mol Sci (2020) 21 3580. DOI: 10.3390/ijms21103580 17. Raghu G, Weycker D, Edelsberg J, Bradford WZ, Oster G. Incidence and prevalence of idiopathic pulmonary fibrosis. Am J Respir Crit Care Med (2006) 174 810-816. DOI: 10.1164/rccm.200602-163OC 18. Rojas A. Epigenetic control of the bone-master Runx2 Gene during osteoblast-lineage commitment by the histone demethylase JARID1B/KDM5B. J Biol Chem (2015) 290 28329-28342. DOI: 10.1074/jbc.M115.657825 19. Saito S, Alkhatib A, Kolls JK, Kondoh Y, Lasky JA. Pharmacotherapy and adjunctive treatment for idiopathic pulmonary fibrosis (IPF). J Thorac Dis (2019) 11 S1740-S1754. DOI: 10.21037/jtd.2019.04.62 20. Salminen A, Kaarniranta K, Kauppinen A. Hypoxia-inducible histone lysine demethylases: impact on the aging process and age-related diseases. Aging Dis (2016) 7 180-200. DOI: 10.14336/AD.2015.0929 21. Sharif R. Overview of idiopathic pulmonary fibrosis (IPF) and evidence-based guidelines. Am J Manag Care (2017) 23 S176-S182. PMID: 28978212 22. Tang R. LncRNA GAS5 attenuates fibroblast activation through inhibiting Smad3 signaling. Am J Physiol Cell Physiol (2020) 319 C105-C115. DOI: 10.1152/ajpcell.00059.2020 23. Tao H, Shi P, Zhao XD, Xuan HY, Ding XS. MeCP2 inactivation of LncRNA GAS5 triggers cardiac fibroblasts activation in cardiac fibrosis. Cell Signal (2020) 74 109705. DOI: 10.1016/j.cellsig.2020.109705 24. Vuorinen K, Gao F, Oury TD, Kinnula VL, Myllarniemi M. Imatinib mesylate inhibits fibrogenesis in asbestos-induced interstitial pneumonia. Exp Lung Res (2007) 33 357-373. DOI: 10.1080/01902140701634827 25. Wang KC, Chang HY. Molecular mechanisms of long noncoding RNAs. Mol Cell (2011) 43 904-914. DOI: 10.1016/j.molcel.2011.08.018 26. Wang X, Morelli MB, Matarese A, Sardu C, Santulli G. Cardiomyocyte-derived exosomal microRNA-92a mediates post-ischemic myofibroblast activation both in vitro and ex vivo. ESC Heart Fail (2020) 7 284-288. PMID: 31981320 27. Wang YC. Exosomal miR-107 antagonizes profibrotic phenotypes of pericytes by targeting a pathway involving HIF-1alpha/Notch1/PDGFRbeta/YAP1/Twist1 axis in vitro. Am J Physiol Heart Circ Physiol (2021) 320 H520-H534. DOI: 10.1152/ajpheart.00373.2020 28. Xie H. Low let-7d exosomes from pulmonary vascular endothelial cells drive lung pericyte fibrosis through the TGFbetaRI/FoxM1/Smad/beta-catenin pathway. J Cell Mol Med (2020) 24 13913-13926. DOI: 10.1111/jcmm.15989 29. Yamaguchi M. Pericyte-myofibroblast transition in the human lung. Biochem Biophys Res Commun (2020) 528 269-275. DOI: 10.1016/j.bbrc.2020.05.091 30. Yang Y. lncRNA ZFAS1 promotes lung fibroblast-to-myofibroblast transition and ferroptosis via functioning as a ceRNA through miR-150-5p/SLC38A1 axis. Aging (albany NY) (2020) 12 9085-9102. DOI: 10.18632/aging.103176 31. Yin W, Han J, Zhang Z, Han Z, Wang S. Aloperine protects mice against bleomycin-induced pulmonary fibrosis by attenuating fibroblast proliferation and differentiation. Sci Rep (2018) 8 6265. DOI: 10.1038/s41598-018-24565-y 32. Yousefi F, Soltani BM, Rabbani S. MicroRNA-331 inhibits isoproterenol-induced expression of profibrotic genes in cardiac myofibroblasts via the TGFbeta/smad3 signaling pathway. Sci Rep (2021) 11 2548. DOI: 10.1038/s41598-021-82226-z 33. Yu QY, Tang XX. Irreversibility of pulmonary fibrosis. Aging Dis (2022) 13 73-86. DOI: 10.14336/AD.2021.0730 34. Zainal Abidin SAI. Myofibroblast transdifferentiation is associated with changes in cellular and extracellular vesicle miRNA abundance. PLoS ONE (2021) 16 e0256812. DOI: 10.1371/journal.pone.0256812 35. Zhang Y. The PHD1 finger of KDM5B recognizes unmodified H3K4 during the demethylation of histone H3K4me2/3 by KDM5B. Protein Cell (2014) 5 837-850. DOI: 10.1007/s13238-014-0078-4 36. Zhang L, Zhao S, Zhu Y. Long noncoding RNA growth arrest-specific transcript 5 alleviates renal fibrosis in diabetic nephropathy by downregulating matrix metalloproteinase 9 through recruitment of enhancer of zeste homolog 2. FASEB J (2020) 34 2703-2714. DOI: 10.1096/fj.201901380RR 37. Zhang YY, Tan RZ, Yu Y, Niu YY, Yu C. LncRNA GAS5 protects against TGF-beta-induced renal fibrosis via the Smad3/miRNA-142-5p axis. Am J Physiol Renal Physiol (2021) 321 F517-F526. DOI: 10.1152/ajprenal.00085.2021 38. Zheng YC. Lysine demethylase 5B (KDM5B): a potential anti-cancer drug target. Eur J Med Chem (2019) 161 131-140. DOI: 10.1016/j.ejmech.2018.10.040 39. Zuo B. Yiqi Huayu decoction alleviates bleomycin-induced pulmonary fibrosis in rats by inhibiting senescence. Front Pharmacol (2022) 13 1033919. DOI: 10.3389/fphar.2022.1033919
--- title: Curcumin-dependent phenotypic transformation of microglia mediates resistance to pseudorabies-induced encephalitis authors: - Luqiu Feng - Guodong Luo - Yuhang Li - Chen Zhang - Yuxuan Liu - Yanqing Liu - Hongyue Chen - Daoling He - Yan Zhu - Ling Gan journal: Veterinary Research year: 2023 pmcid: PMC10015794 doi: 10.1186/s13567-023-01149-x license: CC BY 4.0 --- # Curcumin-dependent phenotypic transformation of microglia mediates resistance to pseudorabies-induced encephalitis ## Abstract Pseudorabies virus (PRV) causes viral encephalitis, a devastating disease with high mortality worldwide. Curcumin (CUR) can reduce inflammatory damage by altering the phenotype of microglia; however, whether and how these changes mediate resistance to PRV-induced encephalitis is still unclear. In this study, BV2 cells were infected with/without PRV for 24 h and further treated with/without CUR for 24 h. The results indicated that CUR promoted the polarization of PRV-infected BV2 cells from the M1 phenotype to the M2 phenotype and reversed PRV-induced mitochondrial dysfunction. Furthermore, M1 BV2 cell secretions induced signalling pathways leading to apoptosis in PC-12 neuronal cells, and this effect was abrogated by the secretions of M2 BV2 cells. RNA sequencing and bioinformatics analysis predicted that this phenotypic shift may be due to changes in energy metabolism. Furthermore, Western blot analysis showed that CUR inhibited the increase in AMP-activated protein kinase (AMPK) phosphorylation, glycolysis, and triacylglycerol synthesis and the reduction in oxidative phosphorylation induced by PRV infection. Moreover, the ATP levels in M2 BV2 cells were higher than those in M1 cells. Furthermore, CUR prevented the increase in mortality, elevated body temperature, slowed growth, nervous system excitation, brain tissue congestion, vascular cuffing, and other symptoms of PRV-induced encephalitis in vivo. Thus, this study demonstrated that CUR protected against PRV-induced viral encephalitis by switching the phenotype of BV2 cells, thereby protecting neurons from inflammatory injury, and this effect was mediated by improving mitochondrial function and the AMPK/NF-κB p65-energy metabolism-related pathway. ### Supplementary Information The online version contains supplementary material available at 10.1186/s13567-023-01149-x. ## Introduction Pseudorabies virus (PRV) is a neurotropic virus similar to herpes simplex virus type I [1] that typically infects domestic and wild animals. However, it has recently been reported that PRV infection in humans manifests as respiratory dysfunction and acute encephalitis [2–4], posing a huge challenge to public health. PRV infection of the respiratory tract is followed by viral replication in peripheral tissues, such as the muscle mucosa, which causes infection, invasion of peripheral nerves, and transmission to the central nervous system (CNS) through synapses, leading to viral encephalitis and death [4–6]. Microglia are resident mononuclear macrophages in the CNS and have long pseudopodia [7, 8]. These cells typically play immunological roles in response to infectious pathogens in the CNS and continuously scan the CNS and sense changes in the microenvironment through the sensome. Furthermore, these cells respond rapidly to pathological triggers, such as the invasion of pathogenic microorganisms, neuronal death, and protein aggregation, and eliminate factors through phagocytosis and degradation [9–11]. After detecting pathogenic microorganisms in the CNS, microglia first polarize to the classically activated proinflammatory M1 phenotype, and these cells express CD16, CD32, CD40, and major histocompatibility complex (MHC) class II markers and secrete tumour necrosis factor-α (TNF-α), interleukin (IL)-6, nitric oxide (NO), and reactive oxygen species (ROS) to kill the pathogenic microorganisms [11–13]. After clearing the pathogenic microorganisms, microglia polarize from the classically activated M1 phenotype to the alternately activated M2 phenotype, and these cells express arginase-1 (ARG-1) and mannose receptor (CD206) markers and secrete IL-4, IL-10, and transforming growth factor-β (TGF-β) to promote inflammation resolution [14]. This enables the phenotypic inactivation of proinflammatory cells and plays a role in the repair and maintenance of the CNS to re-establish homeostasis [13, 15, 16]. Acute microglial activation is widely believed to be beneficial under neuroinflammatory conditions by promoting the clearance of neurotoxic agents and restoring tissue homeostasis [17]. However, if inflammation cannot be dissipated in time, the surrounding tissues or cells such as neurons will be lost, and the damaged tissue will develop into an internal stimulus, further exacerbating inflammatory injury [18]. Therefore, it is particularly important to switch from the M1 to M2 phenotype in a timely manner. Curcumin (CUR), which is a natural compound isolated from turmeric, is excellent for treating neuroinflammation and neurological diseases because of its anti-inflammatory properties and ability to cross the blood‒brain barrier [19, 20]. CUR can interact with multiple molecular targets in microglia, thereby promoting phenotypic shifts and exerting anti-inflammatory effects. These molecular targets include nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) [20, 21], Toll-like receptor-4 (TLR-4) [20, 22], haem oxygenase-1 (HO-1) [23], myeloid differentiation primary response 88 (MyD88) [20], phosphoinositide 3-kinase/protein kinase B (PI3K/Akt) [24], and p38 mitogen-activated protein kinase (p38 MAPK) [25]. The phenotypic and functional changes in macrophages are accompanied by dramatic changes in energy metabolism pathways in mitochondria [26, 27]. AMP-activated protein kinase (AMPK) is a key regulator of energy metabolism during inflammation [28]. Proinflammatory M1 microglia mainly rely on aerobic glycolysis and fatty acid synthesis (FAS) to produce adenosine triphosphate (ATP), while anti-inflammatory M2 microglia rely on oxidative phosphorylation (OXPHOS) and fatty acid oxidation (FAO) [26, 29]. In this study, we aimed to assess the role of CUR in mediating resistance to PRV-induced encephalitis. Furthermore, we investigated the contribution of different components, such as mitochondrial function- and AMPK/NF-κB p65-energy metabolism-related pathways, in driving microglial phenotypic transitions. ## Cell culture, experimental design, and drug administration BV2 cells (Rio de Janeiro Cell Bank, Portugal), PC-12 cells, and PK-15 cells (Chinese Academy of Sciences Cell Bank, China) were cultured in Dulbecco’s modified Eagle medium (DMEM; 11,995,065, Gibco, USA) supplemented with $10\%$ foetal bovine serum (FBS; 10099141C, Gibco, USA) and $1\%$ penicillin/streptomycin (15,140,122, Gibco, USA). The cells were grown in a humidified incubator at 37 °C with $5\%$ CO2. BV2 cells were infected with PRV for 24 h, followed by CUR (20 μM) (A600346, Sangon Biotech, China), Compound C (AMPK inhibitor, 2.5 μM; HY-13418A, MCE Med Chem Express, USA), or small interfering RNA (siRNA) treatment separately or in combination in serum-free medium for 24 h. The supernatant of the BV2 cells treated with/without PRV and/or CUR was collected and further used to treat PC-12 cells for 24 h. The morphology of BV2 and PC-12 cells was observed using a light microscope (Olympus, Japan). ## Model organisms Ten healthy BALB/c mice (20–35 g) and 100 Sprague–Dawley rats (6–8 week-old; half male and half female; specific pathogen-free animals; 200 ± 20 g) with no history of PRV infection or immunization were purchased from Chongqing Medical University. All mice and rats were maintained under standard conditions (23 ± 2 °C, 60–$70\%$ relative humidity, 12 h light/12 h dark cycle) with ad libitum access to food and water. ## Primary microglial culture Primary microglia were isolated from postnatal Day 1 BALB/C mice. Briefly, the whole brains of BALB/C mice were minced with ophthalmic scissors and filtered using a 200-mesh pore screen under sterile conditions. The mixed cells were seeded in T25 cell flasks in DMEM/F12 (C11330500BT, Gibco, USA) supplemented with $10\%$ FBS (10099141C, Gibco, USA) and $1\%$ penicillin/streptomycin (15,140,122, Gibco, USA). On Day 14, the mixed cells were detached with $0.25\%$ trypsin (25,200,072, Gibco, USA). The cell suspension was collected in centrifuge tubes and centrifuged at 1500 × g (25 ℃) for 8 min. The concentration of the cell suspension was adjusted to 1.0 × 106/mL with cell culture medium for subsequent analysis. ## Viral infection and titration The PRV strain was purchased from the China Veterinary Microorganism Collection and Management Center (preservation number: CVCCAV25) and propagated as previously described [30]. Briefly, PK-15 cells were grown to $90\%$ confluence and infected with PRV at various multiplicities of infections. After 2 h, the inoculum was removed by aspiration, and the cells were washed twice with PBS. After 3 days, the cytopathic effects on PK-15 cells were observed using an inverted microscope (Olympus, Japan). The median tissue culture infectious dose (TCID50) was calculated using the Reed–Muench method [31]. ## Transient transfection with siRNA Several siRNA fragments for AMPKα1 were designed, and those that efficiently inhibited its translation were selected (sense: 5'-GCCGACCCAAUGAUAUCAUTT-3' and antisense: 5'-AUGAUAUCAUUGGGUCGGCTT-3'; Gene Pharma, China). The designed siRNAs were all modified with 2'-O-methyl and 5'-carboxyfluorescein fluorescent labels to avoid degradation and assess transfection efficiency, respectively. The inhibitory efficiency of the siRNA probes was assessed by Western blot analysis of AMPKα1 protein levels. BV2 cells were cultured for 24 h and grown to approximately $70\%$ confluence. Subsequently, the cells were transfected with scrambled siRNA or AMPKα1 siRNA using GP-transfect-Mate (G04008, Gene Pharma, China) according to the manufacturer’s protocol. All assays were performed 24 h after the transfection of siRNA. ## Cell viability, lactic dehydrogenase (LDH) activity and malondialdehyde (MDA) level assessment BV2 and PC-12 cell viability was assessed using CCK-8 reagent (BS350B, Biosharp, China). BV2 cells were seeded at a density of 5 × 104 cells/well in 96-well plates, incubated overnight and treated with varying concentrations of CUR or Compound C for 24 h. Alternatively, PC-12 cells were seeded at a density of 5 × 104 cells/well in 96-well plates, incubated overnight and exposed to the conditioned media (CM) of different BV2 cell phenotypes. CM was obtained from the supernatant of BV2 cells treated with/without PRV and/or CUR. After the PC-12 cells were disrupted and homogenized by a high-speed disperser, LDH activity and MDA levels were determined by a colorimetric method using LDH and MDA detection kits, respectively (A003-4–1, A020-2–2, Nanjing Jiancheng Biological Company, China) according to the manufacturer’s protocol. ## NO analysis BV2 cells were seeded in 6-well plates at a density of 5 × 105 cells/mL and incubated overnight. BV2 cells were infected with different PRV titres for 24 h, or the cells were infected with 1.66 × 106 TCID50 PRV for 6, 12, and 24 h or 1.66 × 106 TCID50 PRV for 24 h, followed by CUR treatment for 6, 12, and 24 h. The supernatant was collected, and NO levels were estimated using an NO detection kit (A003-4–1, A020-2–2, Nanjing Jiancheng Biological Company, China) according to the manufacturer’s protocol. ## Intracellular reactive oxygen species analysis BV2 cells were seeded in 6-well plates at a density of 5 × 105 cells/mL, incubated overnight and infected with different PRV titres for 24 h, or the cells were infected with 1.66 × 106 TCID50 PRV for 6, 12, and 24 h or 1.66 × 106 TCID50 PRV for 24 h and then treated with CUR for 6, 12, and 24 h. The fluorescent probe 2,7-dichlorofluorescein diacetate (DCFH-DA) (10 μM) (E004-1–1, Nanjing Jiancheng Biological Company, China) was added to the cells and incubated for approximately 1 h in the dark. The cells were then washed once with PBS, and the fluorescence was measured using a microplate reader (Tecan, Switzerland) at an excitation wavelength of 500 nm and an emission wavelength of 525 nm. After the fluorescence was measured, the cells were observed under a fluorescence microscope (Olympus, Japan). Unstained cells were used as blanks to normalize the fluorescence intensity in the different treatment groups. ## Analysis of cytokines The levels of cytokines released from BV2 cells and primary microglia in culture were quantified using the following commercial enzyme-linked immunosorbent assay (ELISA) kits according to the manufacturer’s instructions: mouse TNF-α uncoated ELISA kit (88–7324-22, Thermo Fisher, USA), mouse IL-6 uncoated ELISA kit (88–7064-22, Thermo Fisher, USA), mouse IL-4 uncoated ELISA kit (88–7044-22, Thermo Fisher, USA), and mouse IL-10 uncoated ELISA kit (88–7105-22, Thermo Fisher, USA). ## Flow cytometry BV2 cells and primary microglia were collected after the indicated treatments and stained with fluorescence-conjugated monoclonal antibodies according to the manufacturer’s instructions. Cultured and treated cells were resuspended in PBS solution. Then, a 95-μL cell suspension was obtained, and 5 μL of FITC-anti-mouse CD40 (11–0402-81, Thermo Fisher, USA), APC-anti-mouse CD206 (17–2061-82, Thermo Fisher, USA), or CoraLite® 488 anti-mouse CD11b (CL488-65,055, Proteintech, China) was added. The suspension was incubated for 2 h at 4 °C in the dark, fixed with $2\%$ paraformaldehyde for 30 min and permeabilized with $0.1\%$ Triton X-100 for 15 min. Finally, the cells were blocked with $1\%$ bovine serum albumin (BSA, BS114-100g, Biosharp, China) for 1 h and washed three times with PBS buffer. Subsequently, 5 μL of PE-anti-mouse CD$\frac{16}{32}$ (12–0161-82, Thermo Fisher, USA) or PE-anti-mouse Arg-1 (12–3697-82, Thermo Fisher, USA) and CoraLite® 647 anti-mouse MHC Class II (CL647-65,122, Proteintech, China) were added, and the suspensions were incubated at 4 °C for 2 h in the dark. Then, the cells were washed three times with prechilled $1\%$ BSA solution. Finally, flow cytometry was performed using an Accuri C6 flow cytometer (BD Biosciences, San Jose, CA, USA). The data were analysed using FlowJo™ v10 software (for Windows) Version 10 (Ashland, BD Life Sciences). ## Transmission electron microscopy (TEM) Mitochondrial morphology was examined using TEM. BV2 cells were added to a $2.5\%$ glutaraldehyde fixative solution and stored overnight at 4 °C. Images were captured using TEM (Hitachi-7500, Japan). ## Mitochondrial membrane potential (MMP) measurement BV2 cells were cultured overnight in 12-well plates at a density of 2.5 × 105 cells/well. MMP was measured using a JC-1 assay kit (BL726A, Biosharp, China). BV2 cells were incubated with 10 μM JC-1 dye for 20 min at 37 °C (shielded from light) and washed with PBS prior to being evaluated using a microplate reader (Tecan, Switzerland) and fluorescence microscope (Olympus, Japan). In normal mitochondria, JC-1 forms aggregates that emit red fluorescence (561 nm). Following a decrease or loss of MMP, aggregated JC-1 is released into the cytoplasm in its monomeric form, which emits green fluorescence (488 nm). ## Transcriptome sequencing analysis and real-time PCR RNA-seq analysis was performed by Allwegene (Allwegene, China). RNA (three biological replicates per group) was extracted from BV2 cells using TRIzol reagent (B511311, Sangon Biotech, China) according to the manufacturer’s instructions, and a cDNA library was prepared using a PerfectStart® Uni RT&qPCR Kit (TransGene Biotech, China) with 1000 ng of total RNA. cDNA, primers, PerfectStart Green qPCR SuperMix, and nuclease-free water were combined into a 20 µL reaction system to perform quantitative real-time PCR on a real-time PCR system (QuantStudio 5, Thermo Scientific, USA). RNA-seq was performed using the Illumina HiSeq 2500 platform. A corrected p value of 0.05 and a log2 (fold change) of 1 were set as thresholds for significantly differentially expressed genes (DEGs). Gene Ontology (GO) and Kyoto Encyclopaedia of Genes and Genomes (KEGG) pathway enrichment functional analyses were performed on the DEGs in each group by using GOSeq software and KOBAS software [32, 33]. The following primer pairs were designed to measure transcript abundance relative to β-actin (B661302, Sangon Biotech, China) as an internal reference: malonyl-CoA-acyl carrier protein transacylase (Mcat): TCTGGTTTCTGTCTACTCCAAC (F) and CCTTTCGTATATGGCATGCATC (R); lactate dehydrogenase A (Ldha): AAGACTACTGTGTAACTGCGAA (F) and ACTTGAAGATGTTCACGTTTCG (R); hexokinase 1 (Hk1): ATTAAGAAGCGAGGGGACTATG (F) and CTCCCCATTCCGTGTTAATACA (R); phosphofructokinase (Pfkl): ACGGTATACATCGTGCATGAT (F) and GATGTTGTAGGTGCGGAGATTC (R); and histocompatibility 2, class II, locus Mb1 (H2dmb1): CATGGGCCGAAAATTTTTCAAG (F) and CTCCCTTGTGTTAAAAGGTGTG (R). ## Detection of the oxygen consumption rate (OCR) in BV2 cells The cellular OCR was measured using an OCR assay kit (600,800; Cayman Chemical, USA). BV2 cells were seeded into 96-well plates and incubated overnight. After treatment, the medium was discarded, and 10 μL of the phosphorescent oxygen probe and 100 μL of LHS mineral oil assay reagent (prewarmed at 37 °C) were added to each well. A microplate reader (Tecan, Switzerland) was used, and the ratiometric time-resolved fluorescence (lifetime) measurement mode at an excitation wavelength of 380 nm and emission wavelength of 650 nm was selected for dynamic measurement for 2 h. ## Detection of the extracellular acidification rate (ECAR) in BV2 cells The ECAR was measured using a glycolytic cell-based assay kit (600,450; Cayman Chemical, USA). BV2 cells were seeded into 96-well plates and incubated overnight. After the different groups were treated, the medium was collected and centrifuged at 400 × g for 5 min at 4 °C. An aliquot of the supernatant from each sample was added to the reaction solution and gently shaken for 30 min at 25 °C on an orbital shaker. The absorbance was measured at 490 nm using a microplate reader (Tecan, Switzerland). ## Determination of ATP levels in BV2 cells ATP levels were measured using a firefly luciferase-based ATP detection kit (S0026, Beyotime, China). After being washed with PBS, BV2 cells were lysed using ATP detection lysis buffer, followed by centrifugation at 12 000 × g for 5 min at 4 °C, and the supernatant was collected. Then, 100 µL of the supernatant was mixed with 100 µL of ATP detection solution in a 1.5 mL tube. Luminescence (corresponding to total ATP levels) was immediately measured in relative light units (RLU) (nmol/mg) using a Turner Biosystems luminometer (Tecan, Switzerland). Finally, the ATP level of each sample was determined according to the RLU value of the standard sample and normalized to the protein concentration. ## Western blotting BV2 and PC-12 cells were lysed in ice-cold radioimmunoprecipitation assay buffer (RIPA; BL504A, Biosharp, China), 1 × complete protease inhibitor (BL612A, Biosharp, China), and a phosphatase inhibitor cocktail (BL615A, Biosharp, China). Cell homogenates (20 μg/well) were loaded onto $8\%$, $10\%$, or $12\%$ SDS polyacrylamide gels under denaturing conditions. Proteins were resolved electrophoretically at 100 mA for 90 min and transferred onto a 0.45 μm polyvinylidene fluoride (PVDF) membrane (Millipore, USA) at 200 V for 60 min (Power Pack; Bio-Rad Laboratories, USA). Furthermore, the membrane was blocked with $5\%$ nonfat dry milk or $5\%$ BSA in tris-buffered saline containing Tween-20 (TBST) at 25 ℃ and then incubated overnight at 4 °C with the following antibodies: rabbit anti-AMPK (1:2000, bs-1115R, Bioss, China), phosphor(p)-AMPKThr172 (1:1000, 2531, Cell Signaling, USA), lactate dehydrogenase A (LDHa, 1:1000, WL03271, Wanleibio, China), nuclear factor kappa-B p65 (NF-κB p65, 1:1000, WL01273b, Wanleibio, China), p-NF-κB p65Ser536 (1:10 000, WL02169, Wanleibio, China), glycerol-3-phosphate acyltransferase 4 (GPAT4, 1:1000, bs-15587R, Bioss, China), Bax (1:10 000, 50,599–2-Ig, Proteintech, China), Caspase-3/p17/p19 (1:1000, 19,677–1-AP, Proteintech, China), Bcl-2 (1:4000, 26,593–1-AP, Proteintech, China), and β-actin (1:5000, 20,536–1-AP, Proteintech, China). Then, the membranes were washed with TBST, clipped according to the prestained protein ladder (BL712A, Biosharp, China), and incubated with horseradish peroxidase-conjugated goat anti-rabbit IgG (H + L) (1:10 000, SA00001-2, Proteintech, China). An enhanced chemiluminescence kit (P0018AS, Beyotime, China) was used to visualize the immunoblots. Immunoreactive bands were analysed using ImageJ software [34]. β-actin was used as the loading control. ## Apoptosis assay PC-12 cells were exposed to the CM of BV2 cells with different phenotypes for 24 h. The cells were then washed with PBS, centrifuged, and resuspended in 200 μL of binding buffer. Then, 5 μL of Annexin V-FITC (BL107A, Biosharp, China) was added, mixed gently and incubated at 25 ℃. After 15 min, 10 μL of propidium iodide staining solution was added to the cells, mixed, covered with aluminium foil and incubated for 10–20 min at 25 ℃ (in the dark). Cells were then observed under a fluorescence microscope (Olympus, Japan) for 1 h. ## Groupings and drug administration in rats Forty Sprague–Dawley rats were randomly divided into four groups, with 10 rats in each group (half male and half female). Rats in the PRV-infected groups were intraperitoneally injected with three different titres of TCID50 PRV solution (2.85 × 102, 2.85 × 103, and 2.85 × 104), and each rat was injected with a volume of 0.1 mL. The control group was intraperitoneally injected with the same volume of DMEM. When rats in the PRV group had been infected for 7 days, the infection dose with a survival rate higher than $60\%$ was selected as the follow-up test dose [5]. Sixty Sprague–Dawley rats were randomly divided into six groups with 10 rats in each group (half male and half female), which included the control, PRV, RES, CUR L, CUR M, and CUR H groups. The RES group was injected with 50 mg/kg bw resveratrol, the CUR L group was injected with 25 mg/kg bw curcumin solution, the CUR M group was injected with 50 mg/kg bw curcumin solution, the CUR H group was injected with 100 mg/kg bw curcumin solution, and the control and PRV groups were intraperitoneally injected with the same amount of $0.5\%$ sodium carboxymethyl cellulose solution once per day for a total of 14 days. On the 8th day, each rat in the PRV, RES, CUR L, CUR M, and CUR H groups was intraperitoneally injected with 0.1 mL of 2.85 × 103 TCID50 PRV solution. ## Open-field test Open-field test experiments were performed on the. A transparent glass box (60 × 60 × 40 cm3) was divided into 36 equal squares with a 0.5 cm-wide medical bandage. Then, the rats were placed in the transparent glass box, and a video imager (Xiaomi, China) was used to record the number of grids that the rats in each group moved horizontally within for 5 min (effective movement was considered when more than three legs were in a square) and the number of times the rats stood (taking the forelimb off the ground represented effective standing). ## Measurement of body temperature, body weight, and the viscera index From 1 to 7 days post-infection (dpi), rat body weight and temperature were measured. The organ index was evaluated by measuring the ratio of the weight of the brain tissue to the body weight of the rat at 7 dpi. ## Preparation of pathological sections At 7 dpi, the cerebral cortices of the rats were collected, sectioned into 5 × 5 × 3 mm sections and processed in tissue cassettes according to a standard protocol. Glass slides were prepared for microscopic examinations using 5-μm sections of formalin-fixed paraffin-embedded (FFPE) tissues with routine haematoxylin and eosin (HE) staining. The sections of each brain region were observed under an inverted microscope (Olympus, Japan). ## Statistical analysis Statistical analyses were performed using SPSS 23.0 statistical software (IBM, Armonk, NY, USA). Quantitative data were examined using one-way analysis of variance (ANOVA) followed by a least significant difference (LSD) post hoc test for multiple comparisons among the groups, and a two-tailed independent t test was used to compare the differences between two groups. Differences were considered statistically significant at $P \leq 0.05.$ ## PRV-infected BV2 cells exhibit increased levels of inflammatory markers After BV2 cells were infected with different titres of PRV for different times, the PRV-induced inflammatory response was assessed by measuring the levels of NO, TNF-α, IL-6, and ROS. We observed that NO, TNF-α, IL-6, and ROS levels were significantly increased in 1.66 × 106 TCID50 PRV-infected BV2 cells at 24 h compared to those in the control group (Figures 1A–E). Furthermore, these levels were significantly higher in 1.66 × 106 TCID50 PRV-infected BV2 cells at 6, 12, and 24 h than in the control group (Figures 1F–J). These results indicate that BV2 cells infected with 1.66 × 106 TCID50 PRV produced inflammatory markers after 24 h of infection. Additionally, to exclude the effect of different cell numbers on the levels of inflammatory factors, the CCK-8 assay was used, and no changes in cell viability were observed in BV2 cells infected with 1.66 × 106 TCID50 PRV after 24 h (Additional file 1).Figure 1Pseudorabies virus (PRV)-infected BV2 cells show increased production of inflammatory markers. BV2 cells were infected with 1.66 × 106 TCID50, 1.66 × 105 TCID50, 1.66 × 104 TCID50, and 1.66 × 103 TCID50 PRV for 24 h. The levels of NO (A) in the supernatant were detected using the Griess method, the levels of TNF-α (B) and IL-6 (C) were detected using ELISA, and the levels of intracellular ROS (D, E) were detected by adding a fluorescent probe DCFH-DA. Furthermore, the levels of NO (F), TNF-α (G), IL-6 (H), and ROS (I, J) were measured at 6, 12, and 24 h after infection with 1.66 × 106 TCID50 PRV. Green fluorescence indicates the presence of ROS in BV2 cells; scale bar = 200 μm. All experiments were performed in parallel. The results are presented as the mean ± standard deviation (SD) of four biological replicates ($$n = 4$$). Statistical significance was determined using one-way analysis of variance (ANOVA) followed by a least significant difference (LSD) post hoc test for multiple comparisons among the groups. * $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. ## CUR suppresses and promotes the release of proinflammatory and anti-inflammatory cytokines, respectively, in PRV-infected BV2 cells To determine whether CUR affected the viability of BV2 cells, the CCK-8 assay was performed after the cells were treated with 0–40 μM CUR for 24 h. The results revealed that CUR concentrations less than 20 μM did not induce any detectable cytotoxicity (Figure 2A). To ensure that CUR had the greatest effect, 20 μM CUR was used in the subsequent experiments. Figure 2Curcumin (CUR) suppresses proinflammatory cytokine release and promotes anti-inflammatory cytokine release in PRV-infected BV2 cells. A BV2 cells were treated with different concentrations of CUR for 24 h, and cell viability was determined by the CCK-8 assay. BV2 cells were infected with/without 1.66 × 106 TCID50 PRV for 24 h, followed by dimethyl sulfoxide (DMSO) or 20 μM CUR treatment for 6, 12, and 24 h. The levels of NO (B) in the supernatant were detected using the Griess method; the levels of TNF-α (C) and IL-6 (D), IL-4 (E), and IL-10 (F) were detected using ELISA; and the levels of intracellular ROS (G, H) were detected by adding the fluorescent probe DCFH-DA. Green fluorescence indicates the presence of ROS in BV2 cells; scale bar = 200 μm. All experiments were performed in parallel. The results are presented as the mean ± SD of four biological replicates ($$n = 4$$). Statistical significance was determined using one-way ANOVA followed by an LSD post hoc test for multiple comparisons among the groups. $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. Then, the optimal time point for the beneficial effects of CUR was determined. PRV treatment significantly increased NO, TNF-α, and IL-6 levels, whereas incubation with 20 μM CUR for 24 h significantly inhibited the release of NO (Figure 2B), TNF-α (Figure 2C), and IL-6 (Figure 2D) and significantly increased IL-4 (Figure 2E) and IL-10 (Figure 2F) levels. The addition of dimethyl sulfoxide (DMSO) and CUR alone within the dose range used in this study had no effect on the release of cytokines by BV2 cells compared to those in the control group. Therefore, control experiments with DMSO and CUR alone were not performed further. Subsequently, we measured the levels of ROS in BV2 cells and found that 20 μM CUR for 24 h inhibited the PRV-induced increase in ROS (Figures 2G, H). The transformation of microglia from a ramified morphology to an amoeboid shape is associated with inflammation and neurotoxicity [35]. In this study, PRV-infected BV2 cells had an amoeboid shape with an enlarged cell body and had lost their extended processes. Pretreatment with CUR ameliorated the PRV-induced morphological changes in BV2 cells. Control and CUR only BV2 cells also showed a ramified morphology (Additional file 2). These results suggest that 20 μM CUR for 24 h inhibited PRV-induced proinflammatory cytokine production in BV2 cells. ## CUR promotes the polarization of PRV-infected BV2 cells from the M1 phenotype to the M2 phenotype and reverses PRV-induced mitochondrial dysfunction The expression levels of CD$\frac{16}{32}$, CD40, ARG-1, and CD206 were measured using flow cytometry to determine the effect of CUR on phenotypic switching in BV2 cells. The expression levels of M1 phenotypic markers such as CD$\frac{16}{32}$ and double-positivity for CD$\frac{16}{32}$ and CD40 were markedly upregulated by PRV infection compared with those in the control group (Figure 3A). Furthermore, CUR treatment greatly decreased the expression levels of CD$\frac{16}{32}$ and CD40 and double-positivity for CD$\frac{16}{32}$ and CD40 and markedly increased the expression of ARG-1 and CD206, which are markers of the M2 phenotype, and double-positivity for ARG-1 and CD206 in BV2 cells (Figure 3B). These results indicate that CUR induced the transformation of BV2 cells from the M1 to M2 phenotype. Figure 3CUR promotes polarization and reverses PRV-induced mitochondrial dysfunction in PRV-infected BV2 cells. BV2 cells were infected with/without PRV for 24 h and then treated with/without 20 μM CUR for 24 h. A The expression of the M1 phenotype surface markers CD40 and CD$\frac{16}{32}$ and B the M2 phenotype surface markers CD206 and ARG-1 in BV2 cells was determined using flow cytometry ($$n = 3$$). C Mitochondrial structures in BV2 cells were observed using transmission electron microscopy ($$n = 3$$); scale bar = 500 nm. D Mitochondrial membrane potential (MMP) in BV2 cells. E MMP in BV2 cells was observed using fluorescence microscopy ($$n = 4$$); scale bar = 200 μm. All experiments were performed in parallel. The results are presented as the mean ± SD. Statistical significance was determined using one-way ANOVA followed by an LSD post hoc test for multiple comparisons among the groups. $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. An intact mitochondrial structure is a prerequisite for proper mitochondrial function. Mitochondrial dysfunction prevents the repolarization of inflammatory macrophages [36]. Therefore, to investigate whether CUR contributes to the restoration of mitochondrial function and improves the reprogramming of inflammatory macrophages to anti-inflammatory cells, TEM was used to observe mitochondrial structures. As shown in Figure 3C, most BV2 cells in the control group had clear and complete mitochondria and several long and uniform cristae that were neatly arranged and densely packed in the mitochondrial membrane. In contrast, the mitochondria of BV2 cells in the PRV-infected group were swollen and vacuolated, and cristae in the mitochondrial membrane were dissolved and broken. However, in the PRV + CUR treatment group, some mitochondria in BV2 cells displayed a clear and complete shape, and the dissolution and fragmentation of mitochondrial membrane cristae were weakened. MMP is an important indicator of mitochondrial function. The JC-1 assay kit showed that the Δψm of the PRV group was significantly reduced compared with that of the control group. However, in the PRV + CUR group, the decrease in the Δψm was significantly reversed (Figures 3D, E). These results suggest that CUR reverses PRV-induced mitochondrial dysfunction. ## The secretions of CUR-treated BV2 cells protect neurons from PRV-induced apoptosis Activated microglia release neurotoxic agents, which correlate with the onset and progression of neurological diseases [37]. CUR inhibited the release of proinflammatory cytokines by PRV-infected BV2 cells. Therefore, we further determined whether the CM of CUR-treated cells could protect against PRV-induced neuronal toxicity in BV2 cells using CCK-8, LDH, and MDA assays. PC-12 cells were incubated with CM from BV2 cells for 24 h. Cell viability in the CM_Control and CM_CUR groups was similar to that in the control group. However, cell viability in the CM_PRV group was significantly decreased compared to that in the control group (Figures 4A–C). Notably, the viability of PC-12 cells in the CM_PRV + CUR group was significantly elevated compared to that in the CM_PRV group. Likewise, PC-12 cells in the control, CM_CUR, and CM_Control groups had fine morphology and smooth cell edges, while PC-12 cells exhibited axonal rupture and cell death in the CM_PRV group under a light microscope. However, axonal rupture in PC-12 cells was attenuated in the CM_PRV + CUR group (Additional file 3A). Annexin V-FITC/PI double staining showed that the numbers of apoptotic and dead PC-12 cells in the CM_Control and CM_CUR groups were similar to those in the control group. The CM-PRV group showed a significant increase in both of these cell types compared to the CM_Control and control groups. Furthermore, apoptosis and cell death in the CM_PRV + CUR group were lower than those in the CM_PRV group (Additional file 3B). Activated caspase-3 and Bax are key mediators of neurotoxin-induced neuronal apoptosis, whereas Bcl-2 is an antiapoptotic protein. Figure 4CUR ameliorated neuronal apoptosis in PRV-infected BV2 cells. PC-12 cells containing neuronal properties were treated with the supernatants of different phenotypes of BV2 cells for 24 h. A The viability of PC-12 cells ($$n = 4$$). B LDH activity in PC-12 cells ($$n = 3$$). C MDA levels in PC-12 cells ($$n = 3$$). D Representative Western blot analysis of apoptotic/antiapoptotic proteins; β-actin was used as the loading control ($$n = 3$$). E Relative protein levels of cleaved caspase-3. F Relative protein levels of Bax. G Relative protein levels of Bcl-2. All experiments were performed in parallel. The results are presented as the mean ± SD. Statistical significance was determined using one-way ANOVA followed by an LSD post hoc test for multiple comparisons among the groups. $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. Then, we investigated whether the CM of CUR-treated cells could protect neurons from apoptosis induced by the secretions of PRV-infected BV2 cells by inhibiting caspase-3 activation and measured cleaved caspase-3 expression and activity using Western blotting. Consistent with the results obtained from the Annexin V-FITC/PI double staining assay, exposure of PC-12 cells to the CM of PRV-infected BV2 cells resulted in a substantial increase in cleaved caspase-3 and Bax protein levels and a substantial decrease in Bcl-2 protein levels. However, treatment with CM from PRV-infected BV2 cells treated with CUR significantly attenuated cleaved caspase-3 and Bax protein production and significantly increased Bcl-2 protein levels in PC-12 cells (Figures 4D–G). These results suggest that CUR attenuates neuronal apoptosis, which may be partially dependent on the regulation of microglial polarization and a reduction in inflammatory responses. ## Distinct mRNA signatures of BV2 cells with different phenotypes were identified using RNA-Seq analysis Based on the abovementioned robust regulatory effects of CUR on cytokines in PRV-infected BV2 cells, we used RNA-seq to profile the transcriptomes of BV2 cells with different phenotypes (accession number: GSE201985). To obtain an unsupervised overview of the whole dataset, principal component analysis (PCA) was applied, and biological replicates were found to cluster together according to the PCA score plot (Additional file 4). This* general pattern confirmed the reproducibility of the manipulations and the robustness of the data acquisition. Venn diagrams were used to show the numbers of DEGs in the PRV and control groups and in the PRV + CUR and PRV groups (Figure 5A). The analysis revealed 306 DEGs ($P \leq 0.05$, and fold change ≥ 1) in the PRV group compared with the control group, of which 242 genes were upregulated and 64 genes were downregulated (Figure 5B). Notably, a comparison of the PRV + CUR group with the PRV group identified 5,073 DEGs, of which 2661 genes were upregulated, and 2412 genes were downregulated (Figure 5C).Figure 5Distinct mRNA signatures were identified during phenotypic transformation according to RNA-Seq analysis. BV2 cells were infected with/without PRV for 24 h and then treated with/without 20 μM CUR for 24 h for RNA sequencing and analysis. Differentially expressed genes (DEGs) were detected by RNA-Seq analysis using DESeq2 ($$n = 3$$, $P \leq 0.05$, and fold change > 1). A Venn diagram showing DEGs. B DEG volcano plot of the PRV group vs. the control group. C DEG volcano plot of the PRV + CUR group vs. the PRV group. D GO enrichment analysis was performed on the DEGs in the PRV group vs. control group. E GO enrichment analysis was performed on the DEGs in the PRV + CUR group vs. PRV group. F KEGG enrichment analysis was performed on the DEGs in the PRV group vs. control group. G KEGG enrichment analysis was performed on the DEGs in the PRV + CUR group vs. PRV group. D KEGG enrichment analysis of DEGs. H) qRT‒PCR verification of differentially expressed genes identified in RNA-seq ($$n = 3$$). All experiments were performed in parallel. The results are presented as the mean ± SD. Statistical significance was determined using one-way ANOVA followed by an LSD post hoc test for multiple comparisons among the groups. * $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. We performed GO and KEGG enrichment analyses on 306 DEGs in the PRV group relative to the control group and 5073 DEGs in the PRV + CUR group relative to the PRV group. As shown in Figures 5D and E, the results of GO analysis showed that both PRV vs. Control DEGs and PRV + CUR vs. PRV were significantly enriched in skeletal muscle-related biological processes, which may be related to the morphological plasticity of BV2 cells. According to the KEGG pathway enrichment analysis (Figure 5F, G), both PRV vs. Control and PRV + CUR vs. PRV DEGs were significantly enriched in metabolic pathways, oxidative phosphorylation, Alzheimer’s disease, and glycolysis/gluconeogenesis, suggesting that the phenotypic transformation of BV2 cells may be related to energy metabolism. To verify the reliability of the results, we performed qPCR analysis of some differentially expressed genes from the RNA-seq analysis (Pfkl, Ldha, Hk1, Mcat, and H2dmb1) that are glycolysis, FAS and phenotype-related genes. Our findings suggest that glycolysis and FAS in energy metabolism drive phenotypic shifts in PRV-infected BV2 cells (Figure 5H). ## CUR modulates phenotype-related cytokines in PRV-infected BV2 cells via AMPK-mediated energy metabolism-related pathways During energy stress, activated AMPK directly phosphorylates key factors involved in multiple pathways to restore energy balance [38]. Accumulating evidence suggests that anti-inflammatory drugs can activate AMPK to alter energy metabolism in macrophages and exert anti-inflammatory effects [39, 40]. Therefore, AMPK phosphorylation was first analysed to determine whether the AMPK pathway was involved in the anti-inflammatory effect of CUR on PRV-infected BV2 cells. PRV infection significantly decreased the levels of phosphorylated AMPK. Conversely, CUR treatment abrogated the reduction in phosphorylated AMPKT172 levels (Figures 6A and B). To further assess whether AMPK regulates energy metabolism in PRV-infected microglia treated with CUR, BV2 cells were pretreated with/without the AMPK inhibitor Compound C or AMPK siRNA before the addition of CUR. As shown in Figure 6C, Compound C (≤ 2 μM) had no effect on the viability of BV2 cells. To ensure maximum suppression, 2 μM Compound C was used for further analyses. Fluorescence microscopy showed that the transfection efficiency of siRNA was good using (Additional file 5A), and the band with the best inhibitory effect was screened using Western blot analysis (Additional files 5B, C). Western blot analysis showed that CUR treatment abrogated the decrease in phosphorylated AMPK protein levels caused by PRV infection and inhibited the increase in LDHa (key enzyme in the glycolysis process) and Gpat4 (key enzyme in triacylglycerol synthesis) protein levels induced by PRV infection. Notably, Compound C or siRNA pretreatment reversed the effect of CUR on PRV-infected BV2 cells (Figure 6D–G). CUR enhanced the reductions in the OCR and ATP caused by PRV infection and attenuated the increase in the ECAR. Likewise, pretreatment with Compound C or siRNA reversed these effects of CUR (Figures 6H–J). Taken together, these results suggest that CUR regulates the activities of enzymes related to the energy metabolism pathways by activating AMPK, thereby increasing energy production and reducing the energy-consuming responses of PRV-infected BV2 cells. Figure 6CUR regulates energy metabolism through AMPK-dependent pathways in PRV-infected BV2 cells. A BV2 cells were infected with/without PRV for 24 h and then treated with/without 20 μM CUR for 24 h. Western blot analysis of p-AMPKThr172 protein levels (normalized to the t-AMPK protein). β-Actin was used as the loading control ($$n = 3$$). B Relative protein levels of p-AMPKThr172. C BV2 cells were treated with different concentrations of Compound C for 24 h, and cell viability was determined by the CCK-8 assay to determine the toxicity range ($$n = 4$$). D BV2 cells were infected with/without PRV or were treated with PRV and an AMPK inhibitor (Compound C or siRNA) alone or in combination for 24 h, followed by treatment with/without 20 μM CUR for 24 h. Western blot analysis of p-AMPK (normalized to the t-AMPK protein), t-AMPK, Gpat4, and LDHa levels (Gpat4 and LDHa values were normalized to the β-actin protein); β-actin was used as the loading control ($$n = 3$$). E Relative p-AMPKThr172 protein levels. F Relative LDHa protein levels. G Relative Gpat4 protein levels. H The extracellular acidification rate of BV2 cells after the different treatments (values normalized to the control) ($$n = 4$$). I The OCR of BV2 cells after the different treatments (values normalized to the control) ($$n = 4$$). J BV2 cell ATP levels in the different treatment groups ($$n = 4$$). K The M1 phenotype-related inflammatory factor TNF-α in BV2 cells was detected using ELISA ($$n = 4$$). L The levels of the M1 phenotype-related inflammatory factor IL-6 in BV2 cells ($$n = 4$$). M The levels of the M2 phenotype-related anti-inflammatory factor IL-4 in BV2 cells ($$n = 4$$). N The levels of the M2 phenotype-related anti-inflammatory factor IL-10 in BV2 cells ($$n = 4$$). All experiments were performed in parallel. The results are presented as the mean ± SD. Statistical significance was determined using one-way ANOVA followed by an LSD post hoc test for multiple comparisons among the groups. $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. To investigate the role of AMPK-related pathways in CUR-treated PRV-infected BV2 cells, the levels of microglial phenotype-related cytokines were measured using ELISA. As predicted, CUR inhibited the increases in TNF-α and IL-6 levels in the PRV-infected BV2 cells and promoted the production of IL-4 and IL-10. After pretreatment with Compound C or AMPK-α siRNA, the anti-inflammatory effects of CUR disappeared (Figures 6K–N). To further verify the role of CUR in transforming the phenotype of PRV-infected microglia by activating AMPK-energy metabolism-related pathways, we investigated the effect of Compound C and siRNA on CUR-treated primary cultured microglia. The purity of the microglia was identified by flow cytometry as $95.2\%$ (Additional file 6A). Similar to previous findings, the effect of CUR disappeared after pretreatment with Compound C or siRNA (Additional file 6B). These results show that CUR-treated PRV-infected microglia changed from the M1 phenotype to the M2 phenotype through activation of the AMPK-energy metabolism-related pathway. ## AMPK mediates the regulatory effect of CUR on NF-κB p65 The proinflammatory NF-κB signalling pathway is a key regulator of immune processes, which affects changes in energy metabolism, such as OXPHOS, glycolysis, triglyceride levels, and lipogenesis [41]. Western blot analysis revealed that CUR treatment inhibited the increase in phosphorylated p65 levels in the cytoplasm in PRV-infected BV2 cells. However, pretreatment with Compound C or siRNA suppressed the phosphorylation of p65 (Figures 7A, B), indicating that CUR inhibits the NF-κB signalling pathway through AMPK.Figure 7NF-κB p65 may mediate the effects of CUR on the AMPK-energy response pathway. BV2 cells were infected with/without PRV or treated with PRV and AMPK inhibitors (Compound C or siRNA) alone or in combination for 24 h, followed by treatment with/without 20 μM CUR for 24 h. A Western blot analysis of NF-κB p65 and NF-κB p-p65Ser536 levels; β-actin was used as the loading control ($$n = 3$$). B Relative p-p65Ser536 levels. All experiments were performed in parallel. The results are presented as the mean ± SD. Statistical significance was determined using one-way ANOVA followed by an LSD post hoc test for multiple comparisons among the groups. $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. ## CUR ameliorates poor survival, CNS excitation, hyperthermia, and slow growth in PRV-infected rats To determine the optimal titre of PRV infection, we injected rats with different titres of PRV and recorded the daily mortality. Rats infected with 2.85 × 102 TCID50 PRV died at 6 dpi and had a survival rate of $90\%$. No deaths were observed after 7 dpi. Th Rats infected with 2.85 × 103 TCID50 PRV began to die at 5 dpi, had a survival rate of $90\%$, and the survival rate was reduced to $70\%$ at 6 dpi with no deaths recorded after 7 dpi. Rats infected with 2.85 × 104 TCID50 PRV began to die at 4 dpi and had a survival rate of $90\%$, which decreased to $60\%$ at 5 dpi. Mortality continued at 6 dpi with a reduced survival rate of $40\%$, and there were no further deaths at 7 dpi (Figure 8A). As the rats in the 2.85 × 103 TCID50 PRV-infected group had a survival rate higher than $60\%$ at 7 dpi and showed clinical symptoms, such as hyperactivity, increased body temperature, and weight loss, this titre was used as the infection dose for follow-up experiments. Figure 8CUR ameliorates central nervous system excitation, hyperthermia, slow growth, and poor survival in PRV-infected rats. A After the rats were intraperitoneally injected with 2.85 × 102 TCID50, 2.85 × 103 TCID50 and 2.85 × 104 TCID50 PRV or DMEM solution, the number of deaths in each group was recorded every day ($$n = 10$$). B The rats were intraperitoneally injected with low, medium, and high concentrations of CUR (25, 50, and 100 mg/kg BW) and resveratrol (RES 50 mg/kg BW) once per day for 14 days. On Day 8 (1 dpi), in addition to the control group, 0.1 mL of 2.85 × 103 TCID50 PRV was injected, and the mortality was recorded daily. C The horizontal and vertical motor abilities of the rats were observed on the 8th day (1 dpi). D The horizontal and vertical motor abilities of the rats were observed on Day 11 (4 dpi) ($$n = 3$$). All experiments were performed in parallel. The results are presented as the mean ± SD. Statistical significance was determined using one-way ANOVA followed by an LSD post hoc test for multiple comparisons among the groups. $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. To investigate the protective effect of CUR on PRV-infected rats, we intraperitoneally injected the rats with low, medium, and high concentrations of CUR. Since resveratrol has anti-inflammatory and anti-PRV properties [42], it was selected as a positive control in this study. The results indicated that there was no mortality in any group at 1 and 2 dpi. At 3 dpi, the survival rate in the CUR L group was $80\%$, while that in the other groups was $100\%$. At 4 dpi, the survival rates in the PRV and CUR L groups were $80\%$, that in the CUR H group was $90\%$, and those in the CUR M, RES, and control groups were $100\%$. At 5-7 dpi, the survival rates in the control, PRV, RES, CUR L, CUR M, and CUR H groups were $100\%$, $70\%$, $90\%$, $70\%$, $100\%$, and $90\%$, respectively (Figure 8B). An open-field test was used to explore the effect of CUR on PRV-induced central nervous system excitation in rats. At 1 dpi, there was no significant difference in the line crossing and rearing abilities of rats in each group. At 4 dpi, significant differences were observed in the line crossing and rearing abilities of the rats in the PRV-infected group and the low-dose CUR group (CUR L) compared with those in the control group; the line crossing and rearing abilities of the rats in the RES group, CUR medium-dose group (CUR M), and CUR high-dose group (CUR H) were significantly reduced compared to those in the PRV group (Figure 8C, D). Then, we examined the effect of CUR treatment on body temperature and weight in PRV-infected rats. A prolonged body temperature increase occurred 3–5 days after PRV infection in rats, and low doses of CUR had no significant inhibitory effect on the PRV-induced increase in body temperature. However, treatment with resveratrol, medium-dose CUR, and high-dose CUR significantly inhibited the PRV-induced increase in body temperature (Table 1). PRV-infected rats displayed slow growth, and low-dose CUR did not significantly improve the growth inhibition caused by PRV, while treatment with resveratrol, medium-dose CUR, and high-dose CUR significantly improved PRV-induced slow growth (Table 2).Table 1Daily variations in body temperatureDayGroupControl (℃)PRV (℃)RES (℃)CUR L (℃)CUR M (℃)CUR H (℃)Day 037.92 ± 0.0637.88 ± 0.0738.02 ± 0.0637.98 ± 0.0637.92 ± 0.0737.98 ± 0.04Day 138.00 ± 0.0538.78 ± 0.22#37.97 ± 0.0537.97 ± 0.1138.10 ± 0.1337.97 ± 0.09Day 238.07 ± 0.0638.22 ± 0.1338.00 ± 0.1038.28 ± 0.1837.90 ± 0.1037.85 ± 0.10Day 337.87 ± 0.0638.60 ± 0.09#38.08 ± 0.1238.40 ± 0.10#37.92 ± 0.1437.93 ± 0.09Day 438.05 ± 0.0838.72 ± 0.14#38.55 ± 0.11#38.50 ± 0.06#38.57 ± 0.11#37.90 ± 0.08Day 537.85 ± 0.0638.47 ± 0.08#38.43 ± 0.06#38.52 ± 0.11#38.50 ± 0.06#38.02 ± 0.15Day 638.00 ± 0.0737.70 ± 0.1438.57 ± 0.06#38.08 ± 0.1738.57 ± 0.08#38.62 ± 0.14#Day 737.85 ± 0.1137.13 ± 0.13#37.98 ± 0.0836.67 ± 0.4738.03 ± 0.1538.00 ± 0.15Values represent the mean ± SD of 6 animals per group. #: the difference was significant compared with the control ($p \leq 0.05$), : the difference was significant compared with the PRV group ($p \leq 0.05$).Table 2The change in body weight gain in each groupTimeGroupControl (g)PRV(g)RES(g)CUR L(g)CUR M(g)CUR H(g)Day 13.50 ± 0.553.12 ± 0.653.13 ± 0.253.67 ± 0.533.63 ± 0.703.50 ± 0.79Day 27.60 ± 0.606.75 ± 0.637.30 ± 0.605.70 ± 0.45#8.43 ± 0.738.05 ± 0.62Day 312.83 ± 0.876.08 ± 0.32#9.75 ± 0.85#5.43 ± 0.34#11.50 ± 0.7011.48 ± 0.72Day 417.22 ± 1.053.27 ± 1.03#12.13 ± 1.33#7.81 ± 0.70#14.87 ± 1.2414.12 ± 0.98Day 521.32 ± 1.173.55 ± 0.90#14.65 ± 0.90#8.31 ± 0.95#16.83 ± 1.07#15.00 ± 1.34#Day 624.76 ± 1.033.88 ± 1.02#16.3 ± 1.11#7.57 ± 1.08#19.13 ± 0.88#17.20 ± 1.15#Day 726.67 ± 1.124.77 ± 0.69#17.95 ± 0.77#6.77 ± 0.82#21.72 ± 0.67#18.50 ± 1.58#Values represent the mean ± SD of 6 animals per group. #: the difference was significant compared with the control ($p \leq 0.05$), : the difference was significant compared with the PRV group ($p \leq 0.05$). ## CUR improves brain congestion, organ index changes, and vascular cuffing in PRV-infected rats Rat brain tissues were used to observe the pathological changes caused by PRV infection. At 7 dpi, the brain tissues in the PRV and CUR L groups showed obvious swelling and congestion compared to those in the control group. However, brain tissue swelling and congestion in the CUR M and CUR H groups were obviously improved compared with those in the PRV group (Figure 9A). The CUR M and CUR H groups showed an increase in the brain tissue organ index compared to the PRV group (Figure 9B). Acute encephalitis is usually caused by pseudorabies virus [4]. Therefore, the vascular cuff phenomenon in PRV-infected rats was examined using HE staining. As expected, the PRV infection-induced vascular cuffing was ameliorated in the CUR L, CUR M, and CUR H groups (Figure 9C), indicating that CUR treatment ameliorated the cortical pathological changes and microglial activation induced by PRV infection. Figure 9CUR improves brain congestion, changes in organ indices, and vascular cuffing in PRV-infected rats. Rats were intraperitoneally injected with low, medium, and high doses of CUR or $0.5\%$ sodium carboxymethylcellulose solution (the solvent of CUR) for 7 consecutive days, and on the 8th day, the rats were intraperitoneally injected with/without PRV. On the 14th day (7 dpi), the rats in each group were sacrificed by decapitation, and the brain tissue was harvested for imaging to evaluate the organ index and prepare pathological sections. A The degree of blood congestion in the brain tissue in each group ($$n = 3$$). B Changes in the viscera index in rat brain tissue ($$n = 3$$). C Pathological sections of the cortex in each group, a: Control group, b: PRV group, c: CUR L group, d: CUR M group, e: CUR H group ($$n = 3$$), scale bar = 20 μm. All experiments were performed in parallel. The results are presented as the mean ± SD. Statistical significance was determined using one-way ANOVA followed by an LSD post hoc test for multiple comparisons among the groups. * $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. ## Discussion PRV infection can cause viral encephalitis in animals and humans [4, 5], which is a devastating disease, and survivors often experience severe neurological complications. During viral encephalitis, dysregulated microglia produce cytokines that cause inflammatory damage to neurons, leading to neurocognitive impairment [43]. Therefore, the timely control of microglial activation may be an effective treatment strategy for viral encephalitis. First, we constructed an inflammatory model of PRV infection. Among the many proinflammatory cytokines, TNF-α, IL-6, and NO play major roles in the inflammatory response [44, 45]. Our study showed that PRV infection at 1.66 × 106 TCID50 led to increased levels of TNF-α, IL-6, and NO in BV2 cells, indicating that an inflammatory response was induced. It has been reported that mitochondrial ROS are involved in metabolic changes associated with macrophage activation during inflammation, and M1 microglia generally show higher intracellular ROS levels than M0 and M2 cells [46, 47]. We found that PRV infection increased intracellular ROS levels in BV2 cells, suggesting that BV2 cells may switch from a resting M0 phenotype to an M1 phenotype. Correspondingly, we found that rat body temperature increased and body growth was reduced after PRV infection and microglial activation in vivo, which may be similar to the lipopolysaccharide-mediated induction of microglial inflammation [48]. Next, we investigated the optimal anti-inflammatory time point for CUR and whether it could transform the PRV-infected BV2 cell phenotype. We found that CUR treatment for 24 h reduced the secretion of the inflammatory factors TNF-α, IL-6, and NO, which was consistent with previous reports [20]. Eradicating the overproduction of ROS can mitigate proinflammatory M1 polarization and advance anti-inflammatory M2 polarization [47]. IL-4 and IL-10 can inhibit the production of proinflammatory cytokines, such as IL-8, IL-6, and TNF-α, and reduce the release of NO, thereby preventing neuronal damage induced by pathogenic microorganisms in vitro and in vivo [49]. In our study, CUR treatment attenuated the PRV-induced increase in intracellular ROS in BV2 cells and enhanced the secretion of IL-4 and IL-10, suggesting that CUR could convert PRV-infected BV2 cells from the M1 phenotype to the M2 phenotype. During viral encephalitis, the major group of MHC class II proteins, as well as the costimulatory molecules CD40, CD16, CD32, and CD86, are expressed on the surface of activated M1 microglia following viral infection [43], while M2 microglia typically express CD206 and ARG-1 [13]. Therefore, we hypothesized that CUR could convert the phenotype of PRV-infected BV2 cells, and our findings confirmed this hypothesis. Correspondingly, we found that CUR treatment significantly reversed the PRV infection-induced increase in body temperature, slow growth, and microglial activation in rats in vivo. The proinflammatory molecules TNF-α, IL-1β, CCL2, CCL5, and IL-6 have been reported to induce neuronal death by causing direct and indirect neurotoxicity [43, 50]. Therefore, to confirm that the changes in cytokines associated with the transformation of the microglial phenotype could alter the state of neuronal inflammatory damage, we first examined the effect of microglial supernatant on PC-12 cell activity. As a metabolite of lipid peroxidation, MDA often reflects the severity of ROS-induced cell damage. LDH is involved in glycolysis and is automatically released from the cell after cell damage. The degree of cell damage was assessed by measuring the amount of LDH released. Treatment of PC-12 cells with the supernatants of BV2 cells with different phenotypes for 24 h showed that the supernatant of M2 BV2 cell cells reversed the decrease in cell viability observed in response to M1 BV2 cell supernatant. We hypothesized that the decrease in cell viability may be caused by changes in the levels of apoptotic proteins. As expected, the supernatant of M2 BV2 cell cells reduced the number of apoptotic PC-12 cells, reduced the levels of the apoptotic proteins Bax and cleaved caspase-3, and increased the levels of Bcl-2 induced by the supernatant of M1 BV2 cells. Correspondingly, CUR treatment attenuated the PRV-induced excitation of the CNS, increased the tissue organ index and brain congestion, and improved the survival rates of the PRV-infected rats. The effects of CUR treatment on PRV-induced neuroinflammatory responses further validate the immunomodulatory role of CUR in the brain and provide new insights into the heterogeneous phenotype-specific microglial responses to this active small molecule. Mitochondria are the main sites of energy metabolism and help cells carry out normal energy metabolism. Mitochondrial dysfunction prevents the repolarization of inflammatory macrophages [36, 51]. Therefore, the restoration of mitochondrial function can improve the reprogramming of inflammatory macrophages to anti-inflammatory cells. Our study showed that CUR treatment reversed PRV-induced mitochondrial swelling and vacuole formation, the dissolution of cristae in the mitochondrial membrane, and the increase in the number of fragmented mitochondria, indicating its effect on alleviating mitochondrial damage after infection. MMP is a key indicator of mitochondrial function. Lipopolysaccharide reduces MMP and mitochondrial damage in BV2 cells [51, 52]. Similarly, the present study showed that CUR treatment effectively ameliorated the PRV infection-induced reduction in MMP in BV2 cells. Thus, CUR ameliorated mitochondrial dysfunction in PRV-infected BV2 cells, which is a prerequisite for reprogramming inflammatory microglia into the anti-inflammatory M2 phenotype. To further explore the mechanism by which CUR transforms PRV-infected BV2 cells at the transcriptomic level, we performed RNA-seq analysis of BV2 cells with M0, M1, and M2 phenotypes. Substantial* gene expression changes occurred after PRV infection and CUR treatment. Compared with those in BV2 cells in the control group (M0 phenotype), glycolysis and proinflammatory-related genes (e.g., Pfkl, Ldha, and H2dmb1) were significantly upregulated in PRV-infected BV2 cells (M1 phenotype). However, CUR treatment significantly reduced the elevated levels of glycolysis and FAS-related genes (e.g., Ldha, Hk-1, Gpat4, and Mcat) induced by PRV infection (M2 phenotype). Furthermore, KEGG enrichment analysis of the gene subsets provided insight into the possible signalling pathways involved in the molecular mechanisms of phenotypic transformation. We found that the differentially expressed genes were mainly enriched in pathways related to energy metabolism (e.g., glycolysis, OXPHOS, and FAS). Therefore, we hypothesized that the change in the energy metabolism pathway in BV2 cells was closely related to phenotypic transitions. RT‒qPCR further verified the expression levels of genes associated with the AMPK pathway. AMPK is a highly conserved sensor of cellular energy status that can be activated by phosphorylation of its subunit at Thr172, which in turn affects energy metabolism [53] by regulating the expression of glycolysis pathway-related genes such as Ldha, Hk1, and Pfkl [54, 55] and fatty acid synthesis pathway-related genes such as Gpat4 and Mcat [56–58]. It has been reported that p-AMPK levels are reduced during microglial inflammation [17]. Similarly, we found that the levels of p-AMPK were decreased after PRV infection, and CUR treatment reversed this decrease. When immune cells are activated, there is a shift in metabolism from OXPHOS to aerobic glycolysis, which is known as the Warburg effect [59]. The metabolism of M1 microglia mainly depends on aerobic glycolysis and the FAS pathway. Among these, two disruptions in the tricarboxylic acid (TCA) cycle lead to the accumulation of itaconic acid and succinic acid, which activate the transcription of glycolytic genes, thereby maintaining glycolytic metabolism in M1 microglia [12, 26]. Although ATP production efficiency is relatively low, ATP provides metabolic intermediates for FAS [26]. M2 microglia are more dependent on OXPHOS, and the TCA cycle is intact and provides substrates for complexes in the electron transport chain. This metabolic ATP production efficiency is relatively high, providing energy for the tissue repair and remodelling functions of M2 microglia [29]. LDH is a key enzyme in glycolysis, and its activity indirectly reflects glycolytic output during the metabolic reprogramming of BV2 cells [12]. The OCR reflects cellular OXPHOS levels; GPAT4 is involved in the synthesis of triacylglycerol [60]. Our study showed that CUR treatment reversed the PRV infection-induced increases in the protein levels of LDHa, GPAT4, and ECAR and increased the OCR and ATP levels. However, after AMPK activity was knocked down by an AMPK inhibitor and siRNA, the beneficial effect of CUR was not observed, indicating that the effect of CUR was mediated by AMPK. The NF-κB pathway is a key regulator of immune processes and has long been regarded as a typical proinflammatory signalling pathway. In recent years, this pathway has been reported to affect changes in energy metabolism related to inflammation and immune responses (e.g., OXPHOS and glycolysis) [41]. NF-κB binds to the inhibitor molecule IκB to form a p50-p65-IκB trimer under steady state conditions. In response to inflammatory mediators, the NF-κB subunit p65 is phosphorylated at Ser536 and migrates into the nucleus, upregulating the expression of various inflammation-related genes [61]. We therefore examined whether AMPK regulates energy metabolism through NF-κB p-65. The results showed that CUR reversed the increase in p-p65 levels caused by PRV infection. After AMPK activity was knocked down, the effects of CUR disappeared, suggesting that NF-κB may regulate energy metabolism via AMPK. Finally, we examined inflammatory factors associated with different phenotypes in BV2 cells, validated them in primary microglia, and concluded that CUR was required for phenotypic transition by repairing mitochondrial dysfunction, microglial phenotypic transition was driven by the AMPK energy metabolism pathway, and NF-κB may mediate this process. In conclusion, to our knowledge, this is the first report on CUR-mediated resistance to PRV-induced encephalitis through the modulation of phenotypic transitions mediated by the AMPK/NF-κB-energy metabolism signalling pathway in microglia. Our study provides insight into the beneficial effects of CUR treatment on PRV-induced neuroinflammation. ## Supplementary Information Additional file 1. The effect of 1.66 × 106 TCID50 PRV infection for 24 h on the viability of BV2 cells. The cells were infected with 1.66 × 106 TCID50 PRV for 24 h, and the changes in the survival rates were measured. All experiments were performed in parallel. The results are expressed as the mean ± standard deviation (SD) of four biological replicates ($$n = 4$$). Statistical significance was determined using a two-tailed independent t test to compare the two groups. * $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. Additional file 2. Effect of CUR on the morphology of PRV-infected BV2 cells. ( A) Untreated BV2 cells. ( B) BV2 cells were infected with 1.66 × 106 TCID50 PRV for 24 h, and the cell maintenance medium was replaced. ( C) BV2 cells were infected with 1.66 × 106 TCID50 PRV for 24 h and then treated with 20 μM curcumin (CUR) for 24 h. Morphological changes in BV2 cells were observed under a light microscope (scale bar = 200 μm). All experiments were performed in parallel. Additional file 3. Effects of the supernatants of BV2 cells with different phenotypes on PC-12 cell morphology and apoptosis. The supernatants of BV2 cells with different phenotypes were added to PC-12 cells and incubated for 24 h. (A) Morphological changes in PC-12 cells were observed using a light microscope (scale bar = 200 μm). ( B) Apoptosis in PC-12 cells detected using the Annexin V-FITC/PI kit. Red fluorescence represents late apoptotic and dying cells, while green fluorescence represents early apoptotic cells; scale bar = 200 μm. All experiments were performed in parallel. Additional file 4. Principal component analysis (PCA). ( A) Correlation check of the RNA-seq data using Pearson’s Correlation Coefficient. R2 ≥ 0.8 represents the repeatability of the experiment and the reliability of the evaluation results. Additional file 5. Screening the optimal interference effect of different siRNAs. ( A) The effects of siRNA transfection were observed under a fluorescence microscope. PC-siRNA group, positive control group; NC-FAM-siRNA group, fluorescence-labelled negative control group; FAM-siRNA 1 group, fluorescently labelled small interfering RNA first band group; FAM-siRNA 2 group, fluorescently labelled small interfering RNA second band group; FAM-siRNA 3 group, fluorescently labelled small interfering RNA third band group. ( B) Western blot analysis of t-AMPK protein levels; β-actin was used as a loading control ($$n = 3$$). ( C) Relative protein levels of t-AMPK. All experiments were performed in parallel. The results are presented as the mean ± SD. Statistical significance was determined using one-way analysis of variance (ANOVA) followed by a least significant difference (LSD) post hoc test for multiple comparisons among the groups. $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. Additional file 6. Purity of primary microglia and the effects of curcumin on the secretion of pro- and anti-inflammatory cytokines by PRV-infected microglia. ( A) Microglial purity was identified by flow cytometry; CD11b indicated resting microglia, and MHC Class II indicated activated cells ($$n = 3$$). ( B) M1 phenotype-related inflammatory factors (TNF-α) in primary microglial cells were examined using ELISA ($$n = 4$$). ( C) Levels of M1 phenotype-related inflammatory factors (IL-6) in primary microglial cells ($$n = 4$$). ( D) Levels of M2 phenotype-related anti-inflammatory factors (IL-4) in primary microglial cells ($$n = 4$$). ( E) Levels of M2 phenotype-related anti-inflammatory factors (IL-10) in primary microglial cells ($$n = 4$$). ( $$n = 4$$). All experiments were performed in parallel. The results are presented as the mean ± SD. Statistical significance was determined using one-way ANOVA followed by an LSD post hoc test for multiple comparisons among the groups. $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.001$, and NS, not significant. ## References 1. Pegg CE, Zaichick SV, Bomba-Warczak E, Jovasevic V, Kim D, Kharkwal H, Wilson DW, Walsh D, Sollars PJ, Pickard GE, Savas JN, Smith GA. Herpesviruses assimilate kinesin to produce motorized viral particles. Nature* (2021) 599 662-666. DOI: 10.1038/s41586-021-04106-w 2. Ai JW, Weng SS, Cheng Q, Cui P, Li YJ, Wu HL, Zhu YM, Xu B, Zhang WH. Human endophthalmitis caused by pseudorabies virus infection, China, 2017. Emerg Infect Dis (2018) 24 1087-1090. DOI: 10.3201/eid2406.171612 3. Zheng L, Liu X, Yuan D, Li R, Lu J, Li X, Tian K, Dai E. Dynamic cerebrospinal fluid analyses of severe pseudorabies encephalitis. Transbound Emerg Dis (2019) 66 2562-2565. DOI: 10.1111/tbed.13297 4. Liu Q, Wang X, Xie C, Ding S, Yang H, Guo S, Li J, Qin L, Ban F, Wang D, Wang C, Feng L, Ma H, Wu B, Zhang L, Dong C, Xing L, Zhang J, Chen H, Yan R, Wang X, Li W. A novel human acute encephalitis caused by pseudorabies virus variant strain. Clin Infect Dis (2021) 73 e3690-e3700. DOI: 10.1093/cid/ciaa987 5. Hu S, Liu Q, Zang S, Zhang Z, Wang J, Cai X, He X. Microglia are derived from peripheral blood mononuclear cells after pseudorabies infection in mice. Viral Immunol (2018) 31 596-604. DOI: 10.1089/vim.2018.0064 6. Pomeranz LE, Reynolds AE, Hengartner CJ. Molecular biology of pseudorabies virus: impact on neurovirology and veterinary medicine. Microbiol Mol Biol Rev (2005) 69 462-500. DOI: 10.1128/MMBR.69.3.462-500.2005 7. Prinz M, Jung S, Priller J. Microglia biology: one century of evolving concepts. Cell (2019) 179 292-311. DOI: 10.1016/j.cell.2019.08.053 8. Chen Z, Zhong D, Li G. The role of microglia in viral encephalitis: a review. J Neuroinflammation (2019) 16 76. DOI: 10.1186/s12974-019-1443-2 9. Davalos D, Grutzendler J, Yang G, Kim JV, Zuo Y, Jung S, Littman DR, Dustin ML, Gan W-B. ATP mediates rapid microglial response to local brain injury in vivo. Nat Neurosci (2005) 8 752-758. DOI: 10.1038/nn1472 10. Nimmerjahn A, Kirchhoff F, Helmchen F. Resting microglial cells are highly dynamic surveillants of brain parenchyma in vivo. Science (2005) 308 1314-1318. DOI: 10.1126/science.1110647 11. Hickman SE, Kingery ND, Ohsumi TK, Borowsky ML, Wang L-c, Means TK, El Khoury J. The microglial sensome revealed by direct RNA sequencing. Nat Neurosci (2013) 16 1896-1905. DOI: 10.1038/nn.3554 12. Orihuela R, McPherson CA, Harry GJ. Microglial M1/M2 polarization and metabolic states. Br J Pharmacol (2016) 173 649-665. DOI: 10.1111/bph.13139 13. Shrivastava R, Shukla N. Attributes of alternatively activated (M2) macrophages. Life Sci (2019) 224 222-231. DOI: 10.1016/j.lfs.2019.03.062 14. Ortega-Gómez A, Perretti M, Soehnlein O. Resolution of inflammation: an integrated view. EMBO Mol Med (2013) 5 661-674. DOI: 10.1002/emmm.201202382 15. Franco R, Lillo A, Rivas-Santisteban R, Reyes-Resina I, Navarro G. Microglial adenosine receptors: from preconditioning to modulating the M1/M2 balance in activated cells. Cells (2021) 10 1124. DOI: 10.3390/cells10051124 16. Dubbelaar ML, Kracht L, Eggen BJL, Boddeke E. The kaleidoscope of microglial phenotypes. Front Immunol (2018) 9 1753. DOI: 10.3389/fimmu.2018.01753 17. Jian M, Kwan JS-C, Bunting M, Ng RC-L, Chan KH. Adiponectin suppresses amyloid-β oligomer (AβO)-induced inflammatory response of microglia via AdipoR1-AMPK-NF-κB signaling pathway. J Neuroinflammation (2019) 16 110. DOI: 10.1186/s12974-019-1492-6 18. Xu X, Gao W, Li L, Hao J, Yang B, Wang T, Li L, Bai X, Li F, Ren H, Zhang M, Zhang L, Wang J, Wang D, Zhang J, Jiao L. Annexin A1 protects against cerebral ischemia-reperfusion injury by modulating microglia/macrophage polarization via FPR2/ALX-dependent AMPK-mTOR pathway. J Neuroinflammation (2021) 18 119. DOI: 10.1186/s12974-021-02174-3 19. Patel SS, Acharya A, Ray RS, Agrawal R, Raghuwanshi R, Jain P. Cellular and molecular mechanisms of curcumin in prevention and treatment of disease. Crit Rev Food Sci Nutr (2020) 60 887-939. DOI: 10.1080/10408398.2018.1552244 20. Zhu HT, Bian C, Yuan JC, Chu WH, Xiang X, Chen F, Wang CS, Feng H, Lin JK. Curcumin attenuates acute inflammatory injury by inhibiting the TLR4/MyD88/NF-κB signaling pathway in experimental traumatic brain injury. J Neuroinflammation (2014) 11 59. DOI: 10.1186/1742-2094-11-59 21. Zhang J, Zheng Y, Luo Y, Du Y, Zhang X, Fu J. Curcumin inhibits LPS-induced neuroinflammation by promoting microglial M2 polarization via TREM2/ TLR4/ NF-κB pathways in BV2 cells. Mol Immunol (2019) 116 29-37. DOI: 10.1016/j.molimm.2019.09.020 22. Panaro MA, Corrado A, Benameur T, Paolo CF, Cici D, Porro C. The emerging role of curcumin in the modulation of TLR-4 signaling pathway: focus on neuroprotective and anti-rheumatic properties. Int J Mol Sci (2020) 21 2299. DOI: 10.3390/ijms21072299 23. Parada E, Buendia I, Navarro E, Avendaño C, Egea J, López MG. Microglial HO-1 induction by curcumin provides antioxidant, antineuroinflammatory, and glioprotective effects. Mol Nutr Food Res (2015) 59 1690-1700. DOI: 10.1002/mnfr.201500279 24. Cianciulli A, Calvello R, Porro C, Trotta T, Salvatore R, Panaro MA. PI3k/Akt signalling pathway plays a crucial role in the anti-inflammatory effects of curcumin in LPS-activated microglia. Int Immunopharmacol (2016) 36 282-290. DOI: 10.1016/j.intimp.2016.05.007 25. Tegenge MA, Rajbhandari L, Shrestha S, Mithal A, Hosmane S, Venkatesan A. Curcumin protects axons from degeneration in the setting of local neuroinflammation. Exp Neurol (2014) 253 102-110. DOI: 10.1016/j.expneurol.2013.12.016 26. Viola A, Munari F, Sánchez-Rodríguez R, Scolaro T, Castegna A. The metabolic signature of macrophage responses. Front Immunol (2019) 10 1462. DOI: 10.3389/fimmu.2019.01462 27. Wang F, Zhang S, Jeon R, Vuckovic I, Jiang X, Lerman A, Folmes CD, Dzeja PD, Herrmann J. Interferon gamma induces reversible metabolic reprogramming of M1 macrophages to sustain cell viability and pro-inflammatory activity. EBioMedicine (2018) 30 303-316. DOI: 10.1016/j.ebiom.2018.02.009 28. Banskota S, Wang H, Kwon YH, Gautam J, Gurung P, Haq S, Hassan FMN, Bowdish DM, Kim JA, Carling D, Fullerton MD, Steinberg GR, Khan WI. Salicylates ameliorate intestinal inflammation by activating macrophage AMPK. Inflamm Bowel Dis (2021) 27 914-926. DOI: 10.1093/ibd/izaa305 29. Saha S, Shalova IN, Biswas SK. Metabolic regulation of macrophage phenotype and function. Immunol Rev (2017) 280 102-111. DOI: 10.1111/imr.12603 30. Yang B, Luo G, Zhang C, Feng L, Luo X, Gan L. Curcumin protects rat hippocampal neurons against pseudorabies virus by regulating the BDNF/TrkB pathway. Sci Rep (2020) 10 22204. DOI: 10.1038/s41598-020-78903-0 31. Pizzi M. Sampling variation of the fifty percent end-point, determined by the Reed-Muench (Behrens) method. Hum Biol (1950) 22 151-190. PMID: 14778593 32. Young MD, Wakefield MJ, Smyth GK, Oshlack A. Gene ontology analysis for RNA-seq: accounting for selection bias. Genome Biol (2010) 11 R14. DOI: 10.1186/gb-2010-11-2-r14 33. Mao X, Cai T, Olyarchuk JG, Wei L. Automated genome annotation and pathway identification using the KEGG Orthology (KO) as a controlled vocabulary. Bioinformatics (2005) 21 3787-3793. DOI: 10.1093/bioinformatics/bti430 34. Schneider CA, Rasband WS, Eliceiri KW. NIH Image to ImageJ: 25 years of image analysis. Nat Methods (2012) 9 671-675. DOI: 10.1038/nmeth.2089 35. Schwabenland M, Brück W, Priller J, Stadelmann C, Lassmann H, Prinz M. Analyzing microglial phenotypes across neuropathologies: a practical guide. Acta Neuropathol (2021) 142 923-936. DOI: 10.1007/s00401-021-02370-8 36. Van den Bossche J, Baardman J, Otto NA, van der Velden S, Neele AE, van den Berg SM, Luque-Martin R, Chen HJ, Boshuizen MC, Ahmed M, Hoeksema MA, de Vos AF, de Winther MP. Mitochondrial dysfunction prevents repolarization of inflammatory macrophages. Cell Rep (2016) 17 684-696. DOI: 10.1016/j.celrep.2016.09.008 37. Yin Z, Han Z, Hu T, Zhang S, Ge X, Huang S, Wang L, Yu J, Li W, Wang Y, Li D, Zhao J, Wang Y, Zuo Y, Li Y, Kong X, Chen F, Lei P. Neuron-derived exosomes with high miR-21-5p expression promoted polarization of M1 microglia in culture. Brain Behav Immun (2020) 83 270-282. DOI: 10.1016/j.bbi.2019.11.004 38. Herzig S, Shaw RJ. AMPK: guardian of metabolism and mitochondrial homeostasis. Nat Rev Mol Cell Biol (2018) 19 121-135. DOI: 10.1038/nrm.2017.95 39. Yu Y, Cai W, Zhou J, Lu H, Wang Y, Song Y, He R, Pei F, Wang X, Zhang R, Liu H, Wei F. Anti-arthritis effect of berberine associated with regulating energy metabolism of macrophages through AMPK/ HIF-1α pathway. Int Immunopharmacol (2020) 87 106830. DOI: 10.1016/j.intimp.2020.106830 40. Lan R, Wan Z, Xu Y, Wang Z, Fu S, Zhou Y, Lin X, Han X, Luo Z, Miao J, Yin Y. Taurine reprograms mammary-gland metabolism and alleviates inflammation induced by. Front Immunol (2021) 12 696101. DOI: 10.3389/fimmu.2021.696101 41. Kracht M, Müller-Ladner U, Schmitz ML. Mutual regulation of metabolic processes and proinflammatory NF-κB signaling. J Allergy Clin Immunol (2020) 146 694-705. DOI: 10.1016/j.jaci.2020.07.027 42. Zhao X, Cui Q, Fu Q, Song X, Jia R, Yang Y, Zou Y, Li L, He C, Liang X, Yin L, Lin J, Ye G, Shu G, Zhao L, Shi F, Lv C, Yin Z. Antiviral properties of resveratrol against pseudorabies virus are associated with the inhibition of IκB kinase activation. Sci Rep (2017) 7 8782. DOI: 10.1038/s41598-017-09365-0 43. Chhatbar C, Prinz M. The roles of microglia in viral encephalitis: from sensome to therapeutic targeting. Cell Mol Immunol (2021) 18 250-258. DOI: 10.1038/s41423-020-00620-5 44. Hirano T. IL-6 in inflammation, autoimmunity and cancer. Int Immunol (2021) 33 127-148. DOI: 10.1093/intimm/dxaa078 45. Cinelli MA, Do HT, Miley GP, Silverman RB. Inducible nitric oxide synthase: Regulation, structure, and inhibition. Med Res Rev (2020) 40 158-189. DOI: 10.1002/med.21599 46. Forrester SJ, Kikuchi DS, Hernandes MS, Xu Q, Griendling KK. Reactive oxygen species in metabolic and inflammatory signaling. Circ Res (2018) 122 877-902. DOI: 10.1161/CIRCRESAHA.117.311401 47. Zeng F, Wu Y, Li X, Ge X, Guo Q, Lou X, Cao Z, Hu B, Long NJ, Mao Y, Li C. Custom-made ceria nanoparticles show a neuroprotective effect by modulating phenotypic polarization of the microglia. Angew Chem Int Ed Engl (2018) 57 5808-5812. DOI: 10.1002/anie.201802309 48. Nam HY, Nam JH, Yoon G, Lee JY, Nam Y, Kang HJ, Cho HJ, Kim J, Hoe HS. Ibrutinib suppresses LPS-induced neuroinflammatory responses in BV2 microglial cells and wild-type mice. J Neuroinflammation (2018) 15 271. DOI: 10.1186/s12974-018-1308-0 49. Zhao W, Xie W, Xiao Q, Beers DR, Appel SH. Protective effects of an anti-inflammatory cytokine, interleukin-4, on motoneuron toxicity induced by activated microglia. J Neurochem (2006) 99 1176-1187. DOI: 10.1111/j.1471-4159.2006.04172.x 50. Kaewmool C, Kongtawelert P, Phitak T, Pothacharoen P, Udomruk S. Protocatechuic acid inhibits inflammatory responses in LPS-activated BV2 microglia via regulating SIRT1/NF-κB pathway contributed to the suppression of microglial activation-induced PC12 cell apoptosis. J Neuroimmunol (2020) 341 577164. DOI: 10.1016/j.jneuroim.2020.577164 51. Nair S, Sobotka KS, Joshi P, Gressens P, Fleiss B, Thornton C, Mallard C, Hagberg H. Lipopolysaccharide-induced alteration of mitochondrial morphology induces a metabolic shift in microglia modulating the inflammatory response in vitro and in vivo. Glia (2019) 67 1047-1061. DOI: 10.1002/glia.23587 52. Kou RW, Gao YQ, Xia B, Wang JY, Liu XN, Tang JJ, Yin X, Gao JM. Ganoderterpene A, a new triterpenoid from. J Agric Food Chem (2021) 69 12730-12740. DOI: 10.1021/acs.jafc.1c04905 53. Garcia D, Shaw RJ. AMPK: mechanisms of cellular energy sensing and restoration of metabolic balance. Mol Cell (2017) 66 789-800. DOI: 10.1016/j.molcel.2017.05.032 54. Cai P, Feng Z, Feng N, Zou H, Gu J, Liu X, Liu Z, Yuan Y, Bian J. Activated AMPK promoted the decrease of lactate production in rat Sertoli cells exposed to Zearalenone. Ecotoxicol Environ Saf (2021) 220 112367. DOI: 10.1016/j.ecoenv.2021.112367 55. Chen J, Zou L, Lu G, Grinchuk O, Fang L, Ong DST, Taneja R, Ong CN, Shen HM. PFKP alleviates glucose starvation-induced metabolic stress in lung cancer cells via AMPK-ACC2 dependent fatty acid oxidation. Cell Discov (2022) 8 52. DOI: 10.1038/s41421-022-00406-1 56. Cheng J, Xu D, Chen L, Guo W, Hu G, Liu J, Fu S. CIDEA regulates de novo fatty acid synthesis in bovine mammary epithelial cells by targeting the AMPK/PPARγ axis and regulating SREBP1. J Agric Food Chem (2022) 70 11324-11335. DOI: 10.1021/acs.jafc.2c05226 57. Yan H, Ajuwon KM. Mechanism of butyrate stimulation of triglyceride storage and adipokine expression during adipogenic differentiation of porcine stromovascular cells. PLoS ONE (2015) 10 e0145940. DOI: 10.1371/journal.pone.0145940 58. Arthur CJ, Williams C, Pottage K, Płoskoń E, Findlow SC, Burston SG, Simpson TJ, Crump MP, Crosby J. Structure and malonyl CoA-ACP transacylase binding of streptomyces coelicolor fatty acid synthase acyl carrier protein. ACS Chem Biol (2009) 4 625-636. DOI: 10.1021/cb900099e 59. Liao ST, Han C, Xu DQ, Fu XW, Wang JS, Kong LY. 4-Octyl itaconate inhibits aerobic glycolysis by targeting GAPDH to exert anti-inflammatory effects. Nat Commun (2019) 10 5091. DOI: 10.1038/s41467-019-13078-5 60. Huang YQ, Wang Y, Hu K, Lin S, Lin XH. Hippocampal glycerol-3-phosphate acyltransferases 4 and BDNF in the Progress of obesity-induced depression. Front Endocrinol (Lausanne) (2021) 12 667773. DOI: 10.3389/fendo.2021.667773 61. Cai Y, Zhang Y, Chen H, Sun XH, Zhang P, Zhang L, Liao MY, Zhang F, Xia ZY, Man RY, Feinberg MW, Leung SW. MicroRNA-17-3p suppresses NF-κB-mediated endothelial inflammation by targeting NIK and IKKβ binding protein. Acta Pharmacol Sin (2021) 42 2046-2057. DOI: 10.1038/s41401-021-00611-w
--- title: 'Time-resolved transcriptomic profiling of the developing rabbit’s lungs: impact of premature birth and implications for modelling bronchopulmonary dysplasia' authors: - Matteo Storti - Maria Laura Faietti - Xabier Murgia - Chiara Catozzi - Ilaria Minato - Danilo Tatoni - Simona Cantarella - Francesca Ravanetti - Luisa Ragionieri - Roberta Ciccimarra - Matteo Zoboli - Mar Vilanova - Ester Sánchez-Jiménez - Marina Gay - Marta Vilaseca - Gino Villetti - Barbara Pioselli - Fabrizio Salomone - Simone Ottonello - Barbara Montanini - Francesca Ricci journal: Respiratory Research year: 2023 pmcid: PMC10015812 doi: 10.1186/s12931-023-02380-y license: CC BY 4.0 --- # Time-resolved transcriptomic profiling of the developing rabbit’s lungs: impact of premature birth and implications for modelling bronchopulmonary dysplasia ## Abstract ### Background Premature birth, perinatal inflammation, and life-saving therapies such as postnatal oxygen and mechanical ventilation are strongly associated with the development of bronchopulmonary dysplasia (BPD); these risk factors, alone or combined, cause lung inflammation and alter programmed molecular patterns of normal lung development. The current knowledge on the molecular regulation of lung development mainly derives from mechanistic studies conducted in newborn rodents exposed to postnatal hyperoxia, which have been proven useful but have some limitations. ### Methods Here, we used the rabbit model of BPD as a cost-effective alternative model that mirrors human lung development and, in addition, enables investigating the impact of premature birth per se on the pathophysiology of BPD without further perinatal insults (e.g., hyperoxia, LPS-induced inflammation). First, we characterized the rabbit’s normal lung development along the distinct stages (i.e., pseudoglandular, canalicular, saccular, and alveolar phases) using histological, transcriptomic and proteomic analyses. Then, the impact of premature birth was investigated, comparing the sequential transcriptomic profiles of preterm rabbits obtained at different time intervals during their first week of postnatal life with those from age-matched term pups. ### Results Histological findings showed stage-specific morphological features of the developing rabbit’s lung and validated the selected time intervals for the transcriptomic profiling. Cell cycle and embryo development, oxidative phosphorylation, and WNT signaling, among others, showed high gene expression in the pseudoglandular phase. Autophagy, epithelial morphogenesis, response to transforming growth factor β, angiogenesis, epithelium/endothelial cells development, and epithelium/endothelial cells migration pathways appeared upregulated from the 28th day of gestation (early saccular phase), which represents the starting point of the premature rabbit model. Premature birth caused a significant dysregulation of the inflammatory response. TNF-responsive, NF-κB regulated genes were significantly upregulated at premature delivery and triggered downstream inflammatory pathways such as leukocyte activation and cytokine signaling, which persisted upregulated during the first week of life. Preterm birth also dysregulated relevant pathways for normal lung development, such as blood vessel morphogenesis and epithelial-mesenchymal transition. ### Conclusion These findings establish the 28-day gestation premature rabbit as a suitable model for mechanistic and pharmacological studies in the context of BPD. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12931-023-02380-y. ## Introduction Extremely premature infants (born below 28 weeks of gestation) are at high risk of developing bronchopulmonary dysplasia (BPD) [1, 2]. At delivery, the lungs of extremely premature infants are still in the boundaries between the canalicular (~ 16–24 weeks of pregnancy) and saccular (~ 24–36 weeks) phases of lung development and are morphologically and biochemically immature. At this early stage of development, their lungs still lack alveolar structures and a fully mature surfactant system, which is reflected by their inability to sustain adequate gas exchange [3]. Consequently, extremely premature infants usually require intensive respiratory support to manage their respiratory distress, including assisted ventilation and supplementary oxygen [4]. Preterm birth, perinatal infections and inflammation, and life-saving therapies like post-natal oxygen and mechanical ventilation are strongly associated with the development of BPD [5]. Such triggers, alone or combined, may cause lung inflammation and alter programmed molecular patterns of normal fetal lung development, most notably normal alveolarization and pulmonary vascularization [5, 6], which clinically manifests as a chronic oxygen dependency in BPD infants. The current knowledge on the molecular regulation of lung development mainly derives from animal studies due to the inherently limited availability of lung samples from preterm infants. Mice and rats exposed to postnatal hyperoxia or antenatal inflammation have been widely used to mimic BPD due to their limited maintenance cost, short gestation and large litter size [7, 8]. However, unlike term human neonates, who are born in the alveolar phase, term rodents are delivered in the saccular phase, displaying structurally immature but functionally mature lungs [7, 8]. Large animal models (e.g., non-human primates and lambs) provide more accurate physiological models than small laboratory animals; for instance, alveolarization starts before term birth, and their developmental timeline is comparable to humans [9]. Moreover, they can be delivered prematurely and managed with clinical-like supportive conditions [9, 10]. Nevertheless, large animal models have disadvantages such as long gestation, small litter size, high maintenance costs, and ethical limitations. Rabbit models represent a good compromise between small and large animals. They are cost-effective, have a short gestation period (31 days), have a large litter size, their lung development mirrors the human one and they can be delivered prematurely [11]. Premature rabbits delivered at 28 days of gestation (i.e., at the early saccular phase) display mild to moderate respiratory distress and impaired postnatal lung development compared to age-matched term rabbit pups [12]. The 28-day gestation preterm rabbit model exposed to postnatal hyperoxia has been used in pharmacological studies in the context of BPD [13–17]. In recent years, the combination of advanced analytical methods with bioinformatic tools has enabled high-volume analysis, integration, and interpretation of the molecular pathways involved in lung development. Microarray and proteomic technologies have been used to characterize the biomolecular aspects of lung development in murine models [18–23]. Moreover, RNA sequencing analyses have been performed on lung tissue collected at different developmental stages in mice, piglets and macaques [24–26]. Nevertheless, just a few studies conducted in mice and rhesus macaques have characterized lung developmental stages from early fetal to later postnatal phases [22, 25]. To the best of our knowledge, a time-dependent molecular characterization of the distinct stages of lung development has not yet been performed in rabbits. The aim of the present study is to delve into the translational power of premature rabbits as a suitable model of BPD. For this purpose, we first characterized the normal lung development of the rabbit through the sequential application of histological, transcriptomic, and proteomic analyses of the different stages of lung development (i.e., pseudoglandular, canalicular, saccular and alveolar phases). Then, we investigated the impact of premature birth on the programmed molecular pathways of the lung development, comparing the transcriptomic profiles of preterm rabbits obtained during their first week of life with those from age-matched term pups. Lastly, we compared the transcriptomic results obtained in the present study with similar available records on lung development in mice [22]. ## In vivo protocol and tissue collection All experiments were approved by the intramural Animal Welfare Body and the Italian Ministry of Health (Prot. n° $\frac{744}{2017}$-PR and n° $\frac{899}{2018}$-PR) and comply with the European regulations for animal care. Timed pregnant rabbits (New Zealand White) were provided by Charles River Laboratories (Domaine des Oncins, France) and maintained with food and water ad libitum until the Caesarian section (C-section) or natural birth occurred (31st day of gestation). For the normal lung development study, fetal (F) pups were extracted by C-section, and the lungs were immediately harvested at the 20th, 23rd, 25th, 27th, 28th, and 29th days of gestation ($$n = 3$$–6 per time point, from F20 to F29, Fig. 1). Dams were sedated with intramuscular (i.m.) medetomidine 2 mg/kg (Domitor®, Orion Pharma, Finland). Ten minutes later, they received i.m. 25 mg/kg of ketamine (Imalgene 1000®, Merial, France) and 5 mg/kg of xylazine (Rompun®, Bayer, Germany). Subsequently, dams were euthanized with an overdose of pentothal sodium (50 mg/kg, MSD Animal Health, USA). The abdomen was immediately opened, and the uterus was exposed to extract all pups through hysterectomy. Term (T) rabbits were naturally delivered on the 31st day of gestation (T31) and maintained with their mothers at room air in individual cages up to 11 days of postnatal age. Lung samples ($$n = 3$$–6 per time point) were collected immediately after birth (T31) or 4, 6 and 11 days after birth (T35, T37, T42, respectively).Fig. 1Scheme of the experimental design. Samples were collected at different lung developmental phases (i.e., pseudoglandular, canalicular, saccular and alveolar phases). Fetal (F) rabbit pups were extracted via C-section on the 20th, 23rd, 25th, 27th, 28th and 29th day of gestation (F20–F29); lung samples were harvested immediately after delivery, trying to avoid that pups took their first breath. Term (T) pups were naturally delivered on the 31st day of gestation and maintained with their mothers in room air until lung collection: immediately after birth (T31), and 4 (T35), 6 (T37), and 11 days (T42) after natural birth. Preterm (P) rabbits were delivered through C-section on the 28th day of gestation and maintained in room air until lung collection: immediately after birth (P0), and 3 (P3) and 7 (P7) days after preterm delivery Premature (P) pups were extracted on the 28th day of gestation and kept at room air for up to 7 days. Animal care and feeding protocols have been described in detail elsewhere [12, 15]. The lung samples of premature pups were harvested 1 h (P0), 3 days (P3) and 7 days (P7) after premature delivery ($$n = 3$$ per time point, Fig. 1). All pups were euthanized with a pentothal sodium overdose before lung harvesting. The lungs were immediately surgically dissected to avoid contact with the surrounding environment, weighted, and selectively separated into the right and left lungs. At least three lung pairs were collected at each fetal and postnatal time point, each lung obtained from pups delivered from a different dam. The right lungs were dedicated to transcriptomic analysis, while histological analysis was performed on the left ones. Exclusively for very early fetal time points (20 and 23 gestational days), transcriptomic and histological analyses were performed on whole lungs due to their reduced size. In addition, for proteomic analyses, whole lungs were harvested from littermates delivered from the same mothers of the RNA-seq pups in fetal and term groups (F25, F27, F28, F29, T31 and T35) and stored at − 80 °C. ## Lung tissue histomorphometry Lungs were fixed with $10\%$ formalin buffer (Sigma-Aldrich, Germany) under constant pressure (25 cmH2O). After 48 h, lung samples were dehydrated in graded alcohol solutions, xylene clarified, and paraffin-embedded. Serial 5 µm thick sections were obtained using a rotary microtome and stained with hematoxylin and eosin, Masson’s trichrome or orcein, according to the manufacturer’s specifications (Histo-Line Laboratories). Histological slides were acquired as whole slide images (WSI) by a digital slide scanner (Nanozoomer S-60, Hamamatsu, Japan). Lung development was histomorphometrically analyzed by calculating the tissue density (TD), radial alveolar count (RAC), and medial thickness of pre- and intra-acinar arteries (MT%). The tissue density was determined using an application developed within the Visiopharm image analysis software (Hoersholm, Denmark). The stained tissue area of the WSI was assessed using a threshold-based mask on a lung segment, predicting the exclusion of air spaces and the inclusion of cells and nuclei. The percentage of the stained area was calculated by dividing the stained area by the total selected area (× 100). The RAC was determined by drawing a perpendicular line from the lumen of the terminal bronchiole to the nearest connective tissue septum or pleural margin, and the number of saccules or alveoli crossed by this line was counted [27, 28]. For the evaluation of MT%, ten random peripheric muscularized vessels with an external diameter (ED) of approximately 100 µm, corresponding to the pre-and intra-acinar arteries in rabbits, were selected for each section. Their external (ED) and internal diameter (ID) along the shortest axis of the vessel was measured at 40 × magnification, and MT% was calculated by applying the following formula [29]:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{MT}}\% = \left({{\text{ED}} - {\text{ID}}} \right)/{\text{ED}} \times {1}00$$\end{document}MT%=ED-ID/ED×100 This proportional parameter nullifies the effect of vasodilation, vasoconstriction, and tissue shrinkage. The statistical comparisons of the histomorphometry parameters at different lung developmental phases were performed by one-way ANOVA followed by Tukey’s test (GraphPad Prism 8.4.3, San Diego, CA, USA). A $P \leq 0.05$ was accepted as significant. ## Transcriptomic profiling: mRNA purification, library preparation and sequencing After collection, lungs were immediately placed in RNA Later (Sigma Aldrich, USA) and stored at − 20 °C. Samples were homogenized in QIAzol® Lysis Reagent. Total RNA was extracted with the miRNeasy Mini Kit protocol (QIAGEN, Germany), using an automated method (QIAcube: QIAGEN, Germany) adapted to include DNase I treatment. RNA concentration was measured using Qubit 4 fluorometer (Thermo Fisher Scientific, USA). RNA integrity was assessed by checking the 2:1 ratio of 18S and 28S ribosomal RNA bands and RNA Integrity Number (RIN) by agarose gel electrophoresis and Bioanalyzer RNA 6000 Nano Kit (Agilent, USA), respectively. Lung-derived RNA was suitable for RNA-sequencing (RIN > 8). Libraries for high-throughput RNA sequencing were prepared using the QuantSeq FWD kit (Lexogen, Austria) and sequenced in three different runs with the Illumina NexSeq500 platform, which generated at least 20 million reads for each sample. $96\%$ of the reads were mapped to the rabbit genome. ## Bioinformatic analyses Read quality analysis was performed with the FastQC tool. The adapters’ sequences were trimmed with BBduk from the suite BBtools [30]. RNA-seq reads were aligned to the rabbit genome (Oryctolagus cuniculus, OryCun2.0) with STAR [31], and gene counts were obtained with HTSeq-count [32], using the latest Ensembl rabbit annotation [33]. A custom script was employed to account for reads mapping to unannotated 3’ UTRs of protein-coding genes (accessible at https://github.com/danilotat/UTR_add_extend_GTF). Data are deposited at the Gene Expression Omnibus (GEO) repository under the accession number GSE220843. The batch effect due to the different sequencing runs was removed using removeBatchEffect function in limma [34]. Each sample count was normalized to the sample library size and transformed into Counts Per Million (CPM). Expressed genes were filtered using a threshold of 0.5 CPM in at least 3 samples. A Trimmed Mean of M values (TMM) normalization was applied. Differentially expressed genes (DEGs) were identified with limma’s voom tool (Bioconductor, R package). Genes were deemed to be differentially expressed if the absolute fold change (FC) was > 2 and the adjusted P ≤ 0.05. ## Gene co-expression analysis Modules of co-expressed genes were identified using the weighted gene co-expression network analysis (WGCNA) package in R [35]. This analysis was performed on data collected from fetal and term groups. Only genes with a minimum expression of 0.5 CPM in all samples of at least 3 out of the 4 developmental phases (pseudoglandular–canalicular–saccular–alveolar) were used for module construction. Modules were identified on a signed network using the blockwiseModules function (WGCNA package in R) using default parameters, except for the soft thresholding power set at 10, minimum module size set at 30, and mergeCutHeight set at 0.25. Module eigengenes (ME) were correlated with histological parameters using Pearson correlation. Pathway enrichment analysis was performed using Metascape software [36]. Gene Ontology Biological Processes, KEGG Pathway, Reactome, and Hallmark Gene Sets databases were used. Only terms with q-values ≤ 10–4 were considered significantly enriched. Heat maps were generated using the Morpheus tool [37]. Principal Component Analysis (PCA) was conducted using R and visualized with the package scatterplot3D. ## Proteomic analysis Whole rabbit lungs from the F25, F27, F28, F29, T31 and T35 ($$n = 3$$ per time point) time points were homogenized in PBS (1 mL/100 mg) using a GentleMax homogenizer. Sodium dodecyl sulfate and dithiothreitol were added at final concentrations of $2\%$ (v/v) and 0.1 M, respectively, to the lysis solution. Each sample was digested with trypsin following the filter-aided sample preparation (FASP) protocol [38]. Peptides were labelled with six-plex tandem mass tags (TMT, TMT 6-plex, Thermo Fisher Scientific) following the manufacturer’s instructions. The 18 available samples were divided into three TMT-6-plex groups, distributing one animal at each time point in each of the TMT 6-plex. TMT sets were analyzed with the Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific, Germany) coupled to a Thermo Scientific Dionex Ultimate 3000 nano RSLC, further coupled with the Advion Triversa Nanomate (Advion BioSciences, Ithaca, NY, USA) as the nanoESI source, performing nanoelectrospray through chip technology in data-dependent acquisition (DDA) mode. For the MS3 analyses for TMT quantification, multiple fragment ions from the previous MS2 scan (SPS ions) were co-selected and fragmented by higher energy collisional dissociation (HCD), using a $65\%$ collision energy and a precursor isolation window of 2 Da. Database searches for protein identification were performed with Proteome Discoverer v2.1.0.81 software (Thermo Fisher Scientific) using Sequest HT search engine and UniProt Oryctolagus cuniculus (2020_04) and contaminants. Peptide mass tolerance was 10 ppm for the MS1 analysis, 0.6 for the MS2, and 20 for the MS3. Protein intensities were normalized using the internal reference scaling (IRS) method [39]. Briefly, only proteins detected in all samples were kept, and three normalization steps were applied: sample Loading normalization, IRS normalization, and TMM normalization. ## Distinct lung developmental stages show significantly different histomorphometry parameters The normal lung development was first characterized through histological analyses and using various histomorphometry parameters. Figure 2 displays representative histological microphotographs of the pseudoglandular, canalicular, saccular and alveolar stages of the rabbit´s lung development. The pseudoglandular phase expands from F16 to F23 in the rabbit (5–17 weeks of gestation in humans and from embryonic [E] day 9.5 to E16.6 days in mice) [3, 11, 40]. At F20, the lung looked like a tubular glandular tissue (Fig. 2A), whereas airway differentiation becomes visible in F23 lungs, corresponding to the transition from the pseudoglandular to the canalicular phase. At this stage, some of the epithelial tubes at their distal end widen into airspaces covered by a more flattened epithelium than the conducting airways one (Fig. 2B, arrows). This distinction allows the recognition of the acinus/ventilatory unit for the first time. Fig. 2Histological characterization and histomorphometry data of the rabbit’s normal lung development. Representative microphotographs of the lung parenchyma during the pseudoglandular (A–C), canalicular (D–F), saccular (G–I) and alveolar phases (J–O) of lung development. The first two columns from the left, show sections stained with hematoxylin and eosin obtained at each lung developmental stage. In the right column, Masson’s trichrome staining (C) shows the scarce presence of intercellular matrix during the pseudoglandular phase, and orcein staining (F, I, L, O) shows fine elastic fibers (dotted arrows). The evolution of tissue density (TD), radial alveolar count (RAC) and the medial thickness of pre- and intra-acinar arteries (MT%) are shown in P, Q and R (as mean ± standard error of the mean of at least six lungs per phase of lung development. Asterisks above horizontal lines indicate a significant difference in the comparisons between different lung developmental phases (ANOVA followed by Tukey’s test; $P \leq 0.05$; $P \leq 0.01$; *$P \leq 0.001$) The canalicular phase encompasses F23 to F27 (16–24 weeks of gestation in humans and E16.6 to E17.4 in mice) [3, 11, 40] and is characterized by the widening, elongation and branching of all airway generations. Simultaneously, there was a marked reduction of the pulmonary interstitial connective tissue. Each terminal bronchiole branched into 2–4 wide and straight acinar canals lined by cuboidal epithelial cells, some of which started to flatten and differentiate into type I pneumocytes (asterisks in Fig. 2D). Vascularization of the surrounding mesenchymal tissue progressively increased (Fig. 2D, E). Orcein staining revealed fine elastic fibers within the inter-canalicular septa, arranged along the perimeter of the air spaces (Fig. 2F). The saccular phase expands from F27 to F30 (24–36 weeks of gestation in humans and E17.4 to postnatal day [PND] 5 in mice). The transition from the canalicular to the saccular phase is characterized by the branching of terminal bronchioles into several prospective alveolar ducts, ending in typical groups of enlarged air spaces (denoted by “s” in Fig. 2E, G). Terminal bronchioles appear associated with thick-walled arterioles. The mesenchymal tissue surrounding the terminal sacs condenses to form thick and highly cellular inter-saccular septa or primary septa. Few low ridges start to protrude from the primary septa to develop into secondary septa (arrowheads in Fig. 2H). Orcein staining at F29 reveals that elastic fibers, essential to elongate the crests and divide the primary saccules into smaller alveoli [41], begin to concentrate in the apical portion of the developing secondary crests (dotted arrows in Fig. 2I). From F29 to T31, the changes in the architecture of the lung parenchyma continued but in a less pronounced way; secondary crests are visible (arrowheads in Fig. 2J) and the alveoli became more abundant, attesting to the entry into the alveolar phase. Like in humans, the alveolar phase starts in the late fetal period in rabbits (at F30) and continues postnatally (starts at 36 weeks of gestation in humans and at PND5 in mice) [3, 11, 40]. After birth, alveoli became dilated, thin-walled, and increasingly cup-shaped (Fig. 2K, M). Also, the wall of the arteries accompanying the bronchioles (pre-and intra-acinar arteries, denoted by “a” in Fig. 2N) became increasingly thin. At T31 and T36, elastic fibers appear highly localized in the apical portion of the developing secondary crests (dotted arrows in Fig. 2L, O). The qualitative histological analysis was complemented with a semiquantitative evaluation of several histomorphometry parameters, including TD, RAC, and MT%. The progressive thinning of interalveolar septa and airspaces enlargement along the lung developmental stages was confirmed by a gradual reduction of the TD (Fig. 2P). An initial mean TD percentage of around $85\%$ was observed in the pseudoglandular phase, which was significantly reduced in later phases ($P \leq 0.001$). The most abrupt and wide variation of TD was observed at the transition from the pseudoglandular to the canalicular phase ($P \leq 0.001$) and the transition between the canalicular and saccular phases ($P \leq 0.001$). Instead, the RAC increased significantly in the saccular phase compared with the pseudoglandular and canalicular phases and peaked in the alveolar phase, as expected (Fig. 2Q). The progressive thinning of the pre- and intra-acinar arterioles is denoted by the gradual decrease of MT% (Fig. 2R). ## Time-resolved transcriptomic profiling of the rabbit’s normal lung development A time-resolved transcriptomic analysis was performed to characterize phase-related expression patterns during lung development. In total, 12,506 protein-coding genes were identified, of which 11,482 ($92\%$) had an identified human orthologue. Samples collected at the same lung developmental phase were grouped together after PCA analysis (Fig. 3A). Fetal samples clustered more tightly in different time points than the term ones, suggesting a progressive gene expression variation from fetal to term groups. Fig. 3A Principal component analysis (PCA) highlighting specific clustering of lung samples belonging to the same time-point. The last number in sample names indicates in which RNA-sequencing run the sample was sequenced. B Histogram representation of differentially expressed genes (DEG) identified by comparative transcriptomic profiling of different developmental phases (P = pseudoglandular, C = canalicular, S = saccular, and A = alveolar phases). Up-regulated (UP) and down-regulated (DN) genes are shown in red and blue, respectively The limma-voom analysis identified 4160 unique genes as differentially expressed between different lung developmental phases in at least one comparison. The number of DEGs in each developmental phase comparison is reported in Fig. 3B. As expected, the number of DEGs increased with the temporal distance between the two phases considered for each comparison, and the highest number of DEGs was identified between the most distant phases (i.e., pseudoglandular vs alveolar phases). Conversely, a lower number of DEGs was detected in the comparisons between chronologically close phases (e.g., pseudoglandular vs canalicular; canalicular vs saccular; and saccular vs alveolar). The lowest number of DEGs was found for the saccular to alveolar transition. Weighted* gene co-expression network analysis (WGCNA) was applied in order to visualize the time-dependent modulation of gene expression. Ten modules of co-expressed genes were thus identified (Fig. 4A). For each module, gene expression appears summarized in a representative module eigengene (ME) profile (i.e., genes with similar temporal expression patterns). MEs 1–4 accounted for about $75\%$ of all the analyzed genes. Enrichment pathway analysis was performed on genes belonging to each ME to highlight specific biological processes involved in different phases of lung development (Fig. 4B). In addition, a correlation analysis was performed between individual MEs and the outcomes of the quantitative evaluation of the histomorphometry parameters (TD, RAC, and MT%, Fig. 4C).Fig. 4Developmental phase-dependent gene expression analysis. A Module eigengene (ME) heatmap representation of gene expression data derived from lung samples collected at the indicated (from fetal to post-natal) developmental time-points. Median-normalized expression levels are shown on a low-to-high scale (blue–white–red). B Results of pathway enrichment analysis performed on distinct modules of co-expressed genes. Representative terms, among the most significantly enriched in one or more modules, are reported (modules 8, 9 and 10 are not shown as they are enriched only in marginally statistically significant pathways; the full list of terms and modules is available in Additional file 1). Color saturation corresponds to enrichment significance (− Log q-values). C: Correlation analysis between MEs and histomorphometry parameters (negative correlations are shown in purple and positive correlations in yellow, − Log p-values are indicated in parenthesis) Module 1, which with over 2500 genes accounts for about $20\%$ of all transcribed genes, was characterized by a gene expression trend that increased from the pseudoglandular to the alveolar phase (Fig. 4A). This module, moreover, displayed the strongest negative correlation with TD ($$P \leq 10$$–18). Enrichment pathway analysis revealed several pathways critical for the latest fetal phases of lung development, such as autophagy, epithelial morphogenesis, response to transforming growth factor-β (TGF-β), angiogenesis, epithelium/endothelial cells development, and epithelium/endothelial cells migration processes. In addition, pathways that might be activated as a consequence of natural birth, such as response to oxidative stress, complement cascade, and tumor necrosis factor-α (TNF-α) signaling via NF-κB, were also part of this module (Fig. 4B). Modules 2, 3, and 6 featured an opposite trend, with high gene expression in the pseudoglandular phase and low gene expression in the later phases of lung development. Among these modules, module 2 displayed the highest positive correlation with TD ($$P \leq 3$$ × 10–12). Cell cycle, embryo development, epithelium morphogenesis, collagen biosynthesis, mTOR signaling, RNA processing, oxidative phosphorylation, and WNT signaling pathways were enriched in these modules related to early embryonic morphogenesis. Interestingly, almost $40\%$ of the genes belonging to these modules code for proteins with a predicted nuclear localization, suggesting that at least a fraction of them may correspond to master regulators of lung development. Gene expression levels in module 4 increased from birth (T31) to T42. This module also featured the highest positive correlation with RAC ($$P \leq 4$$ × 10–13). After birth, innate and adaptive immunity, angiogenesis, and reactive oxygen species metabolic pathways were upregulated. Module 5 is characterized by an increase in gene expression levels from the fetal to early alveolar phase (from F25 up to T35), followed by downregulation in the late stages of the alveolar phase (T37 and T42 groups). Autophagy and vesicle-mediated transport are the most represented pathways in this module. Modules 7 and 8 featured a marked upregulation across the canalicular and saccular phases that dampened off upon transition to the alveolar phase. Modules 9 and 10 displayed the opposite trend, with low expression levels in the canalicular and saccular phases and a subsequent upregulation with high gene expression levels in the late stages of the alveolar phase. Module 7 is enriched in translation and oxidative phosphorylation-related genes, while no significant enrichment could be detected in modules 8, 9, and 10 (see Additional file 1 for complete enrichment analysis). ## Proteomic validation of the transcriptomic profiling of the normal rabbit’s lung development A quantitative proteomic analysis was performed to characterize the proteome profile and validate the transcriptomic analysis of the canalicular, saccular and alveolar stages. Collectively, 28,427 peptide groups were generated, which mapped 4367 proteins, of which 2164 proteins were observed in all analyzed samples. Integration of proteomic and transcriptomic profiles revealed 1927 genes common to both analyses. A subset of genes was selected in order to validate the observed gene expression with the encoded protein levels during different lung developmental stages (a manuscript with the full report of the proteomic analysis is under preparation). Histone deacetylases 1 and 2 (HDAC1 and HDAC2) were more expressed both at the transcriptional and proteomic levels from F25 to F27 (Fig. 5). The expression of surfactant proteins A and B (SFTPA1 and SFTPB) increased from F28 with higher protein upregulation from F29 along the alveolar phase. Catenin beta 1 (CTNNB1) and collagen type I alpha 1 and alpha 2 chain proteins (COL1A1 and COL1A2) are mostly expressed in the alveolar phase. Peroxiredoxin 1, 3 and 6 (PRDX1, PRDX3. PRDX6), catalase (CAT), glutathione peroxidase (GPX1), superoxide dismutase 1 (SOD1), thioredoxin (TXN), platelet and endothelial cell adhesion molecule 1 (PECAM), platelet derived growth factor beta (PDGFRB) and transforming growth factor beta induced (TGFBI) showed a similar expression pattern between protein and mRNA, being mostly expressed in the alveolar phase. Fig. 5Comparison of the expression at mRNA and protein levels for selected genes of interest. Z-score-normalized expression levels are indicated on a low-to-high scale (blue–white–red) ## Premature birth induces distinct gene expression patterns in the developing lung compared with the normal lung development The impact of premature birth on gene expression was evaluated by comparing the transcriptomic profiles of preterm (P0 at 1 h, P3 and P7) and age-matched pups for their physiological lung development (F28, T31 and T35) (Fig. 6A). Upregulated genes outnumbered downregulated genes in all comparisons, ranging from $71.7\%$ to $86.0\%$ of the total DEGs (62 genes in P0 vs F28, 51 in P3 vs T31, and 214 in P7 vs T35 comparisons, respectively) (Fig. 6B). Nevertheless, when considering age-matched pups, genes up-regulated in the P7 vs T35 comparison could either be due to a decrease of gene expression during normal development or to an increase in preterm pups. Therefore, DEG lists were created for P0 vs F28, P3 vs T31 and P7 vs T35 comparisons, which were combined with lists of dysregulated genes during normal lung development (i.e., genes dysregulated in the F28 vs T35 comparison) but not dysregulated after preterm birth (P0 vs P7 comparison) and vice versa (Fig. 6C).Fig. 6The impact of premature birth on the molecular pathways of lung development. A Principal component analysis (PCA) shows preterm pups clustering closely to, yet well separated from, term pups of the same gestational age. Samples were sequenced in three different RNA-sequencing runs and batch-effect correction was applied. B Number of differentially expressed genes between preterm rabbit pups and age-matched term pups at different time points. Upregulated and downregulated genes are shown in red and blue, respectively. C Heatmap representing expression levels during normal development or after premature birth. Gene set 1 and gene set 2 refer to genes and pathways respectively upregulated and downregulated in term pups but not modulated in preterm pups; gene set 3 and gene set 4 refer to genes and pathways respectively upregulated and downregulated in preterm pups but not modulated in term pups. The last number in sample names indicates in which RNA-sequencing run the sample was sequenced. Z-score-normalized expression level is indicated on a low-to-high scale (blue–white–red). Main pathways enriched in each gene set are indicated on the right Pathway enrichment analysis revealed a prevalence of upregulated immune system-related genes and pathways in preterm animals compared to age-matched term pups. The significance of these pathways increased as a function of post-natal time (i.e., from P3 vs T31 to P7 vs T35 comparisons) and moreover were specifically upregulated only in preterm pups (i.e., P7 vs P0 comparison, gene set 3). Notably, regarding the P0 vs F28 comparison, up-regulated genes were significantly enriched in TNF-responsive, NF-κB-regulated genes. Other genes characterized by downregulation during normal development, but not downregulated in preterm pups were enriched in hypoxia pathway-related genes. Genes involved in cell adhesion, heme metabolism, blood vessel morphogenesis, extracellular matrix protein expression (matrisome proteins) and epithelial-mesenchymal transition processes were exclusively upregulated in term animals but not in premature pups (gene set 1, T35 vs F28 comparison). ## Comparison of the rabbit and mice lung transcriptomes between late fetal stage and term birth An independent dataset available from Beauchemin et al. ( GEO: GSE74243) [22] was used to compare mouse and rabbit lung developmental transcriptomic profiles. Specifically, DEG lists comparing E18.5 vs PND0 in mice (intra-saccular phase comparison) and F28 vs T31 in rabbits (saccular vs alveolar) were created and subsequently used for pathways enrichment (Fig. 7A). Characteristic pathways of organ development such as cell cycle, regulation of cell division, cell cycle checkpoint, and mitotic processes were upregulated in both mice (E18.5) and rabbits (F28) at late fetal lung development, with comparable Log q-values for each pathway in both species (Fig. 7B). At term (PND0 in mice and T31 in rabbits) both species were enriched in angiogenesis, vasculature development, blood vessel development, humoral immune system, leukocyte migration, and TNFα signalling via NF-κB even if they born at term in different lung developmental phases (Fig. 7C).Fig. 7Developmental gene expression comparison between rabbits and mice. A Comparison between rabbit and mouse physiological lung development. Rabbits are born at term (T31) in the alveolar phase, whereas mice are born at term (PND0) in the saccular phase. Mice enter the alveolar phase only 4–5 days after birth. B Up-regulated pathways in rabbits (blue line) and mice (grey line) at preterm birth (F28 and E18.5 time points, respectively). C Up-regulated pathways in rabbits (blue line) and mice (grey line) at term birth (T31 and PND0 time points, respectively). The identification of processes enrichment for each comparison was performed using the Metascape software. Only processes with q-values ≤ 10–4 were considered significantly enriched ## Discussion Due to the intrinsic limitations in obtaining representative samples from BPD infants, animal models are essential to decipher the intricate molecular pathways involved in BPD. Rats, mice, rabbits, lambs, and non-human primates have been used to recapitulate the pathophysiology of BPD, incorporating into the animal models well-known clinical “triggers” of BPD such as postnatal hyperoxia, perinatal infections and inflammation, and invasive mechanical ventilation [8–11]. Interestingly, however, less attention has been placed on characterizing the impact of premature birth (i.e., without further perinatal insults) on the molecular regulation of lung development, even though it is a common risk factor in all BPD infants. To fill this knowledge gap, we took advantage of the rabbit model, which represents a cost-effective alternative to non-human primates and enables the delivery of premature rabbit pups with respiratory distress. In the first place, we applied histomorphometry and histological analyses to investigate the progression of the rabbit’s normal lung development. Histomorphometry data indicated a high level of significance for TD and MT% in the pseudoglandular-to-canalicular and canalicular-to-saccular transitions, whereas the changes in TD were less significant in the saccular-to-alveolar transition, without significant differences for MT%. Similarly, the striking difference between the saccular phase and the previous pseudoglandular and canalicular phases in terms of RAC was less pronounced (yet still significant, $P \leq 0.05$) in the transition from the saccular to the alveolar phase. This is in line with the histological observation of less evident changes in the lung parenchyma between the saccular and alveolar phases. Overall, the morphological characterization of the normal rabbit’s lung development allowed clear differentiation of the distinct developmental stages, thereby validating the selected time-points as stage-specific for the transcriptomic and proteomic analyses. Next, we performed a time-resolved transcriptomic and proteomic characterization of the normal rabbit’s lung development. Here, the proteomic data consist of a panel of proteins of interest with the purpose of validating the findings from the transcriptomic profiling. Notably, proteins and transcripts obtained from independent experiments showed a fairly common expression pattern. The transcriptomic analysis demonstrated marked differences in the type and number of dysregulated genes along different developmental stages, with the lowest number of dysregulated genes observed for the comparison between the saccular and alveolar phases. The gene expression patterns and pathways involved during physiological lung development were determined by applying WGCNA analysis, which has been recently used to characterize and identify key pathways involved in several pathologies, including BPD [42–44]. Pathway enrichment analysis on co-expressed gene modules identified different expression patterns during normal lung development. For instance, cell cycle and embryo development-related pathways (genes in ME2 and ME3) appear upregulated in the first developmental phases (up until F27), a sign of organ expansion and early embryonic morphogenesis processes. The genes in the ME1 are characterized by a low expression in the pseudoglandular and canalicular phases and their upregulation in the saccular phase that persists during the alveolar phase. Pathway enrichment analysis of these genes revealed molecular processes that have been described to be essential to specialize the lungs for the extrauterine transition, including autophagy, FOXO-mediated transcription, epithelium/endothelium development pathways, and TGF-β response, among others [45–49]. Interestingly, these pathways increase their expression at F28, corresponding with the starting point for the 28-day gestation premature rabbit model of BPD [13, 14, 16]. Furthermore, the respiratory distress observed in the 28-day gestation rabbit model [12] correlates well with our analysis, showing that the expression of SFTPA1 and SFTPB mRNAs and their subsequent translation started at F28, which remained upregulated during the alveolar phase. Surfactant production and innate immune responses are critical and connected processes necessary for breathing and survival after birth [50]. Our results also show an upregulation of the immune system, response to oxidative stress, complement, and TNF-α signaling pathways in the transition between the saccular and alveolar stages during normal lung development. Initiation of breathing activates critical metabolic and cardiorespiratory changes in the newborn to adapt the lungs to the new environment [51]. In this regard, the genes of ME4, which are characterized by an increased expression pattern after birth (T31), show an upregulation of the reactive oxygen species (ROS) metabolic processes and angiogenesis and cell migration (vascular endothelial growth factor [VEGF] signaling). Accordingly, the proteomic analysis revealed that antioxidant enzymes (PRDX1, CAT, PRDX6, GPX1, SOD1, TXN, and PRDX3) are highly expressed and translated at term (T31). Moreover, the proteomic analysis included other proteins that are mainly expressed during late lung development. For instance, PECAM plays a role in angiogenesis and is important for the process of alveolar septation; the loss of PECAM results in increased endothelial sensitivity to apoptotic stress, an altered response in the recruitment of neutrophils, and decreased angiogenesis [52]. PDGFRB is a tyrosine-protein kinase involved in regulating embryonic development and cell proliferation; its receptor also plays an essential role in blood vessel development by promoting proliferation, migration, and recruitment of pericytes and smooth muscle cells to endothelial [53]. TGFBI is involved in early alveolarization and in pulmonary angiogenesis; this downstream target of TGF-β signaling is a crucial component of the distal lung extracellular matrix, which is necessary for normal alveolar secondary septation [54, 55]. Premature birth is associated with significant respiratory morbidity in human preterm neonates. Respiratory distress syndrome (RDS) and BPD are the most prevalent pulmonary conditions affecting premature neonates, and their incidence increases with decreasing gestational age [5, 56]. RDS is primarily caused by a severe surfactant deficiency and is the consequence of incomplete lung maturation at delivery. In the case of BPD, premature birth represents the starting point of the disease, which develops postnatally in response to perinatal inflammatory events that activate not yet fully understood molecular mechanisms of lung injury and repair that disrupt normal lung development [5]. Premature rabbits closely mimic the lung immaturity observed in human neonates. Rabbit pups delivered after 27 days of gestation suffer from severe RDS and show a very limited life span (i.e., a few hours) despite receiving mechanical ventilation and surfactant treatment [57–59]. After 28 days of gestation, premature rabbits can breathe spontaneously and display a moderate RDS at birth. Salaets et al. have recently compared the pulmonary outcomes of premature rabbits delivered at 28 days of gestation and maintained in room air for seven days with age-matched term controls [12]. Their results indicated that prematurity per se may cause delay in lung development, even in the absence of hyperoxia or mechanical ventilation. Motivated by this study, we performed a transcriptomic analysis using a similar experimental design to investigate the impact of premature birth on pulmonary molecular regulation. Our analysis reveals that premature birth causes a significant dysregulation of the inflammatory response. Remarkably, TNF-responsive, NF-κB regulated genes were significantly upregulated just one hour following premature delivery. NF-κB downstream target genes include innate and adaptive immune response factors, such as cytokines, chemokines, and cell adhesion molecules [60]. Hence, the upregulation of TNF-responsive, NF-κB regulated genes immediately after premature birth seems to trigger the upregulation of downstream inflammatory pathways such as leukocyte activation and cytokine signalling that persist during the first week of life. The analysis also revealed genes that appear dysregulated during normal lung development but are not dysregulated in premature animals, including genes involved in blood vessel morphogenesis, epithelial-mesenchymal transition and matrisome proteins (i.e., genes encoding for the extracellular matrix and associated proteins). Conversely, hypoxia genes were significantly downregulated only in term pups. The differential regulation of hypoxia genes in the early postnatal life may be explained by hypoxic episodes in premature rabbits due to their respiratory distress. Hypoxic episodes are relatively common in premature neonates due to their immature respiratory control [61]. Lastly, we compared our transcriptomic data on the rabbit´s lung development with the transcriptomic study conducted by Beauchemin et al. in mice [22]. Newborn rodents exposed to postnatal hyperoxia or perinatal inflammation have been widely used as models of BPD [7, 8]. At term, rodents are in the early saccular phase of lung development and display histological features of premature infants born at less than 30 weeks of gestation [11], albeit they have fully functional lungs. On the contrary, premature rabbits suffer from respiratory distress at birth, although they are also delivered in the early saccular phase. We, therefore, hypothesized that such functional differences at the starting point of both models (i.e., PND0 in mice and F28 in premature rabbits) would be reflected by the regulation of distinct molecular pathways. The comparison between term mice and rabbits showed that normal physiological birth activated common pathways in both species, despite term mice being in the saccular phase and premature rabbits in the alveolar phase. Interestingly, at the starting point of both BPD models, different gene sets were dysregulated in each specie. For instance, term mice showed an upregulation of pathways related to vascular development, whereas these genes were neither upregulated at F28 (Fig. 7B) nor in the first week of postnatal life following premature delivery (Fig. 6C). These findings suggest different molecular maturation degrees at the starting point of both BPD models, which may partly explain different responses to lung injury [62, 63] and pharmacological interventions [16, 64]. The present study has some limitations. In the first place, although we identified many pathways that are known to be involved in the development of BPD, we could not perform a head-to-head comparison with human data. Secondly, we conducted bulk transcriptomics on lung homogenates, which contain the transcripts of all lung cell types. Therefore, we could neither identify the changes nor the role of specific cell types at each developmental stage. Future studies applying single-cell transcriptomics [26] on the BPD rabbit model and on human samples would represent a significant advance in understanding the molecular pathways driving lung development in health and disease and may reveal novel therapeutic targets. In the present study, term rabbit pups were naturally delivered and mother-reared, whereas premature pups were delivered via C-section and fed milk formula, as would be expected in a clinical setting. We acknowledge that differences in delivery and postnatal care may have influenced our transcriptomic analysis. Nevertheless, the recent study by Salaets et al. [ 12] confirmed a significant impairment of lung development in premature rabbits compared to age-matched term rabbits, even when both term and preterm pups were delivered via C-section and fed with milk formula. Lastly, our analysis of the impact of premature birth was limited to one week of postnatal observation. Since the number of dysregulated genes increases along the first week of life in premature rabbits compared with age-matched term pups, studies are warranted to investigate the transcriptomic profiling and the pulmonary outcomes of premature rabbits at longer time points. ## Conclusion We characterized the rabbit’s normal lung development using histological, transcriptomic and proteomic analyses, and investigated the impact of premature birth on the molecular regulation of this process. Histological findings corroborated that the rabbit’s lung development closely resembles the process in humans, showing developmental stage-specific morphological features and the intrauterine initiation of alveolarization. The time-resolved transcriptomic profile demonstrated the high translational power of the 28-day gestation premature rabbit as a model of BPD. At 28 days of gestation (F28), premature rabbits are delivered in the saccular phase, which concurs with the upregulation of genes and pathways involved in the pathophysiology of BPD, many of them essential to specialize the lungs for the extrauterine transition (e.g., angiogenesis and epithelium morphogenesis pathways). The analysis also revealed a significant impact of premature birth per se, without further perinatal insults (e.g., hyperoxia, LPS-induced inflammation), on the dysregulation of inflammatory and other pathways relevant for normal lung development (e.g., blood vessel morphogenesis and epithelial-mesenchymal transition). Altogether, these findings postulate the premature rabbit model, with or without additional insults, as a complementary alternative to rodent models for early stage mechanistic and pharmacological studies in the context of BPD. ## Supplementary Information Additional file 1: Sequencing statistics and pathway analysis during normal rabbit lung development and after premature delivery. The first sheet (sequencing stats) contains the sequencing statistics. Sheets 2 (Development Pathway analysis) and 3 (Pre-term Pathway analysis) contain the pathway analysis of the rabbit’s normal lung development and the pathways analysis alterations due to premature delivery at 28 days of gestation, respectively. ## References 1. Baker CD, Alvira CM. Disrupted lung development and bronchopulmonary dysplasia: opportunities for lung repair and regeneration. Curr Opin Pediatr (2014.0) 26 306-314. DOI: 10.1097/MOP.0000000000000095 2. Day CL, Ryan RM. Bronchopulmonary dysplasia: new becomes old again!. Pediatr Res (2017.0) 81 210-213. DOI: 10.1038/pr.2016.201 3. Bancalari E, Polin RA. Molecular bases for lung development, injury, and repair. Newborn lung neonatol quest controv (2012.0) 3-27 4. Sweet DG, Carnielli V, Greisen G, Hallman M, Ozek E, Te Pas A. European consensus guidelines on the management of respiratory distress syndrome—2019 update. Neonatology (2019.0) 2019 432-450. DOI: 10.1159/000499361 5. Thébaud B, Goss KN, Laughon M, Whitsett JA, Abman SH, Steinhorn RH. Bronchopulmonary dysplasia. Nat Rev Dis Prim (2019.0) 5 78. DOI: 10.1038/s41572-019-0127-7 6. Jobe AH, Bancalari E. Bronchopulmonary dysplasia. Am J Respir Crit Care Med (2001.0) 163 1723-1729. DOI: 10.1164/ajrccm.163.7.2011060 7. O’Reilly M, Thébaud B. Animal models of bronchopulmonary dysplasia. The term rat models. Am J Physiol Cell Mol Physiol. (2014.0) 307 L948-L958. DOI: 10.1152/ajplung.00160.2014 8. Berger J, Bhandari V. Animal models of bronchopulmonary dysplasia. The term mouse models. Am J Physiol Cell Mol Physiol (2014.0) 307 L936-L947. DOI: 10.1152/ajplung.00159.2014 9. Yoder BA, Coalson JJ. Animal models of bronchopulmonary dysplasia. The preterm baboon models. Am J Physiol Cell Mol Physiol. (2014.0) 307 L970-L977. DOI: 10.1152/ajplung.00171.2014 10. Albertine KH. Utility of large-animal models of BPD: chronically ventilated preterm lambs. Am J Physiol Cell Mol Physiol (2015.0) 308 L983-1001. DOI: 10.1152/ajplung.00178.2014 11. Salaets T, Gie A, Tack B, Deprest J, Toelen J. Modelling bronchopulmonary dysplasia in animals: arguments for the preterm rabbit model. Curr Pharm Des (2017.0) 23 5887-5901. DOI: 10.2174/1381612823666170926123550 12. Salaets T, Aertgeerts M, Gie A, Vignero J, de Winter D, Regin Y. Preterm birth impairs postnatal lung development in the neonatal rabbit model. Respir Res (2020.0) 21 59. DOI: 10.1186/s12931-020-1321-6 13. Salaets T, Tack B, Jimenez J, Gie A, Lesage F, de Winter D. Simvastatin attenuates lung functional and vascular effects of hyperoxia in preterm rabbits. Pediatr Res (2020.0) 87 1193-1200. DOI: 10.1038/s41390-019-0711-2 14. Richter J, Jimenez J, Nagatomo T, Toelen J, Brady P, Salaets T. Proton-pump inhibitor omeprazole attenuates hyperoxia induced lung injury. J Transl Med (2016.0) 14 247. DOI: 10.1186/s12967-016-1009-3 15. Salaets T, Gie A, Jimenez J, Aertgeerts M, Gheysens O, Vande VG. Local pulmonary drug delivery in the preterm rabbit: feasibility and efficacy of daily intratracheal injections. Am J Physiol Lung Cell Mol Physiol. (2019.0) 316 L589-L597. DOI: 10.1152/ajplung.00255.2018 16. 16.Aquila G, Regin Y, Murgia X, Salomone F, Casiraghi C, Catozzi C, et al. Daily intraperitoneal administration of rosiglitazone does not improve lung function or alveolarization in preterm rabbits exposed to hyperoxia. Pharmaceutics. 2022;14:1507. 17. Gie AG, Regin Y, Salaets T, Casiraghi C, Salomone F, Deprest J. Intratracheal budesonide/surfactant attenuates hyperoxia-induced lung injury in preterm rabbits. Am J Physiol Cell Mol Physiol (2020.0) 319 L949-L956. DOI: 10.1152/ajplung.00162.2020 18. Conway RF, Frum T, Conchola AS, Spence JR. Understanding human lung development through in vitro model systems. BioEssays (2020.0) 42 e2000006. DOI: 10.1002/bies.202000006 19. Otulakowski G, Duan W, O’Brodovich H. Global and gene-specific translational regulation in rat lung development. Am J Respir Cell Mol Biol (2009.0) 40 555-567. DOI: 10.1165/rcmb.2008-0284OC 20. Xu Y, Wang Y, Besnard V, Ikegami M, Wert SE, Heffner C. Transcriptional programs controlling perinatal lung maturation. PLoS ONE (2012.0) 7 e37046. DOI: 10.1371/journal.pone.0037046 21. Ljungberg MC, Sadi M, Wang Y, Aronow BJ, Xu Y, Kao RJ. Spatial distribution of marker gene activity in the mouse lung during alveolarization. Data Br (2019.0) 22 365-372. DOI: 10.1016/j.dib.2018.10.150 22. Beauchemin KJ, Wells JM, Kho AT, Philip VM, Kamir D, Kohane IS. Temporal dynamics of the developing lung transcriptome in three common inbred strains of laboratory mice reveals multiple stages of postnatal alveolar development. PeerJ (2016.0) 4 e2318. DOI: 10.7717/peerj.2318 23. Moghieb A, Clair G, Mitchell HD, Kitzmiller J, Zink EM, Kim Y-M. Time-resolved proteome profiling of normal lung development. Am J Physiol Cell Mol Physiol (2018.0) 315 L11-24. DOI: 10.1152/ajplung.00316.2017 24. Jin L, Hu S, Tu T, Huang Z, Tang Q, Ma J. Global long noncoding RNA and mRNA expression changes between prenatal and neonatal lung tissue in pigs. Genes (Basel) (2018.0) 9 1-16. DOI: 10.3390/genes9090443 25. Yu X, Feng L, Han Z, Wu B, Wang S, Xiao Y. Crosstalk of dynamic functional modules in lung development of rhesus macaques. Mol Biosyst R Soc Chem (2016.0) 12 1342-1349. DOI: 10.1039/C5MB00881F 26. Hurskainen M, Mižíková I, Cook DP, Andersson N, Cyr-Depauw C, Lesage F. Single cell transcriptomic analysis of murine lung development on hyperoxia-induced damage. Nat Commun (2021.0) 12 1-19. DOI: 10.1038/s41467-021-21865-2 27. Emery JL, Mithal A. The number of alveoli in the terminal respiratory unit of man during late intrauterine life and childhood. Arch Dis Child (1960.0) 35 544-547. DOI: 10.1136/adc.35.184.544 28. Cooney TP, Thurlbeck WM. The radial alveolar count method of Emery and Mithal: a reappraisal 2—intrauterine and early postnatal lung growth. Thorax (1982.0) 37 580-583. DOI: 10.1136/thx.37.8.580 29. Roubliova XI, Deprest JA, Biard JM, Ophalvens L, Gallot D, Jani JC. Morphologic changes and methodological issues in the rabbit experimental model for diaphragmatic hernia. Histol Histopathol (2010.0) 25 1105-1116. PMID: 20607652 30. 30.JGI: Joint Genome Institute. BBTools User Guide. Available from: https://jgi.doe.gov/data-and-tools/software-tools/bbtools/bb-tools-user-guide/. 31. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S. STAR: ultrafast universal RNA-seq aligner. Bioinformatics (2013.0) 29 15-21. DOI: 10.1093/bioinformatics/bts635 32. Anders S, Pyl PT, Huber W. HTSeq—a Python framework to work with high-throughput sequencing data. Bioinformatics (2015.0) 31 166-169. DOI: 10.1093/bioinformatics/btu638 33. 33.e!Ensembl. Available from: http://www.ensembl.org/Oryctolagus_cuniculus/Info/Index. 34. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res (2015.0) 43 e47-e47. DOI: 10.1093/nar/gkv007 35. Langfelder P, Horvath S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinform (2008.0) 9 559. DOI: 10.1186/1471-2105-9-559 36. Zhou Y, Zhou B, Pache L, Chang M, Khodabakhshi AH, Tanaseichuk O. Metascape provides a biologist-oriented resource for the analysis of systems-level datasets. Nat Commun (2019.0) 10 1523. DOI: 10.1038/s41467-019-09234-6 37. 37.Morpheus: versatile matrix visualization and analysis software. Available from: https://software.broadinstitute.org/morpheus/ 38. Wiśniewski JR. Filter-aided sample preparation for proteome analysis. Methods Mol Biol (2018.0) 1841 3-10. DOI: 10.1007/978-1-4939-8695-8_1 39. Plubell DL, Wilmarth PA, Zhao Y, Fenton AM, Minnier J, Reddy AP. Extended multiplexing of tandem mass tags (TMT) labeling reveals age and high fat diet specific proteome changes in mouse epididymal adipose tissue. Mol Cell Proteomics (2017.0) 16 873-890. DOI: 10.1074/mcp.M116.065524 40. Warburton D, El-Hashash A, Carraro G, Tiozzo C, Sala F, Rogers O. Lung organogenesis. Curr Top Dev Biol (2010.0) 90 73-158. DOI: 10.1016/S0070-2153(10)90003-3 41. Mecham RP. Elastin in lung development and disease pathogenesis. Matrix Biol (2018.0) 73 6-20. DOI: 10.1016/j.matbio.2018.01.005 42. Chen M, Yan J, Han Q, Luo J, Zhang Q. Identification of hub-methylated differentially expressed genes in patients with gestational diabetes mellitus by multi-omic WGCNA basing epigenome-wide and transcriptome-wide profiling. J Cell Biochem (2020.0) 121 3173-3184. DOI: 10.1002/jcb.29584 43. Cai Y, Ma F, Qu L, Liu B, Xiong H, Ma Y. Weighted gene co-expression network analysis of key biomarkers associated with bronchopulmonary dysplasia. Front Genet (2020.0) 11 539292. DOI: 10.3389/fgene.2020.539292 44. Liu W, Su Y, Li S, Chen H, Liu Y, Li X. Weighted gene coexpression network reveals downregulation of genes in bronchopulmonary dysplasia. Pediatr Pulmonol (2021.0) 56 392-399. DOI: 10.1002/ppul.25141 45. Wan H, Dingle S, Xu Y, Besnard V, Kaestner KH, Ang S-L. Compensatory roles of Foxa1 and Foxa2 during lung morphogenesis. J Biol Chem (2005.0) 280 13809-13816. DOI: 10.1074/jbc.M414122200 46. Yeganeh B, Lee J, Ermini L, Lok I, Ackerley C, Post M. Autophagy is required for lung development and morphogenesis. J Clin Invest (2019.0) 129 2904-2919. DOI: 10.1172/JCI127307 47. Bartram U, Speer CP. The role of transforming growth factor β in lung development and disease. Chest (2004.0) 125 754-765. DOI: 10.1378/chest.125.2.754 48. Rackley CR, Stripp BR. Building and maintaining the epithelium of the lung. J Clin Invest (2012.0) 122 2724-2730. DOI: 10.1172/JCI60519 49. Yao J, Guihard PJ, Wu X, Blazquez-Medela AM, Spencer MJ, Jumabay M. Vascular endothelium plays a key role in directing pulmonary epithelial cell differentiation. J Cell Biol (2017.0) 216 3369-3385. DOI: 10.1083/jcb.201612122 50. Han SH, Mallampalli RK. The role of surfactant in lung disease and host defense against pulmonary infections. Ann Am Thorac Soc (2015.0) 12 765-774. DOI: 10.1513/AnnalsATS.201411-507FR 51. Hooper SB, Te Pas AB, Lang J, Van Vonderen JJ, Roehr CC, Kluckow M. Cardiovascular transition at birth: a physiological sequence. Pediatr Res (2015.0) 77 608-614. DOI: 10.1038/pr.2015.21 52. DeLisser HM, Helmke BP, Cao G, Egan PM, Taichman D, Fehrenbach M. Loss of PECAM-1 function impairs alveolarization. J Biol Chem Elsevier (2006.0) 281 8724-8731. DOI: 10.1074/jbc.M511798200 53. Noskovičová N, Petřek M, Eickelberg O, Heinzelmann K. Platelet-derived growth factor signaling in the lung. From lung development and disease to clinical studies. Am J Respir Cell Mol Biol (2015.0) 52 263-284. DOI: 10.1165/rcmb.2014-0294TR 54. Ahlfeld SK, Wang J, Gao Y, Snider P, Conway SJ. Initial suppression of transforming growth factor-β signaling and loss of TGFBI causes early alveolar structural defects resulting in bronchopulmonary dysplasia. Am J Pathol (2016.0) 186 777-793. DOI: 10.1016/j.ajpath.2015.11.024 55. Liu M, Iosef C, Rao S, Domingo-Gonzalez R, Fu S, Snider P. Transforming growth factor-induced protein promotes NF-κB-mediated angiogenesis during postnatal lung development. Am J Respir Cell Mol Biol (2021.0) 64 318-330. DOI: 10.1165/rcmb.2020-0153OC 56. St Clair C, Norwitz ER, Woensdregt K, Cackovic M, Shaw JA, Malkus H. The probability of neonatal respiratory distress syndrome as a function of gestational age and lecithin/sphingomyelin ratio. Am J Perinatol (2008.0) 25 473-480. DOI: 10.1055/s-0028-1085066 57. Bongrani S, Fornasier M, Papotti M, Razzetti R, Curstedt T, Robertson B. Dose-response study of surfactant replacement in immature newborn rabbits. Prenat Neonatal Med (1999.0) 4 71-78 58. Ricci F, Murgia X, Razzetti R, Pelizzi N, Salomone F. In vitro and in vivo comparison between poractant alfa and the new generation synthetic surfactant CHF5633. Pediatr Res (2017.0) 81 369-375. DOI: 10.1038/pr.2016.231 59. Ricci F, Catozzi C, Ravanetti F, Murgia X, D’Aló F, Macchidani N. In vitro and in vivo characterization of poractant alfa supplemented with budesonide for safe and effective intratracheal administration. Pediatr Res (2017.0) 82 1056-1063. DOI: 10.1038/pr.2017.171 60. Alvira CM. Nuclear factor-kappa-B signaling in lung development and disease: one pathway, numerous functions. Birth Defects Res A Clin Mol Teratol (2014.0) 100 202-216. DOI: 10.1002/bdra.23233 61. Di Fiore JM, MacFarlane PM, Martin RJ. Intermittent hypoxemia in preterm infants. Clin Perinatol (2019.0) 46 553-565. DOI: 10.1016/j.clp.2019.05.006 62. Rehan VK, Wang Y, Patel S, Santos J, Torday JS. Rosiglitazone, a peroxisome proliferator-activated receptor-γ agonist, prevents hyperoxia-induced neonatal rat lung injury in vivo. Pediatr Pulmonol (2006.0) 41 558-569. DOI: 10.1002/ppul.20407 63. 63.Jiménez J, Richter J, Nagatomo T, Salaets T, Quarck R, Wagennar A, et al. Progressive vascular functional and structural damage in a bronchopulmonary dysplasia model in preterm rabbits exposed to hyperoxia. Int J Mol Sci. 2016;17:1776. 64. Dasgupta C, Sakurai R, Wang Y, Guo P, Ambalavanan N, Torday JS. Hyperoxia-induced neonatal rat lung injury involves activation of TGF-β and Wnt signaling and is protected by rosiglitazone. Am J Physiol Lung Cell Mol Physiol (2009.0) 296 1031-1041. DOI: 10.1152/ajplung.90392.2008
--- title: Arterial stiffness and its associations with left ventricular diastolic function according to heart failure types authors: - Hack-Lyoung Kim - Jaehoon Chung - Seokmoon Han - Hyun Sung Joh - Woo-Hyun Lim - Jae-Bin Seo - Sang-Hyun Kim - Joo-Hee Zo - Myung-A Kim journal: Clinical Hypertension year: 2023 pmcid: PMC10015827 doi: 10.1186/s40885-022-00233-2 license: CC BY 4.0 --- # Arterial stiffness and its associations with left ventricular diastolic function according to heart failure types ## Abstract ### Background Little is known about the characteristics of arterial stiffness in heart failure (HF). This study was performed to compare the degree of arterial stiffness and its association with left ventricular (LV) diastolic function among three groups: control subjects, patients with HF with reduced ejection fraction (HFrEF), and patients with HF with preserved ejection fraction (HFpEF). ### Methods A total of 83 patients with HFrEF, 68 patients with HFpEF, and 84 control subjects were analyzed. All HF patients had a history of hospitalization for HF treatment. Brachial-ankle pulse wave velocity (baPWV) measurement and transthoracic echocardiography were performed at the same day in a stable condition. ### Results The baPWV was significantly higher in patients with both HFrEF and HFpEF compared to control subjects (1,661 ± 390, 1,909 ± 466, and 1,477 ± 296 cm/sec, respectively; $P \leq 0.05$ for each). After adjustment of age, baPWV values were similar between patients with HFrEF and HFpEF ($$P \leq 0.948$$). In the multiple linear regression analysis, baPWV was significantly associated with both septal e′ velocity (β = –0.360, $$P \leq 0.001$$) and E/e′ (β = 0.344, $$P \leq 0.001$$). However, baPWV was not associated with either of the diastolic indices in HFrEF group. The baPWV was associated only with septal e′ velocity (β = –0.429, $$P \leq 0.002$$) but not with E/e′ in the HFpEF group in the same multivariable analysis. ### Conclusions Although arterial stiffness was increased, its association with LV diastolic function was attenuated in HF patients compared to control subjects. The degree of arterial stiffening was similar between HFrEF and HFpEF. ### Supplementary Information The online version contains supplementary material available at 10.1186/s40885-022-00233-2. ## Introduction Heart failure (HF) is a terminal state of almost all heart diseases, and its prevalence continues to increase with age. Mortality and medical costs due to HF are so enormous that they pose a huge burden to our human society [1, 2]. Therefore, it is important to understand the underlying pathophysiology, and to develop a treatment that can prevent the occurrence of HF based on this. For several decades, many effective drugs for HF have been developed, which have greatly improved the survival rate of patients with HF [3–9]. However, since the mortality rate of HF is still very high, similar to that of some cancers, further efforts to treat HF are continuously required [1, 2]. HF is divided into HF with reduced ejection fraction (HFrEF) and HF with preserved ejection fraction (HFpEF) according to left ventricular ejection fraction (LVEF) [10]. Although HFpEF is also a clinically serious disease due to its high prevalence and poor prognosis, as in HFrEF, underlying pathophysiology and effective long-term treatment has not been well elucidated [11]. Recently, emerging evidence has shown that increased arterial stiffness plays an important role in the development of HFpEF [12–14]. However, most of the previous studies that conducted research into this issue analyzed subjects in the stage before clinically overt HF. Additionally, the role of arterial stiffness in HFrEF is still unknown. This study was performed to investigate in the degree of arterial stiffness and its association with left ventricular (LV) diastolic function in patients with HFrEF and HFpEF. We also compared results in patients with HF to control subjects without HF. ## Study patients This study is a cross-sectional study conducted at a general hospital in a large city (Seoul, Republic of Korea). The study was conducted in accordance with the Declaration of Helsinki. The study protocol was reviewed and approved by the Institutional Review Board of Seoul Metropolitan Government Seoul National University Boramae Medical Center (No. 10–2020-313). Written informed consent was obtained for prospectively enrolled subjects and informed consent was waived by Institutional Review Board for retrospectively enrolled subjects. In the HF groups, the eligible study subjects were patients who had a history of hospitalization for the management of new-onset acute HF or acute exacerbation of chronic HF. At the time of hospitalization, the main diagnosis should be HF. HF patients with LVEF < $40\%$ were further stratified as the HFrEF group, and HF patients with LVEF ≥ $50\%$ were further stratified as the HFpEF group [15]. Relatively healthy subjects without HF and other documented cardiovascular disease were enrolled as the control group. At the time of study enrollment, more than 30 days passed since HF hospitalization, and all study subjects were outpatient in a chronic stage with a stable condition. Both transthoracic echocardiography and brachial-ankle pulse wave velocity (baPWV) measurement were performed on the same day. Subjects with the following conditions were excluded: [1] uncontrolled HF symptoms with New York Heart Association dyspnea scale IV; [2] uncontrolled blood pressure with systolic blood pressure ≥ 180 mmHg, diastolic blood pressure ≥ 110 mmHg, or systolic blood pressure < 90 mmHg; [3] uncontrolled arrhythmia; [4] significant valvular dysfunction, moderate degree or more; [5] presence of pericardial effusion, maximal thickness > 10 mm, and [6] ankle-brachial index < 0.9. Initially 60 subjects (20 control, 20 HFrEF, and 20 HFpEF) were enrolled with informed consent between January 2021 and February 2022. Among them, one in the control group and one in the HFrEF group were excluded from the analysis because baPWV measurement was not performed. Additional 177 subjects (65 control, 64 HFrEF, and 48 HFpEF) were enrolled in the study through a retrospective review of their medical records between January and December 2020. Informed consent was not obtained from these subjects as data were collected retrospectively. Finally, 235 subjects (84 control, 83 HFrEF, and 68 HFpEF) were analyzed in this study. Flow chart for study enrollment is demonstrated in Fig. 1.Fig. 1Flow chart for study subject enrollment. HFrEF, heart failure with reduced ejection fraction; HFpEF, heart failure with preserved ejection fraction; baPWV, brachial-ankle pulse wave velocity ## Data collection Body mass index was calculated as weight in kilograms divided by the square of height in meters (kg/m2). Systolic and diastolic blood pressures were recorded at the time of study enrollment using an oscillometric device. Information on cardiovascular risk factors including hypertension, diabetes mellitus, dyslipidemia, cigarette smoking status, coronary artery disease (CAD), and stroke were obtained. Hypertension was defined basis on previous diagnosis, the current use of antihypertensive medications used to control blood pressure, or blood pressure ≥ $\frac{140}{90}$ mmHg. Diabetes mellitus was defined based on previous diagnosis, the current use of antidiabetic medications used to control hyperglycemia, or fasting blood glucose ≥ 126 mg/dL. Dyslipidemia was defined based on previous diagnosis, the current use of antidyslipidemic medications used to control dyslipidemia, or low-density lipoprotein cholesterol ≥ 160 mg/dL. Smokers were defined as those who had smoked during the past year. CAD included myocardial infarction and coronary revascularization. Stroke was defined as a sudden neurological abnormality with cerebral infarction or hemorrhage in imaging studies. After overnight fasting, venous blood was drawn and the blood levels of the following laboratory parameters were obtained: white blood cell count, hemoglobin, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, triglyceride, creatinine, glucose, glycated hemoglobin, and C-reactive protein. Estimated glomerular filtration rate was calculated by the Modification of Diet in Renal Disease study equation. Information on concomitant cardiovascular medications including calcium channel blockers, beta blockers, renin-angiotensin system blockers, statins, and diuretics was also obtained. ## Transthoracic echocardiography Transthoracic echocardiography was performed using commercially available machines (Vivid E9 and E95, GE Healthcare, Horten, Norway; EPIQ 7 and EPIQ CVx, Philips Ultrasound Inc., Bothell, WA, USA). Echocardiography was performed according to standardized protocols based on current guidelines’ recommendations [16, 17]. LV dimension was measured using M-mode echocardiography. LV ejection fraction was calculated using Simpson biplane method. LV mass (g) was calculated using the following formula: 0.8 × [1.04 × {(LV end-diastolic dimension) + (interventricular septal wall thickness) + (posterior wall thickness)}3 – (LV end-diastolic dimension)3] + 0.6. LV mass index was calculated as LV mass / body surface area. In apical four-chamber view, peak velocities of E and A waves of mitral inflow during diastole were obtained using a pulsed wave Doppler, and E/A ratio was calculated. Deceleration time of E wave was also measured. Using the tissue Doppler imaging technique, the peak velocity of mitral septal annulus (e′) was obtained. Left atrial (LA) volume was measured using the biplane disk summation method and indexed to body surface as LA volume index. In modified four-chamber view, the maximal velocity of tricuspid regurgitation (TR Vmax) was obtained using the continuous Doppler method. In this study, we focused on sepal e′ velocity and E/e′ as indicators of LV diastolic function because these indicators are relatively easy to measure and reliable indicators that are recommended first for the evaluation of left ventricular diastolic function [17]. Interobserver agreements of septal e′ and E/e′ were evaluated by Pearson correlation among 50 subjects. Correlation coefficients were 0.96 and 0.92 for e′ and E/e′, respectively, in our laboratory [18]. ## Brachial-ankle pulse wave velocity measurement Arterial stiffness was assessed using baPWV. The baPWV was measured noninvasively using a volume-plethysmography device (VP‐1000; Colin Co., Komaki, Japan) in the supine position 5 to 10 min after resting in an independent space in a quiet state [19, 20]. On the day of the test, cigarette smoking or consumption of beverages containing caffeine was restricted, and medications that were regularly taken were allowed. Arterial pulse wave was measured on both the brachial artery and posterior tibial artery of the subjects. During measurements, pulse volume waveform, blood pressure, and heart rate were recorded simultaneously. The baPWV were calculated as distance between the brachial and posterior tibial arteries divided by time interval. The distance between the brachial and posterior tibial arteries was estimated based on the height of the subject. The baPWV was measured on the left and right sides, and the average value was used in this study. All baPWV measurements were performed by a single trained operator. Coefficient of variation for intraobserver variability was $5.1\%$ in our laboratory [21]. ## Statistical analysis Continuous variables were expressed as mean ± standard deviation and categorical variables were expressed as number (%). Comparisons among three groups (control, HFrEF, and HFpEF) were performed using analysis of variance (ANOVA) for continuous variables and chi-square test for categorical variables. Bonferroni post-hoc analysis was applied to compare the baPWV mean difference between the two groups. The difference in baPWV among the three groups was further compared by correcting for age through the analysis of covariance (ANCOVA). Linear relations between baPWV and diastolic parameters were assessed using Pearson correlation analysis. Scatter plots demonstrated these correlations. To find independent association between echocardiographic diastolic indices and baPWV, multiple linear regression analysis was performed. The following potential confounders were controlled during multivariable analysis: age, sex, and cardiovascular risk factors including hypertension, diabetes mellitus, and dyslipidemia, and smoking status. All analyses were two-tailed, and clinical significance was defined as $P \leq 0.05.$ All statistical analyses were performed with the statistical package IBM SPSS ver. 23.0 (IBM Corp., Armonk, NY, USA). ## Results Comparisons of clinical characteristics among three groups are shown in Table 1. In overall study patients, mean age was 67.0 ± 12.9 years, and 94 ($40.0\%$) were female. The patients in the HFpEF group were oldest, and the proportion of female patients was highest. The HFpEF group had the highest systolic blood pressure as well as the highest prevalence of cardiovascular risk factors including hypertension, diabetes mellitus, dyslipidemia, previous history of CAD, and stroke. In laboratory findings, patients with HFpEF showed better cholesterol profiles and worse renal function compared to those with HFrEF and control groups. The blood levels of glucose, glycated hemoglobin and C-reactive protein were higher in both HFrEF and HFpEF group compared to control group. HF patients were taking more cardiovascular drugs than the control group. Beta blockers, renin-angiotensin system blockers, and diuretics showed the highest frequency of use in the HFrEF group and calcium channel blockers in the HFpEF group, but there was no difference in the statin use rate among the three groups. Table 1Clinical characteristics of study subjectsCharacteristicControl ($$n = 84$$)HFrEF ($$n = 83$$)HFpEF ($$n = 68$$)P-valueAge (yr)61.3 ± 11.164.6 ± 12.777.0 ± 9.2< 0.001Female sex37 (44.0)19 (22.9)38 (55.9)< 0.001Body mass index (kg/m2)24.5 ± 3.222.4 ± 7.324.3 ± 4.90.033Systolic blood pressure (mmHg)126.0 ± 13.0124.0 ± 18.0139.0 ± 22.0< 0.001Diastolic blood pressure (mmHg)74.3 ± 9.476.2 ± 13.476.5 ± 11.60.436Cardiovascular risk factor Hypertension34 (40.5)54 (65.1)55 (80.9)< 0.001 Diabetes mellitus14 (16.7)30 (36.1)28 (41.2)0.002 Dyslipidemia30 (35.7)25 (30.1)33 (48.5)0.062 Cigarette smoking10 (11.9)26 (31.3)3 (4.4)< 0.001 Coronary artery disease013 (15.7)13 (19.1)0.001 Stroke010 (12.0)10 (14.7)0.020Laboratory finding White blood cell count (/µL)6,136 ± 1,6967,057 ± 1,8656,494 ± 2,4630.252 Hemoglobin (g/dL)14.1 ± 1.315.9 ± 8.712.0 ± 2.0< 0.001 Total cholesterol (mg/dL)181.0 ± 41.0166.0 ± 43.0146.0 ± 62.0< 0.001 LDL cholesterol (mg/dL)108.0 ± 40.0103.0 ± 41.087.0 ± 32.00.019 HDL cholesterol (mg/dL)56.2 ± 13.139.7 ± 10.245.7 ± 13.5< 0.001 Triglyceride (mg/dL)114.0 ± 48.0122.0 ± 68.0128.0 ± 79.00.452 eGFR (mL/min/1.73m2)94.6 ± 19.470.8 ± 26.565.9 ± 32.0< 0.001 Glucose (mg/dL)115.0 ± 31.0123.0 ± 38.0118.0 ± 34.00.659 Glycated hemoglobin (%)5.91 ± 0.776.60 ± 1.206.38 ± 1.25< 0.001 C-reactive protein (mg/dL)0.21 ± 0.821.65 ± 4.311.09 ± 3.150.018Cardiovascular medication Calcium channel blocker9 (10.7)22 (26.5)33 (48.5)< 0.001 Beta blocker11 (13.1)68 (81.9)42 (61.8)< 0.001 RAS blocker10 (11.9)70 (84.3)37 (54.4)< 0.001 Statin31 (36.9)27 (32.5)33 (48.5)0.122 Diuretics6 (7.1)46 (55.4)31 (45.5)< 0.001Values are presented as number (%) or mean ± standard deviationHFrEF Heart failure with reduced ejection fraction, HFpEF Heart failure with preserved ejection fraction, LDL Low-density lipoprotein, HDL High-density lipoprotein, eGFR Estimated glomerular filtration rate, RAS Renin-angiotensin system Results of transthoracic echocardiography are demonstrated in Table 2. Patients with HFrEF had the largest LV systolic and diastolic dimensions and LV mass index. The mean LVEF were $67.4\%$ ± $4.6\%$, $29.7\%$ ± $6.4\%$, and $63.6\%$ ± $8.1\%$, in the control, HFrEF, and HFpEF groups, respectively. Compared to the control group, LV diastolic function was more severely impaired in both patients with HFrEF and HFpEF, which was shown by lower septal e′ velocity as well as by higher E/e′, TR Vmax, and LA volume index. Table 2Echocardiographic findings of study subjectsCharacteristicControl ($$n = 84$$)HFrEF ($$n = 83$$)HFpEF ($$n = 68$$)P-valueLV end-diastolic dimension (mm)47.4 ± 3.655.8 ± 7.449.5 ± 5.1< 0.001LV end-systolic dimension, mm)29.6 ± 3.143.8 ± 8.832.7 ± 5.7< 0.001LV ejection fraction (%)67.4 ± 4.629.7 ± 6.463.6 ± 8.1< 0.001LV mass index (g/m2)85.2 ± 17.8144.0 ± 40.0108.0 ± 30.0< 0.001E/A0.80 ± 0.180.94 ± 0.550.89 ± 0.480.026Deceleration time (ms)223.0 ± 50.0172.0 ± 55.0184.0 ± 56.0< 0.001Peak septal e′ velocity (cm/sec)6.27 ± 1.844.79 ± 1.755.56 ± 2.28< 0.001Septal E/e′10.5 ± 3.416.4 ± 7.416.6 ± 7.2< 0.001Left atrial volume index (mL/m2)31.8 ± 9.343.4 ± 18.257.3 ± 23.8< 0.001TR Vmax (m/sec)2.28 ± 0.232.49 ± 0.552.66 ± 0.46< 0.001Values are presented as mean ± standard deviationHFrEF Heart failure with reduced ejection fraction, HFpEF Heart failure with preserved ejection fraction, LV Left ventricular, TR Vmax Maximal velocity of tricuspid regurgitation Comparison of baPWV values among three groups is demonstrated in Fig. 2. Mean baPWV values were 1,477 ± 296, 1,661 ± 390, and 1,909 ± 466 cm/sec, in the control, HFrEF, and HFpEF groups, respectively (ANOVA, $P \leq 0.001$). In post-hoc analysis, baPWV value was significantly higher in patients with HFrEF than in control subjects ($$P \leq 0.006$$). Also, baPWV value was significantly higher in the HFpEF group compared to the control and HFrEF groups ($P \leq 0.05$ for each). Differences in baPWV values between the control group and HFrEF or HFpEF groups were also significant when age-adjusted using ANCOVA analysis (age-adjusted, $P \leq 0.005$). However, after adjusting for age, no difference in baPWV values was observed between the HFrEF and HFpEF groups (age-adjusted, $$P \leq 0.948$$). Without stratification by heart failure type, baPWV was significantly higher in heart failure group than in control group (1,771 ± 439 vs. 1,477 ± 297 cm/sec) even after controlling for age (age-adjusted, $P \leq 0.001$) (Supplementary Figure S1).Fig. 2Brachial-ankle pulse wave velocity (baPWV) values according to heart failure types. HFrEF, heart failure with reduced ejection fraction; HFpEF, heart failure with preserved ejection fraction. * Age-adjusted value through the analysis of covariance Simple linear correlations between baPWV and diastolic parameters are shown in Fig. 3. The baPWV was significantly correlated with septal e′ velocity in all three groups (r = –0.572, $P \leq 0.05$ for the control group; r = –0.226, $$P \leq 0.040$$ for HFrEF group; r = –0.384, $$P \leq 0.001$$ for HFpEF group). baPWV was significantly correlated with E/e′ in control group ($r = 0.551$, $P \leq 0.001$), but not in the HFrEF ($r = 0.049$, $$P \leq 0.657$$) and HFpEF groups ($r = 0.048$, $$P \leq 0.702$$). Without stratification by heart failure type, baPWV was significantly correlated with septal e′ velocity (r = –0.213, $$P \leq 0.009$$) but not with septal E/e′ (r = –0.016, $$P \leq 0.842$$) (Supplementary Figure S2).Fig. 3Linear correlations between brachial-ankle pulse wave velocity (baPWV) and echocardiographic diastolic indices according to heart failure types. HFrEF, heart failure with reduced ejection fraction; HFpEF, heart failure with preserved ejection fraction In the multiple linear regression analysis (Table 3), baPWV was significantly associated with both septal e′ velocity (β = –0.360, $$P \leq 0.001$$) and E/e′ (β = 0.344, $$P \leq 0.001$$) even after controlling for clinical confounders in the control group. However, baPWV was not associated with septal e′ velocity (β = –0.167, $$P \leq 0.177$$) and E/e′ (β = 0.063; $$P \leq 0.631$$) after controlling for confounders in the HFrEF group. The baPWV was associated with septal e′ velocity (β = –0.429, $$P \leq 0.002$$) but not with E/e′ (β = 0.117, $$P \leq 0.435$$) in the HFpEF group in the same multivariable analysis. Without stratification by heart failure type, baPWV was independently associated with septal e′ velocity (β = –0.281, $$P \leq 0.004$$) but not with E/e′ (β = 0.058, $$P \leq 0.551$$) (Supplementary Table S1).Table 3Independent association between brachial-ankle pulse wave velocity and left ventricular diastolic parametersDependent variableβP-valueControl group Septal e′ velocity–0.3600.001 E/e′0.3440.001HFrEF group Septal e′ velocity–0.1670.177 E/e′0.0630.631HFpEF group Septal e′ velocity–0.4290.002 E/e′0.1170.435β and P values are for brachial-ankle pulse wave velocity. Following clinical covariates were controlled as potential confounders: age, sex, and cardiovascular risk factors including hypertension, diabetes mellitus, dyslipidemia, and cigarette smokingHFrEF Heart failure with reduced ejection fraction, HFpEF Heart failure with preserved ejection fraction ## Discussion Main findings of this study are as follows: [1] baPWV was significantly higher in patients with HFrEF and HFpEF compared to control subjects; [2] although univariable comparison showed that baPWV was significantly higher in patients with HFpEF than in those with HFrEF, it was similar between patients with HFrEF and HFpEF after adjusting for age; and [3] baPWV was significantly associated with septal e′ velocity and E/e′ in the control group, had no association with either of the LV diastolic indices in patients with HFrEF, and was associated only with septal e′ velocity, but not with E/e′ in patients with HFpEF. Our results showed that baPWV was significantly higher in patients with HF than in control subjects who had no HF or other documented cardiovascular disease and stroke. The significance of this difference remains even after adjusting for age, a major determinant of arterial stiffness. It may be due to the fact that patients with HF had more various risk factors that could increase arterial stiffness compared to the control group, which was consistent with the characteristics of our study population. We also presented, for the first time, differences in the degree of arterial stiffness according to HF types. Although the baPWV value in the univariable comparison was higher in the HFpEF group than in the HFrEF group, there was no difference between the two groups after age adjustment. Large-scale data are required to verify our findings. As arterial stiffness increases, LV diastolic function deteriorates. In a stiffened artery, the velocity of reflected wave is increased and merges with the forward wave early [22]. This raises systolic blood pressure and lowers diastolic blood pressure. Increased systolic blood pressure causes LV hypertrophy and decreased diastolic blood pressure reduces coronary perfusion. In addition, with the concept of a shared common risk factors, many cardiovascular risk factors related to arterial stiffening also exacerbate LV diastolic dysfunction [23]. Based on this hypothesis, many existing clinical studies have shown a significant association between increased arterial stiffness and LV diastolic dysfunction in the general population as well as in patients with certain diseases [18, 24–31]. However, in studies demonstrating such ventricular-arterial (VA) coupling, the study subjects were mostly restricted to the general population or subjects without documented cardiovascular disease including HF [25–27, 30]. To the best of our knowledge, there was only two studies showing the association between arterial stiffness and LV diastolic function in patients with established HF [28, 31]. Noguchi et al. [ 28] investigated 44 hypertensive patients with normal LVEF and 31 patients with reduced EF, and showed that cardio-ankle vascular index was correlated with septal e′ velocity in both groups. However, in that study, the definition of HFrEF and HFpEF depended only on LVEF and did not take into account clinical aspects such as hospitalization or symptoms. Additionally, the authors did not perform multivariable analysis [28]. More recently, another study of 107 patients with HFpEF revealed that ambulatory arterial stiffness index was correlated with E/e′ [31]. In our study, baPWV was correlated with e′ velocity in patients with HFpEF but not in those with HFrEF. It seems that the results of each individual study are inconsistent due to differences in the basic characteristics of the study subjects, including race and the method of measuring arterial stiffness. Our study is most meaningful in that it showed a relationship between arterial stiffness and LV diastolic function according to HF types and compared it with the control group. In the comparisons between control and HF groups, our results showed that the degree of arterial stiffness is more severe in the HF groups, but the association between arterial stiffness and LV diastolic function was stronger in the control group. This implies that arterial stiffness has a greater impact on the diastolic function of LV in the stage before the HF onset, and that the effect is somewhat weakened when HF has already occurred. Therefore, this may suggest that strategy targeting arterial stiffness to improve LV diastolic function or prevent HF [32] should be implanted as early as possible. In addition, baPWV was associated with septal e′ velocity in patients with HFpEF but not in patients with HFrEF. This may be a consistent finding with the results of previous studies showing the important role of arterial stiffness in the development of HFpEF [32]. Our study also showed that baPWV was not associated with E/e′ in either type of HF. It has been suggested that e′ velocity is less affected by the LV loading condition than E/e′; thus, e′ velocity is a more reliable indicator of LV diastolic function [33]. Septal e′ may be a better indicator for response monitoring than E/e′ in treatment strategies targeting arterial stiffness especially in patients with HFpEF. Our study has several limitations. First, the associations of baPWV with septal e′ velocity and E/e′ were determined with cross-sectional data; therefore, the causal relationship between arterial stiffness and diastolic function could not be confirmed. Second, although carotid-femoral pulse wave velocity (cfPWV) is the gold standard for the non-invasive measurement of arterial stiffness [34], baPWV was used in our study. However, baPWV is more simple to measure and has a good correlation with cfPWV [23]. Third, the unavoidable differences in clinical characteristics among the three groups might affect the study results. In order to overcome this, we enrolled consecutive subjects who visited the same institution during the same period and corrected for important confounding variables through multivariable analysis. Fourth, due to the small number of heart failure patients enrolled in the study, it was difficult to conduct a detailed analysis according to the etiology of heart failure, such as ischemic vs. non-ischemic. Finally, because our study target is limited to Korean adults, it may be difficult to directly apply our findings to other ethnic groups. ## Conclusions Compared to the control subjects, arterial stiffness was increased in HF patients, but the association of the arterial stiffness on LV diastolic function was weaker in HF patients compared to the control subjects. These results suggest early detection and effective intervention for reverse arterial stiffening may limit adverse cardiac remodeling and HF. The degree of arterial stiffness was similar between HFrEF and HFpEF, but the association between arterial stiffness and LV diastolic function was stronger in the HFpEF group. Given that baPWV correlated well with septal e′ velocity in HFpEF, septal e′ velocity could be useful for devising a therapeutic strategy targeting VA coupling. Further large-scale studies are needed to confirm our findings. ## Supplementary Information Additional file 1: Supplementary Figure S1. The difference in baPWV between control and HF group. baPWV was significantly higher in HF patients than in control subjects. Supplementary Figure S2. Associations between baPWV and LV diastolic parameters in control and HF group. The associations of baPWV with septal e′ velocity and septal E/e′ were stronger in control subjects than in HF patients. Supplementary Table S1. Independent association between brachial-ankle pulse wave velocity and left ventricular diastolic parameters. ## References 1. Virani SS, Alonso A, Benjamin EJ, Bittencourt MS, Callaway CW, Carson AP. Heart disease and stroke statistics: 2020 update: a report from the American Heart Association. Circulation (2020) 141 e139-596. DOI: 10.1161/CIR.0000000000000757 2. Timmis A, Townsend N, Gale CP, Torbica A, Lettino M, Petersen SE. European Society of Cardiology: cardiovascular disease statistics 2019. Eur Heart J (2020) 41 12-85. DOI: 10.1093/eurheartj/ehz859 3. Garg R, Yusuf S. Overview of randomized trials of angiotensin-converting enzyme inhibitors on mortality and morbidity in patients with heart failure Collaborative Group on ACE Inhibitor Trials. JAMA (1995) 273 1450-6. DOI: 10.1001/jama.1995.03520420066040 4. Crozier I, Ikram H, Awan N, Cleland J, Stephen N, Dickstein K. Losartan in heart failure. Hemodynamic effects and tolerability Losartan Hemodynamic Study Group. Circulation (1995) 91 691-7. DOI: 10.1161/01.CIR.91.3.691 5. Mazayev VP, Fomina IG, Kazakov EN, Sulimov VA, Zvereva TV, Lyusov VA. Valsartan in heart failure patients previously untreated with an ACE inhibitor. Int J Cardiol (1998) 65 239-246. DOI: 10.1016/S0167-5273(98)00149-1 6. a randomised trial. Lancet (1999) 353 9-13. DOI: 10.1016/S0140-6736(98)11181-9 7. Packer M, Coats AJ, Fowler MB, Katus HA, Krum H, Mohacsi P. Effect of carvedilol on survival in severe chronic heart failure. N Engl J Med (2001) 344 1651-1658. DOI: 10.1056/NEJM200105313442201 8. Metoprolol CR/XL Randomised Intervention Trial in Congestive Heart Failure (MERIT-HF). Lancet (1999) 353 2001-2007. DOI: 10.1016/S0140-6736(99)04440-2 9. McMurray JJ, Packer M, Desai AS, Gong J, Lefkowitz MP, Rizkala AR. Angiotensin-neprilysin inhibition versus enalapril in heart failure. N Engl J Med (2014) 371 993-1004. DOI: 10.1056/NEJMoa1409077 10. WRITING COMMITTEE MEMBERS, Yancy CW, Jessup M, Bozkurt B, Butler J, Casey DE Jr,. ACCF/AHA guideline for the management of heart failure: a report of the American College of Cardiology Foundation/American Heart Association Task Force on practice guidelines. Circulation (2013) 2013 e240-327 11. Fonarow GC, Stough WG, Abraham WT, Albert NM, Gheorghiade M, Greenberg BH. Characteristics, treatments, and outcomes of patients with preserved systolic function hospitalized for heart failure: a report from the OPTIMIZE-HF Registry. J Am Coll Cardiol (2007) 50 768-777. DOI: 10.1016/j.jacc.2007.04.064 12. Kaess BM, Rong J, Larson MG, Hamburg NM, Vita JA, Cheng S. Relations of central hemodynamics and aortic stiffness with left ventricular structure and function: the Framingham Heart Study. J Am Heart Assoc (2016) 5 e002693. DOI: 10.1161/JAHA.115.002693 13. Cauwenberghs N, Knez J, Tikhonoff V, D'hooge J, Kloch-Badelek M, Thijs L. Doppler indexes of left ventricular systolic and diastolic function in relation to the arterial stiffness in a general population. J Hypertens (2016) 34 762-71. DOI: 10.1097/HJH.0000000000000854 14. Weber T. The role of arterial stiffness and central hemodynamics in heart failure. Int J Heart Fail (2020) 2 209-230. DOI: 10.36628/ijhf.2020.0029 15. Ponikowski P, Voors AA, Anker SD, Bueno H, Cleland JG, Coats AJ. ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: the Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC). Developed with the special contribution of the Heart Failure Association (HFA) of the ESC. Eur Heart J (2016) 37 2129-200. DOI: 10.1093/eurheartj/ehw128 16. Lang RM, Badano LP, Mor-Avi V, Afilalo J, Armstrong A, Ernande L. Recommendations for cardiac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. J Am Soc Echocardiogr (2015) 28 1-39. DOI: 10.1016/j.echo.2014.10.003 17. Nagueh SF, Smiseth OA, Appleton CP, Byrd BF, Dokainish H, Edvardsen T. Recommendations for the evaluation of left ventricular diastolic function by echocardiography: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. Eur Heart J Cardiovasc Imaging (2016) 17 1321-1360. DOI: 10.1093/ehjci/jew082 18. Kim HL, Seo JB, Chung WY, Kim SH, Kim MA, Zo JH. Association between invasively measured central aortic pressure and left ventricular diastolic function in patients undergoing coronary angiography. Am J Hypertens (2015) 28 393-400. DOI: 10.1093/ajh/hpu146 19. Kwak S, Kim HL, In M, Lim WH, Seo JB, Kim SH. Associations of brachial-ankle pulse wave velocity with left ventricular geometry and diastolic function in untreated hypertensive patients. Front Cardiovasc Med (2021) 8 647491. DOI: 10.3389/fcvm.2021.647491 20. Kim HL, Lim WH, Seo JB, Kim SH, Zo JH, Kim MA. Improved Prognostic value in predicting long-term cardiovascular events by a combination of high-sensitivity c-reactive protein and brachial-ankle pulse wave velocity. J Clin Med (2021) 10 3291. DOI: 10.3390/jcm10153291 21. Lee HS, Kim HL, Kim H, Hwang D, Choi HM, Oh SW. Incremental prognostic value of brachial-ankle pulse wave velocity to single-photon emission computed tomography in patients with suspected coronary artery disease. J Atheroscler Thromb (2015) 22 1040-1050. DOI: 10.5551/jat.29918 22. Kim HL, Weber T. Pulsatile hemodynamics and coronary artery disease. Korean Circ J (2021) 51 881-898. DOI: 10.4070/kcj.2021.0227 23. Kim HL, Kim SH. Pulse wave velocity in atherosclerosis. Front Cardiovasc Med (2019) 6 41. DOI: 10.3389/fcvm.2019.00041 24. Einarsen E, Gerdts E, Waje-Andreassen U, Naess H, Fromm A, Saeed S. Association of increased arterial stiffness with diastolic dysfunction in ischemic stroke patients: the Norwegian Stroke in the Young Study. J Hypertens (2020) 38 467-473. DOI: 10.1097/HJH.0000000000002297 25. Park KT, Kim HL, Oh S, Lim WH, Seo JB, Chung WY. Association between reduced arterial stiffness and preserved diastolic function of the left ventricle in middle-aged and elderly patients. J Clin Hypertens (Greenwich) (2017) 19 620-626. DOI: 10.1111/jch.12968 26. Abhayaratna WP, Barnes ME, O'Rourke MF, Gersh BJ, Seward JB, Miyasaka Y. Relation of arterial stiffness to left ventricular diastolic function and cardiovascular risk prediction in patients > or =65 years of age. Am J Cardiol (2006) 98 1387-1392. DOI: 10.1016/j.amjcard.2006.06.035 27. Coutinho T, Borlaug BA, Pellikka PA, Turner ST, Kullo IJ. Sex differences in arterial stiffness and ventricular-arterial interactions. J Am Coll Cardiol (2013) 61 96-103. DOI: 10.1016/j.jacc.2012.08.997 28. Noguchi S, Masugata H, Senda S, Ishikawa K, Nakaishi H, Tada A. Correlation of arterial stiffness to left ventricular function in patients with reduced ejection fraction. Tohoku J Exp Med (2011) 225 145-151. DOI: 10.1620/tjem.225.145 29. Shah AS, Gidding SS, El Ghormli L, Tryggestad JB, Nadeau KJ, Bacha F. Relationship between arterial stiffness and subsequent cardiac structure and function in young adults with youth-onset type 2 diabetes: results from the TODAY study. J Am Soc Echocardiogr (2022) 35 620-628. DOI: 10.1016/j.echo.2022.02.001 30. Kim HL, Lim WH, Seo JB, Chung WY, Kim SH, Kim MA. Association between arterial stiffness and left ventricular diastolic function in relation to gender and age. Medicine (Baltimore) (2017) 96 e5783. DOI: 10.1097/MD.0000000000005783 31. Zhang H, Hu W, Wang Y, Liu J, You L, Dong Q. The relationship between ambulatory arterial stiffness index and left ventricular diastolic dysfunction in HFpEF: a prospective observational study. BMC Cardiovasc Disord (2022) 22 246. DOI: 10.1186/s12872-022-02679-6 32. Chi C, Liu Y, Xu Y, Xu D. Association between arterial stiffness and heart failure with preserved ejection fraction. Front Cardiovasc Med (2021) 8 707162. DOI: 10.3389/fcvm.2021.707162 33. Ommen SR, Nishimura RA, Appleton CP, Miller FA, Oh JK, Redfield MM. Clinical utility of Doppler echocardiography and tissue Doppler imaging in the estimation of left ventricular filling pressures: a comparative simultaneous Doppler-catheterization study. Circulation (2000) 102 1788-1794. DOI: 10.1161/01.CIR.102.15.1788 34. Laurent S, Cockcroft J, Van Bortel L, Boutouyrie P, Giannattasio C, Hayoz D. Expert consensus document on arterial stiffness: methodological issues and clinical applications. Eur Heart J (2006) 27 2588-2605. DOI: 10.1093/eurheartj/ehl254
--- title: Sex differences in fetal intracranial volumes assessed by in utero MR imaging authors: - Paul D. Griffiths - Deborah Jarvis - Cara Mooney - Michael J. Campbell journal: Biology of Sex Differences year: 2023 pmcid: PMC10015831 doi: 10.1186/s13293-023-00497-9 license: CC BY 4.0 --- # Sex differences in fetal intracranial volumes assessed by in utero MR imaging ## Abstract ### Background The primary aim of the study is to test the null hypothesis that there are no statistically significant differences in intracranial volumes between male and female fetuses. Furthermore, we have studied the symmetry of the cerebral hemispheres in the cohort of low-risk fetuses. ### Methods 200 normal fetuses between 18 and 37 gestational weeks (gw) were included in the cohort and all had in utero MR, consisting of routine and 3D-volume imaging. The surfaces of the cerebral ventricles, brain and internal table of the skull were outlined manually and volume measurements were obtained of ventricles (VV), brain parenchyma (BPV), extraaxial CSF spaces (EAV) and the total intracranial volume (TICV). The changes in those values were studied over the gestational range, along with potential gender differences and asymmetries of the cerebral hemispheres. ### Results BPV and VV increased steadily from 18 to 37 gestational weeks, and as a result TICV also increased steadily over that period. TICV and BPV increased at a statistically significantly greater rate in male relative to female fetuses after 24gw. The greater VV in male fetuses was apparent earlier, but the rate of increase was similar for male and female fetuses. There was no difference between the genders in the left and right hemispherical volumes, and they remained symmetrical over the age range measured. ### Conclusions We have described the growth of the major intracranial compartments in fetuses between 18 and 37gw. We have shown a number of statistically different features between male and female fetuses, but we have not detected any asymmetry in volumes of the fetal cerebral hemispheres. ## Highlights In utero magnetic resonance imaging can be used to calculate intracranial volumes in fetuses. In this paper, we present volume data of the major intracranial compartments (cerebral ventricular system (VV), brain parenchymal volume (BPV), extraaxial CSF volume (EAV) and total intracranial volume (TICV) in normal male and female fetuses over a wide gestational age. We have shown that BPV and TICV increase at a statistically significant greater rate in male fetuses when compared with female fetuses. As a result, male fetuses have statistically significant larger BPV and TICV than female fetuses after 24 gestational weeks. There was no statistical evidence for asymmetry of hemispheric volumes in neither male nor female fetuses at any gestational age studied. ## Background In utero magnetic resonance imaging (iuMRI) of the fetal brain is now a widely accepted clinical tool when used as an adjunct to ante-natal ultrasonography because of proven advantages in terms of improved diagnostic accuracy and counselling [1, 2]. Advances in iuMRI technology and post-acquisition data processing now allows calculation of the volumes of the major intracranial compartments of both normal [3–5] and abnormal fetuses [6]. Specifically, it is possible to measure the volume of the cerebral ventricular system (VV), the brain parenchymal volume (BPV), the extraaxial CSF volume (EAV) and, by summation of those three, the total intracranial volume (TICV). Post-acquisition data processing using the same methodology can also be used to produce surface representations of those compartments. In this study, we have used iuMRI to measure the intracranial compartmental volumes of a large cohort of normal fetuses over a wide gestational range in the second and third trimesters. This allows us to study changes in intracranial volumes during the course of pregnancy and the primary aim of the study is to test the null hypothesis that there are no statistically significant differences in intracranial volumes between male and female fetuses. We also evaluate the alternative hypothesis that male fetuses have larger intracranial volumes and if so, when do the differences become apparent. Furthermore, we take the opportunity to study the symmetry of the cerebral hemispheres in the cohort, looking for possible asymmetry related to maturity and sex of the fetuses. ## Participants All the pregnant women whose fetuses are reported in this paper were recruited into the MERIDIAN study (magnetic resonance imaging to enhance the diagnosis of fetal developmental brain abnormalities in utero) [1, 2]. Specifically, they were part of an additional study to examine iuMRI scans of brains of fetuses considered to be normal on ultrasound [7]. Ethical approval was obtained from Yorkshire and the Humber/South Yorkshire ethics committee (11-YH-0006) and each woman provided fully informed, written consent. Fetuses are considered, a priori, to be normal because they were from a low-risk pregnancy, had no abnormalities on ante-natal ultrasonography (brain or somatic) and had normal brains on iuMR imaging. A total of 200 women with singleton pregnancies between 18 and 37 gestational weeks (gw) were scanned. As reported previously, two of the original 200 fetuses had unexpected brain abnormalities on iuMR [7] and for the purpose of this study, they were replaced by two further normal fetuses with gestational ages matched to the two fetuses with abnormalities. Some results of volumetric analyses on the cohort have been published previously in order to describe normative data by gestational age. However, that analysis did not investigate the effect of the sex of the fetus [5]. ## MRI data acquisition and processing The iuMRI protocol has been reported in full elsewhere [7], but is summarised here. All pregnant women were scanned on the same 1.5-T whole body scanner (Signa HDx, GE Healthcare, Milwaukee) at the University of Sheffield’s magnetic resonance facilities. Routine iuMRI of the fetal brain imaging consisted of T2-weighted single-shot fast spin echo sequences in the three natural orthogonal planes, and T1- and diffusion-weighted imaging both in the axial plane only. Those imaging studies were used to confirm normality of the brain following review by a pediatric neuroradiologist with over 18 years’ experience of fetal neuroimaging (PDG). In addition, volumetric brain imaging was acquired using a balanced steady-state imaging sequence (Fast Imaging Employing Steady-state Imaging—FIESTA, GE Healthcare, Milwaukee) in the axial plane. Those datasets were processed by a senior research MR radiographer with over 8 years’ experience of the technique (DJ) using ‘3D Slicer’ software (www.slicer.org). The surfaces of the cerebral ventricles, brain and internal table of the skull were outlined manually and absolute volume measurements were obtained by multiplying the number of voxels by the voxel size in each segmented compartment. This allowed the direct measurement of VV, BPV and EAV and TICV was derived by adding VV, BPV and EAV. We have previously described good intra- and inter-observer reproducibility of this technique [3, 4] (see discussion). The investigators did not know the sex of the fetus at the time the volume measurements were made as that information was collected post-natally, so the measurements were not biased by knowledge of the sex of the fetus. The final assessments were designed to look at the symmetry of the fetal cerebral hemispheres in terms of volume of the brain parenchyma following division of the BPV datasets into three components, brainstem/cerebellum and two cerebral hemispheres. The brain stem/cerebellum regions were divided from the supratentorial structures on sagittal imaging using an arbitrary, but consistent, construction line extending from the vein of Galen to the posterior clinoid processes. The cerebral hemispheres were then separated using imaging in the axial plane allowing the volumes for the two cerebral hemispheres to be assessed independently. The symmetry of the cerebral hemispheres was assessed with knowledge of which was the left or which was the right hemisphere. This was done by comparing the brain volume data (which only included the fetal head) with the routine 2D imaging in the coronal plane, on which the chest and abdomen are visible. The laterality of the hemispheres determined on the assumption of situs solitus by reference to the position of the thoracic and abdominal organs (left atrium, spleen and stomach on the left side and the liver is on the right side). ## Data analysis The data were analysed using Stata (Statacorp, College Station, Texas). The volumetric data were plotted against age to examine their relationship using lowess smoothing plots (bandwidth 0.6) to show the shape of the data. The relationship of VV, BPV and TICV with age appeared linear after 24 weeks and so linear regression models were fitted after this age to assess the rate of growth of the brain and to test whether these rates differ by sex of the fetus. Robust standard errors were used because the standard deviations were observed to increase as the mean volumes increase. In order to examine the whole age range, the data were grouped in quarters by gestational age to enable means by age and sex group to be calculated. The data were also log transformed for more detailed statistical analysis of asymmetry of the cerebral hemispheres. ## Results Table 1 shows the distribution of fetuses by age and sex. There were 109 males and 91 females. All age groups from 18 to 37 weeks were represented with a modal age for both males and females of 29 weeks. Representative images of the compartmental volumes at three gestational ages are shown in Figs. 1 and 2 with the lowess smoothing plots for VV, BPV, EAV and TICV plotted against gestational age are shown in Fig. 3. There are different relationships between the volume of the four intracranial compartments and gestational age. The increase in VV is small and close to linear (increasing at approximately 0.3 cm3/week) over the study period. The increase in EAV is modest between 23 and 32gw (approximately 10 cm3/week) before levelling off after 32gw. The relationship between both BPV and TICV with gestational age is linear after 24gw when the growth rates were approximately 20 cm3/week for BPV and 25 cm3/week for TICV. Accordingly, the contribution of BPV to the TICV changed substantially over the study period, e.g. BPV accounted for approximately $50\%$ of TICV before 22gw and approximately $65\%$ after 34gw. Table 1Distribution of fetuses in the study by age and sexGestational age (wks)n Femalen MaleTotal18–2322335524–2726224828–3028294631–37152551Total91109200Fig. 1Images of the intracranial compartments studied in this paper in a male fetuses at 22gw. The images are surface representations of the cerebral ventricles (a lateral projection, b frontal projection and c superior projection. The same order is used for the surface representations of the brain parenchyma (d–f) and the extraaxial CSF spaces (g–i)Fig. 2The same format as Fig. 1 for a 34gw male fetusFig. 3Lowess plots of the four compartmental voulmes studied plotted against gestational age. VV ventricular volume, EAV extraaxial volume, BPV brain parnchymal volume, TICV total intracranial volume ## Analysis of compartmental volumes in relation to sex of the fetus Figure 4 shows the data plotted by gender in four separate plots with lowess smoothing plots included. Table 2 shows the regression slopes against age and the interaction with sex for the 145 fetuses ≥ 24gw. A number of statistically significant differences between the male and female fetuses are shown. One of the main differences in terms of the slopes between male and female fetuses is for BPV with a greater rate of growth for males of 3.15 cm3/week ($95\%$ CI 1.27 to 5.02). Because BPV is the major contributor to TICV in more mature fetuses, there is a similar pattern of difference in slopes for TICV, which is also statistically significant. There was no statistically significant differences in growth rates between males and females for VV and EAV. The absolute volumes of male and female fetuses (as opposed to growth rates) were studied after the data grouped into quarters by gestational age as shown in Table 3. There is little difference in BPV and TICV between the sexes until 24 weeks, after which male fetuses have statistically significant larger BPV and TICV. For example, after 31gw male fetuses have approximately $7\%$ larger BPV and $8\%$ larger TICV when compared with female fetuses. In contrast, male fetuses have larger VV when compared with females through all of the gestational range studied with the largest difference occurring in the 24–27gw range (approximately $24\%$ larger). The relationship between EAV and gender was not as straightforward, with males having statistically significant larger EAV between 24 and 27gw only. Fig. 4Raw data and smoothing plots for ventricular volume (VV—3a), extraaxial volume (EAV—3b), brain parnchymal volume (BPV—3c) and total intracranial volume (TICV—3d) by gender (males blue, females red)Table 2Regression slopes in relation to gestational age and the interaction with sex for 145 fetuses imaged ≥ 24gwFemaleMaleDifference in slopesP value for differenceaSlopeSESlopeSEVV0.370.080.390.080.020.84EAV7.700.768.120.820.420.71BPV15.70.5818.90.753.20.001TICV23.81.1027.41123.60.024VV ventricular volume, EAV extraaxial volume, BPV brain parnchymal volume, TICV total intracranial volumeSlopes statistically significant ($P \leq 0.001$)aUsing standard errors robust to variance heterogeneityTable 3Absolute volumes of the four intracranial compartments in male and female fetuses after the data were grouped into quarters by gestational ageVariable (cm3)Gestational age (wks)FemaleMaleMale–femalePMeanSDMeanSDMeanVV18–233.540.764.141.400.600.07024–274.611.475.881.871.270.01128–306.162.016.822.000.660.2731–377.702.368.912.751.210.12EAV18–2331.88.4631.28.53− 0.60.7924–2760.214.573.617.213.40.00528–3094.512.5102.417.87.90.08831–37128.723.5138.824.010.10.15BPV18–2344.114.344.113.20.070.9824–2791.619.3104.020.012.40.03428–30140.415.5150.222.29.80.08731–37236.136.4256.238.118.10.12TICV18–2379.422.579.520.70.10.9924–27156.432.4183.537.027.10.01028–30241.023.1259.435.018.30.04031–37360.754.8390.852.230.10.056VV ventricular volume, EAV extraaxial volume, BPV brain parnchymal volume, TICV total intracranial volume ## Symmetry of the cerebral hemispheres The difference between left and right hemispheres is shown in Fig. 5 with the regression slopes (the log of the data is shown as this stabilises the variance). The mean maximum asymmetry between the hemispheres is reasonably constant across the gestational range studied (approximately $2\%$). The largest asymmetry between the cerebral hemispheres in a fetus was $8.8\%$ and there were only three fetuses with asymmetries > $6\%$. Accordingly, there is no statistical evidence that the volumes of the fetal cerebral hemispheres differ in size (mean − 0.0019, $95\%$ CI − 0.0053 to 0.0015) $$p \leq 0.27$$). There is no statistical evidence to suspect the differences in hemispheric volumes is affected by gender (difference boys minus girls = 0.0038978, $95\%$ CI − 0.0031 to 0.011, $$p \leq 0.28$$) or gestational age ($r = 0.056$, $$p \leq 0.332$$). In addition, the standard deviations of the differences of the logged data were very similar (0.024 for female fetuses and 0.025 for male fetuses).Fig. 5Difference in log hemispheric brain volumes (right–left) by gender (blue male, red female) ## Discussion In this paper, we have studied volumes of the intracranial compartments of normal fetuses between 18 and 37gw, a period when there is considerable growth and maturation of the fetal brain. Intracranial compartmental anatomy is a complex subject and our approach in this study has involved considerable simplification because of the limitations of iuMRI in its present form in terms of anatomical and contrast resolution. However, the heavily T2-weighted images produced by the 3D acquisition used in this study allow good delineation of the major intracranial compartments because of the favourable contrast differences between the brain, skull and CSF. VV measured in this study also includes the non-ventricular fluid-containing structures (primarily the cavum septum pellucidum and cavum vergae) and the CSF-producing structures (choroid plexi). As well as CSF the EAV contains the majority of the intracranial vascular compartment (large/medium arteries, large cerebral veins and venous sinuses) as they cannot be delineated from the CSF using current imaging methods. It is important to restate that three compartmental volumes were measured directly in this study (VV, BPV and EAV), whereas TICV was derived from summation of the three compartmental volumes. The technique we have used relies on manual segmentation of the three intracranial compartments and we have previously shown high reproducibility of the method [3, 4], as shown by a mean inter-observer difference of $1.27\%$ (standard deviation ± $4.8\%$, full range 0.05 to $9.31\%$). The ‘compartmental’ approach in assessing the intracranial contents of the fetus we describe in this paper is not used in routine clinical practice at present using either ultrasonography or MR imaging, primarily because of the technical challenges in obtaining and processing the data. Instead, linear measurements of the fetal skull are taken by way of head circumference, bi-parietal diameter and/or occipito-frontal diameter. An assumption is made that, in some way, increasing skull size reflects brain growth. This may be true in many cases, but there are relatively common fetal neuropathologies that will interfere with that relationship as discussed later in this section. With the appropriate technology in place, we aimed to study a large cohort of 200 normal fetuses between 18 and 37gw and measure the intracranial compartmental volumes in order to comment on the normal pattern of growth, sex difference and asymmetry of the cerebral hemispheres. We have shown that there is a close association between brain growth (as indicated by TBV) and head/calvarial growth (as indicated by TICV), particularly after 24gw when the growth trajectories for both compartments are linear and virtually parallel. The close association between BPV and TICV is supportive of the thesis that growth of the brain stimulates mesenchymal development and growth of the skull particularly the bones that develop by intramembranous ossification. The BPV/TICV ratio increases over the gestational range studied because the increase in the volume of the CSF spaces (VV and EAV) were modest in comparison to BPV and TICV. In particular, VV only shows a minor increase in volume whereas the growth of the EAV is modest and plateaus after 32gw. We also had the opportunity to evaluate potential difference in compartmental volumes in relation to the sex of the fetus. Again, BPV and ICV show very similar changes over the time period studied in both sexes but there are differences between male and female fetuses. Although BPV and TICV are very similar in male and female fetuses between 18 and 23gw, male fetuses subsequently grow at rates that are statistically significantly larger (by 7–$8\%$) when compared with female fetuses. As already stated, clinical assessment of skull size uses linear rather than volume measurements and it should be appreciated that an $8\%$ difference in TICV is commensurate with a $2.6\%$ difference in linear dimensions only. Ultrasound-derived biometric charts of normal fetuses used in clinical practice do not distinguish between male and female fetuses (e.g. intergrowth 21.tghn.org/fetal-growth (based on Papageorghiou et al. [ 8]), but biometric charts of babies born prematurely do (e.g. intergrowth21.ndog.ox.ac.uk). Analysis of those charts shows small differences between male and female babies although males are consistently larger, e.g. head circumference median (standard deviation) at 32gw males 29.4 cm (± 1.6 cm), females 29.1 cm (± 1.6 cm); at 36gw males 32.5 cm (± 1.3 cm), females 32.1 cm (± 1.2 cm). It is also widely accepted that boys are slightly larger than girls following delivery at Term, hence the use of gender specific normative charts post-natally. An important recent paper by Galjaard et al. has studied head size in relation to the sex of the fetus [9]. Those authors reported nearly 28,000 fetuses from a low-risk Caucasian population and they showed that male fetuses have larger bi-parietal diameters and head circumferences when compared with female fetuses from 20gw. As predicted by the results of our study, the differences were small (amounting to a 3-day difference between 20−24gw) but were statistically significant at the $p \leq 0.001$ level. Another interesting and potentially clinically relevant finding is found in the gender difference in VV, which were larger in male fetuses (ranging from 11 to $17\%$ bigger through the gestational range). In a comparable fashion to the TICV discussion, those differences equate to only a $5\%$ difference in linear measurements of the ventricles, such as the width of the trigone of the lateral ventricles that is used in routine assessments of the fetus. A fetus is considered to have ventriculomegaly if the trigone measurement is > 10 mm at any stage of pregnancy on the basis that 10 mm is approximately 4sd above the mean value [10]. Between 1 and $\frac{3}{1000}$ of unselected fetuses have trigone measurements ≥ 10 mm, described as ventriculomegaly in clinical practice. The results of the present study show two important caveats when using the accepted current dogma about trigone size. First, VV normally increases with gestational age and although the absolute changes are small there is a more than doubling in VV between 20 and 36gw. It may not be appropriate, therefore to have the same 10-mm cut-off for all gestational ages. Secondly, the statistically significant larger VV in normal male fetuses is likely to explain the consistent observation of an excess of male fetuses diagnosed with isolated ventriculomegaly, which ranges from 1.3:1 to 3.5:1 [11–14]. We also investigated the symmetry of the fetal cerebral hemispheres. There are many descriptions of asymmetry of the mature human cerebral hemispheres and those are generally considered to be important because of functional localisation of the cerebral hemispheres and cerebral ‘dominance’, particularly for language. In children and adults, it is estimated that approximately $90\%$ of right-handed people are left hemisphere dominant on the basis of language localisation and slightly fewer left-handed people are left hemisphere dominant. However, in spite of marked ‘functional asymmetries’ the mature cerebral hemispheres do not show major structural asymmetry. In fact, the only relatively consistent asymmetries in the post-natal human cerebral hemispheres are in the superior temporal lobes and ‘Yakovlevian anti-clockwise torque [15]’. The planum temporale, on the posterior aspect of the superior temporal lobe is involved in language processing and is usually larger in right-handed people. Yakovlevian anti-clockwise torque refers to an apparent twisting of the cerebral hemispheres (anti-clockwise if looking at the brain from above—see Le May 1976 [16]) so that right frontal lobe extends across the midline to the left and the left occipital lobe extends across the midline posteriorly. We have not studied whether this feature is present in fetuses formally in this paper but our initial impression that it is not. In spite of that arrangement, the human mature cerebral hemispheres are not considered to be different in terms of weight or volume [15]. Similarly, there was no evidence of asymmetry in volume of the human fetal cerebral hemispheres in our present study or, more accurately, no asymmetry could be detected in relation to the precision of technique (with an inter-observer error of $1.27\%$). This is true for the entire cohort and when the cohort was divided in terms of gender. We have not studied differences in sulcation/gyration patterns in our cohort, but a previous iuMR study showed more advanced sulcation in the left temporal lobe when compared with the right [17]. It is important to consider the rationale for obtaining normal values for the volumes of the intracranial compartments in second and third trimester fetuses. Some insight can be gained by considering the expected changes in compartmental volumes in some of the commoner causes of fetal ventriculomegaly. For example, ventriculomegaly is the commonest abnormal intracranial finding on ante-natal ultrasonography and hence the commonest referral for iuMR neuroimaging. In most cases fetal ventriculomegaly is an isolated finding, and such fetuses have very good prospects of a normal neurodevelopmental outcome. In those fetuses the expected pattern of intracranial volumes change would be—increased VV but BPV, EAV and TICV within normal ranges. Alternatively, if fetal ventriculomegaly is secondary to non-communicating hydrocephalus (due to a blockage in CSF flow in the ventricles) the VV will be increased (often massively) and the associated raised intraventricular pressure will cause effacement of the extraaxial CSF spaces (hence reduced EAV). The bones of the calvarium are not fused in the fetus, so the raised intraventricular/intracranial pressure will also cause increased TICV, whilst BPV will often be normal or slight reduced. In contrast, if ventriculomegaly is due to a destructive process of the brain (such as the result of trans-placental viral infection or in utero hypoxic ischaemic injury) VV and EAV are increased by an ex vacuo mechanism secondary to reduced BPV. We have confirmed that brain growth is a major driver for head growth in this paper, therefore reduced TICV should be expected in fetuses with destructive brain pathology. Hence, knowledge of the compartmental volumes can aid diagnosis of fetal neuropathology as well as understanding normal development. There are several limitations in the current work, first in relation to the cohort size. The size of the sample reported in this paper was determined from the original purpose of the study, namely to determine if fetuses which are normal on ultrasound had any abnormalities on iuMRI. A sample size of 200, in which no abnormalities were observed would mean that the upper $95\%$ limit of the likely rate of abnormalities in fetuses with a normal ultrasound is $1.5\%$. As such, this activity reported here was not formally powered to show differences in sex. Also, it was not possible to control the content of the cohort in order to have, for example, the same number of cases at each gestational age or to obtain equal numbers of male and female fetuses. We also recognise the potential weaknesses in the technical performance of our methodology in producing the volume measurements. Although we have shown good intra- and inter-observer reproducibility, we cannot compare our measurements with the ‘real’ volumes of the fetal brains, hence we cannot report the accuracy of the technique. In addition, we have described the slopes of our measured volumes in relation to gestational age, which we have implied indicates growth rates of those compartments. Information about fetal growth can only be done formally if repeat measurements are made in the same individual and we stress that there was no capacity to do a longitudinal assessment in our study. ## Perspectives and significance We have shown that BPV and VV increased steadily over the gestational age 18 to 37 weeks, and as a result TICV also increased steadily over that period. TICV and BPV increased at a statistically significantly greater rate in male relative to female fetuses after 24gw. The greater VV in male fetuses was apparent earlier, but the rate of increase was similar for male and female fetuses. For EAV the mean volumes for males and females appeared to diverge and then converge, resulting in similar linear slopes after 24 weeks. There was no difference between the genders in the left and right hemispherical volumes, and they remained symmetrical over the age range measured. ## References 1. Griffiths PD, Bradburn M, Campbell MJ. Use of MRI in the diagnosis of fetal brain abnormalities in utero (MERIDIAN): a multicentre, prospective cohort study. Lancet* (2017) 389 538-546. DOI: 10.1016/S0140-6736(16)31723-8 2. Griffiths PD, Bradburn M, Campbell M. MRI in the diagnosis of fetal developmental brain abnormalities: the MERIDIAN diagnostic accuracy study. Health Technol Assess (2019) 23 1-144. DOI: 10.3310/hta23490 3. Paddock M, Akram R, Jarvis DA. The assessment of fetal brain growth in diabetic pregnancy using in utero magnetic resonance imaging. Clin Radiol. (2017) 72 427e1-8. DOI: 10.1016/j.crad.2016.12.004 4. Jarvis D, Akram R, Paddock M. Quantification of total fetal brain volume using 3D MR imaging data acquired in utero. Prenat Diagn (2016) 36 1225-1232. DOI: 10.1002/pd.4961 5. Jarvis D, Finney CR, Griffiths PD. Normative volume measurements of the fetal intracranial compartments using 3D volume in utero MR imaging and potential clinical applications. Eur Radiol (2019) 29 3488-3495. DOI: 10.1007/s00330-018-5938-5 6. Jarvis D, Griffiths PD. Clinical applications of 3D volume MR imaging of the fetal brain in utero. Prenat Diagn (2017) 37 556-565. DOI: 10.1002/pd.5042 7. Griffiths PD, Bradburn M, Mandefield L. The rate of brain abnormalities on in utero MR studies in fetuses with normal ultrasound examinations of the brain and calculation of indicators of diagnostic performance. Clin Radiol (2019) 74 527-533. DOI: 10.1016/j.crad.2019.03.010 8. Papageorghiou AT, Ohuma EO, Altman DG. International standards for fetal growth based on serial ultrasound measurements: the Fetal Growth Longitudinal Study of the INTERGROWTH-21st Project. Lancet (2014) 384 869-879. DOI: 10.1016/S0140-6736(14)61490-2 9. Galjaard S, Ameye L, Lees CC. Sex differences in fetal growth and immediate birth outcomes in a low-risk Caucasian population. Biol Sex Differ (2019) 10 48. DOI: 10.1186/s13293-019-0261-7 10. Cardoza JD, Goldstein RB, Filly RA. Exclusion of fetal ventriculomegaly with a single measurement: the width of the lateral ventricular atrium. Radiology (1988) 169 711-714. DOI: 10.1148/radiology.169.3.3055034 11. Patel MD, Filly AL, Hersh DR, Goldstein RB. Isolated mild fetal cerebral ventriculomegaly: clinical course and outcome. Radiology (1994) 192 759-764. DOI: 10.1148/radiology.192.3.7520183 12. Gilmore JH, van Tol J, Kliewer MA. Mild ventriculomegaly detected in utero with ultrasound: clinical associations and implications for schizophrenia. Schizophr Res (1998) 33 133-140. DOI: 10.1016/S0920-9964(98)00073-5 13. Ouahba J, Luton D, Vuillard E. Prenatal isolated mild ventriculomegaly: outcome in 167 cases. Br J Obstet Gynecol (2006) 113 1072-1079. DOI: 10.1111/j.1471-0528.2006.01050.x 14. Griffiths PD, Jarvis D, Connolly DJA. Predicting neurodevelopmental outcomes in fetuses with isolated mild ventriculomegaly. Arch Dis Child Fetal Neonatal Ed (2022) 107 431-436. DOI: 10.1136/archdischild-2021-321984 15. Toga AW, Thompson PM. Mapping brain asymmetry. Nat Rev Neurosci (2003) 4 37-48. DOI: 10.1038/nrn1009 16. Le May M. Morphological cerebral asymmetries of modern man, fossil man and non-human primate. Ann NY Acad Sci (1976) 280 471-476 17. Kasprian G, Langs G, Brugger PC. The prenatal origin of hemispheric asymmetry: an in utero neuroimaging study. Cereb Cortex (2011) 21 1076-1083. DOI: 10.1093/cercor/bhq179
--- title: Transcriptomics and metabolomics analysis reveal the anti-oxidation and immune boosting effects of mulberry leaves in growing mutton sheep authors: - Xiaopeng Cui - Yuxin Yang - Minjuan Zhang - Shuang Liu - Hexin Wang - Feng Jiao - Lijun Bao - Ziwei Lin - Xinlan Wei - Wei Qian - Xiang Shi - Chao Su - Yonghua Qian journal: Frontiers in Immunology year: 2023 pmcid: PMC10015891 doi: 10.3389/fimmu.2022.1088850 license: CC BY 4.0 --- # Transcriptomics and metabolomics analysis reveal the anti-oxidation and immune boosting effects of mulberry leaves in growing mutton sheep ## Abstract ### Introduction Currently, the anti-oxidation of active ingredients in mulberry leaves (MLs) and their forage utilization is receiving increasing attention. Here, we propose that MLs supplementation improves oxidative resistance and immunity. ### Methods We conducted a trial including three groups of growing mutton sheep, each receiving fermented mulberry leaves (FMLs) feeding, dried mulberry leaves (DMLs) feeding or normal control feeding without MLs. ### Results Transcriptomic and metabolomic analyses revealed that promoting anti-oxidation and enhancing disease resistance of MLs is attributed to improved tryptophan metabolic pathways and reduced peroxidation of polyunsaturated fatty acids (PUFAs). Furthermore, immunity was markedly increased after FMLs treatment by regulating glycolysis and mannose-6-phosphate pathways. Additionally, there was better average daily gain in the MLs treatment groups. ### Conclusion These findings provide new insights for understanding the beneficial effects of MLs in animal husbandry and provide a theoretical support for extensive application of MLs in improving nutrition and health care values. ## Introduction Currently, a number of medicinal plants are widely used as functional foods and alternative medicine to prevent and treat chronic diseases [1]. This is due to the numerous bioactive components with anti-oxidant capabilities, such as phenolic compounds and flavonoids [2], which may help improve immunity by coordinating the metabolism of the body. Mulberry leaves (MLs) have been used as feed for silk worms for hundreds of years and also as a traditional Chinese medicine, according to the classical medicine books. Owing to their anti-oxidative, anti-inflammatory, anti-bacterial, and anti-hyperlipidemic properties, mulberry leaves are gaining increasing attention for use in Chinese herbal medicines [3]. To date, the hypoglycemic and lipid-lowering effects of extracts or active ingredients in MLs are well established, against diabetes, fatty liver, and some similar diseases related to disorders in glucose and lipid metabolism. In addition, accumulating evidence have validated the promotion of growth and rumen development, anti-oxidant properties, and improvement in milk production by MLs or their active ingredients in livestock (3–6). However, the underlying mechanism of MLs, as an unconventional feed with both nutritive and medicinal properties, on anti-oxidation and immunity in livestock remains poorly understood. In recent years, an explosion has occurred in the acquisition of biological data through the use of so-called ‘omics’ techniques. Whilst many different omics technologies are now featured in the literature, the most frequently used omics are genomics, transcriptomics, proteomics and metabolomics [7]. Thus, the aim of the present study was to evaluate the roles of MLs in growth promoting and animal welfare improving aspects of mutton sheep in terms of antioxidant and immune properties and explore the mechanism by methods of transcriptomic and metabolomic. Our study provides novel insights into the role of MLs in livestock yield and the application of natural functional fodder. ## Materials and methods The experiment was conducted in accordance with the Chinese Guidelines for Animal Welfare and Experimental Protocols, and approved by the Animal Care and Use Committee of the Institute of Northwest A & F University. ## Preparation and chemical indexes measurement of fermented mulberry leaves and dried mulberry leaves MLs (species 707) are harvested in July 2021 at the Institute of Sericulture and Silk in Zhouzhi, Shaanxi Province, China. One half is sun-dried for seven days, next well-sealed in woven bags after a little rubbing and then stored in a dry, dark place for acquiring DMLs for use in feeding experiment. The other half with $65.03\%$ moisture content after wilted by sun-shine for a half day, is a little smashed and vacuum sealed in fermentation-special bags to ferment with $5\%$ Lactobacillus plantarum inoculation at room temperature (27.5-28°C) for thirty days in a dry, dark place for preparation for FMLs. Here $5\%$ is adding 5 mL of bacterial culture suspension to 100 grams of MLs and the concentration of bacterial culture suspension is 1×108 CFU/mL. Lactobacillus plantarum (CICC 23941) purchased from the China Center of Industrial Culture Collection (www.china-cicc.org). Before feeding experiments, the pH, crude protein, crude fiber as well as gross energy of FMLs and DMLs are determined according to standard methods of AOAC. And their contents are shown in Table 1. FMLs are deemed qualified without aflatoxin B1 detected at a minimum checked value of 0.1 μg/kg by Huayan Testing Group Co., Ltd in Xi’an City, Shaanxi Province (Detection number: SP202115450). Table 1 \| Items \| DM loss/% \| pH \| Crude protein/% \| Crude fiber/% \| Gross energy/(MJ/kg) \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| FMLs \| 4.6 \| 3.98 \| 14.69 \| 8.35 \| 15.92 \| \| DMLs \| \| 6.2 \| 15.15 \| 9.72 \| 14.28 \| ## Experimental design and feeding diets Animal experiments are conducted on six-month-old healthy female mutton sheep (white-headed Suffolk sheep♂×Hu sheep♀) weighing 30.41kg at average without genetic modification in Gansu Qinghuan Meat Sheep Seed Production Co. Ltd (Huan County, Qingyang City, Gansu Province, China). The animals were randomly assigned to group Con feeding a normal control diet ($$n = 18$$), group TR1 feeding an experimental diet with FMLs ($$n = 18$$) and group TR2 feeding an experimental diet with DMLs ($$n = 18$$) and then treated for an experiment of fifty days. Each group had 6 replicates with 3 sheep per replicate. Before the feeding experiment, animals undergo an acclimatization period of six days to obtain an appropriate feed intake, during which they were allowed unlimited access to their corresponding experimental diet and tap water. Experimental sheep were housed in sheepfold and given self-help feeding in three groups every day. The ingredients and chemical composition of three experimental diets are shown in Table 2. The chemical compositions of three experimental diets are determined by Ulanqab Yima Agriculture and Animal Husbandry Technology Co., Ltd (Ulanqab City, Inner Mongolia, China). Table 2 \| Items \| Con \| TR1 \| TR2 \| \| --- \| --- \| --- \| --- \| \| Ingredients \| Ingredients \| Ingredients \| Ingredients \| \| DMLs/% \| 0 \| 0 \| 7.11 \| \| FMLs/% \| 0 \| 16.59 \| 0 \| \| Oat hay/% \| 12.43 \| 24.88 \| 17.77 \| \| Corn silage/% \| 29.00 \| 8.29 \| 24.88 \| \| Corn/% \| 19.34 \| 16.59 \| 16.59 \| \| Wheat/% \| 20.72 \| 17.77 \| 17.77 \| \| Concentrate/% \| 17.68 \| 15.17 \| 15.17 \| \| Limestone/% \| 0.83 \| 0.71 \| 0.71 \| \| Total/% \| 100 \| 100 \| 100 \| \| Nutrients (based on dry matter) \| Nutrients (based on dry matter) \| Nutrients (based on dry matter) \| Nutrients (based on dry matter) \| \| Dry matter/% \| 72.50 \| 66.10 \| 66.40 \| \| Crude protein/% \| 16.40 \| 16.70 \| 16.60 \| \| Metabolizable Energy/(MJ/kg) \| 10.46 \| 10.63 \| 10.30 \| \| Crude fat/% \| 3.00 \| 3.40 \| 3.00 \| \| Crude ash/% \| 9.58 \| 9.24 \| 10.59 \| \| Acid detergent fiber/% \| 16.80 \| 17.50 \| 15.30 \| \| Neutral detergent fiber/% \| 27.80 \| 28.47 \| 25.76 \| \| Lignin/% \| 4.30 \| 4.40 \| 4.80 \| ## Weighing and sample collection Prior to the experiments, all sheep are driven to be weighed by an automatic weighing system to obtain the initial body weights. Afterwards, body weights on day 25th and 50th are weighted to calculate daily gains. On the 50th day, blood from the jugular vein was collected and placed in 5mL vacuum negative-pressure tubes with yellow cap containing separation gels for serum separation and then leave to set for one to two hours to collect rough 2.5mL serum, which is immediately stored in liquid nitrogen and taken to the lab for further analysis. At the end of the experimental period, 6 sheep per group which were representative in terms of average weight (inclusion criteria) of group were selected and slaughtered for tissue sample collection. Tissue samples (about 0.5×0.5×0.5cm3) of longissimus dorsi muscle and subcutaneous fat from the left side of the carcass are packed into 2 mL cryopreserved tube and frozen in liquid nitrogen immediately within 20 min of slaughter for biochemical indexes and omics analysis. Weighting samples contain 18 biological repeats of each group. ## Analysis of biochemical indexes Growth hormone (GH), total antioxidant capacity (TAOC), superoxide dismutase (SOD), catalase (CAT), glutathione peroxidase (GSH-Px) and malondialdehyde (MDA) are determined using commercially available kits (HY-60021, HY-60001, HY-M0018, HY-60005, HY-60003; Beijing Huaying Biotechnology Research Institute, Beijing, China). Serum immunoglobulin A, M, G (IgA, IgM, IgG) and tumor necrosis factor (TNFα) were determined by commercially available kits (HY-N0048, HY-N0049, HY-N0050, HY-H0019; Beijing Huaying Institute of Biotechnology Research Institute, Beijing, China). Immunoglobulin (IG) is the sum of immunoglobulins IgA, IgM and IgG. Muscle tissue samples contain 5 biological repeats (group Con), 6 biological repeats (group TR2) and 5 biological repeats (group TR2), respectively. Adipose tissue samples contain 5 biological repeats (group Con), 5 biological repeats (group TR2) and 5 biological repeats (group TR2), respectively. Serum samples contain 8 biological repeats (group Con), 7 biological repeats (group TR2) and 8 biological repeats (group TR2), respectively. Technical repetition is no less than 2 for all samples. No data point from the analysis is excluded. ## Widely target metabolomics analysis Tissue samples of muscle are extracted by Metware according to standard procedures. The sample extracts were analyzed using an LC-ESI-MS/MS system (UPLC, ExionLC AD, https://sciex.com.cn/; MS, QTRAP® System, https://sciex.com/). LIT and triple quadrupole (QQQ) scans were acquired on a triple quadrupole-linear ion trap mass spectrometer (QTRAP), QTRAP® LC-MS/MS System, equipped with an ESI Turbo Ion-Spray interface, operating in positive and negative ion mode and controlled by Analyst 1.6.3 software (Sciex). Instrument tuning and mass calibration were performed with 10 and 100 μmol/L polypropylene glycol solutions in QQQ and LIT modes, respectively. A specific set of MRM transitions were monitored for each period according to the metabolites eluted within this period. Significantly regulated metabolites between groups were determined by variable importance in projection (VIP)≥1 and absolute Log2FC (fold change)≥1. VIP values were extracted from OPLS-DA result, which also contain score plots and permutation plots, was generated using R package MetaboAnalystR. The data was log transform (log2) and mean centering before OPLS-DA. In order to avoid overfitting, a permutation test (200 permutations) was performed. ## Transcriptomic analysis RNA-extract and RNA-seq of muscle are conducted according to standard procedures of Majorbio with the Illumina HiSeq xten/NovaSeq 6000 sequencer (2×150bp read length). The raw paired end reads were trimmed and quality controlled by SeqPrep (https://github.com/jstjohn/SeqPrep) and Sickle (https://github.com/najoshi/sickle) with default parameters. Then clean reads were separately aligned to reference genome with orientation mode using HISAT2 (http://ccb.jhu.edu/software/hisat2/index.shtml) software [8]. The mapped reads of each sample were assembled by StringTie (https://ccb.jhu.edu/software/stringtie/index.shtml?t=example) in a reference-based approach [9]. To identify DEGs (differential expression genes) between two different samples, the expression level of each transcript was calculated according to the transcripts per million reads (TPM) method. RSEM (http://deweylab.biostat.wisc.edu/rsem/) [10] was used to quantify gene abundances. Essentially, differential expression analysis was performed using the DESeq2 [11]/DEGseq [12]/EdgeR [13] with Q value ≤ 0.05, DEGs with \|log2FC\|>1 and Q value ≤ 0.05(DESeq2 or EdgeR)/Q value ≤ 0.001(DEGseq) were considered to be significantly different expressed genes. The transcriptomic sequence data have been deposited in the NCBI database (Accession No. PRJNA898816). ## Statistical analysis Statistical analysis was performed by the SPSS 19.0 software (IBM-SPSS Statistics, IBM Corp., Armonk, NY, United States). Data were evaluated using a one-way ANOVA followed by Turkey’s multiple range tests for physiological and biochemical indexes. Significance was declared if $p \leq 0.05.$ Additionally, omics sequencing data are analyzed using online platforms for data analysis, including Metware cloud tools (https://cloud.metware.cn/#/tools/tool-list) and Majorbio cloud platform (https://cloud.majorbio.com/). Histograms and metabolic pathway maps are drawn respectively using Graphpad Prism 8 and Adobe Illustrator CS6. ## Growth performance Throughout the trial, no significant differences were detected in daily gain (0–25d) (Con<TR2<TR1) and feed to gain ratio (F/G) (Con>TR2>TR1) ($p \leq 0.05$, Table 3), although group TR1 which were fed with FMLs demonstrated a little increase in daily gain (0–25d) and a slight decrease in F/G. Apparently, treatments with FMLs and DMLs (group TR2) generated an obvious increase in ADFI during the overall raising period ($p \leq 0.05$), which suggests MLs are a delicious feed for promotion. In addition, ADG, daily gain (25–50d) and serum growth hormone levels were significantly improved in both MLs-treatment groups ($p \leq 0.05$) in the study. Further, FMLs feeding resulted in a significant increase in the final body weight ($p \leq 0.05$). Table 3 \| Items \| Con \| TR1 \| TR2 \| SEM \| P-value \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Initial BW/kg \| 30.36 \| 30.36 \| 30.50 \| 0.589 \| 0.994 \| \| Final BW/kg \| 38.03b \| 42.05a \| 40.86ab \| 0.696 \| 0.05 \| \| ADFI/kg \| 1.79b \| 2.31a \| 2.35a \| 0.033 \| 0.0 \| \| Daily gain (0-25d)/g \| 95.56 \| 125.56 \| 97.64 \| 6.999 \| 0.145 \| \| Daily gain (25-50d)/g \| 211.11b \| 342.22a \| 317.64a \| 16.33 \| 0.001 \| \| ADG (0-50d)/g \| 156.11b \| 233.89a \| 205.56a \| 7.876 \| 0.001 \| \| F/G \| 13.38 \| 10.26 \| 11.79 \| 1.026 \| 0.483 \| \| GH \| 4.81c \| 5.84b \| 7.28a \| 0.265 \| 0.0 \| ## Anti-oxidant properties As shown in Table 4, SOD (superoxide dismutase), CAT (catalase), GSH-Px (glutathione peroxidase) and TAOC (total antioxidant activity) in serum and muscle were significantly increased in the MLs treatment group, especially in the FMLs treatment group ($p \leq 0.05$); SOD and GSH-Px in adipose tissue also increased significantly ($p \leq 0.05$), CAT and TAOC tended to increase (Con<TR2<TR1, $p \leq 0.05$). In addition, feeding MLs significantly decreased the content of MDA in serum and muscle of mutton sheep ($p \leq 0.05$), the content of MDA in adipose tissue was Con>TR2>TR1 ($p \leq 0.05$). Table 4 \| Items \| Con \| TR1 \| TR2 \| SEM \| P-value \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Serum \| Serum \| Serum \| Serum \| Serum \| Serum \| \| SOD \| 58.72c \| 77.94a \| 66.56b \| 2.021 \| 0.000 \| \| CAT \| 32.80c \| 58.95a \| 45.22b \| 2.489 \| 0.000 \| \| GSH-PX \| 358.13c \| 543.05a \| 476.41b \| 17.737 \| 0.000 \| \| TAOC \| 7.37c \| 10.84a \| 8.49b \| 0.366 \| 0.000 \| \| MDA \| 5.09a \| 4.05b \| 4.60ab \| 0.150 \| 0.013 \| \| Muscle \| Muscle \| Muscle \| Muscle \| Muscle \| Muscle \| \| SOD \| 5.55c \| 9.64a \| 7.67b \| 0.475 \| 0.000 \| \| CAT \| 2.37c \| 4.49a \| 3.56b \| 0.241 \| 0.000 \| \| GSH-PX \| 23.12c \| 31.81a \| 27.35b \| 1.029 \| 0.000 \| \| TAOC \| 3.18c \| 5.10a \| 4.47b \| 0.226 \| 0.000 \| \| MDA \| 4.33a \| 3.31b \| 3.98a \| 0.134 \| 0.001 \| \| Adipose \| Adipose \| Adipose \| Adipose \| Adipose \| Adipose \| \| SOD \| 2.18c \| 5.23a \| 3.58b \| 0.371 \| 0.000 \| \| CAT \| 0.72 \| 1.68 \| 0.98 \| 0.201 \| 0.126 \| \| GSH-PX \| 6.01c \| 11.17a \| 8.20b \| 0.691 \| 0.003 \| \| TAOC \| 0.98 \| 2.05 \| 1.26 \| 0.254 \| 0.212 \| \| MDA \| 1.32 \| 1.17 \| 1.00 \| 0.185 \| 0.809 \| To further explore how MLs cause a differences in promoting oxidation resistance, muscle widely target metabolomics was applied. A total 43 significant differential metabolites (DEMs), including 19 upregulated and 24 downregulated DEMs after FMLs treatment, were filtered according to the criteria that the metabolite contents were within FC≥2 or FC ≤ 0.5, and VIP≥1 (Figure 1A). On this basis, the p-value is listed ascending order and absolute value of log2FC is listed in descending order of 43 DEMs to further obtain the leading 20 DEMs, shown in Figure 1B. These top-ranking DEMs were mainly involved in lipid, carbohydrate, amino acid, and organic acid metabolism. As the heatmap shows, anti-oxidant properties were negatively correlated (dark blue) with products of lipid metabolism (8-iso Prostaglandin F2α, 11β-Prostaglandin F2α, 8-iso Prostaglandin F2β, (±)8-HETE, 13-HOTrE, Carnitine C8:1), and were positively correlated (dark red) with D-Glucose 6-Phosphate, D-Fructose 6-Phosphate-Disodium Salt, D-Fructose-1,6-Biphosphate-Trisodium Salt, and D-Mannose 6-phosphate, which are related to carbohydrate metabolism (Figure 1C). This suggests that lipid metabolism and carbohydrate metabolism in FMLs treatment regulate the anti-oxidant process. More importantly, the correlation analysis suggests that the increased expression of 5-Hydroxy-L-Tryptophan and indoleacrylic acid produced by tryptophan metabolism may play crucial roles in anti-oxidant regulation. Figure 2A exhibited the DEMs from Con vs. TR2,including 16 upregulated and 39 downregulated, 55 in total DEMs, based on the same screening criteria as FMLs treatment. Similarly, the top-ranking 20 DEMs in Figure 2B obtained in the same method, are also mainly involved in lipid metabolism (8-iso Prostaglandin F2α, 11β-Prostaglandin F2α, 8-iso Prostaglandin F2β, 13-HOTrE, Carnitine C8:1), carbohydrate metabolism (D-Mannose 6-phosphate, D-Glucose 6-Phosphate, D-Fructose 6-Phosphate-Disodium Salt), amino acid metabolism and organic acid metabolism. However, few correlation relationships (yellow in Figure 2C) between carbohydrate metabolism and anti-oxidant properties in heatmap analysis show that DMLs supplementation might slightly, or not facilitate oxidation resistance by regulating carbohydrate metabolism. Figure 1: DEMs from Con vs. TR1. Upregulated, downregulated and total numbers of DEMs from Con vs. TR1 (A), the 20 leading DEMs from Con vs. TR1 (B), a correlation heat map between the 20 leading DEMs from Con vs. TR1 and their indexes of antioxidant performance (C). Figure 2: DEMs from Con vs. TR2. Upregulated, downregulated and total numbers of DEMs from Con vs. TR2 (A), the 20 leading DEMs from Con vs. TR2 (B), a correlation heat map between the 20 leading DEMs from Con vs. TR2 and their indexes of antioxidant performance (C). Considering the 20 DEMs and antioxidant performance indexes between Con vs. TR1 and Con vs. TR2, It is not too difficult to discover the importance of lipid metabolism, especially the peroxidation of polyunsaturated fatty acids (PUFAs), amino acid metabolism (mainly tryptophan metabolism) for MLs treatment, and carbohydrate metabolism (mainly glycolysis and mannose 6-phosphate pathway) only for FMLs treatment in promoting oxidation resistance. The tryptophan metabolism and peroxidation of PUFAs could be promising MLs-dependent biomarkers of the anti-oxidant metabolism pathway. Indoleacrylic acid and 5-hydroxy tryptophan (5-HTP) obtained from the two routes of tryptophan metabolism (Figure 3) were significantly upregulated. Indolelactic acid, an upstream metabolite of indoleacrylic acid, was also significantly increased in FMLs treatment ($$p \leq 0.015$$). Moreover, the markedly decreased 8-HETE, 13-HOTrE, 8-iso Prostaglandin F2α, 11β-Prostaglandin F2α, and 8-iso prostaglandin F2β levels and increased carnitine C8:1 are present after MLs treatment. Figure 3: Peroxidation of PUFAs and tryptophan metabolism. Elevated metabolites are highlighted in red, reduced metabolites are shown in blue; the contents of painted green or red metabolites from top 20 DEMs and antioxidant biochemical indexes are displayed in heat map (p<0.05, *p<0.01, and ** are TR1 or TR2 compared to Con). (MLs, mulberry leaves; FMLs, fermented mulberry leaves; DMLs, dried mulberry leaves; LP, Lactobacillus plantarum; LAB, lactic acid bacteria; PUFA, polyunsaturated fatty acids; ALA, α linolenic acid; ARA, arachidonic Acid; CPT1/CPT2, carnitine palmitoyltransferase 1/2; NFA, medium-chain fatty acid; 5-HTP, 5-hydroxytryptophan; 5-HT, serotonin; AAAD, aromatic amino acid decarboxylase; TPH1/2, tryptophan hydroxylase 1/2).* Transcriptome analysis was performed to further verify that reducing the peroxidation of PUFAs could indeed promote oxidation resistance. All filtered sequenced genes were used for weighted gene co-expression network analysis (WGCNA) analysis. Correlation analysis of different module genes and grouping factors and six DEMs related to PUFAs metabolism (8-iso Prostaglandin F2α, 11β-Prostaglandin F2α, 8-iso Prostaglandin F2β, (±)8-HETE, 13-HOTrE, carnitine C8:1) are shown in Figure 4A. Three module genes (underlined module in red in the Figure 4A) with almost the same correlation with the grouping factors and DEMs were integrated for further analysis. Subsequently, 14 target DEGs were obtained from the integrated genes with two criteria that their p-value must be less than 0.05, and absolute log2FC (Con vs. TR1) value must be not less than 1. Subsequently, they were gathered with six DEMs for network map analysis (Figure 4B). The relative expression levels of these 14 target DEGs in the three groups are shown in Figure 4C. Relative expression levels of GCNT1, IFITM10, EXTL1, RILP, BBC3, RAB9A, MOB3B, SESN1, CDH4, MEIS1, RAB9A, NUDT7, FMO2 and NUDT7 was decreased siginificantly ($p \leq 0.05$). Figure 4: DEGs related with peroxidation of PUFAs. Module analysis of DEMs related with PUFAs metabolism and all filtered genes (Underlined modules in red represent selective modules; A, carnitine C8:1; B, 8-iso-prostaglandin F2α; C, 11β-prostaglandin F2α; D, 8-iso-prostaglandin F2β; E, (±) 8-HETE; F, 13-HOTrE) (A), network map analysis of selective twelve DEGs and DEMs related with PUFAs metabolism (circle size represents absolute log2FC (Con vs TR1) value; blue, red and green divisions in every circle are on behalf of contents of some DEGs or DEMs in Con, TR1,TR2 in turn; The thickness of the connecting wire represents the degree of connectivity) (B) and the relative expression levels of selective twelve target DEGs in three groups (represents p<0.05, ** represents p<0.01, ns represents no differences) (C).* ## Immune response Serum immuno globulin G (IgG) and total immuno globulin (Ig) levels increased in the FMLs ($p \leq 0.05$) remarkably and DMLs fed groups ($p \leq 0.05$). The pro-inflammatory tumor necrosis factor-α (TNF-α) was dramatically reduced by both MLs treatments (Figure 5A) ($p \leq 0.05$). However there is a distinctive decrease in immuno globulin M (IgM) following FMLs treatment ($p \leq 0.05$). The decreased IgM following FMLs treatment is related to a transition in antibody class from IgM to IgG, over the course of an immune response [14] (Figure 5B). Figure 5: Indexes of immune properties and DEGs related to immune response. Immune indexes of serum (A) a transition from IgM to IgG in immune B cells over the course of immune response (Heat maps show the relative amounts of substances in group Con, TR1 and TR2 from left to right; represent p<0.05, indicative of the significant difference by comparing TR1 or TR2 to Con) (B), kegg enrichment analysis of all DEGs from TR1 vs. Con (C), Circular correlation analysis of six selective DEGs from top 4 kegg pathway and immune indexes of serum (D), Relative expression levels of six selective DEGs related to immune response (E) (* represents p<0.05, ** represents p<0.01, ns represents no differences in histograms).* As reported by Wu et al. [ 15], muscles support a strong immune response. To validate the promotion of the immune process of FMLs, all DEGs of FMLs treatment in muscle analyzed by transcriptomics were applied for enrichment analysis, and the top eight KEGG pathways are represented in a histogram (Figure 5C). The four leading enriched pathways (arrow’s place in Figure 5C) are closely related to apoptosis and immune processes. Subsequently, a total of six annotated DEGs from the leading four pathways and immune indices were combined to analyze the relevance and a clear relationship was shown in the circular map (Figure 5D). After MLs treatment, the relative expressions of the DEGs, including DOCK2, BBC3, MYO10, PIK3R3, PLA2G4D, GADD45A, were markedly altered (Figure 5E). Previously, we found that FMLs improves carbohydrate metabolism, and glycolysis is one of the key processes. It has been found that it can provide biosynthetic intermediates and reducing power for the growth and proliferation of immune cells. MLs treatments raise levels of glucose, the central substrate of glycolysis and FMLs supplementation significantly increases the contents of glucose-6-P, glyceraldehyde-3-P ($p \leq 0.05$) and almost significantly increases fructose-6-P ($$p \leq 0.054$$) (Figure 6). Additionally, there are significantly increased D-mannose in DMLs treatment ($p \leq 0.05$) and mannose-6-P in FMLs treatment ($p \leq 0.05$) (Figure 6). These two are both derived from mannose-6-P pathway. Figure 6: Glycolysis and the mannose-6-P pathway in immune function. Heat maps show the relative amounts of substances in group Con, TR1 and TR2 from left to right; represent p<0.05, indicative of the significant difference by comparing TR1 or TR2 to Con; M1, type 1 macrophages; M2, type 2 macrophages; Treg, regulatory T cells; P, phosphatase.* According to Sundling et al. [ 52], secreted antibodies confer immune protection by first attaching to foreign antigens through the paired variable regions of their immunoglobulin heavy and light chains. Immunity was enhanced with increased Ig, IgG and reduced TNFα in both MLs treatments. In addition, the FMLs induced maximum immunity in animals with a transition in antibody class from IgM to IgG, over the course of an immune response [14], During which, early low-affinity IgM antibodies are progressively replaced by more-effective, high-affinity IgG antibodies [53] to achieve effective serological immunity [52]. DOCK2 regulates the migration of certain subsets of immune cells via Rac activation [54] and plays an important anti-inflammatory role in the development of various inflammatory diseases [55]. BBC3 is a transcriptional apoptotic target gene and participates in the activation of cell death processes [56]. Pozo et al. [ 57] reported that pro-inflammatory MYO10 mediates inflammation in cancer by regulating genomic stability. Studies have shown that PIK3R3 is a multifunctional gene related to inflammatory diseases, livestock coat color, and cell proliferation (58–60). Shao et al. [ 61] clarified that PLA2G4D, a major pro-inflammatory factor, facilitates CD1a expression, which can be recognized by lipid-specific CD1a-reactive T cells, leading to the production of IL-22 and IL-17A. According to Ehmsen et al. [ 62] and Jiang et al. [ 63], the increased expression of GADD45A, a cell cycle regulator, can ameliorate liver fibrosis in rats and is a protective modifier of neurogenic skeletal muscle atrophy. Collectively, MLs supplementation improves muscle immune response and disease resistance. Glycolysis is a critical process closely related to the immune response, as well as provides biosynthetic intermediates and reducing power for cell growth and proliferation of immune cells [64]. The pentose phosphate pathway (PPP) from glucose-6-P to glyceraldehyde-3-P provides immune cells with key metabolites for immune function, such as reducing power for the synthesis of ROS and antioxidants in phagocytic cells and for phospholipid synthesis in dendritic cells. The hexosamine biosynthesis pathway, originating from fructose-6-P, provides substrates for the glycosylation of lipids and proteins that are important for Treg and M2 macrophage lineages [64]. Thus FMLs treatment may enhance immunity by glycolysis, which in turn provides key metabolites for immune function. D-mannose serves a vital function in T cell immune responses and is currently receiving increasing attention, although its normal physiological blood concentration is less than one-fiftieth of that of glucose. Zhang et al. [ 65] recognized that D-mannose induces regulatory T cells and suppresses immunopathology both in vivo and in vitro. Mannose-6-phosphate metabolized by D-mannose is a novel regulator of T cell immunity [66] and a promising target ligand in cancer therapy, as well as confers a better efficacy and lower toxicity in healthy tissues [67]. Moreover, mannose-6-P not only plays a crucial role in lysosomal functions (such as autophagy) but also in regulating lysosome biogenesis [68]. Thus, significantly increased D-mannose in DMLs treatment ($p \leq 0.05$) and mannose-6-P in FMLs treatment ($p \leq 0.05$) (Figure 6) via the mannose-6-P pathway enhances T cell immunity and likely regulates the lysosome biogenesis in autophagy. Taken together, FMLs supplementation could improve the immune response via glycolysis and the mannose-6-P pathway and induce class switch from low-affinity IgM to high-affinity IgG antibodies. ## Discussion Numerous studies have shown the diverse growth-promoting effects of MLs [4, 16]. Our study also proves this point. In addition, this study also found that FMLs are superior to FMLs in palatability and growth promotion, which is a rare feed additive, and its application prospects in animal husbandry production appear considerable. ## Anti-oxidation activity Oxidation in biological systems is mainly mediated by a series of redox enzymes. Peroxidation caused by free radical chain reactions may lead to oxidative stress [17]. SOD, CAT and GSH-Px are common enzymatic antioxidants. SOD can convert free radicals (O2−•) generated in the body’s peroxidation reaction into H2O2 [18], and H2O2 can then be converted into H2O by CAT and GSH-Px to reduce the damage resulting from free radical to the body and improve antioxidant performance. MDA is one of the representative end products under non-enzymatic lipid peroxidation, indicating the extent of lipid peroxidation [19]. Meanwhile, MDA is also an important indicator of membrane damage and body aging, and one of the toxic substances produced by the increase of ROS [20]. The increase of SOD, CAT, GSH-Px, TAOC and the decrease of MDA in this study all indicate that MLs can improve the antioxidant performance of the body, which is consistent with the results of previous studies (3, 21–23). The reason is that MLs are rich in bioactive ingredients. In addition, this study also found that FMLs have the strongest antioxidant properties, mainly because of their higher active ingredients than DMLs [24]. Indoleacrylic acid derived from tryptophan metabolism, has been shown to have significant anti-inflammatory effects in vitro and vivo [25] and also have beneficial effects on the intestinal epithelial barrier function [26]. Indolelactic acid, an upstream metabolite of indoleacrylic acid has been shown to possess antimicrobial, anti-oxidative, anti-inflammatory activities [26, 27] and can potentially modulate immune function [28]. L-5-hydroxytryptophan (5-HTP) is a monoamine neurotransmitter involved in the modulation of mood, cognition, reward, learning, memory, sleep, and numerous other physiological processes [29], and can also suppress inflammation and arthritis by decreasing the production of pro-inflammatory mediators [30]. Overall, MLs, especially FMLs, must endow anti-bacterial, anti-oxidant, anti-inflammation, and immunity-enhancing properties via tryptophan metabolism. Linoleic acid (LA), arachidonic acid (ARA), eicosapentaenoic acid (EPA) and α-linolenic acid (ALA) are representative of the main PUFAs, and the major metabolic pathways of peroxidation described in mammals are both enzymatic (cyclooxygenase, COX; lipoxygenase, LOX; cytochrome P450, CYP) and non-enzymatic [31] oxidation. 8-HETE and 13-HOTrE are all oxylipins, a group of oxidized metabolites derived from PUFAs [32]. Generally, the synthesis of oxylipins fluctuates with the changes of physiological or pathological states [33].13-HOTrE is derived from ALA via the COX enzymatic pathway. Studies have revealed that 13-HOTrE levels are significantly increased in some diseases [34, 35], such as acute liver injury. Therefore, it is generally thought to be a proinflammatory factor. HETEs are derived from ARA through COX catalysis. Hayashi et al. [ 36] reported that several ARA-derived (18-HETE/20-HETE) and ALA-derived (13-HOTrE) oxylipins tend to increase in bovine mastitic milk. In this study, MLs treatments reduced the 8-HETE contents. Meanwhile Ma et al. [ 37] also reported that 8-HETE is relevant for the efficacy of Zuojin pill treatment in chronic nonatrophic gastritis, as the level of 8-HETE was higher before treatment than after treatment. Thus, decreased oxylipins in this study with MLs treatments probably improve the antioxidant performance and immunity of the body and will be promising markers for livestock welfare. 8-iso Prostaglandin F2α, as a final product of lipid peroxidation, is generated from ARA interacting with ROS through nonenzymatic routes and is a robust oxidative stress biomarker of some diseases [32, 38]. 8-iso Prostaglandin F2β is a constitutional isomer of 8-iso Prostaglandin F2α. Oliveira et al. [ 39] found that 8-iso Prostaglandin F2β has much lower potency than 8-iso Prostaglandin F2α with an α-configuration. 11β-Prostaglandin F2α, as a metabolite of 8-iso Prostaglandin F2α, have been found to be associated with levels of oxidative stress in specific diseases [40]. Thus, the markedly decreased 8-iso Prostaglandin F2α, 11β-Prostaglandin F2α, and 8-iso prostaglandin F2β levels after MLs treatment indicate a decline in the peroxidation of PUFAs, which will produce benificial effects on lowering oxidative stress and enhancing disease resistance. Carnitine plays a key role not only in fatty acid β-oxidation, but also in immunity enhancement and disease resistance. Guo et al. [ 41] found that carnitine C8:1 was significantly decreased the in non-alcoholic steatohepatitis group, and this could be profoundly reversed after luteolin treatment. Studies have reported decreased serum acyl-carnitine concentrations in patients with cancer [42]. It can be hypothesized that increased carnitine C8:1 levels altered by MLs supplementation might accelerate mitochondrial β-oxidation [43] thereby enhancing immunity and disease resistance. Previous studies have shown that increased GCNT1, IFITM10, EXTL1, RILP, BBC3, RAB9A, MOB3B, SESN1, CDH4, MEIS1, RAB9A, NUDT7, and FMO2 are related to immune deficiency, autophagy inhibition, disease sensitivity and oxidative stress. Therefore, after treatment of MLs, the decreased peroxidation of PUFAs (the decreasing in peroxidation products) reduced the expression of the above genes, thus improving the immune and antioxidant properties. In addition, Shumar et al. [ 44] and Kerr et al. [ 45] suggest that decreased NUDT7 may reduce the accumulation of peroxisome through regulating the β-oxidation of peroxisome fatty acids, thus improving the antioxidant performance; Ge et al. [ 46] found that the decrease of NUDT7 enhanced the immune defense response; Taniguchi et al. [ 47] and Liu et al. [ 48] reported that NUDT7 with low expression may up-regulate heme biosynthesis and contribute to meat-redness enrichment. Therefore, the addition of MLs can not only reduce the peroxidation of PUFAs, enhancing the antioxidant capacity and immunity of the body, but also improve meat redness. Overall, oxidation resistance is closely related to the immune response. Our study proved that MLs supplementation is effective in promoting oxidation resistance and disease resistance, which is attributed to its function in reducing peroxidation of PUFAs and increasing tryptophan metabolism. Additionally, as lipid oxidation products affect the shelf life [49], sensory characteristics [50], and nutritional composition of meat [51], the role of MLs in reducing the peroxidation of PUFAs is speculated to be linked to the improvements in meat quality. ## Data availability statement The original contributions presented in the study are publicly available. This data can be found in the NCBI repository under accession number: PRJNA898816 [https://www.ncbi.nlm.nih.gov/search/all/?term=PRJNA898816] and in the MetaboLights repository under accession number: MTBLS6516. ## Ethics statement The animal study was reviewed and approved by Animal Care and Use Committee of the Institute of Northwest A & F University. Written informed consent was obtained from the owners for the participation of their animals in this study. ## Author contributions XC and YY contributed to the conception and design of the study. XC performed the statistical analysis and wrote the first draft of the manuscript. YY, MZ, FJ, LB, CS, and YQ revised the manuscript. SL, HW, ZL, XW, WQ and XS helped with the experimental sections. CS and YQ provided financial support for the manuscript. All authors contributed to manuscript revision, and read and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## References 1. Thaipitakwong T, Numhom S, Aramwit P. Mulberry leaves and their potential effects against cardiometabolic risks: a review of chemical compositions, biological properties and clinical efficacy. Pharm Biol (2018) 56. DOI: 10.1080/13880209.2018.1424210 2. Insang S, Kijpatanasilp I, Jafari S, Assatarakul K. Ultrasound-assisted extraction of functional compound from mulberry (Morus alba l.) leaf using response surface methodology and effect of microencapsulation by spray drying on quality of optimized extract. Ultrason Sonochem (2022) 82 1-11. DOI: 10.1016/j.ultsonch.2021.105806 3. Liu Y, Li Y, Xiao Y, Peng Y, He J, Chen C. Mulberry leaf powder regulates antioxidative capacity and lipid metabolism in finishing pigs. Anim Nutr (2021) 7. DOI: 10.1016/j.aninu.2020.08.005 4. Ding Y, Jiang X, Yao X, Zhang H, Song Z, He X. Effects of feeding fermented mulberry leaf powder on growth performance, slaughter performance, and meat quality in chicken broilers. Anim (Basel) (2021) 11 1-15. DOI: 10.3390/ani11113294 5. Li M, Hassan FU, Tang Z, Peng L, Liang X, Li L. Mulberry leaf flavonoids improve milk production, antioxidant, and metabolic status of water buffaloes. Front Vet Sci (2020) 7 1-12. DOI: 10.3389/fvets.2020.00599 6. Ouyang J, Wang M, Hou Q, Feng D, Pi Y, Zhao W. Effects of dietary mulberry leaf powder in concentrate on the rumen fermentation and ruminal epithelium in fattening hu sheep. Anim (Basel) (2019) 9 1-11. DOI: 10.3390/ani9050218 7. Cavill R, Jennen D, Kleinjans J, Briede JJ. Transcriptomic and metabolomic data integration. Brief Bioinform (2016) 17 891-901. DOI: 10.1093/bib/bbv090 8. Kim D, Langmead B, Salzberg SL. HISAT: a fast spliced aligner with low memory requirements. Nat Methods (2015) 12. DOI: 10.1038/nmeth.3317 9. Pertea M, Pertea GM, Antonescu CM, Chang TC, Mendell JT, Salzberg SL. StringTie enables improved reconstruction of a transcriptome from RNA-seq reads. Nat Biotechnol (2015) 33 1-20. DOI: 10.1038/nbt.3122 10. Li B, Dewey CN. RSEM: accurate transcript quantification from RNA-seq data with or without a reference genome. BMC Bioinf (2011) 12 1-16. DOI: 10.1186/1471-2105-12-323 11. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol (2014) 15 1-21. DOI: 10.1186/s13059-014-0550-8 12. Wang L, Feng Z, Wang X, Wang X, Zhang X. DEGseq: an r package for identifying differentially expressed genes from RNA-seq data. Bioinformatics (2010) 26. DOI: 10.1093/bioinformatics/btp612 13. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics (2010) 26. DOI: 10.1093/bioinformatics/btp616 14. Best J, Banatvala JE, Watson D. SERUM IgM AND IgG RESPONSES IN POSTNATALLY ACQUIRED RUBELLA. Lancet (1969) 294. DOI: 10.1016/S0140-6736(69)92386-1 15. Wu J, Weisshaar N, Hotz-Wagenblatt A, Madi A, Ma S, Mieg A. Skeletal muscle antagonizes antiviral CD8 +T cell exhaustion. Sci Adv (2020) 6 1-11. DOI: 10.1126/sciadv.aba3458 16. Islam MR, Siddiqui MN, A. Khatun MNAS, Rahman MZ, Selim ASM. Dietary effect of mulberry leaf (. SAARC J Agri (2014) 12 79-89. DOI: 10.3329/SJA.V12I2.21920 17. Huang D, Ou B, Prior RL. The chemistry behind antioxidant capacity assays. J Agric Food Chem (2005) 53. DOI: 10.1021/jf030723c 18. Coulombier N, Jauffrais T, Lebouvier N. Antioxidant compounds from microalgae: A review. Mar Drugs (2021) 19. DOI: 10.3390/md19100549 19. Wang W, Zhang Z, Liu X, Cao X, Wang L, Ding Y. An improved GC-MS method for malondialdehyde (MDA) detection: Avoiding the effects of nitrite in foods. Foods (2022) 11 1-14. DOI: 10.3390/foods11091176 20. Zhang J, Yang Z, Zhang S, Xie Z, Han S, Wang L. Investigation of endogenous malondialdehyde through fluorescent probe MDA-6 during oxidative stress. Anal Chim Acta (2020) 1116 9-15. DOI: 10.1016/j.aca.2020.04.030 21. Jin Y, Tu J, Han X, Zhuo J, Liu G, Han Y. Characteristics of mulberry leaf powder enriched with gamma-aminobutyric acid and its antioxidant capacity as a potential functional food ingredient. Front Nutr (2022) 9 1-13. DOI: 10.3389/fnut.2022.900718 22. Li R, Zhu Q, Wang X, Wang H. Mulberry leaf polyphenols alleviated high-fat diet-induced obesity in mice. Front Nutr (2022) 9 1-9. DOI: 10.3389/fnut.2022.979058 23. Shi Y, Zhong L, Fan Y, Zhang J, Zhong H, Liu X. The protective effect of mulberry leaf flavonoids on high-Carbohydrate-Induced liver oxidative stress, inflammatory response and intestinal microbiota disturbance in monopterus albus. Antioxidants (Basel) (2022) 11 1-18. DOI: 10.3390/antiox11050976 24. Cui X, Yang Y, Zhang M, Jiao F, Gan T, Lin Z. Optimized ensiling conditions and microbial community in mulberry leaves silage with inoculants. Front Microbiol (2022) 13 1-14. DOI: 10.3389/fmicb.2022.813363 25. Zhang B, Wan Y, Zhou X, Zhang H, Zhao H, Ma L. Characteristics of serum metabolites and gut microbiota in diabetic kidney disease. Front Pharmacol (2022) 13 1-17. DOI: 10.3389/fphar.2022.872988 26. Wlodarska M, Luo C, Kolde R, d'Hennezel E, Annand JW, Heim CE. Indoleacrylic acid produced by commensal peptostreptococcus species suppresses inflammation. Cell Host Microbe (2017) 22 25-37. DOI: 10.1016/j.chom.2017.06.007 27. Siedler S, Balti R, Neves AR. Bioprotective mechanisms of lactic acid bacteria against fungal spoilage of food. Curr Opin Biotechnol (2019) 56. DOI: 10.1016/j.copbio.2018.11.015 28. Laursen MF, Sakanaka M, von Burg N, Morbe U, Andersen D, Moll JM. Bifidobacterium species associated with breastfeeding produce aromatic lactic acids in the infant gut. Nat Microbiol (2021) 6. DOI: 10.1038/s41564-021-00970-4 29. Maffei ME. 5-hydroxytryptophan (5-HTP): Natural occurrence, analysis, biosynthesis, biotechnology, physiology and toxicology. Int J Mol Sci (2020) 22 1-25. DOI: 10.3390/ijms22010181 30. Yang TH, Hsu PY, Meng M, Su CC. Supplement of 5-hydroxytryptophan before induction suppresses inflammation and collagen-induced arthritis. Arthritis Res Ther (2015) 17 1-12. DOI: 10.1186/s13075-015-0884-y 31. Kuhn MJ, Mavangira V, Gandy JC, Zhang C, Jones AD, Sordillo LM. Differences in the oxylipid profiles of bovine milk and plasma at different stages of lactation. J Agric Food Chem (2017) 65. DOI: 10.1021/acs.jafc.7b01602 32. Samarra I, Masdevall C, Foguet-Romero E, Guirro M, Riu M, Herrero P. Analysis of oxylipins to differentiate between organic and conventional UHT milks. Food Chem (2021) 343 1-14. DOI: 10.1016/j.foodchem.2020.128477 33. Chavan-Gautam P, Rani A, Freeman DJ. Chapter six - distribution of fatty acids and lipids during pregnancy. Advances in clinical chemistry (2018) 84. DOI: 10.1016/bs.acc.2017.12.006 34. Peng S, Shen Y, Wang M, Zhang J. Serum and CSF metabolites in stroke-free patients are associated with vascular risk factors and cognitive performance. Front Aging Neurosci (2020) 12 1-13. DOI: 10.3389/fnagi.2020.00193 35. Zhan Z, Zhang T, Dai F, Wen X, Chen Y, Jiang H. Effect of oridonin on oxylipins in the livers of mice with acute liver injury induced by d-galactosamine and lipopolysaccharide. Int Immunopharmacol (2022) 102 1-8. DOI: 10.1016/j.intimp.2021.108387 36. Hayashi A, Fujii S, Nakamura T, Kobayashi K, Sakatani M, Endo M. Production of lipid mediators in mastitic milk of cow. Anim Sci J (2019) 90 999-1007. DOI: 10.1111/asj.13222 37. Ma X, Xie S, Wang R, Wang Z, Jing M, Li H. Metabolomics profiles associated with the treatment of zuojin pill on patients with chronic nonatrophic gastritis. Front Pharmacol (2022) 13 1-11. DOI: 10.3389/fphar.2022.898680 38. Atrosht.F. P, Kangasniemi. R, Usterman. T. Milk prostaglandins and electrical conductivity in bovine mastitis. Veterinary Res Commun (1987) 11 15-22. DOI: 10.1007/BF00361322 39. Oliveira L, Stallwood NA, Crankshaw DJ. Effects of some isoprostanes on the human umbilical artery. Br J Phar m acology (2000) 129. DOI: 10.1038/sj.bjp.0703083 40. O'Brien JW, Choi PM, Li J, Thai PK, Jiang G, Tscharke BJ. Evaluating the stability of three oxidative stress biomarkers under sewer conditions and potential impact for use in wastewater-based epidemiology. Water Res (2019) 166 1-7. DOI: 10.1016/j.watres.2019.115068 41. Guo W, Luo L, Meng Y, Chen W, Yu L, Zhang C. Luteolin alleviates methionine-choline-deficient diet-induced non-alcoholic steatohepatitis by modulating host serum metabolome and gut microbiome. Front Nutr (2022) 9 1-14. DOI: 10.3389/fnut.2022.936237 42. Malaguarnera M, Risino C, Gargante MP, Oreste G, Barone G, Tomasello AV. Decrease of serum carnitine levels in patients with or without gastrointestinal cancer cachexia. World J Gastroenterol (2006) 12. DOI: 10.3748/wjg.v12.i28.4541 43. Ruiying C, Zeyun L, Yongliang Y, Zijia Z, Ji Z, Xin T. A comprehensive analysis of metabolomics and transcriptomics in non-small cell lung cancer. PloS One (2020) 15 1-16. DOI: 10.1371/journal.pone.0232272 44. Shumar SA, Kerr EW, Fagone P, Infante AM, Leonardi R. Overexpression of Nudt7 decreases bile acid levels and peroxisomal fatty acid oxidation in the liver. J Lipid Res (2019) 60. DOI: 10.1194/jlr.M092676 45. Kerr EW, Shumar SA, Leonardi R. Nudt8 is a novel CoA diphosphohydrolase that resides in the mitochondria. FEBS Lett (2019) 593. DOI: 10.1002/1873-3468.13392 46. Ge X, Li GJ, Wang SB, Zhu H, Zhu T, Wang X. AtNUDT7, a negative regulator of basal immunity in arabidopsis, modulates two distinct defense response pathways and is involved in maintaining redox homeostasis. Plant Physiol (2007) 145. DOI: 10.1104/pp.107.103374 47. Taniguchi M, Hayashi T, Nii M, Yamaguchi T, Fujishima-Kanaya N, Awata T. Overexpression of NUDT7, a candidate quantitative trait locus for pork color, downregulates heme biosynthesis in L6 myoblasts. Meat Sci (2010) 86. DOI: 10.1016/j.meatsci.2010.05.045 48. Liu Y, Liu X, Zheng Z, Ma T, Liu Y, Long H. Genome-wide analysis of expression QTL (eQTL) and allele-specific expression (ASE) in pig muscle identifies candidate genes for meat quality traits. Genet Sel Evol (2020) 52 1-11. DOI: 10.1186/s12711-020-00579-x 49. Domínguez R, Pateiro M, Gagaoua M, Barba FJ, Zhang W, Lorenzo JM. A comprehensive review on lipid oxidation in meat and meat products. Antioxidants (2019) 8. DOI: 10.3390/antiox8100429 50. Chen H, Cao P, Li B, Sun D, Wang Y, Li J. Effect of water content on thermal oxidation of oleic acid investigated by combination of EPR spectroscopy and SPME-GC-MS/MS. Food Chem (2017) 221. DOI: 10.1016/j.foodchem.2016.11.008 51. Damerau A, Ahonen E, Kortesniemi M, Puganen A, Tarvainen M, Linderborg KM. Evaluation of the composition and oxidative status of omega-3 fatty acid supplements on the Finnish market using NMR and SPME-GC-MS in comparison with conventional methods. Food Chem (2020) 330 1-11. DOI: 10.1016/j.foodchem.2020.127194 52. Sundling C, Lau AWY, Bourne K, Young C, Laurianto C, Hermes JR. Positive selection of IgG(+) over IgM(+) b cells in the germinal center reaction. Immunity (2021) 54 988-1001. DOI: 10.1016/j.immuni.2021.03.013 53. Mäkelä O, Rouslahti E, Seppälä IJT. Affinity of IgM and IgG antibodies. Immunochemistry (1970) 7. DOI: 10.1016/0019-2791(70)90053-4 54. Chen Y, Meng F, Wang B, He L, Liu Y, Liu Z. Dock2 in the development of inflammation and cancer. Eur J Immunol (2018) 48. DOI: 10.1002/eji.201747157 55. Xu X, Su Y, Wu K, Pan F, Wang A. DOCK2 contributes to endotoxemia-induced acute lung injury in mice by activating proinflammatory macrophages. Biochem Pharmacol (2021) 184 1-15. DOI: 10.1016/j.bcp.2020.114399 56. Magalhaes-Novais S, Bermejo-Millo JC, Loureiro R, Mesquita KA, Domingues MR, Maciel E. Cell quality control mechanisms maintain stemness and differentiation potential of P19 embryonic carcinoma cells. Autophagy (2020) 16. DOI: 10.1080/15548627.2019.1607694 57. Pozo FM, Geng X, Tamagno I, Jackson MW, Heimsath EG, Hammer JA. MYO10 drives genomic instability and inflammation in cancer. Sci Adv (2021) 7 1-17. DOI: 10.1126/sciadv.abg6908 58. Dlamini NM, Dzomba EF, Magawana M, Ngcamu S, Muchadeyi FC. Linkage disequilibrium, haplotype block structures, effective population size and genome-wide signatures of selection of two conservation herds of the south African nguni cattle. Anim (Basel) (2022) 12 1-23. DOI: 10.3390/ani12162133 59. Ibrahim S, Zhu X, Luo X, Feng Y, Wang J. PIK3R3 regulates ZO-1 expression through the NF-kB pathway in inflammatory bowel disease. Int Immunopharmacol (2020) 85 1-8. DOI: 10.1016/j.intimp.2020.106610 60. Polini B, Carpi S, Doccini S, Citi V, Martelli A, Feola S. Tumor suppressor role of hsa-miR-193a-3p and -5p in cutaneous melanoma. Int J Mol Sci (2020) 21 1-18. DOI: 10.3390/ijms21176183 61. Shao S, Chen J, Swindell WR, Tsoi LC, Xing X, Ma F. Phospholipase A2 enzymes represent a shared pathogenic pathway in psoriasis and pityriasis rubra pilaris. JCI Insight (2021) 6 1-15. DOI: 10.1172/jci.insight.151911 62. Ehmsen JT, Kawaguchi R, Kaval D, Johnson AE, Nachun D, Coppola G. GADD45A is a protective modifier of neurogenic skeletal muscle atrophy. JCI Insight (2021) 6 1-16. DOI: 10.1172/jci.insight.149381 63. Jiang Y, Xiang C, Zhong F, Zhang Y, Wang L, Zhao Y. Histone H3K27 methyltransferase EZH2 and demethylase JMJD3 regulate hepatic stellate cells activation and liver fibrosis. Theranostics (2021) 11. DOI: 10.7150/thno.46360 64. Andrejeva G, Rathmell JC. Similarities and distinctions of cancer and immune metabolism in inflammation and tumors. Cell Metab (2017) 26 49-70. DOI: 10.1016/j.cmet.2017.06.004 65. Zhang D, Chia C, Jiao X, Jin W, Kasagi S, Wu R. D-mannose induces regulatory T cells and suppresses immunopathology. Nat Med (2017) 23. DOI: 10.1038/nm.4375 66. Ara A, Ahmed KA, Xiang J. Mannose-6-phosphate receptor: a novel regulator of T cell immunity. Cell Mol Immunol (2018) 15. DOI: 10.1038/s41423-018-0031-1 67. Dalle Vedove E, Costabile G, Merkel OM. Mannose and mannose-6-Phosphate receptor-targeted drug delivery systems and their application in cancer therapy. Adv Healthc Mater (2018) 7 1-37. DOI: 10.1002/adhm.201701398 68. Zhang W, Yang X, Li Y, Yu L, Zhang B, Zhang J. GCAF(TMEM251) regulates lysosome biogenesis by activating the mannose-6-phosphate pathway. Nat Commun (2022) 13 1-17. DOI: 10.1038/s41467-022-33025-1
--- title: 'Study on positive psychology from 1999 to 2021: A bibliometric analysis' authors: - Feifei Wang - Jia Guo - Guoyu Yang journal: Frontiers in Psychology year: 2023 pmcid: PMC10015893 doi: 10.3389/fpsyg.2023.1101157 license: CC BY 4.0 --- # Study on positive psychology from 1999 to 2021: A bibliometric analysis ## Abstract ### Objective Positive psychology is a revolution in the science of psychology as well as a new milestone in the development of human society. The purpose of the study was to use bibliometrics and visual analysis to assess the current state and trends in positive psychology research. ### Methods The Web of Science Core Collection was searched for 4,378 papers on positive psychology between 1999 and 2021. The features of positive psychology research were analyzed using Microsoft Excel 2013, VOSviewer (1.6.17), and CiteSpace (5.8.R1). ### Results The findings demonstrate a steady growth in positive psychology publications from 1999 to 2021. The United States [1,780] and Harvard University [104], respectively, were the most productive nations and organizations in this subject. Frontiers in Psychology was the most productive journal [288], while the Journal of Personality and Social Psychology had the most co-citations [8,469]. Seligman was the most influential author, with 3,350 citations and 5,020 co-citations. The top ten co-cited references, in terms of citation explosion, suggesting that these papers provide the foundation for the growth of this discipline. The systematic review, character strengths, positive psychology intervention, language pleasure, and the COVID-19 pandemic are the focal points of research and development developments in this discipline. ### Conclusion These findings have helped researchers in positive psychology find new ways to collaborate with partners, hot topics, and research frontiers. ## 1. Introduction Positive psychology is a vibrant field of research, and concepts about the components of well-being predate the positive psychology movement. Diener [1984] evaluated the literature on subjective well-being (SWB) in 1984, then created and validated the Life Satisfaction Scale (SWLS) (Diener et al., 1985). Ryff [1989, 2014, 2022] has studied psychological well-being for over 30 years in an effort to determine its fundamental components, find what conditions encourage or hinder it, and investigate how it impacts health. In 1998, Seligman was elected president of the American Psychological Society. He advocated that psychologists and practitioners concentrate on enjoyment rather than illness reduction. Many psychologists advocate a greater emphasis on positive psychological development. It discusses how to develop attributes such as imagination, optimism, foresight, interpersonal talents, moral judgment, patience, humor, and fearlessness, as well as how to promote pleasure and life satisfaction (Gillham and Seligman, 1999). The Millennium issue of American Psychologist focuses on the emerging science of positive psychology. Psychologists are beginning to consider what advantages humans have at the end of the twentieth century. Positive Psychology, published in 2000 by Seligman and Csikszentmihalyi [2000], marked the formal beginning of positive psychology. So this manuscript will focus on the development of positive psychology after its formal beginning. Therefore, Seligman [2019] is often called the “Father of Positive Psychology.” The application of psychological ideas, research, and intervention strategies to comprehend the good, adaptable, imaginative, and emotionally satisfying elements of human behavior is known as positive psychology (Seligman and Csikszentmihalyi, 2000). It shifts the research’s focus to “ordinary people” (Sheldon and King, 2001). Positive psychologists, like other natural or social scientists, seek to understand psychological structure, phenomena, and functions. There has been significant progress in the theory of positive psychology, as evidenced by PERMA (the five elements of well-being; Seligman, 2011), the dual-factor model of mental health (Suldo and Shaffer, 2008), the complete state model of health (Keyes, 2005), and the broaden-and-build theory of positive emotions (Fredrickson, 2001). Subsequently, positive psychology was widely used, including national psychological accounts of well-being (Diener and Seligman, 2018), positive psychotherapy (Seligman et al., 2006), a classification of strength and virtue (Snow, 2018), comprehensive soldier fitness (Lester et al., 2022), positive education (Seligman et al., 2009), and so on. With the advancement of positive psychology studies, positive psychology intervention has attracted researchers’ attention. Positive psychological intervention (PPI) is defined as “building its intervention on positive psychology theory and employing its coherent theoretical model to achieve the objective of promoting happiness” (Carr et al., 2020). Many positive psychology interventions have been shown to significantly boost well-being and minimize depressive symptoms (Sin and Lyubomirsky, 2009; Bolier et al., 2013; Carr et al., 2020). As the COVID-19 global health crisis unfolds, positive psychology is critical for sustaining mental wellness (Waters et al., 2021). After more than two decades of development, many research papers have been published in the field of positive psychology research. Researchers systematically reviewed 1,336 articles published between 1999 and 2013 and found that positive psychology is a growing and dynamic subfield within the broader discipline of psychology (Donaldson et al., 2014). However, the field has grown rapidly in recent years, adding a large body of literature that requires us to use scientometric methods for analysis. Therefore, we must pay more attention to the research hotspots and trends in positive psychology. Scientometrics is a powerful tool for identifying emerging trends and hotspots in the subject (Chen, 2017; Hou et al., 2018). Bibliometric analysis is more objective and efficient than standard qualitative analysis approaches. In recent years, the advancement of scientific mapping techniques has increased (Cobo et al., 2011). Scientific mapping technologies generally capture a bibliographic record of a group of study fields and create an overview of the underlying knowledge domains. Typical tools are CiteSpace (Chen, 2006) and VOSviewer (van Eck and Waltman, 2010; Ding and Yang, 2020). Computationally aided literature reviews are not intended to replace expert-written reviews; rather, they are intended to provide an additional point of reference with some advantages (Chen et al., 2014b). Bibliometric analysis is currently used extensively in a variety of research areas, including depression (You et al., 2021; Zhou Y. et al., 2021); mindfulness research (Baminiwatta and Solangaarachchi, 2021); COVID-19 (Chen, 2020; Yu et al., 2020); and so on. The use of bibliometric analysis in clinical practice has become increasingly popular. Unlike typical expert-compiled evaluations, scientometrics covers a broader and more diverse range of critical issues. A collection of scholarly literature reflecting positive psychology research was used as input for this research. As a result, the purpose of this research is to conduct a bibliometric analysis of positive psychology research from 1999 to 2021 using the software packages CiteSpace and VOSviewer in order to better understand the field’s current condition, hotspots, and developmental trends. ## 2.1. Data acquisition The study’s data source was the Web of Science database, which is the world’s most reliable citation database and has numerous high-quality papers (Huertas González-Serrano et al., 2019; Luo et al., 2021). The data is gathered mostly from the Web of Science Core Collection (WoSCC), which includes SCI-Expanded, SSCI, ESCI, and A&HCI. Due to the WoSCC literature retrieval database’s ongoing updating, we only conducted a single search on April 18, 2022. The elements of the field of positive psychology were initially outlined by Seligman [2019] in 1998. Additionally, the APA Thesaurus included the index phrase “positive psychology” (Gallagher Tuleya, 2007) in June 2003. Positive psychology topics are also reflected by a variety of additional thesaurus phrases (for example, “well-being,” “life satisfaction,” “positive emotions,” “happiness,” and so on). It was challenging to establish the eligibility of the literature due to the abundance of terminology connected to the issue of positive psychology. This study focuses on the exact term “positive psychology” (Schui and Krampen, 2010), as well as the time period 1999–2021. ## 2.2. Inclusion criteria The search yielded approximately 5,374 publications. The language utilized is English, and the type of literature is limited to “article” or “review.” This advanced search process excludes 996 articles. Finally, 4,378 publications were obtained and analyzed. ## 2.3. Analysis methods Analytics used in the study include Microsoft Excel 2013; VOSviewer (1.6.17); and CiteSpace (5.8.R1). We use Microsoft Excel 2013 to analyze the changes and trends in the number of documents (Zhou T. et al., 2021). VOSviewer is a document analysis software package that has been developed by Van Eck and Waltman (van Eck and Waltman, 2010). It has been proven to have excellent visualization and analysis results and is widely used for document analysis (van Eck and Waltman, 2010; Liao et al., 2018). In this study, the software is used to analyze the features of positive psychology studies, such as Countries/Regions, institutions, journals, and authors. Using the method of full counting to construct a bibliometric network (Perianes-Rodriguez et al., 2016). The size of the nodes represents the number of publications, while the overall connection strength value illustrates the degree of collaboration between a subject and others (Zou and Sun, 2019). CiteSpace is bibliometric analysis software (Chen, 2006), which can be used for the analysis of co-citations, burst detection, and emerging research trends in the literature (Chen et al., 2012; Hou et al., 2018). By developing a collection of visual knowledge maps, CiteSpace explores the states, hotspots, frontiers, and evolution processes in a scientific field. The parameters of CiteSpace are as follows: time split between January 1999 and December 2021 (each slice is 1 year), the analysis items are selected as references, one node type is selected at a time, the selection criteria [g-index ($k = 35$)], and pruning (Pathfinder). A visual knowledge map is created using nodes and connections. In the map, each node represents one reference. The size of the nodes reveals the frequency of reference, while different colors of nodes stand for different years. In the center, the burst node as a red circle represents the number of co-occurrences or references that grow with time. Purple nodes represent the centrality and important knowledge exhibited by the data (Chen, 2012). The line of connection between nodes is taken to be a co-occurrence or co-cited relation; the thickness of the line signifies the strength of the relationship, and the color corresponds to the time of the first node (Liu and Chen, 2012). Cold to warm colors represent the early to recent. Betweenness centrality is another name for centrality. Nodes with high mental quality (>0.1) are frequently considered paradigmatic or pivotal moments in a discipline. An explosion of references to citations explores the trend and shows if the relevant writers have gotten significant attention on this subject (Chen et al., 2014a). Researchers may use this map to better understand new trends and identify hotspots by using burst detection and analysis (Chen et al., 2014a; Shi et al., 2022). ## 3.1. Time trend analysis of publication outputs The search yielded a total of 4,378 publications, of which 4,021 were articles and 357 were reviews. This annual publication may demonstrate the trend in this field of research, which we portray as a broken line chart (Figure 1). Figure 1 shows that the number of papers published between 1999 and 2021 continues to increase, indicating that related research areas are increasingly attracting academic interest. In particular, the literature on positive psychology has grown substantially in the last 3 years. The number of publications is expected to continue to grow. Figure 1: Trends in annual publications in positive psychology research. ## 3.2. Analysis of countries/regions Between 1999 and 2021, 96 countries/regions published research on positive psychology. We use the parameters of the number of publications (≥10) and the strength of the lines (≥1), generating 46 nodes and 428 links in the network of partner countries/regions (Figure 2). In Figure 2, we can see that the United States had the most publications [1,780], accounting for $40.66\%$ (1,$\frac{780}{4}$,378), much outnumbering the rest, followed by England [420], Australia [388], the People’s Republic of China [361], and Canada [298]. The table compares the top 10 countries/regions in terms of the number of publications, WoS citations, citations per study, and overall link strength (Table 1). The United States was the leading country, ranking first in terms of publications, WoS citations, citations per article, and total link strength, indicating that the United States is absolutely dominant in the field of positive psychology. Figure 2: Map of countries/regions in positive psychology research. TABLE_PLACEHOLDER:Table 1 ## 3.3. Analysis of institutions We use the parameters of the number of publications (≥20) and the strength of the line (≥1), generating 61 nodes and 268 links in the network of partner universities (Figure 3). The research institution knowledge map assists us in understanding the key research institutions in this subject as well as their collaborative links. Harvard University, the University of Michigan, the University of Pennsylvania, and the University of Melbourne are all prominently displayed in Figure 3. In terms of the number of publications, 11 universities rank in the top 10. Table 2 shows that each organization participated in at least 43 studies related to positive psychology. Seven of them are from the United States, with the others coming from Australia, Switzerland, South Africa, and Canada. Harvard University was placed first among these universities, with 104 studies completed, followed by the University of Michigan ($$n = 87$$) and the University of Pennsylvania ($$n = 84$$). The top 10 institutions, as shown in Table 2, produced $15.50\%$ of all publications. Among these institutions, the University of Michigan had the most WoS citations [13,234] and citations per article (152.11). The top 10 producing institutions are listed in Table 2. Figure 3: Map of institutions in positive psychology research. TABLE_PLACEHOLDER:Table 2 ## 3.4. Analysis of journals and co-cited journals In Figure 4, we use the parameters of the number of publications (≥20) and the strength of the line (≥1), generating 21 nodes and 156 links in the journal citation map. The top 10 academic journals that publish articles on positive psychology research are shown in Table 3. These publications varied in IF from 0.917 to 4.614 (average IF: 3.228), and they are specialized journals in this field. Of these, the International Journal of Environmental Research and Public Health has the highest factor of influence (4.614). The 10 journals published a total of 986 papers connected to positive psychology research, accounting for $22.52\%$ of the 4,378 studies collected. At least 138 papers were published in the top three journals. In terms of link strength, the Journal of Positive Psychology ranked first ($$n = 1$$,431), followed by the Journal of Happiness Studies ($$n = 1$$,130) and Frontiers in Psychology ($$n = 1$$,096). On this subject, they are quite important. Figure 4: Map of journals in positive psychology research. TABLE_PLACEHOLDER:Table 3 Table 3 displays the top 10 co-cited journals. The journals with the highest academic power and important positions in the field are those with a high co-citation count. In terms of IF, these journals ranged from 2.935 to 23.027 (average IF: 8.214), and they are professional journals in this field. The most influential of these is Psychological Bulletin (23.227), followed by American Psychologist (16.358) and Journal of Applied Psychology (11.802). The Journal of Personality and Social Psychology had the highest number of co-citations [8,469], followed by American Psychologist [7,341], and then the Journal of Positive Psychology [4,057]. As a result of the examination of the co-citation count, the Journal of Personality and Social Psychology has been recognized as the core journal in the positive psychology research area. ## 3.5. Analysis of authors and co-cited authors For authors who posted more than 10 publications, generating a co-author map using VOSviewer resulted in 72 nodes and 107 links (Figure 5). The largest network of partnerships we found included Huffman JC, Proyer RT, Fredrickson BL, Ruch W, and other lead authors (Figure 6). In terms of publications, Ruch W published most of the research [51], followed by Huffman JC [40], Celano CM [30], and Proyer RT [30]. In terms of citations, the top three authors are Seligman MEP [3,350], Wood AM [3,284], and Maltby J [2,432]. Author Maltby J published most of the citations per paper (221.09) in terms of the number of citations per paper, followed by Wood AM (218.93) and Seligman MEP (159.52). Table 4 displays the top 10 authors in terms of publications, citations, and citations per paper in positive psychology research. They are well-known and active authors in this discipline. The top 10 co-cited authors are also shown in Table 4. The most co-cited author is Seligman MEP [5,020], followed by Diener E [2,809] and Fredrickson BL [2,434]. These authors have made remarkable contributions to the field of positive psychology. Figure 5: The authors’ map of positive psychology research. Figure 6: The largest author collaboration network in positive psychology research. TABLE_PLACEHOLDER:Table 4 ## 3.6. Analysis of co-cited references Essentially, science is a dynamic accumulation process. This means that when scholars write scientific papers, they need to cite others’ academic works (Shafique, 2013). The basis of this field of study is represented by the co-citations, which relate to the references that are also listed in the reference lists of other works. CiteSpace allows for automatic labeling of clustering, greatly reducing the subjectivity of the study of search bounds (Hou et al., 2018). Figure 7 depicts a cluster visualization of the CiteSpace software-generated coreference network, which was split into 28 clusters, only the 11 largest of which were retrieved from the references based on indexing terms and determined by a log-likelihood ratio algorithm. They are depicted in the image with various convex hulls, including systematic review (cluster #0), character strength (cluster #1), positive psychology intervention (cluster #2), level matrix model (cluster #3), positive psychology perspective (cluster #4), foreign language enjoyment (cluster #5), adolescent athlete well-being (cluster #6), coronary heart disease (cluster #7), employee well-being (cluster #8), the second wave (cluster #9), and subjective wellbeing questionnaire (cluster #10). The authors of each node in the map are identified in red, and each node indicates a referenced reference. The reference co-citation map’s cluster representation is shown in Figure 7. Table 5 displays the characteristics of the top 11 reference clusters in the co-citation network. The configuration’s overall clarity is stronger the closer each cluster’s silhouette score is to one (Chen, 2020). Each cluster had an average silhouette greater than 0.9348 and an overall Q-value of 0.8649, indicating that the quality of the cluster was extremely reliable. Figure 7: Reference co-citation network analysis of publications in positive psychology research. Cluster visualization of the reference co-citation map. TABLE_PLACEHOLDER:Table 5 We can map the same group of words to the same horizontal axis in the timeline view, with the document located below the horizontal axis; the closer to the left, the newer the document (Figure 8). The figure allows us to see the temporal characteristics of each cluster and the extent of clustering. A total of seven clusters lasting until 2020 are shown in the figure, including systematic review (cluster #0), character strength (cluster #1), foreign language enjoyment (cluster #5), health behavior (cluster #11), COVID-19 pandemic (cluster #14), theoretical model (cluster #18), and momentary blip (cluster #27). Indicating that these fields of research are still receiving scholarly attention. Figure 8: Timeline view in positive psychology research. The top five references in positive psychology research were listed according to their characteristics (see Supplementary material for details). They are regarded as the cornerstone research for the field of positive psychology. The top-ranked paper was published by Seligman MEP, with 1,619 co-citations. The articles with the most co-citations are usually key foundational works in this discipline. Positive Psychology: an Introduction received the most co-citations, making it the most important reference. The research concludes by defining the positive psychology scientific framework, pointing out knowledge gaps, and predicting the future of science careers in the twenty-first century. Furthermore, the authors noted that it enables individuals, communities, and society to thrive (Seligman and Csikszentmihalyi, 2000). While the strongest reference citation burst is considered the primary knowledge of the trend. As can be seen, Seligman MEP led the reference burst in 2000, and the burst was 52.03. The top 30 most potent citation bursts from 1999 through 2021 are shown in Figure 9. Additionally, the number of red squares corresponds to the time of the epidemic in the literature, and each red square indicates a year. Figure 9 lists a few important works of literature. It demonstrates that the red arrow’s target reference is a crucial one with a powerful explosion. Seligman and Csikszentmihalyi [2000], Peterson [2004], Seligman et al. [ 2005], Sin and Lyubomirsky [2009], Seligman [2011], and Lyubomirsky and Layous [2013] are a few examples of references. Hayes [2017] and Rashid [2014] are still bursting and need our high attention. Figure 9: Top 30 references with strong citation bursts in positive psychology research. Important references are also included in Table 6. Bolier et al. [ 2013] is the highest-ranked object per burst in Cluster #2, with bursts of 57.41. The second is Seligman et al. [ 2005] in Cluster #4, with bursts of 56.59. The third is Seligman and Csikszentmihalyi [2000] in Cluster #248, with bursts of 52.03. The 4th is Sin and Lyubomirsky [2009] in Cluster #2, with bursts of 49.35. Peterson [2004] is the 5th in Cluster #3, with a burst number of 37.29. More detailed information is shown in Table 6. Table 6 \| Bursts \| Reference \| DOI \| Cluster-ID \| \| --- \| --- \| --- \| --- \| \| 57.41 \| Bolier et al. (2013) \| 10.1186/1471-2,458-13-119 \| 2 \| \| 56.59 \| Seligman et al. (2005) \| 10.1037/0003-066X.60.5.410 \| 4 \| \| 52.03 \| Seligman and Csikszentmihalyi (2000) \| 10.1037/0003-066X.55.1.5 \| 248 \| \| 49.35 \| Sin and Lyubomirsky (2009) \| 10.1002/jclp.20593 \| 2 \| \| 37.29 \| Peterson (2004) \| \| 3 \| \| 31.72 \| Lyubomirsky and Layous (2013) \| 10.1177/0963721412469809 \| 2 \| \| 26.25 \| Seligman (2011) \| \| 9 \| \| 22.07 \| Hayes (2017) \| \| 14 \| \| 21.84 \| Fredrickson (2001) \| 10.1037/0003-066X.56.3.218 \| 4 \| \| 21.3 \| Lyubomirsky (2011) \| 10.1037/a0022575 \| 2 \| ## 4.1. General information in positive psychology research In this study, a bibliometric analysis of positive psychology research was conducted from 1999 to 2021. Based on the overall analysis of publications, productive countries/regions, institutions, journals, and authors, we can provide further research suggestions for researchers. The summary is as follows: ## 4.2. Emerging trends and hotpots in positive psychology research The co-cited references are what make up the knowledge base. A key component of CiteSpace is co-citation analysis. The time slice used in this study was 1 year; the selection criteria used a modified g index ($k = 35$); and the period is from 1999 to 2021 years. CiteSpace describes the trends and patterns of change in the co-citation reference map, which can be used to capture the research focus of the prospective scientific community. The node in Figure 7 represents a single reference, and the line indicates that the two references are connected in some way. The most representative articles of the cluster list in Supplementary material. For example, in cluster #0, the papers cited that are most related to the cluster are Hendriks et al. [ 2020], which cited $19\%$ of the contributions of the cluster; Neumeier et al. [ 2017], Moskowitz et al. [ 2020], Hausler et al. [ 2017], which cited $17\%$; and Job and Williams [2020], which cited $14\%$ of the literature. ## 4.2.1. Clustering visualization of the reference co-citation map CiteSpace divides the co-citation network into clusters of many co-citation references so that connections across clusters are weak but strong inside each cluster. The 11 major clusters are listed in Table 5 according to their size, or the total number of persons in each cluster. Large-membership clusters are displayed. If the silhouette score is close to 1, it indicates that the cluster has better homogeneity or coherence. Table 5 shows that all clusters had a high silhouette score, indicating better homogeneity or coherence. Based on the labels selected for the clusters by the log-likelihood ratio test method (LLR) (Chen et al., 2010). The three largest clusters were analyzed, and the results are as follows: With 150 members and a silhouette value of 0.872, cluster #0 is the biggest cluster. All references were across 14 years, from 2007 to 2020, and the median year was 2014. It is labeled “systematic review” by LLR. Hendriks et al. [ 2020] published an article that cited the most references in cluster #0. Through a detailed review and meta-analysis, this paper aims to determine if MPPIs are effective. These findings show that MPPI is successful in enhancing mental health. Further good research in different populations is needed to strengthen the claim for the effectiveness of MPPI. While the label for this cluster is “systematic review,” which states that the article type is “review,” the focus of the article is primarily on positive psychology interventions. Neumeier et al. recently developed an online intervention program aimed at improving employees’ well-being (Neumeier et al., 2017). Moskowitz et al. reviewed emotion measurement in positive psychology interventions (Moskowitz et al., 2020), and Job and Williams reviewed the role of online positive psychology interventions in sexual and gender minorities (SGM) (Job and Williams, 2020). From many reviews, it is also found that there has been a lot of research on positive psychology intervention, which is a hot spot in the field of positive psychology. With 117 members and a silhouette value of 0.898, cluster #1 is the second-largest cluster. For all references across 10 years (from 2011 to 2020), the median year was 2014. It is labeled as a character strength by LLR. Waters et al. [ 2021] published an article that cited the most references in cluster #1. The authors underline that strengthening mental health during COVID-2019 and developing positive processes and capacities will benefit the future development of mental health. This study covers the research and applied positive psychology themes of meaning, coping, self-compassion, courage, gratitude, personality advantage, positive emotion, a positive interpersonal process, and high-quality connection to help individuals cope with the epidemic. Apart from this paper, other papers also satisfy the clustering theme of “character strength.” Miglianico et al. [ 2019] review the literature on the use and development of strengths in the workplace. Strecker et al. [ 2020] will discover the circumstances for the use of individual character strengths in the workplace, resulting in enhanced job engagement and well-being. Mayerson [2020] proposed a model for the role of character strengths in the success of individuals, groups, and species. Hausler et al. [ 2017] examined the individual relationships between 24 different aspects of personality strengths, subjective well-being (SWB), and psychological well-being (PWB). Overall, the correlation between “good personality” and PWB was significantly stronger than that of SWB. As can be seen in Cluster #1, not only is “character strength” research emphasized, but more researchers are paying attention to “character strength” in the workplace. With 104 members and a silhouette value of 0.886, cluster #2 is the third-largest cluster. All references were across 14 years, from 2005 to 2018, and the median year was 2011. It is labeled as a positive psychology intervention by LLR. Bolier et al. [ 2013] published an article that cited the most references in cluster #2. The goal of this meta-analysis is to look into the effectiveness of a positive psychology intervention on the general population as well as on those who have specific psychosocial issues. Overall, the paper shows that positive psychology interventions can successfully improve subjective and psychological well-being while also assisting in the reduction of depressive symptoms. Figure 7 shows that clusters #2 and #0 partially overlap, there are some similarities between the two clusters, and there are many papers cited to investigate positive psychology intervention. Schueller and Parks [2014] summarized the current state of positive psychology interventions as they relate to self-help and that the next stage in research necessitates the application of these tactics in ways that allow them to be used in real-world circumstances. Gander et al. [ 2012] conclude that some “strengths-based” therapies can improve happiness. Likewise, Proyer et al. [ 2014] conducted a positive psychology intervention on older people over 50 years of age, and the results showed that the intervention was effective. The results of the Hone et al. [ 2014] meta-analysis show that positive psychology interventions have a clear effect on promoting well-being. To maximize the potential of PPI to promote population health, there is a need to extend the efficacy trial report in the future. Positive psychology intervention is a hotspot for positive psychology research. A great deal of research has been conducted by researchers from the perspectives of meta-analysis, literature reviews, and intervention experiments. ## 4.2.2. Co-citation clusters timeline map Figure 8 depicts the age span of the literature in each cluster. The clusters are placed vertically in decreasing size order, and each cluster is presented from left to right (Chen, 2017). Based on the timeline map, we should focus on larger and more recent clusters. The three largest clusters are #0, #1, and #2, which were analyzed earlier in this paper. Clusters #5, #14, #18, and #27 are relatively recent in terms of time and require further attention. Since clusters #18 and #27 contain few articles and are not representative, we retain only clusters #5 and #14 for the analysis. The results are as follows: Cluster #5 consists of 79 members with a silhouette value of 0.981. All references covered over 9 years, from 2012 to 2020, and the median year was 2017. It is labeled “foreign language enjoyment” by LLR. Elahi Shirvan et al. [ 2021] published an article that cited the most references in cluster #5. In response to the dynamic change in the SLA domain and the necessity for the creation of appropriate methodologies to evaluate the dynamics of developing notions in the field such as grit and pleasure, the current study sought to investigate the rise of foreign language enjoyment (FLE) and L2 grit over time. Wang X. et al. [ 2021] investigate Chinese university students’ enjoyment of a web-based language learning environment. In L2 education, Li and Xu [2019] found that an intervention focused on emotional intelligence has a good effect on promoting positive emotions. At the same time, as a result of the positive impact of second language acquisition (SLA) on the promotion of academic achievement and language learners’ well-being (Guo, 2021). Recently, Wang Y. L. et al., [ 2021] reviewed the role of positive psychology in promoting second language learning. Accordingly, until recently, with the emergence and rapid development of positive psychology in general education (Dewaele, 2014), there has been a clear positive revival in the area of L2 education (Lake, 2015; Kruk, 2019), which has also emerged as a hotspot for study and a trend in the discipline of positive psychology. Cluster #14 consists of 31 members and has a silhouette value of 1. This cluster has the highest homogeneity or coherence, indicating that the degree of coherence in the literature in this cluster is the highest. All references were over 9 years from, 2012 to 2020, and the median year was 2018. It is labeled as a COVID-19 pandemic by LLR. Waters et al. [ 2021] published an article that cited the most references in cluster #14. Although almost all of the papers cited within the cluster were published recently, the topics covered were COVID-19 and positive psychology, demonstrating that positive psychology plays a crucial role in the epidemic (Quiroga-Garza et al., 2021). The COVID-19 epidemic had a significant impact on people’s lives and mental health (Luo et al., 2020). Researchers should concentrate more on applying positive psychology to COVID-19. ## 5. Conclusion In conclusion, our bibliometric analysis of positive psychology found that positive psychology is a rapidly growing discipline with some achievements that warrant further research. In this study, Microsoft Excel 2013, VOS viewer (1.6.17), and CiteSpace (5.8.R1) software were used to analyze the annual number of documents, cooperation networks (countries/regions, institutions, journals and cited journals, authors and cited authors), and total cited documents. By analyzing data from the large-scale literature, we can gain a comprehensive understanding of the development of the field of positive psychology research and the research trends in the field. We can understand the general information in this field. Firstly, the number of papers published on this topic continues to grow, indicating that it is a research hotspot in the field of psychology. Secondly, in the analysis of the cooperation network, we can find that the United States and the institutions of the United States occupy a dominant position in this field; in journals, we should pay attention to several major journals, such as Frontiers in Psychology, Journal of Positive Psychology, Journal of Happiness Studies, and so on; In terms of an author analysis, authors such as Ruch W, Huffman JC, Celano CM, Proyer RT, Fredrickson BL have more output, while authors such as Seligman MEP, Diener, ED, Fredrickson BL, Peterson C, Snyder CR have been cited more and have a greater impact. Analysis of the cited literature allows us to understand the research base and research frontier in this field. First, jointly cited literature forms the research base for a research field. Documents such as Bolier et al. [ 2013], Seligman et al. [ 2005], Seligman and Csikszentmihalyi [2000], Sin and Lyubomirsky [2009], and Peterson [2004] can be found to be highly explosive in nature, indicating that these papers are the foundation for the development of this field, to which we must pay close attention. Second, through cluster analysis of co-citations, we can find research hotspots and development trends in this field. The systematic review, character strengths, positive psychology intervention, language enjoyment, and the COVID-19 pandemic are the foci of research and developmental trends in this field that need our high attention. ## 6. Strengths and limitations This is the first large-scale data analysis of positive psychology papers utilizing CiteSpace and VOSviewer software. Furthermore, our findings offer a clear visual analysis, and so forth, of positive psychology publications. In addition, the co-citation analysis can also capture the research base and hot trends in this field, providing a reference for researchers to fully understand this field. This study used the scientometric method for literature analysis, which has objectivity but also some limitations. First, the results of the software analysis are somewhat mechanical and require us to select meaningful results. At the same time, there is a possibility of ignoring some meaningful literature. For example, in considering the role of positive psychology in global issues, some researchers suggest that positive psychology may benefit from the integration of spirituality to better support people’s well-being (Bellehumeur et al., 2022); others have found that positive psychology has a considerable impact on employees’ green behavior (Meyers and Rutjens, 2022). Second, we did not conduct an in-depth assessment of the literature, only those in WoSCC; other databases, such as Scopus, MEDLINE, and PubMed, are available. We analyzed the type of literature and selected only papers and reviews, ignoring other types of literature. We also analyzed only English literature and ignored literature in other languages, which may have led to a lack of attention to other cultures. Related to this is the need to recognize the parochial nature of positive psychology, which seems to be US-centric, especially in terms of leadership (Ryff, 2022). All of these may have biased the data. For example, in studies across cultures, Appiah et al. [ 2020] found that positive psychology interventions promote mental health among rural Ghanaian adults; a study in Hong Kong, China, found that a multifaceted positive psychology program was effective in reducing adolescent anxiety and increasing subjective well-being (Kwok et al., 2022). Finally, potential bias in the data may be caused by duplicate author names of authors, or the use of different names by the same author; or by irregularities in literature citation, where different authors cite the same literature in different formats in the analysis of co-cited literature. This is where the practice of some researchers is worthy of consideration; for example, Donaldson et al. [ 2014] coded all articles in their review by raters using a systematic coding scheme. ## Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. ## Author contributions FW and JG contributed to the study design, acquisition of research data, and drafted the manuscript. FW conducted the data analysis. GY contributed to critical revising of the manuscript. All authors contributed to the article and approved the submitted version. ## Funding This study was supported by the National Social Science Fund Project of China (No. 19XSH018). ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg.2023.1101157/full#supplementary-material ## References 1. Appiah R., Wilson-Fadiji A., Schutte L., Wissing M. P.. Effects of a community-based multicomponent positive psychology intervention on mental health of rural adults in Ghana. Appl. Psychol. Health Well Being (2020) 12 828-862. DOI: 10.1111/aphw.12212 2. Baminiwatta A., Solangaarachchi I.. Trends and developments in mindfulness research over 55 years: A bibliometric analysis of publications indexed in web of science. Mindfulness (2021) 12 2099-2116. DOI: 10.1007/s12671-021-01681-x 3. Bellehumeur C. R., Bilodeau C., Kam C.. Integrating positive psychology and spirituality in the context of climate change. Front. Psychol. (2022) 13 970362. DOI: 10.3389/fpsyg.2022.970362 4. Bolier L., Haverman M., Westerhof G. J., Riper H., Smit F., Bohlmeijer E.. Positive psychology interventions: a meta-analysis of randomized controlled studies. BMC Public Health (2013) 13 119. DOI: 10.1186/1471-2458-13-119 5. Cameron K. S., Bright D., Caza A.. Exploring the relationships between organizational virtuousness and performance. Am. Behav. Sci. (2004) 47 766-790. DOI: 10.1177/0002764203260209 6. Carr A., Cullen K., Keeney C., Canning C., Mooney O., Chinseallaigh E.. Effectiveness of positive psychology interventions: a systematic review and meta-analysis. J. Posit. Psychol. (2020) 16 749-769. DOI: 10.1080/17439760.2020.1818807 7. Chen C. M.. CiteSpace II: detecting and visualizing emerging trends and transient patterns in scientific literature. J. Am. Soc. Inf. Sci. Technol. (2006) 57 359-377. DOI: 10.1002/asi.20317 8. Chen C. M.. Predictive effects of structural variation on citation counts. J. Am. Soc. Inf. Sci. Technol. (2012) 63 431-449. DOI: 10.1002/asi.21694 9. Chen C. M.. Science mapping: A systematic review of the literature. J. Data Inform. Sci. (2017) 2 1-40. DOI: 10.1515/jdis-2017-0006 10. Chen C.. A glimpse of the first eight months of the COVID-19 literature on Microsoft academic graph: themes, citation contexts, and uncertainties. Front. Res. Metr. Anal. (2020) 5 607286. DOI: 10.3389/frma.2020.607286 11. Chen C. M., Dubin R., Kim M. C.. Emerging trends and new developments in regenerative medicine: a scientometric update (2000 - 2014). Expert. Opin. Biol. Ther. (2014a) 14 1295-1317. DOI: 10.1517/14712598.2014.920813 12. Chen C. M., Dubin R., Kim M. C.. Orphan drugs and rare diseases: a scientometric review (2000-2014). Expert Opin. Orphan Drugs (2014b) 2 709-724. DOI: 10.1517/21678707.2014.920251 13. Chen C. M., Hu Z. G., Liu S. B., Tseng H.. Emerging trends in regenerative medicine: a scientometric analysis in CiteSpace. Expert. Opin. Biol. Ther. (2012) 12 593-608. DOI: 10.1517/14712598.2012.674507 14. Chen C. M., Ibekwe-Sanjuan F., Hou J. H.. The structure and dynamics of Cocitation clusters: A multiple-perspective Cocitation analysis. J. Am. Soc. Inf. Sci. Technol. (2010) 61 1386-1409. DOI: 10.1002/asi.21309 15. Cobo M. J., López-Herrera A. G., Herrera-Viedma E., Herrera F.. Science mapping software tools: review, analysis, and cooperative study among tools. J. Am. Soc. Inf. Sci. Technol. (2011) 62 1382-1402. DOI: 10.1002/asi.21525 16. Dewaele J. M.. The two faces of Janus? Anxiety and enjoyment in the foreign language classroom. SSLLT (2014) 4 237-274. DOI: 10.14746/ssllt.2014.4.2.5 17. Diener E.. Subjective well-being. Psychol. Bull. (1984) 95 542-575. DOI: 10.1037/0033-2909.95.3.542 18. Diener E., Emmons R. A., Larsen R. J., Griffin S.. The satisfaction with life scale. J. Pers. Assess. (1985) 49 71-75. DOI: 10.1207/s15327752jpa4901_13 19. Diener E., Seligman M. E. P.. Beyond money: Progress on an economy of well-being. Perspect. Psychol. Sci. (2018) 13 171-175. DOI: 10.1177/1745691616689467 20. Ding X., Yang Z.. Knowledge mapping of platform research: a visual analysis using VOSviewer and CiteSpace. Electron. Commer. Res. (2020) 22 787-809. DOI: 10.1007/s10660-020-09410-7 21. Donaldson S. I., Dollwet M., Rao M. A.. Happiness, excellence, and optimal human functioning revisited: examining the peer-reviewed literature linked to positive psychology. J. Posit. Psychol. (2014) 10 185-195. DOI: 10.1080/17439760.2014.943801 22. Elahi Shirvan M., Taherian T., Shahnama M., Yazdanmehr E.. A longitudinal study of foreign language enjoyment and L2 grit: A latent growth curve modeling. Front. Psychol. (2021) 12 720326. DOI: 10.3389/fpsyg.2021.720326 23. Fredrickson B. L.. The role of positive emotions in positive psychology. The broaden-and-build theory of positive emotions. Am. Psychol. (2001) 56 218-226. DOI: 10.1037/0003-066X.56.3.218 24. Gallagher Tuleya L.. Thesaurus of Psychological Index Terms (2007) 25. Gana K., Bailly N., Saada Y., Joulain M., Trouillet R., Herve C.. Relationship between life satisfaction and physical health in older adults: A longitudinal test of cross-lagged and simultaneous effects. Health Psychol. (2013) 32 896-904. DOI: 10.1037/a0031656 26. Gander F., Proyer R. T., Ruch W., Wyss T.. Strength-based positive interventions: further evidence for their potential in enhancing well-being and alleviating depression. J. Happiness Stud. (2012) 14 1241-1259. DOI: 10.1007/S10902-012-9380-0 27. Gillham J. E., Seligman M. E.. Footsteps on the road to a positive psychology. Behav. Res. Ther. (1999) 37 S163-S173. DOI: 10.1016/S0005-7967(99)00055-8 28. Guo Y.. Exploring the dynamic interplay between foreign language enjoyment and learner engagement with regard to EFL achievement and absenteeism: A sequential mixed methods study. Front. Psychol. (2021) 12 766058. DOI: 10.3389/fpsyg.2021.766058 29. Hausler M., Strecker C., Huber A., Brenner M., Hoge T., Hofer S.. Distinguishing relational aspects of character strengths with subjective and psychological well-being. Front. Psychol. (2017) 8 1159. DOI: 10.3389/fpsyg.2017.01159 30. Hayes A. F.. Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach (2017) 31. Hendriks T., Schotanus-Dijkstra M., Hassankhan A., De Jong J., Bohlmeijer E.. The efficacy of multi-component positive psychology interventions: A systematic review and meta-analysis of randomized controlled trials. J. Happiness Stud. (2020) 21 357-390. DOI: 10.1007/s10902-019-00082-1 32. Hone L. C., Jarden A., Schofield G. M.. An evaluation of positive psychology intervention effectiveness trials using the re-aim framework: A practice-friendly review. J. Posit. Psychol. (2014) 10 303-322. DOI: 10.1080/17439760.2014.965267 33. Hou J. H., Yang X. C., Chen C. M.. Emerging trends and new developments in information science: a document co-citation analysis (2009-2016). Scientometrics (2018) 115 869-892. DOI: 10.1007/s11192-018-2695-9 34. Huertas González-Serrano M., Jones P., Llanos-Contrera O.. An overview of sport entrepreneurship field: a bibliometric analysis of the articles published in the web of science. Sport Society (2019) 23 296-314. DOI: 10.1080/17430437.2019.1607307 35. Job S. A., Williams S. L.. Translating online positive psychology interventions to sexual and gender minorities: A systematic review. Psychol. Sex. Orientat. Gend. Divers. (2020) 7 455-503. DOI: 10.1037/sgd0000365 36. Keyes C. L. M.. Mental illness and/or mental health? Investigating axioms of the complete state model of health. J. Consult. Clin. Psychol. (2005) 73 539-548. DOI: 10.1037/0022-006X.73.3.539 37. Kruk M.. Dynamicity of perceived willingness to communicate, motivation, boredom and anxiety in second life: the case of two advanced learners of English. Comput. Assist. Lang. Learn. (2019) 35 190-216. DOI: 10.1080/09588221.2019.1677722 38. Kwok S. Y. C. L., Gu M., Tam N. W. Y.. A multiple component positive psychology intervention to reduce anxiety and increase happiness in adolescents: the mediating roles of gratitude and emotional intelligence. J. Happiness Stud. (2022) 23 2039-2058. DOI: 10.1007/s10902-021-00487-x 39. Lake J.. Positive Psychology and Second Language Motivation: Empirically Validating a Model of Positive L2 self (2015) 40. Lester P. B., Stewart E. P., Vie L. L., Bonett D. G., Seligman M. E. P., Diener E.. Happy soldiers are highest performers. J. Happiness Stud. (2022) 23 1099-1120. DOI: 10.1007/s10902-021-00441-x 41. Li C., Xu J.. Trait emotional intelligence and classroom emotions: A positive psychology investigation and intervention among Chinese EFL learners. Front. Psychol. (2019) 10 2453. DOI: 10.3389/fpsyg.2019.02453 42. Liao H., Tang M., Luo L., Li C., Chiclana F., Zeng X.-J.. A bibliometric analysis and visualization of medical big data research. Sustainability (2018) 10 166. DOI: 10.3390/su10010166 43. Liu S. B., Chen C. M.. The proximity of co-citation. Scientometrics (2012) 91 495-511. DOI: 10.1007/s11192-011-0575-7 44. Luo H. F., Cai Z. L., Huang Y. Y., Song J. T., Ma Q., Yang X. W.. Study on pain catastrophizing from 2010 to 2020: A bibliometric analysis via CiteSpace. Front. Psychol. (2021) 12 12. DOI: 10.3389/fpsyg.2021.759347 45. Luo M., Guo L., Yu M., Jiang W., Wang H.. The psychological and mental impact of coronavirus disease 2019 (COVID-19) on medical staff and general public - A systematic review and meta-analysis. Psychiatry Res. (2020) 291 113190. DOI: 10.1016/j.psychres.2020.113190 46. Luthans F., Avolio B. J.. The “point” of positive organizational behavior. J. Organ. Behav. (2009) 30 291-307. DOI: 10.1002/job.589 47. Lyubomirsky S., Dickerhoof R., Boehm J. K., Sheldon K. M.. Becoming happier takes both a will and a proper way: An experimental longitudinal intervention to boost well-being. Emotion (2011) 11 391-402. DOI: 10.1037/a0022575 48. Lyubomirsky S., Layous K.. How do simple positive activities increase well-being?. Curr. Dir. Psychol. Sci. (2013) 22 57-62. DOI: 10.1177/0963721412469809 49. Magyar-Moe J. L., Owens R. L., Conoley C. W.. Positive psychological interventions in counseling: What every counseling psychologist should know. Couns. Psychol. (2015) 43 508-557. DOI: 10.1177/0011000015573776 50. Mayerson N. H.. The character strengths response: an urgent call to action. Front. Psychol. (2020) 11 2106. DOI: 10.3389/fpsyg.2020.02106 51. Meyers M. C., Rutjens D.. Applying a positive (organizational) psychology lens to the study of employee green behavior: A systematic review and research agenda. Front. Psychol. (2022) 13 840796. DOI: 10.3389/fpsyg.2022.840796 52. Miglianico M., Dubreuil P., Miquelon P., Bakker A. B., Martin-Krumm C.. Strength use in the workplace: A literature review. J. Happiness Stud. (2019) 21 737-764. DOI: 10.1007/s10902-019-00095-w 53. Moskowitz J. T., Cheung E. O., Freedman M., Fernando C., Zhang M. W., Huffman J. C.. Measuring positive emotion outcomes in positive psychology interventions: A literature review. Emot. Rev. (2020) 13 60-73. DOI: 10.1177/1754073920950811 54. Neumeier L. M., Brook L., Ditchburn G., Sckopke P.. Delivering your daily dose of well-being to the workplace: a randomized controlled trial of an online well-being programme for employees. Eur. J. Work Organ. Psy. (2017) 26 555-573. DOI: 10.1080/1359432X.2017.1320281 55. Perianes-Rodriguez A., Waltman L., Van Eck N. J.. Constructing bibliometric networks: A comparison between full and fractional counting. J. Informet. (2016) 10 1178-1195. DOI: 10.1016/j.joi.2016.10.006 56. Peterson C.. Character Strengths and Virtues: A Handbook and Classification (2004) 57. Proyer R. T., Gander F., Wellenzohn S., Ruch W.. Positive psychology interventions in people aged 50-79 years: long-term effects of placebo-controlled online interventions on well-being and depression. Aging Ment. Health (2014) 18 997-1005. DOI: 10.1080/13607863.2014.899978 58. Quiroga-Garza A., Cepeda-Lopez A. C., Villarreal Zambrano S., Villalobos-Daniel V. E., Carreno D. F., Eisenbeck N.. How having a clear why can help us cope with almost anything: meaningful well-being and the COVID-19 pandemic in Mexico. Front. Psychol. (2021) 12 648069. DOI: 10.3389/fpsyg.2021.648069 59. Rashid T.. Positive psychotherapy: A strength-based approach. J. Posit. Psychol. (2014) 10 25-40. DOI: 10.1080/17439760.2014.920411 60. Ruhl H. A., Priede I. G.. Open up monitoring of deep-sea drilling. Nature (2011) 473 154. DOI: 10.1038/473154b 61. Ryff C. D.. Happiness is everything, or is it-explorations on the meaning of psychological well-being. J. Pers. Soc. Psychol. (1989) 57 1069-1081. DOI: 10.1037/0022-3514.57.6.1069 62. Ryff C. D.. Psychological well-being revisited: advances in the science and practice of eudaimonia. Psychother. Psychosom. (2014) 83 10-28. DOI: 10.1159/000353263 63. Ryff C. D.. Positive psychology: looking back and looking forward. Front. Psychol. (2022) 13 840062. DOI: 10.3389/fpsyg.2022.840062 64. Schueller S. M., Parks A. C.. The science of self-help translating positive psychology research into increased individual happiness. Eur. Psychol. (2014) 19 145-155. DOI: 10.1027/1016-9040/a000181 65. Schui G., Krampen G.. Bibliometric analyses on the emergence and present growth of positive psychology. Appl. Psychol. Health Well Being (2010) 2 52-64. DOI: 10.1111/j.1758-0854.2009.01022.x 66. Seligman M. E. P.. Flourish (2011) 67. Snow N. E.. Positive psychology, the classification of character strengths and virtues, and issues of measurement. J. Posit Psychol. (2018) 14 20-31. DOI: 10.1080/17439760.2018.1528376 68. Seligman M. E. P., Widiger T., Cannon T. D.. Positive psychology: A personal history. Annual Review of Clinical Psychology (2019) Vol. 15 1-23 69. Seligman M. E., Csikszentmihalyi M.. Positive psychology. An introduction. Am. Psychol. (2000) 55 5-14. DOI: 10.1037/0003-066X.55.1.5 70. Seligman M. E. P., Ernst R. M., Gillham J., Reivich K., Linkins M.. Positive education: positive psychology and classroom interventions. Oxf. Rev. Educ. (2009) 35 293-311. DOI: 10.1080/03054980902934563 71. Seligman M. E. P., Rashid T., Parks A. C.. Positive psychotherapy. Am. Psychol. (2006) 61 774-788. DOI: 10.1037/0003-066X.61.8.774 72. Seligman M. E., Steen T. A., Park N., Peterson C.. Positive psychology progress: empirical validation of interventions. Am. Psychol. (2005) 60 410-421. DOI: 10.1037/0003-066X.60.5.410 73. Shafique M.. Thinking inside the box? Intellectual structure of the knowledge base of innovation research (1988-2008). Strateg. Manag. J. (2013) 34 62-93. DOI: 10.1002/smj.2002 74. Sheldon K. M., King L.. Why positive psychology is necessary. Am. Psychol. (2001) 56 216-217. DOI: 10.1037/0003-066X.56.3.216 75. Shi Y. M., Luo J. M., Wang X. Q., Zhang Y. Q., Zhu H., Su D. S.. Emerging trends on the correlation between neurotransmitters and tumor progression in the last 20 years: A bibliometric analysis via CiteSpace. Front. Oncol. (2022) 12 800499. DOI: 10.3389/fonc.2022.800499 76. Sin N. L., Lyubomirsky S.. Enhancing well-being and alleviating depressive symptoms with positive psychology interventions: a practice-friendly meta-analysis. J. Clin. Psychol. (2009) 65 467-487. DOI: 10.1002/jclp.20593 77. Strecker C., Huber A., Hoge T., Hausler M., Hofer S.. Identifying thriving workplaces in hospitals: work characteristics and the applicability of character strengths at work. Appl. Res. Qual. Life (2020) 15 437-461. DOI: 10.1007/s11482-018-9693-1 78. Suldo S. M., Shaffer E. J.. Looking beyond psychopathology: the dual-factor model of mental health in youth. Sch. Psychol. Rev. (2008) 37 52-68. DOI: 10.1080/02796015.2008.12087908 79. Van Eck N. J., Waltman L.. Software survey: VOSviewer, a computer program for bibliometric mapping. Scientometrics (2010) 84 523-538. DOI: 10.1007/s11192-009-0146-3 80. Wang Y. L., Derakhshan A., Zhang L. J.. Researching and practicing positive psychology in second/foreign language learning and teaching: the past, current status and future directions. Front. Psychol. (2021) 12 10. DOI: 10.3389/fpsyg.2021.731721 81. Wang X., Wang Y., Yang Y., Wang L.. Investigating Chinese university students’ enjoyment in a web-based language learning environment: validation of the online foreign language enjoyment scale. Percept. Mot. Skills (2021) 128 2820-2848. DOI: 10.1177/00315125211041714 82. Waters L., Algoe S. B., Dutton J., Emmons R., Fredrickson B. L., Heaphy E.. Positive psychology in a pandemic: buffering, bolstering, and building mental health. J. Posit. Psychol. (2021) 17 303-323. DOI: 10.1080/17439760.2021.1871945 83. Wissing M. P., Khumalo I. P., Chigeza S. C.. Meaning as perceived and experienced by an African student group. J. Psychol. Afr. (2014) 24 92-101. DOI: 10.1080/14330237.2014.904101 84. Wood A. M., Froh J. J., Geraghty A. W. A.. Gratitude and well-being: A review and theoretical integration. Clin. Psychol. Rev. (2010) 30 890-905. DOI: 10.1016/j.cpr.2010.03.005 85. You Y. W., Wang D. Z., Wang Y. N., Li Z. P., Ma X. D.. A Bird's-eye view of exercise intervention in treating depression among teenagers in the last 20 years: A bibliometric study and visualization analysis. Front. Psych. (2021) 12 17. DOI: 10.3389/fpsyt.2021.661108 86. Yu Y., Li Y., Zhang Z., Gu Z., Zhong H., Zha Q.. A bibliometric analysis using VOSviewer of publications on COVID-19. Ann. Transl. Med. (2020) 8 816. DOI: 10.21037/atm-20-4235 87. Zhou T., Qu J. L., Sun H. P., Xue M. X., Shen Y. J., Liu Y. B.. Research trends and hotspots on Montessori intervention in patients with dementia from 2000 to 2021: A bibliometric analysis. Front. Psych. (2021) 12 737270. DOI: 10.3389/fpsyt.2021.737270 88. Zhou Y., Zhu X. P., Shi J. J., Yuan G. Z., Yao Z. A., Chu Y. G.. Coronary heart disease and depression or anxiety: A bibliometric analysis. Front. Psychol. (2021) 12 669000. DOI: 10.3389/fpsyg.2021.669000 89. Zou L. X., Sun L.. Global diabetic kidney disease research from 2000 to 2017: A bibliometric analysis. Medicine (Baltimore) (2019) 98 e14394. DOI: 10.1097/MD.0000000000014394
--- title: Prevalence and associated factors of incident and persistent loneliness among middle-aged and older adults in Thailand authors: - Supa Pengpid - Karl Peltzer journal: BMC Psychology year: 2023 pmcid: PMC10015912 doi: 10.1186/s40359-023-01115-4 license: CC BY 4.0 --- # Prevalence and associated factors of incident and persistent loneliness among middle-aged and older adults in Thailand ## Abstract ### Background The aim of the study was to assess the prevalence and associated factors of incident and persistent loneliness in a prospective cohort study among middle-aged and older adults (≥ 45 years) in Thailand. ### Methods Longitudinal data from the Health, Aging, and Retirement in Thailand (HART) study in 2015 and 2017 were analysed. Loneliness was assessed with one item from the Center for Epidemiological Studies Depression scale. Logistic regression was used to calculate predictors of incident and persistent loneliness. ### Results In total, at baseline $21.7\%$ had loneliness, 633 of 3696 participants without loneliness in 2015 had incident loneliness in 2017 ($22.2\%$), and 239 of 790 adults had persistent loneliness (in both 2015 and 2017) ($30.3\%$). In adjusted logistic regression analysis, low income (aOR: 1.27, $95\%$ CI: 1.03 to 1.57), poor self-rated physical health status (aOR: 1.64, $95\%$ CI: 1.27 to 2.12), hypertension (aOR: 1.34, $95\%$ CI: 1.09 to 1.65), depressive symptoms (aOR: 1.97, $95\%$ CI: 1.11 to 3.49), and having three or chronic conditions (aOR: 1.76, $95\%$ CI: 1.19 to 2.60) were positively associated and a higher education (aOR: 0.74, $95\%$ CI: 0.55 to 0.98) and living in the southern region of Thailand (aOR: 0.43, $95\%$ CI: 0.30 to 0.61) were inversely associated with incident loneliness. Poor self-rated physical health status (aOR: 1.91, $95\%$ CI: 1.26 to 2.88), and having three or more chronic diseases (aOR: 1.78, $95\%$ CI: 1.07 to 2.98), were positively associated, and living in the southern region (aOR: 0.40, $95\%$ CI: 0.25 to 0.65) was inversely associated with persistent loneliness. ### Conclusion More than one in five ageing adults had incident loneliness in 2 years of follow-up. The prevalence of incident and/or persistent loneliness was higher in people with a lower socioeconomic status, residing in the central region, poor self-rated physical health status, depressive symptoms, hypertension, and a higher number of chronic diseases. ## Introduction Loneliness may interfere with the quantity and quality of one’s social relationships [1] and may be common among older adults as their social relations may decline [2, 3]. Loneliness affects negatively physical and mental health in old age, including mortality [4–6]. The prevalence of loneliness among older adults in high-income countries was $28.5\%$ [7]. Among middle-aged and older adults in low- and middle-income countries, the prevalence of loneliness was $33.8\%$ in India [8], in South Africa $9.9\%$ [9], $10\%$ in Indonesia [10], in Malaysia $32.5\%$ [11], in Mexico $12.3\%$ [12], and in Myanmar ($31.7\%$) [14]. In local studies in Thailand in the Ko Rai Subdistrict, Chachoengsao, $24\%$ of older adults reported loneliness [13], among older Adults in Chon Buri Province, a low level of loneliness was found [15], and in a qualitative study in Thailand contributing factors to loneliness included life transitions, socio-demographic factors, socio-economic and environmental factors, poor health status and health resources [16]. The prevalence of incident loneliness (4 years) among middle-aged and older adults in Indonesia was $15.1\%$ [17], and in the English Longitudinal Study of Ageing (ELSA) $38.4\%$ (length of follow-up 6.86 years), and the Health and Retirement Study (HRS) in USA $30.2\%$ (length of follow-up 8.18 years) [18]. Considering that previous local studies on loneliness in Thailand were cross-sectional, the prevalence of incident and persistent loneliness symptoms among the ageing population in Thailand is unclear, as well as the prospective relationships between baseline indicators and incident and persistent loneliness. A greater understanding of the prevalence and associated factors of incident and persistent loneliness may help in better identifying and treating modifiable risk factors in people with loneliness. In a systematic review of longitudinal studies in high-income countries among older adults found the following risk factors for loneliness, not married, low social activity, poor self-rated health; and depression [19]. In a systematic review of mainly cross-sectional studies among older adults, female sex, older age, lower socioeconomic status, non-married status, living alone, functional disability, poor self-report physical and mental health, poor cognitive functioning, and adverse life events were associated with loneliness [20]. There is a lack of longitudinal studies in Southeast Asia investigating the determinants of incident and persistent loneliness. To address this research gap, our objective was to investigate the prevalence and associated factors of incident and persistent loneliness in a prospective cohort study among ageing adults (≥ 45 years) in Thailand. ## Sample and procedure Secondary longitudinal data from the Health, Aging, and Retirement in Thailand (HART) study in 2015 and 2017 were analysed. In a national multi-stage sampling (regions, provinces, blocks or villages, households) one individual (≥ 45 years) was randomly selected per household [21, 22]. The 2015 study ($$n = 5$$,616), and the 2017 study ($$n = 3$$,708) had a $72.3\%$ response rate and $66.0\%$ the retention rate; reasons for loss to follow-up were had moved away ($$n = 1$$,554), declined ($$n = 270$$) and died ($$n = 192$$). Participants were interviewed with a structured questionnaire in 2015 and with computer-assisted personal interview (CAPI) in 2017. The “Ethics Committee in Human Research, National Institute of Development Administration – ECNIDA (ECNIDA $\frac{2020}{00012}$)” granted approval, and participants gave written informed consent. ## Outcome variable Loneliness was measured with one item from the “Center for Epidemiologic Studies Depression (CES-D-10) scale,” [23], “In the past week, how often did you experience feeling lonely?” defined as “almost always (5–7 days), often (3–4 days) or sometimes (1–2 days)”=1 and “very rarely (less than one day) or none”=0. Single-item measures of loneliness have shown adequate reliability and have a high correlation with multi-item measures [24–26]. ## Covariates Sociodemographic variables included marital status, highest level of education (no formal education illiterate/ no formal education literate, or elementary school, coded as “≤elementary school”, and middle school, high school, vocational diploma, Bachelor degree, or higher than Bachelor degree, coded as “>elementary school”), sex, age, region, religion, and personal annual income. The personal annual income was calculated based from “employment, own business, agricultural/livestock/fishing business, short-term or contract work, financial support from family, remuneration/pension income from the government fund, occupational pension fund, private pension fund, social security/welfare income, income from government living allowance, veteran’s welfare benefit, other welfare assistance income, and income from other sources, and classified low = less than 50,000 Baht, and high = 50,000 or more Thai Baht (Average exchange rate in 2015: 1 US = 34.2 Baht)” [22]. Social participation (at least one social activity in the past month) was sourced from 6 items [22, 27]. Substance use included alcohol use and smoking (never, past, or current). “ Have you ever smoked cigarettes?” ( response options: “1 = yes, and still smoke now, 2 = yes, but quit smoking, and 3 = never”); “Have you ever drunk alcoholic beverages such as liquor, beer, or wine?” ( response options: “1 = yes, and still drinking now, 2 = yes, but do not drink now, and 3 = never)”. Physical activity was classified as 0–149 min/week exercise, and ≥ 150 min/week exercise [28, 29]. Body Mass Index (BMI), based on self-reported height and weight: “underweight (< 18.5 kg/m2), normal weight (18.5–22.9 kg/m2), overweight (23–24.9 kg/m2), and obesity (25 + kg/m2).” [ 30]. Activities of daily living (ADL) disability was defined as the inability to do any of the four ADL (eating, bathing, dressing, and washing) [31]. History of accidents (injury) was assessed with the question, “In the last 2 years, were you involved in an accident that affected your physical health?” ( Yes/No). Fear of falling (FOF) with activity avoidance was defined as “so worried about falling down that I have refrained from doing certain activities.” The self-rated physical health status (ranging from 0 = very poor to 100 excellent) was defined as 0–50 low and 60–100 high. Depressive symptoms (≥ 10 scores) was measured with the “Center for Epidemiologic Studies Depression (CES-D-10) scale,” [23], excluding the loneliness item [32, 33]; (Cronbach’s alpha 0.76). Chronic diseases were based on self-reported conditions that had been diagnosed by a health care professional, including: “1) hypertension, 2) diabetes, 3) vascular diseases, heart disease or heart failure, 4) rheumatism or arthritis, 5) bone diseases, low bone density or osteoporosis, 6) kidney diseases, 7) lung diseases/emphysema, 8) cancer, 9) liver diseases, 10) brain diseases/Alzheimer’s disease 11) visual impairment and 12) hearing impairment.” The 12 chronic diseases were classified into nine groups: [1] cardiovascular disease, heart disease, heart failure, [2] hypertension, [3] endocrine (diabetes), [4] musculoskeletal (bone diseases, rheumatism, low bone density, arthritis, and osteoporosis), [5] liver or kidney diseases, [6] respiratory (lung diseases/emphysema), [7] Cancer, [8] sensorial (visual impairment and/or hearing impairment), and [9] neurological (brain diseases/Alzheimer’s disease) [34]. ## Statistical analysis Frequencies and percentages of incident and persistent loneliness were calculated. The first longitudinal logistic regression model estimated incident loneliness in 2017, excluding those with loneliness in 2015, and the second model estimated persistent loneliness (in both 2015 and 2017). Models were adjusted by chronic diseases, sociodemographic factors, lifestyle factors, social participation, depressive symptoms, and adverse life events; confounders were included based on literature review [19, 20]. Variables found significant at < 0.05 in univariate analyses were included in the multivariable models; $p \leq 0.05$ was considered statistically significant. Missing data were discarded. Statistical analyses were conducted with StataSE 15.0 (College Station, TX, USA). ## Sample characteristics In total, at baseline $21.7\%$ had loneliness, 633 of 3696 participants without loneliness in 2015 had incident loneliness in 2017 ($22.2\%$), and 239 of 790 adults had persistent loneliness (in both 2015 and 2017) ($30.3\%$). In addition, the details of the sample are shown in Table 1. Table 1Analytic sample characteristics by incident and persistent loneliness, Thailand, 2015–2017Baseline variablesSubcategoriesBaseline sampleIncident lonelinessPersistent lonelinessTotal N36962855790N (%)N (%)N (%)All633 (22.2)239 (30.3)Age (in years)45–64≥ 65 or more1685 (45.6)2011 (54.4)246 (18.1)387 (25.9)90 (29.8)149 (30.5)SexFemaleMale1971 (53.3)1725 (46.7)356 (23.8)277 (20.3)154 (34.3)85 (24.9)Education≤Elementary>Elementary3100 (84.1)588 (15.9)555 (23.4)75 (15.8)213 (31.2)26 (24.5)Annual incomeHighLow1875 (50.7)1821 (49.3)286 (18.4)347 (26.6)86 (28.0)153 (31.7)Marital statusMarried/cohabitingDivorced/sep./never marriedWidowed2194 (60.1)352 (9.6)1105 (30.3)352 (19.8)56 (22.3)213 (26.8)105 (27.2)27 (27.8)103 (34.7)ReligionMuslim or otherBuddhist304 (8.2)3390 (91.8)36 (19.4)597 (22.4)28 (24.8)211 (31.2)RegionBangkokCentral (excl. Bangkok)NorthNortheastSouth287 (7.8)899 (24.3)1126 (30.5)659 (17.8)725 (19.6)66 (28.7)170 (23.1)207 (23.4)128 (23.1)62 (13.7)21 (38.2)66 (42.0)71 (31.8)33 (35.1)48 (18.4)Social participationNoYes241 (6.5)3450 (93.5)46 (24.0)587 (22.0)16 (34.0)223 (30.1)Alcohol useNeverPastCurrent2988 (80.8)266 (7.2)442 (12.0)523 (22.8)47 (23.6)63 (17.5)205 (31.7)17 (25.8)17 (22.1)Smoking tobacco useNeverPastCurrent293 (79.9)292 (7.9)451 (12.2)515 (22.7)45 (20.2)73 (20.2)199 (31.1)19 (28.8)21 (25.0)Physical activity≥ 150 min/week< 150 min/week605 (16.4)3091 (83.6)100 (20.2)533 (22.6)38 (36.5)201 (29.3)Body mass indexNormalUnderOverweightObesity1247 (37.6)354 (10.7)661 (19.9)1057 (31.8)226 (23.1)70 (27.1)112 (20.6)171 (20.6)81 (31.9)29 (31.5)23 (21.1)73 (33.2)Activities of Daily Living disabilityNoYes3522 (96.9)112 (3.1)606 (21.8)18 (37.5)218 (30.4)19 (29.7)AccidentNoYes3249 (87.9)447 (12.1)544 (21.6)89 (26.4)201 (29.5)38 (35.2)Fall worry refrain from social activitiesNoYes3425 (92.7)271 (7.3)587 (21.7)46 (30.7)203 (30.3)36 (29.8)Self-rated physical health statusHighLow3024 (83.4)600 (16.6)489 (20.3)137 (33.8)162 (27.7)74 (40.7)Depressive symptomsNoYes3118 (91.6)285 (8.4)576 (22.0)22 (40.0)147 (29.6)71 (31.0) Chronic conditions Cardiovascular diseaseNoYes3499 (94.7)197 (5.3)590 (21.8)43 (29.1)219 (29.5)20 (42.6)HypertensionNoYes2379 (64.4)1317 (35.6)358 (19.1)275 (28.0)136 (29.2)103 (31.8)Endocrine (diabetes)NoYes3139 (84.9)557 (15.1)527 (21.6)106 (25.5)192 (29.4)47 (34.6)MusculoskeletalNoYes3434 (92.9)262 (7.1)573 (21.4)60 (33.1)211 (29.6)28 (35.9)Liver or kidney diseasesNoYes3608 (97.6)88 (2.4)620 (22.2)13 (21.7)232 (30.4)7 (25.9)RespiratoryNoYes3660 (99.0)36 (1.0)627 (22.2)6 (22.2)236 (30.2)3 (33.3)CancerNoYes3664 (99.1)32 (0.9)626 (22.1)7 (31.8)235 (30.1)4 (40.0)SensorialNoYes3129 (84.7)567 (15.3)530 (21.5)103 (26.5)189 (28.9)59 (35.3)NeurologicalNoYes3665 (99.2)31 (0.8)626 (22.1)7 (43.8)234 (30.2)5 (35.7) Number of chronic conditions Chronic conditions0123 or more1716 (46.4)1101 (29.8)600 (16.2)279 (7.5)246 (17.8)205 (24.3)122 (27.5)60 (33.0)80 (26.4)76 (31.3)45 (30.2)38 (40.0)Sep.= separated ## Associations with incident loneliness In adjusted logistic regression analysis, low income (aOR: 1.27, $95\%$ CI: 1.03 to 1.57), poor self-rated physical health status (aOR: 1.64, $95\%$ CI: 1.27 to 2.12), depressive symptoms (aOR: 1.97, $95\%$ CI: 1.11 to 3.49), hypertension (aOR: 1.34, $95\%$ CI: 1.09 to 1.65) and having three or chronic conditions (aOR: 1.76, $95\%$ CI: 1.19 to 2.60) were positively associated and a higher education (aOR: 0.74, $95\%$ CI: 0.55 to 0.98) and living in the southern region of Thailand (aOR: 0.43, $95\%$ CI: 0.30 to 0.61) were inversely associated with incident loneliness. Furthermore, in the unadjusted analysis, older age, widowed, ADL disability, having had an accident in the past two years, fear of falling with refraining from social activities, cardiovascular disease, musculoskeletal problem, sensory and neurological conditions were positively associated with incident loneliness, while male sex, social participation, and current alcohol use were negatively associated with incident loneliness (see Table 2). Table 2Association between independent variables and incident loneliness, HART (2015–2017)Baseline variablesSubcategoryCOR ($95\%$ CI)AOR ($95\%$ CI)Age (in years)45–64≥ 65 or more1 (Reference)1.58 (1.32 to 1.89)**1 (Reference)1.16 (0.93 to 1.45)SexFemaleMale1 (Reference)0.82 (0.68 to 0.97)1 (Reference)0.98 (0.79 to 1.23)Education≤Elementary>Elementary1 (Reference)0.62 (0.47 to 0.80)**1 (Reference)0.74 (0.55 to 0.98)Annual incomeHighLow1 (Reference)1.61 (1.35 to 1.92)**1 (Reference)1.27 (1.03 to 1.57)Marital statusMarried/cohabitingDivorced/sep./never marriedWidowed1 (Reference)1.16 (0.84 to 1.60)1.48 (1.21 to 1.80)*1 (Reference)1.07 (0.75 to 1.51)1.13 (0.89 to 1.43)ReligionMuslim or otherBuddhist1 (Reference)1.20 (0.83 to 1.75)---RegionCentralNorthNortheastSouth1 (Reference)0.95 (0.76 to 1.17)0.93 (0.72 to 1.190.49 (0.36 to 0.66)1 (Reference)0.88 (0.70 to 1.12)0.97 (0.74 to 1.27)0.43 (0.30 to 0.61)Social participationNoYes1 (Reference)0.74 (0.62 to 0.90)1 (Reference)0.96 (0.67 to 1.40)Alcohol useNeverPastCurrent1 (Reference)1.05 (0.75 to 1.48)0.72 (0.54 to 0.96)1 (Reference)0.95 (0.64 to 1.41)0.86 (0.62 to 1.20)Smoking tobacco useNeverPastCurrent1 (Reference)0.86 (0.61 to 1.21)0.86 (0.66 to 1.14)---Physical activity≥ 150 min/week< 150 min/week1 (Reference)1.16 (0.91 to 1.47)---Body mass indexNormalUnderOverweightObesity1 (Reference)1.24 (0.91 to 1.70)0.86 (0.67 to 1.11)0.87 (0.69 to 1.09)---Activities of Daily Living disabilityNoYes1 (Reference)2.15 (1.19 to 3.88)1 (Reference)1.51 (0.78 to 2.91)AccidentNoYes1 (Reference)1.60 (1.17 to 2.19)*1 (Reference)1.22 (0.92 to 1.63)Fall worry refrain from social activitiesNoYes1 (Reference)1.60 (1.12 to 2.29)1 (Reference)1.12 (0.75 to 1.87)Self-rated physical health statusHighLow1 (Reference)2.01 (1.60 to 2.52)*1 (Reference)1.64 (1.27 to 2.12)Depressive symptomsNoYes1 (Reference)2.36 (1.36 to 4.08)1 (Reference)1.97 (1.11 to 3.49) Chronic conditions Cardiovascular diseaseNoYes1 (Reference)1.47 (1.02 to 2.12)1 (Reference)0.99 (0.65 to 1.50)HypertensionNoYes1 (Reference)1.65 (1.37 to 1.97)1 (Reference)1.34 (1.09 to 1.65)Endocrine (diabetes)NoYes1 (Reference)1.24 (0.98 to 1.58)---MusculoskeletalNoYes1 (Reference)1.82 (1.32 to 2.51)**1 (Reference)1.61 (1.12 to 2.30)Liver or kidney diseasesNoYes1 (Reference)0.97 (0.52 to 1.81)---RespiratoryNoYes1 (Reference)1.00 (0.40 to 2.45)---CancerNoYes1 (Reference)1.65 (0.67 to 4.05)---SensorialNoYes1 (Reference)1.32 (1.03 to 1.68)1 (Reference)1.05 (0.79 to 1.40)NeurologicalNoYes1 (Reference)2.75 (1.02 to 7.41)1 (Reference)1.60 (0.48 to 5.39) Number of chronic conditions Chronic conditions0123 or more1 (Reference)1.48 (1.20 to 1.83)1.76 (1.37 to 2.26)2.28 (1.62 to 3.19)*1 (Reference)a1.28 (1.02 to 1.61)1.40 (1.06 to 1.85)1.76 (1.19 to 2.60)Sep.= separated; COR = Crude Odds Ratio; AOR = Adjusted Odds Ratio; $p \leq 0.001$;$p \leq 0.01$; $p \leq 0.05$; aadjusted for all variables except for individual chronic conditions ## Associations with persistent loneliness In adjusted logistic regression analysis, poor self-rated physical health status (aOR: 1.91, $95\%$ CI: 1.26 to 2.88), and having three or more chronic conditions (aOR: 1.78, $95\%$ CI: 1.07 to 2.98), were positively associated and living in the southern region (aOR: 0.40, $95\%$ CI: 0.25 to 0.65) was negatively associated with persistent loneliness. In addition, in univariable analysis, widowed was positively associated and male sex, and living in the northern region were negatively associated with persistent loneliness (see Table 3). Table 3Associations between independent variables and persistent loneliness, HART (2015–2017)Baseline variablesSubcategoryCOR ($95\%$ CI)AOR ($95\%$ CI)Age (in years)45–64≥ 65 or more1 (Reference)1.04 (0.76 to 1.42)---SexFemaleMale1 (Reference)0.64 (0.47 to 0.87)1 (Reference)0.71 (0.49 to 1.03)Education≤Elementary>Elementary1 (Reference)0.72 (0.45 to 1.14)---Annual incomeHighLow1 (Reference)1.19 (0.87 to 1.63)---Marital statusMarried/cohabitingDivorced/sep./never marriedWidowed1 (Reference)1.03 (0.63 to 1.70)1.42 (1.02 to 1.97)1 (Reference)0.78 (0.44 to 1.38)1.16 (0.79 to 1.70)ReligionMuslim or otherBuddhist1 (Reference)1.38 (0.87 to 2.18)---RegionCentralNorthNortheastSouth1 (Reference)0.67 (0.45 to 0.99)0.77 (0.47 to 1.28)0.32 (0.21 to 0.49)1 (Reference)0.74 (0.47 to 1.14)0.87 (0.49 to 1.54)0.40 (0.25 to 0.65)Social participationNoYes1 (Reference)0.83 (0.45 to 1.56)---Alcohol useNeverPastCurrent1 (Reference)0.75 (0.42 to 1.33)0.61 (0.35 to 1.07)---Smoking tobacco useNeverPastCurrent1 (Reference)0.90 (0.51 to 1.57)0.74 (0.44 to 1.24)---Physical activity≥ 150 min/week< 150 min/week1 (Reference)0.72 (0.47 to 1.11)---Body mass indexNormalUnderOverweightObesity1 (Reference)0.98 (0.59 to 1.64)0.57 (0.34 to 0.97)1.06 (0.72 to 1.56)1 (Reference)0.85 (0.49 to 1.46)0.58 (0.33 to 1.01)0.95 (0.63 to 1.44)Activities of Daily Living disabilityNoYes1 (Reference)0.97 (0.55 to 1.69)---AccidentNoYes1 (Reference)1.30 (0.85 to 1.99)---Fall worry refrain from social activitiesNoYes1 (Reference)0.97 (0.64 to 1.48)---Self-rated physical health statusHighLow1 (Reference)1.79 (1.27 to 2.53)*1 (Reference)1.91 (1.26 to 2.88)Depressive symptomsNoYes1 (Reference)1.07 (0.76 to 1.50)--- Chronic conditions Cardiovascular diseaseNoYes1 (Reference)1.77 (0.97 to 2.23)---HypertensionNoYes1 (Reference)1.13 (0.83 to 1.54)---Endocrine (diabetes)NoYes1 (Reference)1.27 (0.86 to 1.88)---MusculoskeletalNoYes1 (Reference)1.33 (0.82 to 2.17)---Liver or kidney diseasesNoYes1 (Reference)0.80 (0.33 to 1.92)---RespiratoryNoYes1 (Reference)1.16 (0.29 to 4.66)---CancerNoYes1 (Reference)1.55 (0.43 to 3.88)---SensorialNoYes1 (Reference)1.34 (0.94 to 1.93)---NeurologicalNoYes1 (Reference)1.29 (0.43 to 3.88)--- Number of chronic conditions Chronic conditions0123 or more1 (Reference)1.27 (0.87 to 1.84)1.21 (0.78 to 1.86)1.86 (1.15 to 3.01)1 (Reference)a1.23 (0.83 to 1.82)1.21 (0.76 to 1.91)1.78 (1.07 to 2.98)Sep.=separated; COR = Crude Odds Ratio; AOR = Adjusted Odds Ratio; *$p \leq 0.001$;$p \leq 0.01$; $p \leq 0.05$; aadjusted for all variables except for individual chronic conditions ## Discussion In this first prospective cohort study in among ageing adults in Thailand, we found that the prevalence of incident loneliness in a 2-year follow-up was $22.2\%$, which is higher than among ageing adults (4-year follow-up) in Indonesia ($15.1\%$) [17], and lower than in ELSA ($38.4\%$) (length of follow-up 6.86 years), and the HRS ($30.2\%$) (length of follow-up 8.18 years) [18]. The cross-sectional prevalence of loneliness was $21.7\%$ in this study, which is lower than in a community study in persons aged 60 years and older in the Ko Rai subdistrict, Chachoengsao, Thailand ($24\%$) [13], in Myanmar ($31.7\%$) [14], Malaysia ($32.5\%$) [11], in India ($33.8\%$) [8] but lower than in Mexico ($12.3\%$) [12] and South Africa ($9.9\%$) [9]. This study showed that loneliness is a significant public health issue in Thailand, calling for intervention programmes to reduce the burden of loneliness. We found that lower economic status, lower education, living in the central region, poor self-rated physical health, depressive symptoms, hypertension, and a higher number of chronic diseases were associated with incident loneliness. In addition, living in the central region, poor self-rated physical health, and having three or more chronic conditions were associated with persistent loneliness. The observed associations were in univariate analysis higher among females, widowed, increased age and those who had no social participation. Consistent with previous longitudinal and cross-sectional reviews [19, 20], we found that lower socioeconomic status, depressive symptoms, and poor self-perceived health were associated with loneliness. People with lower socioeconomic status may have less access to social engagement and activities, which may increase loneliness [10, 20, 35]. Depressed mood and poor self-rated health status may be closely linked with loneliness and its relationship is complex and probably bidirectional [36–38]. In addition, in agreement with previous reviews [19, in high-income countries from 1999 to 2018; 20, except for one study in Nepal, all other studies from high income countries, from 2000 to 2012], in univariate analysis, female sex, widowed, no social participation, ADL disability, adverse life event (accident, fear of falling with activity avoidance) were associated with incident loneliness. We found a lower risk of loneliness among older adults living in the southern region of Thailand. However, when calculating the baseline loneliness prevalence, we found a significantly higher prevalence of loneliness in the southern region ($36.5\%$) than in the central region ($18.0\%$). Meaning that the southern region had a higher baseline prevalence of loneliness than in the other regions, but the southern region had a lower prevalence of incident and persistent loneliness than in the other regions. The high baseline prevalence of loneliness in the southern region may be explained by lower economic indicators [22], lower health care utilization in older rural than urban dwellers [39], and compared to Bangkok older adults in rural areas have a disadvantage in healthcare due to lower socioeconomic capacity and lower healthcare access [40]. On the other hand, the lower prevalence of incident and persistent loneliness in the southern region of Thailand, may be attributed to the southern region having the highest number of Muslims and may be more rural than other regions, while the highest rate of loneliness was in the central region, which is the most urbanized region, all of which are factors that increase loneliness [41]. A previous study found a negative correlation between perceived social support and loneliness among older Muslims [42]. Furthermore, the existence of multiple chronic diseases was found to be associated with incident and persistent loneliness. In a longitudinal study in Germany, the multimorbidity was associated with incident loneliness [43], and in a cross-sectional study in the UK, physical multimorbidity was in a dose response fashion associated with loneliness [44]. The development of loneliness may be explained by limiting participation in activities and dependency feelings due to multimorbidity [9, 10, 45, 46]. In particular, hypertension was found to be associated with incident loneliness in this study. Previous studies showed that loneliness was a risk factor for hypertension [e.g., 47], or it could also be bidirectional [48]. In univariate analysis musculoskeletal conditions, cardiovascular disease, sensory loss and neurological problems were association with incident loneliness. Musculoskeletal conditions are often associated with body pain, and Emerson et al. [ 49] found in a longitudinal study that pain was a risk factor for loneliness among older adults, while in another study bidirectional longitudinal associations were found between loneliness and pain [50]. In a cross-sectional annual survey having had a stroke was associated with higher levels of loneliness [51]. However, in a longitudinal study in England there was an association between coronary heart disease and incident loneliness, but this became non-significant in the adjusted analysis [52]. Persistent loneliness increased the risk of developing dementia and Alzheimer’s disease [53]. In a longitudinal study in China, sensory impairment increased the risk of loneliness [54], and in a systematic review hearing loss increased the odds of loneliness [55]. Ageing adults with impaired vision and/or hearing may be more likely to experience ADL disability and poor social support, which can lead to incident loneliness. Unlike some previous research (in terms of substance use [56] and physical inactivity [57], our study did not find a significant association between smoking, alcohol use, and physical inactivity with incident and persistent loneliness. Similarly to a longitudinal study in Germany [58], this survey was unable to find a significant association between obesity and loneliness. ## Study limitations A limitation of the study was a large proportion of loss of follow-up ($32\%$). Like in many previous studies, we evaluated loneliness with a single item, which, however, has shown high correlations with multi-item loneliness measures [24–26]. Some variables, such as living status and cognitive function, were not assessed, and should become part of future research. Furthermore, the study used a screening questionnaire for depression. The follow-up period (2 years) was relatively short, and longer repeated follow-ups may be needed to identify stronger results. ## Conclusion More than one in five ageing adults had incident loneliness in 2 years of follow-up. The prevalence of incident and/or persistent loneliness was higher in people with a lower socioeconomic status, residing in the central region, poor self-rated physical health status, depressive symptoms, hypertension, and a higher number of chronic diseases. Results show the importance of baseline health status indicators in relation to impacting longitudinal changes in loneliness. Identifying individuals with the identified risk factors can help in providing early interventions to prevent the development of loneliness. ## References 1. Hawkley LC, Cacioppo JT. Loneliness matters: a theoretical and empirical review of consequences and mechanisms. Ann Behav Med* (2010.0) 40 218-27. DOI: 10.1007/s12160-010-9210-8 2. 2.Qualter P, Vanhalst J, Harris R, Van Roekel E, Lodder G, Bangee M et al. Loneliness across the life span.Perspect Psychol Sci2015;10, 250 – 64. 3. Luanaigh CO, Lawlor BA. Loneliness and the health of older people. Int J Geriatric Psychiatry (2008.0) 23 1213-21. DOI: 10.1002/gps.2054 4. Courtin E, Knapp M. Social isolation, loneliness and health in old age: a scoping review. Health Soc Care Community (2017.0) 25 799-812. DOI: 10.1111/hsc.12311 5. Solmi M, Veronese N, Galvano D, Favaro A, Ostinelli EG, Noventa V, Favaretto E, Tudor F, Finessi M, Shin JI, Smith L, Koyanagi A, Cester A, Bolzetta F, Cotroneo A, Maggi S, Demurtas J, De Leo D, Trabucchi M. Factors Associated with loneliness: an Umbrella Review of Observational Studies. J Affect Disord (2020.0) 271 131-8. DOI: 10.1016/j.jad.2020.03.075 6. Rico-Uribe LA, Caballero FF, Martín-María N, Cabello M, Ayuso-Mateos JL, Miret M. Association of loneliness with all-cause mortality: a meta-analysis. PLoS ONE (2018.0) 13 e0190033. DOI: 10.1371/journal.pone.0190033 7. Chawla K, Kunonga TP, Stow D, Barker R, Craig D, Hanratty B. Prevalence of loneliness amongst older people in high-income countries: a systematic review and meta-analysis. PLoS ONE (2021.0) 16 e0255088. DOI: 10.1371/journal.pone.0255088 8. Pengpid S, Peltzer K. Prevalence and correlates of loneliness among a nationally representative population-based sample of middle-aged and older adults in India. Int J Disabil Hum Dev (2022.0) 21 151-8 9. 9.Phaswana-Mafuya N, Peltzer K. Loneliness and health among older adults in South Africa. Glob J Health Sci. 2017;9(12). 10.5539/gjhs.v9n12p1. 10. Peltzer K, Pengpid S. Loneliness correlates and associations with health variables in the general population in Indonesia. Int J Ment Health Syst (2019.0) 13 24. DOI: 10.1186/s13033-019-0281-z 11. Teh JK, Tey NP, Ng ST. Family support and loneliness among older persons in multiethnic Malaysia. ScientificWorldJournal (2014.0) 2014 654382. DOI: 10.1155/2014/654382 12. Peltzer K, Pengpid S. Prevalence and correlates of loneliness among a nationally representative population-based sample of older adults in Mexico. Int J Disabil Hum Dev (2020.0) 19 581-8 13. Phuangcharoen C, Thayansin S. The loneliness of older adults Associated with various types of thai families. JPSS (2022.0) 30 207-21. DOI: 10.25133/JPSSv302022.013 14. Akhter-Khan SC, Wai KM, Drewelies J. Loneliness in Myanmar’s older population: a mixed-methods investigation. J Cross Cult Gerontol (2022.0) 37 315-37. DOI: 10.1007/s10823-022-09459-x 15. Piboon K, Subgranon R, Hengudomsub P, Wongnam P, Louise Callen B. A causal model of depression among older adults in Chon Buri Province, Thailand. Issues Ment Health Nurs (2012.0) 33 118-26. DOI: 10.3109/01612840.2011.630497 16. Aroonsrimorakot S, Laiphrakpam M, Metadilogkul O. Ageing, social isolation, loneliness, Health, Social Care and Longevity: insights from Case Studies in Thailand and India. Ageing Int (2019.0) 44 371-84. DOI: 10.1007/s12126-019-09353-x 17. Akhter-Khan SC, Chua KC, Al Kindhi B, Mayston R, Prina M. Unpaid productive activities and loneliness in later life: results from the Indonesian Family Life Survey (2000–2014). Arch Gerontol Geriatr (2023.0) 105 104851. DOI: 10.1016/j.archger.2022.104851 18. Sutin AR, Luchetti M, Aschwanden D, Lee JH, Sesker AA, Stephan Y, Terracciano A. Sense of purpose in life and concurrent loneliness and risk of incident loneliness: an individual-participant meta-analysis of 135,227 individuals from 36 cohorts. J Affect Disord (2022.0) 309 211-20. DOI: 10.1016/j.jad.2022.04.084 19. Dahlberg L, McKee KJ, Frank A, Naseer M. A systematic review of longitudinal risk factors for loneliness in older adults. Aging Ment Health (2022.0) 26 225-49. DOI: 10.1080/13607863.2021.1876638 20. Cohen-Mansfield J, Hazan H, Lerman Y, Shalom V. Correlates and predictors of loneliness in older-adults: a review of quantitative results informed by qualitative insights. Int Psychogeriatr (2016.0) 28 557-76. DOI: 10.1017/S1041610215001532 21. Anantanasuwong D, Theerawanviwat D, Siripanich P, Gu D, Dupre M. Panel survey and study on health and aging, and retirement in Thailand. Encyclopedia of gerontology and population aging (2019.0) 22. Anantanasuwong D, Pengpid S, Peltzer K. Prevalence and Associated factors of successful ageing among people 50 years and older in a National Community Sample in Thailand. Int J Environ Res Public Health (2022.0) 19 10705. DOI: 10.3390/ijerph191710705 23. Andresen EM, Malmgren JA, Carter WB, Patrick DL. Screening for depression in well older adults: evaluation of a short form of the CES-D. Am J Prev Med (1994.0) 10 77-84. DOI: 10.1016/S0749-3797(18)30622-6 24. Stickley A, Koyanagi A, Leinsalu M, Ferlander S, Sabawoon W, McKee M. Loneliness and health in Eastern Europe: fndings from Moscow, Russia. Public Health (2015.0) 129 403-10. DOI: 10.1016/j.puhe.2014.12.021 25. 25.Mund M, Maes M, Drewke PM, Gutzeit A, Jaki I, Qualter P. Would the real loneliness please stand up? The validity of loneliness scores and the reliability of single-item scores. Assessment. 2022;10731911221077227. 10.1177/10731911221077227. 26. Newmyer L, Verdery AM, Margolis R, Pessin L. Measuring older adult loneliness across Countries. J Gerontol B Psychol Sci Soc Sci (2021.0) 76 1408-14. DOI: 10.1093/geronb/gbaa109 27. Berkman LF, Sekher TV, Capistrant B, Zheng Y, Smith JP, Majmundar M. Social networks, family, and care giving among older adults in India. Aging in Asia: findings from new and emerging data initiatives (2012.0) 261-78 28. Kim JH. Regular physical exercise and its association with depression: a population-based study short title: Exercise and depression. Psychiatry Res (2022.0) 309 114406. DOI: 10.1016/j.psychres.2022.114406 29. 29.World Health Organization (WHO) guidelines on physical activity and sedentary behaviour. Licence: CC BY-NC- SA3.0 IGO, 2020. URL: https://www.who.int/publications/i/item/9789240015128 (accessed 2 April 2022) 30. Wen CP, David Cheng TY, Tsai SP, Chan HT, Hsu HL, Hsu CC. Are Asians at greater mortality risks for being overweight than Caucasians? Redefining obesity for Asians. Public Health Nutr (2009.0) 12 497-506. DOI: 10.1017/S1368980008002802 31. Katz S, Ford AB, Heiple KG, Newill VA. Studies of illness in the aged: recovery after fracture of the hip. J Gerontol (1964.0) 19 285-93. DOI: 10.1093/geronj/19.3.285 32. Yu B, Steptoe A, Niu K, Ku PW, Chen LJ. Prospective associations of social isolation and loneliness with poor sleep quality in older adults. Qual Life Res (2018.0) 27 683-91. DOI: 10.1007/s11136-017-1752-9 33. Van As BAL, Imbimbo E, Franceschi A, Menesini E, Nocentini A. The longitudinal association between loneliness and depressive symptoms in the elderly: a systematic review. Int Psychogeriatr (2022.0) 34 657-69. DOI: 10.1017/S1041610221000399 34. Tey NP, Lai SL, Teh JK. The debilitating effects of chronic diseases among the oldest old in China. Maturitas (2016.0) 94 39-45. DOI: 10.1016/j.maturitas.2016.08.016 35. Algren MH, Ekholm O, Nielsen L, Ersbøll AK, Bak CK, Andersen PT. Social isolation, loneliness, socioeconomic status, and health-risk behaviour in deprived neighbourhoods in Denmark: a cross-sectional study. SSM Popul Health (2020.0) 10 100546. DOI: 10.1016/j.ssmph.2020.100546 36. Cacioppo JT, Hawkley LC, Thisted RA. Perceived social isolation makes me sad: 5-year cross-lagged analyses of loneliness and depressive symptomatology in the Chicago Health, Aging, and Social Relations Study. Psychol Aging (2010.0) 25 453-63. DOI: 10.1037/a0017216 37. Giacco D. Loneliness and mood disorders: consequence, cause and/or unholy alliance?. Curr Opin Psychiatry (2023.0) 36 47-53. DOI: 10.1097/YCO.0000000000000832 38. Ward M, Briggs R, McGarrigle CA, De Looze C, O’Halloran AM, Kenny RA. The bi-directional association between loneliness and depression among older adults from before to during the COVID-19 pandemic. Int J Geriatr Psychiatry (2023.0) 38 e5856. DOI: 10.1002/gps.5856 39. Quashie NT, Pothisiri W. Rural-urban gaps in health care utilization among older Thais: the role of family support. Arch Gerontol Geriatr (2019.0) 81 201-8. DOI: 10.1016/j.archger.2018.12.011 40. Vicerra PMM. Regional disparity of physical function limitation among older adults in Thailand. Online J Health Allied Scs (2022.0) 21 5 41. 41.UNICEF East Asia and Pacific Regional Office. Thailand Case Study in Education, Conflict and Social Cohesion, 2014). URL: https://deepsouthwatch.org/sites/default/files/archives/docs/unicef_thailand_education_conflcit_socialcohesion2014.pdf 42. Dural G, Kavak Budak F, Özdemir AA, Gültekin A. Effect of Perceived Social Support on Self-care Agency and Loneliness among Elderly Muslim People. J Relig Health (2022.0) 61 1505-13. DOI: 10.1007/s10943-021-01377-5 43. Schübbe SF, König HH, Hajek A. Multimorbidity and loneliness. Longitudinal analysis based on the GSOEP. Arch Gerontol Geriatr (2023.0) 105 104843. DOI: 10.1016/j.archger.2022.104843 44. Stickley A, Koyanagi A. Physical multimorbidity and loneliness: a population-based study. PLoS ONE (2018.0) 13 e0191651. DOI: 10.1371/journal.pone.0191651 45. Jessen MAB, Pallesen AVJ, Kriegbaum M, Kristiansen M. The association between loneliness and health—a survey-basedstudy among middleaged and older adults in Denmark. Aging Ment Health (2018.0) 22 1338-43. DOI: 10.1080/13607863.2017.1348480 46. 46.Eckerblad J, Theander K, Ekdahl A, Jaarsma T, Hellstrom I. To adjust and endure: a qualitative study of symptom burden in older people with multimorbidity. Appl Nurs Res. 2015; 28: 322–327. 10.1016/j.apnr.2015.03.008 PMID: 26608433 47. Momtaz YA, Hamid TA, Yusoff S, Ibrahim R, Chai ST, Yahaya N, Abdullah SS. Loneliness as a risk factor for hypertension in later life. J Aging Health (2012.0) 24 696-710. DOI: 10.1177/0898264311431305 48. Turana Y, Tengkawan J, Chia YC, Shin J, Chen CH, Park S, Tsoi K, Buranakitjaroen P, Soenarta AA, Siddique S, Cheng HM, Tay JC, Teo BW, Wang TD, Kario K. Mental health problems and hypertension in the elderly: review from the HOPE Asia Network. J Clin Hypertens (Greenwich) (2021.0) 23 504-12. DOI: 10.1111/jch.14121 49. Emerson K, Boggero I, Ostir G, Jayawardhana J. Pain as a risk factor for loneliness among older adults. J Aging Health (2018.0) 30 1450-61. DOI: 10.1177/0898264317721348 50. Loeffler A, Steptoe A. Bidirectional longitudinal associations between loneliness and pain, and the role of inflammation. Pain (2021.0) 162 930-7. DOI: 10.1097/j.pain.0000000000002082 51. Byrne C, Saville CWN, Coetzer R, Ramsey R. Stroke survivors experience elevated levels of loneliness: a multi-year analysis of the National Survey for Wales. Arch Clin Neuropsychol (2022.0) 37 390-407. DOI: 10.1093/arclin/acab046 52. 52.Smith K, Victor C. Investigating the longitudinal relationship between cardiometabolic conditions and loneliness in older people.Innov Aging. 2018Nov; 2(Suppl 1):964. 53. Akhter-Khan SC, Tao Q, Ang TFA, Itchapurapu IS, Alosco ML, Mez J, Piers RJ, Steffens DC, Au R, Qiu WQ. Associations of loneliness with risk of Alzheimer’s disease dementia in the Framingham Heart Study. Alzheimers Dement (2021.0) 17 1619-27. DOI: 10.1002/alz.12327 54. Wang Q, Zhang S, Wang Y, Zhao D, Zhou C. Dual sensory impairment as a predictor of loneliness and isolation in older adults: National Cohort Study. JMIR Public Health Surveill (2022.0) 8 e39314. DOI: 10.2196/39314 55. Shukla A, Harper M, Pedersen E, Goman A, Suen JJ, Price C, Applebaum J, Hoyer M, Lin FR, Reed NS. Hearing loss, loneliness, and social isolation: a systematic review. Otolaryngol Head Neck Surg (2020.0) 162 622-33. DOI: 10.1177/0194599820910377 56. 56.Chang YC, Lee YH, Chiang T, Liu CT. Associations of Smoking and Alcohol Consumption with loneliness, Depression, and loss of interest among chinese older males and females. Int J Ment Health Addict. 2022;1–16. 10.1007/s11469-022-00912-z. 57. Pengpid S, Peltzer K. Physical activity, health and well-being among a nationally representative population-based sample of middle-aged and older adults in India, 2017–2018. Heliyon (2021.0) 7 e08635. DOI: 10.1016/j.heliyon.2021.e08635 58. Hajek A, König HH. Does obesity lead to loneliness and perceived social isolation in the second half of life? Findings from a nationally representative study in Germany. Geriatr Gerontol Int (2021.0) 21 836-41. DOI: 10.1111/ggi.14246
--- title: Clinical observation of autologous platelet rich fibrin assisted revascularization of mature permanent teeth authors: - Zhaojun Wu - Yao Lin - Xuehong Xu - Zhiqun Chen - Yan Xiang - Lvli Yang - Wei Zhang - Suli Xiao - Xiaoling Chen journal: Head & Face Medicine year: 2023 pmcid: PMC10015916 doi: 10.1186/s13005-023-00350-9 license: CC BY 4.0 --- # Clinical observation of autologous platelet rich fibrin assisted revascularization of mature permanent teeth ## Abstract ### Objective To investigate the clinical observation of autologous platelet-rich fibrin (PRF) assisting the revascularization of mature permanent teeth. ### Methods Twenty patients with mature permanent teeth were divided into experimental group and control group. The control group was treated with classic revascularization, and the experimental group was treated with PRF-assisted mature permanent tooth revascularization. ### Results After treatment, the total effective rate of the experimental group ($100.00\%$) was higher than that of the control group ($50.00\%$); the thickness of the root canal wall of the experimental group was higher than that of the control group, and the crown root length was lower than that of the control group; The bite degree, chewing function, color, overall aesthetic score, and satisfaction rate of the patients were higher, and the difference was statistically significant ($P \leq 0.05$). ### Conclusion Autologous PRF assists in revascularization of mature permanent teeth, which can achieve ideal results, and promote pulp regeneration. ## Introduction As a fully developed tissue, permanent teeth are difficult to recover once damaged [1]. When permanent teeth are fully mature and their development stops, the blood supply to the pulp is insufficient and can only come from the narrow apical foramen [2]. Therefore, traditional root canal treatment is the most common treatment method for mature permanent teeth with carious pulp exposure [3, 4]. The purpose of randomized controlled trial (RCT) is to debridement, chemically and mechanically debride the root canal system, and finally to hermetically fill the root canal system with biocompatible material [5, 6]. However, the filling materials easily discolor the crown, which affects the aesthetics of the patient's teeth [7]. Moreover, in the treated root canal, long-term Ca(OH)2 filling will reduce the flexural resistance of the dentin [8]. Therefore, it is particularly critical to seek an ideal treatment for permanent dental disease. Pulp revascularization is a common method used clinically to treat pulp diseases of permanent teeth. Although pulp revascularization is currently the only clinically approved "Regenerative endodontic treatment (RET)" treatment strategy, it still cannot fully meet the three requirements: the elimination of symptoms and evidence of bony healing, increased root wall thickness and/or increased root length, and positive response to vitality testing [9]. How to further improve the effectiveness of regeneration is still a topic of interest. Dental pulp revascularization forms blood clots in the pulp canal, which provides scaffolds and growth factors. Compared with whole blood, platelet rich fibrin (PRF) theoretically provides higher concentrations of fibrin and growth factors with potentially better therapeutic effects [10]. A recently published meta-analysis indicated that compared with blood clots, PRF has a higher 1-year mean success rate for apical closure or reduction ($85.2\%$ vs $58.8\%$) and root lengthening ($74.1\%$ vs $64.1\%$) [11]. However, there is still a lack of conclusions with significant differences, and more clinical studies are needed to confirm the results. In order to further understand its mechanism of action, through comparative studies, the clinical effect of PRF pulp revascularization in the treatment of mature permanent teeth is analyzed. details as follows. ## Inclusion and exclusion criteria Inclusion criteria: [1] 18–30 years old; [2] Immature necrotic permanent teeth [12]: tooth development is in stage 7, 8 or 9 of Nolla staging. The Nolla staging method is as follows: stage 0: no dental follicle appears; stage 1: imaging of the dental follicle appears; stage 2: beginning of calcification of the tooth tip; stage 3: crown formation of $\frac{1}{3}$; stage 4: crown formation of $\frac{2}{3}$; 5 Stage: the crown is almost formed; stage 6: the crown is formed and the root begins to develop; stage 7: the root is formed $\frac{1}{3}$; stage 8: the root is formed $\frac{2}{3}$; stage 9: the root is almost formed, the apical foramen is not closed; 10 Stage: Tooth root formation, apical foramen closed; [3] Adult permanent teeth with mature roots but with absorption damage to the apex, and the diameter of the apical hole is greater than 1 mm; [4] A restorable tooth; [5] There is no need to leave space for the final post/core restoration; [6] Anterior teeth or premolars with single canal; [7] A cooperative and compliant patient; [8] Patients are not allergic to the drugs and antibiotics which needed to complete treatment; [9] No periodontal disease or root cracking. Exclusion criteria: [1] patients with other serious organ diseases, such as cardiopulmonary, kidney and other major diseases; [2] patients with apical cyst; [3] patients with poor cooperation and those who quit the study halfway. ## General information 20 patients with mature permanent teeth treated in our hospital (may 2019 ~ may 2021) were randomly divided into control group and experimental group, with 10 cases in each group. The general data are shown in Table 1 below. There is no significant difference between the two groups ($p \leq 0.05$). The study protocol was approved by the Ethics Committee of our institution (No. KS20220606001), and it meets the ethical requirement of the Helsinki Declaration. Table 1Comparison of general data between the two groupsgroupnGender (male / female)Age (years)Course of disease (weeks)Follow up time (months)experience group$\frac{105}{524.50}$ ± 6.504.10 ± 1.3721.00 ± 3.00Control group$\frac{107}{324.00}$ ± 6.003.89 ± 1.2021.50 ± 2.50Statistical value-x2 = 0.208t = 0.179t = 0.365t = 0.405Pvalue-0.6480.8600.7200.690 ## Method The experimental group used PRF to assist mature permanent teeth revascularization treatment: [1] At the first visit (Root canal sealing): perfect X-ray, blood routine, and coagulation function examination before operation. Use articaine to local anesthetize the patient’s oral cavity, expose the decayed pulp and uncover the crown with a rubber dam, and construct a crown approach. Use 20 ml of $1.25\%$ NaOCI to wash the root canal repeatedly for 5–10 min, then rinse the root canal with normal saline for 5 min, and then dry the root canal. Ciprofloxacin, metronidazole and minocycline powder were mixed at a ratio of 1:1:1, and saline was added to prepare a 0.1 g/L triple antibiotic paste. Seal the triple antibiotic paste into the root canal, use a conveyor to feed the catheter, cover the mouth of the root canal with a sterile cotton ball, and temporarily seal the cavity with a glass ionomer cement. The patient will follow up within 3–4 weeks after surgery. The paste was removed one week later. If the gums are swollen, painful percussion, etc., root canal disinfection and sealing medicine should be repeated until the patient’s teeth are healed. [ 2] At the second visit (PRF implement): Before drawing blood, confirm that the patient has no symptoms such as red and swollen gums, and the examination has no positive characteristics. Local anesthetize the patient’s oral cavity, remove the temporary sealing material, rinse the root canal with 20 ml of $17\%$ ethylenediamine tetraacetic acid, and dry it with absorbent paper. Remove 5 ml of the patient’s median venous blood and centrifuge. After centrifugation (see Fig. 1a), the middle layer of PRF gel (see Fig. 1b) is taken out, take the supernatant (see Fig. 1c), taken out with sterile tweezers, and the gel is squeezed with sterile gauze to obtain a PRF film. Perform X-ray examination to detect the length of the patient's tooth root. Use a sterile 40# root canal file to puncture the root canal tissue beyond the root tip tissue 3-5 mm to allow blood to flow into the root canal. After that, the PRF membrane was cut into pieces and placed in the root canal (see Fig. 1d). iRoot BP Plus (Innovative Bioceramix Inc., Vancouver, Canada) was placed 4 mm below the enamel bone boundary and no more than 1-2 mm below the enamel cementum boundary (see Fig. 1e). A wet cotton ball was placed above the iRoot BP Plus, and the cavity was temporarily sealed with a glass ionomer cement (see Fig. 1f). X-rays were taken in parallel after operation. Close the crown and review the patient's constant pressure. One day after the operation, the glass ionomer cement was taken out, the hardness of the iRoot BP Plus was checked, and permanent filling was performed with light-cured resin. The patient will be reviewed 3–6 months after surgery. Fig. 1PRF assisted mature permanent teeth revascularization operation diagram. a Centrifugal treatment; b Preparation of PRF: After collecting blood and centrifuging, let it stand for stratification, and take the middle Layer PRF gel; C Take the supernatant; d Put the PRF membrane into the root canal; e Tooth with PDF film implanted; f Closed crown The control group was treated with classic revascularization: the root canal sealing was the same as the experimental group. The root apical hole was pierced, and the blood was drawn so that the blood reached 4 mm below the border of the enamel bone. No PRF was placed in the root canal. The other steps were the same as the experimental group. ## Efficacy evaluation criteria At six months after the operation, the patient had no symptoms such as pain, red and swollen gums, tooth tapping pain and no pain, no sinus in the gums, loose teeth consistent with normal teeth. X-ray examination of the root apex periodontal disease disappeared, the apex was gathered, the root canal cavity was reduced, and the root was extended. It is judged to be remarkable effect. The patient has no symptoms such as pain, red and swollen gums, tooth percussion and no pain, no sinus in the gums, loose teeth consistent with normal teeth. X-ray examination of the root apex periodontal disease disappeared, and the root can not be continued. It is judged to be effective. The patient has gum swelling and pain, hot and cold tingling, etc., gums have sinus. X-ray film shows the existence of apical periodontal disease, and the root can not be extended. It is judged to be invalid [13]. ## Observation indicators [1] The root improvement (root canal wall thickness and crown root length) was recorded 6 months after operation. The root canal wall thickness and crown root length were within the normal range. The greater the root canal wall thickness, the shorter the crown root length, indicating the better effect of the treatment. [ 2] the improvement of tooth function and patient satisfaction score were recorded. ## Sample size calculation In the initial pretrial, three patients who needed mature permanent tooth treatment adopted autologous PRF technology, and $100\%$ achieved effective results after three months. In the same period, 3 patients received treatment without autologous PRF technology, and only 1 patient achieved effective results after three months. The required sample size was calculated based on a two-tailed significance level of 0.05 and a statistical power of 0.8, resulting in a minimum sample size of 6 in each group. The present study protocol further referenced a recently published systematic review related to autologous platelet concentrates for regenerative endodontic treatment [14]. In the included studies, the sample size of each arm was approximately 5 to 15. Therefore, we selected 10 samples from each group in this study. ## Statistical methods The data was analyzed by SPSS18.0 statistical software, the measurement data was described by (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{x} \pm s$$\end{document}x¯±s), and the comparison was performed by t test; the count data was described by percentage (%), and the comparison was performed by χ2 test. $P \leq 0.05$ indicated that the difference was statistically significant. ## Comparison of efficacy between the two groups After treatment, the total effective rate of the experimental group ($100.00\%$) was higher than that of the control group ($50.00\%$), and the difference was statistically significant ($P \leq 0.05$) (see Table 2 and Fig. 2).Table 2Comparison of efficacy between the two groups [n (%)]groupnRemarkable effectEffectiveinvalidTotal effective rateexperience group107 (70.00)3 (30.00)0 (0.00)10(100.00)control group104 (40.00)1 (10.00)5 (50.00)5(50.00)x24.267P0.039Fig. 2Comparison of efficacy between the two groups ## Root canal wall thickness and crown root length before and after surgery After the operation, the thickness of the root canal wall of the experimental group was higher than that of the control group, the length of the crown root was lower than that of the control group, and the difference was statistically significant ($P \leq 0.05$) (see Table 3).Table 3Root canal wall thickness and crown root length before and after surgery (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{x} \pm s$$\end{document}x¯±s,mm)groupnRoot canal wall thicknessCrown root lengthBefore treatmentAfter treatmentBefore treatmentAfter treatmentcontrol group102.08 ± 0.582.10 ± 0.380.69 ± 0.200.97 ± 0.31experience group102.05 ± 0.492.69 ± 0.730.62 ± 0.290.69 ± 0.27t-0.1252.2670.6282.154P-0.9020.0360.5380.045 ## Patient's dental function and satisfaction score After the operation, the teeth occlusion, chewing function, color and overall aesthetic scores of the experimental group were higher than those of the control group, and the satisfaction rate of the experimental group was higher than that of the control group. The difference was statistically significant ($P \leq 0.05$) (see Tables 4, 5, and Fig. 3a, b). X-ray radiography also showed the results of three cases in the experimental group (Fig. 4, a-c).Table 4Comparison of dental function of patients (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{x} \pm s$$\end{document}x¯±s, points)groupnOcclusal degreeMasticatory functioncolor and lustreOverall beautyexperience group108.60 ± 1.208.90 ± 0.808.70 ± 0.908.50 ± 1.20control group107.50 ± 1.106.50 ± 0.407.60 ± 0.507.20 ± 0.90t-2.1378.4853.3792.741P-0.0470.0020.0030.013Table 5Comparison of patient satisfaction scores [n (%)]groupnVery satisfiedsatisfieddissatisfiedSatisfactionexperience group106 (60.00)4 (40.00)0 (0.00)10(100.00)control group103 (30.00)1 (10.00)6 (60.00)4(40.00) × 25.952P0.015Fig. 3Comparison of satisfaction between the two groups (%). a Experimental group; b Control groupFig. 4X-ray radiography showed the results of three cases in the experimental group (A-C, one case in each row) before treatment (left column), six months after treatment (median column), and two years after treatment (right column) ## Discussion Pulp revascularization can sterilize the tooth root, transform the necrotic pulp tissue into a sterile matrix, then stimulate the root tip bleeding, form a blood clot in the root canal, and generate pulp-like tissue to promote the continued development of the tooth root and improve the crown. Root ratio, to improve the strength of tooth roots, presents a better application prospect [15, 16]. PRF can assist in the application of pulp revascularization of mature permanent teeth, and the effect is good [17]. Pulp revascularization can protect the liver cells and active tissues around the roots of the patient's teeth, introduce the patient's own blood, form a biological scaffold, and promote the generation of tissues similar to the pulp [18]. This tissue has the sensation and function of normal pulp, which can enable mature permanent teeth to continue to develop, and eventually reach a level close to that of normal teeth, which is conducive to improving the hardness of the teeth, the thickness of the root canal wall, and the length of the root [19, 20]. Wu Tiantian [21] pointed out in the research that PRF is derived from the body, and the joint action of various components in inflammation regulation, angiogenesis, soft and hard tissue repair and regeneration and other physiological processes play important functions, and it has been gradually applied to young people. Permanent tooth pulp regeneration, apical barrier, delayed replantation and vital pulp preservation treatments, and the effect is good. Relevant data show that [22], pulp revascularization can repair infected or necrotic pulp, allow tooth roots to grow and develop, improve crown-to-root ratio, and increase root strength. The results of this study showed that the total effective rate of the experimental group ($100.00\%$) was higher than that of the control group ($50.00\%$), the thickness of the root canal wall of the experimental group was higher than that of the control group, and the crown root length was lower than that of the control group ($P \leq 0.05$). After the treatment, the thickness of the root canal wall and the length of the crown root have been improved, and most of the patients have achieved good results. This indicates that pulp revascularization promotes the continued development of the tooth root and accelerates the restoration of normal function of the tooth root. The reason is that, on the one hand, PRF provides a good root canal stent, providing sufficient space to store the hard tissue deposits on the inner wall of the root canal; on the other hand, PRF is rich in active factors, including cell chemokines, which promote cell entry and thereby Promote the restoration of dental pulp tissue [23, 24]. This is consistent with the research results of He X [25], which further confirms that PRF can provide a good scaffold material for pulp regeneration and the effect of pulp restoration is ideal. After treatment, the bite degree, chewing function, color, overall aesthetic score, and satisfaction of the experimental group were higher than those of the control group. Zhang Xin and others [26] selected 62 children with pulp necrosis as the research object. The control group underwent conventional pulp revascularization, and the observation group received PRF during the pulp revascularization. The total success rate of the observation group was $96.77\%$, which is significantly higher than $74.19\%$ in the control group ($P \leq 0.05$). It is concluded that the application of PRF to young permanent teeth during pulp revascularization can improve the total success rate of treatment, postoperative root length and root canal wall thickness. The effect is better than that of conventional pulp revascularization surgery. PRF is a fillable fibrin complex composed of platelets, cytokines and white blood cells. Compared to platelet-rich plasma, PRF is more economical and easier to prepare and is feasible in clinical practice [27]. Due to the great potential of PRF in clinical application. Its related technology is also constantly improving [28]. By adjusting the centrifugation procedure, injectable platelet rich fibrin (I-PRF) can be prepared without the use of anticoagulants. I-PRF has a three-dimensional fibrin meshwork while retaining the fluid nature, which has higher antibacterial, anti-inflammatory and regeneration abilities [29, 30]. With reference to the preparation protocol of I-PRF, higher concentrations of platelets and leukocytes were obtained from the buffy coat layer by high-speed centrifugation, which was named concentrated PRF (C-PRF). The growth factor release from C-PRF was then significantly increased and showed greater potential for cell migration and proliferation [31]. According to the "Low-Speed Centrifugation Concept", the preparation of PRF was further modified. An important product is Advanced-PRF (A-PFR), which leads to an increase in the number and distribution of platelets and leukocytes in the fibrin meshwork [32]. A-PRF is a variant of standard PRF that contains more growth factors with better regeneration potential and is commonly used in periodontal regeneration and implant surgery. The abovementioned materials provide a variety of therapeutic materials for dental pulp revascularization. ## Limitations There are limitations in the literature. The evaluation indicators of this study are still less, and more indicators, especially quantitative results based on radiological tests, are still needed. This study was performed in a single center. Due to the differences in medical technology and equipment conditions in different hospitals, a multicenter study is needed to confirm the effectiveness of autologous PRF technology. ## Recommendations for future Although this study confirmed the effectiveness of autologous PFR, PRF was still a complex mixture of multiple cytokines, growth factors, platelets, and various white blood cells. Furthermore, it is necessary to identify the major components that are beneficial for pulp revascularization based on omics research. The concentration and content of such beneficial components can be increased by adding exogenous active components, molecular ultrafiltration, etc., to further improve the therapeutic effect. ## Conclusion In summary, autologous platelet-rich fibrin assists in revascularization of mature permanent teeth, can achieve ideal results, promote pulp regeneration, and can maximize the thickness of the root canal wall and crown root length within the normal range, and improve the treatment effect. It is worthy of further clinical promotion. ## References 1. Chirichella R, De Marinis AM, Pokorny B, Apollonio M. Dentition and body condition: tooth wear as a correlate of weight loss in roe deer. Front Zool (2021) 18 47. DOI: 10.1186/s12983-021-00433-w 2. Park JC, Yang JH, Jo SY, Kim BC, Lee J, Lee W. Giant complex odontoma in the posterior mandible: a case report and literature review. Imaging Sci Dent (2018) 48 289-293. DOI: 10.5624/isd.2018.48.4.289 3. Asgary S, Verma P, Nosrat A. Treatment outcomes of full pulpotomy as an alternative to tooth extraction in molars with hyperplastic/irreversible pulpitis: a case report. Iran Endod J (2017) 12 261-265. PMID: 28512498 4. Sadaf D. Success of coronal Pulpotomy in permanent teeth with irreversible pulpitis: an evidence-based review. Cureus (2020) 12 e6747. PMID: 32133269 5. Machut K, Zoltowska A, Pawlowska E, Derwich M. Plasma rich in growth factors in the treatment of endodontic periapical lesions in adult patients: case reports. Int J Mol Sci (2021) 22 9458. DOI: 10.3390/ijms22179458 6. Bottino MC, Albuquerque MTP, Azabi A, Münchow EA, Spolnik KJ, Nör JE, Edwards PC. A novel patient-specific three-dimensional drug delivery construct for regenerative endodontics. J Biomed Mater Res B Appl Biomater (2019) 107 1576-1586. DOI: 10.1002/jbm.b.34250 7. Yan J, Chen H, Guo J. Status and prevention strategies of periodontal diseases in military personnel. J Endod Periodontol (2018) 28 5 8. Xu Y, Liu X, Yan W. Ibuprofen Palio sustained-release gel combined with platelet-rich fibrin for replantation of dislocated young permanent teeth. Hebei Med (2021) 43 5 9. Xie Z, Shen Z, Zhan P, Yang J, Huang Q, Huang S. Functional dental pulp regeneration: basic research and clinical translation. Int J Mol Sci (2021) 22 8991. DOI: 10.3390/ijms22168991 10. Ramachandran N, Singh S, Podar R, Kulkarni G, Shetty R, Chandrasekhar P. A comparison of two pulp revascularization techniques using platelet-rich plasma and whole blood clot. J Conserv Dent (2020) 23 637-643. DOI: 10.4103/JCD.JCD_221_20 11. Murray PE. Platelet-rich plasma and platelet-rich fibrin can induce apical closure more frequently than blood-clot revascularization for the regeneration of immature permanent teeth: a meta-analysis of clinical efficacy. Front Bioeng Biotechnol (2018) 6 139. DOI: 10.3389/fbioe.2018.00139 12. Zhang G. Diagnosis and prevention of pulp disease. Chin Mod Med Appl (2015) 9 2 13. Liu J, Chen F, Kuang J. The clinical efficacy of platelet-rich fibrin (PRF) combined with autogenous bone in immediate transplantation of autogenous teeth during tooth extraction. Chin J Oral Maxillofac Surg (2016) 14 5 14. Metlerska J, Fagogeni I, Nowicka A. Efficacy of autologous platelet concentrates in regenerative endodontic treatment: a systematic review of human studies. J Endod (2019) 45 20-30.e1. DOI: 10.1016/j.joen.2018.09.003 15. Sun H, Zhu C, Feng G, Gao Z. The biological characteristics and clinical application of platelet-rich fibrin containing leukocytes. Basic Med Clinics (2020) 40 5 16. Pradeep AR, Garg V, Kanoriya D, Singhal S. Platelet-rich fibrin with 1.2% Rosuvastatin for treatment of Intrabony defects in chronic periodontitis: a randomized controlled clinical trial. J Periodontol (2016) 87 1468-1473. DOI: 10.1902/jop.2016.160015 17. Kim JH, Woo SM, Choi NK, Kim WJ, Kim SM, Jung JY. Effect of platelet-rich fibrin on odontoblastic differentiation in human dental pulp cells exposed to Lipopolysaccharide. J Endod (2017) 43 433-438. DOI: 10.1016/j.joen.2016.11.002 18. Antunes LS, Salles AG, Gomes CC, Andrade TB, Delmindo MP, Antunes LA. The effectiveness of pulp revascularization in root formation of necrotic immature permanent teeth: a systematic review. Acta Odontol Scand (2016) 74 161-169. DOI: 10.3109/00016357.2015.1069394 19. Cong Z, Han J, Liu Y, Zhao J. Meta analysis of comparison of using platelet-rich plasma or platelet-rich fibrin and traditional blood clot as a scaffold to reconstruct the blood supply of dental pulp. Chin J Biomed Eng (2021) 27 6 20. Zhao JH, Chang YC. Alveolar ridge preservation following tooth extraction using platelet-rich fibrin as the sole grafting material. J Dent Sci (2016) 11 345-347. DOI: 10.1016/j.jds.2016.08.001 21. Wu T, Liu F, Su X, Li Z, Guo Q. Application of platelet-rich fibrin in the treatment of young permanent teeth. Medical Review (2020) 26 6 22. Tunalι M, Özdemir H, Arabacι T, Gürbüzer B, Pikdöken L, Firatli E. Clinical evaluation of autologous platelet-rich fibrin in the treatment of multiple adjacent gingival recession defects: a 12-month study. Int J Periodontics Restorative Dent (2015) 35 105-114. DOI: 10.11607/prd.1826 23. Liu J, Wu S, Han Y, Xu G, Cui L. Clinical observation on the treatment of refractory wounds with autologous platelet-rich fibrin. J Capital Univ Med Sci (2020) 41 6 24. Chen L, Wen T, Li Y. Progress in the application of platelet-rich fibrin in dental pulp treatment. Chin J Geriatr Stomatol (2020) 18 5 25. He X, Wei W, Chen WX. Three-dimensional structure of platelet-rich fibrin gel and its effect on proliferation of human dental pulp cells in vitro. Shanghai J Stomatol (2015) 24 263-268 26. Zhang X, Li N. Application effect of platelet-rich fibrin in pulp revascularization of young permanent teeth. Chin Minkang Med (2021) 33 52-53 27. Pitzurra L, Jansen IDC, de Vries TJ, Hoogenkamp MA, Loos BG. Effects of L-PRF and A-PRF+ on periodontal fibroblasts in in vitro wound healing experiments. J Periodontal Res (2020) 55 287-295. DOI: 10.1111/jre.12714 28. Arshad S, Tehreem F, Rehab Khan M, Ahmed F, Marya A, Karobari MI. Platelet-rich fibrin used in regenerative endodontics and dentistry: current uses, limitations, and future recommendations for application. Int J Dent (2021) 2021 4514598. DOI: 10.1155/2021/4514598 29. Shashank B, Bhushan M. Injectable Platelet-Rich Fibrin (PRF): the newest biomaterial and its use in various dermatological conditions in our practice: a case series. J Cosmet Dermatol (2021) 20 1421-1426. DOI: 10.1111/jocd.13742 30. Farshidfar N, Amiri MA, Jafarpour D, Hamedani S, Niknezhad SV, Tayebi L. The feasibility of injectable PRF (I-PRF) for bone tissue engineering and its application in oral and maxillofacial reconstruction: from bench to chairside. Biomater Adv (2022) 134 112557. DOI: 10.1016/j.msec.2021.112557 31. Fujioka-Kobayashi M, Katagiri H, Kono M, Schaller B, Zhang Y, Sculean A, Miron RJ. Improved growth factor delivery and cellular activity using concentrated platelet-rich fibrin (C-PRF) when compared with traditional injectable (i-PRF) protocols. Clin Oral Investig (2020) 24 4373-4383. DOI: 10.1007/s00784-020-03303-7 32. Jayadevan V, Gehlot PM, Manjunath V, Madhunapantula SV, Lakshmikanth JS. A comparative evaluation of Advanced Platelet-Rich Fibrin (A-PRF) and Platelet-Rich Fibrin (PRF) as a Scaffold in regenerative endodontic treatment of traumatized immature non-vital permanent anterior teeth: a prospective clinical study. J Clin Exp Dent (2021) 13 e463-e472. DOI: 10.4317/jced.57902
--- title: Gut microbiota in a mouse model of obesity and peripheral neuropathy associated with plasma and nerve lipidomics and nerve transcriptomics authors: - Kai Guo - Claudia Figueroa-Romero - Mohamed Noureldein - Lucy M. Hinder - Stacey A. Sakowski - Amy E. Rumora - Hayley Petit - Masha G. Savelieff - Junguk Hur - Eva L. Feldman journal: Microbiome year: 2023 pmcid: PMC10015923 doi: 10.1186/s40168-022-01436-3 license: CC BY 4.0 --- # Gut microbiota in a mouse model of obesity and peripheral neuropathy associated with plasma and nerve lipidomics and nerve transcriptomics ## Abstract ### Background Peripheral neuropathy (PN) is a common complication in obesity, prediabetes, and type 2 diabetes, though its pathogenesis remains incompletely understood. In a murine high-fat diet (HFD) obesity model of PN, dietary reversal (HFD-R) to a low-fat standard diet (SD) restores nerve function and the nerve lipidome to normal. As the gut microbiome represents a potential link between dietary fat intake and nerve health, the current study assessed shifts in microbiome community structure by 16S rRNA profiling during the paradigm of dietary reversal (HFD-R) in various gut niches. Dietary fat content (HFD versus SD) was also correlated to gut flora and metabolic and PN phenotypes. Finally, PN-associated microbial taxa that correlated with the plasma and sciatic nerve lipidome and nerve transcriptome were used to identify lipid species and genes intimately related to PN phenotypes. ### Results Microbiome structure was altered in HFD relative to SD but rapidly reversed with HFD-R. Specific taxa variants correlating positively with metabolic health associated inversely with PN, while specific taxa negatively linked to metabolic health positively associated with PN. In HFD, PN-associated taxa variants, including Lactobacillus, Lachnoclostridium, and Anaerotruncus, also positively correlated with several lipid species, especially elevated plasma sphingomyelins and sciatic nerve triglycerides. Negative correlations were additionally present with other taxa variants. Moreover, relationships that emerged between specific PN-associated taxa variants and the sciatic nerve transcriptome were related to inflammation, lipid metabolism, and antioxidant defense pathways, which are all established in PN pathogenesis. ### Conclusions The current results indicate that microbiome structure is altered with HFD, and that certain taxa variants correlate with metabolic health and PN. Apparent links between PN-associated taxa and certain lipid species and nerve transcriptome-related pathways additionally provide insight into new targets for microbiota and the associated underlying mechanisms of action in PN. Thus, these findings strengthen the possibility of a gut-microbiome-peripheral nervous system signature in PN and support continuing studies focused on defining the connection between the gut microbiome and nerve health to inform mechanistic insight and therapeutic opportunities. Video Abstract ### Supplementary Information The online version contains supplementary material available at 10.1186/s40168-022-01436-3. ## Background Obesity, prediabetes, and type 2 diabetes are global epidemics affecting hundreds of millions of people worldwide [1, 2]. Driven by overconsumption of a “Western diet” rich in saturated fats that promote metabolic dysfunction (i.e., glucose intolerance and dyslipidemia), these metabolic conditions are associated with several health complications, including peripheral neuropathy (PN) [3, 4]. PN is defined as distal-to-proximal peripheral nerve damage that results in poor gait, increased risk of foot ulceration and amputation, and lower quality of life [5]. Despite intense research, PN pathogenesis remains incompletely understood, and management remains suboptimal. The gut microbiome has emerged as a plausible link between dietary intake and metabolic and nerve health. Indeed, a Western diet and high-fat diet (HFD) influence gut microbiota and induce dysbiosis [6–8]. Additionally, obesity and T2D are associated with a perturbed microbial profile [9–11]. In turn, the gut microbiome influences host metabolism by impacting energy utilization, intestinal absorption of macronutrients including lipids, and promoting insulin resistance, hyperglycemia, and dyslipidemia [12–19]. Thus, metabolic symbiosis occurs between the microbiome and host, a relationship modulated by dietary intake. The microbiome likewise influences nerve health through a microbiome-gut-nervous system axis involving metabolite signaling and the immune system in the context of metabolic dysfunction [20, 21]. Fecal transplant from lean donor mice to recipient animals with HFD-induced obesity reverses small fiber PN and hypersensitivity, accompanied by an improved immune cell profile and an increase in circulating short-chain fatty acids, mainly butyrate [22]. HFD also induces enteric neuropathy by decreasing the density of nitrergic myenteric neurons, changes associated with gut flora restructuring [23]. In humans, T2D patients exhibit a distinct microbiota signature linked to PN status and metabolic status (insulin resistance) [24]. The microbiome may also impact pain in PN through various communication pathways, e.g., immune cells and short-chain fatty acids [22, 25, 26]. Despite several known involved pathways, the precise molecular steps precipitating PN remain elusive. However, the studies indicating potential connections between dietary intake and microbiome structure versus host metabolism and nerve health unlock interesting research avenues. We hypothesize that a microbiome-gut-peripheral nerve axis exists, whereby HFD restructures the gut microbiome which triggers systemic and local metabolic changes that negatively impact peripheral nerve function. This HFD-induced microbiome reorganization and PN relationship suggests that these effects could be reversed through dietary changes. Indeed, we previously demonstrated that dietary reversal (HFD-R) from HFD to a low-fat standard diet (SD) in a HFD obesity mouse model rescues PN phenotypes [27, 28]; however, the impact on the microbiome has not been investigated. Herein, our objective was to leverage SD, HFD, and HFD-R mice to closely examine the microbiome and test the correlations between dietary fat content, gut flora community structure, the plasma and sciatic nerve lipidome, the nerve transcriptome, and PN phenotype in order to gain initial insight into a possible gut-microbiome-peripheral nervous system signature of PN. ## Study design, phenotyping, and biospecimen collection Mice in the current study (Fig. 1A) represent a subset from a previous larger study [27] that also underwent microbiome assessments and plasma lipidomics. Briefly, a 4-week-old male C57BL/6 J mice (cat. no. 000664, The Jackson Laboratory, Bar Harbor, ME, USA) were split into 3 groups (two groups of $$n = 16$$/group; one group of $$n = 8$$/group) and fed SD, deriving $10\%$ kcal from fat (cat. no. D12450B, research diets: $10\%$ kcal fat, $20\%$ Kcal protein, $70\%$ kcal carbohydrate, $3.82\%$ energy density), ad libitum for 1 week to allow habituation. At 5 weeks of age, one group ($$n = 16$$) was maintained on SD, while the other two groups were switched to HFD, deriving $60\%$ kcal from fat (cat. no. D12492, research diets: $60\%$ kcal fat, $20\%$ kcal protein, $20\%$ kcal carbohydrate, $5.21\%$ energy density). At 16 weeks of age, one HFD group ($$n = 8$$) underwent dietary reversal (HFD-R) and was switched back to SD for the remainder of the study until 24 weeks of age. These timelines are consistent with established protocols for generation of mouse models with HFD-induced obesity, prediabetes, and PN [29–31]. Mice were maintained and housed at the University of Michigan in a pathogen-free suite following the Committee on Use and Care of Animals guidelines. Fig. 1Dietary reversal normalizes metabolic and PN phenotypes in mice. A Experimental design depicts standard diet (SD), high-fat diet (HFD), and HFD with dietary reversal (HFD-R) interventions up to 24 weeks (wks) of age in a subset of mice from our larger previously reported study [27]. Asterisks denote datasets from the previous study. B and E Dietary reversal corrects metabolic and neuropathic phenotypes in HFD mice, including B body weight ($$n = 8$$–16), C fasting blood glucose (FBG; $$n = 7$$–8), D motor nerve conduction velocity (NCV; $$n = 7$$–8), and E sensory NCV ($$n = 7$$–8) at 24 weeks of age. One-way ANOVA followed by Tukey’s post hoc test for multiple group comparisons; a, adjusted P-value < 0.05 between HFD versus SD; b, adjusted p-value < 0.05 between HFD-R versus SD; c, adjusted P-value < 0.05 between HFD-R versus HFD All mice underwent metabolic and neuropathy phenotyping at 16 and 24 weeks of age in accordance with guidelines by the Diabetic Complications Consortium (www.diacomp.org), per our standard protocol [27]. Metabolic parameters included body weights (BW) and fasting blood glucose (FBG) levels, while neuropathy phenotyping data consisted of sciatic-tibial motor and sural sensory nerve conduction velocities (NCVs) and analysis of intraepidermal nerve fiber density (IENFD). Additional metabolic and neuropathy phenotype data, including glucose tolerance tests (GTT), oxidized low-density lipoprotein (oxLDL), plasma insulin, cholesterol, and triglyceride lipoprotein profiles, and thermal latency, are previously reported [27]. Fecal pellets were additionally collected directly from animals into sterile Eppendorf tubes at 8, 10, 12, 16, 18, 20, 22, and 24 weeks of age, and data from the 8-, 16-, 18-, and 24-week time points are reported. Mice were sacrificed by lethal pentobarbital (Vortech Pharmaceutical, Dearborn, MI, USA) injection at 16 weeks of age ($$n = 8$$ SD; $$n = 6$$ HFD) or 24 weeks of age ($$n = 8$$ SD, HFD, HFD-R) to collect plasma and content from ~ 3 mm of ileum, cecum, or colon under sterile conditions (Fig. 1A, Table S1). ## Microbiome profiling and analysis Collected fecal samples and intestinal content were seeded in a PowerMag Glass Bead Plate (MO BIO Laboratories, Carlsbad, CA, USA) to isolate bacterial DNA using a MagAttract PowerMicrobiome DNA/RNA Kit (Qiagen) and epMotion 5075 liquid handling system. Amplification of the V4 region of the bacterial 16S rRNA gene was performed on an Illumina MiSeq at the University of Michigan Microbiome Core, as previously reported [32]. Raw sequencing reads were filtered with the dada2 R package [33] and then de-replicated and de-noised using derepFastq function with default parameters. After building an amplicon sequence variant (ASV) table and removing chimeras, taxonomy was assigned against the SILVA database (v132) [34] natively implemented in dada2. Uncharacterized ASVs that were not assigned to any known species at the phylum level were classified as not assigned (NA). ASVs with NA or appearing in less than three samples were removed using a prevalence threshold < number of samples × 0.05. Alpha diversity within the samples was measured using different metrics implemented in the phyloseq package [35]. Principal coordinate analysis (PCoA) based on Bray–Curtis dissimilarity metrics was performed with the proportional normalized data to reveal differences between various groups or time points. A permutational analysis of variance (PERMANOVA) using the Adonis function as part of the vegan package (https://CRAN.R-project.org/package=vegan) was performed to test the effect of treatment as a continuous variable on group differences. Differential abundance analysis was performed with the DESeq2 package [36], and significant ASVs were identified with a p-value < 0.05. Functional profiling was calculated using the Tax4Fun2 package [37]. Multiple testing of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway abundance according to sample groups was performed with the most updated KEGG database. Butyrate-producing bacteria were identified by compiling a taxonomy file that contained the major commensal butyrate-producing bacterial species from the literature [38–41]. The discovered ASVs from 16S rRNA sequencing were classified against this curated taxonomy file to identify the butyrate-producing bacteria in our dataset. ## Untargeted and targeted plasma lipidomics Plasma ($$n = 10$$/group) collected for this study was analyzed by both untargeted and targeted lipidomics at the University of Michigan Regional Comprehensive Metabolomics Resource Center. Lipids were identified from LC/MS/MS untargeted lipidomics data using the LIPIDBLAST program package (http://fiehnlab.ucdavis.edu/projects/LipidBlast) and quantified using MultiQuant software (AB-SCIEX). Missing values were imputed using a K-nearest neighbor (KNN) algorithm, and data were normalized using internal standards. Lipids were measured in both positive and negative ion modes and then merged by their mean values. Differential lipids were identified by unpaired t-test between groups with an adjusted Benjamini–Hochberg p-value < 0.05 as the significance cutoff. Targeted lipidomics was conducted on plasma and sciatic nerve samples collected from the subset of mice ($$n = 10$$) from a previous larger study [27]. All 10 plasma samples were pooled for a total of 350 µl, and lipids were extracted using organic solvents, as previously reported [42]. Triglycerides were then separated on a thin-layer chromatography plate (Merck, Darmstadt, Germany) using hexane:diethyl ether:acetic acid (80:20:1, v/v), as before [27]. Phospholipids, including sphingomyelin, were separated using a solvent mixture of chloroform:methanol:acetic acid:H2O (100:40:12:4, v/v) [43]. Sciatic nerve tissues were pooled, homogenized, and analyzed as described previously [27]. ## Previous datasets Longitudinal metabolic measures, as well as PN phenotyping and sciatic nerve transcriptomic and lipidomic datasets at 16 and 24 weeks of age, were collected from SD, HFD, and HFD-R mice in our previous study [27]. Subsets of these data corresponding to mice also followed for the current study were used for correlation analyses with the newly collected microbiome and plasma lipidomic data. ## Correlation analysis Spearman’s correlations were calculated between differential abundance of PN-associated ASVs, which overlapped between HFD versus SD and HFD-R versus HFD from the four gut microbiome niches (ileum, cecum, colon, pellets), to metabolic state and PN, sciatic nerve transcriptomics, and plasma and sciatic nerve lipidomics. Pathway analysis using KEGG and Gene Ontology (GO) annotations were performed on correlated genes in the correlation analysis of sciatic nerve transcriptomics to PN-associated ASVs. ## Statistical analysis All statistical analyses were performed using R software environment (v4.0.1). ## Dietary reversal normalizes metabolic and PN phenotypes in mice Here, we employed a HFD mouse model that develops obesity (Fig. 1B; HFD, green) and prediabetes (Fig. 1C; defined by a FBG between 150 and 180 mg/dL) versus SD mice (red). HFD mice also robustly and consistently develop PN, evidenced by slowed motor (Fig. 1D) and sensory (Fig. 1E) NVCs relative to control SD mice. Placing HFD mice at 16 weeks back on SD diet, i.e., HFD-R mice (blue), reverses metabolic and neuropathic deficits. At the 24-week time point, there were no significant differences in FBG (Fig. 1C) or motor and sensory NVCs (Fig. 1 D–E) between HFD-R and SD mice, indicating rescue of metabolic and nerve dysfunction. These data from the current subset of mice parallel those reported for the full larger study cohort [27]. ## Dietary reversal shifts microbial structure in obese PN mice In our first assessment of microbial community structure, we evaluated intragroup alpha diversity. We observed that alpha diversity was significantly higher in the HFD versus HFD-R and SD versus HFD-R groups, as assessed by the Shannon index in all samples (ileum, cecum, colon, pellets) (Fig. 2A). Examining samples by microbial niche, i.e., ileum, cecum, colon, and pellets, independent of time point, revealed consistent alpha diversity differences, especially marked by the lowest alpha diversity in the small intestine in all diet groups (Figure S1A). When samples were combined by time point and diet, independent of niche, diversity decreased within the SD group at the last time point and within the HDF-R group after dietary reversal at 18 and 24 weeks (Figure S1B); on the other hand, there were no differences in alpha diversity under the HFD intervention. Fig. 2Dietary reversal shifts microbial structure in obese PN mice. A Intragroup microbial diversity is significantly higher in HFD (green; $$n = 135$$) versus SD (red; $$n = 140$$) and HFD-R groups (blue; $$n = 84$$) as assessed by alpha diversity using Shannon index. One-way ANOVA; $P \leq 0.05$, $P \leq 0.01.$ B–C Inter-group microbial diversity assessed by beta diversity by ASV clustering by principal coordinate analysis. B Differences between gut microbiota community structure in SD (red), HFD (green), and HFD-R (blue) in the ileum, cecum, colon, and fecal pellets. Samples from HFD-R mice prior to dietary reversal from 8 to 16 weeks (wks) of age clustered with HFD samples. All time points are shown (16 and 24 weeks for ileum, cecum, and colon; biweekly samples from 8 to 12 and 16 to 24 weeks for fecal pellets). C The microbial community structure in fecal pellets at different time points (8, 16, 18, and 24 weeks of age) rapidly responds to HFD, which is reversed by dietary reversal within a short timeframe (SD, $$n = 7$$–16, red; HFD, $$n = 7$$–16, green; and HFD-R, $$n = 6$$–8, blue). D Stacked bar plot of the relative abundance of the most abundant taxa at the phylum level in fecal pellets at 8, 16, 18, and 24 weeks of age (SD, $$n = 7$$–16; HFD, $$n = 7$$–16; HFD-R, $$n = 6$$–8). E Stacked bar plot of the relative abundance of butyrate-producing bacteria at 24 weeks of age Next, we examined gut inter-group diversity between samples, assessed by beta diversity of filtered ASVs. Gut microbiome clustered SD from HFD samples, but HFD-R samples following dietary reversal clustered with SD reflecting similar microbial composition (Fig. 2B, Table S2). Longitudinal analysis of fecal pellets indicated microbial communities rapidly adjust to dietary changes (Fig. 2C). Only 3 weeks after mice were initiated on HFD (i.e., 8 weeks of age), there was a distinct shift in microbial community structure versus SD mice ($$P \leq 0.001$$). Similarly, dietary reversal for 2 weeks (i.e., 18 weeks of age) already restored altered HFD microbial community structure closer to the SD group ($$P \leq 0.001$$). Significant clustering differences between SD and HFD-R versus HFD remained at 24 weeks of age ($$P \leq 0.001$$). Notably, although close, SD and HFD-R remained individually clustered at 24 weeks of age. Ileum, cecum, and colon microbiota showed similar shifts in beta diversity between 16 and 24 weeks (Table S2). Dietary fat content also altered the relative abundance of the most abundant bacterial phyla, which included Actinobacteria, Bacteroidetes, Firmicutes, Proteobacteria, Tenericutes, and Verrucomicrobia, in all gut microbiota niches investigated. Specifically, we observed a decrease in Actinobacteria, Bacteroidetes, and Tenericutes under HFD and a concomitant increase in Firmicutes at all time points, which was reversed by dietary reversal (Fig. 2D). At 24 weeks of age, Proteobacteria was drastically lower in the cecum, colon, and fecal pellets of HFD mice, but was not reversed to SD levels in HFD-R samples. Interestingly, we observed increased butyrate-producing bacteria under HFD compared to SD or HFD-R at 24 weeks of age in the small and large intestines (Fig. 2E). However, analysis of HFD-associated butyrate-producing bacterial composition shows that most are pathogenic species. ## Dietary reversal shifts microbial taxa signature in obese PN mice To identify the microbial mediators driving dietary fat-induced changes in gut microbial differences in mice, we identified the most highly differential ASVs in ileum, cecum, colon, and fecal pellets at 16 weeks for HFD versus SD and at 24 weeks for HFD versus SD and HFD-R versus HFD ($P \leq 0.05$) (Fig. 3A, Figure S2). We then identified taxa signatures that were restored upon changes in dietary fat composition, i.e., dietary reversal, at 24 weeks of age (overlap between HFD versus SD and HFD-R versus HFD). There were 6 ASVs in the ileum, 35 in the cecum, 25 in the colon (Figure S2), and 21 in the fecal pellets (Fig. 3B). These taxa were defined as diet-sensitive ASVs and were used for subsequent analyses in the study. Although each microbiota niche had a unique signature at 24 weeks of age, we identified two ASVs at the genus level, Enterorhabdus (ASV94) and Bifidobacterium (ASV5) of the phylum Actinobacteria, and one ASV at the family level, Muribaculaceae (ASV3) of the phylum Bacteroidetes, as shared gut bacteria sensitive to dietary fat (Fig. 3D).Fig. 3Dietary reversal shifts microbial taxa signature in obese PN mice. A Analysis to identify differentially abundant fecal pellet bacteria with DESeq2 between HFD versus SD at 16 weeks of age or between HFD versus SD, HFD-R versus HFD, or HFD-R versus SD at 24 weeks of age (adjusted p-value < 0.05). B Overlap between gut microbial mediators in fecal pellets driving differences between dietary fat at 24 weeks (HFD versus SD, orange; HFD-R versus HFD, purple) represented as a bar plot of log2 fold change (log2FC). ASVs are listed with corresponding genus or family () level. C Overlap between KEGG pathway (level 1) mediators in fecal pellets driving differences between dietary fat at 24 weeks (HFD versus SD, orange; HFD-R versus HFD, purple) represented as a bar plot of log2FC. D–E Venn diagrams of unique and common (D) bacterial taxa and E KEGG pathways, which change with dietary fat intake in the mouse gut. Multiple t-testing was used for pathway comparisons, and significant pathways were identified by FDR-adjusted p-value < 0.05. ASV, amplicon sequence variant; wks, weeks Changes in functional gut microbiota composition between dietary interventions in 24-week-old mice were determined by KEGG enrichment analysis. We first identified statistically significant enriched pathways (false discovery rate [FDR] < 0.05) between HFD versus SD and HFD-R versus HFD. We identified 6 overlapping KEGG pathways in the ileum, 46 in the cecum, 19 in the colon (Figure S2), and 25 in the fecal pellets (Fig. 3C). The altered functional categories across microbial niches were mainly related to amino acid biosynthesis and metabolism, insulin resistance/secretion, adipocytokine signaling, bile secretion, and neurotransmitters. The microbiota consistently altered throughout the gut and shared by all niches were the KEGG biological pathways of tyrosine metabolism, arginine and proline metabolism, carbapenem biosynthesis, tropane, piperidine, and pyridine alkaloid biosynthesis, and legionellosis (Fig. 3E). ## Diet-sensitive gut bacteria correlate with metabolic state and PN To correlate the relative abundance of microbiota associated with dietary fat to PN features, we performed correlation analysis (FDR < 0.05) of the diet-sensitive ASVs to metabolic (BW, FBG) and PN (motor and sensory NCVs) phenotyping at 24 weeks of age (Fig. 4). Unique and common correlation patterns were observed between ASVs and phenotyping among the ileum, cecum, colon, and fecal pellets. Overall, taxa correlated with metabolic and neuropathic phenotypes in opposite directions, i.e., increased metabolic parameters with decreased nerve function and decreased metabolic parameters with increased nerve function. When screening for ASVs correlating with all parameters in at least one gut niche, we found nine ASVs had a positive correlation to metabolic measurements (BW, FBG) and negative correlation with nerve function (motor and sensory NCVs), and they increased with HFD and decreased with SD (Fig. 4). The ASVs correspond to the genera Family_XII_AD3011_group (ASV111), Lachnospiraceae_UCG-006 (ASV57), Lachnoclostridium (ASV89), Anaerotruncus (ASV106, ASV196), Enterorhabdus (ASV94), Lactobacillus (ASV120), and Candidatus_Stoquefichus (ASV98) and the family Ruminococcaceae (ASV64). On the other hand, three ASVs, including Lachnoclostridium (ASV142), Lachnospiraceae_NK4A136_group (ASV41), and Bifidobacterium (ASV5), decreased in HFD and increased in SD mice and correlated positively with healthy metabolic and nerve phenotypes. Fig. 4Diet-sensitive gut bacteria correlate with metabolic state and PN at 24 weeks of age. A Spearman’s correlation heatmap between gut microbial relative abundance at the genus level with metabolic phenotypes [body weight (BW; $$n = 22$$–24), fasting blood glucose (FBG; $$n = 14$$)] and nerve function [motor nerve conduction velocity (MNCV; $$n = 21$$–23), sensory NVC (SNCV; $$n = 21$$–23)]. Only ASVs correlating with all parameters in at least one gut niche are shown. ASVs are listed with corresponding genus or family () level. Blue ASVs are lower with HFD and correlate negatively with metabolic parameters but positively with PN. Black ASVs are higher with HFD and correlate positively with metabolic phenotypes but negatively with PN. B Summary for (A); + Met/ − PN, correlates positively with metabolic parameters and negatively with PN parameters; − Met/ + PN, correlates negatively with metabolic parameters and positively with PN parameters. Red ASVs are unique for the specified tissues; black ASVs are shared in different tissues. ASV, amplicon sequence variant; C, colon; Ce, cecum; I, ileum; P, pellet ## Microbial communities associate with plasma and nerve lipidomics and transcriptomics Next, we were interested in the correlation between long-term systemic and nerve hyperlipidemia to PN and diet-sensitive microbiota. To address this, we performed untargeted lipidomics on plasma that had been banked from a larger cohort of animals whose sciatic nerves were previously analyzed by lipidomics [27]. In plasma, we identified 578 lipids (25 major lipid class) from 76 samples. Lipid-class-based clustering visualized by heatmaps shows a clear increase in sphingomyelins in HFD versus SD mice in plasma at 16 and 24 weeks of age (Fig. 5A). This pattern was reversed in the HFD-R 24-week-old mice which had undergone dietary reversal. On the contrary, plasma triglycerides decreased in the HFD groups but increased after the mice underwent dietary reversal to SD. These observations were confirmed by lipid class aggregates (Fig. 5B) and targeted lipidomics analysis in a separate cohort of animals under the same dietary paradigm (Fig. 5C). The findings in plasma are opposite to what we previously observed in fat surrounding sciatic nerve; in nerve, triglycerides were elevated in HFD but dropped in response to dietary reversal [27], while sphingomyelins were lower in HFD and increased upon reversal to SD (Fig. 5C).Fig. 5Plasma lipidomics in an obesity PN mouse model and upon dietary reversal. A Clustering of lipid classes identified by untargeted lipidomics of plasma from 16- (left, $$n = 20$$) to 24 (right, $$n = 28$$)-week-old SD, HFD, or HFD-R mice represented in heatmaps. Purple rectangles outline areas with the most striking differences. B Levels of each lipid species from (A) were Z-score transformed to generate lipid class aggregates from plasma at 16 (top) and 24 (bottom) weeks of age, represented in bar plots of log2(value). C The sum* of plasma and sciatic nerve (SCN) [27]-targeted lipidomics (TLC-GC) shows higher sphingomyelins and lower triglycerides versus sciatic nerve [27] from HFD compared to SD and HFD-R animals at 24 weeks of age ($$n = 10$$ plasma samples; $$n = 10$$ sciatic nerves). D Analysis to identify differentially altered lipids (DALs) between HFD versus SD at 16 weeks of age or between HFD versus SD, HFD-R versus HFD, and SD versus HFD-R at 24 weeks of age (adjusted $P \leq 0.05$). E Overlapping DALs from (D) show direction of change for HFD versus SD (orange) and HFD-R versus HFD (purple), represented in a bar plot of log2 fold change (log2FC). Shared lipid species between plasma and sciatic nerve are listed in red text. CE, cholesteryl esters; CL, cardiolipins; CerP, N-hexadecanolsphingosine 1-phosphate; DG, diglycerides; FFA, free fatty acids; MG, 1-acyl-sn-glycerol; MGDG, monogalactosyldiacylglycerol; C24:1 SM, N-15Z-tetracosenoyl-sphing-4-enine; PI-Cer(d18:$\frac{1}{22}$:0), N-docosanoyl-sphing-4-enine; Cer(d18:$\frac{1}{20}$:0), N-eicosanoyl-sphing-4-enine; Cer(d18:$\frac{1}{16}$:0), N-hexadecanoyl-sphing-4-enine; Cer(d18:$\frac{1}{18}$:0), N-octadecanoyl − sphing-4-enine; Cer(d18:$\frac{1}{24}$:0), N-tetracosanoyl-sphing-4-enine; PA, phosphatidic acids; PC, phosphatidylcholines; PE, phosphatidylethanolamines; PG, phosphatidylglycerols; PI, phosphatidylinositols; PS, phosphatidylserines; SM, sphingomyelins; TG, triglycerides; LPC, lysophosphatidylcholines; LPE, lysophosphatidylethanolamines; pPC, plasmenyl-phosphatidylcholines; pPE, plasmenyl-phosphatidylethanolamines Individual circulating differentially altered lipids (DALs) were identified by comparing HFD to SD and HFD-R to HFD in 24-week-old mice (Fig. 5 D–E). The levels of 62 DALs changed direction upon dietary reversal; 47 increased and 15 decreased in HFD versus SD, which was reversed in HFD-R versus HFD groups. Lipid species belonging to diglycerides, lysophosphatidylethanolamines, phosphatidylinositols, plasmenyl-phosphatidylethanolamines, and sphingomyelins all increased under HFD and decreased when the animals were switched from HFD to SD. Next, to determine whether PN-associated bacteria sensitive to dietary fat correlate with altered host lipid profiles, we performed a series of correlation analyses (FDR < 0.05). We first determined the correlation between the relative abundance of ASVs identified in Fig. 4 to plasma lipids that changed upon dietary reversal (Fig. 6A; full correlation Figure S3A). Gut bacteria increasing in HFD, such as Lactobacillus (ASV120), Lachnoclostridium (ASV89), and Anaerotruncus (ASV196), directly correlated with many plasma DALs that increased in HFD, especially sphingomyelins. These ASVs showed inverse correlations with lipids that decreased in response to HFD. A similar pattern was observed with Enterorhabdus (ASV94), Ruminococcaceae (ASV64), Family_XIII_AD3011_group (ASV111), Anaerotruncus (ASV106), and Candidatus stoquefichus (ASV98), although correlations were limited to fewer lipid species. On the other hand, ASVs that increased with SD, such as Bifidobacterium (ASV5) and Lachnospiraceae_NK4A136_group (ASV41), correlated positively with decreasing plasma lipids (triglycerides, free fatty acids, phosphatidylcholines) and negatively with increasing lipids (mostly sphingomyelins) in the cecum, colon, and pellets samples (Fig. 6A). Lachnoclostridium (ASV142) also negatively correlated with increased lipids (mostly triglycerides) but had no correlation with decreasing lipids. Fig. 6Microbial communities associate with plasma and nerve lipidomics and transcriptomics. A–C Spearman’s correlation analysis heatmaps (FDR < 0.05) of relative abundance of PN-associated gut microbiota sensitive to dietary fat at 24 weeks of age with A plasma differentially altered lipids (DALs) ($$n = 11$$ animals), B sciatic nerve DALs [27] ($$n = 11$$ animals), and C sciatic nerve differentially expressed genes (DEGs) [27] ($$n = 11$$ animals), which are increased (up) or decreased (down) in HFD versus SD. ASVs at the genus or family () level that are higher in HFD are listed in black, and ASVs higher in SD are listed in blue. Correlation scale (red, positive; green, negative) is the same for (A–C). D Functional enrichment analysis of 64 increasing and 2 decreasing sciatic nerve DEGs [27] correlating with PN-associated bacteria using gene ontology (GO; left) and Kyoto Encyclopedia of Genes and Genomes (KEGG; right) ($P \leq 0.05$). The top ten biological pathways are shown. Bar plots indicate the proportion of DEGs assigned to each term (rich factor), with number of genes in each category indicated. CE, cholesteryl esters; CL, cardiolipins; DG, diglycerides; FFA, free fatty acids; lysoPC, lysophosphatidylcholines; lysoPE, lysophosphatidylethanolamines; PI-Cer(d18:$\frac{1}{22}$:0), N-docosanoyl-sphing-4-enine; Cer(d18:$\frac{1}{20}$:0), N-eicosanoyl-sphing-4-enine; Cer(d18:$\frac{1}{18}$:0), N-octadecanoyl-sphing-4-enine; PC, phosphatidylcholines; PE, phosphatidylethanolamines; PI, phosphatidylinositols; pPE, plasmenyl-phosphatidylethanolamines; SM, sphingomyelins; TG, triglycerides. ASV, amplicon sequence variant; FDR, false discovery rate; C, colon; Ce, cecum; I, ileum; P, pellets To determine whether diet-dependent microbial changes also correlate with lipid species surrounding the sciatic nerve at 24 weeks, we performed correlation analysis between the relative abundance of the PN-associated bacteria and sciatic nerve DALs at 24 weeks of age (Fig. 6B; full correlation Figure S3B) [27]. In contrast to plasma lipidomics, triglycerides and diglycerides around the sciatic nerve were higher in HFD and decreased in response to dietary reversal [27]. We observed positive correlations of bacteria species, such as Lactobacillus, Lachnoclostridium, and Anaerotruncus, to elevated triglycerides and diglycerides in HFD sciatic nerve. Sphingomyelin 41:2 and lysophosphatidylcholine 16:1 were the only lipid species that changed with dietary intervention and overlapped between plasma and sciatic nerve at 24 weeks; however, only lysophosphatidylcholine 16:1 changed in the same direction in both tissues (downregulated in HFD animals and upregulated in SD animals) and negatively correlated with Candidatus stoquefichus* (ASV98), Lactobacillus (ASV120), and Anaerotruncus (ASV196) in both plasma and sciatic nerve. Finally, we performed correlation analysis between the relative abundance of PN-associated bacteria to sciatic nerve transcriptomics profile at 24 weeks of age (Fig. 6C; full correlation Figure S4) [27]. Lactobacillus (ASV120), Lachnoclostridium (ASV89), Family_XIII_AD3011_group (ASV111), Candidatus stoquefichus (ASV98), Ruminococcaceae (ASV64), and Enterorhabdus (ASV94) correlated positively with upregulated genes and negatively with downregulated genes in HFD sciatic nerve. Bifidobacterium (ASV5), Lachnospiraceae_NK4A136_group (ASV41), and Lachnoclostridium (ASV142) correlated negatively with upregulated genes and positively with downregulated genes in HFD sciatic nerve. GO and KEGG pathway analysis of DEGs that correlated with PN-associated microbiota was enriched in inflammatory response driven by disintegrin and metalloproteinase domain-containing protein 8 (Adam8), haptoglobin (Hp), interleukin-1 receptor antagonist (Il1rn), serum amyloid A 3 (Saa3), and serpin family A member 3 (Serpina3n) (Fig. 6D). Also, lipid and bile metabolism and antioxidant defense pathways were represented by cytochrome P450, family 2, subfamily c, polypeptide 70 (Cyp2c70), gamma-glutamyl transferase 6 (Ggt6), phospholipase A2 group IIE (Pla2g2e), lysophosphatidylglycerol acyltransferase 1 (Lpgat1), and monoacylglycerol O-acyltransferase 2 (Mogat2). These genes mostly correlated positively with ASVs upregulated in HFD, indicating inflammatory, lipid and bile metabolism, and antioxidant defense pathways are linked with PN. ## Discussion In the current study, we leveraged our HFD obesity PN model with a paradigm of dietary reversal to investigate a putative association between the microbiome-gut-peripheral nerve axis and dietary fat intake. We report that even a short duration of high-fat feeding altered microbiome structure in mice, which rapidly reversed when animals were placed back on SD. These changes occurred through microbial ASVs linked to various metabolic and biosynthetic pathways in all four gut niches, i.e., ileum, cecum, colon, and fecal pellets. Correlation analysis further identified specific microbiome signatures linked with metabolic health and nerve function. Correlations between PN-associated ASVs from HFD animals, lipidomics, and transcriptomics data revealed that Lactobacillus, Lachnoclostridium, and Anaerotruncus taxa variants positively correlated with several lipid species, particularly elevated plasma sphingomyelins and sciatic nerve triglycerides. Relationships were also identified between specific PN-associated taxa variants to expression of genes in neuropathic nerves related to pathways involved in PN pathogenesis, including inflammation, lipid metabolism, and antioxidant defense pathways. These data link HFD-mediated PN to the gut microbiome. Using our HFD mouse model of obesity, prediabetes, and PN and established time points for dietary alteration and phenotyping [29–31], we observed a dynamic gut microbial community structure throughout the intestine that was rapidly reshaped within weeks by dietary fat content. Similar findings are reported in humans and rodents within the same timeframe [44, 45], including in response to a $60\%$ kcal HFD [46]. As previously published [47], alpha diversity was lowest in the ileum across gut niches for all diets, i.e., SD, HFD, and HFD-R, compared to large intestine. Across diets, we found alpha diversity was highest in HFD metabolically unhealthy obese mice versus SD and HFD-R. This likely represents community structure shifts in the HFD microbiome, which increase the diversity of deleterious bacteria [48], as observed in HFD mouse gut or fecal pellets by Shannon index [46, 49]. Indeed, closer examination of the individual bacterial composition of the HFD microbiota revealed increased abundance of bacteria belonging to Lachnospiraceae, Oscillospiraceae, and Clostridiaceae, families which contain pathogenic bacteria [50–54]. When we assessed beta diversity, SD samples clustered separately from HFD samples, indicating distinct microbiome structure. HFD-R samples clustered closely with SD samples, but did not ever fully reverse, in line with previously observed reports [7, 23, 46]. Our results examining phyla abundance linked to diet indicate that HFD promotes Firmicutes and reduces Bacteroidetes, i.e., high Firmicutes/Bacteroidetes ratio. Similarly, HFD enhances the proportion of butyrate-producing bacteria. While elevated Firmicutes/Bacteroidetes ratio and butyrate-producing bacteria are generally linked to a healthful status [11], several studies have noted, like our findings, elevated Firmicutes/Bacteroidetes ratio [7, 23, 46, 55, 56] and butyrate-producing metagenes [49] in HFD. Discrepancies among studies may arise from differences in host genetic background [49, 57] or species of Firmicutes phylum [58]. Additionally, the families of bacteria we observed in our bacterial composition analysis (Lachnospiraceae, Oscillospiraceae, and Clostridiaceae) produce butyrate [59–61] and thus also likely contribute to the observed increase in butyrate-producing bacteria in HFD mice. We next investigated specific ASVs, which were identified down to the family or genus level. Several ASVs across the four gut niches were sensitive to diet, i.e., differentially abundant in HFD versus SD and HFD-R versus HFD. Enterorhabdus and Bifidobacterium, of the phylum Actinobacteria, and Muribaculaceae, of the phylum Bacteroidetes, were shared by all gut niches, whereas Clostridium_sensu_stricto_1 was prominent in colon and fecal pellets. In the literature, Bifidobacterium [7, 55, 62] and Clostridium_sensu_stricto genera [55] are lower and Enterorhabdus [62] higher in mouse HFD gut, although, conversely, various Clostridium species, e.g., Clostridium CAG:58 and Clostridium orbiscindens, instead correlate positively with obesity (BMI, visceral fat) [63] and animal-based diet [44] in human microbiome. Thus, species-level differences in diet-induced gut microbial restructuring may be occurring. We further considered the functional implications of these differential ASV abundances by performing KEGG pathway analysis comparing HFD to SD. Recurrent among the gut niches were tyrosine metabolism, arginine and proline metabolism, carbapenem biosynthesis, tropane, piperidine and pyridine alkaloid biosynthesis, and legionellosis. In the large intestine, the most significant pathway was insulin resistance, whereas in fecal pellets, neurotransmitter pathways and serotonergic and dopaminergic synapse featured prominently. Several studies have shown correlation of insulin resistance with distinct gut microbial communities [15, 16, 56, 63]. Related to insulin resistance, and also represented in the KEGG analyses on cecum, colon, and fecal pellets, were adipokine signaling (same fold-change direction) and insulin secretion (opposite direction). The literature also underscores the relevance of gut microbiome metabolism related to neurotransmitter biosynthesis, e.g., tyrosine, tryptophan, phenylalanine, and glutamate metabolism and serotonin and dopamine signaling, which are central to nervous system health [20]. Although the impact of HFD on gut microbiome structure is well established [7, 23, 44, 46, 49, 55–58, 63], the effect on peripheral nerve health is less investigated. Thus, after analyzing dietary fat content-induced gut flora changes, we assessed the correlation of microbial ASVs with metabolic parameters and PN phenotypes. Generally, HFD increases BW and FBG and decreases motor and sensory NCVs [27, 28], leading to an inverse relationship in these metabolic parameters to PN phenotype. In the current study, nine ASVs correlated positively with metabolic measurements (BW, FBG) and negatively with nerve function and were elevated in HFD and reduced in SD, while three ASVs that associated negatively with metabolic parameters and positively with nerve function were decreased in HFD and increased in SD. Thus, nerve health correlates with distinct microbiome signatures. Other studies have also noted a microbiome signature of PN in humans [24] and enteric neuropathy in mice [23, 55]. Although ASVs were not provided for direct comparison, similar genera to those identified herein, such as Lactobacillus, Bifidobacterium, and Lachnoclostridium, were among those that differentiated enteric neuropathy [23, 55] and PN [24], among others. To establish a potential link between diet, PN, and microbiome to specific lipid species, we conducted lipidomics analysis of plasma from 24-week-old SD, HFD, and HFD-R mice, which we combined with our published sciatic nerve lipidomics dataset under the same diet paradigms [27]. Our observations for triglycerides and sphingomyelin showing inverse levels between plasma and sciatic nerve lipids, i.e., in HFD versus SD and HFD-R, agree with reports of elevated plasma/serum sphingomyelins in HFD mice [64, 65] and obese humans [66, 67]. In our correlation analysis of PN-associated ASVs with plasma and sciatic nerve lipids, the most salient associations emerged between elevated circulating sphingomyelins and lower triglycerides in HFD mice, which were negatively linked to certain microbiota species in the cecum, colon, and pellets samples. Bifidobacterium, a beneficial bacterial genus used in probiotics supplements, correlates with improved gut microbiome structure [68–70]. Additionally, Bifidobacterium pseudolongum supplements decrease plasma triglyceride levels in HFD mice [70], which would improve metabolic profile. We found gut bacteria increasing in HFD, such as Lactobacillus, Lachnoclostridium, and Anaerotruncus, correlated negatively with several free fatty acid and complex lipid species, including LysoPC 16:1, in both plasma and sciatic nerve. In type 2 diabetes patients, Lachnoclostridium correlated positively with total cholesterol and low-density lipoprotein cholesterol and a poorer metabolic profile [24]. We finally examined microbial correlations to nerve transcriptome. Most PN-associated gut microbiota positively correlated with upregulated sciatic DEGs, although two downregulated DEGs (Acsm3 and Jun) negatively correlated with PN-associated microbiota. Acsm3 is an important enzyme in butyrate metabolism, as it activates medium-chain fatty acids towards mitochondrial β-oxidation [71]. Downregulated Acsm3 in HFD sciatic nerve may potentially be a compensatory mechanism to slow butyrate metabolism in attempts to maintain nerve butyrate levels [22]. Similarly, we observed downregulated Jun, encoding c-Jun, a protein highly expressed in injured Schwann cells [72] and downregulated during myelination in vivo [73]. All other sciatic nerve DEGs correlated positively with gut microbiota and PN phenotypes, indicating an important link between the microbiota and PN. Pathway analysis of the correlated sciatic DEGs in HFD was related to inflammation (included genes Adam8, Saa3, Il1rn, Serpina3n) and lipid, bile, and antioxidant metabolism (Cyp2c70, Ggt6, Pla2g2e, Lpgat1, Mogat2). Notably, IL1rn is involved in granulocyte adhesion and is associated with PN in diabetes db/db murine models [74], and Saa3 is an inflammatory marker of Schwann cell injury in peripheral nerves [75]. Both IL1rn and Saa3 are markers of sterile inflammation, are modulated by gut microbiota [76, 77], and are associated with PN in both ob/ob and db/db mouse models [27, 78–80]. Mice with Serpin3 deficiency exhibit neuropathic pain, which can be reversed by exposure to exogenous Serpin3 [81]. We also identified Adam8, which stimulates axonal extension, as a novel target linking the gut microbiota to sciatic PN [82]. Many of the DEGs related to lipid metabolism (Pla2g2e, Lpgat1, Mogat2) have been previously identified in the sciatic nerve of HFD mice and are involved in linoleic acid, phospholipid, and neutral lipid metabolism [27], suggesting a link between gut microbiota and sciatic lipid metabolism. This study has limitations. First, it was a correlative study, not a causative one, which would require fecal transplants or antibiotic treatment. However, fecal transplants from lean donor mice to recipient mice with HFD-induced obesity reverse PN phenotypes and immune profiles [22], suggesting possible causality between microbiome and PN. Second, our study only involved male mice, which may not identify important findings due to sex differences in lipids [66, 83, 84] and the microbiome [20, 85]. Finally, our HFD model is based on a homogenous mouse C57BL/6 background in an experimental setting. However, genetics influences microbiome [49, 57] and metabolic phenotype [28]; thus, in the real-world setting, intraindividual variation is likely to moderate the relationships identified in this study. ## Conclusions Overall, we report the presence of a PN-associated microbiome signature in response to dietary fat. In correlation analyses, Lactobacillus, Lachnoclostridium, and Anaerotruncus ASVs positively correlated with several lipid species, particularly elevated plasma sphingomyelins and sciatic nerve triglycerides in HFD mice with PN. In sciatic nerve transcriptome, PN-associated ASVs were linked to gene expression related to inflammation, lipid metabolism, and antioxidant defense, intimating a potential gut-microbiome-peripheral nerve system. These findings underscore the importance of microbiota in PN pathogenesis. The identified HFD-associated microbial species could potentially serve as biomarkers to predict PN susceptibility in obese, prediabetic, and diabetic individuals, and clinical studies are warranted to test the correlation between these microbial species and human PN. Additionally, our assessment of dietary impact on microbiota composition shows that a HFD is associated with pathogenic microbiota, while a SD is associated with more beneficial microbiota. This offers insight into novel therapeutic strategies for PN that focus on diets with low-fat content and beneficial microbiota, supplied either via probiotics or fecal microbial transplant. Importantly, manipulation of microbiota has been successfully applied in obese, prediabetic, and diabetic individuals to improve their health and quality of life [86–90]. Our data support the contention that shifting away from a Western diet can delay PN and identify the microbiome as a potential target for therapeutic intervention in PN. Continuing studies focused on defining the connection between the gut microbiome and nerve health are thus warranted. ## Supplementary Information Additional file 1: Supplementary information: Supplementary figures: Figure S1. Alpha diversity across microbiome samples. Figure S2. Dietary reversal shifts microbial taxa signature in obese PN mice. Figure S3. Full correlation analysis for microbial communities to plasma and sciatic nerve lipidomics. Figure S4. Full correlation analysis for microbial communities to plasma and sciatic nerve transcriptomics. Supplementary tables: Table S1. Study microbiome samples. Table S2. Beta diversity in microbiome samples. ## Authors’ information Not applicable/section optional. ## References 1. Blüher M. Obesity: global epidemiology and pathogenesis. Nat Rev Endocrinol (2019.0) 15 288-298. DOI: 10.1038/s41574-019-0176-8 2. Saeedi P, Petersohn I, Salpea P, Malanda B, Karuranga S, Unwin N, Colagiuri S, Guariguata L, Motala AA, Ogurtsova K. Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: results from the International Diabetes Federation Diabetes Atlas, 9(th) edition. Diabetes Res Clin Pract (2019.0) 157 107843. DOI: 10.1016/j.diabres.2019.107843 3. Callaghan BC, Reynolds E, Banerjee M, Chant E, Villegas-Umana E, Feldman EL. Central obesity is associated with neuropathy in the severely obese. Mayo Clin Proc (2020.0) 95 1342-1353. DOI: 10.1016/j.mayocp.2020.03.025 4. Kazamel M, Stino AM, Smith AG. Metabolic syndrome and peripheral neuropathy. Muscle Nerve (2021.0) 63 285-293. DOI: 10.1002/mus.27086 5. Feldman EL, Callaghan BC, Pop-Busui R, Zochodne DW, Wright DE, Bennett DL, Bril V, Russell JW, Viswanathan V. Diabetic neuropathy. Nat Rev Dis Primers (2019.0) 5 41. DOI: 10.1038/s41572-019-0092-1 6. Bibbò S, Ianiro G, Giorgio V, Scaldaferri F, Masucci L, Gasbarrini A, Cammarota G. The role of diet on gut microbiota composition. Eur Rev Med Pharmacol Sci (2016.0) 20 4742-4749. PMID: 27906427 7. He C, Cheng D, Peng C, Li Y, Zhu Y, Lu N. High-fat diet induces dysbiosis of gastric microbiota prior to gut microbiota in association with metabolic disorders in mice. Front Microbiol (2018.0) 9 639. DOI: 10.3389/fmicb.2018.00639 8. 8.Hernandez-Baixauli J, Puigbò P, Torrell H, Palacios-Jordan H, Ripoll VJR, Caimari A, Del Bas JM, Baselga-Escudero L, Mulero M. A pilot study for metabolic profiling of obesity-associated microbial gut dysbiosis in male Wistar rats. Biomolecules. 2021;11(2):303. 9. Torres-Fuentes C, Schellekens H, Dinan TG, Cryan JF. The microbiota-gut-brain axis in obesity. Lancet Gastroenterol Hepatol (2017.0) 2 747-756. DOI: 10.1016/S2468-1253(17)30147-4 10. Gurung M, Li Z, You H, Rodrigues R, Jump DB, Morgun A, Shulzhenko N. Role of gut microbiota in type 2 diabetes pathophysiology. EBioMedicine (2020.0) 51 102590. DOI: 10.1016/j.ebiom.2019.11.051 11. Qin J, Li Y, Cai Z, Li S, Zhu J, Zhang F, Liang S, Zhang W, Guan Y, Shen D. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature (2012.0) 490 55-60. DOI: 10.1038/nature11450 12. Ko CW, Qu J, Black DD, Tso P. Regulation of intestinal lipid metabolism: current concepts and relevance to disease. Nat Rev Gastroenterol Hepatol (2020.0) 17 169-183. DOI: 10.1038/s41575-019-0250-7 13. Martinez-Guryn K, Hubert N, Frazier K, Urlass S, Musch MW, Ojeda P, Pierre JF, Miyoshi J, Sontag TJ, Cham CM. Small intestine microbiota regulate host digestive and absorptive adaptive responses to dietary lipids. Cell Host Microbe (2018.0) 23 458-469.e455. DOI: 10.1016/j.chom.2018.03.011 14. Turnbaugh PJ, Ley RE, Mahowald MA, Magrini V, Mardis ER, Gordon JI. An obesity-associated gut microbiome with increased capacity for energy harvest. Nature (2006.0) 444 1027-1031. DOI: 10.1038/nature05414 15. Wang H, Lu Y, Yan Y, Tian S, Zheng D, Leng D, Wang C, Jiao J, Wang Z, Bai Y. Promising treatment for type 2 diabetes: fecal microbiota transplantation reverses insulin resistance and impaired islets. Front Cell Infect Microbiol (2019.0) 9 455. DOI: 10.3389/fcimb.2019.00455 16. Liu Y, Wang Y, Ni Y, Cheung CKY, Lam KSL, Xia Z, Ye D, Guo J, Tse MA, Panagiotou G. Gut microbiome fermentation determines the efficacy of exercise for diabetes prevention. Cell Metab (2020.0) 31 77-91.e75. DOI: 10.1016/j.cmet.2019.11.001 17. Zhang PP, Li LL, Han X, Li QW, Zhang XH, Liu JJ, Wang Y. Fecal microbiota transplantation improves metabolism and gut microbiome composition in db/db mice. Acta Pharmacol Sin (2020.0) 41 678-685. DOI: 10.1038/s41401-019-0330-9 18. Le Roy T, Lécuyer E, Chassaing B, Rhimi M, Lhomme M, Boudebbouze S, Ichou F, Haro Barceló J, Huby T, Guerin M. The intestinal microbiota regulates host cholesterol homeostasis. BMC Biol (2019.0) 17 94. DOI: 10.1186/s12915-019-0715-8 19. Org E, Blum Y, Kasela S, Mehrabian M, Kuusisto J, Kangas AJ, Soininen P, Wang Z, Ala-Korpela M, Hazen SL. Relationships between gut microbiota, plasma metabolites, and metabolic syndrome traits in the METSIM cohort. Genome Biol (2017.0) 18 70. DOI: 10.1186/s13059-017-1194-2 20. Cryan JF, O'Riordan KJ, Cowan CSM, Sandhu KV, Bastiaanssen TFS, Boehme M, Codagnone MG, Cussotto S, Fulling C, Golubeva AV. The microbiota-gut-brain axis. Physiol Rev (2019.0) 99 1877-2013. DOI: 10.1152/physrev.00018.2018 21. Fung TC, Olson CA, Hsiao EY. Interactions between the microbiota, immune and nervous systems in health and disease. Nat Neurosci (2017.0) 20 145-155. DOI: 10.1038/nn.4476 22. Bonomo RR, Cook TM, Gavini CK, White CR, Jones JR, Bovo E, Zima AV, Brown IA, Dugas LR, Zakharian E. Fecal transplantation and butyrate improve neuropathic pain, modify immune cell profile, and gene expression in the PNS of obese mice. Proc Natl Acad Sci U S A (2020.0) 117 26482-26493. DOI: 10.1073/pnas.2006065117 23. Nyavor Y, Brands CR, May G, Kuther S, Nicholson J, Tiger K, Tesnohlidek A, Yasuda A, Starks K, Litvinenko D. High-fat diet-induced alterations to gut microbiota and gut-derived lipoteichoic acid contributes to the development of enteric neuropathy. Neurogastroenterol Motil (2020.0) 32 e13838. DOI: 10.1111/nmo.13838 24. Wang Y, Ye X, Ding D, Lu Y. Characteristics of the intestinal flora in patients with peripheral neuropathy associated with type 2 diabetes. J Int Med Res (2020.0) 48 300060520936806. DOI: 10.1177/0300060520936806 25. Lin B, Wang Y, Zhang P, Yuan Y, Zhang Y, Chen G. Gut microbiota regulates neuropathic pain: potential mechanisms and therapeutic strategy. J Headache Pain (2020.0) 21 103. DOI: 10.1186/s10194-020-01170-x 26. Cai TT, Ye XL, Yong HJ, Song B, Zheng XL, Cui BT, Zhang FM, Lu YB, Miao H, Ding DF. Fecal microbiota transplantation relieve painful diabetic neuropathy: a case report. Medicine (Baltimore) (2018.0) 97 e13543. DOI: 10.1097/MD.0000000000013543 27. O'Brien PD, Guo K, Eid SA, Rumora AE, Hinder LM, Hayes JM, Mendelson FE, Hur J, Feldman EL. Integrated lipidomic and transcriptomic analyses identify altered nerve triglycerides in mouse models of prediabetes and type 2 diabetes. Dis Model Mech (2019.0) 13 dmm042101. DOI: 10.1242/dmm.042101 28. Hinder LM, O'Brien PD, Hayes JM, Backus C, Solway AP, Sims-Robinson C, Feldman EL. Dietary reversal of neuropathy in a murine model of prediabetes and metabolic syndrome. Dis Model Mech (2017.0) 10 717-725. PMID: 28381495 29. Rumora AE, Guo K, Hinder LM, O'Brien PD, Hayes JM, Hur J, Feldman EL. A high-fat diet disrupts nerve lipids and mitochondrial function in murine models of neuropathy. Front Physiol (2022.0) 13 921942. DOI: 10.3389/fphys.2022.921942 30. Eid SA, Feldman EL. Advances in diet-induced rodent models of metabolically acquired peripheral neuropathy. Dis Model Mech (2021.0) 14 dmm049337. DOI: 10.1242/dmm.049337 31. Sajic M, Rumora AE, Kanhai AA, Dentoni G, Varatharajah S, Casey C, Brown RDR, Peters F, Hinder LM, Savelieff MG. High dietary fat consumption impairs axonal mitochondrial function in vivo. J Neurosci (2021.0) 41 4321-4334. DOI: 10.1523/JNEUROSCI.1852-20.2021 32. Seekatz AM, Theriot CM, Molloy CT, Wozniak KL, Bergin IL, Young VB. Fecal microbiota transplantation eliminates Clostridium difficile in a murine model of relapsing disease. Infect Immun (2015.0) 83 3838-3846. DOI: 10.1128/IAI.00459-15 33. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJ, Holmes SP. DADA2: high-resolution sample inference from Illumina amplicon data. Nat Methods (2016.0) 13 581-583. DOI: 10.1038/nmeth.3869 34. Quast C, Pruesse E, Yilmaz P, Gerken J, Schweer T, Yarza P, Peplies J, Glöckner FO. The SILVA ribosomal RNA gene database project: improved data processing and web-based tools. Nucleic Acids Res (2013.0) 41 D590-596. PMID: 23193283 35. McMurdie PJ, Holmes S. phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One (2013.0) 8 e61217. DOI: 10.1371/journal.pone.0061217 36. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol (2014.0) 15 550. DOI: 10.1186/s13059-014-0550-8 37. Wemheuer F, Taylor JA, Daniel R, Johnston E, Meinicke P, Thomas T, Wemheuer B. Tax4Fun2: prediction of habitat-specific functional profiles and functional redundancy based on 16S rRNA gene sequences. Environ Microbiome (2020.0) 15 11. DOI: 10.1186/s40793-020-00358-7 38. 38.Vital M, Karch A, Pieper DH. Colonic butyrate-producing communities in humans: an overview using omics data. mSystems. 2017;2(6):e00130-17. 39. Barcenilla A, Pryde SE, Martin JC, Duncan SH, Stewart CS, Henderson C, Flint HJ. Phylogenetic relationships of butyrate-producing bacteria from the human gut. Appl Environ Microbiol (2000.0) 66 1654-1661. DOI: 10.1128/AEM.66.4.1654-1661.2000 40. Baxter NT, Schmidt AW, Venkataraman A, Kim KS, Waldron C, Schmidt TM. Dynamics of human gut microbiota and short-chain fatty acids in response to dietary interventions with three fermentable fibers. MBio (2019.0) 10 e02566-e2518. DOI: 10.1128/mBio.02566-18 41. Louis P, Flint HJ. Diversity, metabolism and microbial ecology of butyrate-producing bacteria from the human large intestine. FEMS Microbiol Lett (2009.0) 294 1-8. DOI: 10.1111/j.1574-6968.2009.01514.x 42. Bligh EG, Dyer WJ. A rapid method of total lipid extraction and purification. Can J Biochem Physiol (1959.0) 37 911-917. DOI: 10.1139/o59-099 43. Skipski VP, Peterson RF, Barclay M. Quantitative analysis of phospholipids by thin-layer chromatography. Biochem J (1964.0) 90 374-378. DOI: 10.1042/bj0900374 44. David LA, Maurice CF, Carmody RN, Gootenberg DB, Button JE, Wolfe BE, Ling AV, Devlin AS, Varma Y, Fischbach MA. Diet rapidly and reproducibly alters the human gut microbiome. Nature (2014.0) 505 559-563. DOI: 10.1038/nature12820 45. Walker AW, Ince J, Duncan SH, Webster LM, Holtrop G, Ze X, Brown D, Stares MD, Scott P, Bergerat A. Dominant and diet-responsive groups of bacteria within the human colonic microbiota. Isme J (2011.0) 5 220-230. DOI: 10.1038/ismej.2010.118 46. Shang Y, Khafipour E, Derakhshani H, Sarna LK, Woo CW, Siow YL, Karmin O. Short term high fat diet induces obesity-enhancing changes in mouse gut microbiota that are partially reversed by cessation of the high fat diet. Lipids (2017.0) 52 499-511. DOI: 10.1007/s11745-017-4253-2 47. Sekirov I, Russell SL, Antunes LC, Finlay BB. Gut microbiota in health and disease. Physiol Rev (2010.0) 90 859-904. DOI: 10.1152/physrev.00045.2009 48. Jeffery IB, Lynch DB, O'Toole PW. Composition and temporal stability of the gut microbiota in older persons. Isme J (2016.0) 10 170-182. DOI: 10.1038/ismej.2015.88 49. Xiao L, Sonne SB, Feng Q, Chen N, Xia Z, Li X, Fang Z, Zhang D, Fjaere E, Midtbo LK. High-fat feeding rather than obesity drives taxonomical and functional changes in the gut microbiota in mice. Microbiome (2017.0) 5 43. DOI: 10.1186/s40168-017-0258-6 50. Carter GP, Cheung JK, Larcombe S, Lyras D. Regulation of toxin production in the pathogenic Clostridia. Mol Microbiol (2014.0) 91 221-231. DOI: 10.1111/mmi.12469 51. Kameyama K, Itoh K. Intestinal colonization by a Lachnospiraceae bacterium contributes to the development of diabetes in obese mice. Microbes Environ (2014.0) 29 427-430. DOI: 10.1264/jsme2.ME14054 52. Kim K, Lee S, Park SC, Kim NE, Shin C, Lee SK, Jung Y, Yoon D, Kim H, Kim S. Role of an unclassified Lachnospiraceae in the pathogenesis of type 2 diabetes: a longitudinal study of the urine microbiome and metabolites. Exp Mol Med (2022.0) 54 1125-1132. DOI: 10.1038/s12276-022-00816-x 53. 53.Moore RJ, Lacey JA. Genomics of the pathogenic Clostridia. Microbiol Spectr. 2019;7(3):GPP3-0033. 54. Rauz S, Low L, Suleiman K, Bassilious K, Rossiter A, Acharjee A, Loman N, Murray PI, Wallace GR. OP-10 GUT microbiota dysbiosis as a driver of inflammation in ocular mucous membrane pemphigoid. BMJ Open Ophthalmol (2022.0) 7 A3. PMID: 36161805 55. Nyavor Y, Estill R, Edwards H, Ogden H, Heideman K, Starks K, Miller C, May G, Flesch L, McMillan J. Intestinal nerve cell injury occurs prior to insulin resistance in female mice ingesting a high-fat diet. Cell Tissue Res (2019.0) 376 325-340. DOI: 10.1007/s00441-019-03002-0 56. 56.Reilly AM, Yan S, Huang M, Abhyankar SD, Conley JM, Bone RN, Stull ND, Horan DJ, Roh HC, Robling AG, et al. A high-fat diet catalyzes progression to hyperglycemia in mice with selective impairment of insulin action in Glut4-expressing tissues. J Biol Chem. 2021;298(1):101431. 57. Zhang C, Zhang M, Wang S, Han R, Cao Y, Hua W, Mao Y, Zhang X, Pang X, Wei C. Interactions between gut microbiota, host genetics and diet relevant to development of metabolic syndromes in mice. Isme J (2010.0) 4 232-241. DOI: 10.1038/ismej.2009.112 58. Turnbaugh PJ, Backhed F, Fulton L, Gordon JI. Diet-induced obesity is linked to marked but reversible alterations in the mouse distal gut microbiome. Cell Host Microbe (2008.0) 3 213-223. DOI: 10.1016/j.chom.2008.02.015 59. Flaiz M, Baur T, Brahner S, Poehlein A, Daniel R, Bengelsdorf FR. Caproicibacter fermentans gen. nov., sp. nov., a new caproate-producing bacterium and emended description of the genus Caproiciproducens. Int J Syst Evol Microbiol (2020.0) 70 4269-4279. DOI: 10.1099/ijsem.0.004283 60. Van den Abbeele P, Belzer C, Goossens M, Kleerebezem M, De Vos WM, Thas O, De Weirdt R, Kerckhof FM, Van de Wiele T. Butyrate-producing Clostridium cluster XIVa species specifically colonize mucins in an in vitro gut model. ISME J (2013.0) 7 949-961. DOI: 10.1038/ismej.2012.158 61. Vital M, Howe AC, Tiedje JM. Revealing the bacterial butyrate synthesis pathways by analyzing (meta)genomic data. mBio (2014.0) 5 e00889. DOI: 10.1128/mBio.00889-14 62. Li H, Liu F, Lu J, Shi J, Guan J, Yan F, Li B, Huo G. Probiotic mixture of Lactobacillus plantarum strains improves lipid metabolism and gut microbiota structure in high fat diet-fed mice. Front Microbiol (2020.0) 11 512. DOI: 10.3389/fmicb.2020.00512 63. Asnicar F, Berry SE, Valdes AM, Nguyen LH, Piccinno G, Drew DA, Leeming E, Gibson R, Le Roy C, Khatib HA. Microbiome connections with host metabolism and habitual diet from 1,098 deeply phenotyped individuals. Nat Med (2021.0) 27 321-332. DOI: 10.1038/s41591-020-01183-8 64. Eisinger K, Liebisch G, Schmitz G, Aslanidis C, Krautbauer S, Buechler C. Lipidomic analysis of serum from high fat diet induced obese mice. Int J Mol Sci (2014.0) 15 2991-3002. DOI: 10.3390/ijms15022991 65. 65.Shon JC, Kim WC, Ryu R, Wu Z, Seo JS, Choi MS, Liu KH. Plasma lipidomics reveals insights into anti-obesity effect of Chrysanthemum morifolium Ramat leaves and its constituent luteolin in high-fat diet-induced dyslipidemic mice. Nutrients. 2020;12(10):2973. 66. Beyene HB, Olshansky G, Smith AAT, Giles C, Huynh K, Cinel M, Mellett NA, Cadby G, Hung J, Hui J. High-coverage plasma lipidomics reveals novel sex-specific lipidomic fingerprints of age and BMI: evidence from two large population cohort studies. PLoS Biol (2020.0) 18 e3000870. DOI: 10.1371/journal.pbio.3000870 67. Im SS, Park HY, Shon JC, Chung IS, Cho HC, Liu KH, Song DK. Plasma sphingomyelins increase in pre-diabetic Korean men with abdominal obesity. PLoS One (2019.0) 14 e0213285. DOI: 10.1371/journal.pone.0213285 68. 68.Azad M, Kalam A, Sarker M, Li T, Yin J. Probiotic species in the modulation of gut microbiota: an overview. BioMed Res Int. 2018;2018:9478630. 69. Ma T, Jin H, Kwok L-Y, Sun Z, Liong M-T, Zhang H. Probiotic consumption relieved human stress and anxiety symptoms possibly via modulating the neuroactive potential of the gut microbiota. Neurobiol Stress (2021.0) 14 100294. DOI: 10.1016/j.ynstr.2021.100294 70. Bo T-b, Wen J, Zhao Y-c, Tian S-j, Zhang X-y, Wang D-h. Bifidobacterium pseudolongum reduces triglycerides by modulating gut microbiota in mice fed high-fat food. J Steroid Biochem Mol Biol (2020.0) 198 105602. DOI: 10.1016/j.jsbmb.2020.105602 71. De Preter V, Arijs I, Windey K, Vanhove W, Vermeire S, Schuit F, Rutgeerts P, Verbeke K. Impaired butyrate oxidation in ulcerative colitis is due to decreased butyrate uptake and a defect in the oxidation pathway. Inflamm Bowel Dis (2012.0) 18 1127-1136. DOI: 10.1002/ibd.21894 72. Hutton EJ, Carty L, Laurá M, Houlden H, Lunn MP, Brandner S, Mirsky R, Jessen K, Reilly MM. c-Jun expression in human neuropathies: a pilot study. J Peripher Nerv Syst (2011.0) 16 295-303. DOI: 10.1111/j.1529-8027.2011.00360.x 73. Parkinson DB, Bhaskaran A, Arthur-Farraj P, Noon LA, Woodhoo A, Lloyd AC, Feltri ML, Wrabetz L, Behrens A, Mirsky R. c-Jun is a negative regulator of myelination. J Cell Biol (2008.0) 181 625-637. DOI: 10.1083/jcb.200803013 74. Hinder LM, Murdock BJ, Park M, Bender DE, O'Brien PD, Rumora AE, Hur J, Feldman EL. Transcriptional networks of progressive diabetic peripheral neuropathy in the db/db mouse model of type 2 diabetes: an inflammatory story. Exp Neurol (2018.0) 305 33-43. DOI: 10.1016/j.expneurol.2018.03.011 75. Jang SY, Shin YK, Lee HY, Park JY, Suh DJ, Kim JK, Bae YS, Park HT. Local production of serum amyloid a is implicated in the induction of macrophage chemoattractants in Schwann cells during wallerian degeneration of peripheral nerves. Glia (2012.0) 60 1619-1628. DOI: 10.1002/glia.22382 76. Reigstad CS, Bäckhed F. Microbial regulation of SAA3 expression in mouse colon and adipose tissue. Gut Microbes (2010.0) 1 55-57. DOI: 10.4161/gmic.1.1.10514 77. Rogier R, Ederveen THA, Boekhorst J, Wopereis H, Scher JU, Manasson J, Frambach S, Knol J, Garssen J, van der Kraan PM. Aberrant intestinal microbiota due to IL-1 receptor antagonist deficiency promotes IL-17- and TLR4-dependent arthritis. Microbiome (2017.0) 5 63. DOI: 10.1186/s40168-017-0278-2 78. Hur J, Dauch JR, Hinder LM, Hayes JM, Backus C, Pennathur S, Kretzler M, Brosius FC, Feldman EL. The metabolic syndrome and microvascular complications in a murine model of type 2 diabetes. Diabetes (2015.0) 64 3294-3304. DOI: 10.2337/db15-0133 79. O'Brien PD, Hur J, Hayes JM, Backus C, Sakowski SA, Feldman EL. BTBR ob/ob mice as a novel diabetic neuropathy model: neurological characterization and gene expression analyses. Neurobiol Dis (2015.0) 73 348-355. DOI: 10.1016/j.nbd.2014.10.015 80. Pande M, Hur J, Hong Y, Backus C, Hayes JM, Oh SS, Kretzler M, Feldman EL. Transcriptional profiling of diabetic neuropathy in the BKS db/db mouse: a model of type 2 diabetes. Diabetes (2011.0) 60 1981-1989. DOI: 10.2337/db10-1541 81. Vicuña L, Strochlic DE, Latremoliere A, Bali KK, Simonetti M, Husainie D, Prokosch S, Riva P, Griffin RS, Njoo C. The serine protease inhibitor SerpinA3N attenuates neuropathic pain by inhibiting T cell-derived leukocyte elastase. Nat Med (2015.0) 21 518-523. DOI: 10.1038/nm.3852 82. Naus S, Richter M, Wildeboer D, Moss M, Schachner M, Bartsch JW. Ectodomain shedding of the neural recognition molecule CHL1 by the metalloprotease-disintegrin ADAM8 promotes neurite outgrowth and suppresses neuronal cell death. J Biol Chem (2004.0) 279 16083-16090. DOI: 10.1074/jbc.M400560200 83. 83.Torretta E, Barbacini P, Al-Daghri NM, Gelfi C. Sphingolipids in obesity and correlated co-morbidities: the contribution of gender, age and environment. Int J Mol Sci. 2019;20(23):5901. 84. Santosa S, Jensen MD. The sexual dimorphism of lipid kinetics in humans. Front Endocrinol (Lausanne) (2015.0) 6 103. DOI: 10.3389/fendo.2015.00103 85. Jašarević E, Morrison KE, Bale TL. Sex differences in the gut microbiome-brain axis across the lifespan. Philos Trans R Soc Lond B Biol Sci (2016.0) 371 20150122. DOI: 10.1098/rstb.2015.0122 86. Barthow C, Hood F, Crane J, Huthwaite M, Weatherall M, Parry-Strong A, Krebs J. A randomised controlled trial of a probiotic and a prebiotic examining metabolic and mental health outcomes in adults with pre-diabetes. BMJ Open (2022.0) 12 e055214. DOI: 10.1136/bmjopen-2021-055214 87. Ding D, Yong H, You N, Lu W, Yang X, Ye X, Wang Y, Cai T, Zheng X, Chen H. Prospective study reveals host microbial determinants of clinical response to fecal microbiota transplant therapy in type 2 diabetes patients. Front Cell Infect Microbiol (2022.0) 12 820367. DOI: 10.3389/fcimb.2022.820367 88. 88.Hasain Z, Raja Ali RA, Ahmad HF, Abdul Rauf UF, Oon SF, Mokhtar NM. The roles of probiotics in the gut microbiota composition and metabolic outcomes in asymptomatic post-gestational diabetes women: a randomized controlled trial. Nutrients. 2022;14(18):3878. 89. Su L, Hong Z, Zhou T, Jian Y, Xu M, Zhang X, Zhu X, Wang J. Health improvements of type 2 diabetic patients through diet and diet plus fecal microbiota transplantation. Sci Rep (2022.0) 12 1152. DOI: 10.1038/s41598-022-05127-9 90. Ziegler MC, Garbim Junior EE, Jahnke VS, Lisboa Moura JG, Brasil CS, Schimitt da Cunha PH, Lora PS, Gemelli T. Impact of probiotic supplementation in a patient with type 2 diabetes on glycemic and lipid profile. Clin Nutr ESPEN (2022.0) 49 264-269. DOI: 10.1016/j.clnesp.2022.04.002
--- title: The toxic effects of anabolic steroids “nandrolone decanoate” on cardiac and skeletal muscles with the potential ameliorative effects of silymarin and fenugreek seeds extract in adult male albino rats authors: - Dalia Abd Elwahab Hassan - Sherien S. Ghaleb - Amr reda Zaki - Ahmed Abdelmenem - Shimaa Nabil - Mostafa Abdallah Abdel Alim journal: BMC Pharmacology & Toxicology year: 2023 pmcid: PMC10015925 doi: 10.1186/s40360-023-00658-x license: CC BY 4.0 --- # The toxic effects of anabolic steroids “nandrolone decanoate” on cardiac and skeletal muscles with the potential ameliorative effects of silymarin and fenugreek seeds extract in adult male albino rats ## Abstract ### Background Anabolic steroids (AS) are commonly abused by body builders and athletes aiming to increase their strength and muscle mass but unfortunately, the long-term use of AS may lead to serious side effects. Nandrolone Decanoate is one of the Class II anabolic androgenic steroids which quickly spread globally and used clinically and illicitly. Our research was directed to assess the toxic effects of anabolic steroids on cardiac and skeletal muscles in male albino rats and to evaluate the potential ameliorative effects of fenugreek seeds extract and silymarin. ### Methods Our research was done on 120 male albino rats that were allocated into 6 groups; group I: Served as a control group, group II: Received the anabolic steroid Nandrolone Decanoate, group III: Received silymarin orally, group IV: Received fenugreek seeds extract orally, group (V): Received the anabolic steroid Nandrolone Decanoate and silymarin and group (VI): Received the anabolic steroid Nandrolone Decanoate and fenugreek seeds extract. By the end of the study, rats were sacrificed, and blood samples were collected for biochemical analysis and autopsy samples for histopathological examination. ### Results The anabolic steroids toxic effects on rats showed a significant decrease in serum High Density Lipoprotein (HDL) level and increase in cholesterol, triglycerides, and Low-Density Lipoprotein (LDL) levels. There was a significant elevation in cardiac troponin I level. As regards to histopathological examination of the cardiac and skeletal muscles, the study showed marked degenerative changes and necrosis. Both silymarin and fenugreek seeds extract provided a protective effect on the biochemical and histopathological changes. The antioxidant effects of silymarin and fenugreek seeds extract were evaluated on the heart, skeletal muscles and showed that, the tissue levels of Superoxide dismutase (SOD), Catalase and reduced glutathione (GSH) decreased in AS treated rats compared to the control group. On the other hand, the tissue Malondialdehyde (MDA) levels were elevated. ### Conclusions Anabolic steroids have a toxic effect on the cardiac and skeletal muscles of albino rats with improvement by treatment with fenugreek seeds extract and silymarin. ## Background Anabolic steroids can be used for many therapeutic purposes; however, their usage is usually related with several adverse effects. These side effects are generally associated with the dose as therapeutic doses appears to have little side effects while supra-physiological doses are associated with severe and serious side effects [1]. Nandrolone Decanoate is one of the Class II anabolic androgenic steroids which quickly spread globally and used clinically and illicitly and composed of 19-nortestosterone-derivates [2]. Nandrolone Decanoate is mainly metabolized by the enzyme 5α-reductase, into 5α-dihydronandrolone, 19-norandrosterone, and 19- noretiocholanolone. These metabolites can be detected in urine [3]. Nandrolone Decanoate is used clinically in burns, radiation therapy, surgery, trauma, and various forms of anemia [4]. Also, used for the treatment of chronic kidney disease, osteoporosis in postmenopausal women [5]. Nandrolone Decanoate is used in inoperable breast cancer, patients on long-term corticosteroid therapy, assistant to therapy for conditions characterized by a negative nitrogen balance [6]. Several studies were performed to evaluate the pharmacological and therapeutic benefits of fenugreek in treating different conditions like diabetes, dyslipidemia, indigestion and flatulence, inflammation, aging and cancer with variable results and mechanisms [7]. Fenugreek also known as Trigonella foenum-graecum is a plant of Fabaceae family widely cultivated in Asia and Southern Europe. The whole plant is rich in protein, vitamins and minerals so used as a nutritious healthy vegetable. In folk medicine, it is used in treatment of various conditions like diabetes, fever, and epilepsy [8]. Also, silymarin extracted from Silybum marianum has been used as a healing plant in folk medicine against different medical conditions such as liver disorders, rheumatic diseases, kidney problems, gastrointestinal tract disorders, cardiac disorders, and fever [9]. Our research was directed to assess the toxic effects of AS on some biochemical, histopathological parameters of cardiac and skeletal muscles in male albino rats in addition to evaluation of the potential ameliorative effects of Silymarin and fenugreek seeds extract. ## Methods The research was done according to the rules established by the Local Research ethical Committee for the care and use of laboratory animals present in Beni-Suef University on 120 male albino rats from the animal house of Al Nahda University, faculty of Pharmacy, Beni-Suef Governorate, Egypt. ## Experimental design The weight of the rats ranged from 150–200 gm and were stabilized for 8 weeks in the animal house before the start of the experiments. Plastic cages were used for animals housing within a room of optimum temperature (22 ± 2 °C) and the humidity level were adjusted to be (50 ± $5\%$). All animals were exposed to a 12-h cycles of light and dark with free access to food and water. Assessment of the animals’ weight was done then they were randomly allocated into one of six groups each contain 20 rats: (I)Rats received corn oil orally, 1 mg/kg/day “Control group”.(II)Rats received intramuscular (IM) Nandrolone Decanoate at a dose of 20 mg/kg/week.(III)Rats received oral Silymarin 20 mg/kg/day.(IV)Rats received the extract of fenugreek seeds orally 450 mg/kg/day orally.(V)Rats received IM Nandrolone Decanoate and oral silymarin with the same previous doses.(VI)Rats received IM Nandrolone Decanoate and the extract of fenugreek seeds orally with the same previous doses. ## Chemicals The anabolic steroid used was in the form of Nandrolone decanoate (Nandurabolin 50 mg/ml from El Nile Pharmaceutical company. Each ampule is 1 ml oily solution prepared for intramuscular injection. The dose used was 20 mg/kg/week for 8 weeks. According to the animal weight, the dose was adjusted every week [10]. Silymarin used was in the form of oral suspension (Hepaticum 50 mg/5 ml) from Medical Union Pharma Company (MUP). The silymarin dose used was 20 mg/kg/day for 8 weeks. According to the animals’ weight, the dose was adjusted every week [11]. The crude of fenugreek seeds extract was done according to Sakr and Shalaby, 2014 [12]. Semi dried seeds obtained from a local market were washed by distilled water then dried at 40 °C in an oven. The seeds were then ground by the lab grinder and passed through a sieve to obtain the raw material powder. Aqueous extract was obtained by heating the powder for 5 min after soaking in drinking water followed by filtration. The freshly prepared extract was used orally with a dose 450 mg/kg/day [13]. At the end of the experiment all rats were sacrificed via decapitation with light ether anesthesia inhalation. ## Biochemical assessment Blood samples were then collected for estimation of cardiac enzymes (cardiac troponin I) and Lipid profile (Total cholesterol, HDL, LDL, and triglycerides). The tissues were used for assessment of the antioxidant capacity by measuring Catalase, reduced glutathione (GSH), (SOD) and (MDA) in cardiac and skeletal muscles. Malonaldehyde (MDA) concentration was calculated calorimetrically according to [14] Uchiyama and Mihara [1978]. Also, reduced glutathione (GSH) concentration was estimated calorimetrically using Ellman's reagent according to [15] Sedlak and Lindsay [1968]. Further, superoxide dismutase (SOD) concentration was measured using [16] Paoletti et al. [ 1986]. Phosphate buffered saline solution at pH 7.4 used to perfuse the tissues. A blood clots and RBCs were removed from tissues using 0.16 mg/ml heparin added to the solution. For each gram of tissues, 5–10 ml cold buffer solution was used for tissue homogenization. The cold buffer solution was formed from1 mM EDTA, 1 mL/L Triton X-100 and 50 mM potassium phosphate at pH 7.4. Following tissue homogenization, at 4000 rpm centrifugation was done for 15 min at 4 °C with the supernatant removed for analysis. The samples were preserved by freezing at -80 °C that made them stable for at least one month. ## Histopathological examination For histopathological assessment, Autopsy samples were dissected and removed from the rats in all groups. Heart and Skeletal muscles samples fixed by $10\%$ neutral formalin then washed to be dehydrated with ascending grades of alcohol. Samples were then cleared using Xylen to be embedded in hard paraffin and serially sectioned with 5–6 μ thickness mounted on albumenized slides. Before staining, the slides were kept at 37 °C to dry for 24 h. Hematoxylin and Eosin (H&E) stains used then examination by the light microscope was done. ## Statistical analysis Analysis of data was performed using SPSS v. 25 (Statistical Package for Social science) for Windows. Description of variables was presented as follows: Description of quantitative variables was in the form of mean, standard deviation (SD) for normally distributed variables. All variables were explored for normality and showed that they were normally distributed. One-way ANOVA test was used to detect the difference between the four groups regarding the scale variables and Tukey post hoc high significant degree was conducted for multiple comparisons between groups. The significance of the results was assessed in the form of P-value that was differentiated into:Non-significant when P-value > 0.05.Significant when P-value ≤ 0.05. ## Cardiac troponin I level in the studied groups (Table 1) Table 1 \| Serum Troponin I \| Group I \| Group II \| Group III \| Group IV \| Group V \| Group VI \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Mean ± SD \| 0.8 ± 0.1a \| 5.5 ± 0.5b \| 0.8 ± 0.1a \| 0.5 ± 0.1a \| 0.9 ± 0.1a \| 0.7 ± 0.1a \| As shown in Table 1, There was a significant difference between group II and other groups (P-value < 0.001). The level of troponin I was increased in group II with mean ± SD (5.5 ± 0.5) while decreased in group IV with mean ± SD (0.5 ± 0.1). ## Lipid profile in the studied groups (Table 2) Table 2 \| Mean ± SD \| Group I \| Group II \| Group III \| Group IV \| Group V \| Group VI \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Serum cholesterol \| 76.6 ± 4.1a \| 103.3 ± 3.9b \| 87.3 ± 3.45c \| 73.4 ± 3.4a \| 91.4 ± 3.6d \| 81.8 ± 2.1a \| \| Serum triglycerides \| 78.7 ± 3.9a \| 94 ± 2.6b \| 81.7 ± 1.5a \| 72.5 ± 2.6c \| 81.9 ± 2.7a \| 77.9 ± 2.3a \| \| Serum LDL \| 36.3 ± 1.2a \| 46 ± 2b \| 37.9 ± 1.1a \| 30.6 ± 1.8c \| 38.5 ± 1.5a \| 37.1 ± 1.2a \| \| Serum HDL \| 48.3 ± 1.1a \| 40.6 ± 1.6b \| 52.3 ± 1.8c \| 56.5 ± 1.9d \| 52.1 ± 2.7e \| 54.4 ± 1.8f \| As shown in Table 2, There was a statistically significant differences between the six groups regarding the serum cholesterol, triglycerides, HDL, and LDL levels (P-value < 0.001). The level of serum total cholesterol was increased in group II with mean ± SD (103.3 ± 3.9) while decreased in group IV with mean ± SD (73.4 ± 3.4). The level of serum triglycerides was increased in group II with mean ± SD (94 ± 2.6) while dcreased in group IV with mean ± SD (72.5 ± 2.6). The level of serum LDL increased in group II with mean ± SD (46 ± 2) while decreased in group IV with mean ± SD (30.6 ± 1.8). The level of serum HDL increased in group IV with mean ± SD (56.5 ± 1.9) while the decreased in group II with mean ± SD (40.6 ± 1.6). ## Changes in the oxidative stress and antioxidant marker of the heart in the studied groups (Table 3) Table 3 \| Mean ± SD \| Group I \| Group II \| Group III \| Group IV \| Group V \| Group VI \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| SOD \| 2.35 ± 0.11a \| 2.08 ± 0.06b \| 2.37 ± 0.16a \| 3.03 ± 0.17c \| 2.38 ± 0.08a \| 2.84 ± 0.15d \| \| Catalase \| 1.66 ± 0.13a \| 1.27 ± 0.11b \| 1.68 ± 0.08a \| 1.69 ± 0.09a \| 1.65 ± 0.08a \| 1.62 ± 0.09a \| \| MDA \| 2.22 ± 0.09a \| 2.39 ± 0.14a \| 2.11 ± 0.07a \| 2.17 ± 0.10a \| 2.12 ± 0.13a \| 2.18 ± 0.09a \| \| GSH \| 1.64 ± 0.06a \| 1.46 ± 0.14a \| 1.86 ± 0.10b \| 1.77 ± 0.12a \| 1.75 ± 0.06a \| 1.72 ± 0.05a \| As shown in Table 3, There was a statistically significant differences between the six groups regarding the heart SOD, catalase, and reduced glutathione (GSH) levels (P-value < 0.001). The level of SOD increased seen in group IV with mean ± SD (3.03 ± 0.17) while decreased was in group II with mean ± SD (2.08 ± 0.06). The level of catalase increased seen in group IV with mean ± SD (1.69 ± 0.09) while decreased in group II with mean ± SD (1.27 ± 0.11). The level of MDA increased in group II with mean ± SD (2.39 ± 0.14) while decreased in group III with mean ± SD (2.11 ± 0.07). The level of GSH increased seen in group III with mean ± SD (1.86 ± 0.10) while decreased in group III with mean ± SD (1.46 ± 0.14). ## Changes in the oxidative stress and antioxidant marker of the skeletal muscles in the studied groups (Table 4) Table 4 \| Mean ± SD \| Group I \| Group II \| Group III \| Group IV \| Group V \| Group VI \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| SOD \| 2.8 ± 0.1a \| 2.5 ± 0.1a \| 2.9 ± 0.1a \| 3.3 ± 0.2b \| 2.8 ± 0.7a \| 3.1 ± 0.2c \| \| Catalase \| 1.2 ± 0.1a \| 1 ± 0.1b \| 1.3 ± 0.1a \| 1.25 ± 0.1a \| 1.1 ± 0.03a \| 1.1 ± 0.1a \| \| MDA \| 2.4 ± 0.1a \| 2.8 ± 0.2b \| 2.3 ± 0.1a \| 2.54 ± 0.1a \| 2.47 ± 0.1a \| 2.5 ± 0.2a \| \| GSH \| 1.9 ± 0.05a \| 1.5 ± 0.1b \| 2 ± 0.2a \| 2.3 ± 0.1c \| 1.9 ± 0.1a \| 2.2 ± 0.2a \| As shown in Table 4, There was a statistically significant differences between the six groups regarding the skeletal muscles SOD, catalase, MDA and GSH levels (P-value < 0.001). The level of SOD increased seen in group IV with mean ± SD (3.3 ± 0.2) while decreased in group II with mean ± SD (2.5 ± 0.1). The level of catalase increased in group III with mean ± SD (1.3 ± 0.1) while decreased in group II with mean ± SD (1 ± 0.1). The level of MDA increased in group II with mean ± SD (2.8 ± 0.2) while decreased were in group III with mean ± SD (2.3 ± 0.1). The level of GSH increased in group IV with mean ± SD (2.3 ± 0.1) while decreased in group II with mean ± SD (1.5 ± 0.1). ## Histopathological changes in the heart of the studied groups (Fig. 1) Fig. 1: Histopathological changes of cardiac muscle in different groups. a Section in the heart of group I showing normal histological structure (H&E X-100). b Section in the heart of group II showing severe degenerative changes and necrosis (H&E X-100). c Section in the heart of group III showing normal histological structure (H&E X-100). d Section in the heart of group IV showing normal histological structure (H&E X-100). e Section in the heart of group V showing moderate degenerative changes and hyalinosis (H&E X-100). f Section in the heart of group VI showing mild degenerative changes (H&E X-100) Detailed cardiac pathological lesions were showed in (Fig. 1). The cardiac lesions were degenerative changes and necrosis of the cardiac muscles (hyalinosis). In group I, III and IV a normal histological structure of cardiac muscles could be found appearing as long cylindrical striated cells with centrally located one or two large oval nuclei (Fig. 1a), (Figs. 1c) and (Fig. 1d) respectively. Severe degenerative changes and necrosis of cardiac muscles could be found in group II (Fig. 1b). Group V showed moderate degenerative changes and hyalinosis of cardiac muscles (Fig. 1e), while group VI was associated with mild degenerative changes of the cardiac muscles in comparison to the other groups (Fig. 1f). ## Histopathological changes in the skeletal muscles of the studied groups (Fig. 2) Fig. 2: Histopathological changes of skeletal muscle in different groups. a Section in the muscle of group I showing normal histological structure (H&E X-100). b Section in the muscle of group II showing severe degenerative changes and nuclear pyknosis of the sarcoplasm (H&E X-100). c Section in the muscle of group III showing normal histological structure (H&E X-100). d Section in the muscle of group IV showing normal histological structure (H&E X-100). e Section in the muscle of group V showing quite improvement of the degenerative changes (H&E X-100). f Section in the muscle of group VI showing quite improvement of the degenerative changes (H&E X-100) Detailed muscular lesions are showed in (Fig. 2). Histopathological examination of the skeletal muscle obtained from group I, III and IV were normal (Fig. 2a), (Fig. 2c) and (Fig. 2d) respectively. Group II exhibited severe pathological lesions (severe degenerative changes with nuclear pyknosis of the sarcoplasm) (Fig. 2b). On the other hand, quite improvement of degenerative changes of the musculature mainly in group V and VI. ( Fig. 2e) (Fig. 2f) respectively. ## Discussion The use of AS showed a dramatic increase in the recent years especially by young adults aiming to increase their power, their body mass and weight and to have a better appearance improving their self-esteem. Also, AS have been used for decades by athletes and body builders for the same purpose. On the other hand, many side effects were reported with their illicit abuse [17]. Our work was performed to assess the toxic effects of AS on the heart and the skeletal muscles and assessment of the ameliorative effects of silymarin and fenugreek seeds extract. As shown in Table 1, There was a significant difference between group II and other groups. Highest level of troponin I was seen in group II (5.5 ± 0.5) while the lowest level was in group IV (0.5 ± 0.1). The AS treated rats showed a significant elevation in troponin levels. Also, there were non-significant decrease in Silymarin and fenugreek seeds extract. When, silymarin and fenugreek seeds extract were given with AS the levels of serum troponin decreased with no significant differences. Similarly, Kulaksiz and Sefa, 2019 [18] reported an increase in the serum levels of troponin in testosterone treated rats. Cardiac troponin I (cTnI) is a cardiac enzyme which is released when ether is a damage in the myocytes and considered a sensitive factor of these damages. The levels of this enzyme showed no significant changes in research done on mice treated with silymarin alone [19]. Silymarin could keep the membrane integrity, limiting the leakage of enzymes revealing the cardio-protective effect [20]. The activity of cardiac enzyme biomarkers including troponin showed a significant decrease with fenugreek seeds by $27\%$ indicating the cardio-protective effect of fenugreek seeds [21]. Kamble and Bodhankar, 2014 [22] explained that the cardio-protective effect of fenugreek seed could be due to trigonelline, saponins, 4 hydroxyisoleucine presence, and high fiber contents. Also, may be due to fenugreek scavenging properties of the free radical. These properties are attributed to the active hydrogen-donating ability of the hydroxyl substitutions [12]. As regards to the lipid profile, rats received AS showed a significant elevation in serum cholesterol, triglycerides and LDL levels and a significant decrease in serum HDL levels. Silymarin treated rats showed a significant increase in serum cholesterol and HDL levels. Also, there was elevation in serum triglycerides and LDL levels with no significant differences. In rate treated with fenugreek seeds extract, there was a significant decrease in the serum triglycerides and LDL levels. Also, the serum cholesterol levels decreased but with no significant differences. On the other hand, serum HDL levels increased with significant difference. Like the results of this study was that reported by Silva et al., 2018 [23] who demonstrated that AS could significantly raise the serum triglycerides, cholesterol, and LDL levels. Unlike, their effect on HDL that showed a significant reduction in serum level. The abuse of AS in supra-physiological doses is strongly associated with abnormal levels of plasma lipoproteins showing a decreased level of HDL and increased levels of LDL and cholesterol levels [24]. Treatment with fenugreek seeds extract showed improvement in lipid profiles, including decrease in serum triglycerides, total cholesterol, and LDL levels. Also, there was a significant elevation in HDL levels [25]. This effect could be because of sapogenins agent that play a role in decreasing the synthesis of cholesterol and fatty acids. Also, they increase the excretion of cholesterol in bile lowering the serum cholesterol levels [26]. Also, Roberts, 2011 [27] demonstrated that saponins could decrease the cholesterol levels by inhibiting the absorption of cholesterol and enhance its excretion in bile by forming large particles with bile salts. The excess soluble fibers in fenugreek extract may lead to delay in the absorption of fat and carbohydrate adding to the hypolipidemic effect. Also, the mannose presents in the extract reduce the synthesis of cholesterol. Metwally et al., 2009 [28] reported increase in serum cholesterol, LDL, triglycerides, and HDL levels regarding the effect of silymarin on lipid profile. Gobalakrishnan et al., 2016 [29] explained the different mechanisms for silymarin induced hypo-cholesterolemic effect. Among these effects are the increased the production of LDL receptors on hepatic cells increasing the clearance of plasma LDL. Also, silymarin increases the conversion of cholesterol into bile acid. However, silymarin has no effect on the absorption of cholesterol. The antioxidant effects of silymarin and fenugreek seeds extract were evaluated on the heart, skeletal muscles and showed that, the tissue levels of SOD, Catalase and reduced glutathione decreased in AS treated rats compared to the control group. On the other hand, the tissue MDA levels were elevated. Similarly, previous studies [30] and [31] reported that an increase of the lipid peroxidation and decrease of the enzymatic activity of glutathione reductase and SOD of different tissues in adult male rats because of exogenous testosterone intake. Frankenfeld et al., 2014 [32] showed a decrease in the activity of catalase enzyme after AS (nandrolone decanoate) intake. High doses of nandrolone decanoate are metabolized by cytochrome P450 mono-oxygenases. This results in production of reactive oxygen species leading to the upregulation followed by exhaustion of the antioxidants enzymes activity [33]. Furthermore, anabolic steroids enhance the activity of lipase enzyme which in turn increases the rate of lipolysis [34]. which in turn increases the availability of long chain fatty acids for mitochondrial oxidation and production of ATP leading to development of the lipid peroxidation and generation of ROS [35]. As regards to silymarin, our results were in consistence with the study performed by Avci et al., 2017 [36] about the effect of silymarin on MDA and SOD levels in the heart of rats. They demonstrated an increase in the activity of SOD and decrease in MDA levels denoting a protective effect against lipid peroxidation and oxidative damage. Similarly, Aktas and Ozgocmen, 2020 [37] revealed in their study an increase in the activity of glutathione enzyme and decrease in MDA levels in cardiac tissue of rats treated by silymarin. Surai, 2015 [38] reported that silymarin is a potent antioxidant inhibiting the lipid peroxidation and prevents the reduction of glutathione enhancing the activity of antioxidants enzymes. This effect could be clarified by its effect on maintaining the integrity of cell membranes against the oxidative damage of ROS [39]. This antioxidant properties probably due to the presence of silydianin, silybin, silychristin and flavolignans which are chemical bioactive compounds present in silymarin [40]. Regarding fenugreek seeds extract, Bafadam et al., 2021 [41] revealed an increase in the activity of SOD and catalase in heart tissue of rats and decrease in lipid peroxidation augmented by the decreased MDA levels in hearts tissues of rats treated with germinated fenugreek seed. Also, Arshadi et al., 2015 [42] showed an increase in the activity of cardiac antioxidant enzymes like reduced glutathione, catalase, and SOD. Fenugreek seeds extract antioxidant properties are mainly because of the presence of the biologically active compounds flavonoids and polyphenols. Its major active antioxidant compounds are isovitexin and flavones of vitexin [43]. Histopathological examination of cardiac muscle in our study showed that, the most observed lesions were degenerative changes and cardiac muscles necrosis (hyalinosis) which was severe in rats treated with AS. On the other hand, normal histological structure of cardiac muscles could be found in rats of the control group and those received silymarin and fenugreek seeds extract. Improvement of the toxic effects induced by AS was observed with treatment of silymarin and fenugreek seeds extract with mild degenerative changes were seen with treatment by fenugreek seeds extract and moderate degenerative changes with treatment by silymarin. Histopathological examination of the skeletal muscle in control group rats and rats treated with silymarin and fenugreek seeds extract were normal. Severe degenerative changes associated with nuclear pyknosis of the sarcoplasm were seen in rats treated with nandrolone decanoate. The toxic effect of AS could be improved with treatment by silymarin and fenugreek seeds extract. Also, Kahal and Allem, 2018 [44] showed in their study elongation, severe degeneration and may be rupture of the cardiac muscle. Also, these results coincide with Hassan et al. [ 2009] [45]“AS sustaining induces severe ischemic necrosis and degeneration of the cardiac muscle fibers in male albino rats.” Abdelhafez, 2014 [46] showed highly degenerated muscle fibers with areas of hemorrhage and widened endomysium. Also, she demonstrated a numerous pyknotic and karyolytic nuclei. Also, Elgendy et al. ,2018 [10] reported hypertrophy and degeneration of both cardiac and skeletal muscles and explained this by its effect on the androgen receptors that are widely distributed in different types of muscles. The defensive effect of silymarin against cardiac damage was studied by Avci et al. ,2017 [36] who reported that silymarin improve the cardiotoxic histopathological effects induced by cyclophosphamide. Other study performed by Aktas and Ozgocmen, 2020 [37] showed that silymarin preserves the histological structure of the heart and effectively promotes the antioxidant defense system against valproic acid induced injuries. Also, El-Shitany et al., 2008 [47] demonstrated an improvement in the histopathological effects in Adriamycin induced cardiotoxicity when combined with silymarin and explained this by the antioxidant activity of silymarin inhibiting the lipid peroxidation and promoting antioxidant enzymes activity. Mendoza et al., 2020 [48] demonstrated a protective effect of silymarin on cardiac and skeletal muscle injuries due to its potent antioxidant effects. ## Conclusions Anabolic Steroids have a toxic effect on cardiac and skeletal muscles associated with alteration of the biochemical markers, oxidative stress, and histopathological changes. However, fenugreek seeds extract and silymarin improved the toxic effects of AS on the cardiac and skeletal muscles with better results of fenugreek seeds extract. ## References 1. Reyes-Vallejo L. Current use and abuse of anabolic steroids. Actas Urológicas Españolas (English Edition) (2020.0) 44 309-313. DOI: 10.1016/j.acuroe.2019.10.007 2. 2.Patanè FG, Liberto A, Maria Maglitto AN, Malandrino P, Esposito M, Amico F, et al. Nandrolone decanoate: use, abuse and side effects. Medicina. 2020:1–26. 3. Monda V, Salerno M, Fiorenzo M, Villano I, Viggiano A, Sessa F, Triggiani AI, Cibelli G, Valenzano A, Marsala G. Role of sex hormones in the control of vegetative and metabolicfunctions of middle-agedwomen. Europe PMC (2017.0). DOI: 10.3389/fphys.2017.00773 4. Pardridge WM. 4 Serum bioavailability of sex steroid hormones. Clin Endocrinol Metab (1986.0) 15 259-278. DOI: 10.1016/s0300-595x(86)80024-x 5. Llewellyn W. Anabolics Mol Nutr (2011.0) 193–194 402-412 6. Basaria S, Wahlstrom JT, Dobs AS. Clinical review 138: Anabolic-androgenic steroid therapy in the treatment of chronic diseases. J Clin Endocrinol Metab (2001.0) 86 5108-5117. DOI: 10.1210/jcem.86.11.7983 7. 7.Srivastava A, Singh Z, Verma V, Choedon T. Potential health benefits of fenugreek with multiple pharmacological properties. In Ethnopharmacological Investigation of Indian Spices, IGI Global; 2020. p. 137–53. 8. Badi HN, Mehrafarin A, Mustafavi SH, Labbafi M. Exogenous arginine improved fenugreek sprouts growth and trigonelline production under salinity condition. Ind Crops Prod (2018.0) 122 609-616. DOI: 10.1016/j.indcrop.2018.06.042 9. Marmouzi I, Bouyahya A, Ezzat SM, El Jemli M, Kharbach M. The food plant Silybum marianum (L.) Gaertn.: Phytochemistry, Ethnopharmacology and clinical evidence. Journal of Ethnopharmacology. (2020.0) 265 113303. DOI: 10.1016/j.jep.2020.113303 10. Elgendy HAE, Alhawary AAE, El-Shahat MAE, Ali AT. Effect of anabolic steroids on the cardiac and skeletal muscles of adult male rats. International Journal of Clinical and Developmental Anatomy (2018.0) 4 1. DOI: 10.11648/j.ijcda.20180401.11 11. Mwaheb MA, Mohammed ARS, Al-Galad GM, Abd-Elgayd AA, Al-hamboly HM. Effect of nandrolone decanoate (anabolic steroid) on the liver and kidney of male albino rats and the role of antioxidant (antox-silymarin) as adjuvant therapy. J Drug Metab Toxicol (2017.0) 8 1-11. DOI: 10.4172/2157-7609.1000224 12. Sakr SA, Shalaby SY. Effect of fenugreek seed extract on carbofuran-inhibited spermatogenesis and induced apoptosis in albino rats.". Journal of Infertility and Reproductive Biology (2014.0) 2 36-42 13. Hind B, Zineb M, Elbachir H, Najat EA, Siham A, Driss R. Evaluation of potential effects of the aqueous extract of fenugreek seeds on fertility in male rats. Journal of Ayurvedic and Herbal Medicine (2017.0) 3 210-215. DOI: 10.31254/jahm.2017.3408 14. Uchiyama M, Mihara M. Determination of malonaldehyde precursor in tissues by thiobarbituric acid test. Anal Biochem (1978.0) 86 271-278. DOI: 10.1016/0003-2697(78)90342-1 15. Sedlak J, Lindsay RH. Estimation of total, protein-bound, and nonprotein sulfhydryl groups in tissue with Ellman's reagent. Anal Biochem (1968.0) 25 192-205. DOI: 10.1016/0003-2697(68)90092-4 16. Paoletti F, Aldinucci D, Mocali A, Caparrini A. A sensitive spectrophotometric method for the determination of superoxide dismutase activity in tissue extracts. Anal Biochem (1986.0) 154 536-541. DOI: 10.1016/0003-2697(86)90026-6 17. Tasgin E, Lok S, Demir N. Combined usage of testosterone and nandrolone may cause heart damage. Afr J Biotech (2011.0) 10 3766-3768 18. Kulaksiz Ö, Sefa LÖK. Investigating the effect of testosterone supplement on heart and muscle damage in rats applied with swimming exercise. Türk Spor ve Egzersiz Dergisi (2019.0) 21 170-174 19. 19.Taghiabadi E, Imenshahidi M, Abnous K, Mosafa F, Sankian M, Memar B, Karimi G. Protective effect of silymarin against acrolein-induced cardiotoxicity in mice. Evid Based Complement Alternat Med. 2012:1–14. 20. Alabdan MA. Silymarin ameliorates metabolic risk factors and protects against cardiac apoptosis in streptozotocin-induced diabetic rats. Biomed Biotechnol (2015.0) 3 20-27 21. Pradeep SR, Srinivasan K. Alleviation of Cardiac Damage by Dietary Fenugreek (Trigonella foenum-graecum) Seeds is Potentiated by Onion (Allium cepa) in Experimental Diabetic Rats via Blocking Renin-Angiotensin System. Cardiovasc Toxicol (2018.0) 18 221-231. DOI: 10.1007/s12012-017-9431-1 22. Kamble HV, Bodhankar SL. Cardioprotective effect of concomitant administration of trigonelline and sitagliptin on cardiac biomarkers, lipid levels, electrocardiographic and heamodynamic modulation on cardiomyopathy in diabetic Wistar rats. Biomedicine & Aging Pathology (2014.0) 4 335-342. DOI: 10.1016/j.biomag.2014.07.009 23. Silva VAP, Boaventura GT, Abboud RS, Ribas JAS, Chagas MA. Consumption of Green Tea (Camellia sinensis) Improves Lipid, Hepatic, and Hematological Profiles of Rats That Are Submitted to Long-Term Androgenic Stimulation. American Journal of Sports Science (2018.0) 6 7 24. Albano GD, Amico F, Cocimano G, Liberto A, Maglietta F, Esposito M, Rosi GL, Di Nunno N, Salerno M, Montana A. January. Adverse Effects of Anabolic-Androgenic Steroids: A Literature Review. In Healthcare: Multidisciplinary Digital Publishing Institute. (2021.0) 9 97. DOI: 10.3390/healthcare9010097 25. Al-Chalabi SM, Abdul-Lattif RF, Al-Mahdawi FA, Abud HN. Effect of Fenugreek (Trigonella foenum graecum) Seed Aqueous Extract on Blood Glucose, Lipid Profile and Some Hormonal Assay in Streptozotocin-induced Diabetic Male Albino Rats. International Journal of Drug Delivery Technology (2019.0) 9 19-25 26. Hannan JMA, Rokeya B, Faruque O, Nahar N, Mosihuzzaman M, Khan AA, Ali L. Effect of soluble dietary fibre fraction of Trigonella foenum graecum on glycemic, insulinemic, lipidemic and platelet aggregation status of Type 2 diabetic model rats. J Ethnopharmacol (2003.0) 88 73-77. DOI: 10.1016/S0378-8741(03)00190-9 27. Roberts KT. The potential of fenugreek (Trigonella foenum-graecum) as a functional food and nutraceutical and its effects on glycemia and lipidemia. J Med Food (2011.0) 14 1485-1489. DOI: 10.1089/jmf.2011.0002 28. Metwally MAA, El-Gellal AM, El-Sawaisi SM. Effects of silymarin on lipid metabolism in rats. World Appl Sci J (2009.0) 6 1634-1637 29. Gobalakrishnan S, Asirvatham SS, Janarthanam V. Effect of silybin on lipid profile in hypercholesterolaemic rats. Journal of clinical and diagnostic research: JCDR (2016.0) 10 1 30. Krępa SE, Kłapcińska B, Jagsz S, Sobczak A, Chrapusta SJ, Chalimoniuk M, Grieb P, Poprzęcki S, Langfort J. High-dose testosterone propionate treatment reverses the effects of endurance training on myocardial antioxidant defenses in adolescent male rats. Cardiovasc Toxicol (2011.0) 11 118-127. DOI: 10.1007/s12012-011-9105-3 31. Tothova L, Celec P, Ostatníková D, Okuliarová M, Zeman M, Hodosy J. Effect of exogenous testosterone on oxidative status of the testes in adult male rats. Andrologia (2013.0) 45 417-423. DOI: 10.1111/and.12032 32. Frankenfeld SP, Oliveira LP, Ortenzi VH, Rego-Monteiro IC, Chaves EA, Ferreira AC, Leitão AC, Carvalho DP, Fortunato RS. The anabolic androgenic steroid nandrolone decanoate disrupts redox homeostasis in liver, heart and kidney of male Wistar rats. PLoS ONE (2014.0) 9 102699. DOI: 10.1371/journal.pone.0102699 33. Pey A, Saborido A, Blázquez I, Delgado J, Megías A. Effects of prolonged stanozolol treatment on antioxidant enzyme activities, oxidative stress markers, and heat shock protein HSP72 levels in rat liver. The Journal of steroid biochemistry and molecular biology (2003.0) 87 269-277. DOI: 10.1016/j.jsbmb.2003.09.001 34. Langfort J, Jagsz S, Dobrzyn P, Brzezinska Z, Klapcinska B, Galbo H, Gorski J. Testosterone affects hormone-sensitive lipase (HSL) activity and lipid metabolism in the left ventricle. Biochem Biophys Res Commun (2010.0) 399 670-676. DOI: 10.1016/j.bbrc.2010.07.140 35. Niki E, Yoshida Y, Saito Y, Noguchi N. Lipid peroxidation: mechanisms, inhibition, and biological effects. Biochem Biophys Res Commun (2005.0) 338 668-676. DOI: 10.1016/j.bbrc.2005.08.072 36. Avci H, Epikmen ET, İpek E, Tunca R, Birincioglu SS, Akşit H, Sekkin S, Akkoç AN, Boyacioglu M. Protective effects of silymarin and curcumin on cyclophosphamide-induced cardiotoxicity. Exp Toxicol Pathol (2017.0) 69 317-327. DOI: 10.1016/j.etp.2017.02.002 37. Aktas I, Ozgocmen M. The treatment effect of silymarin on heart damage in rats. Annals of Medical Research (2020.0) 27 948-954. DOI: 10.5455/annalsmedres.2019.10.643 38. Surai PF. Silymarin as a natural antioxidant: an overview of the current evidence and perspectives. Antioxidants (2015.0) 4 204-247. DOI: 10.3390/antiox4010204 39. Nencini C, Giorgi G, Micheli L. Protective effect of silymarin on oxidative stress in rat brain. Phytomedicine (2007.0) 14 129-135. DOI: 10.1016/j.phymed.2006.02.005 40. Yaman T, Uyar A, Kaya MS, Keles ÖF, Uslu BA, Yener Z. Protective effects of silymarin on methotrexate-induced damages in rat testes. Braz J Pharm Sci (2018.0) 54 17529. DOI: 10.1590/s2175-97902018000117529 41. Bafadam S, Mahmoudabady M, Niazmand S, Rezaee SA, Soukhtanloo M. Cardioprotective effects of Fenugreek (Trigonella foenum-graceum) seed extract in streptozotocin induced diabetic rats. Journal of Cardiovascular and Thoracic Research (2021.0) 13 28. DOI: 10.34172/jcvtr.2021.01 42. Arshadi S, Bakhtiyari S, Haghani K, Valizadeh A. Effects of fenugreek seed extract and swimming endurance training on plasma glucose and cardiac antioxidant enzymes activity in streptozotocin-induced diabetic rats. Osong public health and research perspectives (2015.0) 6 87-93. DOI: 10.1016/j.phrp.2014.12.007 43. Khole S, Chatterjee S, Variyar P, Sharma A, Devasagayam TPA, Ghaskadbi S. Bioactive constituents of germinated fenugreek seeds with strong antioxidant potential. Journal of functional foods (2014.0) 6 270-279. DOI: 10.1016/j.jff.2013.10.016 44. Kahal A, Allem R. Reversible effects of anabolic steroid abuse on cyto-architectures of the heart, kidneys and testis in adult male mice. Biomed Pharmacother (2018.0) 106 917-922. DOI: 10.1016/j.biopha.2018.07.038 45. Hassan NA, Salem MF, Sayed MAEL. Doping and effects of anabolic androgenic steroids on the heart: histological, ultrastructural, and echocardiographic assessment in strength athletes. Hum Exp Toxicol (2009.0) 28 273-283. DOI: 10.1177/0960327109104821 46. Abdelhafez HM. Histological, histochemical and ultrastructural studies on the effect of Deca-Durabolin and whey protein isolate on cardiac muscle in adult male albino rats. International Journal. (2014.0) 2 164-187 47. El-Shitany NA, El-Haggar S, El-Desoky K. Silymarin prevents adriamycin-induced cardiotoxicity and nephrotoxicity in rats. Food Chem Toxicol (2008.0) 46 2422-2428. DOI: 10.1016/j.fct.2008.03.033 48. Mendoza VN, Ángeles-Valencia M, Madrigal-Santillán EO, Morales-Martínez M, Tirado-Lule JM, Solano-Urrusquieta A, Madrigal-Bujaidar E, Álvarez-González I, Fregoso-Aguilar T, Morales-González Á, Morales-González JA. Effect of Silymarin Supplementation on Physical Performance, Muscle and Myocardium Histological Changes, Bodyweight, and Food Consumption in Rats Subjected to Regular Exercise Training. Int J Mol Sci (2020.0) 21 7724. DOI: 10.3390/ijms21207724
--- title: 4-Octyl itaconate attenuates glycemic deterioration by regulating macrophage polarization in mouse models of type 1 diabetes authors: - Sunyue He - Yuchen Zhao - Guoxing Wang - Qiaofang Ke - Nan Wu - Lusi Lu - Jiahua Wu - Shuiya Sun - Weihua Jin - Wenjing Zhang - Jiaqiang Zhou journal: Molecular Medicine year: 2023 pmcid: PMC10015936 doi: 10.1186/s10020-023-00626-5 license: CC BY 4.0 --- # 4-Octyl itaconate attenuates glycemic deterioration by regulating macrophage polarization in mouse models of type 1 diabetes ## Abstract ### Background Pancreatic beta cell dysfunction and activated macrophage infiltration are early features in type 1 diabetes pathogenesis. A tricarboxylic acid cycle metabolite that can strongly activate NF-E2-related factor 2 (Nrf2) in macrophages, itaconate is important in a series of inflammatory-associated diseases via anti-inflammatory and antioxidant properties. However, its role in type 1 diabetes is unclear. We used 4-octyl itaconate (OI), the cell-permeable itaconate derivate, to explore its preventative and therapeutic effects in mouse models of type 1 diabetes and the potential mechanism of macrophage phenotype reprogramming. ### Methods The mouse models of streptozotocin (STZ)-induced type 1 diabetes and spontaneous autoimmune diabetes were used to evaluate the preventative and therapeutic effects of OI, which were performed by measuring blood glucose, insulin level, pro- and anti-inflammatory cytokine secretion, histopathology examination, flow cytometry, and islet proteomics. The protective effect and mechanism of OI were examined via peritoneal macrophages isolated from STZ-induced diabetic mice and co-cultured MIN6 cells with OI-pre-treated inflammatory macrophages in vitro. Moreover, the inflammatory status of peripheral blood mononuclear cells (PBMCs) from type 1 diabetes patients was evaluated after OI treatment. ### Results OI ameliorated glycemic deterioration, increased systemic insulin level, and improved glucose metabolism in STZ-induced diabetic mice and non-obese diabetic (NOD) mice. OI intervention significantly restored the islet insulitis and beta cell function. OI did not alter the macrophage count but significantly downregulated the proportion of M1 macrophages. Additionally, OI significantly inhibited MAPK activation in macrophages to attenuate the macrophage inflammatory response, eventually improving beta cell dysfunction in vitro. Furthermore, we detected higher IL-1β production upon lipopolysaccharide stimulation in the PBMCs from type 1 diabetes patients, which was attenuated by OI treatment. ### Conclusions These results provided the first evidence to date that OI can prevent the progression of glycemic deterioration, excessive inflammation, and beta cell dysfunction predominantly mediated by restricting macrophage M1 polarization in mouse models of type 1 diabetes. ### Supplementary Information The online version contains supplementary material available at 10.1186/s10020-023-00626-5. ## Introduction Type 1 diabetes is a progressive autoimmune disease characterized by inflammatory cell infiltration into islets and subsequently results in absolute insulin deficiency (Eizirik et al. 2009; Lucier and Weinstock 2022). Autoimmunity pathogenesis is difficult to reverse once the progression to type 1 diabetes begins. There has been much focus in the past decade on new approaches such as islet allografting (Pepper et al. 2018; Ridler 2016), gene therapy (Chellappan et al. 2018), stem cell therapy (Chen et al. 2020), and immunotherapies (Ni et al. 2019). However, these therapies are limited due to various reasons and have not been widely applied or promoted in clinical practice. On this note, a new paradigm targeting multiple pathogenic pathways is urgently needed for preventing or treating type 1 diabetes. The autoimmune response triggering type 1 diabetes relies on the crosstalk between pancreatic beta cells and the immune system. Histology insulitis results obtained from experimental models of type 1 diabetes and type 1 diabetes patients demonstrate that macrophages and dendritic cells are the first cells that infiltrate into the islets (Nagy et al. 1989; Dahlén et al. 1998; Jörns et al. 2014; Walker et al. 1988; Hanenberg et al. 1989). Macrophage polarization is defined into two broad subsets: the classically activated M1 macrophages and alternatively activated M2 macrophages. The M1-type macrophages initiate insulitis and beta cell destruction whereas the M2-type macrophages suppress the immune response and are protective against type 1 diabetes (Espinoza-Jiménez et al. 2012). Therefore, the balance between M1 and M2 macrophage activation and polarization is crucial for disease progression. The cellular metabolic adaptation to immune responses, immunometabolism is important in regulating the immune function of cells (Diskin and Palsson-McDermott 2018). Metabolic reprogramming is important in macrophage phenotype transition. Increased glycolysis and decreased oxidative phosphorylation are the main metabolic characteristics of M1-type macrophages (Torres et al. 2016). Alongside the enhanced glycolysis, tricarboxylic acid (TCA) cycle interruption is another significant characteristic of inflammatory macrophages, where metabolic intermediates accumulate and function as regulatory mediators of inflammatory responses (Ryan et al. 2019; Murphy and O'Neill 2018). A derivative diverted from the TCA cycle, itaconate has recently become the focus of the immunometabolism field due to its anti-inflammatory properties that negatively regulate cytokine production and the inflammatory response (Mills et al. 2018). In accumulating studies, it was reported that itaconate is important in inflammatory-associated diseases via anti-inflammatory and antioxidant properties, such as acute lung injury (Xin et al. 2021; Li et al. 2020; Liu et al. 2021), autoimmune hepatitis (Yang et al. 2022), vitiligo (Xie et al. 2022), ischemia–reperfusion injury (Yi et al. 2020; Cordes et al. 2020), osteoclast-related diseases (Sun et al. 2019), cancer (Zhan et al. 2022), and renal fibrosis (Tian et al. 2020). The functions attributed to itaconate predominantly include aerobic glycolysis inhibition and transcription factor NF-E2-related factor 2 (Nrf2) activation (Mills et al. 2018; Liao et al. 2019). Notably, Nrf2 signaling activation contributed to ameliorating inflammation-mediated autoimmune disorders. Recently, it was proven that systemic activation of Nrf2 signaling delayed the onset of type 1 diabetes in spontaneous non-obese diabetic (NOD) mice (Yagishita et al. 2019), which suggested that Nrf2 is a potential target for preventing and treating type 1 diabetes. The cell-permeable itaconate derivative 4-octyl itaconate (OI) inhibited proinflammatory cytokine production in macrophages and reprogrammed them into an M2-like phenotype with a significant anti-inflammatory property (Lampropoulou et al. 2016; Tang et al. 2018). However, the effect of OI on inflammation in type 1 diabetes remains unclear. The critical effect of macrophage phenotype transition in innate immunity and the strong influence of OI on immune responses through Nrf2 activation prompted this investigation of the use of OI as a preventative, or even therapeutic, modality to deter type 1 diabetes progression in mouse models and exploration of the potential mechanism of macrophage phenotype reprogramming. ## Animals Six-week-old male C57BL/6 mice (19–22 g) and 4-week-old female NOD mice (15–18 g) were obtained from the Model Animal Research Center (Nanjing, China). All mice were housed in a standard 12-h light–dark cycle and had free access to a normal chow diet with water ad libitum. The animal experiments were performed following national and institutional guidelines for the care and use of animals. ## Type 1 diabetes models and OI treatment A male C57BL/6 mouse model of type 1 diabetes was established by multiple low doses of streptozotocin (STZ) (ESM Methods: Type 1 diabetes induction). In the prevention mouse model, OI (25 mg/kg, China) was dissolved in 2-hydroxypropyl-β-cyclodextrin (HBPCD, $45\%$ wt/vol, China) in PBS and administered intraperitoneally 5 days before the first STZ dose and for 6 weeks (Fig. 1A). The OI and STZ injections were administered 3 h apart. To confirm the effect of OI on diabetes prevention, the incidence of spontaneous autoimmune diabetes was assessed in female NOD mice. OI or HBPCD (control, CTR) was administered intraperitoneally once daily for up to 42 weeks (starting at the age of 4 weeks). To investigate whether OI could reverse the hyperglycemia in type 1 diabetes, the STZ-induced diabetic mice received OI daily, which was initiated after hyperglycemia onset and continued for 12 weeks (Fig. 4A). ## Glucose tolerance test (GTT) All C57BL/6 mice underwent the oral GTT (OGTT). Briefly, after overnight fasting with water ad libitum, mice were gavaged with a glucose solution in saline (1 g/kg using $20\%$ dextrose solution). The NOD mice underwent the intraperitoneal GTT (ipGTT). Briefly, after overnight fasting with water ad libitum, the mice were injected intraperitoneally with a glucose solution in saline (2 g/kg using $20\%$ dextrose solution). The blood glucose level from the tail vein was measured at baseline and at 15, 30, 60, and 120 min after glucose administration using a OneTouch Ultra blood glucose analyzer. ## Histological analysis The mouse pancreases were paraffin-embedded and sectioned to 4 μm thickness. The sections were stained with Mayer’s hematoxylin to evaluate the mononuclear cell infiltration in the islets (ESM Methods: Insulitis score). For the TUNEL assay, the sections were stained with TUNEL assay (In Situ Cell Death Detection Kit, Roche). For immunofluorescence staining, the sections were assayed with DAPI as a counterstain. The primary antibodies are listed in Additional file 1: Table S1. Immunohistochemical analysis of inducible nitric oxide synthase (iNOS), F$\frac{4}{80}$, high-mobility group box 1 protein (HMGB1), and Nrf2 was performed as previously described (Yang et al. 2022). The relative content and positive cells in the islets were determined by ImageJ. ## Islet isolation and proteomics The mouse pancreatic islets were isolated by collagenase type V digestion and purified using a density gradient as described previously (O'Dowd and Stocker 2020). The islets were washed in Hanks’ balanced salt solution (HBSS) and carefully hand-selected. The islet proteomics analysis in this study was performed by Jingjie PTM BioLabs (Hangzhou, China) (ESM Methods: Proteomics). To analyze the proteomics data, we used Venn diagrams and Mfuzz analysis to better visualize and identify the differentially expressed proteins (DFEs) between the three groups. To investigate the effect of OI on STZ-induced diabetes, we created a volcano plot and heat map of the pancreatic beta cell function in the STZ and OI groups. Moreover, we used the KEGG pathway database to annotate the DFEs pathways. ## Cell isolation and culture Spleen and pancreatic lymph nodes (PLNs) were mechanically disassociated and passed through a 40-μm cell strainer to harvest single-cell suspensions. Splenocytes were acquired after removing erythrocytes with red blood cell (RBC) lysis buffer. Peritoneal macrophages were isolated from the control and STZ mice by i.p. injection of 2 ml $3\%$ thioglycolate per mouse for 2 consecutive days. On day 4, the cells were harvested by peritoneal lavage of 10 ml DMEM containing $0.1\%$ penicillin and streptomycin. After 10-min centrifugation at 400×g, the cells were washed with PBS once, resuspended in culture medium, and seeded in 6-well plates at 5 × 106 cells/well. After adherence overnight in the incubator, non-adherent cells were removed by washing with PBS and changed to fresh medium before the intervention. Murine MIN6 beta cells were cultured in DMEM containing $10\%$ FBS, 50 μM β-mercaptoethanol, and $0.1\%$ penicillin and streptomycin. Peripheral blood mononuclear cells (PBMCs) were isolated from a 5 ml fresh whole blood sample using density gradient separation with Histopaque®-1077 (Sigma Aldrich). The cells were then plated in 12-well plates at 5 × 107 cells/well. After adhering for 3 h in the incubator, the PBMCs that failed to adhere were removed by washing with cold PBS and changed to fresh medium before the intervention. ## Flow cytometry The macrophage phenotype in the mouse PLNs was evaluated by flow cytometry using the following antibodies: CD$\frac{16}{32}$, CD45, CD11b, CD206 (all from BioLegend), and CD11c (Ebioscience). After 20-min blocking with CD$\frac{16}{32}$ at 4 °C, the surface markers were subsequently stained for 30 min at 4 °C. The cells were acquired on a BD LSRFortessa and analyzed by FlowJo. The data were first gated on live cells through viability dye, then further gated according to the required analysis. ## Glucose-stimulated insulin secretion (GSIS) Peritoneal macrophages were obtained by peritoneal lavage. MIN6 cells were co-cultured with inflammatory macrophages or OI-pretreated macrophages for 24 h in transwell chambers (12-well plates) containing 0.4-μm pore filters (Fig. 7H). The MIN6 cells were washed with pre-warmed PBS, then pre-incubated in pre-warmed KRB containing 2.8 mM glucose for 0.5 h in a humidified atmosphere of $5\%$ CO2 at 37 °C. Subsequently, the cells were incubated with pre-warmed KRB containing either 2.8 mM or 25 mM glucose for another 1 h. The collected supernatant was centrifuged at 400×g for 10 min to remove cells and debris. Then, the insulin content was measured using an ELISA kit (Ezassy, China) following the manufacturer’s instructions and normalized with the protein content. ## Participants Nine non-diabetes participants aged > 18 years were included in the study. Ten type 1 diabetes patients were recruited from the diabetic ward of Sir Run Run Shaw Hospital of Zhejiang University. The patients comprised eight men and two women (mean age: 41.7 ± 15.2 years). Of these, 3 patients presented positive auto-immunity antibody. All patients were accompanied by ketosis or ketoacidosis. The fasting C-peptide in type 1 diabetes patients was below the lower limit of normal range. The clinical characteristic of the type 1 diabetes patients is summarized in Additional file 1: Table S3. ## Peripheral blood mononuclear cell (PBMC) stimulation The PBMCs from non-diabetes participants and type 1 diabetes patients were pre-treated with 125 μM OI or NT for 2 h, followed by treatment with 200 ng/ml lipopolysaccharide (LPS), 125 μM OI + 200 ng/ml LPS (OI + LPS), or NT for another 24 h. The collected supernatant was centrifuged at 400×g for 10 min to remove cells and debris, and the cytokine concentrations were analyzed with an ELISA kit (Multisciences, China) following the manufacturer’s instructions. ## Statistical analysis The data are expressed as the mean ± SEM for animal experiments or mean ± SD for cell experiments. The number of independent animal experiments is shown in the related figure legends, where n refers to the number of mice per group. All cell experiments were repeated at least three times. Statistical differences were analyzed using Student’s t-test and analysis of variance (ANOVA) (GraphPad Prism 9). A p-value < 0.05 indicated statistical significance. The difference in diabetes onset incidence between two groups of NOD mice was determined using the Mantel-Cox log-rank test. ## OI intervention improved glucose metabolism in mouse models of type 1 diabetes In the prevention mouse model, OI was started 5 days before the first STZ dose and was administered for 6 weeks. Although the animals receiving OI exhibited similar body weight gain to the STZ-induced diabetic mice at the end of the study (Fig. 1B) OI prevented glycemic deterioration and improved the islet quantity of the animals that received STZ (Fig. 1C, I). During the OGTT, the OI mice exhibited significantly lower blood glucose levels than the STZ-alone mice at all time points (Fig. 1D, E). The improved glucose tolerance in the OI group was primarily attributed to the elevated plasma insulin level (Fig. 1G). Importantly, we did not detect adverse effects or liver and kidney function impairment after OI treatment (Additional file 1: Figures S1, S2). Besides, the spontaneous autoimmune diabetes model was used for further study. Five of 14 control NOD mice developed diabetes at 14–41 weeks of age. In contrast, none of the OI intervention mice were diabetic by the end of the study, which was significantly different ($$p \leq 0.012$$, Mantel-Cox log-rank test) (Fig. 3A). During the ipGTT, the NOD mice maintained on OI administration had better glucose homeostasis than the control NOD mice (Fig. 3B, C). The serum insulin level was significantly increased in the NOD mice with OI intervention compared to the control (Fig. 3D). To investigate whether OI could reverse the hyperglycemia in type 1 diabetes, OI treatment was initiated after hyperglycemia onset, where similar results of improved glucose homeostasis were observed in the treated type 1 diabetic mouse model (Fig. 4C, D).Fig. 1OI prevented glycemic deterioration, improved glucose metabolism and restored systemic inflammation in STZ-induced diabetes. A Schematic representation of the experimental protocol. B Mouse body weight change ($$n = 15$$–19 mice per group). C Random blood glucose measurements were performed throughout the study ($$n = 15$$–19 mice per group). D Week 5 OGTT results. E The AUC from the OGTT results in D ($$n = 6$$–13 mice per group). F Fasting blood glucose ($$n = 15$$–18 mice per group). G Fasting serum insulin concentration ($$n = 8$$ mice per group). H HOMA-β index ($$n = 7$$–8 mice per group). I Islet quantity from the mouse pancreas ($$n = 6$$–8 mice per group). J–O Serum cytokine concentrations ($$n = 5$$–8 mice per group). P The proportion of neutrophils (Neu) and lymphocyte (Lym) in mouse peripheral blood ($$n = 12$$–17 mice per group). All values are shown as the mean ± SEM. * $p \leq 0.05$, $p \leq 0.01$, $p \leq 0.001$, and **$p \leq 0.0001.$ CTR, control ## OI intervention restored the cytokine secretion profiles in STZ-induced diabetes OI intervention significantly decreased the proinflammatory factors (IL-1β, TNF-α, IL-2, IL-4, IFN-γ) in the blood of STZ-induced diabetic mice but increased the levels of IL-10, which is an anti-inflammatory characteristic (Fig. 1J–O). In the treated type 1 diabetes mouse model, serum TNF-α, IL-12, and IFN-γ secretion were greatly reduced in the OI group, whereas CXCL1 and IL-10 production was restored (Additional file 1: Figure S7). ## OI intervention reduced pancreatic insulitis and restored beta cell function Pancreatic H&E staining revealed that most islets in the STZ-induced diabetic mice exhibited severe insulitis and unclear margins, while the OI-treated group had a sharply reduced proportion of severe and invasive insulitis and restored islet structure (Fig. 2A). In NOD mouse pancreas, excessive inflammatory cells surrounded or invaded into islets, which OI significantly decreased (Fig. 3E). TUNEL assay demonstrated that the number of apoptotic cells was obviously increased in the STZ-induced diabetic mouse islets, which OI attenuated (Additional file 1: Figure S3). The islets in the STZ-induced diabetic animals contained very few insulin-positive cells and strong centralization of glucagon-staining cells (Fig. 2B). In marked contrast, OI administration increased the presence of functional insulin-positive beta cells and reduced glucagon-positive alpha cells similar to that in healthy islets. Principal component analysis of islets proteomics revealed distinct sample group clustering in the three groups (control, STZ, OI), with OI administration exerting less pronounced effects than STZ as compared with the control group (Fig. 2C). The DFEs in the OI vs. STZ groups are depicted in a volcano plot (Fig. 2D). The DFEs contained proteins relating to beta cell function, including MFN2, PRKCA, SLC2A2, PCSK1, INS1, INS2, GLP1R, and TSC1, which are summarized in a heat map (Fig. 2E). In the NOD mouse model, the insulin-positive cells were markedly destroyed by immune cells whereas they were restored by OI intervention (Fig. 3F). Similar findings of attenuated beta cell damage were also observed in the treated type 1 diabetic mouse model (Fig. 4E, F).Fig. 2OI attenuated insulitis, improved beta cell function, and suppressed inflammatory macrophage infiltration in STZ-induced diabetes. A Representative images of pancreatic histology evaluated by H&E staining ($$n = 5$$–8 mice per group). The scoring criteria were as described in the ESM. B Representative images of pancreatic immunofluorescence staining for insulin (red), glucagon (green), and DAPI (blue). Insulin expression per islet was determined by analyzing fluorescence intensity with ImageJ. The percentage of glucagon-positive cells per islet was calculated in a bar graph ($$n = 3$$–5 mice per group). C Principal component analysis plot demonstrating the islet sample distribution in the control, STZ, and OI groups. The plot displays three distinct clusters ($$n = 3$$ mice per group). D Volcano plot depicting the 130 DFEs identified in the STZ and OI islets. E Heat map analysis illustrating the DFEs associated with islet beta cell function. The color change from blue to orange indicates low to high expression values. F List of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways significantly associated with the DFEs identified in the STZ and OI groups. G Immunohistochemistry analysis* of Nrf2, HMGB1, F$\frac{4}{80}$, and iNOS expression in the mouse pancreas. Nrf2 expression was analyzed by ImageJ and presented as the Average Optical Density (AOD). The percentage of HMGB1-positive nuclei per islet was quantified in a bar graph. The F$\frac{4}{80}$- and iNOS-positive cells per islet area were calculated in a bar graph ($$n = 3$$–6 mice per group). H Flow cytometry assessment of the proportion of CD45+CD11b+, CD11b+CD11c+, and CD11b+CD206+ macrophages in PLNs along with representative dot plots ($$n = 4$$ mice per group). I F$\frac{4}{80}$, Cd11c, and Mrc1 gene expression in splenocytes ($$n = 8$$ mice per group). All values are the mean ± SEM. * $p \leq 0.05.$ All values are shown as the mean ± SEM. $p \leq 0.01$, $p \leq 0.001$, and **$p \leq 0.0001$Fig. 3OI treatment reduced diabetes onset in spontaneous non-obese diabetic (NOD) mice. A Kaplan–Meier survival curves demonstrating the percentages of diabetes-free NOD mice ($$n = 14$$–15 mice per group). B Week 22 ipGTT results of NOD mice. C The AUC of the ipGTT results in B ($$n = 12$$–15 mice per group). D Serum insulin concentration of NOD mice ($$n = 12$$–13 mice per group). E Representative images of pancreatic histology evaluated by H&E staining ($$n = 11$$–12 mice per group). F Representative images of pancreatic immunofluorescence staining for insulin (red), glucagon (green), and DAPI (blue). G Flow cytometry assessment and the proportion of CD45+CD11b+, CD11b+CD11c+, and CD11b+CD206+ macrophages in PLNs and the representative dot plots ($$n = 8$$ mice per group). All values are shown as the mean ± SEM. $p \leq 0.05$ and *$p \leq 0.01.$ CTR, NOD-controlFig. 4The therapeutic effect of OI on glucose homeostasis and inflammation in STZ-induced diabetes. A Schematic representation of the experimental protocol. C57BL/6 mice with STZ-induced diabetes received OI (25 mg/kg) daily after hyperglycemia onset and for 12 weeks. B Mouse body weight change ($$n = 6$$–9 mice per group). C Random blood glucose measurements were performed before the study and at the end of OI treatment. D Week 11 OGTT results. E Representative images of pancreatic histology evaluated by H&E staining ($$n = 5$$–7 mice per group). F Representative images of pancreatic immunofluorescence staining for insulin (red), glucagon (green), and DAPI (blue). Insulin expression per islet was determined by analyzing fluorescence intensity with ImageJ. The percentage of glucagon-positive cells per islet was calculated in a bar graph ($$n = 3$$ mice per group). G Immunohistochemistry analysis* of Nrf2, HMGB1, F$\frac{4}{80}$, and iNOS expression in mouse pancreas. Nrf2 expression was analyzed by ImageJ and presented as the AOD. The percentage of HMGB1-positive nuclei per islet was quantified in a bar graph. The F$\frac{4}{80}$- and iNOS-positive cells per islet area were calculated in a bar graph ($$n = 3$$–5 mice per group). All values are shown as the mean ± SEM. $p \leq 0.01$, $p \leq 0.001$, and **$p \leq 0.0001$ ## OI intervention attenuated oxidative stress and suppressed proinflammatory macrophage infiltration The protective effect of OI predominantly relied on its powerful activation of Nrf2, which is critical in oxidative stress responses (Mills et al. 2018). The results demonstrated greatly decreased Nrf2 expression after the mice received STZ, whereas OI intervention significantly reversed the islet Nrf2 content (Fig. 2G). In some of the infiltrated immune cells, HMGB1 translocated from the nucleus to the extracellular spaces, indicating HMGB1 active secretion (Zhang et al. 2009). Compared with diabetic mice, the islet cell nuclei of the OI-treated mice had a much higher proportion of HMGB1 expression, which implied that HMGB1 secretion was greatly inhibited (Fig. 2G). Although the STZ animals had much higher F$\frac{4}{80}$-positive cell infiltration in the islets compared to the controls, those cells were not significantly changed among the diabetes and OI groups. Surprisingly, the OI treatment mice had much lower numbers of iNOS-positive macrophages associated with the islets (Fig. 2G), which was also observed in the treated type 1 diabetes mouse model (Fig. 4G). Phenotype analysis of PLN cells demonstrated that the STZ-induced diabetic mice had a greatly increased macrophage population (CD45+CD11b+) and percentage of CD11b+CD11c+ (proinflammatory M1) macrophages and a decreased proportion of CD11b+CD206+ (anti-inflammatory M2) macrophages as compared to the control. OI intervention obviously lowered the percentage of M1 macrophages in the PLNs of STZ-induced diabetic mice (Fig. 2H). Furthermore, OI significantly decreased Cd11c gene expression in the STZ-induced diabetic mouse splenocytes and increased Mrc1* gene expression (Fig. 2I). In the NOD mouse model, the OI group had a markedly lower proportion of CD11b+CD11c+ (proinflammatory M1) macrophages in the PLNs as compared to the control (Fig. 3G). ## Effect of OI intervention in vitro on the peritoneal macrophage phenotype and mitogen-activated protein kinase (MAPK) pathway in STZ-induced diabetes Islets proteomics demonstrated that DFEs in the OI vs. STZ groups were significantly enriched in diabetes, insulin secretion, insulin resistance, and in the HIF1-α, mTOR, and MAPK signaling pathways (Fig. 2F). Furthermore, we observed MAPK pathway activity in the peritoneal macrophages from diabetic mice. Consistently, the MAPK pathway was greatly activated in macrophages, which presented as higher levels of phosphorylated (p)-ERK, p-p38, and p-JNK in macrophages from the diabetic animals than in the controls (Fig. 5B, D). Macrophages from the STZ-induced diabetic mice produced more NOD-like receptor thermal protein domain associated protein 3 (NLRP3) and iNOS than the cells from the control mice regardless of whether they were stimulated by LPS (Fig. 5A, C). As expected, OI intervention sharply elevated Nrf2 and its downstream antioxidative molecules (GCLC, GCLM, HO-1, NQO-1). Consistent with this, OI treatment attenuated p-ERK and p-p38 activation but did not affect p-JNK. Notably, OI decreased NLRP3 and iNOS production such that it was even lower than the baseline level in the control mice. Fig. 5The effect of OI on peritoneal macrophages from STZ-induced diabetic mice. The peritoneal macrophages from healthy and diabetic mice were stimulated with OI or NT for 24 h. A, C Representative western blots and protein quantification of Nrf2, HO-1, NQO-1, GCLC, GCLM, NLRP3, and iNOS in peritoneal macrophage lysates ($$n = 4$$ mice per group). The internal control was α-tubulin. Band intensity was measured by ImageJ. The band intensities were quantified and normalized to α-tubulin. B, D Representative western blots and activity quantification of the phosphorylated and total levels of ERK, p38, and JNK in peritoneal macrophage lysates. The internal control was α-tubulin. The results are shown as the mean ± SD. * $p \leq 0.05$, $p \leq 0.01$, $p \leq 0.001$, and **$p \leq 0.0001$ ## Effect of OI intervention in vitro on antioxidative gene expression and macrophage polarization under LPS stimulation LPS induces monocyte/macrophage polarization to the M1 phenotype and was used in our study to activate macrophages. We challenged the diabetic and non-diabetic peritoneal macrophages with 100 ng/ml LPS and examined the expression of the M1/M2-related genes and antioxidative genes. LPS stimulation significantly upregulated IRG1, HO-1, IL-1β, IL-6, TNF-α, iNOS, NLRP3, COX2, SOCS3, and IL-10 and downregulated Nrf2, NQO1, GCLC, GCLM, Mrc1, YM1, and Mgl1* gene expression in the macrophage of control as seen in diabetic mice (Fig. 6). Notably, IL-1β, IL-6, iNOS, NLRP3, and COX2 gene expression was much higher in the diabetic macrophages upon LPS stimulation, whereas TNF-α and IL-10 gene expression was slightly and clearly decreased, respectively. The OI-pre-treated LPS-stimulated peritoneal macrophages had significantly increased HO-1, NQO1, GCLC, GCLM, Mrc1, YM1, and Mgl1 gene expression and reduced IL-1β, IL-6, TNF-α, iNOS, NLRP3, and SOCS3 gene expression, but OI did not affect Nrf2 transcription in the control or diabetic mice. Unexpectedly, OI treatment only attenuated COX2 and IL-10 gene expression in the healthy mouse peritoneal macrophages after LPS challenge. Fig. 6The effect of OI on oxidative stress gene expression and macrophage polarization. The peritoneal macrophages from healthy and diabetic mice were collected and were pretreated with 125 μM OI or NT for 2 h, followed by LPS, OI + LPS, or NT treatment. After 8-h treatment, total RNA was extracted and the expression of the M1-, M2-, and oxidative-related genes were measured by real-time PCR ($$n = 4$$ mice per group). The loading control was β-actin. A M1-related genes expression. B M2-related genes expression. C Oxidative-related genes expression. The results are shown as the mean ± SD. * $p \leq 0.05$, $p \leq 0.01$, $p \leq 0.001$, and **$p \leq 0.0001$ ## Peritoneal macrophages from the type 1 diabetic mice demonstrated an inflammatory state attenuated by OI treatment We examined the proinflammatory cytokine concentrations in the supernatant of LPS-stimulated peritoneal macrophages with or without OI intervention. OI inhibited the release of proinflammatory factors such as IL-1β, IL-6, and TNF-α in the control and diabetic macrophages (Fig. 7B–D). Nitric oxide (NO) generation was more pronounced in the LPS-treated diabetic mice than in the controls, which was markedly suppressed to the basal level in both models (Fig. 7E). Similar to the above gene expression results, the diabetic macrophages contained higher IL-1β, iNOS, NLRP3, and COX2 protein levels compared with the control macrophages upon LPS stimulation, which were all attenuated by OI except for COX2 in diabetic mice (Fig. 7F).Fig. 7Peritoneal macrophages from diabetic mice demonstrated an inflammatory state that was attenuated by OI treatment. The peritoneal cells from healthy and diabetic mice were collected and were pretreated with 125 μM OI or NT for 2 h, followed by LPS, OI + LPS, or NT treatment ($$n = 4$$ mice per group). A Western blot analysis of the phosphorylated and total levels of ERK, p38, and JNK after 0.5-h treatment. B–D The cells and supernatant were collected after 24-h treatment. IL-1β, IL-6, and TNF-α levels were determined by ELISA and corrected with the protein content. E NO generation in the supernatant was measured by the Griess reaction. F Cell lysates underwent western blotting with Nrf2, HO-1, NQO-1, GCLC, GCLM, NLRP3, iNOS, IL-1β, COX2, and α-tubulin antibodies. The internal control was α-tubulin. G PBMCs collected from non-diabetes (CTR) and type 1 diabetic patients were adhered for 3 h, then pretreated with 125 μM OI or NT for 2 h, followed by LPS, OI + LPS, or NT treatment. After 24-h treatment, IL-1β levels were in the supernatant ($$n = 9$$–10 per group). H Peritoneal macrophages were treated with or without LPS and OI for 8 h, then co-cultured with MIN6 cells for another 24 h. The supernatant was collected and GSIS was analyzed. The results are shown as the mean ± SD. $p \leq 0.05$, $p \leq 0.01$, $p \leq 0.001$, and *$p \leq 0.0001.$ I Schematic depicting the regulatory role of OI on macrophage polarization in protection against type 1 diabetes progression After dissociating from KEAP1, Nrf2 accumulates and enters the nucleus to induce antioxidant protein expression (Mills et al. 2018). In this study, OI predominantly increased Nrf2, HO-1, NQO1, GCLC, and GCLM production in the LPS-treated control and diabetic macrophages (Fig. 7F). We also evaluated the MAPK pathway in macrophages after LPS stimulation. As expected, LPS treatment sharply increased the p-ERK, p-p38, and p-JNK levels in the macrophages. In both mouse models, OI reduced the p-ERK and p-p38 levels but not the p-JNK levels (Fig. 7A). ## OI intervention modulated IL-1β production in monocytes derived from type 1 diabetes patients in vitro The above results prompted the investigation of whether OI exerts a similar effect on the human monocyte inflammatory response. The monocytes from type 1 diabetes patients exhibited higher IL-1β secretion potential in response to LPS than the monocytes from non-diabetic donors. Notably, the OI + LPS monocytes presented attenuated IL-1β secretion compared with LPS-only-treated cells from the non-diabetes or type 1 diabetes participants (Fig. 7G). Therefore, OI exhibited an anti-inflammatory impact on the monocytes of type 1 diabetes patients. ## Activated macrophages induced beta cell dysfunction and were restored by OI treatment We designed a co-culture system to clarify the protective effect of OI on beta cell function via regulating macrophage polarization. Co-culture with LPS-stimulated activated macrophages greatly impaired the insulin release of MIN6 cells upon glucose stimulation as compared to co-culture with unstimulated macrophages. Surprisingly, macrophages pre-treated with OI before LPS stimulation significantly restored the impaired GSIS of the MIN6 cells. Moreover, OI pre-treatment before co-culture with activated macrophages improved glucose-stimulated insulin secretion in the MIN6 cells (Fig. 7H). ## Discussion In this study, we demonstrated that the cell-permeable itaconate derivative OI ameliorated glycemic deterioration, improved impaired insulin secretion, and reduced insulitis in STZ-induced diabetic mice and delayed autoimmune diabetes onset in NOD mice. Notably, OI suppressed systemic inflammatory cytokine levels and attenuated pancreatic beta cell damage by inhibiting M1 macrophage activation through the Nrf2–MAPK pathway. Moreover, the macrophages from diabetic mice and monocytes from type 1 diabetes patients presented an inflammatory status that was attenuated by OI treatment. It has been suggested in an increasing number of studies that OI is a novel and promising anti-inflammatory metabolite in restricting immunopathology and inflammation. However, its role in type 1 diabetes disease progression is unclear. Accordingly, we used multiple low-dose STZ-induced diabetes and spontaneous autoimmune diabetes to evaluate the effect of OI on type 1 diabetes prevention and treatment and explore the potential mechanism of macrophage phenotype reprogramming. OI significantly restored pancreatic beta cell function, which was consistent with improved glucose metabolism and less immune cell infiltration within or near the islets as compared to the diabetic mice. These results were also reported in a previous study, where the Nrf2 activator dimethyl fumarate protected beta cells against oxidative stress and reduced the incidence of spontaneous autoimmune diabetes in female NOD mice by attenuating insulitis and the level of circulating proinflammatory cytokines (Li et al. 2021). Cytokines are crucial in orchestrating complex multicellular interactions between pancreatic beta cells and immune cells in type 1 diabetes development. Proinflammatory cytokines are thought to lead to type 1 diabetes onset and progression. By contrast, cytokines that induce regulatory functions are thought to generate feedback regulation of diverse immune responses and protect against beta cell destruction (Lu et al. 2020). Our data demonstrated that the serum proinflammatory cytokines IL-1β, TNF-α, IL-2, IFN-γ, and IL-4 and the anti-inflammatory cytokine IL-10 were dysregulated in diabetes and improved by OI intervention. The effectiveness of OI in ameliorating the glucose metabolism of type 1 diabetes was related to reduced inflammation in the pancreas. Despite the adaptive immune system playing a central role in the inflammation associated with type 1 diabetes, pancreatic beta cell dysfunction and activated M1 macrophage infiltration are early features in the disease pathogenesis (Burg and Tse 2018). Activated macrophages produce a series of inflammatory cytokines, such as IL-1β, TNF-α, and IFN-γ, which are critical in beta cell dysfunction and apoptosis (Burg and Tse 2018; Delmastro and Piganelli 2011). Histological analysis of pancreatic sections from both patients with type 1 diabetes and mouse models of autoimmune diabetes revealed an influx of recruited macrophages to the islets (Hänninen et al. 1992; Itoh et al. 1993; Jansen et al. 1993; Kolb-Bachofen et al. 1992; Roep et al. 1992). Zhang et al. reported that more inflammatory macrophages infiltrated into the islet cells in STZ-induced diabetic mice and that enhancing M1 macrophage activation further exacerbated pancreas injury (Zhang et al. 2020). Inhibiting macrophage infiltration into the islet cells or restricting macrophage M1 polarization in diabetic mice would be helpful for maintaining pancreas function and preventing type 1 diabetes progression. Notably, our results provided evidence supporting the hypothesis that the primary site of action of OI protection against diabetes might be the infiltrating macrophages and the subsequent cascade of local inflammatory events. Indeed, the macrophage infiltration was not significantly different among the diabetic and OI groups in our study. Interestingly, the OI mice had lower numbers of islet-associated M1 macrophages and a lower proportion of M1 macrophages in different peripheral compartments, which indicates a possible direct effect of OI on macrophage polarization. However, the role of OI in macrophage phenotype reprogramming in diabetic mice remains complex. Metabolic alterations followed by diabetic progression are also associated with macrophage polarization. Hyperglycemia in diabetes induces epigenetic changes that reprogram the macrophage phenotype and modify the subsequent cellular response upon stimulus (Ratter et al. 2018). Macrophages derived from diabetic mice presented more abundant NLRP3 and iNOS protein expression and higher inflammatory cytokine secretion and NO production after LPS stimulation compared with cells from non-diabetic mice (Davanso et al. 2021). Moreover, BMDM from alloxan-induced diabetes impaired signaling pathways, which involved alterations at both PI3K–Akt and MAPK levels (Galvao Tessaro et al. 2020). Accompanying this, both BMDM and peritoneal macrophages from diabetic mice demonstrated dysregulated cytokine secretion profiles (Galvao Tessaro et al. 2020). Consistent with these literature data, our results supported the idea that OI reduced NLRP3 and iNOS expression and decreased NO production and proinflammatory cytokine release in macrophages upon LPS administration, which implied that OI significantly restricted macrophage M1 polarization. Although the anti-inflammatory and antioxidative effects of OI have been reported, its distinct role in regulating the MAPK signaling response to inflammation is unknown. Mounting evidence suggested that the MAPK family is critical for regulating proinflammatory cytokines and mediators. MAPK pathway activation phosphorylates various downstream targets to induce inflammation mediator production in macrophages. It was proven in several studies that Nrf2 exerted anti-inflammatory and antioxidative functions by regulating MAPK pathway activity. It was suggested that Nrf2 activation mitigated oxidative stress and inflammation in mesangial cells caused by high glucose through inhibiting MAPK signaling (Yao et al. 2022). Moreover, sulforaphane exerted anti-neuroinflammatory effects on LPS-activated microglia through Nrf2–HO-1 pathway activation and JNK–AP-1–NF-κB pathway inhibition (Subedi et al. 2019). It is widely recognized that OI inhibits LPS-induced proinflammatory cytokine secretion, NF-κB activation, and oxidative stress. Here, our data indicate another pathway behind OI regulation of the inflammatory response in macrophages by attenuating p-ERK and p-p38 MAPK activation in diabetic macrophages and response to LPS stimulation. This result was consistent with our proteomics analysis results, where MAPK pathway-related proteins were greatly enriched in the DFEs between the OI intervention and diabetic mouse islets. It has been reported in several studies that monocytes from new-onset type 1 diabetes patients have increased IL-1β basal levels (Meyers et al. 2010) and a more pronounced response to LPS stimulation in vitro (Davanso et al. 2021). Similarly, during the in vitro treatment of LPS-stimulated PBMCs isolated from type 1 diabetes patients, we observed substantially increased IL-1β secretion compared to the non-diabetes population. Our findings supported the idea that type 1 diabetes patients present a basal inflammatory status. Surprisingly, OI treatment of the monocytes attenuated IL-1β production compared with LPS-only stimulation in both non-diabetes and type 1 diabetes participants. The anti-inflammatory effect was consistent with that reported in a previous study, where OI activated Nrf2 signaling to inhibit proinflammatory cytokine secretion in the PBMCs of systemic lupus erythematosus patients (Tang et al. 2018). As mentioned above, pancreatic beta cell dysfunction is closely related to activated M1 macrophage infiltration in type 1 diabetes pathogenesis. Apart from presenting antigens to autoreactive T cells in the initiation and effector phases of type 1 diabetes, activated M1 macrophages produce proinflammatory cytokines and NO to induce beta cell dysfunction and apoptosis (Burg and Tse 2018). In the present study, we regulated macrophage polarization in a co-culture system to investigate the protective effect of OI on beta cell function. The activated macrophages caused beta cell dysfunction whereas OI treatment significantly restored beta cell function. In addition, not only did OI regulate macrophage polarization, it might have distinctly improved beta cell function, which warrants further investigation. Overall, our results demonstrated that OI was a potent activator of the Nrf2-mediated antioxidative response in macrophages and subsequently inhibited M1 polarization through the MAPK pathway. These actions might be crucial to the crosstalk between innate immunity and beta cell function and establish the basis for the emerging therapeutic implications of OI in type 1 diabetes progression. ## Conclusions We provided the first evidence to date that the OI can impede islet inflammation progression and improve glucose metabolism by regulating macrophage phenotype reprogramming in mouse models of type 1 diabetes. Furthermore, OI significantly inhibited MAPK activation in macrophages to alleviate the inflammatory response of the pancreas, eventually improving beta cell dysfunction. The findings indicated that elevating endogenous itaconate levels might attenuate systemic inflammation, which provided potential new insight into a feasible adjuvant therapy for preventing and treating type 1 diabetes. Although this tentative investigation provided an important basic finding toward the development of a new drug class target for preventing and treating type 1 diabetes, significant hurdles remain and in-depth studies are warranted, as are further supporting studies in other models of diabetes. ## Supplementary Information Additional file 1. The experimental protocols of prevention model and treatment model were presented in Fig. 1A and Fig. 4A, separately. ## References 1. Burg AR, Tse HM. Redox-sensitive innate immune pathways during macrophage activation in type 1 diabetes*. Antioxid Redox Signal* (2018.0) 29 1373-1398. DOI: 10.1089/ars.2017.7243 2. Chellappan DK, Sivam NS, Teoh KX, Leong WP, Fui TZ, Chooi K. Gene therapy and type 1 diabetes mellitus. Biomed Pharmacother (2018.0) 108 1188-1200. DOI: 10.1016/j.biopha.2018.09.138 3. Chen S, Du K, Zou C. Current progress in stem cell therapy for type 1 diabetes mellitus. Stem Cell Res Ther (2020.0) 11 275. DOI: 10.1186/s13287-020-01793-6 4. Cordes T, Lucas A, Divakaruni AS, Murphy AN, Cabrales P, Metallo CM. Itaconate modulates tricarboxylic acid and redox metabolism to mitigate reperfusion injury. Mol Metab (2020.0) 32 122-135. DOI: 10.1016/j.molmet.2019.11.019 5. Dahlén E, Dawe K, Ohlsson L, Hedlund G. Dendritic cells and macrophages are the first and major producers of TNF-alpha in pancreatic islets in the nonobese diabetic mouse. J Immunol (1998.0) 160 3585-3593. DOI: 10.4049/jimmunol.160.7.3585 6. Davanso MR, Crisma AR, Braga TT, Masi LN, de Amaral CL, Leal VNC. Macrophage inflammatory state in Type 1 diabetes: triggered by NLRP3/iNOS pathway and attenuated by docosahexaenoic acid. Clin Sci (lond) (2021.0) 135 19-34. DOI: 10.1042/CS20201348 7. Delmastro MM, Piganelli JD. Oxidative stress and redox modulation potential in type 1 diabetes. Clin Dev Immunol (2011.0) 2011 593863. DOI: 10.1155/2011/593863 8. Diskin C, Palsson-McDermott EM. Metabolic modulation in macrophage effector function. Front Immunol (2018.0) 9 270. DOI: 10.3389/fimmu.2018.00270 9. Eizirik DL, Colli ML, Ortis F. The role of inflammation in insulitis and beta-cell loss in type 1 diabetes. Nat Rev Endocrinol (2009.0) 5 219-226. DOI: 10.1038/nrendo.2009.21 10. Espinoza-Jiménez A, Peón AN, Terrazas LI. Alternatively activated macrophages in types 1 and 2 diabetes. Mediators Inflamm (2012.0) 2012 815953. DOI: 10.1155/2012/815953 11. Galvao Tessaro FH, Ayala TS, Bella LM, Martins JO. Macrophages from a type 1 diabetes mouse model present dysregulated Pl3K/AKT, ERK 1/2 and SAPK/JNK levels. Immunobiology (2020.0) 225 151879. DOI: 10.1016/j.imbio.2019.11.014 12. Hanenberg H, Kolb-Bachofen V, Kantwerk-Funke G, Kolb H. Macrophage infiltration precedes and is a prerequisite for lymphocytic insulitis in pancreatic islets of pre-diabetic BB rats. Diabetologia (1989.0) 32 126-134. DOI: 10.1007/BF00505185 13. Hänninen A, Jalkanen S, Salmi M, Toikkanen S, Nikolakaros G, Simell O. Macrophages, T cell receptor usage, and endothelial cell activation in the pancreas at the onset of insulin-dependent diabetes mellitus. J Clin Invest (1992.0) 90 1901-1910. DOI: 10.1172/JCI116067 14. Itoh N, Hanafusa T, Miyazaki A, Miyagawa J, Yamagata K, Yamamoto K. Mononuclear cell infiltration and its relation to the expression of major histocompatibility complex antigens and adhesion molecules in pancreas biopsy specimens from newly diagnosed insulin-dependent diabetes mellitus patients. J Clin Invest (1993.0) 92 2313-2322. DOI: 10.1172/JCI116835 15. Jansen A, Voorbij HA, Jeucken PH, Bruining GJ, Hooijkaas H, Drexhage HA. An immunohistochemical study on organized lymphoid cell infiltrates in fetal and neonatal pancreases. A comparison with similar infiltrates found in the pancreas of a diabetic infant. Autoimmunity (1993.0) 15 31-38. DOI: 10.3109/08916939309004836 16. Jörns A, Arndt T, MeyerzuVilsendorf A, Klempnauer J, Wedekind D, Hedrich HJ. Islet infiltration, cytokine expression and beta cell death in the NOD mouse, BB rat, Komeda rat, LEW.1AR1-iddm rat and humans with type 1 diabetes. Diabetologia (2014.0) 57 512-521. DOI: 10.1007/s00125-013-3125-4 17. Kolb-Bachofen V, Schraermeyer U, Hoppe T, Hanenberg H, Kolb H. Diabetes manifestation in BB rats is preceded by pan-pancreatic presence of activated inflammatory macrophages. Pancreas (1992.0) 7 578-584. DOI: 10.1097/00006676-199209000-00011 18. Lampropoulou V, Sergushichev A, Bambouskova M, Nair S, Vincent EE, Loginicheva E. Itaconate links inhibition of succinate dehydrogenase with macrophage metabolic remodeling and regulation of inflammation. Cell Metab (2016.0) 24 158-166. DOI: 10.1016/j.cmet.2016.06.004 19. Li Y, Chen X, Zhang H, Xiao J, Yang C, Chen W. 4-Octyl itaconate alleviates lipopolysaccharide-induced acute lung injury in mice by inhibiting oxidative stress and inflammation. Drug Des Devel Ther (2020.0) 14 5547-5558. DOI: 10.2147/DDDT.S280922 20. Li S, Vaziri ND, Swentek L, Takasu C, Vo K, Stamos MJ. Prevention of Autoimmune Diabetes in NOD Mice by Dimethyl Fumarate. Antioxidants (basel). (2021.0) 10 2 21. Liao ST, Han C, Xu DQ, Fu XW, Wang JS, Kong LY. 4-Octyl itaconate inhibits aerobic glycolysis by targeting GAPDH to exert anti-inflammatory effects. Nat Commun (2019.0) 10 5091. DOI: 10.1038/s41467-019-13078-5 22. Liu G, Wu Y, Jin S, Sun J, Wan BB, Zhang J. Itaconate ameliorates methicillin-resistant Staphylococcus aureus-induced acute lung injury through the Nrf2/ARE pathway. Ann Transl Med (2021.0) 9 712. DOI: 10.21037/atm-21-1448 23. Lu J, Liu J, Li L, Lan Y, Liang Y. Cytokines in type 1 diabetes: mechanisms of action and immunotherapeutic targets. Clin Transl Immunology (2020.0) 9 e1122. DOI: 10.1002/cti2.1122 24. Meyers AJ, Shah RR, Gottlieb PA, Zipris D. Altered Toll-like receptor signaling pathways in human type 1 diabetes. J Mol Med (berl) (2010.0) 88 1221-1231. DOI: 10.1007/s00109-010-0666-6 25. Mills EL, Ryan DG, Prag HA, Dikovskaya D, Menon D, Zaslona Z. Itaconate is an anti-inflammatory metabolite that activates Nrf2 via alkylation of KEAP1. Nature (2018.0) 556 113-117. DOI: 10.1038/nature25986 26. Murphy MP, O'Neill LAJ. Krebs cycle reimagined: the emerging roles of succinate and itaconate as signal transducers. Cell (2018.0) 174 780-784. DOI: 10.1016/j.cell.2018.07.030 27. Nagy MV, Chan EK, Teruya M, Forrest LE, Likhite V, Charles MA. Macrophage-mediated islet cell cytotoxicity in BB rats. Diabetes (1989.0) 38 1329-1331. DOI: 10.2337/diab.38.10.1329 28. Ni Q, Pham NB, Meng WS, Zhu G, Chen X. Advances in immunotherapy of type I diabetes. Adv Drug Deliv Rev (2019.0) 139 83-91. DOI: 10.1016/j.addr.2018.12.003 29. O'Dowd JF, Stocker CJ. Isolation and Purification of Rodent Pancreatic Islets of Langerhans. Methods Mol Biol (clifton, NJ) (2020.0) 2076 179-184. DOI: 10.1007/978-1-4939-9882-1_9 30. Pepper AR, Bruni A, Shapiro AMJ. Clinical islet transplantation: is the future finally now?. Curr Opin Organ Transplant (2018.0) 23 428-439. DOI: 10.1097/MOT.0000000000000546 31. Ratter JM, Tack CJ, Netea MG, Stienstra R. Environmental signals influencing myeloid cell metabolism and function in diabetes. Trends Endocrinol Metab (2018.0) 29 468-480. DOI: 10.1016/j.tem.2018.04.008 32. Ridler C. Diabetes: Islet transplantation for T1DM. Nat Rev Endocrinol (2016.0) 12 373. PMID: 27174023 33. Roep BO, Kallan AA, De Vries RR. Beta-cell antigen-specific lysis of macrophages by CD4 T-cell clones from newly diagnosed IDDM patient. A putative mechanism of T-cell-mediated autoimmune islet cell destruction. Diabetes (1992.0) 41 1380-1384. DOI: 10.2337/diab.41.11.1380 34. Ryan DG, Murphy MP, Frezza C, Prag HA, Chouchani ET, O'Neill LA. Coupling Krebs cycle metabolites to signalling in immunity and cancer. Nat Metab (2019.0) 1 16-33. DOI: 10.1038/s42255-018-0014-7 35. Subedi L, Lee JH, Yumnam S, Ji E, Kim SY. Anti-Inflammatory Effect of Sulforaphane on LPS-Activated Microglia Potentially through JNK/AP-1/NF-kappaB Inhibition and Nrf2/HO-1 Activation. Cells (2019.0) 8 2. DOI: 10.3390/cells8020194 36. Sun X, Zhang B, Pan X, Huang H, Xie Z, Ma Y. Octyl itaconate inhibits osteoclastogenesis by suppressing Hrd1 and activating Nrf2 signaling. FASEB J (2019.0) 33 12929-12940. DOI: 10.1096/fj.201900887RR 37. Tang C, Wang X, Xie Y, Cai X, Yu N, Hu Y. 4-Octyl itaconate activates Nrf2 signaling to inhibit pro-inflammatory cytokine production in peripheral blood mononuclear cells of systemic lupus erythematosus patients. Cell Physiol Biochem (2018.0) 51 979-990. DOI: 10.1159/000495400 38. Tian F, Wang Z, He J, Zhang Z, Tan N. 4-Octyl itaconate protects against renal fibrosis via inhibiting TGF-beta/Smad pathway, autophagy and reducing generation of reactive oxygen species. Eur J Pharmacol (2020.0) 873 172989. DOI: 10.1016/j.ejphar.2020.172989 39. Torres A, Makowski L, Wellen KE. Immunometabolism: Metabolism fine-tunes macrophage activation. Life (2016.0) 5 e14354 40. Walker R, Bone AJ, Cooke A, Baird JD. Distinct macrophage subpopulations in pancreas of prediabetic BB/E rats Possible role for macrophages in pathogenesis of IDDM. Diabetes (1988.0) 37 1301-1304. DOI: 10.2337/diab.37.9.1301 41. Xie Y, Chen Z, Wu Z. Four-octyl itaconate attenuates UVB-induced melanocytes and keratinocytes apoptosis by Nrf2 activation-dependent ROS inhibition. Oxid Med Cell Longev (2022.0) 2022 9897442. DOI: 10.1155/2022/9897442 42. Xin Y, Zou L, Lang S. 4-Octyl itaconate (4-OI) attenuates lipopolysaccharide-induced acute lung injury by suppressing PI3K/Akt/NF-kappaB signaling pathways in mice. Exp Ther Med (2021.0) 21 141. DOI: 10.3892/etm.2020.9573 43. Yagishita Y, Uruno A, Chartoumpekis DV, Kensler TW, Yamamoto M. Nrf2 represses the onset of type 1 diabetes in non-obese diabetic mice. J Endocrinol (2019.0) 8 67 44. Yang W, Wang Y, Zhang P, Wang T, Li C, Tong X. Hepatoprotective role of 4-octyl itaconate in concanavalin a-induced autoimmune hepatitis. Mediators Inflamm (2022.0) 2022 5766434. DOI: 10.1155/2022/5766434 45. Yao H, Zhang W, Yang F, Ai F, Du D, Li Y. Discovery of caffeoylisocitric acid as a Keap1-dependent Nrf2 activator and its effects in mesangial cells under high glucose. J Enzyme Inhib Med Chem (2022.0) 37 178-188. DOI: 10.1080/14756366.2021.1998025 46. Yi Z, Deng M, Scott MJ, Fu G, Loughran PA, Lei Z. Immune-Responsive Gene 1/Itaconate Activates Nuclear Factor Erythroid 2-Related Factor 2 in Hepatocytes to Protect Against Liver Ischemia-Reperfusion Injury. Hepatology (2020.0) 72 1394-1411. DOI: 10.1002/hep.31147 47. Zhan Z, Wang Z, Bao Y, Liu W, Hong L. OI inhibits development of ovarian cancer by blocking crosstalk between cancer cells and macrophages via HIF-1alpha pathway. Biochem Biophys Res Commun (2022.0) 606 142-148. DOI: 10.1016/j.bbrc.2022.03.106 48. Zhang S, Zhong J, Yang P, Gong F, Wang CY. HMGB1, an innate alarmin, in the pathogenesis of type 1 diabetes. Int J Clin Exp Pathol (2009.0) 3 24-38. PMID: 19918326 49. Zhang C, Han X, Yang L, Fu J, Sun C, Huang S. Circular RNA circPPM1F modulates M1 macrophage activation and pancreatic islet inflammation in type 1 diabetes mellitus. Theranostics (2020.0) 10 10908-10924. DOI: 10.7150/thno.48264 50. Lucier J, Weinstock RS. Diabetes Mellitus Type 1. StatPearls. Treasure Island (FL): StatPearls Publishing Copyright © 2022, StatPearls Publishing LLC.; 2022.
--- title: 'Critical windows of exposure to air pollution and gestational diabetes: assessing effect modification by maternal pre-existing conditions and environmental factors' authors: - Marcel Miron-Celis - Robert Talarico - Paul J. Villeneuve - Eric Crighton - David M. Stieb - Cristina Stanescu - Éric Lavigne journal: Environmental Health year: 2023 pmcid: PMC10015960 doi: 10.1186/s12940-023-00974-z license: CC BY 4.0 --- # Critical windows of exposure to air pollution and gestational diabetes: assessing effect modification by maternal pre-existing conditions and environmental factors ## Abstract ### Background Ambient air pollution has been associated with gestational diabetes (GD), but critical windows of exposure and whether maternal pre-existing conditions and other environmental factors modify the associations remains inconclusive. ### Methods We conducted a retrospective cohort study of all singleton live birth that occurred between April 1st 2006 and March 31st 2018 in Ontario, Canada. Ambient air pollution data (i.e., fine particulate matter with a diameter ≤ 2.5 μm (PM2.5), nitrogen dioxide (NO2) and ozone (O3)) were assigned to the study population in spatial resolution of approximately 1 km × 1 km. The Normalized Difference Vegetation Index (NDVI) and the Green View Index (GVI) were also used to characterize residential exposure to green space as well as the Active Living Environments (ALE) index to represent the active living friendliness. Multivariable Cox proportional hazards regression models were used to evaluate the associations. ### Results Among 1,310,807 pregnant individuals, 68,860 incident cases of GD were identified. We found the strongest associations between PM2.5 and GD in gestational weeks 7 to 18 (HR = 1.07 per IQR (2.7 µg/m3); $95\%$ CI: 1.02 – 1.11)). For O3, we found two sensitive windows of exposure, with increased risk in the preconception period (HR = 1.03 per IQR increase (7.0 ppb) ($95\%$ CI: 1.01 – 1.06)) as well as gestational weeks 9 to 28 (HR 1.08 per IQR ($95\%$ CI: 1.04 –1.12)). We found that women with asthma were more at risk of GD when exposed to increasing levels of O3 (p- value for effect modification = 0.04). Exposure to air pollutants explained $20.1\%$, $1.4\%$ and $4.6\%$ of the associations between GVI, NDVI and ALE, respectively. ### Conclusion An increase of PM2.5 exposure in early pregnancy and of O3 exposure during late first trimester and over the second trimester of pregnancy were associated with gestational diabetes whereas exposure to green space may confer a protective effect. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12940-023-00974-z. ## Introduction Gestational diabetes is one of the most common pregnancy complications, affecting approximately 3–$5\%$ of pregnancies [1]. It is also a substantial contributor to both maternal and neonatal morbidity and mortality in most countries [2–5]. The specific disease pathways involved in the pathophysiology of gestational diabetes are yet to be clearly elucidated, but it is commonly argued that these conditions are multifactorial — often involving genetic, social and environmental factors [6]. On the environmental front, ambient air pollution has been associated with gestational diabetes [7]. While the underlying mechanism(s) of this association are not well understood, evidence suggests that exposure to air pollution during pregnancy can lead to oxidative stress and inflammatory processes that increase the likelihood of gestational diabetes [7]. Recent evidence has also highlighted the importance of critical air pollution exposure windows, with maternal exposure in the 21st to 24th gestational weeks posing particular risks [8]. Few studies, however, have assessed the preconception period as a vulnerable window for gestational diabetes [7]. In addition, associations between ambient air pollution and gestational diabetes among those with pre-existing medical conditions are not well understood [9, 10]. This is important because the presence of inflammation is also a common characteristic among individuals with pre-existing medical conditions such as asthma and hypertension. The role of other environmental factors including features of the natural and built environments (and their interrelationship with ambient air pollution) in the etiology of gestational diabetes requires further examination. While there is a growing body of evidence pointing to the relationships between natural and built environments in determining diverse health outcomes including birth weight and preterm delivery [11, 12], few studies have examined their relationship to gestational diabetes or in particular how these environments may interact with exposure to ambient air pollution during pregnancy [13, 14]. For instance, there is a considerable research gap relating to how natural and built environment characteristics (e.g., access to green environment and neighborhood friendliness to active living – “walkability”) may modify associations between ambient air pollution exposure and gestational diabetes. Greenness exposure during pregnancy has been explored extensively for its association with fetal growth, birthweight, gestational age, preterm birth and head circumference, and it has generally been found to have protective effects against adverse outcomes, though results are not entirely consistent [15]. However, only a small amount of researches has investigated the interrelationship between green space, air pollution and gestational diabetes [13, 14]. For example, one can hypothesize that green space may confer a protective effect against gestational diabetes by reducing exposure to air pollution, which may be associated to a possible mediation effect that has been observed on other maternal and pregnancy outcomes [16]. Greenness has also been associated to a reduction of stress, an increase of opportunity for physical activity and their combined effect could reduce the risk of gestational diabetes. Associations between exposures to air pollution and gestational diabetes could also be impacted by neighborhood walkability, in which highly walkable neighbourhoods are associated with higher levels of physical activity, but likely more exposure to air pollution [17]. These factors may also reduce allostatic load [18–20]. Therefore, this study sought to investigate whether ambient air pollution increases the risk of gestational diabetes, accounting for pre-existing maternal health conditions, and assessing variations in risk across different exposure periods. In addition, the study investigates the extent to which associations between exposures to air pollution and gestational diabetes are modified by neighbourhood green space and active living friendliness. Given its impact on maternal and neonatal health, there is an important need to understand the etiological pathways of gestational diabetes to mitigate adverse impacts in both the mother and child. ## Study design and population We conducted a retrospective cohort study using data on singleton live births that occurred between April 1st 2006 and March 31st 2018 in the Province of Ontario, Canada. Mother-infant pairs were obtained from the Better Outcomes Registry & Network (BORN) Ontario, a province wide birth registry that captures perinatal health information (https://www.bornontario.ca/en/about-born/). Data pertaining to each hospital birth in Ontario are collected from patient charts by hospital staff from clinical forms, and patient interviews, and then entered into the BORN information system. The registry contains information on maternal demographics, health behaviours (e.g. smoking, alcohol use), reproductive history, and clinical information related to pregnancy, labour, birth, and foetal and neonatal outcomes. Formal training of data collectors and ongoing data validation programs ensure the database is maintained with high quality data. We used the Postal Code Conversion File Plus (PCCF +) to obtain the geographic coordinates of maternal place(s) of residence based on residential postal code(s) reported in health administrative data. Pregnancies with postal codes of residence outside Ontario were excluded from the analysis. Subjects with pre-gestational diabetes (type 1 or type 2), without a valid health card number, missing date of birth, missing information on the sex of the new-born, postal code value, or mothers who did not have continuous residence in Ontario, Canada for their respective gestational period were excluded. A flow chart describing the exclusion process in presented in Supplementary Fig. 1. As well, we had no information on birth outcomes for some women who were pregnant at the same time as participants of our study, but gave birth before the study started or after the study ended. In order to account for the non-inclusion of these women, which has been described previously as the “fixed cohort bias” [21], we included only births with estimated conception dates ranging from 20 weeks (i.e. shortest pregnancy) before the study started to 44 weeks (i.e. longest pregnancy) before it ended. ## Outcome ascertainment Incident cases of gestational diabetes were obtained from the BORN registry [22] between April 1st, 2006 and March 31st, 2018. From 2006 through 2013, gestational diabetes was diagnosed in Ontario using the 2003 and 2008 Canadian Diabetes Association’s guidelines, while the updated 2013 guidelines were used from 2013 to 2018 [23]. The 2003 and 2008 guidelines recommend universal screening for gestational diabetes using a 50 g glucose challenge test (GCT) at 24–28 weeks gestation, and when positive (i.e., > 7.8 mmol/L), a subsequent 75 g oral glucose tolerance test (OGTT) was needed to confirm the presence of gestational diabetes (positive thresholds: fasting, ≥ 5.3 mmol/L; 1 h, ≥ 10.6 mmol/L; 2 h, ≥ 8.9 mmol/L). Gestational diabetes was diagnosed when the results showed ≥ 2 positive OGTT results or a GCT result ≥ 10.3 mmol/L. The updated 2013 guidelines proposed two diagnostic methods to ascertain gestational diabetes. The first method was nearly identical to the $\frac{2003}{2008}$ guidelines with only a slight increase in some of the positive threshold values (i.e., 50 g GCT ≥ 11.1 mmol/L and 2 h OGTT ≥ 9.0 mmol/L). The second method was a one-step approach involving only the 75 g OGTT with the updated positive threshold values. ## Exposure ascertainment Ambient air pollution during pregnancy was the primary exposure of interest. Residential exposures to fine particulate matter with a diameter ≤ 2.5 μm (PM2.5), nitrogen dioxide (NO2) and ozone (O3) were assigned to the study population at the geographic centre of each 6-digit postal code area. This assignment was facilitated by the Postal Code Conversion File Plus which was used to convert residential postal codes into geographic coordinates [24]. The mother’s residences during pregnancy were used for determining exposure assignment during pregnancy. We used air pollution surfaces available at spatial resolutions of approximately 1-km2. The PM2.5 surface was derived using satellite-based estimates that were combined with ground-level monitor information and chemical transport models, as described by van Donkelaar et al. [ 25]. NO2 was assessed based on a national land-use regression (LUR) model, using data from the Canadian National Air Pollution Surveillance (NAPS) monitoring network, combined with information on satellite-derived NO2 estimates, road lengths within 10 km (km), area of industrial land use within 2 km and the mean summer rainfall [26]. O3 was assessed based on a surface that represents an average of daily 8 h maximum concentrations in the warm seasons (May 1st to October 31st) using an optimal interpolation technique described previously [27]. PM2.5 levels are described in micrograms per cubic meter (µg/m3) while O3 and NO2 levels are described in parts per billion (ppb). Air pollution estimates were available on a weekly level from April 1st 2006 until March 31st 2018, based on temporal scaling previously described [28]. Therefore, exposures were assigned for each week of pregnancy and for the preconception period (i.e. 12 weeks before estimated conception). Other environmental exposure variables were also obtained from CANUE, namely, residential exposure to green space, noise and neighbourhood active living friendliness. Detailed description of the ascertainment to residential exposure to green space using the Normalized Difference Vegetation Index (NDVI) and the Green View Index (GVI) are provided in supplementary material. Estimation procedures for noise are described in detail elsewhere [29]. Noise is reported in A-weighted decibels (dBA). The Active Living Environments (ALE) index represents the active living friendliness of Canadian communities on a scale from 1 (very low) to 5 (very high). A negative value indicates below average active living friendliness, a positive value indicates above average active living friendliness, and a value of zero indicates average active living friendliness. We also extracted data on daily average ambient temperature throughout the study period from the Daymet dataset at a 1 km × 1 km grid spatial resolution across Canada [30]. The data were then converted into weekly averages to match the air pollution data. The corresponding values of all exposure variables were assigned to each cohort member using the centroid geographical coordinates of the home address postal code. The exposure data were linked to the study cohort and analyzed by the Institute of Clinical Evaluative Science (ICES). ICES is an independent, non-profit research institute funded by an annual grant from the Ontario Ministry of Health and Long-Term Care (MOHLTC). As a prescribed entity under Ontario’s privacy legislation, ICES is authorized to collect and use health care data for the purposes of health system analysis, evaluation and decision support. Secure access to these data is governed by policies and procedures that are approved by the Information and Privacy Commissioner of Ontario. ## Covariates Covariates were available from BORN and included maternal age at delivery, maternal cigarette smoking anytime during pregnancy, parity, pre-pregnancy body mass index, month of birth and year of birth. We also captured gestational age, which was determined from the mother’s last menstrual period and ultrasound dating. Several additional covariates were also derived based on individual’s postal code(s) of residence during pregnancy: [1] a dichotomous variable classifying Ontario into the Greater Toronto Area, a densely-populated urban mega-region, and all other areas; [2] a categorical variable classifying the size of the community where individuals lived; [3] area-level deprivation based on the Ontario Marginalization Index, which quantifies the degree of marginalization between areas and inequalities in health and social well-being in Ontario and includes deprivation quintiles, instability quintiles, ethnic quintiles and dependency quintiles [31]. A directed acyclic graph (DAG) was conceptualized using previous knowledge on potential confounders. Using this approach the following covariates were included in all statistical models: maternal age, parity, maternal smoking status, pregnancy body mass index, weekly ambient mean temperatures, month of birth, year of birth, residence in the Greater Toronto Area, community size, and the Ontario Marginalization Index. The conceptual DAG showing the pathways through which these variables may influence the exposures and the outcomes of interest is shown in Supplementary Fig. 2. Pre-pregnancy health conditions (i.e. conditions present before pregnancy) among pregnant individuals considered as potential effect modifiers in the investigated associations included asthma, and hypertension, Information on pre-existing health conditions was captured from BORN. ## Statistical analysis Multivariable Cox proportional hazards regression models were used to evaluate the associations between each of the three air pollutants as continuous variables and gestational diabetes. We used gestational weeks of pregnancy as the underlying timescale in the Cox models. Follow-up was conducted from the 20th week of gestation until gestational diabetes diagnosis, delivery, still birth, maternal death or loss of eligibility for provincial health insurance. Results are expressed as the hazard ratio (HR) and $95\%$ confidence interval (CI) corresponding to an increase across the interquartile range (IQR) of the pollutant of interest. An extension of the distributed lag non-linear model (DLNM) was used to simultaneously investigate exposure by preconception weeks as well as by each of the first 37 weeks during pregnancy [32]. This method allows for identification of critical windows of exposure for complex exposure–response relationships [33]. To select the appropriate model, different lag structures (natural and B splines) and number of knots (2–5 knots) were used to define the crossbasis of pregnancy exposures. The crossbasis that minimized the Akaike Information Criterion (AIC) was selected as the final model. Estimates of association were obtained by calculating the cumulative hazard over the preconception period, pregnancy period and critical windows identified. Next, effect modification by residential green space, ALE and maternal pre-existing health conditions were tested by including an interaction term between each air pollutant of interest and these variables. Wald’s method was used to assess the presence of interaction on the multiplicative scale. Effect modification was considered statistically significant if the interaction term p-value was less than 0.05. We also conducted a sensitivity analysis by limiting the follow-up to the end of the 28th week of pregnancy since most women are being tested for gestational diabetes in Canada by the end of that week. We also additionally adjusted models for noise pollution. A mediation analysis was also conducted to assess whether the effects of exposure to green spaces and ALE might be mediated by air pollution. We reported natural direct, indirect and total effect of the impact of NDVI, GVI and ALE on gestational diabetes. Statistical analyses were conducted using R version 3.0.1,(R Core Team, 2019) using the survival (version 2.42–3), dlnm (version 2.1.3) and medflex packages. ## Results A total of 1,310,807 pregnant individuals were included in the study cohort. From April 1st, 2006 to March 31st, 2018, a total of 68,860 incident cases of gestational diabetes were identified in the province of Ontario. The complete baseline characteristics of the study population are shown in detail in Table 1. Some differences on key demographic characteristics can be noted among those with vs. without incident gestational diabetes. Namely, individuals with gestational diabetes tended to be slightly older, have a higher pre-pregnancy BMI and a higher parity. Additionally, those with gestational diabetes had a lower prevalence of smoking during pregnancy and were more likely to reside in the Greater Toronto Area and live within a larger community. There were also notable differences in socio-economic status between those with and without incident gestational diabetes, as shown by the deprivation and ethnic concentration quintiles. The mean concentrations of PM2.5 and NO2 were slightly higher among women with gestational diabetes, while levels were similar across the two groups for O3.Table 1Descriptive characteristics of the study population ($$n = 1$$,310,807) at birth stratified by disease statusCharacteristicsPresence of gestational diabetesN = 68,860Absence of gestational diabetesN = 1,241,947Demographic & behavioural factors Maternal age, years (Mean ± SD)32.72 ± 5.0630.14 ± 5.43 Gestational age, weeks (Mean ± SD)38.22 ± 1.6038.91 ± 1.78 Prepregnancy BMI (Mean ± SD)28.50 ± 8.1025.71 ± 7.09Parity 016,359 ($38.7\%$)555,150 ($44.7\%$) 115,564 ($36.9\%$)452,069 ($36.4\%$) ≥ 210,304 ($24.4\%$)234,728 ($18.9\%$)Smoking Status During Pregnancy Missing1,079 ($1.6\%$)12,996 ($1.0\%$) No62,135 ($90.2\%$)1,093,557 ($88.1\%$) Yes5,646 ($8.2\%$)135,394 ($10.9\%$)Maternal pre-existing conditions Asthma2,249 ($3.3\%$)44,090 ($3.6\%$) Hypertension1,269 ($1.8\%$)7,889 ($0.6\%$)Neighbourhood socio-economic factors Instability Missing1,052 ($1.5\%$)14,516 ($1.2\%$) 1st quintile17,540 ($25.5\%$)264,609 ($21.3\%$) 2nd quintile11,721 ($17.0\%$)231,123 ($18.6\%$) 3rd quintile10,751 ($15.6\%$)215,826 ($17.4\%$) 4th quintile12,072 ($17.5\%$)232,501 ($18.7\%$) 5th quintile15,724 ($22.8\%$)283,372 ($22.8\%$)Dependency Missing1,052 ($1.5\%$)14,516 ($1.2\%$) 1st quintile25,234 ($36.6\%$)403,903 ($32.5\%$) 2nd quintile15,262 ($22.2\%$)265,568 ($21.4\%$) 3rd quintile11,168 ($16.2\%$)214,494 ($17.3\%$) 4th quintile8,889 ($12.9\%$)183,931 ($14.8\%$) 5th quintile7,255 ($10.5\%$)159,535 ($12.8\%$) Deprivation Missing1,052 ($1.5\%$)14,516 ($1.2\%$) 1st quintile10,351 ($15.0\%$)236,802 ($19.1\%$) 2nd quintile10,952 ($15.9\%$)228,773 ($18.4\%$) 3rd quintile12,521 ($18.2\%$)231,539 ($18.6\%$) 4th quintile14,502 ($21.1\%$)237,554 ($19.1\%$) 5th quintile19,482 ($28.3\%$)292,763 ($23.6\%$) Ethnic Concentration Missing1,052 ($1.5\%$)14,516 ($1.2\%$) 1st quintile5,725 ($8.3\%$)162,917 ($13.1\%$) 2nd quintile6,656 ($9.7\%$)180,767 ($14.6\%$) 3rd quintile8,532 ($12.4\%$)205,411 ($16.5\%$) 4th quintile13,180 ($19.1\%$)260,678 ($21.0\%$) 5th quintile33,715 ($49.0\%$)417,658 ($33.6\%$)Geographical & Environmental factors Community Size ≥ 1 500 00040,033 ($58.1\%$)564,514 ($45.5\%$) 500 000–1 499 9998,001 ($11.6\%$)167,535 ($13.5\%$) 100 000–499 99913,100 ($19.0\%$)292,069 ($23.5\%$) 10 000–99 9993,075 ($4.5\%$)91,510 ($7.4\%$) < 10 0004,621 ($6.7\%$)125,945 ($10.1\%$) Missing30 ($0.0\%$)374 ($0.0\%$) Greater Toronto Area residence Yes28,008 ($66.3\%$)690,523 ($55.6\%$) No14,219 ($33.7\%$)551,424 ($44.4\%$) Active Living Environment (Mean ± SD)1.25 ± 3.520.69 ± 2.96 PM2.5, µg/m3 (Mean ± SD)8.09 ± 1.587.97 ± 1.72 NO2, ppb (Mean ± SD)13.23 ± 5.5911.36 ± 5.40 O3, ppb (Mean ± SD)48.32 ± 4.9348.45 ± 4.92 NDVI (Mean ± SD)0.69 ± 0.080.69 ± 0.08 Segmented GVI (Mean ± SD)13.45 ± 8.0513.95 ± 8.44 Ambient temperature8.59 ± 3.068.30 ± 2.97 Noise, dB(A) (Mean ± SD)59.38 ± 5.1159.31 ± 5.02 The IQRs for PM2.5, NO2 and O3 over the entire preconception and gestational periods were 2.7 μg/m3, 10.02 ppb and 7.0 ppb, respectively (Supplementary Table 1). During the entire pregnancy, exposure to PM2.5 was weakly correlated with exposures to NO2 ($r = 0.44$) and O3 (r = -0.25) and the correlation between exposure to NO2 and O3 (r = -0.27) was weakly negative (all three significant at $p \leq 0.001$) (Supplementary Table S2). As well, there was a weak correlation between the two different green space metrics (i.e. GVI and NDVI) ($r = 0.14$). Associations between weekly exposures to air pollutants and gestational diabetes for identifying potential sensitive windows are presented in Fig. 1. Associations between PM2.5 and gestational diabetes appeared to be strongest and most highly statistically significant from weeks 7 to 18 during the gestational period. The cumulative HR for those weeks of gestation was 1.07 per IQR (2.7 µg/m3) increase in PM2.5 ($95\%$ CI: 1.02 – 1.11) (Table 2). We did not identify a sensitive window for weekly exposures to NO2. The cumulative HRs over the preconception and gestational periods for NO2 exposure were 1.05 per IQR (10.0 ppb) increase ($95\%$ CI: 0.91, 1.21) and 0.99 ($95\%$ CI: 0.85, 1.16), respectively (Table 2). For O3, we found two sensitive windows of exposure, with statistically significant increased risk in the preconception period as well as gestational weeks 9 to 28. The cumulative HR for the sensitive window during the preconception period was 1.03 per IQR increase (7.0 ppb) ($95\%$ CI: 1.01 – 1.06). The cumulative HR for the sensitive window of 9–28 weeks of gestation was 1.08 per IQR increase ($95\%$ CI: 1.04, 1.12) (Table 2). For all three pollutants, the cumulative HR over the whole gestational period was not statistically significant. We did not find meaningful differences in the HRs when adjusting only for individual-level covariates (Supplementary Table 2) as opposed to adding neighborhood level covariates in the models as presented in Table 2.Fig. 1Weekly associated hazard ratios (HRs) associated with weekly PM2.5, NO2, and O3 exposures over the preconception period and the gestational period with risk of gestational diabetes in the overall cohort ($$n = 1$$,310,807). Gray shade indicates $95\%$ confidence intervals; dashed vertical line demarcate preconception and post conception weeks. All the models were adjusted for maternal age, parity, maternal smoking status, prepregnancy body mass index, weekly ambient temperatures, month of birth, year of birth, residing in the Greater Toronto Area, community size, deprivation quintiles, instability quintiles, dependency quintiles and ethnic quintilesTable 2Adjusteda cumulative hazard ratios (HRs) and $95\%$ confidence intervals (CIs) of gestational diabetes per interquartile range (IQR) increase in PM2.5, NO2, and O3 for the preconception period, entire pregnancy and DLM-identified sensitive windowsPollutantHR ($95\%$ CI)PM2.5 (per IQR = 2.7 µg/m3 increase) Preconception period1.01 (0.98 – 1.03) Pregnancy period1.05 (1.00 – 1.09) Sensitive windows1.07 (1.02 – 1.11)NO2 (per IQR = 10.0 ppb increase) Preconception period1.07 (0.88 – 1.23) Pregnancy period1.00 (0.85 – 1.16) Sensitive windows-O3 (per IQR = 7.0 ppb increase) Preconception period1.03 (1.01 – 1.06) Pregnancy period1.04 (1.00 – 1.08) Sensitive windows1.08 (1.04 – 1.12)aAdjusted for maternal age, parity, maternal smoking status, prepregnancy body mass index, weekly ambient temperatures, month of birth, year of birth, residing in the Greater Toronto Area, community size, deprivation quintiles, instability quintiles, dependency quintiles and ethnic quintiles We found evidence of effect modification by maternal asthma status for exposure to O3 over the sensitive window from the 9th to 28th weeks of gestation (p value for effect modification = 0.041) (Table 3). For instance, a cumulative HR of 1.12 ($95\%$ CI: 1.07 – 1.15) per IQR increase (7.0 ppb) for gestational diabetes was observed among women with asthma. In comparison, women without asthma had a cumulative HR of 1.04 ($95\%$ CI: 1.01 – 1.07). We also found that the cumulative HR for gestational diabetes for the sensitive window of exposure to PM2.5 was higher among women with asthma, but the effect modification was not statistically significant (p value for effect modification = 0.099). We did not find differences across other characteristics investigated, although cumulative HRs appeared higher in neighbourhoods with elevated levels of ALE index. Table 3Adjusteda cumulative hazard ratios (HRs) and $95\%$ confidence intervals (CIs) of gestational diabetes per interquartile range (IQR) increase in PM2.5 and O3 for the DLM-identified sensitive windows, stratified by potential effect modifiersPotential effect modifiersPM2.57th to 18th weeks of gestationO3Preconception weeksO39th to 28th weeks of gestationHR ($95\%$ CI)HR ($95\%$ CI)HR ($95\%$ CI)Asthma Presence1.21 (1.05 – 1.35)0.96 (0.88 – 1.04)1.12 (1.07 – 1.15) Absence1.06 (1.01 – 1.10)1.03 (1.01 – 1.06)1.04 (1.01 – 1.07) P value for effect modification0.0990.3310.041Chronic hypertension Presence0.95 (0.80 – 1.14)0.96 (0.86 – 1.08)1.03 (0.88 – 1.21) Absence1.07 (1.02 – 1.11)1.00 (0.99 – 1.02)1.08 (1.04 – 1.12) P value for effect modification0.7540.8520.842NDVI 1st tertile1.10 (1.06 – 1.14)1.01 (0.98 – 1.04)1.05 (1.02 – 1.09) 2nd tertile1.07 (1.02 – 1.11)1.03 (1.01 – 1.06)1.08 (1.04 – 1.12) 3rd tertile1.04 (1.00 – 1.08)1.00 (0.98 – 1.03)1.04 (1.01 – 1.06) P value for effect modification0.4230.7320.891GVI 1st tertile1.12 (1.07 – 1.16)1.00 (0.97 – 1.03)1.02 (0.98 – 1.06) 2nd tertile1.06 (1.02 – 1.10)1.01 (0.98 – 1.04)1.06 (1.02 – 1.10) 3rd tertile1.07 (1.02 – 1.11)1.03 (1.00 – 1.06)1.09 (1.04 – 1.13) P value for effect modification0.5210.7890.652Active Living Environment 1st tertile1.06 (1.01 – 1.10)1.00 (0.97 – 1.04)1.03 (0.99 – 1.07) 2nd tertile1.02 (0.98 – 1.06)1.01 (0.99 – 1.04)1.06 (1.02 – 1.10) 3rd tertile1.12 (1.08 – 1.16)1.03 (1.00 – 1.05)1.08 (1.04 – 1.12) P value for effect modification0.4890.9210.356aAdjusted for maternal age, parity, maternal smoking status, prepregnancy body mass index, weekly ambient temperatures, month of birth, year of birth, residing in the Greater Toronto Area, community size, deprivation quintiles, instability quintiles, dependency quintiles and ethnic quintiles In sensitivity analyses, we investigated the mediating effects of PM2.5, NO2 and O3 in the associations between independent exposures to GVI, NDVI and ALE on gestational diabetes (Supplementary Table 4). We found that exposure to air pollutants explained $20.1\%$, $1.4\%$ and $4.6\%$ of the associations between GVI, NDVI and ALE, respectively. In addition, adjusting for noise exposure at the place of residence of pregnant individuals did not modify substantially the HRs (data not shown). ## Discussion Findings from this large population-based birth cohort study showed that exposures to PM2.5 and O3 during early to mid-pregnancy increased the risk of gestational diabetes. Preconception exposure to O3 appeared to increase the risk of gestational diabetes. We also found evidence that the presence of pre-pregnancy maternal asthma increased susceptibility to the impact of exposure to O3 during pregnancy on the incidence of gestational diabetes. We did not find evidence that pre-pregnancy maternal hypertension or the investigated environmental factors modified susceptibility to air pollution for gestational diabetes. Several epidemiological studies have examined the associations between ambient air pollution and the risk of gestational diabetes [7, 10, 34]. However, few studies have investigated critical windows of exposure on a weekly level during preconception and gestational periods. In a recent study applying similar methods to ours conducted in China, Chen et al. found that exposures to PM2.5 among 4174 pregnant women during the 21st to 24th gestational weeks was the most critical window of exposure for increasing the risk of gestational diabetes [35]. In a meta-analysis of 11 epidemiological studies, authors found that second trimester PM2.5 exposure was associated with increased gestational diabetes risk (OR = 1.04, $95\%$ CI: 1.01 – 1.09, per 10 μg/m3 increase in PM2.5) [7]. In fact, a recent study conducted in California, which used monthly estimates of ambient air pollutants, found stronger associations of PM2.5 exposure during the second trimester with gestational diabetes, except for black carbon which was more strongly associated with gestational diabetes during early pregnancy [9]. In our study, we found that PM2.5 exposure during the 7th to 18th gestational weeks (i.e. overlapping late first trimester and early second trimester) appeared to be the most important critical window. We also observed positive associations between O3 exposures during the preconception period as well as during gestational weeks 9 to 28, and incidence of gestational diabetes. Results from the meta-analysis by Zhang et al. that included 13 epidemiological studies showed that prepregnancy O3 exposure was inversely associated with gestational diabetes (OR = 0.98, $95\%$ CI: 0.98–0.99) while no associations were observed for trimester 1 or 2 exposures [34]. Evidence for exposure to NO2 during different trimesters has been inconclusive according to recent meta-analyses [7, 34], except for first trimester exposure to nitrogen oxides (NOx) which appeared to be associated with gestational diabetes [7]. However, in a study among 395,927 pregnancies in southern California, authors found that NO2 was the pollutant most strongly associated with gestational diabetes [9]. In our study, we did not find any association with exposure to NO2. Further research is needed, in particular in understanding specific components of particulate matter and mixtures of pollutants driving those risks during finely resolved (i.e. weekly) critical windows of exposure. In terms of potential biological mechanisms, previous evidence in animals has shown that PM2.5 can affect glucose homeostasis, metabolic inflammatory responses, the production of reactive oxygen species, insulin resistance and glucose tolerance [36]. Exposure to PM2.5 during the later part of the first trimester and part of the second trimester could induce increases in fasting plasma glucose which can increase the likelihood of gestational diabetes diagnosis [37]. The fact that we found some effects during the preconception period could be explained by the fact that previous studies have shown that air pollution exposure before conception in rodents has led to adipose tissue inflammation and the generation of reactive oxygen species which may result in insulin resistance [38]. We observed a higher risk of gestational diabetes among pregnant individuals with asthma exposed to O3 during late first trimester and throughout the second trimester. Prior literature has shown that inhalation of gaseous pollutants can induce pro-inflammatory processes during pregnancy [39]. Inflammation is also a characteristic feature of the pathophysiology of asthma [40]. It is therefore biologically plausible that inflammation from exposure to air pollution during pregnancy combined with inflammation due to maternal presence of asthma increases the risk of gestational diabetes. These findings require further investigation. A mediation analysis was also done to explore the etiological pathways of the green space metrics (i.e. GVI and NDVI) and a measure of neighbourhood active living friendliness (i.e. ALE). The results showed that air pollution exposure explained $20.1\%$, $1.4\%$ and $4.6\%$ of the effects of GVI, NDVI and the ALE on the development of gestational diabetes, respectively. Evidence from a study conducted in Wuhan, China, showed that exposure to PM2.5 also mediated the association between residential green space exposure during pregnancy and development of gestational diabetes [14]. In our study, we found that of neighbourhood built environment measures, the effect of GVI on incidence of gestational diabetes was most strongly mediated by air pollution. The GVI metric could potentially capture exposure to trees to a better extent than NDVI, which may have a stronger impact on reducing air pollution levels. In fact, assessing green space exposure with street view images is a novel method and its advantages are being identified. Similarly to correlations found in this study, Larkin and colleagues found low correlations between GVI and NDVI [41]. This preliminary evidence requires further investigation. One important strength of this study is its very large sample size, as it allows for greater sensitivity to detect the effects of specific exposures while allowing adjustment for numerous potential confounders. Secondly, the rich individual-level covariate data strengthen the internal validity of the study and renders the results less prone to residual confounding. Thirdly, our methodology allowed the identification of critical windows of exposure rather than averaging exposures by trimesters in order to account for potentially different periods of vulnerability during pregnancy. Some limitations should also be considered. Some of the data used for this study came from administrative sources, which may be less accurate than clinical data. Additionally, estimates for exposures of interest were not ascertained at the level of full address of residence, but rather at the six character postal code level, potentially introducing exposure measurement error. Finally, the medical diagnostic criteria of gestational diabetes considered in this study changed during the 12-year study period, which could influence the incidence rates of the outcomes and affect the results in unpredictable ways. ## Conclusion In summary, this study has shown that increased PM2.5 exposure in early pregnancy and O3 exposure during late first trimester and over the second trimester of pregnancy were associated with incidence of gestational diabetes. Effects of O3 were stronger among pregnant individuals with asthma. Exposure to green space may confer protective effects in incidence of gestational diabetes through reductions in ambient air pollution. Prevention strategies aiming to reduce impacts of air pollution through increased access to green space during pregnancy should be considered. A more definitive characterization of the windows of susceptibility, especially in subgroups of the population and across mixtures of pollutants, will enhance insight into underlying mechanisms. ## Supplementary Information Additional file 1: Supplementary Table 1. Descriptive statistics of environmental factors. Supplementary Table 2. Coefficient of correlation between continuous variables of interest. Supplementary Table 3. Adjusted cumulative hazard ratios (HRs) for individual-level covariates only and $95\%$ confidence intervals (CIs) of gestational diabetes per interquartile range (IQR) increase in PM2.5, NO2, and O3 for the preconception period, entire pregnancy and DLM-identified sensitive windows. Supplementary Table 4. Adjusted mediating effects of exposures to air pollution (PM2.5, NO2 and O3) on the associations between the environmental exposures of interest and gestational diabetes. Supplementary Figure 1. Flow chart of participants exclusion. Supplementary Figure 2. Directed acyclic graph. ## References 1. 1.Lapolla A, Metzger BE (eds). Gestational Diabetes. A Decade after the HAPO Study. Front Diabetes. Basel, Karger. 2020;28:1–10. 10.1159/000480161. 2. Chen L, Mayo R, Chatry A, Hu G. Gestational Diabetes Mellitus: Its Epidemiology and Implication beyond Pregnancy. Curr Epidemiol Rep (2016.0) 3 1-11. DOI: 10.1007/s40471-016-0063-y 3. Khan KS, Wojdyla D, Say L, Gülmezoglu AM, Van Look PF. WHO analysis of causes of maternal death: a systematic review. Lancet (2006.0) 367 1066-1074. DOI: 10.1016/S0140-6736(06)68397-9 4. Kuklina EV, Ayala C, Callaghan WM. Hypertensive disorders and severe obstetric morbidity in the United States. Obstet Gynecol (2009.0) 113 1299-1306. DOI: 10.1097/AOG.0b013e3181a45b25 5. McLennan NM, Hazlehurst J, Thangaratinam S, Reynolds RM. ENDOCRINOLOGY IN PREGNANCY: Targeting metabolic health promotion to optimise maternal and offspring health. Eur J Endocrinol (2022.0) 186 R113-R126. DOI: 10.1530/EJE-21-1046 6. Erickson AC, Arbour L. The shared pathoetiological effects of particulate air pollution and the social environment on fetal-placental development. J Environ Public Health (2014.0) 2014 901017. DOI: 10.1155/2014/901017 7. Hu CY, Gao X, Fang Y, Jiang W, Huang K, Hua XG, Yang XJ, Chen HB, Jiang ZX, Zhang XJ. Human epidemiological evidence about the association between air pollution exposure and gestational diabetes mellitus: Systematic review and meta-analysis. Environ Res (2020.0) 180 108843. DOI: 10.1016/j.envres.2019.108843 8. Chen G, Sun X, Wang J, Dong M, Ye Y, Liu X, Sun J, Xiao J, He G, Hu J, Guo L, Li X, Rong Z, Zeng W, Zhou H, Chen D, Li J, Ma W, Bartashevskyy M, Wen X, Liu T. The association between maternal exposure to fine particulate matter (PM2.5) and gestational diabetes mellitus (GDM): a prospective birth cohort study in China. Environ Res Lett (2021.0) 16 5 055004. DOI: 10.1088/1748-9326/abe4f8 9. Sun Y, Li X, Benmarhnia T, Chen JC, Avila C, Sacks DA, Chiu V, Slezak J, Molitor J, Getahun D, Wu J. Exposure to air pollutant mixture and gestational diabetes mellitus in Southern California: Results from electronic health record data of a large pregnancy cohort. Environ Int (2022.0) 158 106888. DOI: 10.1016/j.envint.2021.106888 10. Bai W, Li Y, Niu Y, Ding Y, Yu X, Zhu B, Duan R, Duan H, Kou C, Li Y, Sun Z. Association between ambient air pollution and pregnancy complications: A systematic review and meta-analysis of cohort studies. Environ Res (2020.0) 185 109471. DOI: 10.1016/j.envres.2020.109471 11. Yang BY, Zhao T, Hu LX, Browning MHEM, Heinrich J, Dharmage SC, Jalaludin B, Knibbs LD, Liu XX, Luo YN, James P, Li S, Huang WZ, Chen G, Zeng XW, Hu LW, Yu Y, Dong GH. Greenspace and human health: An umbrella review. Innovation (Camb) (2021.0) 2 100164. DOI: 10.1016/j.xinn.2021.100164 12. Hu CY, Yang XJ, Gui SY, Ding K, Huang K, Fang Y, Jiang ZX, Zhang XJ. Residential greenness and birth outcomes: A systematic review and meta-analysis of observational studies. Environ Res (2021.0) 193 110599. DOI: 10.1016/j.envres.2020.110599 13. Qu Y, Yang B, Lin S, Bloom MS, Nie Z, Ou Y, Mai J, Wu Y, Gao X, Dong G, Liu X. Associations of greenness with gestational diabetes mellitus: The Guangdong Registry of Congenital Heart Disease (GRCHD) study. Environ Pollut (2020.0) 266 115127. DOI: 10.1016/j.envpol.2020.115127 14. Liao J, Chen X, Xu S, Li Y, Zhang B, Cao Z, Zhang Y, Liang S, Hu K, Xia W. Effect of residential exposure to green space on maternal blood glucose levels, impaired glucose tolerance, and gestational diabetes mellitus. Environ Res (2019.0) 176 108526. DOI: 10.1016/j.envres.2019.108526 15. Zhan Y, Liu J, Lu Z, Yue H, Zhang J, Jiang Y. Influence of residential greenness on adverse pregnancy outcomes: A systematic review and dose-response meta-analysis. Sci Total Environ (2020.0) 718 137420. DOI: 10.1016/j.scitotenv.2020.137420 16. Banay RF, Bezold CP, James P, Hart JE, Laden F. Residential greenness: current perspectives on its impact on maternal health and pregnancy outcomes. Int J Womens Health (2017.0) 9 133-144. DOI: 10.2147/IJWH.S125358 17. Lang JJ, Pinault L, Colley RC, Prince SA, Christidis T, Tjepkema M, Crouse DL, de Groh M, Ross N, Villeneuve PJ. Neighbourhood walkability and mortality: Findings from a 15-year follow-up of a nationally representative cohort of Canadian adults in urban areas. Environ Int (2022.0) 161 107141. DOI: 10.1016/j.envint.2022.107141 18. Egorov AI, Griffin SM, Converse RR, Styles JN, Sams EA, Wilson A, Jackson LE, Wade TJ. Vegetated land cover near residence is associated with reduced allostatic load and improved biomarkers of neuroendocrine, metabolic and immune functions. Environ Res (2017.0) 158 508-521. DOI: 10.1016/j.envres.2017.07.009 19. Egorov AI, Griffin SM, Converse RR, Styles JN, Klein E, Scott J, Sams EA, Hudgens EE, Wade TJ. Greater tree cover near residence is associated with reduced allostatic load in residents of central North Carolina. Environ Res (2020.0) 186 109435. DOI: 10.1016/j.envres.2020.109435 20. Haluza D, Schönbauer R, Cervinka R. Green perspectives for public health: a narrative review on the physiological effects of experiencing outdoor nature. Int J Environ Res Public Health (2014.0) 11 5445-5461. DOI: 10.3390/ijerph110505445 21. 21.Strand LB, Barnett AG, Tong S: Methodological challenges when estimating the effects of season and seasonal exposures on birth outcomes. BMC Med Res Methodol. 2011;11:49–2288–11–49; doi:10.1186/1471-2288-11-49 22. Murphy MSQ, Fell DB, Sprague AE, Corsi DJ, Dougan S, Dunn SI, Holmberg V, Huang T, Johnson M, Kotuba M, Bisnaire L, Chakraborty P, Richardson S, Teitelbaum M, Walker MC. Data Resource Profile: Better Outcomes Registry & Network (BORN) Ontario. Int J Epidemiol (2021.0) 50 1416-1417h. DOI: 10.1093/ije/dyab033 23. Mussa J, Meltzer S, Bond R, Garfield N, Dasgupta K. Trends in National Canadian Guideline Recommendations for the Screening and Diagnosis of Gestational Diabetes Mellitus over the Years: A Scoping Review. Int J Environ Res Public Health (2021.0) 18 1454. DOI: 10.3390/ijerph18041454 24. Wilkins R, Peters P. PCCF+ Version 5K User’s Guide. Automated Geographic Coding Based on the Statistics Canada Postal Code Conversion Files, Including Postal Codes through May 2011 (2012.0) 25. van Donkelaar A, Martin RV, Li C, Burnett RT. Regional Estimates of Chemical Composition of Fine Particulate Matter Using a Combined Geoscience-Statistical Method with Information from Satellites, Models, and Monitors. Environ Sci Technol (2019.0) 53 2595-2611. DOI: 10.1021/acs.est.8b06392 26. Hystad P, Setton E, Cervantes A, Poplawski K, Deschenes S, Brauer M, van Donkelaar A, Lamsal L, Martin R, Jerrett M, Demers P. Creating national air pollution models for population exposure assessment in Canada. Environ Health Perspect (2011.0) 119 1123-1129. DOI: 10.1289/ehp.1002976 27. Robichaud A, Ménard R. Multi-year objective analyses of warm season ground-level ozone and PM2.5 over North America using real-time observations and Canadian operational air quality models. Atmospheric Chem Phys (2014.0) 14 4 1769-1800. DOI: 10.5194/acp-14-1769-2014 28. Elten M, Benchimol EI, Fell DB, Kuenzig ME, Smith G, Chen H, Kaplan GG, Lavigne E. Ambient air pollution and the risk of pediatric-onset inflammatory bowel disease: A population-based cohort study. Environ Int (2020.0) 138 105676. DOI: 10.1016/j.envint.2020.105676 29. Liu Y, Goudreau S, Oiamo T, Rainham D, Hatzopoulou M, Chen H, Davies H, Tremblay M, Johnson J, Bockstael A, Leroux T, Smargiassi A. Comparison of land use regression and random forests models on estimating noise levels in five Canadian cities. Environ Pollut (2020.0) 256 113367. DOI: 10.1016/j.envpol.2019.113367 30. 30.Thornton PE, Shrestha R, Thornton M, Kao SC, Wei Y, Wilson BE: Gridded daily weather data for North America with comprehensive uncertainty quantification. Sci Data. 2021;8 1:190–021–00973–0; doi:10.1038/s41597-021-00973-0 [doi]. 31. Matheson FI, Dunn JR, Smith KL, Moineddin R, Glazier RH. Development of the Canadian Marginalization Index: a new tool for the study of inequality. Can J Public Health (2012.0) 103 S12-S16. DOI: 10.1007/BF03403823 32. Elten M, Benchimol EI, Fell DB, Kuenzig ME, Smith G, Kaplan GG, Chen H, Crouse D, Lavigne E. Residential Greenspace in Childhood Reduces Risk of Pediatric Inflammatory Bowel Disease: A Population-Based Cohort Study. Am J Gastroenterol (2021.0) 116 347-353. DOI: 10.14309/ajg.0000000000000990 33. Wilson A, Chiu YM, Hsu HL, Wright RO, Wright RJ, Coull BA. Potential for Bias When Estimating Critical Windows for Air Pollution in Children's Health. Am J Epidemiol (2017.0) 186 1281-1289. DOI: 10.1093/aje/kwx184 34. Zhang H, Wang Q, He S, Wu K, Ren M, Dong H, Di J, Yu Z, Huang C. Ambient air pollution and gestational diabetes mellitus: A review of evidence from biological mechanisms to population epidemiology. Sci Total Environ (2020.0) 719 137349. DOI: 10.1016/j.scitotenv.2020.137349 35. 35.Chen G, Sun X, Wang J, Dong M, Ye Y, Liu X, Sun J, Xiao J, He G, Hu J, Guo L, Li X, Rong Z, Zeng W, Zhou H, Chen D, Li J, Ma W, Bartashevskyy M, Liu T: The association between maternal exposure to fine particulate matter (PM 2.5 ) and gestational diabetes mellitus (GDM): A prospective birth cohort study in China. Environmental Research Letters. 2021;16:; doi:10.1088/1748-9326/abe4f8. 36. 36.Yi L, Wei C, Fan W: Fine particulate matter (PM(2.5)), a risk factor of rat gestational diabetes with altered blood glucose and pancreatic GLUT2 expression. Gynecol Endocrinol. 2017:1–6; doi:10.1080/09513590.2017.1318368. 37. Cheng X, Ji X, Yang D, Zhang C, Chen L, Liu C, Meng X, Wang W, Li H, Kan H, Huang H. Associations of PM(2.5) exposure with blood glucose impairment in early pregnancy and gestational diabetes mellitus. Ecotoxicol Environ Saf (2022.0) 232 113278. DOI: 10.1016/j.ecoenv.2022.113278 38. Zhang M, Wang X, Yang X, Dong T, Hu W, Guan Q, Tun HM, Chen Y, Chen R, Sun Z, Chen T, Xia Y. Increased risk of gestational diabetes mellitus in women with higher prepregnancy ambient PM(2.5) exposure. Sci Total Environ (2020.0) 730 138982. DOI: 10.1016/j.scitotenv.2020.138982 39. Vadillo-Ortega F, Osornio-Vargas A, Buxton MA, Sanchez BN, Rojas-Bracho L, Viveros-Alcaraz M, Castillo-Castrejon M, Beltran-Montoya J, Brown DG, O'Neill MS. Air pollution, inflammation and preterm birth: a potential mechanistic link. Med Hypotheses (2014.0) 82 219-224. DOI: 10.1016/j.mehy.2013.11.042 40. Litonjua AA, Carey VJ, Burge HA, Weiss ST, Gold DR. Parental history and the risk for childhood asthma. Does mother confer more risk than father?. Am J Respir Crit Care Med (1998.0) 158 176-181. DOI: 10.1164/ajrccm.158.1.9710014 41. Larkin A, Hystad P. Evaluating street view exposure measures of visible green space for health research. J Expo Sci Environ Epidemiol (2019.0) 29 447-456. DOI: 10.1038/s41370-018-0017-1
--- title: Clinical characteristics of women with gestational diabetes - comparison of two cohorts enrolled 20 years apart in southern Brazil authors: - Angela Jacob Reichelt - Letícia Schwerz Weinert - Livia Silveira Mastella - Vanessa Gnielka - Maria Amélia Campos - Vânia Naomi Hirakata - Maria Lúcia Rocha Oppermann - Sandra Pinho Silveiro - Maria Inês Schmidt journal: São Paulo Medical Journal year: 2017 pmcid: PMC10015997 doi: 10.1590/1516-3180.2016.0332190317 license: CC BY 4.0 --- # Clinical characteristics of women with gestational diabetes - comparison of two cohorts enrolled 20 years apart in southern Brazil ## ABSTRACT ### CONTEXT AND OBJECTIVE: The prevalence and characteristics of gestational diabetes mellitus (GDM) have changed over time, reflecting the nutritional transition and changes in diagnostic criteria. We aimed to evaluate characteristics of women with GDM over a 20-year interval. ### DESIGN AND SETTING: Comparison of two pregnancy cohorts enrolled in different periods, in university hospitals in Porto Alegre, Brazil: 1991 to 1993 ($$n = 216$$); and 2009 to 2013 ($$n = 375$$). ### METHODS: We applied two diagnostic criteria to the cohorts: International Association of Diabetes and Pregnancy Study Groups (IADPSG)/World Health Organization (WHO); and National Institute for Health and Care Excellence (NICE). We compared maternal-fetal characteristics and outcomes between the cohorts and within each cohort. ### RESULTS: The women in the 2010s cohort were older (31 ± 7 versus 30 ± 6 years), more frequently obese ($29.4\%$ versus $15.2\%$), with more hypertensive disorders ($14.1\%$ versus $5.6\%$) and at increased risk of cesarean section (adjusted relative risk 1.8; $95\%$ confidence interval: 1.4 - 2.3), compared with those in the 1990s cohort. Neonatal outcomes such as birth weight category and hypoglycemia were similar. In the 1990s cohort, women only fulfilling IADPSG/WHO or only fulfilling NICE criteria had similar characteristics and outcomes; in the 2010s cohort, women only diagnosed through IADPSG/WHO were more frequently obese than those diagnosed only through NICE (33 ± 8 kg/m2 versus 28 ± 6 kg/m2; $P \leq 0.001$). ### CONCLUSION: The epidemic of obesity seems to have modified the profile of women with GDM. Despite similar neonatal outcomes, there were differences in the intensity of treatment over time. The IADPSG/WHO criteria seemed to identify a profile more associated with obesity. ## CONTEXTO E OBJETIVO: Prevalência e características do diabetes mellitus gestacional (DMG) modificaram-se com o tempo, refletindo transição nutricional e diferentes critérios diagnósticos. Nosso objetivo foi avaliar características de gestações com DMG em intervalo de 20 anos. ## TIPO DE ESTUDO E LOCAL: Comparação de duas coortes gestacionais arroladas em diferentes períodos, em hospitais universitários de Porto Alegre, Brasil: 1991 a 1993 ($$n = 216$$) e 2009 a 2013 ($$n = 375$$). ## MÉTODOS: Aplicamos dois critérios diagnósticos às coortes: International Association of Diabetes and Pregnancy Study Groups (IADPSG)/Organização Mundial de Saúde (OMS); e National Institute for Health and Care Excellence (NICE). Comparamos características e desfechos materno-fetais entre as coortes e dentro de cada uma. ## RESULTADOS: Na coorte dos anos 2010, as mulheres eram mais velhas (31 ± 7 versus 30 ± 6 anos), obesas (29,$4\%$ versus 15,$2\%$), apresentaram mais distúrbios hipertensivos (14,$1\%$ versus 5,$6\%$) e risco aumentado de cesariana (risco relativo ajustado 1,8; intervalo de confiança de $95\%$ 1,4 - 2,3), comparadas às da coorte de 1990. Desfechos neonatais, como categoria do peso ao nascer e hipoglicemia, foram semelhantes. Na coorte de 1990, essas características e desfechos foram semelhantes nas mulheres que preenchiam apenas um dos critérios; na de 2010, mulheres diagnosticadas apenas pelo IADPSG/OMS eram mais obesas (33 ± 8 kg/m2 versus 28 ± 6 kg/m2, $P \leq 0$,001) do que as diagnosticadas apenas pelo NICE. ## CONCLUSÃO: A epidemia de obesidade parece ter modificado o perfil de mulheres com DMG. Embora desfechos neonatais sejam semelhantes, houve diferenças na intensidade de tratamento ao longo do tempo. O critério da IADPSG/OMS parece identificar um perfil mais associado à obesidade. ## INTRODUCTION Gestational diabetes (GDM), initially defined as the highest glycemic distribution values, has been surrounded by controversy, as detailed in the World Health Organization (WHO) position in 20131 and illustrated in a timeline.2 From the 1980s to 2010, two general procedures were in vogue, one based on a 2 h/75 g oral glucose tolerance test (OGTT) with two plasma glucose values and diagnostic criteria similar to those used outside of pregnancy, and another one based on a 3 h/100 g OGTT, with four pregnancy-specific plasma glucose cutoffs.1 Screening for gestational diabetes in Brazil was infrequent before the 1990s, but both OGTT procedures were increasingly adopted thereafter. The 2 h/75 g OGTT gained wider acceptance after a 1997 consensus meeting3 at which GDM was defined using the intermediate hyperglycemic cutoffs that are used outside of pregnancy (fasting ≥ 110 mg/dl; 2 h ≥ 140 mg/dl). This definition was validated using data from the Brazilian Gestational Diabetes Study (Estudo Brasileiro de Diabetes Gestacional, EBDG)4 and remained the main diagnostic criterion used in Brazil, usually with two-step screening based on fasting values.3 In 2010, the International Association of Diabetes and Pregnancy Study Group (IADPSG) made new recommendations based on a 75 g OGTT and using data from the Hyperglycemia and Adverse Pregnancy Outcome (HAPO) study.1 Their recommendations have been endorsed by various entities, but new controversies arose. Perhaps the most important of these was the observed increase in GDM prevalence, especially when applied universally.5,6 This led other bodies to maintain the previous two-step diagnostic and screening procedures.1 In 2013, the World Health Organization (WHO) recommended the IADPSG criteria,1 although it warned of possible difficulties in implementing them. Alternatives to aid implementation were also proposed.7 In 2015, the British National Institute for Health and Care Excellence (NICE)8 made new recommendations. These are quite similar to the 1999 WHO1 criteria for the 2 h value (140 mg/dl), but specifically define a lower fasting plasma glucose cutoff (100 mg/dl) that matches the cutoff for impaired fasting glucose established by the American Diabetes Association in 2004.9 This strategy had been previously suggested by the Latin American Diabetes Association, in 2007,10 and resembles the one adopted in Brazil in 1997, although at that time, it was based on the impaired fasting glucose cutoff in vogue for use outside of pregnancy (110 mg/dl). Although the 1997 diagnostic criteria are still used in Brazil, the new IADPSG/WHO criteria are increasingly being adopted. The question that arises is whether the clinical profile of women detected through the IADPSG/WHO criteria differs from the profile of those detected through the NICE criteria. Moreover, it can be asked whether these profiles have changed over time. ## OBJECTIVE The aim of the study was to evaluate changes in clinical characteristics and maternal and offspring outcomes over 20 years, between two Brazilian cohorts of women with GDM, and compare them when classified through the IADPSG/WHO or NICE criteria. ## METHODS This study is a comparison of two cohorts of pregnant women in Brazil. We studied two cohorts of women with GDM who had singleton pregnancies and at least one prenatal appointment in two university hospitals. A 75 g OGTT with two or three glucose measurements was available for 560 ($94.8\%$) women, while confirmatory fasting plasma glucose data was available for 31 ($5.2\%$). We applied two recent criteria for GDM to both cohorts: the IADPSG/WHO criteria: FPG ≥ 92 mg/dl or 1-h plasma glucose ≥ 180 mg/dl or 2-h plasma glucose ≥ 153 mg/dl;1 andthe NICE criteria: fasting plasma glucose (FPG) ≥ 100 mg/dl or 2 h plasma glucose ≥ 140 mg/dl.8 The first cohort was composed of 216 women who met either of the two contemporary criteria for GDM (IADPSG/WHO or NICE). This cohort was derived from a cohort of 1031 women who were enrolled between 1991 and 1993, in general prenatal clinics of two university hospitals in Porto Alegre, which was one of the centers of the EBDG study.4 In the original cohort, cases with known pre-gestational diabetes had been excluded at the time of booking, and only the cases that reached diabetes levels outside of pregnancy had been treated.11 The second cohort was recruited between November 2009 and December 2013 and was composed of 375 women who had been referred to a high-risk pregnancy prenatal clinic at a public university hospital located in the southernmost state of the country, which provides medical care through the Brazilian National Health System (Sistema Único de Saúde, SUS). In 2013, around 3,800 babies were delivered at the hospital and the cesarean rate was $35.16\%$.12 All eligible women with singleton pregnancies who had a diagnosis of GDM through either of the two GDM criteria were included and cared for by a multidisciplinary team. The dietary counseling differed according to the individuals’ BMI and gestational stage, and emphasized low glycemic index and carbohydrates, along with high intake of fiber-rich foods. We collected information on sociodemographic characteristics, medical history and pregnancy outcomes. The pre-gestational weight was obtained through self-reporting. Weights and heights were measured with the subjects wearing light clothes and no shoes. Use of diets, insulin and oral medications (metformin or glyburide) were considered to be “any treatment”. Data on pregnancy follow-up, delivery and maternal and newborn outcomes were retrieved from medical files. A positive family history of diabetes was defined as among first-degree relatives, and gravidity, as the number of pregnancies including the current one. Pre-gestational BMI was calculated as the informed pre-pregnancy weight divided by the square of the height and categorized according to the current WHO classification.13 Total weight gain was calculated as the difference between the last registered weight (measured at delivery or at the last prenatal appointment) and the informed pre-pregnancy weight. The 2009 Institute of Medicine recommendations were used to classify weight gain adequacy: for underweight women, 12.5 to 18 kg; normal BMI, 11.5 to 16 kg; overweight, 7 to 11 kg; and obese, 5 to 9 kg.14 Hypertensive-related disorders of pregnancy were a composite of gestational hypertension, preeclampsia and eclampsia, as defined by the International Society for the Study of Hypertension in Pregnancy (ISSHP).15 We used the Alexander birth weight chart16 to classify newborns as small for gestational age (SGA) or as large for gestational diabetes (LGA), according to birth weight and gestational age. The latter was based on the first day of a reliable last menstrual period or on first-trimester ultrasonography. Macrosomia was defined as birth weight ≥ 4,000 g at term, and preterm birth, as delivery at less than 37 gestational weeks.17 The ethics committees of both hospitals approved the study protocols (number 90-058 for the 1990s cohort and number 10-0364 for the 2010s cohort). Informed consent was obtained from all individual participants included in the study. ## Statistical analysis The data are presented as means (with standard deviation) or proportions (%). Student’s t test and Pearson’s χ2 test (with the Z test for comparison of proportions and Bonferroni’s correction) were used to compare the two GDM groups. Kappa statistics were used to calculate the level of agreement between the two diagnostic criteria. For adjustment of outcomes, we performed Poisson regression with robust variance and, in the models, we included the mothers’ baseline characteristics that were significant in univariable analyses. The outcomes assessed were: hypertensive disorders, cesarean section, preterm delivery, birth weight, frequencies of SGA and LGA, macrosomia, malformation, hypoglycemia and perinatal death. The 1990s cohort was taken to be the as reference and the results were presented as crude and adjusted relative risk (RR) and $95\%$ confidence interval (CI). The statistical analyses were performed using the SPSS software, version 18.8. Statistical significance was set at 0.05, and was taken to be two-sided. ## RESULTS The main characteristics of the two cohorts are shown in Table 1. Age, schooling and gravidity were greater in the recent cohort, while living with a partner and smoking decreased, the latter in a remarkable way (33.3 to $9.6\%$). The nutritional characteristics also changed importantly, such that the women of the 2010s cohort were notably more obese (45.1 versus $15.2\%$) before becoming pregnant and reached a higher weight at delivery (86 ± 18 versus 74 ± 12 kg). Accordingly, the plasma glucose values for the 2010s cohort were higher, based on fasting, 1 h and 2 h values. Additionally, the diagnosis of GDM was reached slightly earlier for the 2010s cohort and treatment was notably more frequent. The women of the 2010s cohort reported having markedly greater family histories of diabetes and having had a previous pregnancy with GDM. Although not statistically significant, a trend towards higher frequency of chronic hypertension was also observed, with slightly higher levels of diastolic blood pressure. Table 1:Characteristics of women in two gestational diabetes cohorts, 20 years apartBMI = body mass index; BP = blood pressure; GDM = gestational diabetes mellitus; OGTT = oral glucose tolerance test. Gestational diabetes diagnosed through the criteria of either IADPSG/WHO (International Association of Diabetes and Pregnancy Study Groups/World Health Organization) or NICE (National Institute for Health and Care Excellence); †Means (with standard deviation, SD) were compared using Student’s t test; proportions (%) were compared using Pearson’s χ2 test, with the Z test for proportions, adjusted using Bonferroni’s correction. ‡Includes one woman with BMI < 25 kg/m2 in each cohort; §Any GDM treatment: diet for the 1990s cohort and diet + oral drug or insulin for the 2010s cohort. As seen in Table 2, the women of the 2010s cohort ended their pregnancies with a slightly shorter duration and higher frequency of cesarean section. Pregnancy-related hypertension was more frequent but total gestational weight gain and adequacy of gestational weight gain did not differ much between the two cohorts. The main offspring outcomes were similar. Although not statistically significant, perinatal mortality decreased from $\frac{33}{1}$,000 to $\frac{18}{1}$,000. The adjusted relative risks of the main outcomes showed that there was higher risk of cesarean section in the 2010s cohort (Table 2). Although the difference in gestational age at delivery was significant in univariable analyses (5 days less in the 2010s cohort), the rates of preterm delivery were similar between the cohorts (14.9 versus $16.3\%$; $$P \leq 0.744$$). Table 2:Maternal and offspring outcomes in two gestational diabetes cohorts, 20 years apart$95\%$ CI = $95\%$ confidence interval; AGA = adequate for gestational age; aRR = adjusted relative risk; LGA = large for gestational age; RR = relative risk; SGA = small for gestational age. Gestational diabetes diagnosed through the criteria of either IADPSG/WHO (International Association of Diabetes and Pregnancy Study Groups/World Health Organization) or NICE (National Institute for Health and Care Excellence). †Means (with standard deviation, SD) were compared using Student’s t test; proportions (%) were compared using Pearson’s χ2 test, with the Z test for proportions, adjusted using Bonferroni’s correction. ‡Poisson regression with robust variance; §See comment in Results section; \|\|Adjusted for center, age, schooling, gravidity, smoking, previous gestational diabetes mellitus (GDM) and pre-gestational body mass index (BMI). In the 1990s cohort, the NICE criteria would label $51.4\%$ of women as having GDM, while the IADPSG/WHO criteria would label $94.5\%$ as having GDM. In the 2010s cohort, $87.0\%$ would meet the NICE criteria and $90.9\%$, the IADPSG/WHO criteria. The overall agreement between the two diagnostic criteria, examining the two cohorts together, was $68\%$ ($95\%$ CI: 66-$70\%$) but, as shown in Figure 1, the rate of agreement was greater for the 2010s cohort ($43.5\%$ in the 1990s cohort and $77.8\%$ in the 2010s cohort). The proportion of the remaining cases that would be detected through only one of the two criteria decreased over time for those only meeting the IADPSG/WHO criteria ($48.0\%$ versus $13.1\%$) but not for those only meeting the NICE criteria. Figure 1:Overlap of NICE criteria and IADPSG/WHO criteria in two gestational diabetes cohorts 20 years apart We then compared the clinical characteristics and outcomes for women only meeting the NICE criteria or only meeting the IADPSG/WHO criteria for the two cohorts (Table 3). Although the numbers became small, it was apparent that women only meeting the IADPSG/WHO criteria had higher BMI and pre-gestational weight and showed a trend towards excessive gestational weight gain and delivery of heavier babies, but showed less neonatal hypoglycemia. Conversely, women only meeting the NICE criteria had higher rates of neonatal hypoglycemia. For both cohorts, the mean fasting plasma glucose was higher and the 2 h plasma glucose was lower for women who only met the IADPSG/WHO criteria. Table 3:Characteristics and pregnancy outcomes of two gestational diabetes cohorts defined only through the NICE criteria or only through the IADPSG/WHO criteriaBMI = body mass index; GWG = gestational weight gain; IADPSG/WHO = International Association of Diabetes and Pregnancy Study Groups/World Health Organization; NICE = National Institute for Health and Care Excellence; OGTT = oral glucose tolerance test. Student’s t test for glycemic cutoffs: Means (with standard deviation, SD) were compared using Student’s t test; proportions (%) were compared using Pearson’s χ2 test, with the Z test for proportions, adjusted using Bonferroni’s correction; †$$P \leq 0.598$$; ‡$P \leq 0.999$; §$P \leq 0.999$; \|\|\|$$P \leq 0.006.$$ ## DISCUSSION The women in the more recent cohort of GDM were more obese, had higher plasma glucose values at diagnosis, higher frequency of pregnancy-related hypertension and higher adjusted risk of cesarean section than the previous cohort, which had been assembled about 20 years earlier. They were also more frequently treated for GDM. Newborn outcomes were similar over time, except for a downward trend in perinatal mortality. We found a very good overlap (Figure 1) between those diagnosed through the IADPSG/WHO and through the NICE criteria in the 2010s cohort. Women only meeting the IADPSG/WHO cutoffs showed a profile more associated with the effects of the ongoing obesity epidemic. The differences observed between the two cohorts may reflect the nationwide public policies that have been adopted, which have resulted in better social indicators, as revealed by an increasing Human Development Index (from 0.608 in 1990 to 0.755 in 2014).18 This attainment is reflected in the higher schooling levels, later pregnancies (surprisingly, in contrast to higher gravidity) and better health indicators (such as lower rates of smoking)19 that have been seen in the whole Brazilian population.18 Furthermore, implementation of a national health system,19 which has enabled almost universal access to diagnosis and treatment for gestational diabetes, may have contributed, at least partly, to the differences found between the two cohorts. However, the effects of the obesity epidemic have hampered these successes. Brazil moved up from 9th position, in 1975, to 5th position, in 2014, in the ranking of female obesity.20 Maternal obesity increased remarkably, revealed here through an average pre-gestational weight increase of 10 kg, in just 20 years between the two GDM cohorts. On average, the women in the 1990s cohort began pregnancy within the overweight category, whereas those in the 2010s cohort did so within the obesity category. Given that maternal obesity confers important adverse outcomes for both the mother and the child, and possibly for future generations,21 this epidemic rise in obesity threatens the progress in pregnancy outcomes that has already achieved. It also puts at risk the attainability of the goals for reducing the burden of non-communicable diseases by 2025, a challenge faced by Brazil and all other nations. Hyperglycemia and obesity share common metabolic pathways and characteristics, and thus lead to consequences that are probably indissoluble, with additive effects on GDM outcomes.22 It is* apparent that the effects of the obesity epidemic were fully manifested in our current cohort: the women were remarkably more obese, presented pregnancy hypertension more often and were at higher risk of cesarean section. Birth weight and the large-for-gestational-age rate among the newborns did not differ between the two cohorts, perhaps because of the more widespread treatment for GDM in the recent cohort. Within this scenario, over the last few years, we have faced the challenge of adopting new diagnostic criteria, following the new recommendations from IADPSG in 2010 and WHO in 2013. Our main concern is that these criteria are likely to increase the prevalence of gestational diabetes,23 both as a result of the epidemic of maternal obesity and as a consequence of only requiring one altered cutoff for a diagnosis of GDM. Previous estimates indicated that changing from the 1997 Brazilian criteria to the new IADPSG/WHO criteria would raise the frequency of GDM from $7.6\%$ to $18.0\%$, i.e. a 2.5-fold increase.24 As illustrated in Figure 1, by applying each criterion to the diagnostic test for women with GDM, the IADPSG/WHO criteria labeled a higher number of women in both cohorts as presenting GDM, although the rate of disagreement between the two criteria was lower in the 2010s cohort (down from $56.5\%$ to $22.2\%$). The rate of agreement between the two different criteria varies across studies, from $49.7\%$25 or $50.6\%$26 to $65.6\%$.27 This partial overlap suggests that these studies probably reflect distinct GDM profiles. In one study that compared the NICE and the IADPSG/WHO criteria, $55.1\%$ of the women with GDM would be detected by both criteria, which was lower than the overlap that we found in the 2010s cohort.28 It is possible that differences we found for the 2010s cohort concerning maternal weight and birth weight reflected the effects of the current obesity epidemic and associated factors, particularly in relation to those only meeting the IADPSG/WHO criteria. In a recent study comparing obese women with and without GDM in the first trimester of pregnancy, obesity markers such as insulin resistance and higher BMI were more frequent in those with GDM, along with higher glucose levels. It was suggested that application of IADPSG/WHO to the DALI cohort had “identified a profile akin to the metabolic syndrome”.29 Moreover, to be worthwhile, adoption of a GDM criterion that enhances prevalence should also increase the detection rate of relevant clinical outcomes. Given the low attributable fractions relating to hyperglycemia ($6.7\%$ for large for gestational age and $3.5\%$ for preeclampsia, based on the IADPSG/WHO criteria),24 increased detection of relevant outcomes is likely to be small. The main strength of our study is that it enables comparison between the features of a recent GDM cohort with those of an old one. We were able to document the important effect of the obesity epidemic over this 20-year interval. The major limitation of our study relates to the source of the cohorts: the 1990s cohort was derived from a large sample and had little intervention for treatment, and although the study was directed from university hospitals, the women were attending general prenatal care. On the other hand, for the 2010s cohort, enrollment was at a specialized clinic of a university hospital and women with greater severity of hyperglycemia may have been included. These women more frequently presented histories of family diabetes and previous GDM. This could have biased our results; nevertheless, diabetes rates are also increasing worldwide30 and this trend could potentially explain these findings. Intensive treatment in the 2010s cohort limited interpretation of pregnancy outcomes. Finally, only a few of our cases met only one criterion or the other, which limited the extrapolation of our data. Even so, some subtle differences were revealed. ## CONCLUSION Important effects reflecting the nutritional transition over time were documented through evaluation of these two GDM cohorts separated by a 20-year interval, and some differences in applying two different GDM criteria were apparent. Women only meeting the IADPSG/WHO criteria presented pregnancy features that were often linked to obesity, while those meeting the NICE criteria presented worse neonatal outcomes, here represented by hypoglycemia. Further studies focusing on the combined effects of the obesity epidemic and hyperglycemia will help to clarify similarities and differences, and whether these are real, in the profile of pregnancies diagnosed through these two currently used GDM criteria. ## References 1. 1 World Health Organization Diagnostic criteria and classification of hyperglycaemia first detected in pregnancy 2013 Geneva World Health Organization Available from: http://apps.who.int/iris/bitstream/10665/85975/1/WHO_NMH_MND_13.2_eng.pdf Accessed in 2017 (Apr 18). Diagnostic criteria and classification of hyperglycaemia first detected in pregnancy (2013.0) 2. Reichelt AJ, Oppermann MLR. Gestational diabetes diagnosis landmarks. A timeline 3. Reichelt AJ, Oppermann MLR, Schmidt MI. Recomendações da 2a reunião do grupo de trabalho em diabetes e gravidez [Guidelines of the 2nd meeting of the diabetes and pregnancy task force]. Arq Bras Endocrinol Metab (2002.0) 46 574-581 4. Schmidt MI, Duncan BB, Reichelt AJ. Gestational diabetes mellitus diagnosed with a 2-h 75-g oral glucose tolerance test and adverse pregnancy outcomes. Diabetes Care (2001.0) 24 1151-1155. PMID: 11423494 5. Kanguru L, Bezawada N, Hussein J, Bell J. The burden of diabetes mellitus during pregnancy in low- and middle-income countries: a systematic review. Glob Health Action (2014.0) 7 23987-23987. PMID: 24990684 6. Schneider S, Bock C, Wetzel M, Maul H, Loerbroks A. The prevalence of gestational diabetes in advanced economies. J Perinat Med (2012.0) 40 511-520. PMID: 23120759 7. Colagiuri S, Falavigna M, Agarwal MM. Strategies for implementing the WHO diagnostic criteria and classification of hyperglycaemia first detected in pregnancy. Diabetes Res Clin Pract (2014.0) 103 364-372. PMID: 24731475 8. 8 National Collaborating Centre for Women's and Children's Health (UK) Diabetes in Pregnancy: Management of Diabetes and Its Complications from Preconception to the Postnatal Period 2015 London RCOG Press. Diabetes in Pregnancy: Management of Diabetes and Its Complications from Preconception to the Postnatal Period (2015.0) 9. Standards of medical care in diabetes--2014. Diabetes Care (2014.0) 37 S14-S80. PMID: 24357209 10. Salzberg S, Alvariñas J, López G. Guías de diagnóstico y tratamiento de diabetes gestacional. ALAD 2016. Revista de la ALAD (2016.0) 6 155-169 11. Second report. World Health Organ Tech Rep Ser (1980.0) 646 1-80. PMID: 6771926 12. 12 Ministério da Educação (MEC). Secretaria de Educação Superior (SESu). Hospital de Clínicas de Porto Alegre (HCPA) Prestação de Contas Ordinárias Anual. Relatório de Gestão do Exercício de 2013 Porto Alegre 2014 Available from: http://www.hcpa.edu.br/downloads/Publicacoes/relatorio_gestao_hcpa_2013.pdf Accessed in 2017 (Apr 18). Prestação de Contas Ordinárias Anual. Relatório de Gestão do Exercício de 2013 (2014.0) 13. Physical status: the use and interpretation of anthropometry. World Health Organ Tech Rep Ser (1995.0) 854 1-452. PMID: 8594834 14. Rasmussen KM, Yaktine AL. Weight Gain During Pregnancy: Reexamining the Guidelines (2009.0) 15. Tranquilli AL, Dekker G, Magee L. The classification, diagnosis and management of the hypertensive disorders of pregnancy: A revised statement from the ISSHP. Pregnancy Hypertens (2014.0) 4 97-104. PMID: 26104417 16. Alexander GR, Himes JH, Kaufman RB, Mor J, Kogan M. A United States national reference for fetal growth. Obstet Gynecol (1996.0) 87 163-168. PMID: 8559516 17. Feig DS, Corcoy R. Diabetes in pregnancy outcomes: a systematic review and proposed codification of definitions. Diabetes Metab Res Rev (2015.0) 31 680-690. PMID: 25663190 18. 18 Human Development Report 2015 Human Development New York United Nations Development Programme 2015 Available from: https://s3.amazonaws.com/hdr2015report/2015_human_development_report.pdf Accessed in 2017 (Apr 18). Human Development Report 2015 (2015.0) 19. Victora CG, Barreto ML, do Carmo Leal M. Health conditions and health-policy innovations in Brazil: the way forward. Lancet (2011.0) 377 2042-2053. PMID: 21561659 20. Trends in adult body-mass index in 200 countries from 1975 to 2014: a pooled analysis of 1698 population-based measurement studies with 19.2 million participants. Lancet (2016.0) 387 1377-1396. PMID: 27115820 21. Godfrey KM, Reynolds RM, Prescott SL. Influence of maternal obesity on the long-term health of offspring. Lancet Diabetes Endocrinol (2017.0) 5 53-64. PMID: 27743978 22. Catalano PM, McIntyre HD, Cruickshank JK. The hyperglycemia and adverse pregnancy outcome study: associations of GDM and obesity with pregnancy outcomes. Diabetes Care (2012.0) 35 780-786. PMID: 22357187 23. Zhu Y, Zhang C. Prevalence of Gestational Diabetes and Risk of Progression to Type 2 Diabetes: a Global Perspective. Curr Diab Rep (2016.0) 16 7-7. PMID: 26742932 24. Trujillo J, Vigo A, Duncan BB. Impact of the International Association of Diabetes and Pregnancy Study Groups criteria for gestational diabetes. Diabetes Res Clin Pract (2015.0) 108 288-295. PMID: 25765668 25. Egan AM, Heerey AM, Carmody L, Dunne FP. The changing diagnosis of gestational diabetes mellitus: does anyone miss out?. Diabetes Res Clin Pract (2014.0) 106 e53-e55. PMID: 25467618 26. Zhu W, Yang H, Wei Y. Comparing the diagnostic criteria for gestational diabetes mellitus of World Health Organization 2013 with 1999 in Chinese population. Chin Med J (Engl) (2015.0) 128 125-127. PMID: 25563325 27. Sagili H, Kamalanathan S, Sahoo J. Comparison of different criteria for diagnosis of gestational diabetes mellitus. Indian J Endocrinol Metab (2015.0) 19 824-828. PMID: 26693435 28. Meek CL, Lewis HB, Patient C, Murphy HR, Simmons D. Diagnosis of gestational diabetes mellitus: falling through the net. Diabetologia (2015.0) 58 2003-2012. PMID: 26071759 29. Harreiter J, Simmons D, Desoye G. IADPSG and WHO 2013 Gestational Diabetes Mellitus Criteria Identify Obese Women With Marked Insulin Resistance in Early Pregnancy. Diabetes Care (2016.0) 39 e90-e92. PMID: 27208336 30. Worldwide trends in diabetes since 1980: a pooled analysis of 751 population-based studies with 4.4 million participants. Lancet (2016.0) 387 1513-1530. PMID: 27061677
--- title: 'Time trends in physical activity of adult users of the Brazilian National Health System: 2010-2014. Longitudinal study' authors: - Bruna Camilo Turi - Jamile Sanches Codogno - Rômulo Araújo Fernandes - Kyle Robinson Lynch - Eduardo Kokubun - Henrique Luiz Monteiro journal: São Paulo Medical Journal year: 2017 pmcid: PMC10016000 doi: 10.1590/1516-3180.2017.0025190317 license: CC BY 4.0 --- # Time trends in physical activity of adult users of the Brazilian National Health System: 2010-2014. Longitudinal study ## ABSTRACT ### CONTEXT AND OBJECTIVE: In this longitudinal study, we aimed to describe time trends of physical activity (PA) in different domains from 2010 to 2014 among users of the Brazilian National Health System, taking into account the effects of sex, age and economic status (ES). ### DESIGN AND SETTING: Longitudinal study conducted in five primary care units in Bauru (SP), Brazil. ### METHODS: The sample was composed of 620 men and women who were interviewed in 2010, 2012 and 2014. The same group of researchers conducted the interviews, using the questionnaire developed by Baecke et al. Scores for occupational, exercise/sport, leisure-time/transportation and overall PA were considered in this longitudinal survey. Time trends of PA over the four years of follow-up were assessed according to sex, age and ES. ### RESULTS: We found that after four years of follow-up, the reduction in overall PA (-$13.6\%$; $95\%$ confidence interval, CI = -11.9 to -15.3) was statistically significant. Additionally, declines in the occupational domain and exercise/sports participation were affected by age, while the reduction in overall PA was affected by sex, age and ES. ### CONCLUSIONS: Overall PA decreased significantly from 2010 to 2014 among these outpatients of the Brazilian National Health System, and age and male sex were important determinants of PA in its different domains. ## CONTEXTO E OBJETIVO: Neste estudo longitudinal, o objetivo foi descrever as tendências temporais de atividade física (AF) em diferentes domínios de 2010 a 2014 entre usuários do Sistema Único de Saúde, levando em conta o efeito do sexo, idade e condição econômica (CE). ## TIPO DE ESTUDO E LOCAL: Estudo longitudinal realizado em cinco Unidades Básicas de Saúde em Bauru (SP), Brasil. ## MÉTODOS: A amostra foi composta de 620 homens e mulheres que foram entrevistados em 2010, 2012 e 2014. O mesmo grupo de pesquisadores realizou as entrevistas utilizando o questionário desenvolvido por Baecke et al. Escores da AF ocupacional, exercícios/esportes, lazer/transporte e AF global foram considerados neste estudo longitudinal. Tendências temporais de AF nos quatro anos de seguimento foram avaliados de acordo com sexo, idade e CE. ## RESULTADOS: Verificou-se que, após quatro anos de seguimento, a redução da AF total (-13,$6\%$; intervalo de confiança, IC $95\%$ = -11,9 a -15,3) foi estatisticamente significativa. Além disso, o declínio no domínio ocupacional e no exercício/participação esportiva foram afetados pela idade, enquanto a redução na AF total foi afetada pelo sexo, idade e CE. ## CONCLUSÕES: A AF total diminuiu significativamente de 2010 para 2014 em pacientes ambulatoriais do Sistema Único de Saúde, e idade e sexo masculino foram importantes determinantes de AF em seus diferentes domínios. ## INTRODUCTION Development of new technologies and characteristics within the environment have continued to reduce the amount of energy expenditure on a daily basis.1 Physical inactivity is an important risk factor for health, contributing substantially to the worldwide epidemic of non-communicable diseases (NCDs). These cause 5.3 million deaths per year2 and lead to significant economic losses in developed and developing countries.3,4,5,6,7 An active lifestyle is important for promoting good health, and monitoring of lifestyles at population level is critical for decision-making regarding public health. Recent data show that, worldwide, one third of adults do not reach the recommended levels of physical activity (PA).8 Additionally, data from the Global Burden of Disease (GBD) study published in 2015 found discordant trends for low physical activity according to sex, such that the overall summary exposure value for men increased by $2.4\%$, whereas the same indicator for women declined by $1.5\%$.9 In Brazil, a national surveillance system was implemented in 2006 to annually collect data on risk factors for non-communicable diseases, including low levels of PA.10 This initiative should be recognized as a step forward for public health. However, its limitations, such as the facts that the survey is carried out by telephone, it only involves adults living in the state capital cities and the participants of the sample are not the same every year, need to be acknowledged. In developing countries, most studies have a cross-sectional design or only assess total or leisure-time PA.11,12,13 Therefore, they are vulnerable to missing crucial information regarding other PA domains (e.g. occupational, active transportation and sports data). Hence, longitudinal data are important for describing the patterns of habitual PA over time. These make it possible to understand the burden of physical inactivity on health outcomes.14 Monitoring of PA within primary care constitutes an important preventive action. This makes it possible to reach a wide portion of the overall population, thereby averting occurrences of diseases relating to physical inactivity and subsequent economic losses.3,15,16,17 ## OBJECTIVE The objective of this study was to describe time trends of PA in different domains from 2010 to 2014, among users of the Brazilian National Health System, taking into account the effects of sex, age and economic status (ES). ## Sample This longitudinal study was conducted from August 2010 to December 2014, in the city of Bauru, which has around 300,000 inhabitants and is located in the state of São Paulo, the most industrialized region of Brazil. Prior to implementation, the study was approved by the Ethics Committee of Universidade Estadual de São Paulo (UNESP), Bauru campus (procedural number $\frac{1046}{46}$/$\frac{01}{10}$), and all participants gave written and verbal informed consent. The sample size was estimated based on the percentage of the Brazilian population that is covered only by the Brazilian National Health System ($60\%$).18 The parameters used in making the estimate were: $3.8\%$ error (arbitrary because there were no other similar studies), $5\%$ statistical significance and $50\%$ design effect. Therefore, at the baseline, a minimum sample size of 960 participants was estimated to be representative, i.e. at least 192 subjects in each primary healthcare unit (PHU) in Bauru. The primary care of the Brazilian National Health System in Bauru is organized into 17 PHUs, spread out across all geographical regions of the city. To recruit participants for this longitudinal study, we stratified the metropolitan region of the city into five geographical regions in 2010. The biggest PHU of each geographical region was selected to take part in the study. The Municipal Health Department provided lists with the names of all patients attended at these PHUs over the preceding six months. Taking these lists into account, 1,915 patients were randomly selected for telephone contact. This overall number of potential participants was estimated considering that there would be one refusal per two subjects invited to take part in the study. During the telephone contact, the inclusion criteria were checked. Participants needed to be ≥ 50 years of age and to have lived in the area covered by that specific PHU for at least one year. Patients who fulfilled the inclusion criteria were invited to attend an interview and assessment at their own PHU. ## Physical activity assessment The same group of researchers ($$n = 3$$) conducted the interviews using the questionnaire developed by Baecke et al. ,19 at the specific BHU, in a quiet room reserved for the study. The version of the questionnaire used in this cohort study had previously been validated for use in Brazilian Portuguese.20 The questionnaire comprises 16 questions that are scored on a five-point Likert scale, ranging from never to very often/always. It addresses three domains of PA: occupational, leisure-time/locomotion and exercise/sport participation. The occupational domain is composed of eight questions that take into account behavior adopted during work activities, such as sitting, standing, walking, lifting heavy loads, sweating and feeling tired. The exercise/sport participation domain is composed of one question stratified into three sections that take into account the intensity, weekly duration (in hours) and previous length of practice (in months) of the activities performed by the interviewee. The leisure-time/transportation domain is composed of seven questions that take into account behavior during leisure-time/locomotion, like playing sports, watching television, walking and cycling. The PA level is calculated by means of specific equations and is expressed as scores for each PA domain (higher score denotes higher PA). The sum of all domains constitutes the overall PA. Changes in PA scores (considering all PA domains) from 2010 to 2014 were calculated and then expressed as z-scores. In the present study, z-scores ≤ -1.5 were treated as significant reductions in PA over the follow-up (dependent variable), and the sample was split up as ≤ -1.5 or > -1.5 for all PA domains. ## Independent variables The following data were obtained through interviews at the baseline and were structured as categorical variables: Sex (male or female);Age (< 65 years old or ≥ 65 years old); andEconomic status (ES; low or middle/high income). The questionnaire used to estimate ES was a previously validated Brazilian questionnaire,21 which specifies the following income groups: low (classes C, D and E, with family income of US$ 76.94-966.38 per month); and middle/high (classes B and A, with family income of US$ 1,823.33-2,703.61 per month at 3.60 currency exchange rate at the time of the study). The questionnaire estimates the income based on data about education attainment, possession of appliances and physical characteristics of the house (e.g. number of toilets). Additionally, body mass index (BMI) was calculated using measurements of weight and height in each patient and was obtained by dividing weight by squared height (kg/m2). ## Statistical analyses Descriptive statistics were presented as means, medians, standard deviations, interquartile ranges and $95\%$ confidence intervals ($95\%$ CI). The effects of sex, age and ES on PA levels were assessed using analysis of variance (ANOVA) for repeated measurements. Mean values were adjusted using the covariance explained by BMI (baseline) and PHU, thus generating estimated means and standard error means. Ordinal data were expressed as percentages, while Cox regression (expressed in terms of hazard ration [HR] and its $95\%$ CI) was used in the multivariate model adopted, to assess associations with ordinal data (reduction of PA was treated as an outcome). All multivariate models generated using Cox regression were simultaneously adjusted according to sex, age, EC, schooling and PHU. All statistical procedures were conducted using the BioEstat software, version 5.2 (BioEstat, Teffe, Amazonas, Brazil) and statistical significance was set at $P \leq 0.05.$ ## RESULTS We called each of the 1,915 individuals identified in the PHU lists to invite them for a baseline assessment between August and December 2010; 963 ($50.3\%$) agreed to take part in the longitudinal study. Two years after the baseline assessment (August-December 2012), we approached these subjects again. We were able to trace 802 participants: 161 subjects could not be reached or declined to participate and 25 had died. Finally, between August and December 2014, we attempted to locate all participants again and were able to interview 695 of them: 237 subjects could not be reached or declined to participate and another 34 had died. Taking all three assessment periods into account, we thus had data on 620 participants. It should be noted that this is an ongoing cohort study, in which the main purpose is to investigate the relationship between healthcare costs at primary care level and behavioral variables (treated as independent variables). During all three years (2010, 2012 and 2014), the medical records of all participants were assessed ($$n = 904$$, excluding 59 deaths), but only the participants for whom information about PA was available at all three assessments (2010, 2012 and 2014) were included in this study ($$n = 620$$). The final sample was composed of 620 participants (454 women; $73.2\%$) with the three evaluations and its baseline characteristics according to sex are presented in Table 1. Table 1:Characteristics of the sample at baseline according to sex (Bauru, SP, Brazil; $$n = 620$$)SD = standard deviation; IR = interquartile range; $95\%$ CI = $95\%$ confidence interval. Regarding the occupational domain, older people’s PA decreased from 2010 to 2014 (HR = 2.23; $95\%$ CI = 1.29 to 3.86; Table 2). Sex and ES were not associated with modification of occupational PA. Regarding exercise/sports participation, older people presented $90\%$ less likelihood of decreasing their PA in this domain from 2010 to 2014 (HR = 0.10; $95\%$ CI = 0.01 to 0.84; Table 2). On the other hand, sex and ES were not associated with PA modification in this domain. It should be noted that no independent variable was associated with modifications in the leisure-time/transportation domain. Table 2:Changes in physical activity from 2010 to 2014 among adults attended through the Brazilian National Health System ($$n = 620$$)HR = hazard ratio; $95\%$ CI = $95\%$ confidence interval; Cox regression simultaneously adjusted for sex, age, economic condition, primary healthcare unit and schooling. Taking into account the overall PA score, women had $69\%$ less chance of decreased PA from 2010 to 2014 (HR = 0.31/ $95\%$ CI = 0.15 to 0.64; Table 2), while older people presented increased likelihood of reduced PA after four years of follow-up (HR = 2.23; $95\%$ CI = 1.03 to 4.81; Table 2). Figure 1 describes the overall PA during the follow-up, according to sex (Figure 1, Panel A), age (Figure 1, Panel B) and ES (Figure 1, Panel C). Overall PA decreased over time independently of the variables (ANOVA parameter “time”), but women (Figure 1A; P-value = 0.001) and younger adults (Figure 1B; P-value = 0.001) were more active than men and older adults. In the overall sample, after four years of follow-up, the reductions in overall PA (-$13.6\%$; -11.9 to -15.3) and leisure-time/transportation PA (-$28.3\%$; -26.1 to -30.5) were statistically significant, while exercise/sports participation increased ($13.3\%$; 10.8 to 15.7) and occupational PA did not decrease significantly (-$0.9\%$; -6.1 to +4.2). Figure 1:Overall physical activity from 2010 to 2014 according to sex, age and economic status (Bauru, SP, Brazil; $$n = 620$$). ## DISCUSSION In this four-year longitudinal study, PA among outpatients of the Brazilian National Health System showed a significant decrease over time, mostly among men and older subjects. Regarding the occupational domain, we observed that older people decreased their PA over the years more than the younger groups, while sex and ES were not associated with significant changes. This finding was expected when considering the age range adopted in this study for the older group, which was higher than the average retirement age in Brazil (55.1 years for men and 52.2 years for women in 2010).22 Moreover, in order to encompass retired people who maintain PA at home, in our study we also considered household activities in the occupational domain. In agreement with our findings, recent data show that the largest absolute decline in PA among Brazilians is in the occupational domain, but the largest relative decline is in domestic PA.1 The reduction of energy expenditure in this domain is associated with increased overweight/obesity.23 Thus, given the growth in prevalence of obesity and its associated morbidity, mortality and economic impact, remaining physically active within the occupational/domestic domain is a key factor for public health policies. Concerning exercise/sports participation, our results showed that older adults were less likely to decrease their PA in this domain from 2010 to 2014, and no changes according to sex or ES were observed. Considering the age range of the population in our study, it is reasonable to think that the older group (age ≥ 65 years old) was less likely to present modifications to their routines, mainly because most of them had already retired. Moreover, the PA scores in this domain are usually lower in older groups than in younger groups and thus less prone to modifications. Finally, in early old age (65-75 years), there may be a modest increase in physical activity, in an attempt to fill free time resulting from retirement.24 Regarding overall PA, women had $69\%$ less chance of decreased PA from 2010 to 2014, while older people presented increased likelihood of reduced PA after four years of follow-up. In this sample, overall PA decreased by $13\%$ during the follow-up period. In general, women spend less time doing exercise/sports activities, but the time spent doing household activities is substantially higher, which increases the overall score in comparison with men (household activities are performed daily, while exercise/sports during leisure time are not). Regarding the decline in overall PA with age, it is well established in the scientific literature that PA decreases from adolescence to adulthood,25 and is expected to continue decreasing during the aging process. The declining trend in overall PA usually increases the percentage of body fat and reduces muscle strength, agility, flexibility and endurance, thus compromising the ability to remain physically fit.26 However, it is well known that the prevalence and incidence of NCDs increase with age, which emphasizes the importance of becoming physically active or maintaining the existing levels of PA. The biggest strength of our study consisted of its description of longitudinal patterns of PA in different domains according to sociodemographic characteristics. Thus, considering that one of the biggest public health challenges is to encourage individuals to be physically active, and that lifetime PA contains occupational, domestic, sports, leisure-time and transportation domains, there are many possibilities for people to reach the guideline goals of PA for health. Nonetheless, greater availability of environments favorable for PA (which includes schools, workplaces, commuting and the built environment) is urgently needed.27,28 Moreover, our findings indicate that development of public policies in Brazil for promoting leisure-time PA for adults and elderly people, particularly to make up for the decline of occupational PA, has not been entirely effective. This means that the amount of money that fails to be spent on public policies to promote PA might be proportional to the increase in healthcare expenditure for this population. If this indicator were to be adjusted, the country could boost its levels of total PA and, consequently, decrease the burden of undesirable health outcomes. Some limitations should be taken into account in interpreting our results. Firstly, there are no other national studies describing time trends of PA using the same questionnaire, and therefore we were unable to make comparisons. Secondly, PA was self-reported. Although methods of greater accuracy for determining PA levels exist (and these could improve the quality of information), the costs involved and time required to conduct large studies using more direct measurement tools would be greater. Thus, use of a questionnaire seemed more appropriate for this study. Moreover, the same staff members conducted the interviews at all stages of the analysis. Finally, we did not include any information about comorbidities. ## CONCLUSIONS In summary, overall PA decreased significantly from 2010 to 2014 among these outpatients of the Brazilian National Health System, while age and male sex were important determinants of PA in its different domains. The large decrease in overall PA (more than $10\%$) is of concern, especially in this specific age group, in which PA has an impact on prevention and treatment of diseases. PA decreased from 2010 to 2014 for people aged under 65 and among and women. ## References 1. Ng SW, Popkin BM. Time use and physical activity: a shift away from movement across the globe. Obes Rev (2012) 13 659-680. PMID: 22694051 2. Lee IM, Shiroma EJ, Lobelo F. Effect of physical inactivity on major non-communicable diseases worldwide: an analysis of burden of disease and life expectancy. Lancet (2012) 380 219-229. PMID: 22818936 3. Codogno JS, Turi BC, Kemper HC. Physical inactivity of adults and 1-year health care expenditures in Brazil. Int J Public Health (2015) 60 309-316. PMID: 25680327 4. Bielemann RM, Silva BG, Coll CV, Xavier MO, Silva SG. Impacto da inatividade física e custos de hospitalização por doenças crônicas [Burden of physical inactivity and hospitalization costs due to chronic diseases]. Rev Saúde Pública (2015) 49 75-75. PMID: 26487291 5. Cadilhac DA, Cumming TB, Sheppard L. The economic benefits of reducing physical inactivity: an Australian example. Int J Behav Nutr Phys Act (2011) 8 99-99. PMID: 21943093 6. Scarborough P, Bhatnagar P, Wickramasinghe KK. The economic burden of ill health due to diet, physical inactivity, smoking, alcohol and obesity in the UK: an update to 2006-07 NHS costs. J Public Health (Oxf) (2011) 33 527-535. PMID: 21562029 7. Zhang J, Chaaban J. The economic cost of physical inactivity in China. Prev Med (2013) 56 75-78. PMID: 23200874 8. Hallal PC, Andersen LB, Bull FC. Global physical activity levels: surveillance progress, pitfalls, and prospects. Lancet (2012) 380 247-257. PMID: 22818937 9. Global, regional, and national comparative risk assessment of 79 behavioural, environmental and occupational, and metabolic risks or clusters of risks, 1990-2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet (2016) 388 1659-1724. PMID: 27733284 10. 10 Brasil. Ministério da Saúde. Secretaria de Vigilância em Saúde. Departamento de Análise de Situação de Saúde Plano de ações estratégicas para o enfrentamento das doenáas crônicas nãoo transmissíveis (DCNT) no Brasil 2011-2022 Brasília Ministério da Saúde 2011 Available from: http://actbr.org.br/uploads/conteudo/918_cartilha_dcnt.pdf Accessed in 2017 (Apr 25). Plano de ações estratégicas para o enfrentamento das doenáas crônicas nãoo transmissíveis (DCNT) no Brasil 2011-2022 (2011) 11. Del Duca GF, Malta DC, Hallal PC. Indicadores da atividade física em adultos de uma capital do Sul do Brasil comparação entre pesquisas telefônicas e face a face [Physical activity indicators in adults from a state capital in the South of Brazil: a comparison between telephone and face-to-face surveys]. Cad Saúde Pública (2013) 29 2119-2129. PMID: 24127105 12. Matsudo VK, Matsudo SM, Araújo TL. Time trends in physical activity in the state of São Paulo, Brazil: 2002-2008. Med Sci Sports Exerc (2010) 42 2231-2236. PMID: 20404769 13. Mielke GI, Hallal PC, Malta DC, Lee IM. Time trends of physical activity and television viewing time in Brazil: 2006-2012. Int J Behav Nutr Phys Act (2014) 11 101-101. PMID: 25124462 14. Brown W, Blair SN. Good news, good news: occupational and household activities are important for energy expenditure, but sport and recreation remain the best buy for public health. Br J Sports Med (2012) 46 702-703. PMID: 22869787 15. Turi BC, Codogno JS, Fernandes RA. Accumulation of Domain-Specific Physical Inactivity and Presence of Hypertension in Brazilian Public Healthcare System. J Phys Act Health (2015) 12 1508-1512. PMID: 25710729 16. Bielemann RM, Knuth AG, Hallal PRC. Atividade física e redução de custos por doenças crônicas ao Sistema Único de Saúde [Physical activity and cost savings for chronic diseases to the Sistema Único de Saúde]. Revista Brasileira de Atividade Física & Saúde (2015) 15 9-14 17. de Rezende LF, Rabacow FM, Viscondi JY. Effect of physical inactivity on major noncommunicable diseases and life expectancy in Brazil. J Phys Act Health (2015) 12 299-306. PMID: 24769913 18. Kilsztajn S, Silva DF, Camara MB, Ferreira VS. Grau de cobertura dos planos de saúde e distribuição regional do gasto público em saúde [Level of private health insurance coverage and regional distribution of public health expenditure]. Saúde Soc (2001) 10 35-45 19. Baecke JA, Burema J, Frijters JE. A short questionnaire for the measurement of habitual physical activity in epidemiological studies. Am J Clin Nutr (1982) 36 936-942. PMID: 7137077 20. Florindo AA, Latorre MRDO. Validation and reliability of the Baecke questionnaire for the evaluation of habitual physical activity in adult men. Rev Bras Med Esporte (2003) 9 129-135 21. 21 Associação Brasileira de Empresas de Pesquisa Critério de Classificação Econômica Brasil 2008 Available from: https://www.google.com.br/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&ved=0ahUKEwjz2PGryOTPAhVDDZAKHXj8B0MQFggeMAA&url=http%3A%2F%2Fwww.abep.org%2FServicos%2FDownload.aspx%3Fid%3D07&usg=AFQjCNH2G9V5iYOmckiA4iLqDpCE1EYDsQ&sig2=gC9N3UnxmXEYRw9y9vHIrg Accessed in 2017 (Apr 25). Critério de Classificação Econômica Brasil (2008) 22. Camarano AM, Kanso S, Fernandes D. Saída do mercado de trabalho: qual é a idade?. Mercado de trabalho (2012) 51 19-28 23. Singer RH, Stoutenberg M, Gellman MD. Occupational Physical Activity and Body Mass Index: Results from the Hispanic Community Health Study/Study of Latinos. PLoS ONE (2016) 11. PMID: 27031996 24. Shephard RJ. Physical Activity and Aging (1987) 25. Twisk JW, Kemper HC, van Mechelen W. Tracking of activity and fitness and the relationship with cardiovascular disease risk factors. Med Sci Sports Exerc (2000) 32 1455-1461. PMID: 10949012 26. Milanovic Z, Pantelic S, Trajkovic N. Age-related decrease in physical activity and functional fitness among elderly men and women. Clin Interv Aging (2013) 8 549-556. PMID: 23723694 27. Ratzlaff CR. Good news, bad news: sports matter but occupational and household activity really matter - sport and recreation unlikely to be a panacea for public health. Br J Sports Med (2012) 46 699-701. PMID: 22411766 28. Sallis JF, Cervero RB, Ascher W. An ecological approach to creating active living communities. Annu Rev Public Health (2006) 27 297-322. PMID: 16533119
--- title: 'Analysis of quality of life among asthmatic individuals with obesity and its relationship with pulmonary function: cross-sectional study' authors: - Letícia Baltieri - Luiz Claudio Martins - Everton Cazzo - Débora Aparecida Oliveira Modena - Renata Cristina Gobato - Elaine Cristina Candido - Elinton Adami Chaim journal: São Paulo Medical Journal year: 2017 pmcid: PMC10016004 doi: 10.1590/1516-3180.2016.0342250217 license: CC BY 4.0 --- # Analysis of quality of life among asthmatic individuals with obesity and its relationship with pulmonary function: cross-sectional study ## ABSTRACT ### CONTEXT AND OBJECTIVE: The combined effect of obesity and asthma may lead to significant impairment of quality of life (QOL). The aim here was to evaluate the prevalence of asthma among obese individuals, characterize the severity of impairment of quality of life and measure its relationship with pulmonary function. ### DESIGN AND SETTING: Observational cross-sectional study in public university hospital. ### METHODS: Morbidly obese individuals (body mass index > 40 kg/m2) seen in a bariatric surgery outpatient clinic and diagnosed with asthma, were included. Anthropometric data were collected, the Standardized Asthma Quality of Life Questionnaire (AQLQ(S)) was applied and spirometry was performed. The subjects were divided into two groups based on the median of the score in the questionnaire (worse < 4 and better ≥ 4) and were compared regarding anthropometric data and pulmonary function. ### RESULTS: Among the 4791 individuals evaluated, 219 were asthmatic; the prevalence of asthma was $4.57\%$. Of these, 91 individuals were called to start multidisciplinary follow-up during the study period, of whom 82 answered the questionnaire. The median score in the AQLQ(S) was 3.96 points and, thus, the individuals were classified as having moderate impairment of their overall QOL. When divided according to better or worse QOL, there was a statistically difference in forced expiratory flow (FEF) 25-$75\%$, with higher values in the better QOL group. ### CONCLUSION: The prevalence of asthma was $4.57\%$ and QOL was impaired among the asthmatic obese individuals. The worst QOL domain related to environmental stimuli and the best QOL domain to limitations of the activities. Worse QOL was correlated with poorer values for FEF 25-$75\%$. ## CONTEXTO E OBJETIVO: O efeito combinado de obesidade e asma pode levar a um comprometimento significativo da qualidade de vida (QV). O objetivo foi avaliar a prevalência de asma entre obesos, caracterizar a gravidade do comprometimento da QV e verificar sua relação com a função pulmonar. ## TIPO DE ESTUDO E LOCAL: Estudo transversal observacional em hospital universitário público. ## MÉTODOS: Foram incluídos indivíduos obesos mórbidos (indice de massa corporal > 40 kg/m2), acompanhados num ambulatório de cirurgia bariátrica e diagnosticados com asma. Foram coletados dados antropométricos e aplicado o Standardized Asthma Quality of Life Questionnaire (AQLQ (S)), bem como a espirometria. Os indivíduos foram divididos em dois grupos com base na mediana obtida no escore do questionário (pior < 4 e melhor ≥ 4) e os grupos foram comparados aos dados antropométricos e função pulmonar. ## RESULTADOS: Dos 4.791 indivíduos avaliados, 219 eram asmáticos; a prevalência de asma foi de 4,$57\%$. Destes, 91 indivíduos foram chamados para iniciar o acompanhamento multidisciplinar no período do estudo, sendo que 82 responderam ao questionário. A pontuação mediana do AQLQ (S) foi de 3,96 pontos, portanto, classificados com prejuízo moderado na QV global. Quando divididos por melhor ou pior QV, houve diferença estatística no fluxo expiratório forçado (FEF) 25-$75\%$, com maior valor no grupo com melhor QV. ## CONCLUSÃO: A prevalência da asma na população estudada foi de 4,$57\%$ e há prejuízos na QV de obesos asmáticos, sendo o pior domínio de QV relacionado aos estímulos ambientais e o melhor domínio de QV relacionado às limitações das atividades. A pior QV se relacionou a piores valores de FEF 25-$75\%$. ## INTRODUCTION Asthma is a chronic inflammatory disease of the airways that is associated with hyper-responsivity. It leads to recurrent episodes of wheezing, dyspnea, sensation of chest tightness and coughing, particularly at night or in the early morning. The obstruction to the airflow may be reversed spontaneously or by means of treatment. About 300 million individuals worldwide present asthma. The factors associated with the disease include environmental factors relating to allergies, occupation, smoking, infections, pollution and diet; and endogenous factors relating to genetics, gender and obesity.1 Asthma is diagnosed based on the symptoms and is confirmed through pulmonary function tests, such as spirometry and expiratory flow peak measurement. These enable evaluation of the severity of the limitation to the airflow and its reversibility and variability.1 In a meta-analysis by Beuther et al. ,2 it was observed that obese individuals were more likely to develop asthma than were lean individuals. The exact mechanism for development of asthma is uncertain, but the inflammation mediators produced by the adipose tissue may contribute towards a low-grade systemic inflammatory state and promote changes to pulmonary function, thus leading to episodes of bronchospasm. Today, obesity has reached epidemic levels and has become a public health concern. In 2014, more than 1.9 billion adult individuals ($39\%$) were at least overweight, and of these, more than 600 million were obese.3 Obesity is defined as body mass index (BMI) greater than or equal to 30 kg/m2 and considered to be a multifactorial disease.4 Its probable causes are a combination of genetic, endocrine, behavioral, socioeconomic, psychological and environmental imbalances, and it leads to several comorbidities.5 Follow-up for asthmatic patients is necessary, with the aims of controlling the condition and avoiding exacerbations and the need for in-hospital assistance, especially when it is associated with obesity. Assessment of this information by means of questionnaires is useful within clinical practice and scientific research, since this allows standardization and reproducibility of measurements at low cost. ## OBJECTIVE The aims of this study were to evaluate the prevalence of asthma in the obese population, characterize its severity of impairment of the quality of life of asthmatic obese individuals and measure its influence on pulmonary function. ## Study design and setting This was an observational cross-sectional study conducted at the bariatric surgery outpatient clinic of our university’s teaching hospital. It was submitted for evaluation and was then approved by the local ethics review board (289.425). The laws and norms regarding studies on humans were followed, in accordance with resolution $\frac{196}{96}$ of the National Health Council and all the participants in the study signed an informed consent statement. ## Sampling and participants The power of the sample was calculated based on the global AQLQ(S) (Standardized Asthma Quality of Life Questionnaire) and a sample power of $88\%$ was obtained. The inclusion criteria were that the subjects needed to: present morbid obesity (BMI ≥ 40 kg/m2);be candidates for bariatric surgery;have a clinical diagnosis of asthma in accordance with the Global Initiative for Asthma consensus statement1 and/or antecedents of any episode of bronchospasm at any time during their lives and/or current or previous use of medication to treat asthma. The exclusion criteria were the presence of: smoking habit;cognitive impairment that could impede performance of the clinical tests and completion of the questionnaire;respiratory diseases other than asthma;congestive heart failure or cardiovascular ischemic disease. The recruitment period for the participants was from February 2015 to April 2016. Adult individuals were screened at the time of registration to enter the outpatient clinic and become candidates for bariatric surgery. On this occasion, they filled out a registration form that asked for information about the presence of asthma. Those who reported having asthma or experiencing episodes of bronchospasm without an ultimate diagnosis, and who fulfilled the other criteria, were then informed about the procedures of the study and were invited to take part in it. The procedures would involve clinical confirmation of the diagnosis of asthma by means of consultations with a physiotherapist, evaluation with a pneumologist physician and performance of spirometry. ## Pulmonary function tests and asthma diagnosis Asthma was investigated based on the symptoms that individuals reported having had over their whole lifetime, such as episodes of bronchospasm, breathlessness, sensation of chest tightness and coughing,1,6 or in situations in which individuals were routinely using medications for asthma, in accordance with the Global Initiative for Asthma consensus statement.1 Once diagnosed, these individuals would undergo pulmonary function tests to assess the severity of the disease. Other respiratory diseases were excluded based on anamnesis and pulmonary function test. Spirometry was performed at the Pulmonary Function Laboratory under supervision by a technical team and the norms of the American Thoracic Society (ATS) and European Respiratory Society (ERS)7 were followed. To evaluate measurements of pulmonary volumes and flows, two maneuvers were performed: slow vital capacity and forced vital capacity. The maneuvers were performed repeatedly until three acceptable curves were obtained, of which two needed to be reproducible. The total number of trials could not exceed eight. The subjects rested for 10 minutes before the test and received appropriate orientations during the test. The maneuvers were performed at two times: before and after using a bronchodilator (salbutamol, 200-400 µg) to observe the increase in the forced expiratory volume in the first second (FEV1) and/or the peak expiratory flow (PEF). Asthma is diagnosed when there is a $12\%$ or 200 ml increase in FEV1 and a $20\%$ or 60 liter/min increase in PEF, in relation to the pre-bronchodilator values. The subjects were instructed to suspend their use of bronchodilator for 8-12 hours before the test.1,6 ## Evaluation procedures and outcome measurements Antropometric data were collected and the quality of life was assessed follows. The following anthropometric data were collected: weight, height and BMI. Weight was measured by means of a digital weighing machine (Filizola ID-1500, Brazil), with a capacity of 300 kg capacity and precision of 0.1 kg. Height was measured by means of a wall-mounted stadiometer, with a capacity of 2 meters and precision of 0.1 cm. Body mass index (BMI) was calculated by means of Quetelet’s formula,8 i.e. weight/(height2). Quality of life was assessed by means of the Standardized Asthma Quality of Life Questionnaire (AQLQ(S)), which is a self-applicable questionnaire consisting of 32 questions that evaluate the last two weeks within four separate domains (impairment of activities, symptoms, emotions and environmental stimuli). It was developed by Juniper et al.9 and was validated and standardized by Juniper et al.10 It has been translated into Portuguese for use in Brazil, as well as into more than 30 other languages. The Brazilian Portuguese version was validated and was considered to have good reproducibility and characteristics similar to those of the original instrument.11 Thus, it could be used for the population of the present study. The questionnaire scores are calculated from the means of each domain; the scores range from 1 to 7. The higher the score is, the better the quality of life is. The questionnaire contains specific questions relating to asthma and respiratory symptoms that are triggered in specific activities and, therefore, assesses these conditions without connection with obesity. ## Statistical analysis The data were encoded for the SPSS 13.0 software and descriptive analysis was performed. The individuals were divided into two groups based on the scores obtained from the questionnaire (better or worse quality of life). The cutoff value for defining the groups was obtained through descriptive analysis on the overall AQLQ(S), which found a median score of 3.96 points. Thus, the cutoff value of 4 was used. In addition, according to Juniper et al. ,9 4.0 is an intermediate score in the questionnaire and therefore separates between worse and better quality of life. In this manner, the subjects were then divided between group 1 (worse QOL; score < 4) and group 2 (better QOL; score ≥ 4). These groups were compared regarding their anthropometric data and pulmonary function results, by means of the Mann-Whitney test. The significance level used was $5\%$ (P-value < 0.05). ## RESULTS During the study period, there were three inscription events to enlist candidates for bariatric surgery at our service. On the first occasion (March 2014), 1,782 individuals were registered and, of these, 82 ($4.6\%$) were reported as asthmatic; on the second occasion (December 2014), 1,781 were registered and, of these, 61 ($3.42\%$) were reported as asthmatic; and on the third occasion (November, 2015), 1,228 individuals were registered and, of these, 76 ($6.18\%$) were reported as asthmatic. Hence, out of an overall population of 4,791 individuals with obesity, 219 ($5.57\%$) reported having asthma symptoms. Figure 1 shows a graphic representation of this phase of the recruitment. Figure 1.Patient recruitment flow in the study. For patients to be called up to begin the preoperative program, the criteria used were their severity of obesity and comorbidities, their position on the waiting list and the availability of preoperative examinations and surgical vacancies. Up to the end of the study period, 91 asthmatic individuals were called up to enter the program and, of these, 82 adequately filled out the proposed questionnaire. All the individuals in our sample who reported in this file that they had suggestive symptoms and who were called up for the program were confirmed as having a clinical diagnosis of asthma. Anthropometric data were collected from these 82 individuals and are presented in Table 1. After analysis on the sample, they were stratified as having “better” or “worse” QOL, according to the scores obtained in the questionnaire. The features that significantly differed between the groups were height ($$P \leq 0.004$$) and pre-bronchodilator pulmonary function test values, which presented a significant difference regarding forced expiratory flow (FEF) 25-$75\%$, which was higher in the group with better QOL group ($$P \leq 0.043$$). Table 2 shows the characteristics of both groups. Table 1.Overall characteristics of the study population. Data expressed as means and standard deviations (SD), and as medians and quartilesBMI = body mass index; AQLQ(S) = Standardized Asthma Quality of Life Questionnaire; FVC = forced vital capacity; FEV1 = forced expiratory volume in 1st second; PEF = peak expiratory flow; FEF = forced expiratory flow; FIVC = forced inspiratory vital capacity; PIF = peak inspiratory flow; L = liter. Table 2.Comparison of the pre-bronchodilator pulmonary function tests between the “better” and “worse” quality of life (QOL) groups (cutoff value = 4)Statistically significant P value; BMI = body mass index; AQLQ(S) = Standardized Asthma Quality of Life Questionnaire; FVC = forced vital capacity; FEV1 = forced expiratory volume in 1st second; PEF = peak expiratory flow; FEF = forced expiratory flow; FIVC = forced inspiratory vital capacity; PIF =: peak inspiratory flow; L = liter. ## DISCUSSION In a meta-analysis conducted by Beuther et al. ,2 obese individuals presented higher risk of developing asthma than did lean subjects. The prevalence of asthma in the present study was $4.57\%$ in a population of 4,791 individuals. This prevalence is low in comparison with what was found in the study by Melo et al. ,12 which was $18.5\%$ in a population of 363 obese individuals. However, in our study, asthma was reported in our subjects’ registration files for their entry to the preoperative program for bariatric surgery, i.e. before contact with the multiprofessional team or detailed clinical interview. It is known that asthma may be underdiagnosed in low-income obese populations for several reasons, such as poor access to information or to specific healthcare services that provide diagnosis and management of asthma. Moreover, individuals may interpret their own episodes of wheezing as physical tiredness caused by obesity, which would remit without use of medications or medical evaluation. In such situations, they might not provide this information at the time that the registration file is filled out. However, all the individuals in our sample who reported in this file that they had suggestive symptoms and who were called up for the program were confirmed as having a clinical diagnosis of asthma, determined through the reported clinical history. All the individuals in this study present grade III obesity (BMI 40-49.9 kg/m2), with a mean BMI of 45.27 ± 6.79 kg/m2. Grade III obesity causes severe changes to pulmonary function due to several factors, such as fat deposition around the thorax and abdomen, which limits adequate movements of the thorax13 and changes pulmonary compliance.14,15 This leads to microatelectasis in the pulmonary inferior lobes16,17 and reduces functional capacity,13,18 which compromises performance of simple daily activities, due to early tiredness. Furthermore, the low-grade systemic inflammatory state caused by fat tissue has the capacity to influence the lung parenchyma,19 thereby leading to episodes of bronchospasm. Besides changes to pulmonary function, obesity may lead to physical limitations, postural changes and joint overload,20 which gives rise to joint pain and impairment of walking ability and daily activities. Such impairments, both pulmonary and physical, directly affect the QOL of these individuals, and weight loss is strongly recommended. Hence, the individuals called up for the study were instructed to begin preparations for the preoperative assessment for bariatric surgery, which favors a healthy lifestyle, especially regarding diet and physical activity. The QOL data obtained demonstrated that the individuals scored in the medium band of the score scale from 1-7 (median = 3.96 points) and, thus, presented moderately compromised QOL in all the domains evaluated. The domain with the best final score related to limitations on activities (median = 4.13 points) and the worst related to environmental stimuli (median = 2.62 points). The domain with the best score (albeit still denoting moderate impairment), relating to limitations on activities, comprised questions on specific daily activities that may cause episodes of bronchospasm and breathlessness and the degree of limitation that these cause to the individual (such as walking, running, practicing exercises, working, socializing etc.). These were not necessarily physical limitations, but could also be limitations relating to fear of exposure to risky situations. The worst-scoring domain related to environmental stimuli, which comprised specific questions on symptoms caused by smoke, dust, foul weather, pollution and perfume fragrances. External environmental stimuli may potentiate systemic pulmonary inflammation, thus leading to hyperresponsivity of the airways and episodes of bronchospasm. When the individuals were stratified into two groups according to their asthma-related QOL, it was observed that the individuals with better QOL also presented significantly higher FEF 25-$75\%$ ($$P \leq 0.043$$). FEF represents the mean forced expiratory flow in the intermediate band of forced vital capacity (FVC), i.e. between 25 and $75\%$ of the FVC curve.21 FEF 25-$75\%$ depends on the elastic retraction force of the lungs, the permeability of the small airways and the muscle strength. Its measurement provides information on the permeability of the small airways and is unrelated to the patient’s collaboration.22 Thus, all of the mechanical and inflammatory changes present in the lungs of morbidly obese individuals may lead to changes in the permeability of low-caliber airways, which is mirrored in measurements of FEF 25-$75\%$.12 Although the FEV1/FVC% ratio is the measurement that best represents obstructive disorders,21 it was normal in our study, albeit at the lower limit. According to Pereira,21 patients with established chronic obstructive pulmonary disease (COPD) tend to show much more surprising changes in FEF 25-$75\%$ than in the FEV1/FVC% ratio. However, because of the correlation between FEF 25-$75\%$ and FEV1/FVC%, the FEF 25-$75\%$ measurement becomes redundant when the FEV1/FVC% ratio is abnormal. Therefore, if the FEV1/FVC ratio is borderline, a reduction in FEF 25-$75\%$ or other terminal flows indicates airflow obstruction in individuals with symptomatic respiratory disorders. According to Lebecque et al. ,22 for mild asthma, FEF 25-$75\%$ appeared to be more sensitive than the FEV1/FVC ratio for indicating the presence of small-caliber airway obstruction. In the present study, although no relationship was found for other spirometric variables, it could be seen that the values of FEV1, FEF and FEF $25\%$ also were below the normal range when the non-stratified sample was analyzed, this finding is expected in asthmatic individuals.21,23 Some studies in which pulmonary function tests were performed on obese individuals without pulmonary abnormalities showed significant reductions in functional residual capacity (FRC)13,24 and expiratory reserve volume (ERV)13,25,26 that were attributable to the mechanical changes that fat tissue causes to the thorax. Nonetheless, the changes in FEF 25-$75\%$ was attributed by Sood13 to inflammatory changes that occurred in the lungs of obese individuals, thereby leading to premature closure of the small airways during forced expiration. This might explain the relationship between the severity of asthma and the observed values of FEF 25-$75\%$. Such changes in the small airways may contribute towards situations in which low effort or low environmental stimuli provoke episodes of bronchospasm and breathlessness, thus compromising the QOL of these individuals. There is recent evidence highlighting the burdens on QOL caused by certain situations, such as asthma. In a study that used the same QOL evaluation questionnaire as in the present study, Rocha27 observed that asthma had a significant impact on QOL even when partly controlled. Furthermore, in a review of literature conducted by Araújo et al. ,28 it was concluded that the QOL and sleep quality of asthmatic individuals were compromised. On the other hand, in a study by Pereira et al.29 that used the Saint George’s Respiratory Questionnaire (SGRQ) to assess the QOL of individuals with asthma and chronic obstructive pulmonary disease (COPD), it was observed that, when the disease was classified as mild to moderate and was adequately treated, there was no impairment of QOL. In studies that evaluated the impact of obesity on QOL, there was evidence that obesity led to impairment of QOL. Weight loss might improve the overall QOL within this group.30,31,32 Hence, since impairments of QOL occur in both diseases, an association between them would be expected to cause even more damage. This explains the importance of measuring QOL in these cases, in such a way that therapeutic strategies and goals can be designed. Since we identified that worse QOL in the present study was related to greater impairment of pulmonary function, it is possible for the attending physician to identify individuals with disease of greater severity by means of a simple questionnaire that may be self-applicable. This would reduce the need for additional pulmonary function tests and, thus, minimize the cost of therapy for these individuals, since improvement of the symptoms and QOL should be the ultimate goal. Therefore, the possibility of classifying the QOL of asthmatic obese individuals by means of a questionnaire may provide attending physicians with significant information on the degree of impairment of pulmonary function in these individuals and make it possible to define strategies for better and individualized therapy. ## Limitations Although the Brazilian Portuguese version of the questionnaire has many properties similar to the original instrument, and is a valid instrument for this population according to the authors who validated it, these authors mentioned in their validation study that hardly any study can claim to provide full validation. Therefore, studies that validate the questionnaire more appropriately would be required, and this might constitute a form of bias for research that uses the instrument. Nonetheless, the original questionnaire was translated into Brazilian Portuguese in accordance the internationally accepted methodology. ## CONCLUSION The prevalence of asthma in the study population was $4.57\%$. The QOL of individuals with asthma and obesity was impaired. The worst QOL domain related to environmental stimuli and the best QOL domain related to the limitations of the activities. Worse QOL correlated with lower values for FEF 25-$75\%$ in the pulmonary function test. ## References 1. 1 Global Initiative for Asthma Global strategy for asthma management and prevention 2017 http://ginasthma.org/2017-gina-report-global-strategy-for-asthma-management-and-prevention Accessed in 2017 (Apr 6). Global strategy for asthma management and prevention* (2017.0) 2. Beuther DA, Sutherland ER. Overweight, obesity, and incident asthma: a meta-analysis of prospective epidemiologic studies. Am J Respir Crit Care Med (2007.0) 175 661-666. PMID: 17234901 3. 3 World Health Organization. Media centre Obesity and overweight http://www.who.int/mediacentre/factsheets/fs311/en/ Accessed in 2017 (Apr 6). Obesity and overweight 4. 4 World Health Organization Global strategy on diet, physical activity and health. World Health Organization Geneva World Health Organization 2003 http://apps.who.int/gb/ebwha/pdf_files/WHA57/A57_R17-en.pdf Accessed in 2017 (Apr 6). Global strategy on diet, physical activity and health. World Health Organization (2003.0) 5. Yurcisin BM, Gaddor MM, DeMaria EJ. Obesity and bariatric surgery. Clin Chest Med (2009.0) 30 539-553. PMID: 19700051 6. Diretrizes da Sociedade Brasileira de Pneumologia e Tisiologia para o Manejo da Asma - 2012. Jornal Brasileiro de Pneumologia (2012.0) 38 S1-S46 7. Miller MR, Hankinson J, Brusasco V. Standardisation of spirometry. Eur Respir J (2005.0) 26 319-338. PMID: 16055882 8. Quetelet AD. Antropométrie ou Mésure des Différences Facultés de l’Homme (1871.0) 9. Juniper EF, Guyatt GH, Epstein RS. Evaluation of impairment of health related quality of life in asthma: development of a questionnaire for use in clinical trials. Thorax (1992.0) 47 76-83. PMID: 1549827 10. Juniper EF, Buist AS, Cox FM, Ferrie PJ, King DR. Validation of a standardized version of the Asthma Quality of Life Questionnaire. Chest (1999.0) 115 1265-1270. PMID: 10334138 11. Silva LMC, Silva LCC. Validação do questionário de qualidade de vida em asma (Juniper) para o português brasileiro [Validation of asthma quality of life questionnaire (Juniper) to brazilian portuguese]. Revista da AMRIGS (2007.0) 51 31-37 12. Melo SMD, Alves AJ, Menezes RS, Melo VA. Prevalência e gravidade de asma brônquica em adultos obesos com indicação de cirurgia bariátrica [Prevalence and severity of asthma in obese adult candidates for bariatric surgery]. J Bras Pneumol (2011.0) 37 326-333. PMID: 21755187 13. Sood A. Altered resting and exercise respiratory physiology in obesity. Clin Chest Med (2009.0) 30 445-454. PMID: 19700043 14. Dumont L, Mattys M, Mardirosoff C. Changes in pulmonary mechanics during laparoscopic gastroplasty in morbidly obese patients. Acta Anaesthesiol Scand (1997.0) 41 408-413. PMID: 9113188 15. Pelosi P, Croci M, Calappi E. Prone positioning improves pulmonary function in obese patients during general anesthesia. Anesth Analg (1996.0) 83 578-583. PMID: 8780285 16. Eichenberger AS, Proietti S, Wicky S. Morbid obesity and postoperative pulmonary atelectasis: an underestimated problem. Anesth Analg (2002.0) 95 1788-1792. PMID: 12456460 17. Baltieri L, Peixoto-Souza FS, Rasera-Junior I. Análise da prevalência de atelectasia em pacientes submetidos à cirurgia bariátrica [Analysis of the prevalence of atelectasis in patients undergoing bariatric surgery]. Rev Bras Anestesiol (2016.0) 66 577-582. PMID: 27639505 18. McCallister JW, Adkins EJ, O’Brien JM. Obesity and acute lung injury. Clin Chest Med (2009.0) 30 495-508. PMID: 19700048 19. Thyagarajan B, Jacobs DR, Smith LJ. Serum adiponectin is positively associated with lung function in young adults, independent of obesity: the CARDIA study. Respir Res (2010.0) 11 176-176. PMID: 21143922 20. Toivanen AT, Heliövaara M, Impivaara O. Obesity, physically demanding work and traumatic knee injury are major risk factors for knee osteoarthritis--a population-based study with a follow-up of 22 years. Rheumatology (Oxford) (2010.0) 49 308-314. PMID: 19946021 21. Pereira CAC. Espirometria. J Pneumol (2002.0) 28 S1-S82 22. Lebecque P, Kiakulanda P, Coates AL. Spirometry in the asthmatic child: is FEF25-75 a more sensitive test than FEV1/FVC?. Pediatr Pulmonol (1993.0) 16 19-22. PMID: 8414736 23. Mallozi MC, Rozov T. O laboratório nas doenças pulmonares [Laboratorial tests in pulmonary diseases]. J Pediatr (Rio J.) (1998.0) 74 S125-S132. PMID: 14685581 24. Guimarães C, Martins MV, Moutinho dos Santos J. Função pulmonar em doentes obesos submetidos a cirurgia bariátrica. Revista Portuguesa de Pneumologia (2012.0) 18 115-119. PMID: 22402178 25. Rasslan Z, Saad R, Stirbulov R, Fabbri RMA, Lima CAC. Avaliação da função pulmonar na obesidade graus I e II [Evaluation of pulmonary function in class I and II obesity]. J Bras Pneumol (2004.0) 30 508-514 26. Baltieri L, Santos LA, Pazzianotto-Forti EM, Montebelo MIL, Rasera-Junior I. Uso da pressão positiva em cirurgia bariátrica e efeitos sobre a função pulmonar e prevalência de atelectasias: estudo randomizado e cego [Use of positive pressure in the bariatric surgery and effects on pulmonary function and prevalence of atelectasis: randomized and blinded clinical trial]. ABCD Arq Bras Cir Dig (2014.0) 27 26-30 27. Rocha CC. Qualidade de vida e inflamação das vias aéreas em diferentes níveis de controle da asma [dissertation] (2013.0) 28. Araújo DL, Souza-Machado A, Souza-Machado C, Salles C. Avaliação da qualidade do sono e da qualidade de vida na asma [Evaluation of quality of sleep and quality of life in asthma]. Braz J Allergy Immunol (2014.0) 2 107-111 29. Pereira EA, Ferreira PR, Araújo MEA, Carvalho STRF, Carvalho LN. Estudo comparativo da qualidade de vida entre pacientes com doença pulmonar obstrutiva crônica e pacientes asmáticos. Revista Ceuma Perspectivas (2016.0) 27 31-42 30. Pimenta GP, Jaudy TR, Moura DN. Avaliação da qualidade de vida tardia após gastroplastia vertical [Long-term quality of life after vertical sleeve gastroplasty]. Rev Col Bras Cir (2013.0) 40 453-457. PMID: 24573622 31. Barros LM, Moreira RAN, Frota NM, Araújo TM, Caetano JA. Qualidade de vida entre obesos mórbidos e pacientes submetidos à cirurgia bariátrica [Quality of life among morbid obese and patients submitted to bariatric surgery]. Revista Eletrônica de Enfermagem (2015.0) 17 312-321 32. Gomes T, Teixeira RAT, Nascimento DC. Qualidade de vida e síndrome metabólica em mulheres brasileiras: análise da correlação com a aptidão aeróbia e a força muscular [Quality of life and metabolic syndrome in Brazilian women: analysis of the correlation with aerobic fitness and muscle strength]. Motri (2015.0) 11 48-61
--- title: Quality of diet plans for weight loss featured in women’s magazines. A cross-sectional descriptive study authors: - Maiara Martinighi - Edina Mariko Koga da Silva journal: São Paulo Medical Journal year: 2017 pmcid: PMC10016006 doi: 10.1590/1516-3180.2016.0301280217 license: CC BY 4.0 --- # Quality of diet plans for weight loss featured in women’s magazines. A cross-sectional descriptive study ## ABSTRACT ### CONTEXT AND OBJECTIVE: Brazil has the fifth largest population of obese individuals in the world. Women’s magazines publish a large number of diet plans, and therefore the objective of this study was to assess the quality of these plans. ### DESIGN: Cross-sectional descriptive study. ### METHODS: We included the Brazilian women’s magazines of highest circulation published between January and June 2014 that advertised diets for weight loss on their covers. We extracted the quantities of macro and micronutrients from each of these diet plans and compared these quantities with the World Health Organization nutritional guidelines for adult women. We also checked the total energy quantities of these plans, and any recommendations about water intake and physical activity. ### RESULTS: We identified 136 potentially eligible magazine issues; 41 were excluded and 95 issues of 6 different magazines were included in the study. We found that $83.1\%$ of the plans had carbohydrate and fiber levels below the recommendations. On the other hand, the protein and saturated fatty acid levels were above the recommendations in $97.8\%$ and $95.7\%$ of the plans, respectively; $75.7\%$ of the diets had inadequate calcium levels and $70.5\%$ had low iron levels. Only 30 plans specified the total daily quantity of dietary energy and in $53.3\%$ of these, the information was inconsistent with our estimates; $20\%$ of the plans had no recommendations on daily water intake and $37.5\%$ did not give recommendations regarding physical activity practices. ### CONCLUSION: The diet plans for weight loss featured in Brazilian women’s magazines are of low quality. ## CONTEXTO E OBJETIVO: O Brasil tem a quinta maior população de obesos do mundo. Revistas femininas publicam um grande número de planos dietéticos. Assim, o objetivo deste estudo foi avaliar a qualidade desses planos. ## TIPO DE ESTUDO: Estudo descritivo transversal. ## MÉTODOS: Incluímos as revistas femininas brasileiras de maior circulação publicadas entre janeiro e junho de 2014 que tivessem uma chamada de dieta para perda de peso em sua capa. De cada plano dietético, foi extraída a quantidade de macro e micronutrientes. Comparamos esses valores com as diretrizes nutricionais da Organização Mundial de Saúde para mulheres adultas. Verificamos também a quantidade total de energia, se o plano recomendava consumo de água e prática de atividade física. ## RESULTADOS: Um total de 136 exemplares potencialmente elegíveis foi identificado; 41 foram excluídos e 95 exemplares de 6 revistas diferentes foram incluídos no estudo. A análise mostrou que 83,$1\%$ dos planos apresentaram valores de carboidratos e de fibras menores do que os recomendados. Por outro lado, o teor de proteína e de ácidos graxos saturados foi maior do que o recomendado em 97,$8\%$ e 95,$7\%$ dos planos, respectivamente, e 75,$7\%$ dos planos tinham teor inadequado de cálcio e 70,$5\%$ de ferro. Apenas 30 planos especificaram a quantidade total de energia diária e em 53,$3\%$ desses, a informação foi discordante com nossas estimativas; $20\%$ dos planos não recomendavam consumo diário de água, 35,$7\%$ não recomendavam a prática de atividade física. ## CONCLUSÃO: A qualidade dos planos dietéticos para perda de peso veiculados em revistas femininas brasileiras é baixa. ## INTRODUCTION There are currently 2.1 billion overweight individuals in the world, which represents a huge increase in relation to the 875 million overweight individuals in the 1980s. Estimates indicate that obesity and overweight caused 3.4 million deaths worldwide in 2010 and reduced life expectancy by $3.9\%$ years.1 Brazil is the country with the fifth largest number of obese individuals in the world, surpassed only by the United States, China, India and Russia.1 In São Paulo, the largest city in Brazil, $18.2\%$ of all women and $17.5\%$ of all men are obese.2 Emotional and genetic factors, along with an overall increase in dietary energy and a sedentary lifestyle, are the main causes of the increased prevalence of obesity.3,4 In parallel with the national increase in the prevalence of obesity, the last three decades stand out as a period during which body image and “being fit” became fundamental values within Brazilian culture. Nowadays, fitness and body image are highly prioritized and a large number of individuals regard any slight weight gain as a major problem. Many people whose weight is within the normal range feel that they are overweight. In this culture, simply accepting one’s body shape is frowned upon since the body is always imperfect and in need of correction or transformation.5,6,7 Despite the many health consequences and the high economic costs of obesity, the search for an efficient diet for weight loss and maintenance is still ongoing. The World Health Organization (WHO) recommends that, in order to promote healthy and gradual weight loss, diet plans should limit dietary energy consumption and encourage physical activity.8,9 There is relentless pursuit of a fit body that conforms to the standards imposed by society, and constant surveillance of food intake according to recommended diet plans that produce a slow and steady weight loss. These coexist with an avalanche of popular diet plans that promise quick and easy ways to slim down. Most modern popular diets encourage reducing or excluding a specific macronutrient or excessive reduction of total dietary energy, and are only efficient over the short term.10,11,12 The media plays an important role in providing information about eating habits, nutrition, health and related matters, including diet plans and fitness programs.12,13,14 Most women are exposed to all kinds of diet plans for weight loss in magazines, which also lead them to be intensely concerned about their weight, appearance and body image. Women of all sizes and shapes are motivated to lose weight because of fashion and not because of the health risks associated with being obese or overweight.11 It is also increasingly common to find nutrition articles in popular magazines that include interviews with celebrities. When public figures (especially actors and athletes) give testimonials about their eating habits, these can have far-reaching and potentially harmful effects on the population because many readers tend to believe any information endorsed by role models.15 Professional nutritionists can and should also use these media channels to reduce the lack of information and confusion that lay people have about nutrition and to disseminate correct facts about adequate eating habits, to promote health. These educational efforts can also be very useful when working with group or individual nutritional counseling.15,16,17,18,19 In Brazil, women buy $60\%$ of the popular magazines.20 Many of these weekly or monthly issues publish a large number of diet plans of unknown quality. Therefore, it is important to assess the content and adequacy of these diet plans from a nutritional perspective. The results from the present study will help to inform the public about the potential harm and benefits of diet plans published in popular magazines. ## OBJECTIVES To assess the quality of diet plans for weight loss published in Brazilian women’s magazines. ## METHOD This was a descriptive cross-sectional study conducted in the postgraduate program of the Federal University of São Paulo and approved by the Research Ethics Committee of the institution under the number 25223. The analysis unit was diet plans for weight loss presented in articles published in popular women’s magazines. The articles included in this study were obtained using a pre-specified sampling strategy. We selected the magazines with the highest circulation between January and June 2014 that were classified as addressing “feminine”, “behavior and beauty” and “quality of life and health” matters, according to the 2014 Brazilian media data registry.20,21 All of the issues of these magazines that advertised diets for weight loss on their covers were considered eligible for inclusion. After reading each potentially eligible full-text article, in printed versions of the journals, we excluded those that did not provide a full diet plan (i.e. that did not allow us to extract the nutritional content of the plan), those that only provided general nutritional recommendations and those that proposed diet plans lasting less than 7 days. We extracted data from all the eligible articles using a form specifically created for the study. Estimation of the total caloric values and nutritional composition of the diet plans of each article included was performed in duplicate by two independent investigators using the DIETPRÓ software, the database of the AVANUTRI software and the Brazilian Food Composition Table.22 We analyzed the adequacy of each plan suggested in the magazine articles by comparing the estimated nutritional macro and micronutrient content of the diet plan with international nutritional recommendations for women between 19 and 50 years of age. For carbohydrates, proteins, fats and saturated fatty acids, we compared the diet plan content according to the latest WHO recommendations.23 Briefly, these recommend that adults should get $55\%$ to $75\%$ of their total daily calories from carbohydrates, $15\%$ to $30\%$ from fats (less than $10\%$ from saturated fatty acids) and $10\%$ to $15\%$ from proteins. We classified diet plans with less than $55\%$ carbohydrates as being below the recommendations, those between $55\%$ and $75\%$ as adequate and those with more than $75\%$ carbohydrates as above the recommendations. The fat content was categorized as low when it represented less than $15\%$ of the total dietary energy, adequate when it was between $15\%$ and $30\%$ and high when it was more than $30\%$. Plans with saturated fatty acid values below $10\%$ were considered adequate and above this were considered high. Plans in which the protein content accounted for between $10\%$ and $15\%$ of the total dietary energy were classified as being adequate; those below and above these figures were categorized as being below and above the recommendations, respectively.23,24 We assessed the fiber content of each diet according to the latest dietary reference intake (DRI) values proposed by the Institute of Medicine (IOM).25 Diet plans with less than 25 g of fibers were categorized as having low fiber content and those with 25 g or more were considered adequate, since there is no upper reference level for fibers. We analyzed the micronutrient content (calcium, iron, vitamin C and sodium) of the diet plan according to the IOM recommendations for vitamins and minerals.26,27,28,29 For calcium, we used the micronutrient apparent adequacy criterion proposed by the DRI.30 The formula used was: Z = Y - EAR/√Vnec + (Vint/n), where Y = average intake of micronutrients obtained from food surveys, or in this case, the diet plans of women’s magazines; EAR = estimated average requirement, i.e. mean micronutrient needs according to age and sex; Vnec = variance of needs, which corresponds to $10\%$ of EAR = 0.1 x EAR; and Vint = intrapersonal variance, according to age and sex. Intrapersonal standard deviation (SD) was obtained.30 We analyzed the vitamin C and iron content, as suggested by the IOM subcommittee that published the DRI.30,31 Values below the estimated average requirement (EAR), which is 60 mg for vitamin C and 8.1 mg for iron, were considered to be lower than the recommendations. Values between the EAR and the recommended dietary allowances (RDA), which are 75 mg and 18 mg for vitamin C and iron, respectively, were considered to represent a risk of inadequacy. Values between the RDA and the tolerable upper intake level (UL), which is 2000 mg for vitamin C and 45 mg for iron, were considered appropriate and values above the UL were considered to be higher than the recommendations.30,31 For sodium, the adequate intake and UL values were used, since there is no EAR value for this mineral. Values between the adequate intake and UL were considered appropriate and values greater than the UL were considered to be higher than the recommendations.28 We also checked whether the article specified the total daily calorie content of the diet plan, and whether it recommended appropriate daily water intake and encouraged readers to practice physical activities along with the diet plan. We calculated the total daily calorie content of each plan and compared this with the values provided in the articles that presented this information. The total dietary energy of the diet plan was considered correct if it was between $90\%$ and $110\%$ of our calculations.18,19,32 Finally, we checked whether the article cited the name of the professional responsible for elaboration of the plan. The results are presented using descriptive statistics. ## RESULTS We identified a total of 53 different women’s magazines (23 weekly and 30 monthly titles) on the website of the National Association of Magazine Editors. Six different women’s magazines (five weekly and one monthly) were included in the study. The numbers of copies sold and types of the magazines were: 188,895 (A, weekly); 137,138 (B, weekly); 125,774 (C, weekly); 67,500 (D, weekly); 27,520 (E, weekly); and 209,772 (F, monthly). A total of 136 issues of these six magazines had diet plans on their covers and were selected for full-text reading; 41 were excluded for various reasons and 95 articles were included in the study (Figure 1). Figure 1.Flowchart of the sample acquisition: January-June 2014. The 95 articles included were published in the six different magazines as follows: 11 articles ($11.5\%$) in magazine A; 21 ($22.1\%$) in magazine B; 17 ($17.8\%$) in magazine C; 25 ($26.3\%$) in magazine D; 15 ($15.7\%$) in magazine E; and 6 ($6.3\%$) in magazine F. Quantitative analysis on the content of the 95 diet plans revealed that 79 plans ($83.1\%$) had carbohydrate levels and 93 ($97.8\%$) had protein levels below those recommended by the reference used. On the other hand, 63 ($66.3\%$) of the plans proposed adequate fat levels and 91 ($95.7\%$) of the plans had saturated fatty acid levels above the recommendations (Table 1). Table 1.Quantitative analysis on the adequacy of provision for carbohydrates, proteins, lipids and saturated fatty acids. Seventy-nine ($83.1\%$) of the plans had a fiber content lower than the recommendations, but 49 ($51.5\%$) and 85 ($89.4\%$) of the plans had adequate sodium and vitamin C levels, respectively. Sixty-seven ($70.5\%$) of the plans were categorized as having an iron level that represented a risk of inadequacy (Tables 2 and 3). Finally, 72 ($75.7\%$) of the plans had a $70\%$ to $98\%$ likelihood of inadequacy in relation to calcium. Table 2.Quantitative analysis on the adequacy of provision for fiber and sodium. Table 3.Quantitative analysis on the adequacy of provision for vitamin C and iron. Less than one third ($$n = 30$$; $31.5\%$) of the diet plans specified the total dietary energy expressed in calories. In 16 ($53.3\%$) of these plans, the total informed by the magazine was discordant with our calculations: in 9 plans ($56.2\%$), our calculation was more than $10\%$ higher than the number specified in the magazine; and in 7 ($43.75\%$), our calculation was more than $10\%$ lower than this number. Only 19 ($20\%$) of the articles provided information on the daily water intake. Approximately one third ($$n = 34$$; $35.7\%$) of the articles encouraged readers to increase their level of physical activity along with the diet for weight loss. Almost all ($$n = 92$$; $96.8\%$) of the articles stated the name of the professional responsible for elaborating the diet plan or who provided advice or comments about it. In most cases ($$n = 84$$; $88.4\%$), this professional was a nutritionist; in 6 articles ($6.3\%$), a physician was named, and in 2 ($2.1\%$), a phytotherapist was named. Three ($3.1\%$) of the diet plans did not present any health professional in charge. However, one of these ($33.3\%$) was based on the method for losing weight proposed by the French nutrologist Pierre Dukan and one ($33.3\%$) was based on the method of the American physician Ian Smith. ## DISCUSSION In view of the large number of women’s magazines that publish diet plans for weight loss, the results from this study raise concerns about the potential impacts on health from following these plans. Less than $20\%$ of the 95 plans proposed diets with adequate carbohydrate content, less than $5\%$ presented adequate saturated fatty acid content and less than $3\%$ proposed diets with appropriate protein intake, according to WHO standards.23,24 Simple exclusion of food sources of carbohydrates such as fruits, vegetables and grains from an individual’s usual diet will typically lead to a calorie deficit of approximately 500 calories per day, thus resulting in a loss of 0.45 to 0.9 kg per week. In turn, diet plans that reduce or restrict carbohydrate consumption lead to a loss of 2 to 3 kg in the first week.10 This extra loss is not due to a change in metabolism, which would lead to an increase in lipolysis, but to increased diuresis induced by the diet. Even by increasing the 24-hour energy expenditure from $2\%$ to $3\%$, this effect is responsible for only a small fraction of the weight loss.10,33 Because low carbohydrate diet plans restrict consumption of fruit and vegetables, they are notoriously deficient in micronutrients.33 A systematic review compared the effects of low carbohydrate/high protein (LC/HP) versus low fat/high carbohydrate (LF/HC) diet plans on weight loss. At six months, there were weight reduction of up to 4.02 kg in favor of the LC/HP plan ($P \leq 0.00001$); but after 12 months, this difference dropped to 1.05 kg ($P \leq 0.05$). Evidence from this systematic review shows that LC/HP diets are more effective than LF/HC diets for weight reduction at 6 months, but their effectiveness decreases by the 12th month.34 A prospective Japanese study involving two cohorts that were followed for 9 to 14 years tested the hypothesis that the intake of saturated fatty acids is inversely associated with the risk of stroke and directly associated with coronary heart disease. The investigators reported that there was a direct association between saturated fatty acid intake and myocardial infarction, mainly among men. On the other hand, they found an inverse association between saturated fatty acid intake and ischemia.35 Only $16.8\%$ of the 95 diet plans provided fiber content within the DRI recommendations.25 Fibers have important physiological effects, including prevention of intestinal diseases, treatment of obesity and reduction of serum lipid levels.11,36,37,38,39 From a meta-analysis on 67 randomized trials assessing the effect of dietary fiber on serum cholesterol levels, it was found that the consumption of 2-10 g of soluble fibers per day was associated with a significant reduction in total cholesterol levels.38 Almost $70\%$ of the diet plans did not specify the total daily quantity of calories, and in over half of those that provided this information, it was incorrect. All diet plans that propose reductions in total daily calorie intake will lead to weight loss. In the absence of physical activity, a plan that provides between 1400 to 1500 calories per day, regardless of the macronutrient percentage, will result in weight loss.40 A previous Brazilian study published in 2004 also assessed diet plans for weight loss that were featured in non-scientific publications. The authors reported that $80\%$ of the diet plans did not provide any information on daily intake of water and less than $25\%$ had appropriate macronutrient distributions. In fact, all of the plans had inadequate protein content, with one third of them recommending that over $15\%$ of the total calorie intake should come from that macronutrient. Moreover, $86\%$ of those plans provided inappropriate calcium content, $92\%$ gave rise to inadequate vitamin E content and $97\%$ did not provide adequate iron content.12 In our study, carried out ten years later, $20\%$ of the diet plans did not provide any information on daily intake of water and less than $4\%$ had appropriate macronutrient distributions. In relation to protein, $97.8\%$ of them recommended that over $15\%$ of the total calorie intake should come from that macronutrient. The vitamin E content of the diet plans was not verified in this study, but $75.7\%$ of the plans presented a probability of being inadequate in relation to calcium and $70.5\%$ did not provide adequate iron content. Less than $13\%$ of the diet plans provided adequate calcium content. Calcium is an essential mineral at different stages of women’s lives, especially with regard to maintaining bone health.26,41 Consumption of high protein diets, especially for long periods, may cause increased urinary calcium loss, thus increasing the risks of osteoporosis.42,43 A prospective Norwegian study reported that there was a significantly higher risk of hip fracture among women with high intake of non-dairy animal protein and low calcium intake (RR 1.96; $95\%$ CI 1.09-3.56).44 The lack of recommendations about water intake in most diet plans for weight loss that are available in popular women’s magazines is worrying. Under normal environmental conditions and energy expenditure, an average adult will need approximately 1 ml of water per calorie. The need will be higher among individuals who have an exercise routine and among those who are on a protein-rich diet. Deficiency in water intake manifests quickly and a change in the body’s water content as small as $1\%$ promptly causes symptoms of dehydration.45 Almost two thirds of the plans included in our study did not encourage physical activity along with the diet, despite the well-known fact that exercise is an important part of any weight loss and maintenance program.44,46,47 Although our study did not evaluate the readers’ actual use of the diets or the possible impact of the dietary plans on their health, it clearly shows that most diets published in popular Brazilian women’s magazines are inadequate. Our results are important for informing the general public about the risks of blindly following the diets published in these magazines and to alert journalists and editors of these publications about the need to check the scientific accuracy and safety of these diet plans before disseminating them to a wide lay audience. ## CONCLUSION All of the 95 diet plans for weight loss published during a six-month period in popular Brazilian women’s magazines failed to follow one or more of the international dietary recommendations. In this light, the results obtained here emphasize that publication of diet plans for weight loss in non-scientific magazines needs to be based on international dietary recommendations and evidence from studies of good methodological quality. Following these diets may cause deleterious effects to health. ## References 1. Ng M, Fleming T, Robinson M. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet (2014.0) 384 766-781. PMID: 24880830 2. 2 Brasil. Ministério da Saúde. Secretaria de Vigilância em Saúde. Departamento de Vigilância de Doenças e Agravos não Transmissíveis e Promoção da Saúde Vigitel Brasil 2013: vigilância de fatores de risco e proteção para doenças crônicas por inquérito telefônico/Ministério da Saúde, Secretaria de Vigilância em Saúde, Departamento de Vigilância de Doenças e Agravos não Transmissíveis e Promoção da Saúde Brasília Ministério da Saúde 2014 http://bvsms.saude.gov.br/bvs/publicacoes/vigitel_brasil_2013.pdf Accessed in 2017 (Apr 5). Vigitel Brasil 2013: vigilância de fatores de risco e proteção para doenças crônicas por inquérito telefônico/Ministério da Saúde, Secretaria de Vigilância em Saúde, Departamento de Vigilância de Doenças e Agravos não Transmissíveis e Promoção da Saúde (2014.0) 3. McManus K, Antinoro L, Sacks F. A randomized controlled trial of a moderate-fat, low-energy diet compared with a low fat, low-energy diet for weight loss in overweight adults. Int J Obes Relat Metab Disord (2001.0) 25 1503-1511. PMID: 11673773 4. Akiyama T, Tachibana I, Shirohara H, Watanabe N, Otsuki M. High-fat hypercaloric diet induces obesity, glucose intolerance and hyperlipidemia in normal adult male Wistar rat. Diabetes Res Clin Pract (1996.0) 31 27-35. PMID: 8792099 5. Alcântra MLB. O corpo do brasileiro: estudos de estética e beleza. Rev Antropol (2001.0) 44 231-234 6. Duke L. Get real!: Cultural relevance and resistance to the mediated feminine ideal. Psychology and Marketing (2002.0) 19 211-233 7. Vasques F, Martins FC, Azevedo AP. Aspectos psiquiátricos no tratamento da obesidade [Psychiatric aspects in the treatment of obesity]. Rev Psiquiatr Clín (2004.0) 31 195-198 8. Gardner CD, Kiazand A, Alhassan S. Comparison of the Atkins, Zone, Ornish, and LEARN diets for change in weight and related risk factors among overweight premenopausal women. the A TO Z Weight Loss Study: a randomized trial. JAMA (2007.0) 297 969-977. PMID: 17341711 9. 9 World Health Organization Programmes and projects. Obesity and overweight. Global Strategy on Diet, Physical Activity and Health Geneva World Health Organization 2012. Programmes and projects. Obesity and overweight. Global Strategy on Diet, Physical Activity and Health (2012.0) 10. Denke MA. Metabolic effects of high-protein, low-carbohydrate diets. Am J Cardiol (2001.0) 88 59-61. PMID: 11423059 11. Ma Y, Pagoto SL, Griffith JA. A dietary quality comparison of popular weight-loss plans. J Am Diet Assoc (2007.0) 107 1786-1791. PMID: 17904938 12. Amancio OMS, Chaud DMA. Weight loss diets advertised in non-scientific publications. Cad Saúde Pública (2004.0) 20 1219-1222. PMID: 15486664 13. Turney J. Public understanding of science. Lancet (1996.0) 347 1087-1090. PMID: 8602062 14. Grunert KG, Wills JM. A review of European research on consumer response to nutrition information on food labels. Journal of Public Health (2007.0) 15 385-399 15. Wansink B. Position of the American Dietetic Association: food and nutrition misinformation. J Am Diet Assoc (2006.0) 106 601-607. PMID: 16639825 16. Buttriss JL. Food and nutrition: attitudes, beliefs, and knowledge in the United Kingdom. Am J Clin Nutr (1997.0) 65 l985S-l995S 17. Redmond EC, Griffith CJ. Consumer perceptions of food safety education sources: implications for effective strategy development. British Food Journal (2005.0) 107 467-483 18. 18 Brasil. Presidência da República. Casa Civil. Subchefia para Assuntos Jurídicos Lei no 8.234, de 17 de setembro de 1991. Regulamenta a profissão de Nutricionista e determina outras providências http://www.planalto.gov.br/ccivil_03/leis/1989_1994/L8234.htm Accessed in 2017 (Apr 5). Lei n 19. 19 Conselho Federal de Nutricionistas Resolução CFN no 380/2005. Dispõe sobre a definição das áreas de atuação do nutricionista suas atribuições, estabelece parâmetros numéricos de referência, por área de atuação, e dá outras providências http://www.cfn.org.br/novosite/pdf/res/2005/res380.pdf Accessed in 2017 (Apr 5). Resolução CFN n 20. 20 Grupo de Mídia São Paulo Mídia Dados Brasil 2014 http://sunflower2.digitalpages.com.br/html/reader/119/38924 Accessed in 2017 (Apr 5). Mídia Dados Brasil 2014 21. 21 National Association of Magazine Editors. Associação Nacional de Editores de Revistas 2014 Dados de Circulação http://aner.org.br/?page_id=50 Accessed in 2017 (Jan 18). Dados de Circulação (2014.0) 22. 22 Taco - Tabela Brasileira de Composição de Alimentos Campinas NEPA-UNICAMP 2011 http://www.unicamp.br/nepa/taco/home.php?ativo=home Accessed in 2017 (Apr 5). Taco - Tabela Brasileira de Composição de Alimentos (2011.0) 23. 23 World Health Organization Interim Summary of Conclusions and Dietary Recommendations on Total Fat & Fatty Acids Geneva World Health Organization 2008 http://www.fao.org/ag/agn/nutrition/docs/Fats%20and%20Fatty%20Acids%20Summary.pdf Accessed in 2017 (Apr 5). Interim Summary of Conclusions and Dietary Recommendations on Total Fat & Fatty Acids (2008.0) 24. 24 World Health Organization Diet, nutrition and the prevention of chronic diseases. WHO Technical Report Series 916 Geneva World Health Organization 2003 http://apps.who.int/iris/bitstream/10665/42665/1/WHO_TRS_916.pdf Accessed in 2017 (Apr 5). Diet, nutrition and the prevention of chronic diseases. WHO Technical Report Series 916 (2003.0) 25. 25 Institute of Medicine. National Academies Press Dietary Reference Intakes for energy, carbohydrate, fiber, fat, fatty acids, cholesterol, protein, and amino acids Washington National Academy Press 2002 https://www.nap.edu/read/10490/chapter/1#ii Accessed in 2017 (Apr 5). Dietary Reference Intakes for energy, carbohydrate, fiber, fat, fatty acids, cholesterol, protein, and amino acids (2002.0) 26. 26 Institute of Medicine. National Academies Press Dietary Reference Intakes for calcium, phosphorus, magnesium, vitamin D and fluoride Washington National Academy Press 1997 https://www.nap.edu/read/5776/chapter/1 Accessed in 2017 (Apr 5). Dietary Reference Intakes for calcium, phosphorus, magnesium, vitamin D and fluoride (1997.0) 27. 27 Institute of Medicine. National Academies Press Dietary Reference Intakes for vitamin A, vitamin K, arsenic, boron, chromium, copper, iodine, iron, manganese, molybdenum, nickel, silicon, vanadium, and zinc Washington National Academy Press 2001 https://www.nap.edu/read/10026/chapter/1 Accessed in 2017 (Apr 5). Dietary Reference Intakes for vitamin A, vitamin K, arsenic, boron, chromium, copper, iodine, iron, manganese, molybdenum, nickel, silicon, vanadium, and zinc (2001.0) 28. 28 Institute of Medicine. National Academies Press Dietary Reference Intakes for water, potassium, sodium, chloride, and sulfate Washington National Academy Press 2005 https://www.nap.edu/read/10925/chapter/1 Accessed in 2017 (Apr 5). Dietary Reference Intakes for water, potassium, sodium, chloride, and sulfate (2005.0) 29. 29 Institute of Medicine. National Academies Press Dietary Reference Intakes for vitamin C, vitamin E, selenium, and carotenoids Washington National Academy Press 2000 https://www.nap.edu/read/9810/chapter/1 Accessed in 2017 (Apr 5). Dietary Reference Intakes for vitamin C, vitamin E, selenium, and carotenoids (2000.0) 30. Marchioni DML, Slater B, Fisberg RM. Aplicação das Dietary Reference Intakes na avaliação da ingestão de nutrientes para indivíduos. Rev Nutr (2004.0) 17 207-216 31. 31 Institute of Medicine (US) Subcommittee on Interpretation and Uses of Dietary Reference Intakes; Institute of Medicine (US) Standing Committee on the Scientific Evaluation of Dietary Reference Intakes DRI Dietary Reference Intakes: Applications in Dietary Assessment Washington National Academies Press 2000. DRI Dietary Reference Intakes: Applications in Dietary Assessment (2000.0) 32. 32 Brasil. Ministério da Saúde. Secretaria de Atenção à Saúde. Departamento de Atenção Básica Guia alimentar para a população brasileira: promovendo a alimentação saudável Brasília Ministério da Saúde 2008 http://bvsms.saude.gov.br/bvs/publicacoes/guia_alimentar_populacao_brasileira_2008.pdf Accessed in 2017 (Apr 5). Guia alimentar para a população brasileira: promovendo a alimentação saudável (2008.0) 33. Astrup A, Meinert Larsen T, Harper A. Atkins and other low-carbohydrate diets: hoax or an effective tool for weight loss?. Lancet (2004.0) 364 897-899. PMID: 15351198 34. Hession M, Rolland C, Kulkarni U, Wise A, Broom J. Systematic review of randomized controlled trials of low-carbohydrate vs. low-fat/low-calorie diets in the management of obesity and its comorbidities. Obes Rev (2009.0) 10 36-50. PMID: 18700873 35. Yamagishi K, Iso H, Kokubo Y. Dietary intake of saturated fatty acids and incident stroke and coronary heart disease in Japanese communities: the JPHC Study. Eur Heart J (2013.0) 34 1225-1232. PMID: 23404536 36. Brown L, Rosner B, Willett WW, Sacks FM. Cholesterol-lowering effects of dietary fiber: a meta-analysis. Am J Clin Nutr (1999.0) 69 30-42. PMID: 9925120 37. Weickert MO, Pfeiffer AF. Metabolic effects of dietary fiber consumption and prevention of diabetes. J Nutr (2008.0) 138 439-442. PMID: 18287346 38. Erkkilä AT, Lichtenstein AH. Fiber and cardiovascular disease risk: how strong is the evidence?. J Cardiovasc Nurs (2006.0) 21 3-8. PMID: 16407729 39. Ma Y, Griffith JA, Chasan-Taber L. Association between dietary fiber and serum C-reactive protein. Am J Clin Nutr (2006.0) 83 760-766. PMID: 16600925 40. Freedman MR, King J, Kennedy E. Popular diets: a scientific review. Obes Res (2001.0) 9 Suppl 1 1S-40S. PMID: 11374180 41. 41 Joint FAO/WHO Expert consultation on human vitamin and mineral requirements Bangkok 1998 http://apps.who.int/iris/bitstream/10665/42716/1/9241546123.pdf Accessed in 2017 (Apr 5). Expert consultation on human vitamin and mineral requirements (1998.0) 42. Lemon PW. Effects of exercise on dietary protein requirements. Int J Sport Nutr (1998.0) 8 426-447. PMID: 9841962 43. Ferrari P, Rinaldi S, Jenab M. Dietary fiber intake and risk of hormonal receptor-defined breast cancer in the European Prospective Investigation into Cancer and Nutrition study. Am J Clin Nutr (2013.0) 97 344-353. PMID: 23269820 44. Meyer HE, Pedersen JI, Løken EB, Tverdal A. Dietary factors and the incidence of hip fracture in middle-aged Norwegians. A prospective study. Am J Epidemiol (1997.0) 145 117-123. PMID: 9006308 45. 45 National Research Council Recommended Dietary Allowances 10 Washington National Academy Press 1989. Recommended Dietary Allowances (1989.0) 46. Greenberg I, Stampfer MJ, Schwarzfuchs D, Shai I. Adherence and success in long-term weight loss diets: the dietary intervention randomized controlled trial (DIRECT). J Am Coll Nutr (2009.0) 28 159-168. PMID: 19828901 47. Catenacci VA, Wyatt HR. The role of physical activity in producing and maintaining weight loss. Nat Clin Pract Endocrinol Metab (2007.0) 3 518-529. PMID: 17581621
--- title: 'Smoking among adolescents is associated with their own characteristics and with parental smoking: cross-sectional study' authors: - Rafaela Campos Cuissi de Andrade - Aline Duarte Ferreira - Dionei Ramos - Ercy Mara Cipulo Ramos - Catarina Covolo Scarabottolo - Bruna Thamyres Ciccotti Saraiva - Luis Alberto Gobbo - Diego Giulliano Destro Christofaro journal: São Paulo Medical Journal year: 2017 pmcid: PMC10016012 doi: 10.1590/1516-3180.2017.0154220717 license: CC BY 4.0 --- # Smoking among adolescents is associated with their own characteristics and with parental smoking: cross-sectional study ## ABSTRACT ### BACKGROUND: This study aimed to analyze the association between smoking during adolescence and the characteristics of smoking and alcohol consumption among their parents. ### DESIGN AND SETTING: Cross-sectional study in Londrina (PR), Brazil. ### METHODS: The subjects comprised 1,231 adolescents aged 14-17 years. The adolescents and their parents answered a self-report questionnaire that asked for sociodemographic information and data on smoking and alcohol consumption. Multiple logistic regression models were used to analyze associations between smoking among adolescents and their characteristics (age, sex, period of the day for attending school, alcohol consumption and socioeconomic level) and their parents’ characteristics (smoking, alcohol consumption, age and education level), adjusted according to the adolescents’ characteristics (sex, age and socioeconomic level). ### RESULTS: The prevalence of smoking among adolescents was $3.4\%$ ($95\%$ confidence interval, CI: 2.4-4.4). Adolescents whose mothers or fathers were smokers were 2.0 and 2.5 times more likely to be smokers, respectively. The prevalence of smoking among adolescents with a smoking mother was $7.1\%$ ($95\%$ CI: 2.6-10.7) and a smoking father, $5.4\%$ ($95\%$ CI: 1.6-8.5). There were significant associations between smoking adolescents and age [$5.2\%$ ($95\%$ CI: 3.3-6.6)], studying at night [$9.6\%$ ($95\%$ CI: 4.0-15.5)] and alcohol consumption [$69.0\%$ ($95\%$ CI: 55.0-83.0)]. It was observed that the number of alcoholic beverage doses consumed was higher among smoking adolescents ($$P \leq 0.001$$). ### CONCLUSION: Adolescent smoking was associated with smoking by their parents, regardless of the gender of the parents or adolescents. Age, alcohol consumption and studying at night are characteristics of adolescents that can contribute towards smoking. ## INTRODUCTION Smoking is considered to be a behavior that puts health at risk.1 Several studies have demonstrated a strong relationship between smoking and various types of diseases in the adult population, such as carotid calcification2 and other cardiovascular problems, like stroke.3 However, this type of behavior has been detected not only among adults, but also in young populations. In a study conducted in Saudi Arabia, Al-Zalabani and Kasim4 observed that the prevalence of smoking was around $15\%$ among the young people who they evaluated. Adolescent smoking is increasing in poorer countries. Smoking in adulthood may start during adolescence, which demonstrates the importance of studies addressing this issue.5 Tavares et al.6 and Barreto et al.7 highlighted that adolescence is a period of great exposure and vulnerability to consumption of substances such as tobacco and alcohol, with frequent experimentation by adolescents. Consequently, determining the factors that could cause this type of behavior among adolescents is important. Another aspect that has been investigated is whether adolescents’ household environment might contribute to such behavior. In a study on adolescents aged 13-18 years, Vázquez-Rodríguez et al.8 reported that parental smoking was associated with smoking among their children. Similar results were observed in some studies in which adolescent smoking was more prevalent among those whose parents were smokers.9,10 Tondowski et al.9 showed that approximately $45\%$ of adolescents who reported frequent tobacco use had fathers or mothers who smoked. In addition, it has been shown that smoking during adolescence may be linked both to use of illicit drugs such as marijuana and to use of licit drugs such as alcohol.11,12,13,14,15 The majority of previous studies have only examined parental smoking as a risk factor. However, other lifestyle variables such as alcohol consumption and sociodemographic characteristics such as the parents’ ages and educational levels, also need to be considered. One of the hypotheses is that the parents’ characteristics other than smoking may also be associated with smoking among adolescents. Moreover, it needs to be emphasized that the characteristics of the adolescents themselves should also be considered in order to eliminate possible confounding factors, since late adolescence16 and being male17 tend to be more associated with smoking, and socioeconomic level may also be associated, depending on the characteristics of each country.18 Studies that investigate lifestyle habits between parents and children can contribute towards health promotion actions, if these relationships are observed in the family environment. Therefore, the aim of the present study was to analyze the association between smoking during adolescence and the lifestyle characteristics (smoking and alcohol consumption) of parents or family members who live with adolescents. ## Sample The sample of this study formed part of a larger study that looked at risk factors for health among adolescents at public schools in the city of Londrina (PR), Brazil, and among their parents. To contact the adolescents, the Londrina Department of Education was first contacted in order to explain the objectives of the study. Subsequently, the Department identified the six largest public schools in the central region, which receive adolescents from different areas of the city (north, south, east, west and central areas). Subsequently, the researchers contacted the principals of the schools that were invited to participate in the study to explain the objectives of the study. After authorization from the schools’ directors, contact was made with all classes of students aged 14-17 years in these schools. To calculate the sample, the prevalence of smoking among adolescents was taken to be $15\%$,4 and a tolerable error of $3\%$ and power of $80\%$ were used. Since the sample was selected through clusters, a design correction of 1.5 was used. To anticipate possible losses from the sample, $10\%$ was added to the sample calculation. Thus, the minimum sample required was 870 adolescents. In the end, the study included 1231 adolescents (716 girls and 515 boys) aged 14-17 years. Adolescents and their parents or family members who agreed to participate in the study signed a free and informed written consent form. This study was approved by the Research Ethics Committee of the institution responsible for this study (procedural number: 0.181.0.268.000-10; register number: 367.801). The adolescents took the consent form home for their parents to sign and thus authorize the adolescent to participate in the study. Along with this consent form, they also took the parents’ questionnaire with them, so that their parents could answer this instrument at home. The parents’ questionnaire contained questions about their lifestyle habits (among them smoking and alcohol consumption) and sociodemographic variables (sex, age and schooling level). In total, 1,202 mothers and 871 fathers answered the questionnaire. Subsequently, the adolescents were evaluated at school. ## Smoking and alcohol Smoking status was ascertained through analysis of participants’ smoking behavior.19 If individuals replied that they had smoked cigarettes within the previous 30 days, they were considered to be smokers. The number of cigarettes that these individuals consumed in a typical week was also established. Alcohol consumption was obtained through questions based on the questionnaire of the Brazilian Center for Psychotropic Drugs (CEBRID),20 which assesses the frequency and quantity of alcoholic drinks consumed. Adolescents and parents or family members who reported consumption of more than 1-2 doses (each dose corresponded to 250 ml of beer or 40 ml of distilled beverages in this study) on more than 1-2 days a week were classified as high consumers. The cutoff points used in this study were adapted from Moreira et al.21 These instruments demonstrate good reproducibility values: kappa = 0.81 for smoking and kappa = 0.83 for alcohol consumption. ## Anthropometric variables The adolescents were measured wearing light clothing and no shoes. Weight was measured using a portable scale (Plenna; precision of 0.100 kg) with a capacity of 150 kg. Height was measured using a portable stadiometer (Sanny; precision of 0.1 cm) with a scale in centimeters. The anthropometric characteristics of the adolescents were evaluated by two previously trained evaluators. The procedures were applied in accordance with the recommendations of Gordon et al.22 To assess the participants’ nutritional status, the body mass index (BMI) was calculated as the ratio between the weight and height squared. The adolescents’ nutritional status was classified in accordance with the values proposed by Cole et al.23 Overweight among the parents was determined based on the cutoff points of the World Health Organization, and adults with BMI greater than or equal to 25 kg/m2 were classified as overweight.24 ## Sociodemographic variables The parents’ educational level was evaluated as the number of years of study reported over the course of their lives. Parental schooling was divided into terciles, such that lower education level was considered to be up to 8 years of study; medium education level, from 8 to 12 years; and higher education level, more than 12 years. Parental age was determined as the difference between the date of data collection and birth date. Subsequently, age was divided into terciles. To define the families’ economic class, the 2011 Brazilian economic classification criteria of the Brazilian Market Research Association (ABEP) were used.25 Householders’ education level and the presence and quantity of certain rooms, assets and domestic employees in the homes analyzed were considered (color TV, VCR or DVD player, radio, number of bathrooms, car, washing machine, housemaids, refrigerator and freezer). At the end of this instrument, a scoring system is provided in which the individual is classified according to economic strata, such that higher scores represent higher economic strata. ## Statistical analysis The data characterizing the sample were presented as means and standard deviations stratified according to smoking status (smoker or nonsmoker). Analysis on the association between the dependent variable (smoking adolescents) and the independent variables was performed using the chi-square test. Subsequently, two multivariate models were created and were analyzed through binary logistic regression. The association between smoking adolescents and their own characteristics (sex, age, period of the day for attending school, day or night, socioeconomic status and alcohol consumption) was analyzed. The association between smoking adolescents and the characteristics of their mothers and fathers was analyzed. In the first model, unadjusted smoking among adolescents was analyzed in relation to their mothers and fathers’ smoking, alcohol consumption, age and education level. In the second model, which was adjusted according to the adolescents’ characteristics, the sociodemographic variables of the adolescents that might be potential confounders were considered (sex, age and socioeconomic status). Although only the adolescents’ ages presented P values lower than 0.200, it was decided that, in analyzing the association with smoking, the adolescents’ sex and socioeconomic status would be inserted as adjustment variables. This was done to ascertain whether the possible associations between smoking and the variables analyzed would be independent of these confounding factors. The significance level used for all analyses was P ≤ $5\%$. The confidence interval (CI) used was $95\%$. The analyses were performed using the Statistical Package for the Social Sciences (SPSS) software, version 15.0. ## RESULTS The prevalence of smoking in the sample of this study was $3.4\%$ ($95\%$ CI: 2.4-4.4), which was equivalent to 42 adolescents. The average number of cigarettes smoked by the adolescents interviewed was 0.29 per month with no difference between boys and girls ($$P \leq 0.351$$). The prevalence of smoking among the mothers was $12.1\%$ ($95\%$ CI: 10.5-14.2) and among the fathers, $17.9\%$ ($95\%$ CI: 15.3-20.4). Smoking mothers consumed more alcohol than did mothers who did not smoke. Fathers who smoked presented lower weight, lower BMI and higher alcohol consumption than did fathers who did not smoke. Mothers and fathers with fewer years of schooling were more likely to be smokers. The prevalence of smoking was higher among fathers and mothers of medium socioeconomic status. Table 1 presents information regarding sample characterization. It can be seen that the highest average of alcoholic beverages consumed in doses were higher among smoking adolescents. Table 1.Characteristics of the sample according to smoking statusSD = standard deviation. Table 2 shows the significant associations between smoking among adolescents and later adolescence, studying at night and alcohol consumption. Older adolescents (16-17 years) presented higher levels of smoking behavior ($5.2\%$; $95\%$ CI: 3.45-6.82) than younger adolescents ($1.6\%$; $95\%$ CI: 0.56-2.61) ($$P \leq 0.002$$). The adolescents who studied in the evenings presented higher prevalence of smoking ($9.6\%$; $95\%$ CI: 4.0-15.5) than those who studied during the day ($3.2\%$; $95\%$ CI: 1.94-3.91). Smoking adolescents presented higher frequency of alcohol consumption: among the 42 adolescents who smoked, 29 ($69.0\%$; $95\%$ CI: 55.0-83.0) consumed alcohol and $7.0\%$ of them ($95\%$ CI: 0.65-14.95) consumed alcohol with a frequency of four times a week. Table 2.Association between adolescents’ smoking habits and their characteristics Table 3 shows the associations between adolescents who smoked and mothers or female guardians who smoked. Adolescents whose mothers were smokers were twice as likely to have this habit. The prevalence of smoking among adolescents with smoking mothers was $7.1\%$ ($95\%$ CI: 2.6-10.7), compared with $2.3\%$ ($95\%$ CI: 1.85-3.86) among adolescents with non-smoking mothers. In both the raw and adjusted analyses on the variables relating to the adolescents, associations between smoking adolescents and alcohol consumption could be seen. There were no significant differences in smoking levels among adolescents between those with older mothers and those with younger mothers, or between those with mothers with lower education levels and those with mothers with higher education levels. Table 3.Association between smoking among adolescents and characteristics of their mothersAdjusted according to the adolescents’ sex, socioeconomic level and age. The prevalence of smoking among adolescents with smoking fathers was $5.4\%$ ($95\%$ CI: 1.6-8.5). Smoking among fathers was also associated with smoking among adolescents: teens whose fathers smoked were 2.5 times more likely to be smokers (Table 4). Table 4.Association between smoking among adolescents and characteristics of their fathersAdjusted according to the adolescents’ sex, socioeconomic level and age. Table 5 presents information on the relationship between smoking among adolescents and the smoking habits of both of their parents. There were no associations between adolescent smoking and both parents smoking. Table 5.Association between smoking among adolescents and smoking among parentsAdjusted according to the adolescents’ sex, socioeconomic level and age. ## DISCUSSION The prevalence of smoking adolescents in this study can be considered low ($3.4\%$) in comparison with other studies.4,26 In a* recent study, Figueiredo et al.16 observed that the prevalence of smoking in a sample of adolescents aged from 12 to 17 years was $5.7\%$, considering several Brazilian cities. Their findings were similar to those of the present study and their prevalence can also be considered low. One reason for this low prevalence appears to be related to restrictions on tobacco advertising on the television and to laws prohibiting tobacco use in public places, along with increased prices for cigarettes and increased activity of smoking cessation programs, as shown by Levy et al.27 There were no significant differences in the nutritional status of adolescent smokers and nonsmokers in this study. This same relationship was observed for the mothers, but smoking fathers presented lower weight and prevalence of overweight than did nonsmoking fathers. Adolescents are at an early stage of life, at which they have probably not yet established a pattern for nutritional status or smoking habits. Considering the difference in the nutritional status among their fathers, one of the reasons for this that can be considered is nicotine levels, which cause several changes to appetite and metabolic rate, thus giving rise to differences between smokers and nonsmokers.28 For both fathers and mothers, those with higher average schooling levels presented lower tobacco consumption than did parents with lower schooling levels, possibly because parents with higher education levels have more knowledge about the harm that cigarette smoking can cause. This study found that adolescents whose mothers or fathers smoked were about 2.0 and 2.5 times as likely, respectively, to have the same kind of behavior, even after various adjustments for potential confounders. It was observed that the habit of smoking among parents was associated with their children’s habits, independent of the abovementioned variables. In a study on young Canadians, O’Loughlin et al.29 also observed that parental smoking was associated with the onset of smoking during their children’s adolescence, regardless of parental schooling levels. Similar relationships have been observed in other studies.9,10 The fact is that teens tend to replicate their parents’ habits. In a recent meta-analysis, Laird et al.30 observed that adolescents with physically active parents were more likely to be physically active. However, this replication of habits does not seem to occur for healthy habits alone, and a similar relationship regarding smoking habits is observed between adolescents and their parents. One of the hypotheses for this is that these young people may have felt more freedom to experiment with smoking because of the example seen in their homes. Having a father and/or mother who smokes seems to represent a permissive attitude for adolescents, based on their parents’ behavior, thus producing an image that smoking is acceptable and possibly contributing towards a process of initiation of smoking. Nonetheless, in our sample, contrary to expectations, there was no association between smoking among adolescents and both parents being smokers. One of the possible reasons for this finding is that the prevalence of occurrences of both parents smoking at home was low: only $4.4\%$. An association was observed between mothers who had the behavior of consuming alcohol and smoking among adolescents. One of the possible reasons for this is the strong relationship between smoking and alcohol. Elicker et al. showed that $39.2\%$ of the adolescents in Porto Velho (RO), Brazil, consumed alcohol for the first time at home.31 This may be one of the reasons, since alcohol consumption at parties or family meetings would not be characterized as a risk factor for health, but as a normal attitude that could be related to smoking. Perhaps parents who are permissive regarding alcohol use may also be permissive regarding to smoking among young people. Among the characteristics of adolescent smoking, there were associations with age (being older), alcohol consumption and the period of the day in which adolescents attended school (night). Khuder et al.32 found that older adolescents were about six times more likely to be smokers than younger adolescents. One reason for this relationship is the transformation that adolescents experience during this stage of life. This is a period in which social relationships are important, and this could contribute towards starting some types of behavior such as smoking, with the aim of achieving acceptance in social groups. Several studies have reported the strong influence that friends have on smoking habits among adolescents.10,33 Regarding alcohol consumption, those who consumed alcoholic beverages were six times more likely to be smokers. Several studies have demonstrated significant relationships between smoking and alcohol consumption among adolescents.13,14,15 Among the substances found in large quantities in cigarettes, nicotine acts in many areas of the brain. It has been hypothesized that neuronal nicotinic acetylcholine receptors act in a specific brain area that also causes higher propensity towards alcohol use.34 Additionally, adolescents who went to school in the evenings were about three times more likely to have a smoking habit than their peers who attended school during the day. This corroborated the findings of Farias Junior et al.35 who observed that young people who attended classes in the evenings presented a greater chance of being smokers. This relationship was also observed among almost 3,000 adolescents in northern Brazil.36 Among the reasons that could explain this relationship, the first is that adolescents who study in the evenings have a higher average age than the adolescents who study during the day. Another characteristic of the adolescents attending school in the evenings is that they tend to work during the day, which aids independence and brings the possibility of buying cigarettes with their own income. The association between adolescents working and smoking has also been reported in another study.37 A further factor to be considered as a hypothesis, but which was not analyzed in this study, is that at night, several bars and nightclubs, which are often close to where schools are located, are open. This may contribute towards this type of behavior among young people who attend these places. The practical application for the present study is that it serves as a reminder to different healthcare agencies regarding the importance of organizing prevention strategies among families, in order to avoid problems caused by smoking in the future. In this regard, the recent findings of West et al.38 are noteworthy. Through a cohort study conducted over a period of more than twenty years, these authors found that children exposed to parental smoking had higher odds of developing carotid atherosclerotic plaque in adulthood. The limitations of the present study were, firstly, that the outcome was assessed using a questionnaire, which was self-administered and may have underestimated the prevalence of smoking, since some adolescents may have omitted the fact that they were smokers. In addition, this was an epidemiological cross-sectional study and it was not possible to quantify serum nicotine levels to confirm the presence of the smoking habit among these adolescents and thus to preclude the limitation of potential underreporting of this habit. Another factor to be mentioned is that the sample was not representative of all schools in the city in which the study was conducted. However, the sample was selected from the six largest schools in the central region of the city of Londrina, and these schools receive students from different areas of the city, with large numbers of students, which made the sample more representative. The strong aspects of this study are its large sample size, and all of the adjustments made in the analysis on the data. It is worth noting that through stratification of the parents according to sex, it became possible to observe the relationships between both the fathers’ and the mothers’ smoking habits and those of the adolescents. ## CONCLUSION The smoking habit among adolescents was associated both with parental and maternal smoking, regardless of the gender of the parents or adolescents. Factors such as age, alcohol consumption and attending school at night were characteristics among these adolescents that may have contributed towards smoking. Health promotion actions need to focus on the family unit and not on strategies that are isolated from each other. ## References 1. Lotufo PA. Smoking and cancer: Brazil and the Global Burden of Disease initiative. Sao Paulo Med J (2015.0) 133 385-387. PMID: 26648425 2. Brito ACR, Nascimento HAR, Freitas DQ. Prevalência de imagens sugestivas de calcificações da artéria carótida em radiografias panorâmicas e sua relação com fatores predisponentes [Prevalence of suggestive images of carotid artery calcifications on panoramic radiographs and its relationship with predisposing factors]. Ciên Saúde Coletiva (2016.0) 21 2201-2208 3. Xu T, Bu X, Li H. Smoking, heart rate, and ischemic stroke: a population-based prospective cohort study among Inner Mongolians in China. Stroke (2013.0) 44 2457-2461. PMID: 23881954 4. Al-Zalabani A, Kasim K. Prevalence and predictors of adolescents’ cigarette smoking in Madinah, Saudi Arabia: a school-based cross-sectional study. BMC Public Health (2015.0) 15 17-17. PMID: 25604704 5. Sims TH. From the American Academy of Pediatrics: Technical report--Tobacco as a substance of abuse. Pediatrics (2009.0) 124 e1045-e1053. PMID: 19841120 6. Tavares BF, Béria JU, Lima MS. Fatores associados ao uso de drogas entre adolescentes escolares [Factors associated with drug use among adolescent students in southern Brazil]. Rev Saúde Pública (2004.0) 38 787-796. PMID: 15608896 7. Barreto SM, Crespo C, Giatti L. Exposição ao tabagismo entre escolares no Brasil [Smoking exposure among school children in Brazil]. Ciên Saúde Coletiva (2010.0) 15 3027-3034 8. Vázquez-Rodríguez CF, Vázquez-Nava F, Vázquez-Rodríguez EM. Smoking in non-student Mexican adolescents with asthma: relation with family structure, educational level, parental approval of smoking, parents who smoke, and smoking friends. Arch Bronconeumol (2012.0) 48 37-42. PMID: 22113156 9. Tondowski CS, Bedendo A, Opaleye ES. Estilos parentais como fator de proteção ao consumo de tabaco entre adolescentes brasileiros [Parenting styles as a tobacco-use protective factor among Brazilian adolescentes]. Cad Saúde Pública (2015.0) 31 2514-2522. PMID: 26872228 10. Abreu MNS, Souza CF, Caiaffa WT. Tabagismo entre adolescentes e adultos jovens de Belo Horizonte, Minas Gerais, Brasil: influência do entorno familiar e grupo social [Smoking among adolescents and young adults in Belo Horizonte, Minas Gerais State, Brazil: the influence of family setting and social group]. Cad Saúde Pública (2011.0) 27 935-943. PMID: 21655844 11. Leatherdale ST, Hammond D, Ahmed R. Alcohol, marijuana, and tobacco use patterns among youth in Canada. Cancer Causes Control (2008.0) 19 361-369. PMID: 18058247 12. Faeh D, Viswanathan B, Chiolero A, Warren W, Bovet P. Clustering of smoking, alcohol drinking and cannabis use in adolescents in a rapidly developing country. BMC Public Health (2006.0) 6 169-169. PMID: 16803621 13. Bonilha AG, Ruffino-Netto A, Sicchieri MP. Correlatos de experimentação e consumo atual de cigarros entre adolescentes [Correlates of experimentation with smoking and current cigarette consumption among adolescentes]. J Bras Pneumol (2014.0) 40 634-642. PMID: 25610504 14. Reed MB, Wang R, Shillington AM, Clapp JD, Lange JE. The relationship between alcohol use and cigarette smoking in a sample of undergraduate college students. Addict Behav (2007.0) 32 449-464. PMID: 16844313 15. Duhig AM, Cavallo DA, McKee SA, George TP, Krishnan-Sarin S. Daily patterns of alcohol, cigarette, and marijuana use in adolescent smokers and nonsmokers. Addict Behav (2005.0) 30 271-283. PMID: 15621398 16. Figueiredo VC, Szklo AS, Costa LC. ERICA: prevalência de tabagismo em adolescentes brasileiros [ERICA: smoking prevalence in Brazilian adolescents]. Rev Saúde Pública (2016.0) 50 12s-12s 17. Yue Y, Hong L, Guo L. Gender differences in the association between cigarette smoking, alcohol consumption and depressive symptoms: a cross-sectional study among Chinese adolescents. Sci Rep (2015.0) 5 17959-17959. PMID: 26639938 18. Liu Y, Wang M, Tynjälä J. Socioeconomic differences in adolescents’ smoking: a comparison between Finland and Beijing, China. BMC Public Health (2016.0) 16 805-805. PMID: 27534849 19. Zanuto EAC, de Lima MCS, de Araújo RG. Distúrbios do sono em adultos de uma cidade do Estado de São Paulo [Sleep disturbances in adults in a city of Sao Paulo state]. Rev Bras Epidemiol (2015.0) 18 42-53. PMID: 25651010 20. Galduróz JCF, Noto AR, Fonseca AM, Carlini EA. V Levantamento nacional sobre o consumo de drogas psicotrópicas entre estudantes de ensino fundamental e médio da rede pública de ensino nas 27 capitais brasileiras (2004.0) 21. Moreira LB, Fuchs FD, Moraes RS. Alcoholic beverage consumption and associated factors in Porto Alegre, a southern Brazilian city: a population-based survey. J Stud Alcohol (1996.0) 57 253-259. PMID: 8709583 22. Gordon CC, Chumlea WC, Roche AF, Lohman TG, Roche AF, Martorel R. Stature, recumbent length and weight. Anthropometric standardization reference manual (1988.0) 23. Cole TJ, Bellizzi MC, Flegal KM, Dietz WH. Establishing a standard definition for child overweight and obesity worldwide: international survey. BMJ (2000.0) 320 1240-1243. PMID: 10797032 24. 24 World Health Organization Division of Noncommunicable Diseases. Programme of Nutrition Family and Reproductive Health. Obesity: Preventing and managing the global epidemic. Report of a WHO consultation on obesity Geneva World Health Organization 1998. Division of Noncommunicable Diseases. Programme of Nutrition Family and Reproductive Health. Obesity: Preventing and managing the global epidemic. Report of a WHO consultation on obesity (1998.0) 25. 25 Associação Brasileira de Empresas de Pesquisa Critério de Classificação Econômica Brasil. Alterações na aplicação do Critério Brasil, válidas a partir de 01/01/2013 http://www.abep.org/Servicos/Download.aspx?id=02 Accessed in 2017 (Aug 18). Critério de Classificação Econômica Brasil. Alterações na aplicação do Critério Brasil, válidas a partir de 01/01/2013 26. Precioso J, Samorinha C, Macedo M, Antunes H. Prevalência do consumo de Tabaco em adolescents escolarizados portugueses por sexo: podemos estar otimistas? [Smoking prevalence in Portuguese school-aged adolescents by gender: Can we be optimistic?]. Revista Portuguesa de Pneumologia (2012.0) 18 182-187. PMID: 22542095 27. Levy D, de Almeida LM, Szklo A. The Brazil SimSmoke policy simulation model: the effect of strong tobacco control policies on smoking prevalence and smoking-attributable deaths in a middle-income nation. PLoS Med (2012.0) 9. PMID: 23139643 28. Audrain-McGovern J, Benowitz NL. Cigarette smoking, nicotine, and body weight. Clin Pharmacol Ther (2011.0) 90 164-168. PMID: 21633341 29. O’Loughlin J, Karp I, Koulis T, Paradis G, Difranza J. Determinants of first puff and daily cigarette smoking in adolescents. Am J Epidemiol (2009.0) 170 585-597. PMID: 19635735 30. Laird Y, Fawkner S, Kelly P, McNamee L, Niven A. The role of social support on physical activity behaviour in adolescent girls: a systematic review and meta-analysis. Int J Behav Nutr Phys Act (2016.0) 13 79-79. PMID: 27387328 31. Elicker E, Palazzo LS, Aerts DRGC, Alves GG, Câmara S. Uso de álcool, tabaco e ouras drogas por adolescentes escolares de Porto Velho-RO, Brasil [Use of alcohol, tobacco and other drugs by adolescent students from Porto Velho-RO, Brazil]. Epidemiol Serv Saúde (2015.0) 24 399-410 32. Khuder SA, Price JH, Jordan T, Khuder SS, Silvestri K. Cigarette smoking among adolescents in Northwest Ohio: correlates of prevalence and age at onset. Int J Environ Res Public Health (2008.0) 5 278-289. PMID: 19190357 33. Huang HW, Lu CC, Yang YH, Huang CL. Smoking behaviours of adolescents, influenced by smoking of teachers, family and friends. Int Nurs Rev (2014.0) 61 220-227. PMID: 24571366 34. Feduccia AA, Chatterjee S, Bartlett SE. Neuronal nicotinic acetylcholine receptors: neuroplastic changes underlying alcohol and nicotine addictions. Front Mol Neurosci (2012.0) 5 83-83. PMID: 22876217 35. Farias JC, Nahas MV, Barros MVG. Comportamentos de risco à saúde em adolescents no Sul do Brasil: prevalência e fatores associados [Health risk behaviors among adolescents in the south of Brazil: prevalence and associated factors]. Rev Panam Salud Publica (2009.0) 25 344-352. PMID: 19531323 36. Silva MP, Silva RMVG, Botelho C. Fatores associados à experimentação do cigarro em adolescents [Factors associated with cigarette experimentation among adolescentes]. J Bras Pneumol (2008.0) 34 927-935. PMID: 19099099 37. Sitrin D, Bishai D. The association between cigarette smoking and work status among Egyptian adolescent males. Int J Tuberc Lung Dis (2008.0) 12 670-676. PMID: 18492335 38. West HW, Juonala M, Gall SL. Exposure to parental smoking in childhood is associated with increased risk of carotid atherosclerotic plaque in adulthood: the Cardiovascular Risk in Young Finns Study. Circulation (2015.0) 131 1239-1246. PMID: 25802269
--- title: 'Engagement in physical education classes and health among young people: does sports practice matter? A cross-sectional study' authors: - Diogo Henrique Constantino Coledam - Philippe Fanelli Ferraiol journal: São Paulo Medical Journal year: 2017 pmcid: PMC10016014 doi: 10.1590/1516-3180.2017.0111260617 license: CC BY 4.0 --- # Engagement in physical education classes and health among young people: does sports practice matter? A cross-sectional study ## ABSTRACT ### CONTEXT AND OBJECTIVE: Physical education classes aim to promote health but it is unknown whether benefits occur independently of sports practice. The purpose of this study was to examine associations between engagement in physical education classes and physical fitness and obesity according to sports practice among Brazilian students. ### DESIGN AND SETTING: Cross-sectional school-based study involving 737 students aged 10-17 years in southern Brazil. ### METHODS: Engagement in physical education classes and sports practice were analyzed using a self-report questionnaire. The health indicators analyzed were cardiorespiratory fitness, muscle strength, obesity and combinations thereof. The covariates were sex, age, socioeconomic status, physical activity and sedentary behavior. Prevalence ratios (PR) adjusted for confounding variables were estimated using Poisson regression. Analyses were stratified according to sports practice. ### RESULTS: Engagement in physical education classes was associated with achievement of health-related criteria for cardiorespiratory fitness (PR = 1.52), muscle strength (PR = 1.55), obesity + cardiorespiratory fitness (PR = 1.51), obesity + muscle strength (PR = 1.70), cardiorespiratory fitness + muscle strength (PR = 2.60) and the three outcomes combined (PR = 2.43), only among non-sports practitioners, all $P \leq 0.05.$ Engagement in physical education classes was not associated with obesity (PR = 1.00, $P \leq 0.05$). No associations were found for sports practitioners ($P \leq 0.05$). ### CONCLUSION: Engagement in physical education classes was associated with health among non-sports practitioners. However, to protect students from obesity and promote additional health benefits for sports practitioners, the conventional physical education program offered to the sample studied should be reformulated. ## INTRODUCTION Physical activity is an important health-related behavior in different age groups. It is associated with mental, cardiovascular and bone health, lower adiposity, greater physical fitness, greater motor skills development and better quality of life among children and adolescents.1 Approximately $70\%$ of Brazilian adolescents are inactive and therefore promotion of physical activity in this age group is necessary.2 In this context, since schools have a responsibility for promoting physical activity, the subject of physical education takes on an important role in relation to public health among young people,3 with the aims of increasing the amount of moderate to vigorous physical activity and decreasing students’ daily sedentary behavior.4 In addition, physical education classes have the goal of providing students with knowledge, skills and confidence so that they can be active throughout their lifetime, thus preventing the emergence of health problems.3,5 Several observational and experimental studies have been conducted to examine the benefits of physical education classes for schoolchildren’s health. Physical fitness and obesity are widely investigated outcomes because of their contribution to cardiovascular, metabolic, musculoskeletal and mental health among young people.6,7 Experimental studies have demonstrated that intervention programs within physical education classes increase muscle strength8,9 and cardiorespiratory fitness,8,9,10 and decrease the body mass index8 and prevalence of overweight11 among children and adolescents. Similarly, an observational study that evaluated 91,236 fifth-grade students in California, United States, found that policies offering physical education classes were associated with better cardiorespiratory fitness.12 In analyzing the relationship between physical education programs and health among young people, sports practice is a variable that needs to be considered for two reasons. Firstly, sports practice is associated with habitual physical activity among young people13,14 and consequently increases their cardiorespiratory fitness and muscle strength,13,15,16 and decreases their overweight and obesity.17 Secondly, it has recently been reported that the effects of physical education classes on cardiorespiratory fitness occur only among young people in a poor physical condition,10 who are probably not sports practitioners. Despite the information provided, studies conducted so far with the aim of examining the relationship between physical education classes and health, as well as those analyzing the effects of intervention programs within physical education classes, have not considered whether the participants were sports practitioners. This limitation prevents knowledge of whether the benefits of physical education classes on health occur among both young people who practice and those who do not practice sports in their leisure time. ## OBJECTIVE The aim of the present study was to examine associations of engagement in physical education classes with physical fitness and obesity, according to sports practice, among Brazilian young people. ## Ethics This study was approved by the Ethics Committee for Research Involving Human Beings of the State University of Londrina (Universidade Estadual de Londrina, UEL), Paraná, Brazil, under protocol $\frac{312}{2011.}$ A parent or legal guardian provided written informed consent through signing a statement in which the aims of the study, details about the procedures, risks and benefits of the study and contact details of the researcher were described. ## Study sample and design This was a cross-sectional study that formed part of a larger epidemiological survey entitled “Physical education and health criteria achievement in Brazilian young people”, which was conducted from May to July 2012. The aim of the larger survey was to investigate the association between engagement in physical education classes and health indicators in a representative sample of students in the city of Londrina, Paraná, Brazil. The study population was composed of students enrolled in state schools in Londrina in 2012. The inclusion criteria were that they needed to: Agree voluntarily to participate in the study;Provide an informed consent statement signed by a legal representative;Be aged between 10 and 17 years;Be enrolled in a state school;Not present any physical or metabolic limitations that would prevent performance of any study procedures; andUndergo all the proposed procedures. For this study, the sample size was estimated considering a population of 55,475, outcome prevalence of $40\%$, confidence interval of $95\%$, design effect of 2, and sample loss of $30\%$, using the Epi Info 7.0 software. The minimum sample size was estimated as 732 students. Out of the 965 students invited to participate in this study, 737 met the eligibility criteria and composed the final sample. The students were aged 10 to 17 years and were probabilistically selected through clusters (school and classrooms) that were stratified according to region of the city (north, south, east, west and center), sex and school year. The sampling procedure was performed in two stages. One school from each region of the city was selected randomly and the proportional number of students in the region was assessed using full classrooms (25-30 students). ## Data collection and variables All procedures were carried out at the school in which the participants were enrolled. The questionnaire was answered and the anthropometric measurements were performed in the classroom. The field tests were carried out in the school’s indoor sports court. All information was collected within a maximum period of three days. The independent variables of the present study were sports practice and engagement in physical education classes. Sports practice was analyzed by means of the following question: “In leisure time activities, do you practice sports?”, with the following response options: never; rarely; sometimes; frequently; or always. The question was taken from the Questionnaire of Habitual Physical Activity.18 Participants who answered “frequently” or “always” were considered to be sports practitioners. Engagement in physical education classes was assessed using two self-report questions: 1. In this semester, did you participate in physical education classes?”, with answer options “no”, “yes, but only one class per week” or “yes, I participated in all classes”. This question presented $90\%$ agreement one month later, through direct observation among 40 students selected from the sample of the present study. 2. “ Generally, during physical education classes, how active were you, i.e. did you play, run, jump and throw balls intensely?” with the following response options: “I didn’t participate in the classes,” “rarely,” “sometimes,” “often” or “always.” This question was adapted from the PAQ-C questionnaire (Physical Activity Questionnaire for Children). The translation and cross-cultural adaptation of PAQ-C into the Portuguese language, and its reproducibility and concurrent validity, have been described elsewhere.19 We tested the validity of question 2 for assessing the intensity of classes, by using a perceived exertion scale20 in eight physical education classes: two classes a week for one month. Students who reported being active during classes presented significantly higher perceived exertion than did those who reported not being active: 4.0 (3.0-5.0) versus 6.5 (4.5-7.5) arbitrary units; $P \leq 0.05.$ Participants who answered that they had participated in all physical education classes and were “often” or “always” active during classes were considered to be engaged in physical education classes. The independent variables presented high reproducibility (agreements of $80\%$ and $93.4\%$). ## Outcomes The outcomes of the study were cardiorespiratory fitness, muscle strength, obesity and combinations of these outcomes. Cardiorespiratory fitness was evaluated by means of the multistage 20-meter shuttle run test.21 This is a progressive test and participants are required to run back and forth over a 20-meter distance. The velocity starts at 8.5 km/h and increases by 0.5 km/h each minute until voluntary exhaustion. Upper-limb muscle strength was estimated using the 90° push-up test.22 The cutoffs used for cardiorespiratory fitness and muscle strength (health fitness zone) were as proposed through Fitnessgram, according to sex and age.22 Nutritional status was assessed through the body mass index (BMI = body mass/height2). Measurements of body mass and height were obtained using a digital scale and a portable stadiometer. The cutoff points used to classify obesity were as proposed by the International Obesity Task Force.23 The above outcomes were analyzed both separately and in combinations. ## Covariates The covariates used to adjust the analysis were sex, age, socioeconomic condition, physical activity and sedentary behavior. Socioeconomic condition was estimated using the questionnaire of the Brazilian Association of Polling Companies. The Questionnaire of Habitual Physical Activity was used to assess physical activity.18 The following question was used to assess sedentary behavior: “How many hours on average do you watch TV, play video games or use the computer,” with the following response options: < 1 hour per day, 1 hour per day, 2 hours per day, 3 hours per day, 4 hours per day or 5 or more hours per day. ## Physical education curriculum In the year during which the study was conducted, the school subject of physical education was taught by a physical education teacher and each student had timetabled classes totaling 100 minutes/week. All schools included in the sample had an indoor sports court. In the state schools of the state of Paraná, the physical education curriculum has the objective of teaching body culture, which is based on the cultural forms of human movement historically produced by humanity. This is based on the assumption that the pedagogical practice of physical education within the school context should turn the different forms of body expression activities into topics, consisting of the following: games, sports, gymnastics, rhythmic activities and martial arts. With the aims of increasing knowledge of reality and establishing relationships between everyday social and cultural phenomena, the curriculum also includes the following articulating elements: body, playfulness, health, world of work, technical and tactical elements, leisure time, diversity and media. This curriculum is the same as in other Brazilian states and details have previously been described.24 ## Statistical analyses Descriptive statistics were produced, comprising relative frequencies and $95\%$ confidence intervals. The chi-square test was used to analyze the bivariate association between engagement in physical education classes and health indicators. Multivariate analysis was performed using Poisson regression to estimate prevalence ratios (PR) and $95\%$ confidence intervals. The analyses were stratified according to sports practice and independent variables were inserted simultaneously in the final model. Because of the complex sample used and stratifications of the analysis according to sports practice, the multivariate analysis was conducted considering the strata, primary sample units and sample weight, using the “survey” (svy) command of STATA 11.0. In all cases, results were considered significant when $P \leq 0.05.$ ## RESULTS The sample loss from the present study was $23.6\%$. This loss arose because some of the students did not perform all the study procedures. However, this did not affect the representativeness of the sample, given that missing values had been anticipated in the sample size estimates, and because the losses did not change the proportions among the participants according to sex, age, socioeconomic status or region of the city. Moreover, the losses did not prevent the study from attaining the minimum sample size required to conduct the analysis. Out of the 737 students, $35\%$ reported practicing sports. Higher proportions of the sports practitioners were males, were aged between 10 and 12 years, were engaged in physical education classes and achieved the health criteria for cardiorespiratory fitness and muscle strength ($P \leq 0.05$). No differences between the proportions of sports practitioners and non-practitioners regarding socioeconomic status or obesity were found ($P \leq 0.05$) (Table 1). Table 1.Descriptive characteristics of the study participants ($$n = 737$$)1Achievement of health status according to the criteria proposed through Fitnessgram; $95\%$ CI = $95\%$ confidence interval of prevalence; P refers to the chi-square test. The bivariate analyses are presented in Table 2. Among the young people who did not practice sports, positive associations were found between engagement in physical education classes and health criterion achievement regarding cardiorespiratory fitness and muscle strength ($P \leq 0.05$). No association was found regarding obesity ($P \leq 0.05$). In analyzing combined outcomes, positive associations were found between engagement in physical education classes and the following variables: obesity + cardiorespiratory fitness, obesity + muscle strength, cardiorespiratory fitness + muscle strength and all outcomes combined ($P \leq 0.05$). No association was found between engagement in physical education classes and achievement of health criteria for any of the outcomes analyzed, for the participants who were sports practitioners ($P \leq 0.05$). Table 2.Bivariate association analysis between engagement in physical education classes and achievement of health criteria among studentsPR = prevalence ratio; % = relative frequency; P = P for chi-square test; 1Achievement of health status according to the criteria proposed through FitnessGram; 2Health criteria combined (cardiorespiratory fitness, muscle strength and obesity). The results described in the bivariate analysis were maintained after adjustment for the confounding variables (Table 3). Young people who reported not practicing sports but being engaged in physical education classes were more likely to achieve the health criteria for cardiorespiratory fitness (PR = 1.52), muscle strength (PR = 1.55), obesity + cardiorespiratory fitness (PR = 1.51), obesity + muscle strength (PR = 1.70), cardiorespiratory fitness + muscle strength (PR = 2.60) and all outcomes combined (PR = 2.43), all with $P \leq 0.05.$ Being engaged in physical education classes was not associated with obesity for those who were not sports practitioners or with any of the outcomes for those who were sports practitioners ($P \leq 0.05$). Table 3.Multivariate analysis on the association between engagement in physical education (PE) classes and achievement of health criteria among studentsPR = prevalence ratio; $95\%$ CI = $95\%$ confidence interval; 1Achievement of health status according to the criteria proposed through FitnessGram; 2Health criteria combined (cardiorespiratory fitness, muscle strength and obesity); 3Adjusted for sex, age, socioeconomic status, obesity, physical activity and sedentary behavior; 4Adjusted for sex, age, socioeconomic status, physical activity and sedentary behavior. ## DISCUSSION The aim of this study was to analyze the association between engagement in physical education classes and some health indicators, according to sports practice among young Brazilians. The novelty of this study was that engagement in physical education classes was associated with cardiorespiratory fitness, muscle strength and combined health outcomes, only among students who did not practice sports. In contrast, engagement in physical education classes was not associated with obesity, independent of sports practice. Although the present study had a cross-sectional design, the results found regarding participants who were not sports practitioners corroborate previous experimental studies that demonstrated increases in cardiorespiratory fitness8,9,10 and muscle strength8,9 after implementation of intervention programs within physical education classes. Likewise, they corroborate an observational study that was carried out on a representative sample in the American state of California. The results from that study demonstrated that adoption of public policies to promote physical education classes was associated with higher cardiorespiratory fitness among schoolchildren.12 Despite the information available regarding the relationship between physical education programs and health, none of the studies listed above considered students’ sports practice, which limits comparison of the results. Differently from previous studies, it was sought in the present study to investigate whether engagement in physical education classes was associated with health indicators, among both young people who practiced sports and those who did not. There was no benefit in engaging in physical education classes for the students who were sports practitioners, in relation to any of the variables analyzed. This can probably be explained in terms of the adaptations resulting from the intensity of sports practice. Although sports practice during leisure time was analyzed in the present study, this type of activity is usually performed at high intensities and results in positive cardiovascular and muscle adaptations.25 Associations between sports practice and cardiorespiratory fitness,13,16 muscle strength15 and protection against overweight and obesity have been described.17 Because young people who practice sports are more active,13,14 they are probably protected from the outcomes analyzed in the present study. Hence, engagement in physical education classes would not present any additional benefit. One previous result that reinforces this statement was the finding that a physical education program only increased cardiorespiratory fitness among young people who were in a poor physical condition,10 who probably were not practicing sports or physical exercise. The results from the present study also indicate that there is a need to estimate sports practice when analyzing the effects of both intervention programs and conventional physical education on the health of young people, in order to better understand the results. However, this methodological procedure is not commonly performed. In the present study, the outcomes were analyzed separately and in combinations. Analysis on combined outcomes is necessary for two reasons. Firstly, in combining low physical fitness with indicators of high adiposity, there is an increase in cardiovascular risk in comparison with the separate outcomes.26,27 Secondly, cardiorespiratory fitness, muscle strength and low adiposity are independently associated with cardiometabolic risk28,29,30 and thus, it is desirable that young people should fulfill all of these health criteria. Thus, the results showed that engagement in physical education classes was associated with higher probability of achievement of combined health criteria among participants who were not sports practitioners. It is worth noting that it was difficult to compare the results found with those previously reported, since it has been unusual for studies investigating relationships between physical education classes and health to use combined outcomes. From a public health point of view, this is an important result, given that sports make a large contribution towards total physical activity.13,14 Students who do not practice sports may perform less daily physical activity and be exposed to the risks of future non-communicable diseases.2 Regarding the results relating to physical fitness and combined outcomes, it had been expected that engagement in physical education classes would be associated with obesity among students who were not sports practitioners, but this did not occur. This result corroborates a previous study that demonstrated that school physical education, as typically offered, does not reduce or prevent obesity.31 Although it has the potential to assist in controlling obesity, there is a need to implement stricter policies to promote physical education.32 Another point that should be highlighted is that strategies for promoting physical activity in isolation do not demonstrate efficacy in reducing obesity. Obesity-related outcomes were found to be improved in intervention programs with two or three components (i.e. physical activity plus nutrition and physical activity plus both education and nutrition),33,34 which did not occur in conventional physical education programs as analyzed in the present study. The present study has limitations that need to be considered in interpreting the results. Although previous studies provided the theoretical basis for demonstrating the effects of physical education classes on students’ health, the design used here was cross-sectional, which prevented inferences regarding causality among the associations identified. The main limitation was that engagement in physical education classes was estimated through a self-report questionnaire, which prevented accurate measurement of the intensity of classes, in comparison with objective measurements. However, this limitation was attenuated, given that the instrument used was valid for detecting students who reported higher perceived exertion during classes. Regarding sports practice, although the instrument used showed correlations with the amount of daily physical activity,35 it presented the limitation of not estimating which types of sports were practiced by these young people, or the volume and intensity of the activities. Despite these limitations, the present study had a representative sample and analyzed a conventional physical education program with outcomes relating to public health, using multivariate analysis. This enables generalization of the results to populations with similar characteristics and similar physical education programs. Physical education plays an important role in health promotion among young people who do not practice sports, because of the protection that it provides against the risk of low physical fitness. However, to provide protection against obesity and obtain additional benefits regarding the health of young sports practitioners, conventional Brazilian physical education programs require improvement. Future studies aiming towards examining the relationship between physical education programs and health should consider sports practice, in order to better understand the benefits among young people. ## CONCLUSION The benefits of engagement in physical education classes regarding cardiorespiratory fitness and muscle strength were only seen among students who did not practice sports. On the other hand, no association was observed regarding obesity. Benefits were also observed when the variables of cardiorespiratory fitness, muscle strength and obesity were combined for analysis. ## References 1. Poitras VJ, Gray CE, Borghese MM. Systematic review of the relationships between objectively measured physical activity and health indicators in school-aged children and youth. Appl Physiol Nutr Metab (2016) 41 S197-S239. PMID: 27306431 2. Bergmann GG, Bergmann ML, Marques AC, Hallal PC. Prevalência e fatores associados à inatividade física entre adolescentes da rede pública de ensino de Uruguaiana, Rio Grande do Sul, Brasil [Prevalence of physical inactivity and associated factors among adolescents from public schools in Uruguaiana, Rio Grande do Sul State, Brazil]. Cad Saúde Pública (2013) 29 2217-2229. PMID: 24233037 3. McKenzie TL, Lounsbery MA. The pill not taken: revisiting Physical Education Teacher Effectiveness in a Public Health Context. Res Q Exerc Sport (2014) 85 287-292. PMID: 25141081 4. Chen S, Kim Y, Gao Z. The contributing role of physical education in youth’s daily physical activity and sedentary behavior. BMC Public Health (2014) 14 110-110. PMID: 24495714 5. Sallis JF, McKenzie TL, Beets MW. Physical education’s role in public health: steps forward and backward over 20 years and HOPE for the future. Res Q Exerc Sport (2012) 83 125-135. PMID: 22808697 6. Ortega FB, Ruiz JR, Castillo MJ, Sjöström M. Physical fitness in childhood and adolescence: a powerful marker of health. Int J Obes (Lond) (2008) 32 1-11. PMID: 18043605 7. Kumar S, Kelly AS. Review of Childhood Obesity: From Epidemiology, Etiology, and Comorbidities to Clinical Assessment and Treatment. Mayo Clin Proc (2017) 92 251-265. PMID: 28065514 8. Erfle SE, Gamble A. Effects of daily physical education on physical fitness and weight status in middle school adolescents. J Sch Health (2015) 85 27-35. PMID: 25440450 9. Mayorga-Vega D, Montoro-Escaño J, Merino-Marban R, Viciana J. Effects of a physical education-based programme on health-related physical fitness and its maintenance in high school students: A cluster-randomized controlled trial. European Physical Education Review (2016) 22 243-259 10. Mayorga-Vega D, Viciana J. Las clases de educación física solo mejoran la capacidad cardiorrespiratoria de los alumnos con menor condición física: un estudio de intervención controlado [Physical education classes only improve cardiorespiratory fitness of students with lower physical fitness: a controlled intervention study]. Nutr Hosp (2015) 32 330-335. PMID: 26262735 11. Klakk H, Chinapaw M, Heidemann M, Andersen LB, Wedderkopp N. Effect of four additional physical education lessons on body composition in children aged 8-13 years--a prospective study during two school years. BMC Pediatr (2013) 13 170-170. PMID: 24131778 12. Sanchez-Vaznaugh EV, Sánchez BN, Rosas LG, Baek J, Egerter S. Physical education policy compliance and children’s physical fitness. Am J Prev Med (2012) 42 452-459. PMID: 22516484 13. Telford RM, Telford RD, Cochrane T. The influence of sport club participation on physical activity, fitness and body fat during childhood and adolescence: The LOOK Longitudinal Study. J Sci Med Sport (2016) 19 400-406. PMID: 26111721 14. Vella SA, Schranz NK, Davern M. The contribution of organised sports to physical activity in Australia: Results and directions from the Active Healthy Kids Australia 2014 Report Card on physical activity for children and young people. J Sci Med Sport (2016) 19 407-412. PMID: 25979479 15. Beets MW, Pitetti KH. Contribution of physical education and sport to health‐related fitness in high school students. J Sch Health (2005) 75 25-30. PMID: 15776877 16. Silva G, Andersen LB, Aires L. Associations between sports participation, levels of moderate to vigorous physical activity and cardiorespiratory fitness in children and adolescents. J Sports Sci (2013) 31 1359-1367. PMID: 23631663 17. Drake KM, Beach ML, Longacre MR. Influence of sports, physical education, and active commuting to school on adolescent weight status. Pediatrics (2012) 130 e196-e304 18. Baecke JA, Burema J, Frijters JE. A short questionnaire for the measurement of habitual physical activity in epidemiological studies. Am J Clin Nutr (1982) 36 936-942. PMID: 7137077 19. Guedes DP, Guedes JERP. Medida da atividade física em jovens brasileiros: reprodutibilidade e validade do PAQ-C e do PAQ-A [Measuring physical activity in brazilian youth: reproducibility and validity of the PAQ-C and PAQ-A]. Rev Bras Med Esporte (2015) 21 425-432 20. Yelling M, Lamb KL, Swaine IL. Validity of a pictorial perceived exertion scale for effort estimation and effort production during stepping exercise in adolescent children. European Physical Education Review (2002) 8 157-175 21. Léger LA, Mercier D, Gadoury C, Lambert J. The multistage 20 metre shuttle run test for aerobic fitness. J Sports Sci (1988) 6 93-101. PMID: 3184250 22. Welk G, Meredith MD. Fitnessgram and Activitygram Test Administration Manual (2010) 23. Cole TJ, Lobstein T. Extended international (IOTF) body mass index cut‐offs for thinness, overweight and obesity. Pediatr Obes (2012) 7 284-294. PMID: 22715120 24. Betti M, Knijnik J, Venâncio L, Sanches L. In search of the autonomous and critical individual: a philosophical and pedagogical analysis of the physical education curriculum of São Paulo (Brazil). Physical Education and Sport Pedagogy (2015) 20 427-441 25. Hammami A, Chamari K, Slimani M. Effects of recreational soccer on physical fitness and health indices in sedentary healthy and unhealthy subjects. Biol Sport (2016) 33 127-137. PMID: 27274105 26. Bergmann GG, de Araújo Bergmann ML, Hallal PC. Independent and combined associations of cardiorespiratory fitness and fatness with cardiovascular risk factors in Brazilian youth. J Phys Act Health (2014) 11 375-383. PMID: 23364256 27. Reuter CP, Silva PT, Renner JD. Dislipidemia Associa-se com Falta de Aptidão e Sobrepeso-Obesidade em Crianças e Adolescentes [Dyslipidemia is Associated with Unfit and Overweight-Obese Children and Adolescents]. Arq Bras Cardiol (2016) 106 188-193. PMID: 26885973 28. Buchan DS, Young JD, Boddy LM, Baker JS. Independent associations between cardiorespiratory fitness, waist circumference, BMI, and clustered cardiometabolic risk in adolescents. Am J Hum Biol (2014) 26 29-35. PMID: 24136895 29. Buchan DS, Boddy LM, Young JD. Relationships between Cardiorespiratory and Muscular Fitness with Cardiometabolic Risk in Adolescents. Res Sports Med (2015) 23 227-239. PMID: 26114326 30. Sasayama K, Ochi E, Adachi M. Importance of both fatness and aerobic fitness on metabolic syndrome risk in Japanese children. PloS One (2015) 10. PMID: 25993528 31. Casazza K, Fontaine KR, Astrup A. Myths, presumptions, and facts about obesity. N Engl J Med (2013) 368 446-454. PMID: 23363498 32. Kahan D, McKenzie TL. The potential and reality of physical education in controlling overweight and obesity. Am J Public Health (2015) 105 653-659. PMID: 25713972 33. Shirley K, Rutfield R, Hall N. Combinations of obesity prevention strategies in US elementary schools: a critical review. J Prim Prev (2015) 36 1-20. PMID: 25288474 34. von Hippel PT, Bradbury WK. The effects of school physical education grants on obesity, fitness, and academic achievement. Prev Med (2015) 78 44-51. PMID: 26163396 35. Coledam DHC, Ferraiol PF, Pires R, dos-Santos JW, Oliveira AR. Prática esportiva e participação nas aulas de educação física: fatores associados em estudantes de Londrina, Paraná, Brasil [Factors associated with participation in sports and physical education among students from Londrina, Paraná State, Brazil]. Cad Saúde Pública (2014) 30 533-545. PMID: 24714943
--- title: 'A study of pulmonary function in end-stage renal disease patients on hemodialysis: a cross-sectional study' authors: - Ashima Sharma - Ashok Sharma - Sushila Gahlot - Pawan Kumar Prasher journal: São Paulo Medical Journal year: 2017 pmcid: PMC10016016 doi: 10.1590/1516-3180.2017.0179150817 license: CC BY 4.0 --- # A study of pulmonary function in end-stage renal disease patients on hemodialysis: a cross-sectional study ## ABSTRACT ### BACKGROUND: The aim here was to study acute effects of hemodialysis among end-stage renal disease (ESRD) patients. ### DESIGN AND SETTING: Prospective study in tertiary-level care center. ### METHODS: Fifty ESRD patients undergoing hemodialysis were studied. Spirometric pulmonary function tests were performed before and after four-hour hemodialysis sessions. ### RESULTS: The patients’ average age was 45.8 ± 10.0 years; $64\%$ were males and $64\%$ had normal body mass index. Anemia ($94\%$) and hypoalbuminemia ($72\%$) were common. Diabetes mellitus ($68\%$), hypertension ($34\%$) and coronary artery disease ($18\%$) were major comorbidities. Forty-five patients ($90\%$) had been on hemodialysis for six months to three years. The patients’ pre-dialysis mean forced vital capacity (FVC) and forced expiratory volume in 1 second (FEV1) were below normal: 45.8 ± $24.9\%$ and 43.5 ± $25.9\%$ of predicted, respectively. After hemodialysis, these increased significantly, to 51.1 ± $23.4\%$ and 49.3 ± $25.5\%$ of predicted, respectively ($P \leq 0.01$). The increase in mean FEV1/FVC, from 97.8 ± $20.8\%$ to 99.3 ± $20.1\%$ of predicted, was not significant ($P \leq 0.05$). The pre-dialysis mean forced expiratory flow 25-$75\%$ was 50.1 ± $31\%$ and increased significantly, to 56.3 ± $31.6\%$ of predicted ($P \leq 0.05$). The mean peak expiratory flow was below normal (43.8 ± $30.7\%$) and increased significantly, to 49.1 ± $29.9\%$ of predicted ($P \leq 0.05$). Males and females showed similar directions of change after hemodialysis. ### CONCLUSIONS: Pulmonary function abnormalities are common among ESRD patients. Comparison of pre and post-hemodialysis parameters showed significant improvements, but normal predicted values were still not achieved. ## INTRODUCTION Chronic kidney disease (CKD) is a worldwide public health problem and comprises the presence of sustained and irreversible abnormality of renal functions and loss of the kidneys’ ability to maintain homeostasis. CKD results from different causes of renal injury and can lead to progressive loss of renal function. It may reach end-stage renal disease (ESRD) after a variable period of time following the initiating injury. ESRD is the situation when kidney function is insufficient to sustain life and there is then a need for hemodialysis (HD), peritoneal dialysis (PD) or kidney transplantation, to substitute for native kidney function.1 ESRD presents not only as progressive and irreversible loss of excretory function of the kidneys but also as a complex syndrome with altered metabolic and endocrine functions. It has effects on almost all body systems, including pulmonary function.2 The relationship between the lungs and the kidneys is clinically important for both health and disease.3 Kidney failure directly and indirectly impacts the mechanical function and ventilation of the lungs, and treatment with drugs and HD are responsible for part of this effect.4 Patients with ESRD require dialysis in the form of HD or peritoneal dialysis for survival, because these can partially replace the impaired kidney function, reverse the uremic symptoms and preserve patients with ESRD, while they await a definitive solution through kidney transplantation, if possible.5 Pulmonary function tests have been compared among individuals on HD and peritoneal dialysis and among kidney transplant recipients, and it was found that pulmonary restrictive defect was the most common dysfunction in all these groups.6 Further spirometric changes were studied by Lang et al. before and after HD using different dialyzer membranes, and they found that there was no significant difference between pre-HD and post-HD vital capacity.7 On the other hand, in a similar study on the acute effects of HD, Rahgoshai et al. demonstrated that pulmonary function, and especially forced vital capacity (FVC), improved after a HD session; while no significant improvements in forced expiratory volume in 1 second (FEV1), FVC or FEV1/FVC ratio were observed.8 Dialysis may have beneficial effects at least in the initial stages of some respiratory disorders among CKD patients without primary lung disease. It may lead to improvement of respiratory symptoms and even pulmonary function test values.9 However, the immune response resulting from contact between blood and bioincompatible dialysis filters may cause complement activation, which can have a deteriorating effect on the respiratory system and can even cause respiratory distress.10 HD-related hypoxemia is another issue among ESRD patients. HD can reduce pulmonary edema around the small airways, and this may lead to dilation of the small airways (decreased closing capacity). It also gives rise to improved basal ventilation and perfusion.11 The effects of HD on patients with CKD relate mainly to changes to the volume of body fluid, thus leading to reduction of the amount of water in the lungs following dialysis. Hence, HD improves respiratory status but it may cause pulmonary complications as well, due to various pulmonary injuries of multifactorial origin. Moreover, the malnutrition and degenerative alterations that can occur in CKD patients persist, thereby worsening muscle loss and predisposing these patients towards fatigue, with increases in respiratory rate and work.12 ## METHODS Design: Our study was prospective and was conducted over a one-year period from November 2011 to December 2012, in a tertiary-level care center. It was an observation study. It was conducted on 50 ESRD patients undergoing HD. The study was approved by our institution’s research ethics committee. Inclusion criteria: Ambulatory, clinically stable patients in the age group of 18-60 years, who had been undergoing HD for more than three months, were included in the study. Exclusion criteria: Patients with histories of smoking (current or previous), acute infection, acute renal failure, chronic lung disease, tuberculosis, skeletal muscle abnormality, decompensated heart failure, arrhythmias or liver cirrhosis, and patients in severe respiratory distress or who were unable to undergo spirometry, as assessed by the clinician administering the treatment, were excluded from the study. The patients were given explanations regarding the purpose of the study and written informed consent was obtained from them. Individuals’ data were recorded on an assessment form. The glomerular filtration rate (GFR) was estimated by using the empirical formula for creatinine clearance (Cockcroft-Gault equation). HD was performed using the Fresenius Medical Care 4008-S, a German machine. While the patients’ blood flow range was variable from 300 to 350 ml/min, the dialysate flow was constant (500 ml/min). Dialysis was done using a biocompatible membrane and bicarbonate buffer. Intra-dialysis ultrafiltration was based on the patients’ condition and on their weight gain during the interdialytic period. The pulmonary function tests were performed using a computerized spirometer (“Medicaid Spirometer”). This automatically corrected all gas volumes to body temperature and pressure, saturated (BTPS), i.e. a set of conditions at body temperature, with ambient pressure and with saturation with water vapor. Spirograms (flow volume and volume-time graphs) were produced, along with numerical data and the predicted percentage values for the spirometric parameters. Spirometric variables were recorded 15 minutes before and after the first HD session of the week. The data obtained were analyzed statistically with the aid of the Statistical Package for the Social Sciences (SPSS) for personal computer, version 11.0, and paired t tests were used for comparative analyses. $P \leq 0.05$ was taken to be significant and $P \leq 0.01$ was taken to be highly significant. ## RESULTS This study was conducted from December 2011 to December 2012. A total of 50 patients met our inclusion criteria and were included in the study. The mean age of the study patients was 45.8 ± 10.0 years (range 24 to 60 years). The majority ($62\%$) of the patients were below fifty years of age. Males formed the predominant group among the patients ($64\%$) and $72\%$ of the patients were from a rural background. The majority ($64\%$) of the patients had a normal body mass index (BMI). BMI did not differ statistically according to gender (Table 1). Table 1:Body mass index status of study patients (in kg/m2)Body mass index (kg/m 2)No. of patients (n)(%)Underweight (< 18.5)1020Normal (18.5-24.99)3264Overweight (25-29.99)816Obese (> 30)00 On investigation, $94\%$ of the patients were anemic (hemoglobin < 12 g%). The hemoglobin values ranged from 5.0 to 12.5 g/dl (mean ± standard deviation, SD: 9.5 ± 1.6) among all the patients. For males, the hemoglobin range was 7.5-12.5 g/dl (10.2 ± 1.3), while for females, the values were 5.0-11.0 g/dl (8.3 ± 1.5). The serum albumin levels ranged from 2.4 to 4.2 gm/dl. The mean serum albumin concentration was 3.28 ± 0.48 and $72\%$ of the patients had hypoalbuminemia. Pre-dialysis serum urea levels ranged from 92.0 to 278.0 mg/dl (163.1 ± 39.1 mg/dl), serum creatinine 4.6-19.7 mg/dl (9.9 ± 3.3 mg/dL) and estimated GFR 3.0-19.8 ml/min/1.73 m2 (7.8 ± 3.4 ml/min/1.73 m2). There were no statistical differences in baseline urea and creatinine levels according to gender, but estimated GFR was significantly higher in males (8.8 ± 3.5 ml/min/1.73 m2) than in females (6.0 ± 2.3 ml/ min/1.73 m2). Forty-five patients ($90\%$) had been on HD for six months to three years and only $10\%$ had been on HD for less than six months. The proportions were identical for males and females (Table 2). The majority ($58\%$) of the study patients ($62\%$ of the males and $50\%$ of the females) were undergoing HD twice a week, while $30\%$ of all the patients ($31\%$ of the males and $28\%$ of the females) were undergoing HD three times a week. Eight percent of the study patients were on hemodialysis once a week and another four percent once every two weeks (Table 3). $68\%$ of the patients were diabetic before they were diagnosed as having CKD, while $34\%$ were known to be hypertensive and $18\%$ had coronary artery disease. Table 2:Duration of hemodialysis among study patientsDuration of hemodialysisMales ($$n = 32$$)Females ($$n = 18$$)Total ($$n = 50$$)< Six months3 ($9.4\%$)2 ($11.1\%$)5 ($10.0\%$)Six months to one year12 ($37.5\%$)7 ($38.9\%$)19 ($38.0\%$)> One year to three years17 ($53.1\%$)9 ($50.0\%$)26 ($52.0\%$) Table 3:Frequency of hemodialysis among study patientsFrequency of hemodialysisTotal ($$n = 50$$)%Three times a week1530Twice a week2958Once a week48Once every two weeks24 The percentages of the predicted spirometric parameters (% pred) and changes to spirometric parameters among our study patients after hemodialysis are presented in Table 4 and Table 5. Table 4:Percentage of predicted spirometric parameters among the study patients (before and after hemodialysis [HD]) ($$n = 50$$)Spirometric parametersBefore HD (mean ± SD)After HD (mean ± SD)Paired differencesMeanSDSEM$95\%$ CIP-valueFVC45.8 ± 24.951.1 ± 23.45.38.61.22.8-7.7< 0.001FEV143.5 ± 25.949.3 ± 25.55.89.41.33.2-8.6< 0.001FEV1/FVC (%)97.8 ± 20.899.3 ± 20.11.518.02.6-3.6-6.60.561FEF 25-$75\%$50.1 ± 31.356.3 ± 31.66.220.93.00.2-12.10.043PEFR43.8 ± 30.749.1 ± 29.95.313.11.91.6-9.10.006SD = standard deviation; SEM = standard error of the mean; CI = confidence interval; P-value from paired t-test (two-tailed); FVC = forced vital capacity; FEV1 = forced expiratory volume in 1 second; FEV1/FVC (%) = ratio between FEV1 and FVC x 100; FEF 25-$75\%$ = forced expiratory flow in 25-$75\%$ of FVC; PEFR = peak expiratory flow rate. Note: The mean FEV1/FVC (%) of the study patients was 97.8 ± $20.8\%$ of the predicted value. Table 5:Change in spirometric parameters among patients after hemodialysisSpirometric parameterMale ($$n = 32$$) Female ($$n = 18$$) P-valueIncrease Increase N%N%FVC2578.11583.30.941FEV12475.01477.80.901FEV1/FVC ratio2165.61055.60.689FEF 25-$75\%$1753.11372.20.307PEFR1959.41161.10.856FVC = forced vital capacity; FEV1 = forced expiratory volume in 1 second; FEF = forced expiratory flow; PEFR = peak expiratory flow rate. FVC: The mean FVC of the study patients was 45.8 ± $24.9\%$ pred, i.e. well below the normal predicted values for pulmonary function (normal is more than $80\%$ of the predicted values), determined through spirometry. After HD, the mean FVC increased to 51.1 ± $23.4\%$ pred, and this increase was statistically highly significant ($P \leq 0.01$) (Table 4). FEV1: The mean FEV1 of the study patients was 43.5 ± $25.9\%$ pred, which was also well below the normal predicted values (normal is more than $80\%$ of the predicted values). After HD, the mean FEV1 increased to 49.3 ± $25.5\%$ pred, and this increase was statistically highly significant ($P \leq 0.01$) (Table 4). FEV1/FVC%: The mean FEV1/FVC% of the study patients was 97.8 ± $20.8\%$ pred. After HD, the mean FEV1/FVC% increased to 99.3 ± $20.1\%$ pred, but this increase was not statistically significant ($P \leq 0.05$) (Table 4). Forced expiratory flow (FEF) 25-$75\%$: The mean FEF 25-$75\%$ of the study patients was 50.1 ± $31\%$ pred. After HD, the mean FEF 25-$75\%$ increased to 56.3 ± $31.6\%$ pred, and this increase was statistically significant ($P \leq 0.05$) (Table 4). Peak expiratory flow rate (PEFR): The mean PEFR of the study patients was 43.8 ± $30.7\%$ pred, i.e. well below the normal range (normal is more than $80\%$ of the predicted values). After HD, the mean PEFR increased to 49.1 ± $29.9\%$ pred, and this increase was statistically significant ($P \leq 0.05$) (Table 4). The overall analysis on pulmonary function in our study revealed that the majority of the patients ($82\%$) had a normal FEV1/FVC ratio (> $70\%$) and low percentages of the predicted FVC value (< $80\%$ pred), which was indicative of restrictive pulmonary disorder. Moreover, $6\%$ had an FEV1/FVC ratio less than $70\%$, which indicates obstructive respiratory disorder. However, because these patients also had low FVC, they could be included in the category of mixed respiratory disorder. Only $12\%$ of the study patients had pulmonary function in the normal range. There were no statistically significant differences in any of the spirometric parameters after HD when compared on the basis of gender. Males and females showed similar directions of change after HD (Table 5). ## DISCUSSION The mean age of our study patients was 45.8 ± 10.0 years and $62\%$ of them were < 50 years, thus suggesting that CKD had emerged as an early complication of various disorders. The preponderance of males ($64\%$) may have reflected either greater prevalence of CKD among males or, alternatively, poor availability of costly HD treatment for female patients, due to various sociocultural and economic constraints. The mean BMI of the study group was 21.6 ± 3.0 kg/m2 and $20\%$ had BMI < 18.5 kg/m2, but the majority ($64\%$) of the patients had normal BMI. This was despite chronic illness, but the patients’ obvious water retention and edema may have led to their normal BMI. Anemia ($94\%$) and hypoalbuminemia ($72\%$) were very prevalent. This may have been due to higher levels of renal dysfunction and poor nutritional status, reflecting both an inflammatory state and poor nutritional status among the patients. The high prevalence of these conditions may also have been due to a hypercatabolic state in CKD, caused by accumulation of proinflammatory cytokines and a combination of factors like uremic toxicity, insulin resistance, and amino acid losses13 during the dialysis procedure, rather than mere lack of a high protein diet.14 ## FVC (forced vital capacity) The mean FVC of the study group was 45.8 ± $24.9\%$ pred, i.e. well below the normal predicted values for pulmonary function. After HD, the mean FVC increased to 51.1 ± $23.4\%$ pred ($P \leq 0.001$). Our findings regarding FVC are in agreement with Mehmood.7,15,16,17 Decreased FVC, restrictive pattern and reduced airflows have been observed through spirometry, in studies by several authors. Chronic subclinical pulmonary edema due to increased capillary permeability and hypoalbuminemia was considered to be the cause for the decreased FVC.11,18 ## FEV1 The mean FEV1 of our study patients was 43.5 ± $25.9\%$ pred. which was also well below the normal predicted values (normal is more than $80\%$ of the predicted values). It showed a statistically significant increase after HD, to 49 ± $25.5\%$ pred ($P \leq 0.01$). However, these low FEV1 values were associated with a normal FEV1/FVC ratio in most of our patients, which suggested that the large airways were not affected in situations of chronic renal failure and that the reduction of FEV1 was primarily due to reduction in FVC, as in restrictive pulmonary disease. Reduced FEV1 as observed in our study patients has also been reported by Maehmood et al. and Nascimento et al.15,16 Inflammation and malnutrition have been found to present significant relationships with reduced pulmonary parameters.19 ## FEV1/FVC percentage ratio Most of the study patients ($82\%$) had a normal FEV1/FVC ratio (> $70\%$) and reduced FVC (i.e. < $80\%$ of the predicted values), which was indicative of restrictive pulmonary disorder, while $6\%$ had an FEV1/FVC ratio of less than $70\%$, thus pointing towards obstructive respiratory disorder. However, the latter patients also presented decreased FVC and could be included in the category of mixed respiratory disorder. Only $12\%$ of the study patients had pulmonary function within the normal range. The mean FEV1/FVC ratio of our patients was 97.8 ± $20.8\%$ pred, and this increased to 99.3 ± $20.1\%$ pred) after HD, but this increase was not statistically significant ($P \leq 0.05$). In contrast to FEV1 and FVC, there were no significant changes overall or among the subgroups of the study patients regarding the FEV1/FVC ratio after HD, because there was corresponding increase in both parameters (FVC and FEV1). No significant change in this ratio was also observed by Navari et al.4 ## FEF 25-75% (forced expiratory flow over the middle part of FVC) FEF 25-$75\%$ is measured from a segment of the FVC that includes flow from medium and small airways. Decreased flows are common in the early stages of obstructive disease. In the presence of borderline values for FEV1/FVC, a low FEF 25-$75\%$ confirms airway obstruction and sometime signals decreased cross-sectional area in the small airways. The mean FEF 25-$75\%$ among our study patients was 50.1 ± $31.3\%$ pred and, after HD, it increased to 56.3 ± $31.6\%$ pred, which was a statistically significant increase ($P \leq 0.05$). Rakovaca et al. found decreased values for FEF 25-$75\%$ among ESRD patients.20 Mahmood et al. also observed decreased values that were indicative of small airway disease, in comparison with their controls.15 The improvement in FEF 25-$75\%$ that was achieved through HD among our study patients showed that there was a situation of reversible obstruction, comprising removal of excess fluid from the lungs that had been compressing the small airways. However, chronic subclinical pulmonary edema leading to peribronchial fibrosis may also contribute towards the persistent abnormalities in the small airways that are reflected in reduced FEF 25-$75\%$ values. ## PEFR (peak expiratory flow rate) Among our study patients, the mean PEFR was 43.8 ± $30.7\%$ pred, i.e. below the normal range. After HD, the PEFR increased to 49.1 ± $29.9\%$ pred, and this increase was statistically significant ($P \leq 0.05$). Mahmood et al. reported that the values among CKD patients on HD were lower than those of normal subjects.15 Reduced PEFR before and even during HD sessions was observed by Davenport, who attributed this to activation of the complement system for neutrophils, monocytes and platelets following blood membrane interaction, thereby resulting in appreciable airway constriction.21 In contrast to the findings from our study, normal PEFR values were observed by Lang et al.7 ## CONCLUSION Pulmonary function abnormalities were common among our study patients, but were significantly ameliorated after HD. The majority of our patients had restrictive and mixed respiratory disorders. In the pulmonary function tests on our ESRD patients, spirometric parameters like FVC, FEV1 and PEFR were less than the normal predicted values (i.e. < $80\%$ of the predicted values) in the majority of these patients. In comparing the pre-HD and post-HD spirometric parameters, there was significant improvement but normal predicted values were still not achieved. ## References 1. 1 US Renal Data System Annual data report: atlas of end-stage renal disease in the United States Bethesda National Institutes of Health. National Institute of Diabetes and Digestive and Kidney Diseases 2002 Available from: https://www.usrds.org/atlas02.aspx Accessed in 2017 (Sep 11). Annual data report: atlas of end-stage renal disease in the United States (2002) 2. Brenner BM, Mackenzie HS, Fauci AS, Braunwald E, Isselbacher KJ. Disturbances of renal function. Harrison's Principles of Internal Medicine (2008) 1761-1771 3. Pierson JD. Respiratory considerations in the patient with renal failure. Respir Care (2006) 51 413-422. PMID: 16563195 4. Navari K, Farshidi H, Pour-Reza-Gholi F. Spirometry parameters in patients undergoing hemodialysis with bicarbonate and acetate dialysates. Iran J Kidney Dis (2008) 2 149-153. PMID: 19377229 5. Parmar MS. Chronic renal disease. BMJ (2002) 325 85-90. PMID: 12114240 6. Karacan O, Tutal E, Colak T. Pulmonary function in renal transplant recipients and end-stage renal disease patients undergoing maintenance dialysis. Transplant Proc (2006) 38 396-400. PMID: 16549130 7. Lang SM, Becker A, Fischer R, Huber RM, Schiffl H. Acute effects of hemodialysis on lung function in patients with end-stage renal disease. Wien Klin Wochenschr (2006) 118 108-113. PMID: 16703255 8. Rahgoshai R, Rahgoshai R, Khosraviani A, Nasiri AA, Solouki M. Acute effects of hemodialysis on pulmonary function in patients with end-stage renal disease. Iran J Kidney Dis (2010) 4 214-217. PMID: 20622309 9. Senatore M, Buemi M, Di Somma A, Sapio C, Gallo GC. [Respiratory function abnormalities in uremic patients]. G ltal Nefrol (2004) 21 29-33 10. Craddock PR, Fehr J, Brigham KL, Kronenberg RS, Jacob HS. Complement and leukocyte-mediated pulmonary dysfunction in hemodialysis. N Engl J Med (1977) 296 769-774. PMID: 840277 11. Zidulka A, Despas PJ, Milic-Emili J, Anthonisen NR. Pulmonary function with acute loss of excess lung water by hemodialysis in patients with chronic uremia. Am J Med (1973) 55 134-141. PMID: 4722852 12. Silva VG, Amaral C, Monterio MB, Nascimento DM, Boschetti JR. Efeitos do treinamento muscular inspiratório nos pacientes em hemodiálise [Effects of inspiratory muscle training in hemodialysis patients]. J Bras Nefrol (2011) 33 62-68. PMID: 21541465 13. Lim VS, Ikizler TA, Raj DS, Flanigan MJ. Does hemodialysis increase protein breakdown? Dissociation between whole-body amino acid turnover and regional muscle kinetics. J Am Soc Nephrol (2005) 16 862-868. PMID: 15716333 14. Adams GR, Vaziri ND. Skeletal muscle dysfunction in chronic renal failure: effect of exercise. Am J Physiol Renal Physiol (2006) 290 F753-F761. PMID: 16527920 15. Mahmoud BL, Abdulkader A, El-Sharkawy MM, Khalil HH. Assessment of pulmonary functions in chronic renal failure patients with different haemodialysis regimens. J Egypt Soc Parasitol (2004) 34 1025-1040. PMID: 15587326 16. Nascimento MM, Quershi RA, Stenvinkel P. Malnutrition and inflammation are associated with impaired pulmonary function in patients with chronic kidney disease. Nephrol Dial Transplant (2004) 19 1823-1828. PMID: 15150347 17. Bark H, Heimer D, Chaimovitz C, Mostoslovski M. Effect of chronic renal failure on respiratory muscle strength. Respiration (1988) 54 153-161. PMID: 3247515 18. Prezant DJ. Effect of uremia and its treatment on pulmonary function. Lung (1990) 168 1-14 19. Yoon SH, Choi NW, Yun SR. Pulmonary dysfunction is possibly a marker of malnutrition and inflammation but not mortality in patients with end-stage renal disease. Nephron Clin Pract (2009) 111 c1-c6 20. Rajkovaca Z, Kovacevic P, Jakovljevic B, Eric Z. Detection of pulmonary calcification in haemodialised patients by whole-body scintigraphy and the impact of the calcification to parameters of spirometry. Bosn J Basic Med Sci (2010) 10 303-306. PMID: 21108612 21. Davenport A, Williams AJ. Fall in peak expiratory flow during haemodialysis in patients with chronic renal failure. Thorax (1988) 43 693-696. PMID: 3194875
--- title: Assessment of Nutritional Status of Under-Five Children in an Urban Area of South Delhi, India journal: Cureus year: 2023 pmcid: PMC10016022 doi: 10.7759/cureus.34924 license: CC BY 3.0 --- # Assessment of Nutritional Status of Under-Five Children in an Urban Area of South Delhi, India ## Abstract Introduction Malnutrition among children continues to be a severe public health problem worldwide, whether in a developing country like India or a developed nation. Correct estimation of the problem is a prerequisite to planning the measures to control it. Objective To estimate the prevalence of undernutrition among children under five years of age by utilizing the Composite Index of Anthropometric Failure and the WHO growth charts. Methods From January to March 2020, 1332 children under the age of five years participated in a facility-based, descriptive, cross-sectional study at Fatehpur Beri, Urban Primary Health Center. An anthropometric assessment for each participant was done as per the WHO criteria. The data were entered into a Microsoft Office Excel spreadsheet (Microsoft Corporation, Redmond, WA) and analyzed with WHO Anthro software (WHO, Geneva, Switzerland) and a licensed version of SPSS 21 (IBM Corp., Armonk, NY). Continuous data were expressed using appropriate measures of central tendency, while categorical data were expressed in either frequency or proportions. Results The mean age of the study participants was 23.04 ± 18.24 months, and males ($53.3\%$) were more than ($46.7\%$) females. The prevalence of being underweight was $24.5\%$ ($\frac{327}{1332}$), of which $24.1\%$ ($\frac{79}{327}$) of children were severely underweight. Of the total study participants, $27.3\%$ ($\frac{362}{1332}$) were stunted, and $17.8\%$ ($\frac{237}{1332}$) were wasted, of which $29.1\%$ ($\frac{69}{237}$) were severely wasted. The prevalence of anthropometric failure was $45\%$. Conclusions According to the findings of this study, the prevalence of undernutrition among the study participants was substantial. Furthermore, considering weight for age as the sole criterion may underestimate the true prevalence of malnutrition. The findings have critical implications for future interventions and initiatives among children in India. ## Introduction Malnutrition among children under five years is a significant public health problem. According to the World Health Organization (WHO), malnutrition means deficiencies, excesses, or imbalances in a person's energy or nutritional consumption [1]. The term malnutrition refers to two distinct groups of conditions. The first is undernutrition, which includes being underweight (low weight for age), stunting (being short for age), wasting (being underweight for height), and nutritional deficiencies or inadequacies such as lack of essential vitamins and minerals. The second aspect refers to individuals being either overweight or obese [1,2]. Worldwide, approximately 1.9 billion adults are overweight, while 462 million are underweight. Overweight or obese children under the age of five years are estimated to number 41 million, with 159 million stunted and 50 million wasted [1]. Undernourished children have a higher risk of death and are more likely to contract childhood illness [3-5]. They are prone to be cognitively impaired, perform worse in school, have lower earning potential, and are at a higher risk of developing non-communicable diseases later in life [6]. The consequences of poor nutrition begin in utero and last for generations [7]. Undernourished women are more likely to have low birth weight babies, who are more likely to have suboptimal growth and development [8]. In response to this evidence, the WHO has set goals to reduce the number of stunted children by $40\%$ and maintain childhood wasting to less than $5\%$ by 2025 [9,10]. The United Nations (UN) adopted the first-ever UN Decade of Action on Nutrition to accelerate this process from 2016 to 2025 [10]. In support of this, goal 2 of the Sustainable Development Goals (SDGs) also purports to end hunger by 2030 [10]. Several nutrition targets were agreed upon in the years running up to 2016. To date, however, most of these targets remain unmet. As per the WHO estimates, by 2025, the number of stunted children worldwide will reach 131 million (27 million above the expected $40\%$ reduction in the target number of stunted children), while the prevalence of wasting will remain well above the $5\%$ target [9]. The majority of the global burden of childhood undernutrition remains concentrated in low-income and lower-middle-income countries [11,12]. In India, according to the National Family Health Survey 5 (NFHS-5), in 2019-2020, $32.1\%$ of children under five years of age were found to be underweight, $19.3\%$ were wasted, and $35.5\%$ were stunted [13]. As per the NFHS-5 data, the prevalence of underweight and stunting in Delhi was $21.2\%$ and $30\%$, respectively [13]. The current under-five mortality rate in India is $\frac{28}{1000}$, and undernutrition is one of the significant contributors to under-five mortality in India [14]. Weight-for-age (WFA), height-for-age (HFA), and weight-for-height (WFH) are the primary indicators used for measuring undernutrition [15]. However, in a population, different degrees of overlap between these indicators are noticed, i.e., some underweight children might also be stunted or wasted, and some children might have all three forms of anthropometric failure, i.e., stunting, wasting, and underweight. To measure the prevalence of undernutrition more precisely and accurately in a population, Svedberg developed the Composite Index of Anthropometric Failure (CIAF) [16]. In India, Nandy et al. were the first to use the concept of the CIAF on the 1998-1999 National Family Health Survey-2 (NFHS-2) data [17]. Hence, to compare the utility of the CIAF and the WHO Z-score classification, the present study was planned to determine the overall prevalence of undernutrition status among children under five years residing in an urban area of Delhi. ## Materials and methods Study setting and duration This was a cross-sectional descriptive study conducted at the Fatehpur Beri Urban Primary Health Center (UPHC) in South Delhi, which caters to a population of approximately 58,000 and falls under the South Delhi Municipal Corporation (SDMC). Children under five years contribute approximately $25\%$ of the monthly outpatient department (OPD) cases at the center. The study duration was of three months (from January 2020 to March 2020). Sampling methods A complete enumeration of all children under five years visiting the UPHC between January 2020 and March 2020, seeking immunization and healthcare services, was done. Children under the age of five years (0-59 months) are periodically monitored for growth parameters at the UPHC in accordance with programmatic guidelines under the Reproductive, Maternal, Newborn, Child, and Adolescent Health (RMNCH+A). Sample size Of the 1450 potential participants identified, 1332 agreed to participate, yielding a response rate of $91.9\%$ ($\frac{1332}{1450}$) during the three-month study period. The inclusion criteria included healthy children under five years of age and participants who were willing to participate, i.e., whose parents gave written informed consent to be a part of the present study. The exclusion criteria included children born with congenital diseases or currently undergoing treatment for chronic illnesses. Data collection Parents of the selected participants were interviewed by a trained team of doctors using a semi-structured questionnaire seeking information on the socio-demographic characteristics of the participants. Anthropometric measurements such as weight and height were recorded following the WHO guidelines [15]. Indicators based on weight, height, and age were further assessed and compared with the WHO growth reference standards [2006] and CIAF to assess the nutritional status of children [16]. Weight The participant's weight was measured in kilograms with a weighing scale to assess their growth and nutritional status using the standard technique to the nearest 0.5 kg. If the child was less than two years old or was unable to stand, tared weighing was performed, and if the child was two years or older, the child was weighed alone using a standardized, recently calibrated analog weighing machine. Height Using the standard technique, the participant's height was measured using a stadiometer/infantometer to the nearest 0.1 cm. If a child was less than two years of age, recumbent length (lying down) was measured using an infantometer; however, if the child was two years or older and could stand, standing height was measured using a stadiometer. Operational definitions Underweight The participant's weight was recorded and compared to the median values; alternatively, the participant's weight was plotted against age on a graph for comparison with the standard curve. A low weight-for-age is termed as underweight, defined as a weight-for-age Z-score (WAZ) of less than -2. Severely underweight is classified if WAZ is less than -3 of the WHO [2006] reference values [15]. Wasting Wasting is an indicator of acute malnutrition and is defined as a weight-for-height Z-score (WHZ) of less than -2. A Z-score between -2 and -3 is classified as moderate wasting. Severe wasting is classified if WHZ is less than -3 according to the WHO [2006] reference standards. The data collected were entered into the WHO Anthro software (WHO, Geneva, Switzerland) for analysis. The prevalence of undernutrition was determined using the CIAF and the WHO Z-scoring systems. Stunting Low height for age indicates stunting and depicts early chronic exposure to undernutrition. Stunting is a height-for-age Z-score (HAZ) of less than -2. A Z-score between -2 and -3 is considered moderate stunting, and severe stunting is classified if HAZ is less than -3 of the WHO [2006] reference standards. CIAF The under-nutritional status of children was also classified based on the CIAF using Nandy et al. 's model of six groups (Table 1) [17]. Table 1 \| Category \| Interpretation \| \| --- \| --- \| \| A \| No failure \| \| B \| Wasting only \| \| C \| Wasting and underweight \| \| D \| Wasting, stunting, and underweight \| \| E \| Stunting and underweight \| \| F \| Stunting only \| \| Y \| Underweight only \| ## Results Socio-demographic characteristics of the study participants A total of 1332 children were assessed to collect baseline data. More than half of the participants were males ($54\%$, $$n = 714$$), and the rest were females ($46\%$, $$n = 618$$) (Figure 1). There were 457 ($34.3\%$) children aged less than one year, 293 ($22\%$) children aged 12 months to 23 months, 187 ($14\%$) children aged between 24 and 35 months, 163 ($12.2\%$) children were in the category of 36-47 months, and 232 ($17.4\%$) children aged between 48 and 59 months (Table 2). Figure 1: Distribution of the study participants by gender ($$n = 1332$$) TABLE_PLACEHOLDER:Table 2 Nutritional assessment according to the WHO Z-scores In the current study, the prevalence of underweight, stunting, and wasting was $24.5\%$, $27.3\%$, and $17.8\%$, respectively (Figure 2). Among those underweight, a quarter ($24.1\%$) of children were severely underweight (WFA z < -3SD) (Table 3). Figure 2: Prevalence of undernutrition among the study participants ($$n = 1332$$) TABLE_PLACEHOLDER:Table 3 Nutritional assessment according to the CIAF On applying the CIAF, almost half ($45\%$) of the study participants were undernourished. As per the CIAF, $55\%$ of children fell in category A (no failure). Exclusive categories like B (wasting only), F (stunting only), and Y (underweight only) had a prevalence of $7.65\%$, $12.8\%$, and $3.75\%$, respectively. Nearly one-fifth ($20.8\%$) of the study participants had different combinations of undernourishment as demonstrated by category C (wasting and underweight), category D (wasting, underweight, and stunted), and category E (stunting and underweight) of the CIAF (Table 4). Table 4 \| CIAF category \| Anthropometric status \| Number of children (%) \| \| --- \| --- \| --- \| \| A \| No failure \| 732 (55%) \| \| B \| Wasting only \| 101 (7.7%) \| \| C \| Wasting and underweight \| 87 (6.5%) \| \| D \| Wasting, underweight, and stunted \| 48 (3.6%) \| \| E \| Stunting and underweight \| 143 (10.7%) \| \| F \| Stunting only \| 171 (12.8%) \| \| Y \| Underweight only \| 50 (3.8%) \| \| CIAF (B + C + D + E + F + Y) \| CIAF (B + C + D + E + F + Y) \| 600 (45%) \| Factors associated with undernutrition Out of the 498 study participants who were >24 months old, 151 ($30.3\%$) were stunted, whereas out of a total of 834 participants between zero and 24 months, 211 ($25.3\%$) were stunted and the difference between the two proportions was statistically significant ($p \leq 0.05$). Significantly, a higher proportion of participants between zero and 24 months when compared to those >24 months in age were found to be wasted. This difference was found to be highly significant for both groups ($p \leq 0.05$). As the age of the study participants increased, the prevalence of stunting increased, peaking at $44.9\%$ among children aged 24-35 months. The prevalence of underweight, wasting, and stunting was higher in males than females; however, this difference in proportion across genders was not found to be statistically significant ($p \leq 0.05$) (Table 5). Table 5 \| Variable \| Underweight \| Underweight.1 \| P-value \| Stunting \| Stunting.1 \| P-value.1 \| Wasting \| Wasting.1 \| P-value.2 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Variable \| Yes, n (%) \| No, n (%) \| P-value \| Yes, n (%) \| No, n (%) \| P-value \| Yes, n (%) \| No, n (%) \| P-value \| \| Age \| Age \| Age \| Age \| Age \| Age \| Age \| Age \| Age \| Age \| \| 0-24 months \| 193 (23.1) \| 641 (76.9) \| 0.12 \| 211 (25.3) \| 623 (74.7) \| 0.04* \| 174 (20.9) \| 660 (79.1) \| <0.01* \| \| >24 months \| 134 (26.9) \| 364 (73.1) \| 0.12 \| 151 (30.3) \| 347 (69.7) \| 0.04* \| 63 (12.7) \| 435 (87.3) \| <0.01* \| \| Gender \| Gender \| Gender \| Gender \| Gender \| Gender \| Gender \| Gender \| Gender \| Gender \| \| Male \| 186 (26.1) \| 528 (73.9) \| 0.17 \| 195 (27.3) \| 519 (72.7) \| 0.90 \| 133 (18.6) \| 581 (814) \| 0.39 \| \| Female \| 141 (22.8) \| 477 (77.2) \| 0.17 \| 167 (27) \| 451 (73) \| 0.90 \| 104 (16.8) \| 514 (83.2) \| \| ## Discussion In the current study, the prevalence of underweight, stunting, and wasting was $24.5\%$, $27.2\%$, and $17.8\%$, respectively. These findings are comparable to the national prevalence of $32.1\%$, $35.5\%$, and $19.3\%$ of underweight, stunted, and wasted, respectively, as per the NFHS-5 data [13]. As per the NFHS-5 data, the prevalence of underweight and stunting for Delhi was $21.2\%$ and $30\%$, respectively; these are comparable to the prevalence rates of the current study. However, wasting was prevalent among $17.8\%$ of the participants in the current study, which is higher than the $11.2\%$ reported by NFHS-5 data in Delhi [13]. The current study also used the CIAF to evaluate the nutritional status of participants. The prevalence of undernutrition, as per the CIAF, was $45\%$. Almost half of the study participants had some form of anthropometric failure, whereas $55\%$ experienced no anthropometric failure (Table 4). The current study's findings are comparable with the study conducted by Ramkumar et al. [ 2018] in Puducherry, wherein the prevalence of CIAF was $48.7\%$ [18]. However, the prevalence in the current study is higher than the $36.1\%$ and $32.7\%$ reported by Roy et al. [ 2014] and Dasgupta et al. [ 2014] from rural areas in West Bengal [19,20]. This difference can be attributed to the fact that the study settings (urban vs. rural) were distinct, and a different set of risk factors may have played a role and contributed to the observed difference from the present study. However, the prevalence of $45\%$ in the current study is much lower than the $62.5\%$ reported by Anwar et al. ( Varanasi) and $58.6\%$ reported by Dhok et al. in Nagpur [21,22]. It was also lower than the prevalence of $63.6\%$ found by Sen et al. in Darjeeling [23]. Damor et al. reported the prevalence of malnutrition to be $54\%$ among children aged between one and five years [24]. However, the difference can be attributed to the different methodologies used to classify malnourished, i.e., Gomez's classification in their study, which took into account only weight for age criteria. The individual criteria for undernutrition in the current study revealed the prevalence of underweight (WFA < -2SD), stunting (HFA < -2SD), and wasting (WFH < -2SD) to be $24.2\%$, $27.5\%$, and $17.4\%$, respectively (Tables 2, 3). The findings imply that the prevalence of undernutrition was substantially lower than that of CIAF when individual criteria were considered, i.e., $45\%$. Similar findings have been reported by prior research. Roy et al. [ 2014], in a rural area of West Bengal, reported CIAF to be $36.1\%$, and the prevalence of stunting, underweight, and wasting was $16.7\%$, $29.2\%$, and $22.2\%$, respectively [19]. Sen et al. ( West Bengal, 2012) found the CIAF to be $63.6\%$; however, individual parameters were $52\%$ (underweight), $43.3\%$ (stunting), and $21.5\%$ (wasting), which is lower when compared to the prevalence of CIAF [23]. Nandy et al. also revealed that the prevalence of underweight was $47.1\%$, while CIAF was $59.9\%$, which was much higher than the individual criteria [17]. Similarly, CIAF was $65.3\%$ for toddlers in a study by Seetharaman et al., and the prevalence of underweight was $46.6\%$, which is much lower than the CIAF [25]. The findings also suggest no significant difference existed in childhood undernutrition across genders. However, children in the younger age group (0-24 months) had a higher risk of being wasted, whereas a higher proportion of children >24 months in age were more likely to be stunted when compared to children in the younger age group. This finding is explained by the fact that wasting is a sign of acute malnutrition, whereas stunting represents chronic malnutrition, which takes time to exhibit symptoms. The CIAF is a comprehensive tool permitting the segregation of undernourished children into different subgroups for further analysis. From Table 4, we observe that $45\%$ of the children suffered from one or the other form of anthropometric failure. We can identify $25\%$ of the children from subgroups C, D, E, and Y by using low weight for age (underweight) as the only criterion for undernourishment; however, we miss the children in subgroups B and F who were stunted and wasted but not underweight in the current study. Therefore, $20\%$ of such children would be missed out as not being undernourished. Similarly, stunting misses groups B, C, and Y ($28\%$ of children), and wasting misses those children in groups E, F, and Y ($27\%$ of children). The present study had the following strengths: as the study was conducted under routine programmatic conditions, a large sample size was collected within a short period. Secondly, comprehensive standard operative procedures were established for undertaking anthropometric measurements reducing measurement bias. Both the CIAF and the WHO growth charts tools were used to assess nutritional status allowing for comparisons and inferences to be drawn simultaneously. All severely undernourished children were referred to the nearest nutritional rehabilitation centers, and follow-up visits were undertaken for the mild to moderately undernourished. Nutritional counseling sessions were conducted for all parents, along with line list preparations by Anganwadi workers (AWWs), Accredited Social Health Activists (ASHAs), and auxiliary nurses and midwives (ANMs). This paper has some limitations. Firstly, because the study was cross-sectional, causal inferences cannot be drawn. Second, the survey was limited to children from a single UPHC in a particular state; therefore, incorporating children from other primary health centers (PHCs) in different locations may yield different results. As a result, these results cannot be generalized to the entire state or country. Third, aside from gender and age, other potential risk factors for malnutrition were not captured. Fourth, the children were not followed up; thus, a longitudinal study may be more beneficial. Researchers verified that all anthropometric measures were precise to prevent measuring bias. Nonetheless, the current study's findings are consistent with previously published literature. ## Conclusions The prevalence of undernutrition was high among the study participants. The CIAF is a comprehensive tool for calculating the total number of undernourished children in a community in terms of a single composite number. The CIAF enables the early identification of children with numerous anthropometric failures by categorizing undernourished children into several different groups. Hence, it enables healthcare providers to prioritize and provide swift treatment and care to those who need it the most. This information can further enable healthcare workers, clinicians, planners, and policymakers better estimate the prevalent problem in the community allowing the implementation of need-specific reforms, interventions, and policies. More studies utilizing the CIAF are required among children from other areas of Delhi, as well as from other parts of India, to acquire a broader representation. These findings will not only allow us to compare the rates of three standard indicators of undernutrition with CIAF but will also aid in establishing CIAF's increased effectiveness and utilization. Future studies should also focus on a comprehensive exploration of risk factors contributing to the burden of malnourishment among this age group. ## References 1. World Health Organization. Malnutrition: key facts. (2022) 2. Maleta K. Undernutrition. Malawi Med J (2006) 18 189-205. PMID: 27529011 3. World Health Organization. Children: improving survival and well-being. (2022) 4. UNICEF. Childhood diseases. (2022) 5. Walson JL, Berkley JA. The impact of malnutrition on childhood infections. Curr Opin Infect Dis (2018) 31 231-236. PMID: 29570495 6. De Sanctis V, Soliman A, Alaaraj N, Ahmed S, Alyafei F, Hamed N. Early and Long-term consequences of nutritional stunting: from childhood to adulthood. Acta Biomed (2021) 92 0 7. Roseboom T, de Rooij S, Painter R. The Dutch famine and its long-term consequences for adult health. Early Hum Dev (2006) 82 485-491. PMID: 16876341 8. Belkacemi L, Nelson DM, Desai M, Ross MG. Maternal undernutrition influences placental-fetal development. Biol Reprod (2010) 83 325-331. PMID: 20445129 9. de Onis M, Dewey KG, Borghi E. The World Health Organization's global target for reducing childhood stunting by 2025: rationale and proposed actions. Matern Child Nutr (2013) 9 6-26 10. United Nations System Standing Committee on Nutrition (UNSCN). United Nations System Standing Committee on Nutrition. UNSCN Discussion Paper - By 2030, end all forms of malnutrition and leave no one behind. (2017) 1-32 11. Romieu I, Dossus L, Barquera S. Energy balance and obesity: what are the main drivers?. Cancer Causes Control (2017) 28 247-258. PMID: 28210884 12. Victora CG, Christian P, Vidaletti LP, Gatica-Domínguez G, Menon P, Black RE. Revisiting maternal and child undernutrition in low-income and middle-income countries: variable progress towards an unfinished agenda. Lancet (2021) 397 1388-1399. PMID: 33691094 13. National Family Health Survey (NFHS-5), 2019-21. Minist Heal Fam Welf Natl (2022) 361 2 14. Singh A. Childhood malnutrition in India. Perspective of Recent Advances in Acute Diarrhea (2020) 15. WHO Multicentre Growth Reference Study Group. WHO child growth standards based on length/height, weight and age. Acta Paediatr Suppl (2006) 450 76-85. PMID: 16817681 16. Svedberg P. Poverty and Undernutrition: Theory, Measurement, and Policy. (2000) 17. Nandy S, Irving M, Gordon D, Subramanian SV, Smith GD. Poverty, child undernutrition and morbidity: new evidence from India. Bull World Health Organ (2005) 83 210-216. PMID: 15798845 18. Ramkumar S, Vijayalakshmi S, Kanagarajan P, Patil R, Lokeshmaran A. Z-score and CIAF-a descriptive measure to determine prevalence of under-nutrition in rural school children, Puducherry, India. J Clin Diagnostic Res (2018) 12 24-27 19. Roy K, Dasgupta A, Roychoudhury N, Bandyopadhyay L, Mandal S, Paul B. Assessment of under nutrition with Composite Index of Anthropometric Failure (CIAF) among under-five children in a rural area of West Bengal, India. Int J Contemp Pediatr (2018) 5 1651 20. Dasgupta A, Parthasarathi R, Prabhakar VR, Biswas R, Geethanjali A. Assessment of under nutrition with Composite Index of Anthropometric Failure (CIAF) among under-five children in a rural area of West Bengal. Ind J Comm Health (2014) 26 132-138 21. Anwar F, Gupta MK, Prabha C, Srivastava R. Malnutrition among rural Indian children: an assessment using web of indices. Int J Public Health Epidemiol (2013) 2 78-84 22. Dhok RS, Thakre SB. Measuring undernutrition by Composite Index of Anthropometric Failure (CIAF): a community-based study in a slum of Nagpur city. Int J Med Sci Public Health (2016) 5 2013-2018 23. Sen J, Mondal N. Socio-economic and demographic factors affecting the Composite Index of Anthropometric Failure (CIAF). Ann Hum Biol (2012) 39 129-136. PMID: 22324839 24. Damor RD, Pradeep P, Lodhiya KK, Mehta JP. A study on assessment of nutritional and immunization status of under five children in urban slums of Jamnagar city, Gujarat. Heal J Indian Assoc Prev Soc Med (2013) 4 35-39 25. Seetharaman N, Chacko TV, Shankar SLR, Mathew AC. Measuring malnutrition -the role of Z scores and the Composite Index of Anthropometric Failure (CIAF). Indian J Community Med (2007) 32 35-39
--- title: Enhanced Endosomal Signaling and Desensitization of GLP-1R vs GIPR in Pancreatic Beta Cells authors: - Yusman Manchanda - Stavroula Bitsi - Shiqian Chen - Johannes Broichhagen - Jorge Bernardino de la Serna - Ben Jones - Alejandra Tomas journal: Endocrinology year: 2023 pmcid: PMC10016038 doi: 10.1210/endocr/bqad028 license: CC BY 4.0 --- # Enhanced Endosomal Signaling and Desensitization of GLP-1R vs GIPR in Pancreatic Beta Cells ## Body Incretin receptors, comprising the glucagon-like peptide-1 receptor (GLP-1R) and the glucose-dependent insulinotropic polypeptide receptor (GIPR), are key components of the glucoregulatory system due to their capacity to prevent postprandial hyperglycemia by amplifying insulin secretion from pancreatic beta cells in a glucose-dependent manner [1]. Since their discovery and cloning in the 1990s [2-4], both receptors have been recognized for their glucose lowering potential. However, while the GLP-1R has successfully been exploited for the treatment of type 2 diabetes (T2D), with several pharmacological GLP-1R agonists currently in use clinically or undergoing clinical trials [5], the GIPR has not until recently been intensively pursued as a T2D treatment target, primarily due to GIP responses being blunted in T2D patients [6] and the perception that GIPR activation leads to weight gain, as inferred from the observation that GIPR knockout (KO) mice are protected against the effects of an obesogenic diet [7]. As a result, antagonizing rather than activating the GIPR has been suggested as a potential therapeutic intervention for diabetes and obesity [8, 9]. However, recent data from preclinical and clinical studies appear to contradict these assumptions regarding the role of GIPR in diabetes, as GIPR agonists have been shown to improve glucose tolerance and reduce body weight in T2D patients [10], and dual GLP-1R/GIPR targeting peptides, such as the recently developed tirzepatide, have demonstrated enhanced efficacy compared with currently approved GLP-1R agonist monotherapies [11]. Previous studies from our group and others have demonstrated that changes in the spatiotemporal regulation of signaling play a crucial role in determining the metabolic outcomes of GLP-1R activation [12] and underlie the enhanced therapeutic effects of partial or biased GLP-1R agonists [13]. However, neither the GIPR trafficking nor the spatiotemporal regulation of signaling by active GIPRs have been well characterized, despite the potential importance of these processes in the paradoxical responses obtained with both GIPR agonists and antagonists. GIPR has seldom been compared directly to GLP-1R in pancreatic beta cell systems, the primary cell type in which these related receptors coexist and exert many of their metabolic effects. In whole islets, both GLP-1R and GIPR stimulate insulin secretion in a glucose-dependent manner, but only GLP-1R retains its insulinotropic action in islets from diabetic models, implying that glucotoxicity specifically impairs GIP-dependent action in beta cells [14]. However, the molecular mechanisms that explain the preservation of insulinotropic actions of GLP-1R but not GIPR in T2D remain unclear, with some suggestions that include reduced GIPR expression [15], increased GIPR degradation [16], and differences in Gαs vs Gαq coupling for each receptor [17]. Of note, GLP-1R and GIPR potentiation of insulin secretion seems to have different dependency on KATP channels [18, 19], suggesting differences in downstream signaling between the 2 incretin receptors. In the present study, we present a dataset describing the effects of GLP-1R vs GIPR activation on target downregulation and compartmentalization of intracellular signaling responses specifically in pancreatic beta cells, unveiling striking differences between the trafficking and signaling signatures from the 2 incretin receptors within their native environment. In particular, we demonstrate that the beta cell GIPR is a slow-internalizing, fast-recycling receptor compared with the GLP-1R. This trafficking pattern is accompanied by a reduced capacity for clustering, endosomal vs plasma membrane signaling, lysosomal targeting, and degradation of GIPR in response to agonist stimulation, as well as significantly reduced GIPR vs GLP-1R coupling to downstream effectors such as Gαs, Gαq, and β-arrestin 2. Paradoxically, however, and despite notably reduced surface levels of GIPR vs GLP-1R in primary islets from healthy mice, GIPR stimulation leads to similar or even enhanced signaling outputs, suggesting a greater degree of signal amplification and reduced desensitization associated with this incretin receptor under nondiabetic conditions. ## Abstract The incretin receptors, glucagon-like peptide-1 receptor (GLP-1R) and glucose-dependent insulinotropic polypeptide receptor (GIPR), are prime therapeutic targets for the treatment of type 2 diabetes (T2D) and obesity. They are expressed in pancreatic beta cells where they potentiate insulin release in response to food intake. Despite GIP being the main incretin in healthy individuals, GLP-1R has been favored as a therapeutic target due to blunted GIPR responses in T2D patients and conflicting effects of GIPR agonists and antagonists in improving glucose tolerance and preventing weight gain. There is, however, a recently renewed interest in GIPR biology, following the realization that GIPR responses can be restored after an initial period of blood glucose normalization and the recent development of dual GLP-1R/GIPR agonists with superior capacity for controlling blood glucose levels and weight. The importance of GLP-1R trafficking and subcellular signaling in the control of receptor outputs is well established, but little is known about the pattern of spatiotemporal signaling from the GIPR in beta cells. Here, we have directly compared surface expression, trafficking, and signaling characteristics of both incretin receptors in pancreatic beta cells to identify potential differences that might underlie distinct pharmacological responses associated with each receptor. Our results indicate increased cell surface levels, internalization, degradation, and endosomal vs plasma membrane activity for the GLP-1R, while the GIPR is instead associated with increased plasma membrane recycling, reduced desensitization, and enhanced downstream signal amplification. These differences might have potential implications for the capacity of each incretin receptor to control beta cell function. ## Peptides Native sequence peptides including GLP-1[7-36]NH2 (referred to as GLP-1) and GIP[1-42] (referred to as GIP), and their fluorescein isothiocyanate (FITC) and tetramethylrhodamine (TMR) conjugates were obtained from Wuxi Apptec at >$90\%$ purity. ## Cell Culture Parental male rat insulinoma INS-1 $\frac{832}{3}$ cells (a gift from Prof. Christopher Newgard, Duke University, USA), INS-1 $\frac{832}{3}$ cells with endogenous GLP-1R or GIPR deleted by CRISPR/Cas9 [20] (a gift from Dr. Jacqueline Naylor, MedImmune), and corresponding multiclonal INS-1 $\frac{832}{3}$ cells stably expressing SNAP-GLP-1R or SNAP-GIPR (generated by transfecting GLP-1R KO or GIPR KO cells with SNAP-GLP-1R or SNAP-GIPR constructs (Cisbio), respectively, followed by selection with 1 mg/mL G418, FACS sorting of the population of SNAP-receptor-expressing cells and maintenance in 0.5 mg/mL G418) were cultured in RPMI-1640 with 11 mM D-glucose, supplemented with $10\%$ FBS, 10 mM HEPES, 1 mM sodium pyruvate, 50 μM β-mercaptoethanol, and $1\%$ penicillin/streptomycin in a 37 °C/$5\%$ CO2 incubator. ## Diffusion-Enhanced Resonance Energy Transfer Internalization Assays The diffusion-enhanced resonance energy transfer (DERET) assay was performed as previously described [21]. The INS-1 $\frac{832}{3}$ SNAP-GLP-1R or SNAP-GIPR cells were labeled in suspension with Lumi4-Tb (40 nM) for 30 minutes in complete media. After washing, cells were resuspended in HBSS containing 24 µM fluorescein and dispensed into 96-well white plates. A baseline read was serially recorded over 5 minutes using a Flexstation 3 instrument at 37 °C in time-resolved fluorescence resonance energy transfer (TR-FRET) mode using the following settings: λex 340 nm, λem 520 and 620 nm, auto-cutoff, delay 400 µs, integration time 1500 µs. Ligands were then added, after which signal was repeatedly recorded for 30 minutes. Fluorescence signals were expressed ratiometrically after first subtracting signal from wells containing 24 µM fluorescein but no cells. Internalization was quantified as area under the curve (AUC) relative to individual well baseline. ## High-content Microscopy Assays for Receptor Internalization and Recycling The assay was performed as previously described [21]. The INS-1 $\frac{832}{3}$ SNAP-GLP-1R or SNAP-GIPR cells were seeded into poly-D-lysine-coated, black 96-well plates. On the day of the assay, labeling was performed with BG-S-S-649 (1 µM), a surface-labeling SNAP-tag probe that can be released on application of reducing agents such as Mesna. After washing, treatments were applied for 30 minutes at 37 °C in complete medium. Ligand was removed and cells washed with cold HBSS and placed on ice for subsequent steps. Mesna (100 mM in alkaline TNE buffer, pH 8.6) or alkaline TNE buffer without Mesna was applied for 5 minutes, and then washed with HBSS. Cells were imaged by widefield microscopy, with both epifluorescence and transmitted phase contrast images acquired. On imaging completion, HBSS was removed and replaced with fresh complete medium, and receptor was allowed to recycle for 60 minutes at 37 °C, followed by a second Mesna application to remove any receptor that had recycled to the plasma membrane, with the plate re-imaged as above. Internalized receptor at each time point was determined from cell-containing regions as determined from the phase contrast image using PHANTAST [22] and used to determine internalization and recycling parameters as previously described [21]. ## Mini-G protein/β-arrestin-2 recruitment NanoBiT assays Here the SmBiT was cloned in frame at the C-terminus of the GLP-1R and the GIPR by substitution of the Tango sequence on FLAG-tagged GLP-1R-Tango or GIPR-Tango (a gift from Prof. Bryan Roth, University of North Carolina, USA; Addgene plasmids #66291 and #66294), respectively. Mini-Gs, mini-Gq, and mini-Gi plasmids, tagged at the N-terminus with LgBiT, were a gift from Prof. Nevin Lambert, Medical College of Georgia, USA. For β-arrestin 2 recruitment assays, β-arrestin 2 fused at the N-terminus to LgBiT (LgBiT-β-arrestin 2; Promega, plasmid no. CS1603B118) was chosen as it has previously been used successfully with other class B GPCRs. The INS-1 $\frac{832}{3}$ GLP-1R KO and GIPR KO cells were seeded in 12-well plates and co-transfected with 0.5 μg each of GLP-1R-SmBiT or GIPR-SmBiT and either LgBiT-mini-Gs, -mini-Gq, -mini-Gi or -β-arrestin 2. ## KRAS/Rab5 bystander NanoBRET assays GLP-1R-NanoLuc was generated in house by polymerase chain reaction (PCR) cloning of the NanoLuciferase sequence from pcDNA3.1-ccdB-NanoLuc (a gift from Prof. Mikko Taipale; Addgene plasmid # 87067) onto the C-terminus end of the SNAP-GLP-1R vector (CisBio), followed by site-directed mutagenesis of the GLP-1R stop codon. GIPR-NanoLuc was subsequently cloned in house by exchanging the GLP-1R for the GIPR in the GLP-1R-NanoLuc construct. KRAS- and Rab5-Venus plasmids were a gift from Prof. Kevin Pfleger, University of Western Australia. The INS-1 $\frac{832}{3}$ GLP-1R KO and GIPR KO cells were seeded in 12-well plates and co-transfected with 0.2 µg KRAS-Venus and 0.1 µg GLP-1R- or GIPR-NanoLuc, respectively, or 0.5 µg Rab5-Venus and 0.1 µg GLP-1R- or GIPR-NanoLuc, respectively. ## Mini-Gs-Venus recruitment NanoBRET assays Mini-Gs-Venus was a gift from Prof. Nevin Lambert, Augusta University, USA. INS-1 $\frac{832}{3}$ GLP-1R KO or GIPR KO cells were seeded in 12-well plates and co-transfected with 0.5 µg mini-Gs-Venus and either 0.5 µg GLP-1R- or GIPR-NanoLuc, respectively. ## Nb37 bystander NanoBiT assays The Nb37 assay constructs were kindly provided by Prof. Asuka Inoue, Tohoku University, Japan. Nb37 (gene synthesized by GenScript with codon optimization) was C-terminally fused to SmBiT with a 15 amino acid flexible linker (GGSGGGGSGGSSSGGG), and the resulting construct referred to as Nb37-SmBiT. The C-terminal KRAS CAAX motif (SSSGGGKKKKKKSKTKCVIM) was N-terminally fused with LgBiT (LgBiT-CAAX). The Endofin FYVE domain (amino acid region Gln739-Lys806) was C-terminally fused with LgBiT (Endofin-LgBiT). Gαs (human, short isoform), Gβ1 (human), Gγ2 (human), and RIC8B (human, isoform 2) plasmids were inserted into pcDNA3.1 or pCAGGS expression plasmid vectors. INS-1 $\frac{832}{3}$ GLP-1R KO or GIPR KO cells were seeded in 6-well plates and co-transfected with 0.1 μg SNAP-GLP-1R or SNAP-GIPR, 0.5 μg Gαs, Gβ1, and Gγ2, 0.1 μg RIC8B, 0.1 μg CAAX-LgBiT or 0.5 μg Endofin-LgBiT with 0.1 μg or 0.5 μg Nb37-SmBiT, respectively, with 0.8 µg pcDNA3.1 added to the former to equalize DNA content. All NanoBiT and NanoBRET readings were obtained in a Flexstation 3 plate reader. Briefly, 24 hours after transfection, cells were detached, resuspended in NanoGlo Live Cell Reagent (Promega) with furimazine (1:20 dilution) and seeded into white 96-well half-area plates. For NanoBiTs, baseline luminescence was recorded for 5 minutes at 37 °C followed by 30 minutes with or without addition of GLP-1 or GIP at 100 nM for G protein and β-arrestin 2 recruitment assays, and at serial doses of up to 1 μM for the Nb37 bystander assays; readings were taken every 30 seconds or every minute, respectively. For NanoBRETs, baseline luminescent signals were recorded every minute at 460 nm (NanoLuc emission peak) and 535 nm (Venus emission peak) over 5 minutes at 37 °C, followed by 30 minutes with or without the addition of 100 nM GLP-1 or GIP. Readings were normalized to well baseline and then to average vehicle-induced signal to establish the agonist-induced effect. AUCs from response curves were calculated for each agonist concentration and fitted to four-parameter curves using Prism 9 (GraphPad). ## Transfections Transient transfection of plasmids was performed using Lipofectamine 2000 (Thermo Fisher) according to the manufacturer's instructions. Experiments were performed 24 hours after transfection unless otherwise indicated. ## High-content microscopy assay The assay was adapted from a previous description [23]. The INS-1 $\frac{832}{3}$ SNAP-GLP-1R or SNAP-GIPR cells were seeded in complete medium in poly-D-lysine-coated black, clear-bottom plates. Once attached, cells were washed twice in PBS and incubated in fresh serum-free medium containing cycloheximide (50 µg/mL) to arrest protein translation. After 2 hours, agonists were added in reverse time order (the longest time point being 8 hours), with the medium replaced for the final 30 minutes of the experiment with complete medium containing 1 µM BG-OG to label total residual SNAP-GLP-1R or SNAP-GIPR. Wells were then washed 3X in HBSS and the microplate imaged by widefield microscopy, with quantification of total cellular receptor at each time point from segmented cell-containing regions as for the high-content internalization and recycling assays described above. ## Degradation assays by immunoblotting INS-1 $\frac{832}{3}$ SNAP-GLP-1R and SNAP-GIPR cells were seeded in 6-well plates (1.5 million per well) and cultured overnight prior to incubation in serum-free medium containing cycloheximide (50 µg/mL) for 2 hours. Cells were then incubated with or without 100 nM GLP-1 or GIP for 6 hours before being lysed in 1X TNE lysis buffer (20 mM Tris, 150 mM NaCl, 1 mM EDTA, $1\%$ NP40, protease and phosphatase inhibitor cocktails) for 10 minutes at 4 °C followed by cell scraping and sonication (3X, 10 seconds each). The lysates were then frozen at −80 °C for 2 minutes, thawed, and centrifuged at 15 000 g for 10 minutes at 4 °C. The supernatants were collected, fractionated by SDS-PAGE in urea loading buffer (200 mM Tris HCl pH 6.8, $5\%$ w/v SDS, 8 M urea, 100 mM DTT, $0.02\%$ w/v bromophenol blue) and analyzed by Western blotting. SNAP-GLP-1R and SNAP-GIPR were detected with an anti-SNAP-tag rabbit polyclonal antibody (P9310S, New England Biolabs, RRID: AB_10631145, $\frac{1}{1000}$) followed by goat anti-rabbit HRP secondary (ab6721, Abcam, RRID: AB_955447, $\frac{1}{2000}$). Post-stripping, tubulin was labeled with anti-α-tubulin mouse monoclonal antibody (T5168, Sigma, RRID: AB_477579, $\frac{1}{5000}$) followed by sheep anti-mouse HRP secondary antibody (ab6808, Abcam, RRID: AB_955441, $\frac{1}{5000}$). Blots were developed with the Clarity Western enhanced chemiluminescence (ECL) substrate system (BioRad) in a Xograph Compact X5 processor and specific band densities quantified in Fiji. ## Measurement of Receptor Clustering by Time-Resolved Fluorescence Resonance Energy Transfer The time-resolved fluorescence resonance energy transfer (TR-FRET) assay was performed as previously described [24]. INS-1 $\frac{832}{3}$ SNAP-GLP-1R or SNAP-GIPR cells were labeled in suspension with 40 nM SNAP-Lumi4-Tb and 1 mM SNAP-Surface 649 (New England Biolabs, Hitchin, UK) for 1 hour at room temperature in complete medium. After washing, cells were resuspended in HBSS, and TR-FRET was monitored before and after addition of 100 nM GLP-1, GIP, or a mixture of GLP-1 and GIP at 37 °C in a Spectramax i3x plate reader in homogeneous time-resolved fluorescence (HTRF) mode. TR-FRET was quantified as the ratio of fluorescent signal at 665 nm to that at 616 nm, after subtraction of background signal at each wavelength. ## Raster Image Correlation Spectroscopy INS-1 $\frac{832}{3}$ SNAP-GLP-1R or SNAP-GIPR cells were seeded onto glass bottom MatTek dishes and surface-labeled with SNAP-Surface 488 (1 mM, 30 minutes at 37 °C). After washing, cells were imaged at the basal plasma membrane in HBSS with 10 mM HEPES at 37 °C either before or 5 minutes after stimulation with 100 nM GLP-1 or GIP, respectively. Time-lapse images of cells were acquired in a Zeiss LSM-780 inverted confocal microscope fitted with a 63x/1.2 NA water immersion objective. SNAP-Surface 488 was excited by a continuous wavelength laser at 488 nm and emission signal collected at 500 to 580 nm. The pinhole was set to one Airy unit. Optimized acquisition was performed to retrieve protein membrane diffusion values as described previously [25, 26]. Images of 256 × 256 pixels at 8-bit depth were collected using 80 nm pixel size and 5 μsec dwell time, for 250 consecutive frames. To characterize the waist of the point spread function (PSF), 200 frames of freely diffusing recombinant EGFP (20 mM) were continuously collected, as described elsewhere [27, 28]. Analysis was performed on images where intensity traces were not decreased continuously by $20\%$ or more over 50 frames to avoid possible bleaching artifacts that would interfere in diffusion coefficient measurements. A moving average (background subtraction) of 10 was applied, so that artifacts due to cellular motion or very slow-moving particles were avoided. The obtained 2-dimensional (2D) autocorrelation map was then fitted, and a surface map obtained with the characterized PSF and the appropriate acquisition values for line time and pixel time. Three different regions of interest (ROI) were analyzed within the same cell, with the corresponding regions drawn employing a 64 × 64-pixel square. Raster image correlation spectroscopy (RICS) analysis was performed using the “SimFCS 4” software (Global Software, G-SOFT Inc., Champaign, IL) as described [29]. RICS analysis was performed in ROIs of 64 × 64 pixels at 4 random cytoplasmic areas per cell using a moving average (background subtraction) of 10 to discard possible artifacts due to cellular motion and slow-moving particles passing through. The autocorrelation 2D map was then fitted to obtain a surface map that was represented as a 3D projection with the residuals on top. As a rule, we focused on those regions with intensity fluctuation events in which the intensity changes were following short increasing or decreasing steps, avoiding abrupt intensity decays or increases. ## Cyclic AMP Homogeneous Time-Resolved Fluorescence Assays INS-1 $\frac{832}{3}$ cells were stimulated with increasing concentrations of GLP-1 or GIP followed by lysis and cyclic AMP (cAMP) homogeneous time-resolved fluorescence (HTRF) immunoassay (cAMP Dynamic 2, 62AM4PEB, Cisbio, Codolet, France) according to the manufacturer's instructions. Results were expressed as basal fold increase responses and fitted to 3-parameter curves using Prism 9 (GraphPad). ## Isolation and Culture of Pancreatic Islets Nondiabetic mice of both sexes were used for islet isolation. Briefly, pancreata were infused via the common bile duct with RPMI-1640 medium containing 1 mg/mL collagenase from Clostridium histolyticum (Nordmark Biochemicals), dissected, and incubated in a water bath at 37 °C for 10 minutes. Islets were subsequently washed and purified using a Histopaque gradient (Histopaque-1119, 11191, Sigma-Aldrich, and Histopaque-1083, 10831, Sigma-Aldrich). Isolated islets were allowed to recover overnight at 37 °C in $5\%$ CO2 in RPMI-1640 supplemented with $10\%$ FBS and $1\%$ penicillin/streptomycin. All procedures were carried out in accordance with the regulations of the UK Home Office Animals (Scientific Procedures) Act and the Imperial College London guidelines for animal care. Animal protocols were approved by the Home Office Animals in Science Regulation Unit (ASRU) under Project License number PP7151519 to Dr. A. Martinez-Sanchez. ## cAMP FRET Assays CAMPER reporter mice [30], with conditional expression of the cAMP fluorescence resonance energy transfer (FRET) biosensor TEPACVV [31], were purchased from Jackson Laboratory (Stock No: 032205) and crossed with Pdx1-CreERT mice (in house) to generate mice with inducible TEPACVV expression from pancreatic beta cells, used to isolate islets for ex vivo cAMP FRET assays. Isolated islets were treated overnight with 4-hydroxytamoxifen to induce biosensor expression prior to Matrigel encasing on MatTek glass bottom dishes and imaging by FRET between CFP (donor) and YFP (acceptor) with CFP excitation and both CFP and YFP emission settings in a Zeiss LSM-780 inverted confocal laser-scanning microscope and a 20X objective to capture time-lapse recordings with image acquisition every 6 seconds, and treatments manually added by pipetting. Specifically, islets were imaged in Krebs-Ringer bicarbonate-HEPES (KRBH) buffer (140 mM NaCl, 3.6 mM KCl, 1.5 mM CaCl2, 0.5 mM MgSO4, 0.5 mM NaH2PO4, 2 mM NaHCO3, 10 mM HEPES, saturated with $95\%$ O2/$5\%$ CO2; pH 7.4) containing $0.1\%$ w/v bovine serum albumin (BSA) and 6 mM glucose (KRBH G6) for 1 minute, then agonist at 100 nM was added and imaged for 10 minutes before addition of 10 μM forskolin + 100 μM isobutyl methylxanthine (IBMX) for the final 2 minutes of the acquisition to record maximal responses. Raw intensity traces for YFP and CFP fluorescence were extracted from whole islet ROIs using Fiji and YFP/CFP ratios calculated for each ROI and time point. Responses were plotted relative to the average fluorescence intensity per islet during the 6 mM glucose baseline period, before agonist addition. ## Calcium Assays Imaging of INS-1 $\frac{832}{3}$ cells or whole-islet Ca2+ dynamics was performed as follows: cells or Matrigel-encased islets from individual animals were loaded with the Ca2+ responsive dye Cal-520 Am (AAT Bioquest), pre-incubated for 1 hour in KRBH G6, and imaged in MatTek glass bottom dishes every 6 second at 488 nm using a Nikon Eclipse Ti microscope with an ORCA-Flash 4.0 camera (Hamamatsu) and Metamorph software (Molecular Devices) while maintained at 37 °C on a heated stage. Raw fluorescence intensity traces from cell-occupied areas or islet ROIs were extracted using Fiji. Responses were plotted relative to the average fluorescence intensity during the 6 mM glucose baseline period, before agonist addition. ## Insulin Secretion Assays INS-1 $\frac{832}{3}$ cells were seeded in a 48-well plate and incubated in 3 mM glucose in full medium overnight before incubation with 11 mM glucose ± GLP-1/GIP at 100 nM in KRBH buffer containing $0.1\%$ w/v BSA at 37 °C. At the end of the treatments, the supernatant containing the secreted insulin was collected, centrifuged at 1000g for 3 minutes, and transferred to a fresh tube. To determine total insulin content, cells were lysed using KRBH buffer + $1\%$ w/v BSA + $1\%$ v/v Triton X-100 (Sigma). The lysates were sonicated 3 × 10 seconds in a water bath sonicator and centrifuged at 10 000g for 10 minutes, and the supernatants collected. The samples were stored at −20 °C until the insulin concentration was determined using an Insulin Ultra-Sensitive HTRF Assay kit (62IN2PEG, Cisbio, Codolet, France) according to the manufacturer's instructions. ## Statistical Analyses All data analyses and graph generation were performed with GraphPad Prism 9.0. The statistical tests used are indicated in the corresponding figure legends. The number of replicates for comparisons represents biological replicates. Technical replicates within biological replicates were averaged prior to statistical tests. Data are represented as mean ± SEM. The P value threshold for statistical significance was set at.05. ## Results We first analyzed the trafficking characteristics of both receptors following stimulation with their cognate full-length endogenous agonists, GLP-1[7-36]NH2 and GIP[1-42], using rat INS-1 $\frac{832}{3}$ beta cells in which the endogenous incretin receptor was deleted and the equivalent SNAP-tagged human receptor exogenously expressed (INS-1 $\frac{832}{3}$ SNAP-GLP-1R or SNAP-GIPR cells). Note that surface expression levels of SNAP-GLP-1R and SNAP-GIPR were similar within these 2 cell models (Supplemental Fig. 1A) [32]. DERET assays, which detect disappearance of the receptor by a loss of TR-FRET signal between the receptor extracellular domain (ECD) and the extracellular buffer [24], revealed stark differences in the degree of internalization between the 2 receptors following stimulation with their native agonists, with the GLP-1R achieving approximately 3 times more internalization in the first hour poststimulation with 100 nM GLP-1 compared with the GIPR for the same stimulation period with 100 nM GIP (Supplemental Fig. 1B and 1C) [32]. Notably, there was negligible GLP-1R internalization in response to GIP, and vice versa, and no significant change to internalization for either receptor when using both agonists combined. Greater internalization of GLP-1R than of GIPR was observed across a wide concentration range (Fig. 1A and Supplemental Fig. 1D) [32]. Analysis of the rate of change of DERET signal indicated that GLP-1R endocytosis was significantly faster (Fig. 1B). We corroborated these findings by high-content microscopy analysis of receptor internalization in the same cells (Fig. 1C), with significantly less internalization of GIPR compared with GLP-1R when stimulated with their respective endogenous agonists. Figure 1.: Beta cell GLP-1R vs GIPR trafficking patterns. (A) Internalization AUC dose response curves from DERET assays in INS-1 832/3 SNAP-GLP-1R vs SNAP-GIPR cells stimulated with the indicated concentrations of GLP-1 or GIP, respectively. Results were fitted to a 3-parameter dose response curve to obtain Emax and logEC50 for both receptors, with comparisons between these parameters included; n = 5. (B) Rates of GLP-1R vs GIPR internalization (k values) derived from (A) by one-phase association (with Y0 = 0) of baseline-deleted DERET data; n = 5. (C) GLP-1R vs GIPR internalization dose response curves measured by high-content microscopy assay (HCA) in the same cells as above. Results fitted and Emax and logEC50 comparisons included as above; n = 5. (D) Confocal microscopy analysis of SNAP-GLP-1R vs SNAP-GIPR (red, middle panels) localization following 30 minutes stimulation with 100 nM GLP-1-FITC or GIP-FITC (green, left panels), respectively, in the same cells as above. (E) Confocal microscopy analysis of isolated intact mouse islets stimulated with fluorescently labeled agonists as indicated: WT islets were imaged following stimulation with 100 nM GLP-1-TMR, while both WT and GIPR−/− (KO) islets were imaged following stimulation with 1 µM GIP-TMR. (F) Quantification of surface GLP-1R vs GIPR levels in WT mouse islets using fluorescent agonist uptake data from (E); data corrected for binding affinity differences between both agonists; n = 4. (G) GLP-1R vs GIPR recycling dose response curves measured by high-content microscopy assay (HCA) in cells from (C); n = 5. (H) Confocal microscopy analysis of SNAP-GLP-1R vs SNAP-GIPR (red, middle panels) colocalization with SNX27-GFP (green, left panels) following 3 hours stimulation with 100 nM GLP-1 or GIP, respectively, in the same cells as above. Nuclei (DAPI), blue. Data are mean ± SEM, compared by paired or unpaired t tests, or two-way ANOVA with Sidak's post hoc test; P < .05, P < .01, P < .001; size bars: 10 µm. In concordance with these results, 30 minutes of stimulation with the fluorescently labeled agonist GLP-1-FITC [33] resulted in the near complete co-internalization of SNAP-GLP-1R and fluorescent ligand, while the equivalent treatment with GIP-FITC led to only partial SNAP-GIPR endocytosis, with the receptor still clearly visible at the plasma membrane (Fig. 1D and Supplemental Fig. 1E) [32]. Moreover, when both receptors harboring different N-terminal tags (so that they could be differentially labeled) were expressed together in wild-type (WT) INS-1 $\frac{832}{3}$ cells, HALO-GLP-1R did again show faster internalization vs SNAP-GIPR following stimulation with a mixture of GLP-1 and GIP (Supplemental Fig. 2) [32]. Finally, using alternative fluorescent conjugates labeled with TMR, the amount of TMR-labeled agonist intracellular accumulation during the first 5 minutes of stimulation for each receptor correlated with the previously shown receptor internalization results in INS-1 $\frac{832}{3}$ SNAP-GLP-1R/SNAP-GIPR cell lines (Supplemental Fig. 3A and 3B) [32]. Consistently, in WT mouse primary islets, GIP-TMR could be detected at the plasma membrane, while GLP-1-TMR was predominantly localized in punctate structures reminiscent of endosomes after 15 minutes of agonist stimulation (Fig. 1E). As a control, GIP-TMR signal at the plasma membrane was absent in islets from Gipr−/− (KO) mice [34] labeled in parallel with the same concentration of GIP-TMR, demonstrating specificity of labeling in WT islets. Of note, signal was significantly lower for GIP- vs GLP-1-TMR (quantified in Fig. 1F), with GIP-TMR signal virtually undetectable in islets at a concentration of 100 nM (not shown). This result suggests reduced levels of endogenous GIPR vs GLP-1R in mouse islets and correlates with previously published RNAseq data indicating higher levels of beta cell Glp1r vs Gipr mRNA expression in those islets [35]. Additional experiments were performed using dispersed mouse islet cells, which also showed markedly reduced that GIP-TMR uptake compared with GLP-1-TMR (Supplemental Fig. 3C) [32]. We also analyzed the level of receptor recycling back to the plasma membrane after internalization in INS-1 $\frac{832}{3}$ SNAP-GLP-1R or SNAP-GIPR cells by high-content microscopy after stimulation with 100 nM GLP-1 or GIP, respectively, and detected significantly increased recycling rates for the GIPR compared with the GLP-1R (Fig. 1G), an observation that correlated with sustained SNAP-GIPR, but not SNAP-GLP-1R, colocalization with the recycling factor SNX27 [36] fused to EGFP (Fig. 1H). To determine the intracellular destination of internalized GLP-1Rs/GIPRs more precisely, we performed concentration response bystander NanoBRET assays using C-terminal NanoLuc-fused SNAP-tagged GLP-1R vs GIPR and either KRAS-Venus (plasma membrane) or Rab5-Venus (early endosome) co-expressed transiently in INS-1 $\frac{832}{3}$ cells (Fig. 2). In these experiments, we again observed increased propensity for plasma membrane retention of the GIPR compared with the GLP-1R, with reduced maximal internalization responses but no changes in potency (Fig. 2A). Agonist-mediated redistribution of GLP-1R to Rab5-positive early endosomes was clearly detectable, but virtually absent for the GIPR in this system (Fig. 2B). In broad agreement with these results, using stable INS-1-SNAP-GLP-1R or -GIPR cells, the majority of SNAP-GLP-1R signal could be detected in Rab5-Venus-positive endosomes after 10 minutes of stimulation with 100 nM GLP-1, while a sizable amount of SNAP-GIPR was still present at the plasma membrane following stimulation with 100 nM GIP; with the fraction of internalized GIPRs nevertheless also localized to Rab5-Venus-positive endosomes (Supplemental Fig. 4) [32]. Figure 2.: Endosomal vs plasma membrane localization of GLP-1R compared with GIPR in beta cells. (A) GLP-1R vs GIPR plasma membrane localization dose response curves from NanoBRET assays performed in INS-1 832/3 GLP-1R KO vs GIPR KO cells transiently expressing KRAS-Venus and GLP-1R- or GIPR-NanoLuc, after stimulation with the indicated concentrations of GLP-1 or GIP, respectively; results were fitted to 3-parameter dose response curves to obtain Emax and logEC50 for both receptors, with comparisons between these parameters included; n = 5. (B) As for (A) but for GLP-1R vs GIPR endosomal localization dose response curves from NanoBRET assays performed in INS-1 832/3 GLP-1R KO vs GIPR KO cells transiently expressing Rab5-Venus and GLP-1R- or GIPR-NanoLuc, after stimulation with the indicated concentrations of GLP-1 or GIP, respectively; n = 5. Data are mean ± SEM, compared by paired t tests or two-way ANOVA with Sidak's test; P < .05.* We have previously found that agonist-induced GLP-1R internalization is preceded by receptor clustering at the plasma membrane [24]. We therefore investigated whether the degree of clustering for each incretin receptor in a beta cell setting would reflect the differences observed in their internalization profiles (Fig. 3). Employing RICS [27], we observed receptor clustering tendencies in INS-1 $\frac{832}{3}$ SNAP-GLP-1R or SNAP-GIPR cells labeled with the SNAP-Surface 488 probe under vehicle conditions as well as after 5 minutes of stimulation with 100 nM GLP-1 or GIP, respectively (Fig. 3A). Quantification of receptor diffusion coefficients revealed that, while the GIPR exhibits slower basal diffusion, suggesting more clustering than the GLP-1R under vehicle conditions [a phenotype that correlates with our previously observed increased propensity for this receptor to segregate to cholesterol-rich lipid nanodomains under basal conditions [24]], agonist stimulation resulted in marked slowing of diffusion for both receptors (Fig. 3B). Additionally, TR-FRET experiments suggested increased GLP-1R clustering when stimulated with GLP-1 (Fig. 3C), an effect that could not be detected following GIP stimulation of the GIPR (Fig. 3D). Moreover, clustering was not detectable when inverting the cognate agonists and no significant increases were detected by co-application of both agonists for each of the INS-1 $\frac{832}{3}$ receptor cell models (Fig. 3C and 3D). Figure 3.: GLP-1R vs GIPR clustering propensities in beta cells. (A) Representative images from RICS analysis of GLP-1R vs GIPR clustering, showing SNAP-Surface 488-labeled GLP-1Rs or GIPRs imaged in the basolateral plane of INS-1 832/3 SNAP-GLP-1R vs SNAP-GIPR cells after 5 minutes treatment in vehicle (Veh) and either 100 nM GLP-1 or GIP. Diffusion coefficients for individual ROIs are indicated on each image, with corresponding intensity traces, as well as 2D and fitted 3D autocorrelation maps for each ROI, are also depicted. (B) Average RICS diffusion coefficients for each receptor and treatment for each cell analyzed from n = 4 experiments. (C) GLP-1R clustering kinetics measured by TR-FRET in INS-1 832/3 SNAP-GLP-1R cells treated with 100 nM agonist as indicated, with vehicle-corrected AUCs included; n = 4. (D) GIPR clustering kinetics measured by TR-FRET in INS-1 832/3 SNAP-GIPR cells treated with 100 nM agonist as indicated, with vehicle-corrected AUCs included; n = 4. Data are mean ± SEM, compared by one-way ANOVA with Sidak's test; **P < .0001; ns: non-significant. We next analyzed the level of receptor degradation and lysosomal localization with a series of assays for both incretin receptors in INS-1 $\frac{832}{3}$ cells. We first assessed the total level of SNAP-GLP-1R vs SNAP-GIPR in INS-1 $\frac{832}{3}$ SNAP-GLP-1R or SNAP-GIPR cells in vehicle conditions or following 3-hour stimulation with 100 nM GLP-1 or GIP by Western blotting (Fig. 4A and 4B). This experiment showed an increased propensity for degradation of the GLP-1R when compared with the GIPR. We next quantified the level of receptor degradation using a high-content microscopy approach in which remaining total cellular SNAP-tag receptor is labeled after agonist incubation using the cell-permeable SNAP-tag probe BG-OG [23], and again observed faster receptor degradation for the GLP-1R vs the GIPR (Fig. 4C). Finally, we also quantified the colocalization between each SNAP-tagged receptor and the lysosomes, finding significantly higher lysosomal targeting for the GLP-1R compared with the GIPR (Fig. 4D and Supplemental Fig. 5) [32], a pattern that correlates with a reduced tendency for the GLP-1R vs the GIPR to localize to Rab11-positive recycling compartments following cognate agonist exposure (Supplemental Fig. 6) [32]. Figure 4.: Beta cell GLP-1r vs GIPR degradation propensities. (A) Western blot assessment of SNAP-GLP-1R or GIPR over tubulin levels in INS-1 832/3 SNAP-GLP-1R or SNAP-GIPR cells with or without stimulation with 100 nM GLP-1 or GIP, respectively, for 6 hours in the presence of the protein synthesis inhibitor cycloheximide; n = 3. (B) Representative Western blot results from (A). Note that the top bands were used to quantify the SNAP-receptor levels, as they correspond to the glycosylated forms of the receptors, known to be biologically active and correctly inserted at the plasma membrane (24). (C) Percentage of GLP-1R vs GIPR, labeled with the cell-permeable SNAP-tag probe BG-OG, and corresponding representative images from INS-1 832/3 SNAP-GLP-1R or SNAP-GIPR cells with or without stimulation with 100 nM GLP-1 or GIP, respectively, for the indicated times in the presence of cycloheximide; n = 4. (D) Percentage of co-localization (Mander's coefficient) and representative images of SNAP-GLP-1R vs -GIPR (labeled with SNAP-Surface 649) with Lysotracker Green in INS-1 832/3 SNAP-GLP-1R or SNAP-GIPR cells stimulated with 100 nM GLP-1 or GIP for 1 hour; n = 5. Data are mean ± SEM, compared by ratio-paired or unpaired t test or two-way ANOVA with Sidak's post hoc test; P < .001, *P < .0001. Having elucidated the main trafficking characteristics of both receptors, we next determined the coupling of each incretin receptor with specific signaling mediators, including Gαs, Gαq, Gαi, and β-arrestin 2 in INS-1 $\frac{832}{3}$ cells using NanoBiT complementation assays (Fig. 5). As previously shown by our group using analogous assays in HEK293T cells [37], the GLP-1R was preferentially coupled to Gαs, followed by Gαq and with minimum coupling to Gαi proteins in response to GLP-1 stimulation (Fig. 5A), while GIPR responses to GIP were markedly reduced for all readouts compared with GLP-1R (Fig. 5B). As the values for Gαs recruitment to the GIPR were almost as low as those obtained for Gαi using this NanoBiT approach, we decided to verify whether the Gαs results would be consistent when using a potentially more sensitive assay, namely a method based on NanoBRET between C-terminal NanoLuc-fused receptor and mini-Gs-Venus in the same cells as above (Fig. 5C). Results with this method again showed significantly reduced Gαs recruitment to the GIPR compared to the GLP-1R, although the difference between both receptors was less pronounced than previously found by NanoBiT complementation. Figure 5.: Beta cell GLP-1R vs GIPR G protein subtype and β-arrestin 2 recruitment characteristics. (A) Kinetics of Gαs, Gαq, Gαi, and β-arrestin 2 recruitment to the GLP-1R assessed by NanoBiT complementation assay in INS-1 832/3 GLP-1R KO cells transiently expressing GLP-1R-SmBiT and the corresponding mini-G protein subtype or β-arrestin 2. Responses to 100 nM GLP-1 normalized to vehicle and corresponding AUCs are shown; n = 5. (B) Kinetics of Gαs, Gαq, Gαi, and β-arrestin 2 recruitment to the GIPR assessed by NanoBiT complementation assay in INS-1 832/3 GIPR KO cells transiently expressing GIPR-SmBiT and the corresponding mini-G protein subtype or β-arrestin 2. Responses to 100 nM GIP normalized to vehicle and corresponding AUCs are shown; n = 5. (C) NanoBRET assessment of GLP-1R vs GIPR recruitment of Gαs, performed in INS-1 832/3 GLP-1R KO vs GIPR KO cells transiently expressing mini-Gs-Venus and either GLP-1R- or GIPR-NanoLuc, after stimulation with 100 nM GLP-1 or GIP, respectively, with corresponding AUCs also shown; n = 4. Data are mean ± SEM, compared by paired t test; *P < .05. Next, to determine the spatiotemporal pattern of Gαs activation elicited by the 2 receptors, we employed a recently developed bystander NanoBiT signaling assay based on the recruitment of activated Gαs-recognizing nanobody 37 (Nb37) to plasma membrane and endosomal locations in response to specific agonist stimulations [38] (Fig. 6 and Supplemental Fig. 7) [32]. While the assay showed no significant differences in the recruitment of Nb37 to active GLP-1Rs or GIPRs within the plasma membrane (Fig. 6A and Supplemental Fig. 7A) [32], there was a clear difference in GLP-1R vs GIPR endosomal activity, including a profoundly reduced GIPR Emax response despite increased potency for endosomal signaling with this receptor compared with the GLP-1R (Fig. 6B and Supplemental Fig. 7B) [32]. Figure 6.: GLP-1R vs GIPR endosomal vs plasma membrane activity in beta cells. (A) GLP-1R vs GIPR plasma membrane activity dose response curves from bystander NanoBiT signaling assays performed in INS-1 832/3 GLP-1R KO vs GIPR KO cells transiently expressing Nb37-SmBiT, LgBiT-CAAX and SNAP-GLP-1R or -GIPR, after stimulation with the indicated concentrations of GLP-1 or GIP, respectively; 30-minute AUC for each agonist concentration tested were fitted to 3-parameter dose response curves to obtain plasma membrane Emax and logEC50 for both receptors, with comparisons between these parameters included; n = 4. (B) As for (A) but for GLP-1R vs GIPR endosomal activity in INS-1 832/3 GLP-1R KO vs GIPR KO cells transiently expressing Nb37-SmBiT, Endofin-LgBiT and SNAP-GLP-1R or -GIPR, after stimulation with the indicated concentrations of GLP-1 or GIP, respectively; n = 4. Data are mean ± SEM, compared by paired t tests; P < .05, **P < .001. Finally, we determined the downstream signaling effects of both incretin receptors when stimulated with their native cognate agonists in beta cells (Fig. 7), and found a near-significant increase in cAMP response to 100 nM GIP compared with GLP-1 stimulation in wild-type INS-1 $\frac{832}{3}$ cells (Fig. 7A), a tendency replicated in primary mouse islet beta cells (Fig. 7B). For intracellular calcium influx in INS-1 $\frac{832}{3}$ cells, while there was an initial trend for a decreased GIPR compared with GLP-1R response during the first minute of stimulation, this tendency was reversed for the following 4 minutes of agonist exposure, resulting in a zero net difference between both receptors (Fig. 7C). However, in primary islets, there was a significant increase in calcium influx in response to GIP vs GLP-1 stimulations (Fig. 7D). Finally, insulin secretion assays performed in INS-1 $\frac{832}{3}$ cells showed a nonsignificant tendency toward improvement following GIP compared with GLP-1 exposure, with responses being specific for either the GIPR (for GIP) or the GLP-1R (for GLP-1), as they were abolished in the absence of each receptor in the corresponding GLP-1R or GIPR KO cells (Fig. 7E). Figure 7.: Functional analysis of GLP-1R vs GIPR signaling in beta cells. (A) cAMP dose response curves to GLP-1 vs GIP assessed in INS-1 832/3 cells by HTRF assay; results were fitted to 3-parameter dose response curves to obtain plasma membrane Emax and logEC50 for both receptors, depicted here combined as log(Emax/EC50) for each receptor; n = 5. (B) cAMP FRET responses to 100 nM GLP-1 or GIP from isolated and 4-hydroxytamoxifen-treated Pdx1-CreERT/CAMPER mouse islets; including agonist AUCs calculated for each receptor; n = 4. (C) INS-1 832/3 calcium responses (using the calcium indicator Cal520-AM) to 100 nM GLP-1 or GIP, including agonist AUCs calculated for 0-1 minute and 1-5 minutes responses for each receptor; n = 4. (D) Calcium responses to 100 nM GLP-1 vs GIP from purified WT mouse islets loaded with Cal520-AM, including agonist AUCs for each receptor; n = 4. (E) Insulin secretion responses to 100 nM GLP-1 vs GIP, expressed as fold increases to 11 mM glucose (G11) secretion levels from INS-1 832/3 WT (n = 3), GLP-1R KO (n = 5) and GIPR KO (n = 4) cells. Data are mean ± SEM, compared by paired t tests; P < .05; ns, nonsignificant.* ## Discussion In this study, we have established the main pattern of spatiotemporal signaling for each incretin receptor in a relevant cellular system, primarily INS-1 $\frac{832}{3}$ rat beta cells (with selected assays in primary mouse islets), where the receptors are expressed endogenously (see Table 1 for a summary of the main results). We have found marked differences in the trafficking and signaling characteristics from the 2 receptors, with GIPR associated with significantly reduced internalization and degradation propensities but increased plasma membrane recycling when compared with GLP-1R by several different techniques, a pattern that correlates with reduced activity from endosomes but no significant differences in plasma membrane activity. While the trafficking of the GIPR has been examined before in heterologous HEK293/HEK293T cells [37, 39], this is to our knowledge the first in-depth examination of these patterns in endogenous receptor-expressing pancreatic beta cells, including spatiotemporal assessment of signaling. Interestingly, while the net effect of these trafficking variations appears to be an increased level of sustained GIPR localization at the plasma membrane, we were nevertheless able to detect some intracellular GIP-FITC signal accumulation after 30 minutes of agonist stimulation (although intracellular GIP-TMR, unlike GLP-1-TMR, was negligible at the shorter time point of 5 minutes poststimulation), pointing toward active GIPRs continuously shuttling in and out of cells by a slow internalizing coupled to a rapidly recycling pathway, depositing their agonist in an intracellular location prior to returning to the plasma membrane, in a mechanism reminiscent to that previously observed by us for the related glucagon receptor (GCGR) [38]. Table 1. \| Unnamed: 0 \| GLP-1R \| GIPR \| \| --- \| --- \| --- \| \| Trafficking \| Trafficking \| Trafficking \| \| Islet surface expression \| + \| + \| \| Internalization \| + \| + \| \| Recycling \| + \| + \| \| Endosome localization \| + \| − \| \| Clustering (vehicle) \| − \| + \| \| Clustering (stimulated) \| + \| + \| \| Degradation \| + \| + \| \| Coupling \| Coupling \| Coupling \| \| Gαs \| + \| + \| \| Gαq \| + \| − \| \| Gαi \| − \| − \| \| β-arrestin 2 \| + \| + \| \| Plasma membrane Gαs \| + \| + \| \| Endosome Gαs \| + \| + \| \| Beta cell responses \| Beta cell responses \| Beta cell responses \| \| cAMP (INS-1 832/3) \| + \| +a \| \| cAMP (islet beta cells) \| + \| +a \| \| Calcium (INS-1 832/3) \| + \| + \| \| Calcium (islets) \| + \| + \| \| Insulin secretion (INS-1 832/3) \| + \| +a \| A previous study from our group performed in HEK293T cells also reported reduced internalization propensity for the GIPR vs the GLP-1R in response to stimulation with their corresponding native agonists, a pattern that correlated with reduced recruitment of β-arrestin 2 to the GIPR [37], an effect already observed before in a prior study from a separate group performed in HEK293 cells [40]. Here, we again observe reduced propensity for β-arrestin 2 recruitment by the GIPR in a beta cell context. The role of β-arrestins on incretin receptor trafficking and signaling has previously been investigated from several angles, for example by the use of biased agonists with different capabilities for β-arrestin recruitment [13, 37], using in vivo conditional β-arrestin 2 knockout mouse models, or with in vitro cell systems with deleted β-arrestin $\frac{1}{2}$ expression [24, 37]. In all these instances, β-arrestin recruitment closely correlated with the degree of incretin receptor internalization, but alterations in β-arrestin expression levels or complete β-arrestin deletion did not lead to significant effects in receptor endocytosis, but rather resulted in the prolongation of cAMP/PKA signaling duration, suggesting that the main effect of this important signaling mediator lies in the steric hindrance caused by its binding to the receptor, leading to reduced access of GαS to its binding pocket and promoting homologous receptor desensitization [41]. Of note, similar reduced GIPR vs GLP-1R β-arrestin recruitment and internalization into a Rab5-positive endosomal compartment were also apparent in a separate study [39], although in this instance the authors focused on the comparison between responses from single and dual agonists such as tirzepatide or MAR709 for each receptor rather than performing a direct comparison of both receptor responses. Also interestingly, we have previously observed that in vivo GIPR responses are less affected by β-arrestin 2 deletion specifically from pancreatic beta cells (unpublished), suggesting a reduced reliance on β-arrestin 2 to regulate GIPR signaling effects in the pancreas. Despite the abovementioned effects, the GIPR seems not only to be associated with reduced β-arrestin 2 recruitment but paradoxically with a general dampening in the recruitment of other downstream effectors including Gαs and Gαq proteins when measured as fold increases to vehicle levels, matching previous observations in HEK293T cells [37]. While the reasons behind this effect are not fully elucidated, it is important to point out that the GIPR has previously been found to have significantly higher levels of basal activity vs the GLP-1R [40]. Accordingly, we found increased basal association of this receptor with cholesterol-rich plasma membrane nanodomains [24], which are signaling hotspots rich in G proteins [42]. Consistently, we now find significantly reduced rates of basal diffusion for GIPR vs GLP-1R, indicative of increased basal clustering, which correlate with an overall reduction in clustering fold increases in response to GIPR stimulation, again suggesting higher basal activity for the GIPR compared with the GLP-1R. This observation could potentially contribute to the disconnect between the overall reduced level of Gαs and Gαq recruitment to the GIPR, measured as fold increases in stimulated over vehicle conditions, and the observed tendency toward increased cAMP, calcium, and insulin secretion responses to GIP vs GLP-1 in INS-1 $\frac{832}{3}$ cells. Also of note, these tendencies are present despite the measured loss of recruitment of active Gαs to endosomal compartments, suggesting that, as previously observed for the GLP-1R when stimulated with biased compounds affecting its capacity for endosomal localization and activity [13, 43], the plasma membrane is the main contributor to the overall signaling output of incretin receptors. This disconnect also highlights the existence of powerful mechanisms of signal amplification associated with the GIPR, potentially related to differential interactions with downstream signaling mediators vs the GLP-1R. It is worth noting, however, that direct comparisons between different recruitment and activity assays are problematic due to potential differences in receptor vs effector expression levels leading to variations in stoichiometry as well as different dynamic ranges for each specific assay which might complicate their interpretation. Another possible limitation of our study relates to the use of clonal INS-1 $\frac{832}{3}$ sublines with either GLP-1R or GIPR inactivation, although preservation of cellular responses to the remaining incretin receptor in GLP-1R and GIPR KO cells, as demonstrated in Fig. 7E, increases our confidence that these clones have retained the necessary machinery for incretin-dependent potentiation of insulin secretion. Nevertheless, it is interesting to note that the GIPR also appears to signal more prominently than the GLP-1R in primary islets, where we measured a nonsignificant tendency toward increased cAMP and a significant increase in calcium responses to GIP vs GLP-1 despite a 12-fold reduction in endogenous surface GIPR vs GLP-1R levels estimated in WT mouse islets in this study by quantification of TMR-labeled agonist uptake. Here it is also important to highlight that while cAMP responses were measured in islets from mice expressing a cAMP biosensor specifically from beta cells, calcium responses were acquired from whole islets and therefore also included responses from other islet endocrine cell types, which might explain the differences in the response observed in INS-1 $\frac{832}{3}$ cells. In summary, we have described here significant differences in the trafficking and spatiotemporal signaling propensities of the 2 incretin receptors following their stimulation with native agonists in beta cells. While the GLP-1R is a rapidly internalizing receptor with increased propensity for β-arrestin 2 recruitment, endosomal localization and activity, and lysosomal degradation, the GIPR is associated with reduced coupling to G proteins and β-arrestin 2, as well as reduced internalization and endosomal activity, increased recycling, and an overall increase in beta cell signaling despite highly reduced levels of endogenous surface receptor expression. These characteristics suggest that GLP-1R and GIPR signaling from beta cells are differentially regulated, and might potentially engage distinct signal transduction and amplification mechanisms, highlighting the rationale for the development of dual agonists eliciting complementary beneficial effects from each receptor, as exemplified by the successful clinical development of the dual GLP-1R—GIPR agonist tirzepatide [11], a ligand which combines reduced internalization and β-arrestin 2 recruitment at the GLP-1R with GIPR stimulation [39]. Our study raises several conceptual issues: why is a receptor with reduced tendency for desensitization such as GIPR more easily exhausted in T2D conditions? We can speculate that the reduced level of expression of GIPR vs GLP-1R in beta cells under normal conditions might result in the selective preservation of GLP-1 responses in a context where expression of both receptors is compromised; alternatively, as our results point to a higher reliance on signal amplification mechanisms for the GIPR, it is possible that beta cell GIP responses are more dependent on conservation of downstream signaling mechanisms that might become dysfunctional in T2D. With regards to the apparent beneficial effects of both GIPR agonists and antagonists in controlling blood glucose levels, it is difficult to infer any direct answers from the present study as antagonist effects might be indirect, potentially involving weight loss or control of glucagon hypersecretion from alpha cells. We can only speculate that any direct effect on beta cells could conceivably be linked to the increased propensity for basal coupling of the GIPR, with antagonists possibly affecting the balance of free and bound effectors available for productive signal transduction. In the future, it would be interesting to test the effect of both GLP-1R and GIPR antagonists in our beta cell systems to further investigate this possibility. The present study highlights profound differences in the behavior of both incretin receptors in beta cells. It is therefore likely that underlying regulatory mechanisms specific for each receptor exist. These might include unique sets of interacting proteins and lipids as well as specific receptor posttranslational modifications that can potentially regulate the trafficking and signaling of each receptor individually. A thorough investigation of these processes is now paramount to design novel treatment strategies for T2D and obesity based on enhancing the individual signaling output of each receptor for maximal effect. ## Disclosures The authors have nothing to disclose. ## Data Availability Original data generated and analyzed during this study are included in this manuscript. ## References 1. Seino Y, Fukushima M, Yabe D. GIP And GLP-1, the two incretin hormones: similarities and differences. J Diabetes Investig (2010.0) 1 8-23 2. Thorens B. Expression cloning of the pancreatic beta cell receptor for the gluco-incretin hormone glucagon-like peptide 1. Proc Natl Acad Sci U S A (1992.0) 89 8641-8645. PMID: 1326760 3. Volz A, Goke R, Lankat-Buttgereit B, Fehmann HC, Bode HP, Goke B. Molecular cloning, functional expression, and signal transduction of the GIP-receptor cloned from a human insulinoma. FEBS Lett (1995.0) 373 23-29. PMID: 7589426 4. Yamada Y, Hayami T, Nakamura K. Human gastric inhibitory polypeptide receptor: cloning of the gene (GIPR) and cDNA. Genomics (1995.0) 29 773-776. PMID: 8575774 5. Nauck MA, Quast DR, Wefers J, Meier JJ. GLP-1 receptor agonists in the treatment of type 2 diabetes—state-of-the-art. Mol Metab (2021.0) 46 6. Tan Q, Akindehin SE, Orsso CE. Recent advances in incretin-based pharmacotherapies for the treatment of obesity and diabetes. Front Endocrinol (Lausanne) (2022.0) 13 7. Boer GA, Keenan SN, Miotto PM, Holst JJ, Watt MJ. GIP Receptor deletion in mice confers resistance to high-fat diet-induced obesity via alterations in energy expenditure and adipose tissue lipid metabolism. Am J Physiol Endocrinol Metab (2021.0) 320 E835-E845. PMID: 33645252 8. Gasbjerg LS, Gabe MBN, Hartmann B. Glucose-dependent insulinotropic polypeptide (GIP) receptor antagonists as anti-diabetic agents. Peptides (2018.0) 100 173-181. PMID: 29412817 9. Yang B, Gelfanov VM, El K. Discovery of a potent GIPR peptide antagonist that is effective in rodent and human systems. Mol Metab (2022.0) 66 10. Campbell JE. Targeting the GIPR for obesity: to agonize or antagonize? Potential mechanisms. Mol Metab (2021.0) 46 11. Min T, Bain SC. The role of tirzepatide, dual GIP and GLP-1 receptor agonist, in the management of type 2 diabetes: the SURPASS clinical trials. Diabetes Ther (2021.0) 12 143-157. PMID: 33325008 12. Manchanda Y, Bitsi S, Kang Y, Jones B, Tomas A. Spatiotemporal control of GLP-1 receptor activity. Curr Opin Endocr Metab Res (2021.0) 16 19-27 13. Jones B, Buenaventura T, Kanda N. Targeting GLP-1 receptor trafficking to improve agonist efficacy. Nat Commun (2018.0) 9 1602. PMID: 29686402 14. Yabe D, Seino Y. Two incretin hormones GLP-1 and GIP: comparison of their actions in insulin secretion and beta cell preservation. Prog Biophys Mol Biol (2011.0) 107 248-256. PMID: 21820006 15. Younan SM, Rashed LA. Impairment of the insulinotropic effect of gastric inhibitory polypeptide (GIP) in obese and diabetic rats is related to the down-regulation of its pancreatic receptors. Gen Physiol Biophys (2007.0) 26 181-193. PMID: 18063845 16. Zhou J, Livak MF, Bernier M. Ubiquitination is involved in glucose-mediated downregulation of GIP receptors in islets. Am J Physiol Endocrinol Metab (2007.0) 293 E538-E547. PMID: 17505054 17. Oduori OS, Murao N, Shimomura K. Gs/Gq signaling switch in beta cells defines incretin effectiveness in diabetes. J Clin Invest (2020.0) 130 6639-6655. PMID: 33196462 18. Miki T, Minami K, Shinozaki H. Distinct effects of glucose-dependent insulinotropic polypeptide and glucagon-like peptide-1 on insulin secretion and gut motility. Diabetes (2005.0) 54 1056-1063. PMID: 15793244 19. Aaboe K, Knop FK, Vilsboll T. KATP Channel closure ameliorates the impaired insulinotropic effect of glucose-dependent insulinotropic polypeptide in patients with type 2 diabetes. J Clin Endocrinol Metab (2009.0) 94 603-608. PMID: 19050053 20. Naylor J, Suckow AT, Seth A. Use of CRISPR/Cas9-engineered INS-1 pancreatic beta cells to define the pharmacology of dual GIPR/GLP-1R agonists. Biochem J (2016.0) 473 2881-2891. PMID: 27422784 21. Fang Z, Chen S, Pickford P. The influence of peptide context on signaling and trafficking of glucagon-like peptide-1 receptor biased agonists. ACS Pharmacol Transl Sci (2020.0) 3 345-360. PMID: 32296773 22. Jaccard N, Griffin LD, Keser A. Automated method for the rapid and precise estimation of adherent cell culture characteristics from phase contrast microscopy images. Biotechnol Bioeng (2014.0) 111 504-517. PMID: 24037521 23. Fang Z, Chen S, Manchanda Y. Ligand-specific factors influencing GLP-1 receptor post-endocytic trafficking and degradation in pancreatic Beta cells. Int J Mol Sci (2020.0) 21 8404. PMID: 33182425 24. Buenaventura T, Bitsi S, Laughlin WE. Agonist-induced membrane nanodomain clustering drives GLP-1 receptor responses in pancreatic beta cells. PLoS Biol (2019.0) 17 25. Morana O, Nieto-Garai JA, Bjorkholm P. Identification of a new cholesterol-binding site within the IFN-gamma receptor that is required for signal transduction. Adv Sci (Weinh) (2022.0) 9 26. Pickford P, Lucey M, Fang Z. Signalling, trafficking and glucoregulatory properties of glucagon-like peptide-1 receptor agonists exendin-4 and lixisenatide. Br J Pharmacol (2020.0) 177 3905-3923. PMID: 32436216 27. Rossow MJ, Sasaki JM, Digman MA, Gratton E. Raster image correlation spectroscopy in live cells. Nat Protoc (2010.0) 5 1761-1774. PMID: 21030952 28. Garcia E, Bernardino de la Serna J. Dissecting single-cell molecular spatiotemporal mobility and clustering at focal adhesions in polarised cells by fluorescence fluctuation spectroscopy methods. Methods (2018.0) 140-141 85-96. PMID: 29605734 29. Compte M, Harwood SL, Munoz IG. A tumor-targeted trimeric 4-1BB-agonistic antibody induces potent anti-tumor immunity without systemic toxicity. Nat Commun (2018.0) 9 4809. PMID: 30442944 30. Muntean BS, Zucca S, MacMullen CM. Interrogating the spatiotemporal landscape of neuromodulatory GPCR signaling by real-time imaging of cAMP in intact neurons and circuits. Cell Rep (2018.0) 22 255-268. PMID: 29298426 31. Klarenbeek JB, Goedhart J, Hink MA, Gadella TW, Jalink K. A mTurquoise-based cAMP sensor for both FLIM and ratiometric read-out has improved dynamic range. PLoS One (2011.0) 6 32. Manchanda Y, Bitsi S, Chen S. DOI: 10.6084/m9.figshare.21992441.v12023 33. Jones BJ, Scopelliti R, Tomas A, Bloom SR, Hodson DJ, Broichhagen J. Potent prearranged positive allosteric modulators of the glucagon-like peptide-1 receptor. ChemistryOpen (2017.0) 6 501-505. PMID: 28794944 34. Samms RJ, Christe ME, Collins KA. GIPR Agonism mediates weight-independent insulin sensitization by tirzepatide in obese mice. J Clin Invest (2021.0) 131 35. DiGruccio MR, Mawla AM, Donaldson CJ. Comprehensive alpha, beta and delta cell transcriptomes reveal that ghrelin selectively activates delta cells and promotes somatostatin release from pancreatic islets. Mol Metab (2016.0) 5 449-458. PMID: 27408771 36. Chandra M, Kendall AK, Jackson LP. Toward understanding the molecular role of SNX27/retromer in human health and disease. Front Cell Dev Biol (2021.0) 9 37. Jones B, McGlone ER, Fang Z. Genetic and biased agonist-mediated reductions in beta-arrestin recruitment prolong cAMP signaling at glucagon family receptors. J Biol Chem (2021.0) 296 38. McGlone ER, Manchanda Y, Jones B. Receptor activity-modifying protein 2 (RAMP2) alters glucagon receptor trafficking in hepatocytes with functional effects on receptor signalling. Mol Metab (2021.0) 53 39. Novikoff A, O'Brien SL, Bernecker M. Spatiotemporal GLP-1 and GIP receptor signaling and trafficking/recycling dynamics induced by selected receptor mono- and dual-agonists. Mol Metab (2021.0) 49 40. Al-Sabah S, Al-Fulaij M, Shaaban G. The GIP receptor displays higher basal activity than the GLP-1 receptor but does not recruit GRK2 or arrestin3 effectively. PLoS One (2014.0) 9 41. van Gastel J, Hendrickx JO, Leysen H. beta-Arrestin based receptor signaling paradigms: potential therapeutic targets for Complex age-related disorders. Front Pharmacol (2018.0) 9 1369. PMID: 30546309 42. Ferre S, Ciruela F, Dessauer CW. G protein-coupled receptor-effector macromolecular membrane assemblies (GEMMAs). Pharmacol Ther (2022.0) 231 43. Marzook A, Chen S, Pickford P. Evaluation of efficacy- versus affinity-driven agonism with biased GLP-1R ligands P5 and exendin-F1. Biochem Pharmacol (2021.0) 190
--- title: The Legacy of Infectious Disease Exposure on the Genomic Diversity of Indigenous Southern Mexicans authors: - Obed A Garcia - Kendall Arslanian - Daniel Whorf - Serena Thariath - Mark Shriver - Jun Z Li - Abigail W Bigham journal: Genome Biology and Evolution year: 2023 pmcid: PMC10016042 doi: 10.1093/gbe/evad015 license: CC BY 4.0 --- # The Legacy of Infectious Disease Exposure on the Genomic Diversity of Indigenous Southern Mexicans ## Abstract To characterize host risk factors for infectious disease in Mesoamerican populations, we interrogated 857,481 SNPs assayed using the Affymetrix 6.0 genotyping array for signatures of natural selection in immune response genes. We applied three statistical tests to identify signatures of natural selection: locus-specific branch length (LSBL), the cross-population extended haplotype homozygosity (XP-EHH), and the integrated haplotype score (iHS). Each of the haplotype tests (XP-EHH and iHS) were paired with LSBL and significance was determined at the $1\%$ level. For the paired analyses, we identified 95 statistically significant windows for XP-EHH/LSBL and 63 statistically significant windows for iHS/LSBL. Among our top immune response loci, we found evidence of recent directional selection associated with the major histocompatibility complex (MHC) and the peroxisome proliferator-activated receptor gamma (PPAR-γ) signaling pathway. These findings illustrate that Mesoamerican populations' immunity has been shaped by exposure to infectious disease. As targets of selection, these variants are likely to encode phenotypes that manifest themselves physiologically and therefore may contribute to population-level variation in immune response. Our results shed light on past selective events influencing the host response to modern diseases, both pathogenic infection as well as autoimmune disorders. ## Introduction Infectious diseases are among the strongest selective pressures acting on the human genome. Indeed, many genes subject to local positive natural selection (e.g., CD40) or balancing natural selection (e.g., CCR5, genes of the MHC complex) are associated with susceptibility to infectious disease (Hughes and Yeager 1998a; Bamshad et al. 2002; Sabeti et al. 2002a). Population genomic studies of global human populations show evidence of population-specific selection at immune response loci (Fan et al. 2016). Despite these advances, there remains a critical gap in our knowledge concerning selection at immune response loci among Indigenous Americans. The diversity of infectious diseases in the Americas prior to colonial contact differed from the infectious diseases of Afro-Eurasia and Oceania. This was an outcome of both geographic isolation and differences in zoonotic biota (e.g., insects and fauna) that served as disease vectors. Therefore, selection likely did not act on the genomes of Indigenous Americans for variants that protected them from the Afro-Eurasian infectious diseases. Rather, pre-colonial populations in the Americas adapted to diseases that were locally prevalent, such as Chagas, tuberculosis, syphilis, and hepatitis (Merbs 1992; Klaus et al. 2010; Bos et al. 2014; Steverding 2014). Indigenous Americans' isolation from Afro-Eurasian infectious diseases ended with devastating effects. Beginning with European colonial contact in the late 15th century, there was a steady influx of novel infectious diseases to the Americas such as variola virus (smallpox) and measles virus—diseases for which indigenous communities across the Americas did not possess specific immunity. Across the Americas, famine, slavery, infectious disease, and warfare, contributed to the population collapse of various Indigenous American societies (Livi-Bacci 2006). Mitochondrial DNA data corroborate these historical accounts by demonstrating a population bottleneck 500 years ago coincident with European contact (O'Fallon and Fehren-Schmitz 2011). Accordingly, the evolutionary pressures for survival were strong. However, our knowledge of Indigenous American genetic variation in general and at loci related to infectious disease and immune response is limited. To date, only a few studies have identified genes under selection in Indigenous American populations (Eichstaedt et al. 2014; Lindo et al. 2016; Crawford et al. 2017; Mychaleckyj et al. 2017; Reynolds et al. 2019; Avila-Arcos et al. 2020). Among Mesoamericans, selection at immune response loci may have been particularly robust given this region's population density and level of urbanization throughout both pre- and post-colonial time periods (Smith 2005; Livi-Bacci 2006; Mummert et al. 2011). Furthermore, historical records from the colonial era indicate that novel infectious disease introduced by European colonizers (e.g., variola virus that causes smallpox) killed upwards of $90\%$ of the indigenous communities in the region (Lockhart 1992; Feldman 1999; Restall et al. 2005; Leon-Portilla 2011). This high mortality rate led us to hypothesize that colonial contact left a strong signature of natural selection in the genomes of Mesoamericans at immune response loci. Here, we interrogated SNP genotype data from Indigenous Mesoamericans for evidence of natural selection. We expected to identify a high proportion of immune response genes and pathways under natural selection given the history of infectious disease exposure among Mesoamericans across time. ## Mesoamerican Population Characteristics Our Mesoamerican cohort included 39 individuals genotyped using the Affymetrix Genome-Wide Human SNP Array 6.0 containing 906,600 SNPs representing 25 Maya from the Yucatan Peninsula of Mexico, two Nahua, seven Mixtec, and five Tlapanec speakers from Guerrero, Mexico previously described in Bigham et al. [ 2010]. Together, these individuals from different linguistic groups form a metapopulation that provides a shared history of selection in the population of Indigenous American particularly considering the much later diversification of languages than the dates calculated for the origin of the haplotypes under selection (Campbell 2000). Similar genetic data show that even though population substructure occurs among linguistic groups, the south forms a cluster with each other in Mesoamerica (García-Ortiz et al. 2021). We carried out statistical analysis using 857,481 autosomal SNPs that passed QC. We removed six individuals from the dataset that were first, second, or third-degree relatives, leaving us with a sample size of $$n = 33$$ individuals (supplementary table S1, Supplementary Material online; Manichaikul et al. 2010). Indigenous Mesoamerican populations are known to exhibit varying degrees of European admixture (Bryc et al. 2010; Magalhaes et al. 2012). We performed a principal component analysis (PCA) in Plink 1.9, to visualize the relationship between our populations (fig. 1A; Purcell et al. 2007; Chang et al. 2015). In order to identify and remove the effects of European admixture from our selection scan, we estimated global ancestry using ADMIXTURE (Alexander and Lange 2011). We tested for four-way admixture including ancestry from the Americas, Europe, Africa, and East Asia (fig. 1B). Individual admixture estimates ranged from a maximum of $100\%$ Indigenous American ancestry to a minimum of $75\%$ Indigenous American ancestry, with most of our cohort possessing Indigenous Ancestry estimates above $90\%$ (supplementary table S1, Supplementary Material online). European admixture was the most common of the three non-American ancestries (fig. 1C). It was detected in 15 individuals, ranging from $1\%$ to $25\%$. Ten individuals had detectable East Asian ancestry ranging from $1\%$ to $6\%$. African Ancestry was detected in two individuals at $3\%$ and $2\%$. Fig. 1.: Individual ancestry estimates. Individual ancestry was estimated for Mesoamerican study participants using ADMIXTURE. (A) Principle component analysis for Mesoamericans, CEPH Europeans (CEU: Northern and Western Europeans from Utah), East Asians (CHB: Han Chinese from Beijing + JPT: Japanese from Tokyo), and Africans (YRI: Yoruba). PC1 explains 41.35% of the variance, while PC2 explains 17.63% of the variance observed. (B) ADMIXTURE global estimates (K = 4) of ancestry for Mesoamericans prior to masking admixture and removing related individuals. (C) All 39 Mesoamerican individuals prior to removing related individuals and correcting for admixture. (D) ADMIXTURE results for N = 33 and K = 4, after removing admixed segments using RFMIX2 and imputing missing genotypes using the unadmixed individuals for that specific chromosome from our cohort. Given the presence of non-Indigenous American ancestry within our final cohort of study participants, we assigned locus-specific ancestry to each chromosomal segment/haplotype using RFMix (Maples et al. 2013). To correct for admixture, which could be incorrectly detected as regions of selection, we set non-Indigenous American ancestry segments to missing and imputed the missing genotypes with SHAPEIT4 using the Indigenous American ancestry tracts from our dataset as the reference population (Delaneau et al. 2019). ADMIXTURE analysis performed on the masked and imputed Mesoamerican dataset indicated that this analysis effectively eliminated European and African ancestry from the Mesoamerican genomes (fig. 1D; supplementary tables S1 and S2, Supplementary Material online). After imputation, only three individuals had detectable East Asian ancestry less than $2\%$. Although recent scholarship such as Rodríguez-Rodríguez et al. [ 2022] has found substantial East Asian ancestry in Southern Mexico, we did not control for it as our IBD analysis also failed to detect any significant East Asian segments, therefore these are more likely due to shared ancestry rather than recent admixture events. ## Mesoamerican Genomes Show Evidence of a Population Bottleneck Mitochondrial DNA and historical records indicate that Indigenous American populations underwent a severe population bottleneck coincident with European contact beginning in the early 1500s (Lockhart 1992; Feldman 1999; Restall et al. 2005; Leon-Portilla 2011; O'Fallon and Fehren-Schmitz 2011). This bottleneck is hypothesized to be in large part caused by the introduction of novel infectious disease into the region. To detect evidence for this bottleneck, we estimated the historical effective population size of our cohort consisting of 33 Mesoamericans using the program AS-IBDne (Browning and Browning 2015; Browning et al. 2018). Our power is limited to reconstruct population effective size only to the first 50 generations as we used array data (Browning and Browning 2015). These data confirm that Mesoamericans went through a recent bottleneck, most likely associated with colonial contact (fig. 2). Fifty generations ago (∼1,250 years ago), Mesoamericans had a population size of roughly 86,400 people ($95\%$ bootstrap CI: 37,400−163,000). The vertex of the curve, or highest population effective size, was 42 generations ago with a population size of 94,100 ($95\%$ bootstrap CI: 36,700–21,0000). The data changes by a factor of 10 (from 104 to 103) between generations 15–16 (∼375–400 years ago). The base of the curve is evident at eight generations ago (∼200 years ago) with an effective population size of 4,800 ($95\%$ bootstrap CI: 3,270–6,130). While the confidence intervals are large throughout the dataset as a function of the small sample size analyzed, the effect of the bottleneck is noticeable with tighter $95\%$ confidence intervals throughout the bottleneck (supplementary table S3, Supplementary Material online). This supports the bottleneck previously observed for individuals of Indigenous American ancestry (Browning et al. 2018; Mooney et al. 2018). Fig. 2.: Effective population size estimates. Effective population sizes were calculated using AS-IBDNe. The y-axis represents the effective population size (Ne). The x-axis represents the generation time. Mesoamericans experienced a bottleneck effect, with the lowest effective population size at eight generations ago (200 years ago, assuming a 25-generation time). ## Mesoamerican Genomes Show Evidence of Selection at Immune Response Loci To detect evidence of positive directional selection in Mesoamericans, we performed a selection scan using 33 Mesoamerican genomes whose non-indigenous chromosomal ancestry tracts were masked and imputed. We identified genomic signals of natural selection using three statistics: 1) locus-specific branch length (LSBL) (Shriver et al. 2004), 2) cross-population extended haplotype homozygosity (XP-EHH) (Sabeti et al. 2007; Pickrell et al. 2009), and 3) integrated haplotype score (iHS) (Voight et al. 2006). The EHH-based haplotype tests, XP-EHH and iHS, were calculated in Selscan (Szpiech and Hernandez 2014). In so doing, we leveraged both allele frequency difference and haplotype homozygosity to identify putatively selected regions of the genome. LSBL was calculated for each SNP in the dataset with a MAF ≥ 0.05 (497,699 SNPs) by comparing Mesoamericans to East Asians and Europeans. We identified 4,976 SNPs falling in the top $1\%$ of the empirical distribution out of 497,699 total SNPs analyzed (fig. 3A). These SNPs exhibited Mesoamerican LSBL values from 0.442 to 0.887. The SNP with the most extreme LSBL value was MRTFA intronic variant rs17425081 located on chromosome 22. XP-EHH and iHS were calculated for non-overlapping windows of 100 kilobase pairs (kb). XP-EHH compared Mesoamericans to East Asians at 826,691 SNPs to look specifically for haplotypes present in Mesoamerican populations that arose after their split from Asian populations. iHS was calculated for 455,845 SNPs after filtering low-frequency variants. We identified 319 and 206 statistically significant 100 kb windows at the $1\%$ level for XP-EHH and iHS, respectively (fig. 3BandC). These windows were scattered across the autosomes. Chromosome 6 contained the most significant windows of any chromosome for both XP-EHH and iHS, with 59 and 28 windows, respectively. Fig. 3.: Manhattan plots of selection-scan statistics. For each plot, the value of the statistic is represented on the y-axis. Chromosome location is depicted along the x-axis. The thick horizontal lines indicate significance thresholds for each test statistic. (A) Distribution of LSBL values across the genome for Mesoamerican branch length calculated using East Asians and CEPH European Americans as outgroups. The line represents the 1% significance. (B) Plot of the absolute value of iHS scores for Mesoamericans. The line indicates scores of 2. The proportion of scores above 2 for each window is taken into consideration for determining the 1% significance. (C) Plot of XP-EHH comparing Mesoamericans to East Asians. Values above indicate directional selection in the Mesoamerican population whereas values below 0 indicate direction selection in East Asians. The line represents the values above or below 2, which Selscan flags as potentially significant above the 5% level. To reduce false positives, we identified regions of the genome showing statistical significance for LSBL and at least one of the two haplotype tests, XP-EHH and iHS. To be considered significant, the XP-EHH and iHS windows were designated to be in the $1\%$ tail by Selscan and that window needed at least one significant LSBL SNP also at the $1\%$ level. Ninety-five significant regions at $P \leq 0.01$ were identified for the LSBL/XP-EHH analysis and 63 for the LSBL/iHS analysis (fig. 4). These regions were scattered across the genome and found on every autosome except chromosome 9 and chromosome 22. For the iHS/LSBL analysis, chromosomes 3 and 6 tied for the most significant regions of any chromosome, with 10 windows falling in the top $1\%$ on each of the two chromosomes. Chromosome 6 contained the most significant regions for the XP-EHH/LSBL analysis with 15 windows, followed by chromosomes 3 and 12 with 10 each. Most of the significant results for chromosome 6 were identified in and around the major histocompatibility complex (MHC), a region essential for the adaptive immune response. Fig. 4.: Genomic distribution of the 1% windows for iHS and XP-EHH when paired with LSBL. Regions that were in the 1% distribution for all three statistical tests are shaded in a lightest color. For both of the combined statistical tests, the majority of windows were found on chromosome 6, followed by chromosome 3. One of the largest contiguous regions of statistical significance was found on chromosome 3 (chr3:12,300,001–12,700,001). This 4 MB region consisted of four, tandem significant 100 kb windows containing the following genes: PPARG, MKRN2, MKRN2OS, TSEN2, and RAF1. Twenty-four of the 64 SNPs genotyped for this region fell in the top $1\%$ of the empirical distribution for LSBL. Here, our most extreme LSBL value was 0.683 (rank 116) for the intronic SNP rs4684106 located in TSEN2 followed by the TSEN2 upstream variant rs17279604 (LSBL = 0.683, rank 117) and intronic variant rs17036821 (LSBL = 0.683, rank 118). There were several other extreme LSBL values including the RAF1 non-coding transcript variant rs1051208 (LSBL = 0.640, rank 274), the MKRN2OS intronic variant rs17036922 (LSBL = 0.593, rank 671), and the PPARG intronic variant rs17793693 (LSBL = 0.581, rank 824). The most extreme iHS value was for the regulatory region variant to PPARG, rs9833097 (iHS = −3.73082). Furthermore, 11 of the 24 SNPs analyzed in this gene fell in the top $1\%$ of the LSBL empirical distribution. A second compelling result was the identification of two related regions on separate chromosomes. The first region was a 500 kb window located on chromosome 16 (chr16:11,000,001–11,500,001) containing the immune response genes SOCS1 and CIITA, along with DEXI, CELC16A, PRM1, PRM2, PRM3, TNP2, MIR548H2, and RMI2. This region, particularly CIITA, is known to directly regulate MHC Class II expression, the second related region for which we found strong evidence of natural selection (Devaiah and Singer 2013; Sarmah et al. 2019). Within the first region, a 300 kb window (11,200,001–11,500,001) was significant for XP-EHH/LSBL and two 100 kb windows (11,000,001–11,100,001 and 11,200,001–11,300,001) were significant for iHS/LSBL. SNP rs4414511, located in the putative uncharacterized protein LOC400499, had the highest LSBL value (LSBL = 0.723, rank 49) for the region. Four additional extreme LSBL outliers, rs17605165, rs40448, rs28769, and rs193773, ranked 64, 68, 69, and 74, respectively, were identified in this region. Each of these four SNPs were in the lincRNA RP11-396B14.2, lying immediately upstream (∼30 kb) of SOCS1. The second related region was a 200 kb window located on chromosome 6 (chr6:33,000,001–33,200,001) containing the genes HLA-DPA1, HLA-DPB1, HLA-DPB2 (pseudogene), COL11A2, HCG24, HSD17B8, MIR219A1, RING1, RXRB, and SLC39A7. This region was significant for XP-EHH/LSBL and included a nested 100 kb window (33,000,001–33,100,001) that was significant for iHS/LSBL. In fact, this region contained the highest number of significant LSBL/XP-EHH and LSBL/iHS windows of any region analyzed. Our highest LSBL value was for rs3128918 (LSBL = 0.752, rank 27), followed by rs3130578, rs3130179, rs3128952, and rs3130180 (ranked 98, 99, 201, and 202, respectively). Eight additional SNPs fell in the $1\%$ XP-EHH/LSBL tail. To resolve which HLA alleles were part of the signature of selection, we imputed the classical HLA alleles for HLA-DPA1 and HLA-DPB1 using the multi-ethnic HLA reference panel in the Michigan Imputation Server (Das et al. 2016; Luo et al. 2021). HLA-DPA1 was resolved to be HLA-DPA101:03 (AF = 0.93, rsq = 0.67) and HLA-DPB1 was resolved to be HLA-DPB104:02 (AF = 0.89, rsq = 0.78). Given the high frequency in our dataset, these alleles form a single long-range haplotype, HLA-DPA101:03/DPB104:02. These allele frequencies and HLA haplotypes (HLA-DPA101:03/DPB104:02) were cross-referenced and concordant with previously reported HLA allele frequencies across Mexico, Central, and South America in the Allele Frequency Net Database (AFND) (Gonzalez-Galarza et al. 2019). To further confirm our imputed HLA allele frequencies and rule out unknown alternatives contributing to our imputation results, we accessed the publicly available high coverage whole genomes for individuals belonging to the Maya, Mixe, Mixtec, and Zapotec populations ($$n = 28$$) in the HGDP and SGDP projects (Mallick et al. 2016; Bergström et al. 2020). We used HLA-LA to call the DPA1 and DPB1 HLA alleles at the G-group resolution level (Dilthey et al. 2019). For DPA1, coverage ranged from 30.4 to 96.4, with an average of 45.5. For DPB1, the coverage ranged from 18.9 to 70.9, with an average of 43.9. In this new dataset, DPA101:03's frequency was $94.33\%$ and DPB104:02's frequency was $74.47\%$. While our DPA101:03 frequency was similar between the two datasets, $93\%$ versus $94.33\%$, our DPB104:02 frequency was higher in our dataset, $89\%$ versus $74.47\%$. However, both DPB104:02 frequencies are high, giving us confidence in our imputation. A third compelling result was a 200 kb window located on chromosome 5 (chr5:153,800,001–154,000,001) containing the genes GALNT10, HAND1, MIR3141, SAP30L, and SAP30L-AS1. For this window, 6 of the 18 SNPs tested for LSBL and all 54 of the SNPs tested for XP-EHH fell in the top $1\%$ of the results. The intergenic SNP, rs4958377, exhibited the highest LSBL value, 0.580 (rank 829), followed by the non-coding transcript exon variant, rs2351485, located in lncRNA region CTB-158E9 (LSBL = 0.560, rank = 1,149). The most extreme XP-EHH value within the window was 4.67 for SNP rs880083. Of note, the haplotype “core” for XP-EHH may be present just outside of the window, where rs7710430 had the max XP-EHH score for that region (chr5:153,797,277, XP-EHH value = 4.761, LSBL = 0.556, rank 1,207). For our combined LSBL-haplotype analysis, we identified several other significant chromosomal regions containing genes involved in immune response pathways that stood out given what is known about Mesoamerican population history. They included regions on chromosomes 2, 5, 6, 8, 12, and 15 that included the immune response genes CHIA, IL18R1, IL18RAP, DOCK2, CYP7A1, IL17F, RPAP3, ENDOU, and TCF12 (table 1). Windows containing these genes displayed LSBL values ranging from 0.791 (rank = 10) for the DOCK2 intronic variant, rs264838, to 0.449 for the CHIA intronic variant, rs1266828 (rank = 4,612). Five of these windows contained LSBL values falling in the top 200 and included the genes DOCK2, TCF12, RPAP3/ENDOU, and CYP7A1. Of note, upstream from the RPAP3/ENDOU window lies an extreme LSBL value for the regulatory region variant, rs2051827 (LSBL = 0.837, rank = 7). Six windows were significant for XP-EHH with values ranging from 2.43 for the regulatory region variant rs10201184 located in the window containing IL18R1/IL18RAP to 4.02 for RPAP3/ENDOU intergenic SNP rs667610. Two windows contained significant iHS scores including DOCK2 (rs155239 = −3.70). Table 1* \| Gene \| Chr \| Window (hg19) \| Hap. test \| Max test imputed \| Max test admixed \| N unadmixed subset \| Max test unadmixed \| Max LSBL \| Ohana LSBL \| Max LSBL SNP ID \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| IL18R1, IL18RAP \| 2 \| 103,000,001–103,100,001 \| XP-EHH \| 2.43 \| 2.55 \| \| \| 0.504 \| 0.493 \| rs4851007 \| \| PPARG, MRKN2, RAF1 \| 3 \| 12,300,001–12,700,001 \| iHS \| −3.73a \| 2.98 \| \| \| 0.683 \| 0.617 \| rs4684106 \| \| GALNT10 \| 5 \| 153,800,001–154,000,001 \| XP-EHH \| 4.76 \| 4.83 \| \| \| 0.58 \| 0.518 \| rs4958377 \| \| DOCK2 \| 5 \| 169,100,001–169,200,001 \| iHS \| −3.7 \| −3.45 \| \| \| 0.791 \| 0.717 \| rs264838 \| \| HLA-DPA1, HLA-DPB1, HLA-DPB2, RING1, RXRB \| 6 \| 33,000,001–33,100,001 \| iHS and XP-EHH \| −4.69 and 3.82 \| −4.41 and 2.56b \| 22.0 \| 3.51 (XP-EHH) \| 0.752 \| 0.576 \| rs3128918 \| \| IL17F \| 6 \| 52,100,001–52,200,001 \| XP-EHH \| 2.89a \| 2.65b \| 25.0 \| 3.36 \| 0.486 \| 0.401 \| rs1266828 \| \| CYP7A1 \| 8 \| 59,400,001–59,500,001 \| XP-EHH \| 3.83 \| 3.49 \| \| \| 0.746 \| 0.669 \| rs12545426 \| \| SOCS1, CIITA \| 16 \| 11,000,001–11,500,001 \| iHS and XP-EHH \| −6.26a and 3.01 \| −5.46 and 2.76 \| \| \| 0.723 \| 0.683 \| rs4414511 \| \| ENDOU \| 12 \| 48,000,001–48,200,001 \| XP-EHH \| 4.1 \| 4.16 \| \| \| 0.765 \| 0.685 \| rs10082722 \| \| CHIA \| 1 \| 111,800,001–111,900,001 \| XP-EHH \| 3.02 \| 3.06 \| \| \| 0.449 \| 0.352 \| rs17027272 \| \| TCF12 \| 15 \| 57,200,001–57,500,001 \| XP-EHH \| 2.78a \| 2.17b \| 27.0 \| 2.69 \| 0.705 \| 0.618 \| rs2585110 \| We performed gene ontology pathway enrichment analysis in DAVID (Huang et al. 2009a, 2009b) to identify overrepresented associations of genes and gene groups. We limited our analysis to the significant regions identified from the LSBL/XP-EHH or the LSBL/iHS analysis. DAVID analysis identified 21 biological processes gene ontology (GO) terms for the LSBL/XP-EHH analysis (supplementary table S4, Supplementary Material online) and 16 biological processes GO terms for the LSBL/iHS analysis (supplementary table S5, Supplementary Material online). We also performed a DAVID analysis using the combined set of significant genes for both LSBL/XP-EHH and LSBL/iHS, identifying 33 biological processes GO terms (supplementary table S6, Supplementary Material online). Several pathways related to immune function were identified including positive regulation of chemokine secretion, immune response, signal transduction, positive regulation of transcription from RNA polymerase II promoter, and positive regulation of interferon-gamma production. None of the fold enrichment P-values remained significant after correcting for multiple tests. In addition to our DAVID analysis, we performed a Reactome pathway analysis in SNP-NEXUS (Chelala et al. 2009; Dayem Ullah et al. 2018; Jassal et al. 2019). Our iHS/LSBL hits falling in the $1\%$ represent categories related to amino acid transport and transcriptional regulation (supplementary table S7, Supplementary Material online). For results in the $1\%$ of the XP-EHH/LSBL analysis, our strongest signals were related to homeostasis, IL-18 signaling, TGF-β signaling, and pro-inflammatory response (supplementary table S7, Supplementary Material online). The combined XP-EHH/iHS/LSBL analysis introduced additional signal transduction categories (supplementary table S7, Supplementary Material online). Expanding the analysis to the $5\%$ cutoff, we see overrepresentation of PPARG-related transcription factors, intracellular signaling by second messengers, and interferon-gamma signaling, among other immune-related categories (supplementary tables S8–S10, Supplementary Material online). To determine whether imputing from our own dataset overly homogenized regions and biased our selection scan results, we generated admixture-corrected allele frequencies using the program Ohana with $K = 4$ to control for any European, East Asian, and African genetic contributions to Mesoamericans (Szpiech and Hernandez 2014; Cheng et al. 2022). Using the Ohana matrix results, we recalculated LSBL on the Mesoamerican admixture-corrected allele frequencies. Our significant results remained significant in this new analysis. For instance, IL18R1 SNP rs4851007, had an LSBL value of 0.504 (P-value 0.0047) for our original calculations and 0.493 (P-value 0.0027) for the calculations using Ohana. Admixture adjusted LSBL values for our top candidate loci are reported in table 1, and they all remained in the $1\%$ tail of LSBL values. We also tested for the effects of imputation on XP-EHH and iHS. To do this, we used the original dataset (with admixture and without imputation) to recalculate XP-EHH/iHS and normalized as described above. Eight of the 11 regions under selection remained in the $1\%$ significance for the haplotype tests (table 1). Three regions that contained HLA-DPA1, HLA-DPB1, HLA-DPB2, RING1, RXRB, IL17F, and TCF12 dropped to the $5\%$ significance level but only in the haplotype tests. For these regions, we re-estimated iHS and XP-EHH using only individuals with Indigenous American ancestry. We find that even though these regions fell to the $5\%$ significance using the admixed individuals, using an unadmixed cohort brings the results back into the $1\%$. Therefore, our methods show that certain important immune loci could be missed when looking at an admixed cohort with a small sample size. To gain insight into the age and spread of putatively selected genomic regions identified here, we generated haplotype age estimates for eleven haplotypes using extended haplotype homozygosity (EHH) scores based on the results of gene grouping related to immune response from the DAVID analysis. To do so, EHH scores were log-transformed and linearly regressed to the distance from the core SNP (Sabeti et al. 2002b; Voight et al. 2006; Szpiech and Hernandez 2014). The haplotypes showing evidence of selection range in age from roughly 4,000 to 10,000 years (table 2). This translates to 162–380 generations when assuming a 25-year generation time. Thus, their introduction predates colonial contact and has implications for natural selection operating on standing variation. For the HLA haplotype, HLA-DPA101:03/DPB104:02, we estimated that it arose 6,000 years ago, and then increased in frequency through natural selection in response to infectious diseases. Table 2 \| Gene \| Chr \| Window (hg19) \| r2 \| Generations ago \| Generations (95% CI) \| Age (25 yr gen) \| Age (95% CI) \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| CHIA \| 1 \| 111,800,001–111,900,001 \| 0.61 \| 162.23 \| 145.25–179.21 \| 4055.85 \| 3,631.31–4,480.25 \| \| IL18R1, IL18RAP \| 2 \| 103,000,001–103,100,001 \| 0.87 \| 335.29 \| 322.24–348.34 \| 8382.2 \| 8,056.00–8,708.50 \| \| PPARG, MRKN2, RAF1 \| 3 \| 12,300,001–12,700,001 \| 0.86 \| 320.46 \| 298.43–342.49 \| 8011.5 \| 7,460.71–8,562.27 \| \| GALNT10 \| 5 \| 153,800,001–154,000,001 \| 0.93 \| 192.58 \| 187.92–197.25 \| 4814.55 \| 4,697.97–4,931.14 \| \| DOCK2 \| 5 \| 169,100,001–169,200,001 \| 0.9 \| 212.81 \| 205.39–220.24 \| 5320.35 \| 5,134.83–5,505.90 \| \| HLA-DPA1, HLA-DPB1, HLA-DPB2, RING1, RXRB \| 6 \| 33,000,001–33,100,001 \| 0.89 \| 240.13 \| 231.13–249.13 \| 6003.28 \| 5,778.34–6,228.23 \| \| IL17F \| 6 \| 52,100,001–52,200,001 \| 0.87 \| 199.11 \| 188.97–209.24 \| 4977.68 \| 4,724.29–5,231.06 \| \| CYP7A1 \| 8 \| 59,400,001–59,500,001 \| 0.97 \| 362.09 \| 353.06–371.11 \| 9052.28 \| 8,826.50–9,278.00 \| \| ENDOU \| 12 \| 48,000,001–48,200,001 \| 0.65 \| 264.2 \| 233.38–295.01 \| 6605.0 \| 5,834.56–7,375.24 \| \| TCF12 \| 15 \| 57,200,001–57,500,001 \| 0.77 \| 380.64 \| 355.14–406.15 \| 9516.0 \| 8,878.515–10,153.65 \| \| SOCS1, CIITA \| 16 \| 11,000,001–11,500,001 \| 0.91 \| 186.63 \| 176.85–196.41 \| 4665.78 \| 4,421.33–4,910.23 \| ## Discussion The pathogenic history of the Americas has undoubtedly impacted the suite of genetic variation present among Indigenous Americans. Nonetheless, our knowledge of Indigenous American genetic variation at immune response loci is incomplete, leaving a critical gap in our understanding of the genomic consequences of infectious disease exposure in the Americas. We hypothesized that genetic variation at immune response loci was shaped by natural selection among Indigenous Americans during their unique history of infectious disease exposure, including exposure to pathogens prevalent in the Americas prior to European contact as well as newly introduced infectious diseases arriving during the contact period. We show that Mesoamerican populations experienced a sizeable population bottleneck coincident with the arrival of Europeans in the Americas. This lends support to the hypothesis that newly introduced infectious diseases shaped extant patterns of genomic variation. To this end, we find evidence of natural selection in regions of the genome involved in the body's immune response in support of this hypothesis. Together, our data highlight the importance of immunity and adaptation among Mesoamerican populations whether deep in our evolutionary past, as recent as colonial contact, or continuously shaped by recent infectious diseases. Leveraging two orthogonal population genomic statistics that detected departures from neutrality, LSBL and XP-EHH/iHS, we identified signatures of natural selection in regions of the genome involved in the body's immune response. We identified 100 and 57 statistically significant windows for the LSBL/XP-EHH and LSBL/iHS analysis, respectively. Of these, three stood out as particularly compelling with respect to immune adaptation. The first was a 4 MB region located on chromosome 3 (chr3:12,300,001− 12,700,001) containing the gene PPARG, or Peroxisome Proliferator-Activated Receptor Gamma. It is a ligand-activated transcription factor that contributes to gene regulation as part of the PPAR-γ signaling pathway, which regulates lipid and glucose metabolism through the expression of cytokines and chemokines (Le Menn and Neels 2018). Importantly, the PPAR-γ signaling pathway activates both pro- and anti-inflammatory macrophages (Chawla 2010). The second was a region on chromosome 5 containing the gene GALNT10. GALNT10 interacts with MHC complex genes as well as various interleukin cytokines (Kakoola et al. 2014) and is responsible for regulating CD4+ T cells infiltration of macrophages and decreasing granzyme B expression in CD8+ T cells (Zhang et al. 2020). CD4+ T cells are crucial to immune memory and CD8+ T cells are essential for protection against viruses, intracellular bacterial infection, and tumor cells (Worthington et al. 2012). It should be cautioned, given the continuous legacy of infectious disease exposure in Mesoamerica, any gene(s) in this region could have been the target of past selection. The third compelling result included two related regions residing on separate chromosomes 6 and 16. The chromosome 6 result was anticipated given the presence of the MHC, a known region of high genomic diversity that contains 224 genes largely related to immunity (Trowsdale 1993; de Bakker et al. 2006). The MHC complex has been identified numerous times in natural selection scans performed in human populations and across other mammalian and aquatic species (Hughes and Yeager 1998b). We hypothesized that the haplotype, HLA-DPA101:03/DPB104:02, is most likely the target of selection given its primary role as a cell surface receptor in antigen-presenting cells—crucial to recognizing foreign pathogens. Hepatitis B (HB) may have driven selection on this haplotype across time given the continuous presence of pre- and post-colonial lineages of the virus. Both HLA-DPA101:03 and HLA-DPB104:02 alleles independently have been shown to be protective for HB infection and known to play a role in developing long-term seroprotective immunity following HB vaccination among East Asian populations (Chung et al. 2019; Ou et al. 2019, 2021; Wang et al. 2019, 2021; Sanchez-Mazas 2020). HB infection previously was thought to have originated in the Americas, but ancient DNA analysis has demonstrated that it most likely co-evolved with humans as we dispersed across the globe (Muhlemann et al. 2018). Therefore, lineages existing in the Americas and novel HB lineages introduced through European contact, in conjunction with shifting social demographics, likely shaped the HLA diversity among Indigenous American populations although uncertain to know for certain (Guzman-Solis et al. 2021). The second related region on chromosome 16 contained the genes, class II, major histocompatibility complex transactivator (CIITA), known to positively regulate chromosome 6 MHC Class II expression, and suppressor of cytokine signaling 1 (SOCS1) (Reith et al. 2005; Krawczyk and Reith 2006; Devaiah and Singer 2013). SOCS1 activation inhibits CIITA activation and therefore subsequent MHC Class II expression as part of the IFN-γ pathway (O'Keefe et al. 2001). We identified a cluster of SNPS exhibiting extreme LSBL values residing in the lincRNA, RP11-396B14.2. The gene targets of this lincRNA are currently unknown, but it lies immediately upstream of SOCS1. This provides evidence for natural selection acting on variation affecting transcription. Together, these two related windows on chromosomes 6 and 16 illustrate the potential importance of selection acting on complimentary regions. Given the lack of publicly available data for larger cohorts of Indigenous American populations, we did not compare Mesoamerican genomes to other Indigenous American genomes to identify region-specific selective events. Therefore, our study design was unable to distinguish if a selective event was specific to Mesoamericans or affected Indigenous American populations more broadly. However, by comparing our results to other research identifying evidence of selection in the Americas, we were able to identify genes or chromosomal regions that overlap across studies or were distinct to our analysis. One particularly noteworthy gene with overlapping evidence of selection in our Mesoamerican cohort as well as among the Amerindian ancestry component of Brazilians from Mychaleckyj et al. [ 2017] was CIITA. In two complementary LSBL analyses performed by Mychaleckyj et al. [ 2017], one promoter region SNP, rs6498115, and one intronic SNP, rs45601437, from this gene where among the topmost differentiated SNPs. rs6498115 was included on the Affymetrix 6.0 array, whereas rs45601437 was not. In our analysis, rs6498115 was among four SNPs in the CIITA region that fell in the top $1\%$ of the empirical distribution for LSBL. This overlap between our studies lends further support to the hypothesis that CIITA variation was the target of selection during the Asia to America migration or during the peopling of the Americas in a population ancestral to both Mesoamericans and Brazilians. Genes in pathways that control the body's response to infectious disease, as well as to climate, altitude, and metabolic traits, show the strongest selection signatures in the human genome (Sabeti et al. 2007; Grossman et al. 2013). Notably, our immune system is highly redundant and compensates for factors such as novel genetic variation that may be detrimental to specific pathways. Furthermore, genetic markers regulating the immune response are general and diverse in function. For these reasons, evolutionary changes in allele frequencies brought about by natural selection to more ancient pathogens are likely to affect the pathogenesis of modern infectious diseases. For instance, CIITA continues to be important by providing resistance factors to modern infectious diseases such as Ebola virus and SARS-CoV-2 (Bruchez et al. 2020), and IL18R1 has been shown to confer protection against more severe clinical dengue phenotypes through IL1α downregulation (Yeo et al. 2014). However, not all genomic variation is protective against modern infectious agents. Variants in SOCS1 increase susceptibility to and disease progression of Influenza A and SARS-CoV-2 (Bhattacharjee and Banerjee 2020; Johnson et al. 2020; Lee et al. 2020). Therefore, we can leverage regions of the genome showing signatures of selection to identify resistance and/or susceptibility loci to modern pathogenic infection. This approach can be a particularly attractive strategy for studies with a limited-sized study population (Werren et al. 2021). In fact, focusing on genes under selection has proven beneficial in smaller sample sizes as demonstrated by several studies taking this approach (Park et al. 2012; Schwarzenbacher et al. 2012; Karlsson et al. 2013; Perry et al. 2014). The immune response genes identified here can provide an excellent starting point for genomic susceptibility studies of infectious diseases burdening modern Mesoamerican populations, while also providing greater statistical power to test fewer variants in smaller cohorts. Furthermore, they may be useful in studies seeking to understand cross-immunity between various infectious diseases of the period. Similarly, targeting immune response genes subject to past natural selection can aid in the study of population-specific variants related to metabolic disease, autoimmune disease, or cancer given that many of the pathways are overlapping. For example, PPARG and CYP7A1 regulate cholesterol homeostasis and metabolism, with documented effects of CYP7A1 polymorphisms on statin metabolism across worldwide populations (Chinetti et al. 2001; Kajinami et al. 2004; Thompson et al. 2005; Baker et al. 2010; Wei et al. 2011; Li et al. 2013; Kadam et al. 2016). The IL-17 pathway, of which IL17F is a part of, is an important target for various autoimmune disorders (Hu et al. 2011), and a variant in GALNT10 is highly associated with asthma susceptibility in a meta-analysis of populations of Latin American ancestry (Torgerson et al. 2011). Follow-up studies to our selection scan using highly differentiated alleles in populations of Mesoamerican ancestry would increase statistical power to identify associations with complex disease as exemplified by the study of Ko et al. [ 2014] on risk alleles involved in dyslipidemia. There were several limitations to our study. First, our results are based on SNP microarray data, which inherently suffers from ascertainment bias. The Affymetrix 6.0 chip was designed to capture the diversity and haplotype structure of the HapMap Project populations, European Americans of northern and western European descent (CEU), East Asians (JPT and CHB), and Yorubans (YRI). Linkage disequilibrium blocks and SNP distribution is expected to differ in Mesoamerican populations. Therefore, our analysis may have failed to identify candidate genes and gene regions for natural selection, but it is unlikely to suffer from a high rate of false positivity. Additionally, this SNP microarray has been used successfully to identify signatures of natural selection in several human populations including Indigenous Americans, cementing its usefulness in population genomics studies of natural selection (Bigham et al. 2010). Second, the haplotype tests used in our analysis required ancestral allele information for each SNP. Any SNP without this information was removed from the XP-EHH and iHS analyses. As a result, several chromosomal windows contained insufficient SNP density for calculations of XP-EHH and iHS. Lastly, our analysis for effective population size was limited due to the number of individuals included in the analysis and the array ascertainment bias, leading to wider confidence intervals. However, a population bottleneck with much smaller confidence intervals was clearly visible, coinciding with the time period of colonial contact. A better designed SNP array or the interrogation of sequencing data would remedy these caveats in future studies. ## Conclusions We present the results of a natural selection scan performed in Indigenous Mesoamerican populations from Mexico. We find evidence for a population bottleneck coincident with the arrival of Europeans to the Americas and natural selection in genes related to both adaptive and innate immunity. We suggest that past selective events influence the host response to modern diseases, both pathogenic infection as well as autoimmune disorders. Therefore, searching for signatures of past natural selection in genes related to immune function is a particularly attractive strategy for identifying host genetic factors influencing both susceptibility and resistance to disease. Together, our findings provide valuable insight into Mesoamerican population history and identify candidate loci for studying localized, biological responses to modern infectious and autoimmune disease. ## Populations Our Mesoamerican cohort included a total of 39 individuals representing the following populations: Twenty-five Maya from the Yucatan Peninsula of Mexico, two Nahua, seven Mixtec, and five Tlapanec speakers from Guerrero, Mexico previously described in Bigham et al. [ 2010]. We obtained publicly available data from The 1000 Genomes Project Consortium for the following control populations: 60 Europeans of Northern and Western European ancestry (CEU), 90 East Asians from Beijing, China (CHB) and Tokyo, Japan (JPT), and 90 Yoruba from Ibadan, Nigeria (YRI) (International HapMap Project 2003; The 1000 Genomes Project Consortium 2012, 2015). ## Genome-wide SNP Data All samples were previously genotyped using the Affymetrix Genome-Wide Human SNP Array 6.0 containing 906,600 SNPs (Bigham et al. 2010). We analyzed autosomal SNPs with call rates >95. The X and Y-chromosome as well as mitochondrial DNA (mtDNA) SNPs were excluded from our analyses as we chose to focus on the autosomes. No SNPs were removed based on departure from Hardy-Weinberg equilibrium (HWE) as this could potentially remove SNPs under selection that would mimic HWE departures. After QC, we carried out statistical analysis using 857,481 autosomal SNPs. ## Phasing and File Manipulation All files were haplotype-phased using SHAPEIT4, processed using PLINK $\frac{1.9}{2.0}$, and manipulated using VCFtools and BCFtools (Purcell et al. 2007; Li et al. 2009; Danecek et al. 2011; Chang et al. 2015; Delaneau et al. 2019). 1000 Genomes Project phase 3 populations were used for phasing the SNP data (The 1000 Genomes Project Consortium 2015). ## Relatedness We calculated relatedness using kinship coefficients as estimated with the Kinship-based INference for Genome-wide association studies (KING) (Manichaikul et al. 2010). We removed six individuals from our dataset that were first, second, and third-degree relatives, leaving us with a sample size of $$n = 33$$ individuals. ## Admixture Analysis PCA was conducted in PLINK 1.9 (Purcell et al. 2007; Chang et al. 2015). Global estimate of admixture for each Mesoamerican individual ($$n = 39$$) was calculated using an unsupervised model in ADMIXTURE ($K = 4$) (Alexander and Lange 2011). Chromosomal segment ancestry was estimated using RFMix2, assuming approximately 20 generations since initial admixture, which corresponds to admixture on Spanish encomiendas in the Yucatan (Machuca Gallegos 2016); (Maples et al. 2013). A 3-population admixture model between Indigenous Americans (SGDP), Africans (YRI), and Europeans (CEU) was assumed, as we believe the segments containing East Asian ancestry were due to shared ancestry and not a result of admixture. We used the Viterbi segment assignments to extract haplotypes demonstrating admixture, set genotypes in these regions as missing using bedtools, and imputed the missing segments on each chromosome with PBWT imputation in SHAPEIT4 using a customized reference panel (Quinlan and Hall 2010; Quinlan 2014; Delaneau et al. 2019). This customized reference panel was designed on a chromosome-by-chromosome basis comprised of our own unadmixed Mesoamerican individuals for that chromosome to impute the “missing” segments (supplementary table S2, Supplementary Material online). Given that the proportions of admixture were small, we imputed off our own dataset as the homogenization of haplotypes would be minimal. ## Ancestral Alleles Ancestral alleles were queried from the 1000 Genomes Project phase 3 VCF files using BCFtools (Li et al. 2009). VCF files were recoded using PLINK 2.0 to preserve phasing information (Purcell et al. 2007; Chang et al. 2015). The dataset used for XP-EHH and iHS contained 841,217 SNPs, as only SNPs with ancestral allele information were used for haplotype testing. ## Estimating Historical Effective Population Size To estimate effective population size in our cohort, and account for admixture, we used the Ancestry-specific Identity by Descent Effective Population size (AS-IBDne) (Browning et al. 2018). We used van Eeden et al. [ 2022]'s adapted snakemake AS-IBDne pipeline to calculate the Indigenous American effective population size in our 33 Mesoamericans (https://github.com/hennlab/AS-IBDNe). The primary deviation from the original AS-IBDne method is that this pipeline incorporates the local ancestry from RFMix2 instead of the RFMix 1.5.4 output. For references, we used 22 Indigenous Americans from the SGDP project, 22 Han Chinese, 22 European Americans, and 22 Yoruba from the HapMap project (International HapMap Project 2003; Mallick et al. 2016). Breaks and gaps in the IBD segments caused by phasing or genotype errors were filtered using the merge-ibd-segments program as part of the Refined IBD suite, setting no more than one discordant homozygote, and removing IBD segments shorter than 0.6 cM (Browning and Browning 2013). Historical effective population size and its $95\%$ confidence were calculated using default parameters with a filter to analyze segments larger than 4 cM as appropriate for array data in IBDne (Browning and Browning 2015). A generation time of 25 years was assumed to transform the generations ago into years before present. Results were restricted to 50 generations before present as IBDne underestimates effective population size for SNP array data (supplementary table S3, Supplementary Material online; Browning and Browning 2015). After QC and filtering our IBD results, we only yielded AS-IBDne results for Indigenous American, European, and African ancestry. No putative East Asian segments passed our filters. As we are primarily interested in the effective population size of our Mesoamerican cohort, we only used the output corresponding to Indigenous American ancestry. ## Selection Scan We employed three statistics to identify regions in the genome showing statistical evidence of natural selection: 1) LSBL (Shriver et al. 2004), 2) XP-EHH (Sabeti et al. 2007; Pickrell et al. 2009), and 3) iHS (Voight et al. 2006). LSBL compared Mesoamericans against European Americans and East Asians. We filtered the dataset for SNPs with an MAF > 0.05, which left us with 497,699 SNPs to analyze for LSBL. Fst values were computed for each SNP using Weir-Cockerham's equation (Weir and Cockerham 1984; Shriver et al. 2004; Akey 2009; Bigham et al. 2010). Statistical significance was determined using an empirical distribution. PE(x) = (number of loci > x)/(total number loci) using a significance threshold of α = 0.01 (Akey et al. 2002). LSBL results were then aggregated into 100 kilobase pair windows, that matched with the XP-EHH and iHS coordinates. XP-EHH and iHS were calculated in Selscan (Szpiech and Hernandez 2014). XP-EHH was calculated for 826,691 autosomal SNPs, whereas iHS was calculated for 455,845 autosomal SNPs after filtering out low-frequency variants as this statistic was not designed to capture low-frequency variant information or alleles near fixation. XP-EHH was genome-wide normalized using the norm function. iHS was standardized based on allele frequency bins, normalizing the SNPs in quantiles organized by similar frequencies, again using the norm function. We grouped each haplotype statistic into non-overlapping windows of 100 kb pairs. We identified regions with the longest haplotypes reaching significance thresholds of α = 0.01. For XP-EHH, we compared Mesoamericans to East Asians to look specifically for haplotypes present in Mesoamerican populations that arose after their split from Asian populations. For iHS, the windows were binned by the number of SNPs for quantile estimation of percentile using Selscan's norm function. iHS scores were not computed for a MAF < 0.05. Windows with fewer than 10 SNPs were dropped from analysis, which included the loss of 18,779 SNPs in the iHS dataset and 24,245 in the XP-EHH dataset. Only windows in the $1\%$ tail of bin distribution were considered. We found 316 XP-EHH windows and 203 iHS windows significant at the $1\%$ level. LSBL was paired the XP-EHH and iHS haplotype tests. Only windows that fell within the $1\%$ tail for both paired tests were considered as candidates for positive selection. Here, we find that only 57 iHS plus LSBL windows and 100 XP-EHH and iHS windows passed this threshold. We tested how admixture and imputing from our own dataset affected our allele frequency analyses by using the program Ohana with our merged dataset consisting of the Mesoamerican cohort plus CEU, YRI, CHB + JPT. This program analyzes population structure and outputs admixture-corrected allele frequencies (Cheng et al. 2022). We conducted a supervised population structure analysis with $K = 4$, and then used the output matrix of admixture-corrected allele frequencies to calculate pairwise Fst and LSBL for each SNP for the same populations used above. These results validated our original LSBL findings for our candidate regions under selection. As our Allele frequency test was unaffected by the imputation, we determined that our imputed dataset had been minimally affected by cryptic population substructure by homogenizing certain regions. To test this, we re-ran iHS and XP-EHH and normalized for the entire dataset with admixed individuals at the locus of interest, if the region was still in the $1\%$ further analysis was not necessary. If the region dropped out of the $1\%$ significance, a separate run was conducted for a subset of individuals who were unadmixed at the window. To determine who was unadmixed at each specific locus, we used a 4-population admixture model between Indigenous Americans (SGDP), Africans (YRI), Han Chinese (CHB), and Europeans (CEU). We considered each of the window assignments, pulling out only those individuals for which an Indigenous American ancestry was reported in both the paternally and maternally inherited chromosomal haplotypes. ## Annotation of Regions For both the selection scan and introgression analysis, windows were annotated for genes using the bedmaps option from BEDOPS tools (Neph et al. 2012). ## HLA Allele Calls HLA allele calls were imputed using the multi-ethnic HLA (version 1.0 2021) reference panel as part of the Michigan Imputation Server HLA-TAPAS pipeline (Das et al. 2016; Luo et al. 2021). QC criteria used were MAF > 0.01 and rsq > 0.3 (Pistis et al. 2015). To confirm accuracy of the HLA calls, we downloaded the high coverage alignment files for the HGDP and SGDP individuals from Central/Southern Mexico through the European Bioinformatics Institute (EMBL-EBI) endpoint on GLOBUS (Mallick et al. 2016; Bergström et al. 2020). These individuals included 21 Maya, 3 Mixe, 2 Mixtec, and 2 Zapotec. HLA calls were generated using HLA-LA, which takes the alignment files and realigns them to a graph genome (Dilthey et al. 2019). One Zapotec sample failed to run and was removed from our analysis. Each individual's HLA-DPA1 and DPB1 calls with coverage and probability are available in the supplementals. ## Estimating Haplotype Ages To estimate the haplotype age, we used a method that employs EHH scores and assumes a starlike phylogeny to assess the age of decay from a core marker (Reich and Goldstein 1999; Sabeti et al. 2002b; Voight et al. 2006). We calculated the EHH statistic for our core SNPs using Selscan v1.3.0 until EHH decay reached 0.05 (Hardwick et al. 2014; Szpiech and Hernandez 2014). Given that EHH≈Pr(Homozygosity), or the probability of homozygosity, we can use the following equation: Pr(Homozygosity)=e−2RG, R = haplotype length in M (morgans), G = generation time marker (Reich and Goldstein 1999; Sabeti et al. 2002b; Voight et al. 2006). This equation can be rearranged and reduced to a simple slope-intercept form (y = mx + b) through the origin ($b = 0$) by taking the natural log of the EHH values and doubling the distance in morgans from the core, which gives us the equation −ln(EHH) = G × 2R. This rearrangement of the data allows us to determine the generation time (slope of equation) since the haplotype arose using a linear regression (Hardwick et al. 2014). We calculated the regression coefficients using a linear model through the origin and generated a $95\%$ confidence in R (Team 2013). The raw output, with residuals and coefficients, is available in the supplementals. To calculate a rough estimate of age in years, we assumed a 25-year generation time. ## Supplementary Material Supplementary data are available at Genome Biology and Evolution online (http://www.gbe.oxfordjournals.org/). ## Data Availability Data will be shared on the author's GitHub page: https://github.com/obedaram/Mesoamerican-Data. ## References 1. An integrated map of genetic variation from 1,092 human genomes. Nature. (2012) 491 56-65. PMID: 23128226 2. A global reference for human genetic variation. Nature. (2015) 526 68-74. PMID: 26432245 3. Akey JM. Constructing genomic maps of positive selection in humans: where do we go from here?. Genome Res. (2009) 19 711-722. PMID: 19411596 4. Akey JM, Zhang G, Zhang K, Jin L, Shriver MD. Interrogating a high-density SNP map for signatures of natural selection. Genome Res. (2002) 12 1805-1814. PMID: 12466284 5. Alexander DH, Lange K. Enhancements to the ADMIXTURE algorithm for individual ancestry estimation. BMC Bioinformatics. (2011) 12 246. PMID: 21682921 6. Avila-Arcos MC. Population history and gene divergence in native Mexicans inferred from 76 human exomes. Mol Biol Evol. (2020) 37 994-1006. PMID: 31848607 7. Baker AD. PPARgamma regulates the expression of cholesterol metabolism genes in alveolar macrophages. Biochem Biophys Res Commun. (2010) 393 682-687. PMID: 20170635 8. Bamshad MJ. A strong signature of balancing selection in the 5′. Proc Natl Acad Sci U S A. (2002) 99 10539-10544. PMID: 12149450 9. Bergström A. Insights into human genetic variation and population history from 929 diverse genomes. Science. (2020) 367 10. Bhattacharjee S, Banerjee M. Immune thrombocytopenia secondary to COVID-19: a systematic review. SN Compr Clin Med. (2020) 2 2048-2058. PMID: 32984764 11. Bigham A. Identifying signatures of natural selection in Tibetan and Andean populations using dense genome scan data. PLoS Genet. (2010) 6 12. Bos KI. Pre-Columbian mycobacterial genomes reveal seals as a source of New World human tuberculosis. Nature. (2014) 514 494-497. PMID: 25141181 13. Browning SR. Ancestry-specific recent effective population size in the Americas. PLoS Genet. (2018) 14 14. Browning BL, Browning SR. Improving the accuracy and efficiency of identity-by-descent detection in population data. Genetics. (2013) 194 459-471. PMID: 23535385 15. Browning SR, Browning BL. Accurate non-parametric estimation of recent effective population size from segments of identity by descent. Am J Hum Genet. (2015) 97 404-418. PMID: 26299365 16. Bruchez A. MHC Class II transactivator CIITA induces cell resistance to Ebola virus and SARS-like coronaviruses. Science. (2020) 370 241-247. PMID: 32855215 17. Bryc K. Colloquium paper: genome-wide patterns of population structure and admixture among Hispanic/Latino populations. Proc Natl Acad Sci U S A. (2010) 107 8954-8961. PMID: 20445096 18. Campbell L. American Indian languages: the historical linguistics of Native America (2000) 19. Chang CC. Second-generation PLINK: rising to the challenge of larger and richer datasets. Gigascience. (2015) 4 7. PMID: 25722852 20. Chawla A. Control of macrophage activation and function by PPARs. Circ Res. (2010) 106 1559-1569. PMID: 20508200 21. Chelala C, Khan A, Lemoine NR. SNPnexus: a web database for functional annotation of newly discovered and public domain single nucleotide polymorphisms. Bioinformatics. (2009) 25 655-661. PMID: 19098027 22. Cheng JY, Stern AJ, Racimo F, Nielsen R. Detecting selection in multiple populations by modeling ancestral admixture components. Mol Biol Evol. (2022) 39 23. Chinetti G. PPAR-alpha and PPAR-gamma activators induce cholesterol removal from human macrophage foam cells through stimulation of the ABCA1 pathway. Nat Med. (2001) 7 53-58. PMID: 11135616 24. Chung S. GWAS Identifying. J Viral Hepat. (2019) 26 1318-1329. PMID: 31243853 25. Crawford JE. Natural selection on genes related to cardiovascular health in high-altitude adapted Andeans. Am J Hum Genet. (2017) 101 752-767. PMID: 29100088 26. Danecek P. The variant call format and VCFtools. Bioinformatics. (2011) 27 2156-2158. PMID: 21653522 27. Das S. Next-generation genotype imputation service and methods. Nat Genet. (2016) 48 1284-1287. PMID: 27571263 28. Dayem Ullah AZ. SNPnexus: assessing the functional relevance of genetic variation to facilitate the promise of precision medicine. Nucleic Acids Res. (2018) 46 W109-W113. PMID: 29757393 29. de Bakker PI. A high-resolution HLA and SNP haplotype map for disease association studies in the extended human MHC. Nat Genet. (2006) 38 1166-1172. PMID: 16998491 30. Delaneau O, Zagury JF, Robinson MR, Marchini JL, Dermitzakis ET. Accurate, scalable and integrative haplotype estimation. Nat Commun. (2019) 10 5436. PMID: 31780650 31. Devaiah BN, Singer DS. CIITA and its dual roles in MHC gene transcription. Front Immunol. (2013) 4 476. PMID: 24391648 32. Dilthey AT. *HLALA-HLA typing from linearly projected graph alignments*. Bioinformatics.* (2019) 35 4394-4396. PMID: 30942877 33. Eichstaedt CA. The Andean adaptive toolkit to counteract high altitude maladaptation: genome-wide and phenotypic analysis of the Collas. PLoS ONE. (2014) 9 34. Fan S, Hansen ME, Lo Y, Tishkoff SA. Going global by adapting local: a review of recent human adaptation. Science. (2016) 354 54-59. PMID: 27846491 35. Feldman LH. The war against epidemics in colonial Guatemala, 1519-1821 (1999) 36. García-Ortiz H. The genomic landscape of Mexican Indigenous populations brings insights into the peopling of the Americas. Nat Commun. (2021) 12 5942. PMID: 34642312 37. Gonzalez-Galarza FF. Allele frequency net database (AFND) 2020 update: gold-standard data classification, open access genotype data and new query tools. Nucleic Acids Res. (2019) 48 D783-D788 38. Grossman SR. Identifying recent adaptations in large-scale genomic data. Cell. (2013) 152 703-713. PMID: 23415221 39. Guzman-Solis AA. Ancient viral genomes reveal introduction of human pathogenic viruses into Mexico during the transatlantic slave trade. Elife. (2021) 10 40. Hardwick RJ. Haptoglobin (HP) and Haptoglobin-related protein (HPR) copy number variation, natural selection, and trypanosomiasis. Hum Genet. (2014) 133 69-83. PMID: 24005574 41. Hu Y, Shen F, Crellin NK, Ouyang W. The IL-17 pathway as a major therapeutic target in autoimmune diseases. Ann N Y Acad Sci. (2011) 1217 60-76. PMID: 21155836 42. Huang DW, Sherman BT, Lempicki RA. Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Res. (2009a) 37 1-13. PMID: 19033363 43. Huang DW, Sherman BT, Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc. (2009b) 4 44-57. PMID: 19131956 44. Hughes AL, Yeager M. Natural selection and the evolutionary history of major histocompatibility complex loci. Front Biosci. (1998a) 3 d509-d516. PMID: 9601106 45. Hughes AL, Yeager M. Natural selection at major histocompatibility complex loci of vertebrates. Annu Rev Genet. (1998b) 32 415-435. PMID: 9928486 46. The International HapMap Project. Nature. (2003) 426 789-796. PMID: 14685227 47. Jassal B. The reactome pathway knowledgebase. Nucleic Acids Res. (2019) 48 D498-D503 48. Johnson HM, Lewin AS, Ahmed CM. SOCS, intrinsic virulence factors, and treatment of COVID-19. Front Immunol. (2020) 11 49. Kadam P, Ashavaid TF, Ponde CK, Rajani RM. Genetic determinants of lipid-lowering response to atorvastatin therapy in an Indian population. J Clin Pharm Ther. (2016) 41 329-333. PMID: 26932749 50. Kajinami K, Brousseau ME, Ordovas JM, Schaefer EJ. Interactions between common genetic polymorphisms in ABCG5/G8 and CYP7A1 on LDL cholesterol-lowering response to atorvastatin. Atherosclerosis. (2004) 175 287-293. PMID: 15262185 51. Kakoola DN, Curcio-Brint A, Lenchik NI, Gerling IC. Molecular pathway alterations in CD4 T-cells of nonobese diabetic (NOD) mice in the preinsulitis phase of autoimmune diabetes. Results Immunol. (2014) 4 30-45. PMID: 24918037 52. Karlsson EK. Natural selection in a Bangladeshi population from the cholera-endemic Ganges river delta. Sci Transl Med. (2013) 5 53. Klaus HD. Tuberculosis on the north coast of Peru: skeletal and molecular paleopathology of late pre-Hispanic and postcontact mycobacterial disease. J Archaeol Sci. (2010) 37 2587-2597 54. Ko A. Amerindian-specific regions under positive selection harbour new lipid variants in Latinos. Nat Commun. (2014) 5 3983. PMID: 24886709 55. Krawczyk M, Reith W. Regulation of MHC class II expression, a unique regulatory system identified by the study of a primary immunodeficiency disease. Tissue Antigens. (2006) 67 183-197. PMID: 16573555 56. Lee PY. Immune dysregulation and multisystem inflammatory syndrome in children (MIS-C) in individuals with haploinsufficiency of SOCS1. J Allergy Clin Immunol. (2020) 146 1194-1200.e1. PMID: 32853638 57. Le Menn G, Neels JG. Regulation of immune cell function by PPARs and the connection with metabolic and neurodegenerative diseases. Int J Mol Sci. (2018) 19 1575. PMID: 29799467 58. Leon-Portilla M. The broken spears 2007 revised edition: the Aztec account of the conquest of Mexico (2011) 59. Li H. The sequence alignment/map format and SAMtools. Bioinformatics. (2009) 25 2078-2079. PMID: 19505943 60. Li T, Francl JM, Boehme S, Chiang JY. Regulation of cholesterol and bile acid homeostasis by the cholesterol 7alpha-hydroxylase/steroid response element-binding protein 2/microRNA-33a axis in mice. Hepatology. (2013) 58 1111-1121. PMID: 23536474 61. Lindo J. A time transect of exomes from a Native American population before and after European contact. Nat Commun (2016) 7 13175. PMID: 27845766 62. Livi-Bacci M. The depopulation of Hispanic America after the conquest. Popul Dev Rev. (2006) 32 199-232 63. Lockhart J. The Nahuas after the conquest: a social and cultural history of the Indians of central Mexico, sixteenth through eighteenth centuries (1992) 64. Luo Y. A high-resolution HLA reference panel capturing global population diversity enables multi-ancestry fine-mapping in HIV host response. Nat Genet. (2021) 53 1504-1516. PMID: 34611364 65. Machuca Gallegos L. El ocaso de la encomienda en Yucatán, 1770-1821. Estud Hist Novohisp. (2016) 54 31-49 66. Magalhaes TR. HGDP and HapMap analysis by Ancestry Mapper reveals local and global population relationships. PLoS ONE. (2012) 7 67. Mallick S. The Simons Genome Diversity Project: 300 genomes from 142 diverse populations. Nature. (2016) 538 201-206. PMID: 27654912 68. Manichaikul A. Robust relationship inference in genome-wide association studies. Bioinformatics. (2010) 26 2867-2873. PMID: 20926424 69. Maples BK, Gravel S, Kenny EE, Bustamante CD. RFMix: a discriminative modeling approach for rapid and robust local-ancestry inference. Am J Hum Genet. (2013) 93 278-288. PMID: 23910464 70. Merbs CF. A new world of infectious disease. Am J Phys Anthropol. (1992) 35 3-42 71. Mooney JA. Understanding the hidden complexity of Latin American population isolates. Am J Hum Genet. (2018) 103 707-726. PMID: 30401458 72. Muhlemann B. Ancient hepatitis B viruses from the Bronze Age to the Medieval period. Nature. (2018) 557 418-423. PMID: 29743673 73. Mummert A, Esche E, Robinson J, Armelagos GJ. Stature and robusticity during the agricultural transition: evidence from the bioarchaeological record. Econ Hum Biol. (2011) 9 284-301. PMID: 21507735 74. Mychaleckyj JC. Genome-wide analysis in Brazilians reveals highly differentiated Native American genome regions. Mol Biol Evol. (2017) 34 msw249-msw574 75. Neph S. BEDOPS: high-performance genomic feature operations. Bioinformatics. (2012) 28 1919-1920. PMID: 22576172 76. O’Fallon BD, Fehren-Schmitz L. Native Americans experienced a strong population bottleneck coincident with European contact. Proc Natl Acad Sci U S A. (2011) 108 20444-20448. PMID: 22143784 77. O’Keefe GM, Nguyen VT, Ping Tang LL, Benveniste EN. IFN-gamma regulation of class II transactivator promoter IV in macrophages and microglia: involvement of the suppressors of cytokine signaling-1 protein. J Immunol. (2001) 166 2260-2269. PMID: 11160280 78. Ou G. Relationship between. J Viral Hepat. (2019) 26 155-161. PMID: 30267609 79. Ou G. Variation and expression of HLA-DPB1 gene in HBV infection. Immunogenetics. (2021) 73 253-261. PMID: 33710355 80. Park DJ. Sequence-based association and selection scans identify drug resistance loci in the. Proc Natl Acad Sci U S A. (2012) 109 13052-13057. PMID: 22826220 81. Perry GH. Adaptive, convergent origins of the pygmy phenotype in African rainforest hunter-gatherers. Proc Natl Acad Sci U S A. (2014) 111 82. Pickrell JK. Signals of recent positive selection in a worldwide sample of human populations. Genome Res. (2009) 19 826-837. PMID: 19307593 83. Pistis G. Rare variant genotype imputation with thousands of study-specific whole-genome sequences: implications for cost-effective study designs. Eur J Hum Genet. (2015) 23 975-983. PMID: 25293720 84. Purcell S. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet. (2007) 81 559-575. PMID: 17701901 85. Quinlan AR. BEDTools: the Swiss-army tool for genome feature analysis. Curr Protoc Bioinformatics. (2014) 47 86. Quinlan AR, Hall IM. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics. (2010) 26 841-842. PMID: 20110278 87. Reich DE, Goldstein DB. Goldstein DB, Schlötterer C, editors. Microsatellites: evolution and applications (1999) 129-138 88. Reith W, LeibundGut-Landmann S, Waldburger JM. Regulation of MHC class II gene expression by the class II transactivator. Nat Rev Immunol. (2005) 5 793-806. PMID: 16200082 89. Restall M, Sousa L, Terraciano K. Mesoamerican voices: native-language writings from Colonial Mexico, Oaxaca, Yucatan, and Guatemala (2005) 90. Reynolds AW. Comparing signals of natural selection between three Indigenous North American populations. Proc Natl Acad Sci U S A. (2019) 116 9312-9317. PMID: 30988184 91. Rodríguez-Rodríguez JE. The genetic legacy of the Manila galleon trade in Mexico. Philos Trans R Soc B Biol Sci. (2022) 377 92. Sabeti P. CD40L Association with protection from severe malaria. Genes Immun. (2002a) 3 286-291. PMID: 12140747 93. Sabeti PC. Detecting recent positive selection in the human genome from haplotype structure. Nature. (2002b) 419 832-837. PMID: 12397357 94. Sabeti PC. Genome-wide detection and characterization of positive selection in human populations. Nature. (2007) 449 913-918. PMID: 17943131 95. Sanchez-Mazas A. A review of HLA allele and SNP associations with highly prevalent infectious diseases in human populations. Swiss Med Wkly. (2020) 150 96. Sarmah N, Baruah MN, Baruah S. Immune modulation in HLA-G expressing head and neck squamous cell carcinoma in relation to human papilloma virus positivity: a study from northeast India. Front Oncol. (2019) 9 58. PMID: 30859089 97. Schwarzenbacher H. Combining evidence of selection with association analysis increases power to detect regions influencing complex traits in dairy cattle. BMC Genomics. (2012) 13 48. PMID: 22289501 98. Shriver MD. The genomic distribution of population substructure in four populations using 8,525 autosomal SNPs. Hum Genomics. (2004) 1 274-286. PMID: 15588487 99. Smith ME. City size in late postclassic Mesoamerica. J Urban Hist. (2005) 31 403-434 100. Steverding D. The history of Chagas disease. Parasit Vectors. (2014) 7 317. PMID: 25011546 101. Szpiech ZA, Hernandez RD. Selscan: an efficient multithreaded program to perform EHH-based scans for positive selection. Mol Biol Evol. (2014) 31 2824-2827. PMID: 25015648 102. Team RC. (2013) 103. Thompson JF. An association study of 43 SNPs in 16 candidate genes with atorvastatin response. Pharmacogenomics J. (2005) 5 352-358. PMID: 16103896 104. Torgerson DG. Meta-analysis of genome-wide association studies of asthma in ethnically diverse North American populations. Nat Genet. (2011) 43 887-892. PMID: 21804549 105. Trowsdale J. Genomic structure and function in the MHC. Trends Genet. (1993) 9 117-122. PMID: 8516845 106. van Eeden G. The recombination landscape of the Khoe-San likely represents the upper limits of recombination divergence in humans. Genome Biol. (2022) 23 172. PMID: 35945619 107. Voight BF, Kudaravalli S, Wen X, Pritchard JK. A map of recent positive selection in the human genome. PLoS Biol. (2006) 4 e72. PMID: 16494531 108. Wang LY. Response to hepatitis B vaccination is co-determined by HLA-DPA1 and -DPB1. Vaccine. (2019) 37 6435-6440. PMID: 31515149 109. Wang WC. Association of HLA-DPA1, HLA-DPB1, and HLA-DQB1 alleles with the long-term and booster immune responses of young adults vaccinated against the hepatitis B virus as neonates. Front Immunol. (2021) 12 110. Wei KK, Zhang LR, Zhang Y, Hu XJ. Interactions between CYP7A1 A-204C and ABCG8 C1199A polymorphisms on lipid lowering with atorvastatin. J Clin Pharm Ther. (2011) 36 725-733. PMID: 21128988 111. Weir BS, Cockerham CC. Estimating. Evolution. (1984) 38 1358-1370. PMID: 28563791 112. Werren EA, Garcia O, Bigham AW. Identifying adaptive alleles in the human genome: from selection mapping to functional validation. Hum Genet. (2021) 140 241-276. PMID: 32728809 113. Worthington JJ, Fenton TM, Czajkowska BI, Klementowicz JE, Travis MA. Regulation of TGFbeta in the immune system: an emerging role for integrins and dendritic cells. Immunobiology. (2012) 217 1259-1265. PMID: 22902140 114. Yeo AS. Lack of clinical manifestations in asymptomatic dengue infection is attributed to broad down-regulation and selective up-regulation of host defence response genes. PLoS ONE. (2014) 9 115. Zhang G. Elevated GALNT10 expression identifies immunosuppressive microenvironment and dismal prognosis of patients with high grade serous ovarian cancer. Cancer Immunol Immunother. (2020) 69 175-187. PMID: 31853576
--- title: 'The differences in drug resistance between drug-resistant tuberculosis patients with and without diabetes mellitus in northeast China: a retrospective study' authors: - Yuanping Pan - Yingying Yu - Yaohui Yi - Xiaofeng Dou - Jiachen Lu - Ling Zhou journal: BMC Infectious Diseases year: 2023 pmcid: PMC10016172 doi: 10.1186/s12879-023-08130-1 license: CC BY 4.0 --- # The differences in drug resistance between drug-resistant tuberculosis patients with and without diabetes mellitus in northeast China: a retrospective study ## Abstract ### Background Diabetes mellitus (DM) and drug-resistant tuberculosis (DR-TB) are serious global public health problems. This study aimed to explore the differences in drug resistance between DR-TB patients with and without DM. Risk factors for developing multidrug-resistant tuberculosis (MDR-TB) were also investigated among DR-TB patients. ### Methods The patient’s basic demographic, clinical characteristics, and drug susceptibility testing (DST) data were collected from the Chinese Disease Control Information System. Descriptive statistics were used to estimate the frequency and proportion of included variables. Categorical variables were compared using the Chi-square test or Fisher’s exact test. Chi-square tests for trends were used to determine changes and trends in MDR-TB and pre-extensively drug-resistantTB (pre-XDR-TB) patterns over time. Univariate and multivariate logistic regression analysis was used to explore the risk factors of MDR-TB. ### Results Compared with DR-TB patients with DM, DR-TB patients without DM had significantly higher rates of mono-resistant streptomycin (SM) and any resistance to kanamycin (KM), but significantly lower rates of any resistance to protionamide (PTO) and mono-resistance to levofloxacin (LFX), and pre-XDR-TB ($P \leq 0.05$). The proportion of resistance to other anti-TB drugs was not statistically different between the DR-TB with and without DM. Among DR-TB patients without and with DM, the proportion of patients with MDR-TB and pre-XDR-TB patterns showed a significant downward trend from 2016 to 2021 ($P \leq 0.05$). Among DR-TB patients without DM, male, previously treated DR-TB cases, and immigration were risk factors for MDR-TB ($P \leq 0.05$). In DR-TB patients with DM, a negative sputum smear is a risk factor for MDR-TB ($P \leq 0.05$). ### Conclusion There was no statistical difference in resistance patterns between DR-TB with and without DM, except in arbitrary resistance to PTO and KM, mono-resistant SM and LFX, and pre-XDR-TB. Great progress has been made in the prevention and control of MDR-TB and pre-XDR-TB. However, DR-TB patients with and without DM differ in their risk factors for developing MDR-TB. ## Background Tuberculosis (TB) is an infectious disease caused by Mycobacterium tuberculosis which seriously damages human health, especially in some low- and middle-income population. In 2021, 10.6 million new cases of TB were diagnosed globally, with an incidence rate of 134 per 100,000 and 1.6 million deaths from TB[1]. The World Health Organization (WHO) proposed a strategy to eliminate TB, which was to reduce TB mortality to less than $95\%$ and incidence to $90\%$ by 2035[2]. However, the emergence of drug-resistant TB (DR-TB) has seriously hindered the prevention and treatment of TB[3]. DR-TB is now becoming the world’s deadliest pathogen, with a quarter of deaths attributed to antimicrobial drug resistance[4]. There are nearly 5000,000 patients with rifampicin-resistance TB (RR-TB) worldwide, of which $78\%$ were multidrug-resistant TB (MDR-TB), defined as resistance to at least isoniazid (INH) and rifampicin (RFP), reported in 2019[5]. Resistance to anti-TB drugs often means fewer treatments, poorer outcomes, and higher medical costs. Knowledge of drug resistance profiles can help us evaluate current TB control measures and develop more effective TB treatments. In countries with a high TB burden, an estimated $15\%$ of TB patients have diabetes mellitus (DM)[6]. With changing lifestyles and aging population, International Diabetes Federation (IDF) estimated that around 537 million adults worldwide will have DM in 2021, and that number is expected to rise to 643 million by 2030[7]. Due to the high prevalence of TB and DM, the double burden of TB and DM constitutes a global public health concern[8]. China has one of the highest TB burdens in the world, and previous studies have shown a significant burden of DR-TB in northeastern China[9, 10]. Moreover, northeastern China has a significant aging population and a high burden of DM[11]. The relationship between DM and TB has been investigated, and previous studies identified it as a risk factor for developing TB[12–14]. In addition, the impact of DM on TB treatment outcomes has been extensively studied and there is broad consensus that the relationship between DM and TB treatment outcomes.[15–17]. However, few studies have looked at the relationship between DR-TB and DM. A study was conducted to explore the profiles of TB patients with different DM statuses in eastern China, and it did not find an association between DM and DR-TB[18]. Conversely, other studies have found that DM is a risk factor for DR-TB in TB patients[8, 19]. Due to DM complicating the treatment of DR-TB, the research on this aspect is relatively insufficient. This study focused on the differences in drug resistance between DR-TB patients with and without DM, and finally explored the risk factors for developing MDR-TB among DR-TB patients with and without DM. ## Study population and data collection This was a retrospective study conducted at a specialized TB hospital in northeastern China. All sociodemographic, clinical, and drug susceptibility data from 1 to 2016 to 31 December 2021 were extracted from the China Disease Control and Prevention Information System which regularly collects information associated with DR-TB surveillance and management. The sociodemographic included sex (female/male), age (14~ $\frac{30}{31}$ ~ $\frac{44}{45}$ ~ $\frac{59}{60}$ aged and above), residence (urban/rural), nationality (Han/others), occupation (employed/unemployed), migrant (yes/no). Clinical data included patient category (primary DR-TB cases / previously treated DR-TB cases), sputum smear status (negative/active), and DM status. Drug susceptibility testing (DST) data included the testing results of first-line drugs and second-line drugs. Patients with negative sputum cultures and nontuberculous mycobacteria were excluded. Patients without DST results were excluded. Since HIV-positive patients are treated in specialized sentinel hospitals, data related to treatment are not available. Therefore, we excluded HIV-positive DR-TB patients. In addition, patients without results of DM also were excluded. A total of 513 patients were included in the final study, including 186 patients with DM and 327 patients without DM (Fig. 1). Fig. 1Flowchart for the inclusion of patientsAbbreviations: DR-TB, drug-resistant tuberculosis; DM, diabetes mellitus; DST, drug susceptibility testing. ## Drug Sensitivity Testing Sputum samples were cultured on Lowenstein-Jensen (LJ) culture media[20]. DST was conducted for first-line anti-TB drugs including INH, RFP, ethambutol (EMB), and streptomycin (SM), and for second-line drugs including protionamide (PTO), para-aminosalicylic acid (PAS), amikacin (AM), capreomycin (CM), kanamycin (KM), levofloxacin (LFX) and ofloxacin (OFX). The following concentrations of anti-TB drugs were administered according to the proportion method: 0.2 µg/ml for INH, 40.0 µg/ml for RIF, 2.0 µg/ml for EMB, 4.0 µg/ml for SM, 30.0 µg/ml for KM, 2.0 µg/ml for OFX, 2.0 µg/ml for LFX, 40.0 µg/ml for CM, 40.0 µg/ml for AM, 1.0 µg/ml for PAS and 40.0 µg/ml for PTO[21]. Resistance to a specific drug was determined if the growth rate was greater than $1.0\%$ compared to the control[21]. ## Definitions According to the report issued by WHO, relevant definition of TB defined[22]. Pre-XDR-TB was defined as being resistant to at least INH and RFP, as well as to either of the three injectable second-line drugs or to ofloxacin. XDR-TB was defined as resistance to at least INH and RFP, combined with resistance to a fluoroquinolone and resistance to one of three injectable second-line drugs. Primary DR-TB cases refer to a patient who has never been treated for TB or has taken TB treatment for less than one month. Previously treated DR-TB cases referred to patients who had received one month or more of TB treatment in the past. Mono-resistant TB (MR-TB) is defined as resistance to one first-line anti-TB drug only. Polydrug-resistant TB (PDR-TB) is defined as resistance to more than one first-line anti-TB drug (other than both INH and RFP).Patients with DM were defined as those with fasting plasma glucose ≥ 126 mg/dL (7 mmol/L) or hemoglobin A1C ≥ $6.5\%$ or who self-reported having been diagnosed with diabetes by a physician[23]. ## Statistical analysis The demographics, clinical, and the results of DST for DR-TB patients with and without DM were depicted using descriptive statistics. The categorical characteristics of DR-TB patients were compared by DM status using Chi-square test or Fisher’s exact. In addition, Chi-square tests for trends were used to determine changes and trends in MDR-TB and pre-XDR-TB patterns over time. To identify the link between a collection of independent variables and dependent variables, multivariate logistic regression was performed on all variables with P-values < 0.25 in univariate logistic regression. Two-sided $P \leq 0.05$ was considered statistically significant. All data analysis was carried out by using SPSS 20.0 statistical package (IBM Corporation, Armonk, State of New York). ## Patients’ characteristics A total of 513 DR-TB patients were included in the present study. Of these patients, 327 ($63.74\%$) patients did not have DM; 186 ($36.26\%$) patients had DM. The percentage of the male was $62.08\%$ and $86.02\%$ among DR-TB patients without and with DM, respectively ($P \leq 0.05$). Of all patients, $34.11\%$ were between 45 and 59 years of age and $27.49\%$ were over 60 years of age. In addition, In terms of overall patients, the highest percentage of patients were 51~60 years old. The highest percentage of non-DM patients with DR-TB was in the age group of 31~40 years, but the highest percentage of DM patients with DR-TB was in the age group of 51~60 years (Fig. 2). Among DR-TB patients without and with DM, the percentage of urban residents was $87.77\%$ and $89.25\%$, respectively. Patients with a history of previous tuberculosis treatment were $64.22\%$ and $60.75\%$ among DR-TB patients without and with DM, respectively. The majority of patients had positive sputum smears at baseline ($81.35\%$ vs. $85.48\%$). Unemployment was $88.07\%$ among DR-TB patients without diabetes and $89.78\%$ among DR-TB patients with diabetes. The vast majority of patients are Han Chinese ($98.47\%$ vs. $97.85\%$). The percentage of migrants was $39.76\%$ and $36.02\%$ among DR-TB patients without and with DM, respectively (Table 1). Figure 2Proportion of DR-TB with and without DM patients in different age groupsAbbreviations: DM, diabetes mellitus. Table 1Demographic and clinical characteristics of DR-TB patients without and with DM.CharacteristicsNo DM, N (%)DM, N (%)Total, N (%)χ2P-valueGender32.847<0.001Male203 (62.08)160 (86.02)363 (70.76)Female124 (37.92)26 (13.98)150 (29.24)Age (years)52.078<0.00114~ 3069 (21.10)5 (2.69)74 (14.42)31 ~ 4487 (26.61)36 (19.35)123 (23.98)45 ~ 5982 (25.08)93 (50.00)175 (34.11)60~89 (27.22)52 (27.96)141 (27.49)Patient residence0.2510.616Urban287 (87.77)166 (89.25)453 (88.30)Rural40 (12.23)20 (10.75)60 (11.70)Patient Category0.6110.434Primary DR-TB cases117 (35.78)73 (39.25)190 (37.04)Previously treated DR-TB cases210 (64.22)113 (60.75)323 (62.96)Sputum smear status1.4290.232Negative61 (18.65)27 (14.52)88 (17.15)Active266 (81.35)159 (85.48)425 (82.85)Occupation0.3460.556Unemployed288 (88.07)167 (89.78)455 (88.69)Employed39 (11.93)19 (10.22)58 (11.31)NationalityOthers5 (1.53)4 (2.15)9 (1.75)0.1970.657Han322 (98.47)182 (97.85)504 (98.25)Migrant0.6990.403No197 (60.24)119 (63.98)316 (61.60)Yes130 (39.76)67 (36.02)197 (38.40)Abbreviations: DM, diabetes mellitus; DR-TB, drug-resistant tuberculosis. Notes: Migrant refers to people who leave their domicile for various reasons and go to other Provinces and cities with registered tuberculosis patients. ## Drug resistance profile As is shown in Table 2, among 327 patients without DM and 186 patients with DM, 44,$95\%$ and $47.85\%$ were any INH resistant, respectively; $97.55\%$ and $99.46\%$ were any RFP resistant respectively. The ratio of mono-resistance to SM in cases with and without DM were $6.73\%$ and $2.69\%$, respectively. As is shown in Table 3, $1.83\%$ and $5.38\%$ were any PTO resistant, respectively among 327 patients without DM and 186 patients with DM. $7.65\%$ and $2.69\%$ were any KM resistant, respectively. The ratio of mono-resistance to LFX among cases with and without DM were $2.75\%$ and $6.45\%$, respectively. Among 327 cases without DM and 186 cases with DM, $7.34\%$ and $13.98\%$ were pre-XDR-TB, respectively Table 2First-line drugs resistance profile DR-TB patients without diabetes mellitus and with diabetes mellitusDrug resistance patternNo DM, N (%)DM, N (%)Any resistanceAny INH147 (44.95)89 (47.85)Any RFP319 (97.55)185 (99.46)Any EMB53 (16.21)31 (16.67)Any SM102 (31.19)59 (31.72)MR-TBINH5 (1.53)1 (0.54)RFP159 (48.62)91 (48.92)EMB0 [0]0 [0]SM22 (6.73)5 (2.69)PDR-TBRFP + SM17 (5.20)5 (2.69)INH + SM12 (3.67)6 (3.23)INH + EMB0 [0]1 (0.54)RFP + EMB + SM3 (0.92)0 [0]MDR-TBINH + RFP130 (39.76)82 (44.09)INH + RFP + EMB46 (14.07)29 (15.59)INH + RFP + SM42 (12.84)25 (13.44)INH + RFP + EMB + SM27 (8.26)22 (11.83)Abbreviations: INH, isoniazid; RFP, rifampicin; SM, streptomycin; EMB, ethambutol; MR-TB, mono-resistant tuberculosis; PDR-TB, polydrug-resistant tuberculosis; MDR-TB, multidrug-resistant tuberculosis; DM, diabetes mellitus. Table 3Second-line drugs resistance profile DR-TB patients without and with DM.Drugs resistance patternNo DM, N (%)DM, N (%)Any resistance to second-line drugsAny PTO6 (1.83)10 (5.38)Any PAS19 (5.81)8 (4.30)Any AM23 (7.03)6 (3.23)Any CM13 (3.98)9 (4.84)Any KM25 (7.65)5 (2.69)Any LFX50 (15.29)35 (18.82)Any OFX51 (15.60)25 (13.44)Monoresistant second-line drugsPTO1 (0.31)3 (1.61)PAS4 (1.22)4 (2.15)AM1 (0.31)0 [0]CM1 (0.31)3 (1.61)KM2 (0.61)2 (1.08)LFX9 (2.75)12 (6.45)OFX10 (3.06)7 (3.76)Pre-XDR-TB24 (7.34)26 (13.98)XDR-TB16 (4.89)5 (2.69)Abbreviations: PTO, protionamide; PAS, para-aminosalicylic acid; AM, amikacin; CM, capreomycin; KM, kanamycin; LFX, levofloxacin; OFX, ofloxacin; XDR-TB, extensively drug-resistant TB; DM, diabetes mellitus. ## Annual drug resistance trends for MDR and pre-XDR-TB from 2016 to 2021 Among the 513 patients, the rate of MDR-TB decreased from $61.96\%$ in 2016 to $27.71\%$ in 2021, with an average annual decrease of $14.86\%$ ($P \leq 0.05$). The rate of pre-XDR-TB decreased from $25.00\%$ in 2016 to $18.07\%$ in 2021, with an average annual decrease of $6.28\%$ ($P \leq 0.05$). Among 327 patients without DM, the rate of MDR-TB decreased from $63.08\%$ in 2016 to $22.92\%$ in 2021, with an average annual decrease of $18.33\%$ ($P \leq 0.05$). The rate of pre-XDR-TB decreased from $20.00\%$ in 2016 to $14.58\%$ in 2021, with an average annual decrease of $6.12\%$ ($P \leq 0.05$). Among 186 patients with DM, the rate of MDR-TB decreased from $59.26\%$ in 2016 to $34.29\%$ in 2021, with an average annual decrease of $10.37\%$ ($P \leq 0.05$). The rate of pre-XDR-TB decreased from $37.04\%$ in 2016 to $22.86\%$ in 2021, with an average annual decrease of $9.20\%$ ($P \leq 0.05$)(Fig. 3 and Fig. 4) Figure 3Trends in the proportion of MDR among DR-TB patients without and with DM from 2016 to 2021Abbreviations: DM, diabetes mellitus. Figure 4Trends in the proportion of pre-XDR-TB among DR-TB patients without and with DM from 2016 − 2021Abbreviations: DM, diabetes mellitus. ## Risk of MDR-TB in patients with DR-TB combined without and with DM In the multivariable analysis, among patients without DM, compared with women, men had a 1.71-fold risk of developing MDR-TB (OR = 1.71, $95\%$ CI: 1.04–2.82, $P \leq 0.05$); previously treated DR-TB cases had a 1.70-fold risk of developing MDR-TB compared with primary DR-TB cases (OR = 1.70, $95\%$ CI: 1.04–2.79, $P \leq 0.05$). Immigrants are 2.08 times more likely to develop MDR-TB than native residents (OR = 2.08, $95\%$ CI: 1.28–3.39, $P \leq 0.05$) (Table 4). However, among patients with DM, negative sputum smear alone was an independent risk factor for developing MDR-TB (OR = 2.65, $95\%$ CI: 1.04–6.76, $P \leq 0.05$) (Table 5). Table 4Risk factors of MDR-TB among DR-TB patients without DM.CharacteristicsNone-MDR N(%)MDR N(%)OR($95\%$CI)P-valueAOR($95\%$CI)P-valueGenderFemale66 (35.11)58 (41.73)11Male122 (64.89)81 (58.27)0.76 (0.48–1.19)0.2231.71 (1.04–2.82)0.034Age (years)14~ 3039 (20.74)30 (21.58)1131–4460 (31.91)27 (19.42)0.59 (0.30–1.13)0.1100.56 (0.28–1.11)0.09845–5942 (22.34)40 (28.78)1.24 (0.65–2.36)0.5151.45 (0.73–2.91)0.2906047 (25.00)42 (30.22)1.16 (0.62–2.19)0.6421.42 (0.71–2.83)0.319Patient residenceRural24 (12.77)16 (11.51)1--Urban164 (87.23)123 (88.49)1.13 (0.57–2.21)0.732-Patient CategoryPrimary DR-TB cases76 (40.43)41(29.50)11Previously treated DR-TB cases112 (59.57)98 (70.50)1.62 (1.02–2.59)0.0421.70 (1.04–2.79)0.033OccupationUnemployed162 (86.17)126 (90.65)11Employed26 (13.83)13 (9.35)0.63 (0.31–1.30)0.2200.78 (0.36–1.66)0.512NationalityHan184 (97.87)138 (99.28)1--Others4 (2.13)1 (0.72)3.00 (0.33–27.14)0.328-MigrantNo127 (67.55)70 (50.36)11Yes61 (32.45)69 (49.64)2.05(1.31–3.22)0.0022.08 (1.28–3.39)0.003Sputum smear statusActive161 (85.64)105 (75.54)11Negative27 (14.36)34 (24.46)1.93 (1.0-3.39)0.0221.72 (0.94–3.17)0.079Abbreviations: CI, Confidence Interval; OR, Odds Ratio; AOR, Adjust Odds Ratio, MDR-TB, multi-drug resistant tuberculosis. Notes: Migrant refers to people who leave their domicile for various reasons and go to other Provinces and cities with registered tuberculosis patients. Table 5Risk factors of MDR-TB among DR-TB patients with DM.CharacteristicsNone-MDR, N (%)MDR, N (%)OR ($95\%$CI)P-valueAOR ($95\%$CI)P-valueGenderFemale14 (14.29)12 (13.64)1--Male84 (85.71)76 (86.36)1.06 (0.46–2.42)0.899-Age (years)14~ 302 (2.04)3 (3.41)1--31–4418 (18.37)18 (20.45)0.67 (0.10–4.48)0.677-45–5954 (55.10)39 (44.32)0.48 (0.08–3.02)0.435-6024 (24.49)28 (31.82)0.78 (0.12–5.05)0.792-Patient residenceRural11 (11.22)9 (10.23)1--Urban87 (88.78)79 (89.77)1.11 (0.44–2.82)0.827-Patient CategoryPrimary DR-TB cases45 (45.92)28 (31.82)11Previously treated DR-TB cases53 (54.08)60 (68.18)1.82 (1.00-3.31)0.0501.85(1.00-3.42)0.051OccupationUnemployed86 (87.76)81 (92.05)1--Employed12 (12.24)7 (7.95)0.62 (0.23–1.65)0.338-NationalityHan97 (98.98)85 (96.59)1--Others1 (1.03)3 (3.41)3.00 (0.33–27.14)0.328-MigrantNo70 (71.43)49 (55.68)11Yes28 (28.57)39 (44.32)2.05 (1.31–3.22)0.0021.58 (0.83–3.02)0.168Sputum smear statusActive90 (8.16)69 (21.59)11Negative8 (91.84)19 (78.41)3.10 (1.28–7.50)0.0122.65 (1.04–6.76)0.042Abbreviations: CI, Confidence Interval; OR, Odds Ratio; AOR, Adjust Odds Ratio, MDR-TB, multi-drug resistant tuberculosis. Notes: Migrant refers to people who leave their domicile for various reasons and go to other provinces and cities with registered tuberculosis patients. ## Discussion This is the first study on the difference in drug resistance profile and trends of DR-TB patients with and without DM in northeast China. In our study, we found that patients with DM were older and were more likely to be men than patients without DM. These findings were similar to studies conducted in Georgia and China[24, 25]. Some reasons can be explained. On the one hand, male and older age are independent risk factors for DM, as many studies have confirmed[26–28]. In addition, men tend to be more likely to develop DR-TB than women[29]. However, some studies have also reported that women are at high risk for DR-TB[30, 31]. This difference may be due to socioeconomic and cultural differences and regional differences. On the other hand, the population of northeast China is seriously aging, and in 2020, the proportion of elderly people over 60 years old would exceed $20\%$. In this study, patients aged 60 and over accounted for $27.49\%$, which may be the reason for the larger proportion of elderly patients in patients without and with DM. Except for mono-resistant SM, the drug resistance profile of other first-line anti-TB drugs in patients with and without diabetes was not statistically different. This result is similar to that of a previous study[18], but studies in other countries have come to the opposite conclusion[8, 19]. We speculate that may be due to different study designs and smaller samples. The difference in mono-resistant SM between the DR-TB with and without DM implies that our health care providers are aware of this difference when dealing with patients with and without DM when administering clinical drug therapy. In total, the proportion of patients with DM was higher for arbitrary PTO resistance, and for mono-resistant of LFX than for patients without diabetes. However, among patients with without DM, arbitrary KM resistance, and mono-resistant of SM were higher than that of patients with DM. In addition to this, we also found that the proportion of pre-XDR-TB was significantly higher in patients with DM compared to those without DM. In other studies, some opposite findings were found[18]. Molecular epidemiology should be further developed to examine the reasons for this discrepancy. Our study found that MDR-TB and pre-XDR-TB with and without DM showed a downward trend from 2016 to 2021 ($P \leq 0.05$). We speculate that this may be since two-way screening for TB and DM was carried out according to the requirements of the National Health Commission of China, allowing more patients to receive active treatment and preventing the spread of DR-TB. However, it is worth noting the increasing trend of MDR and pre-XDR-TB resistance rates from 2020 to 2021, which may be due to the impact of the COVID-19 pandemic. Among patients without DM, male, a history of TB treatment and migration are independent risk factors for the development of MDR-TB. Males are more likely to smoke and abuse alcohol than women, both of which are risk factors for developing MDR-TB[32]. In addition, men are more likely than women to be infected with drug-resistant Mycobacterium tuberculosis because their work involves more social interaction[33]. Patients with a history of TB treatment may have their sensitive Mycobacterium tuberculosis eliminated and drug-resistant Mycobacterium tuberculosis left behind due to multiple treatments[34]. Immigrants are not often able to regularize their treatment due to their unstable work and place of residence, which eventually leads to the development of drug resistance[35]. Among patients with DM, sputum-smear-negative patients were more likely to develop MDR-TB than sputum-smear-positive patients. This may be since patients with negative sputum smears receive less attention and guidance from medical staff than patients with positive sputum smears, leading to irregular medication and more likely to develop MDR-TB. There are some limitations. First, relatively few patients were included in our study, especially those with DR-TB and DM. Second, because our data came from the Chinese Disease Control Information System, some important information such as smoking, drinking, and adverse reactions were not found in this system. Third, some important anti-tuberculosis drugs, such as moxifloxacin and pyrazinamide, were not included due to many deletions. Fourth, the exclusion of HIV-positive patients may not explore the underlying status of drug resistance among HIV-positive drug-resistant patients. Nonetheless, our study identified differences in drug resistance patterns and trends among DM and non-DM patients with DR-TB, as well as risk factors for developing MDR-TB. These findings provide important information for healthcare workers to combat DR-TB. ## Conclusion Except for arbitrary resistance to PTO and KM, mono-resistance to SM and LFX, and pre-XDR-TB, there was no statistical difference in resistance patterns between patients with and without DM. Among patients with and without DM, the highest proportion of drug resistance was RFP. Great progress has been made in the prevention and treatment of DR-TB in patients with and without DM. Male gender, history of TB treatment, and migrant are risk factors for MDR-TB in non-DM patients with DR-TB. However, in patients with DR-TB and DM, negative sputum smear is a geographic risk factor for developing MDR-TB. The development of targeted measures among DR-TB patients with and without DM would be beneficial for the control of DR-TB. ## References 1. 1.WHO. Global tuberculosis report 2022. In. World Health Organization; 2022. 2. Uplekar M, Weil D, Lonnroth K, Jaramillo E, Lienhardt C, Dias HM, Falzon D, Floyd K, Gargioni G, Getahun H. WHO’s new end TB strategy. Lancet (2015.0) 385 1799-801. DOI: 10.1016/S0140-6736(15)60570-0 3. Pan Y, Yu Y, Lu J, Yi Y, Dou X, Zhou L. Drug resistance patterns and Trends in patients with suspected drug-resistant tuberculosis in Dalian, China: a retrospective study. Infect Drug Resist (2022.0) 15 4137-47. DOI: 10.2147/IDR.S373125 4. Furin J, Cox H, Pai M. Tuberculosis. Lancet (2019.0) 393 1642-56. DOI: 10.1016/S0140-6736(19)30308-3 5. 5.WHO. Global tuberculosis report 2020. In. World Health Organization; 2020. 6. Loennroth K, Castro KG, Chakaya JM, Chauhan LS, Floyd K, Glaziou P, Raviglione MC. Tuberculosis control and elimination 2010-50: cure, care, and social development. Lancet (2010.0) 375 1814-29. DOI: 10.1016/S0140-6736(10)60483-7 7. Sun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB, Stein C, Basit A, Chan JC, Mbanya JC. IDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. Diabetes Res Clin Pract (2022.0) 183 109119. DOI: 10.1016/j.diabres.2021.109119 8. Song W-m, Shao Y, Liu J-y, Tao N-n, Liu Y, Zhang Q-y, Xu T-t, Li S-j, Yu C-B, Gao L. Primary drug resistance among tuberculosis patients with diabetes mellitus: a retrospective study among 7223 cases in China. Infect Drug Resist (2019.0) 12 2397-407. DOI: 10.2147/IDR.S217044 9. Zhang J, Gou H, Hu X, Hu X, Shang M, Zhou J, Zhou Y, Ye Y, Song X, Lu X. Status of drug-resistant tuberculosis in China: a systematic review and meta-analysis. Am J Infect Control (2016.0) 44 671-6. DOI: 10.1016/j.ajic.2015.12.042 10. Alene KA, Xu Z, Bai L, Yi H, Tan Y, Gray DJ, Viney K, Clements ACA. Spatiotemporal patterns of tuberculosis in Hunan Province, China. Int J Environ Res Public Health (2021.0) 18 6778. DOI: 10.3390/ijerph18136778 11. Yang L, Shao J, Bian Y, Wu H, Shi L, Zeng L, Li W, Dong J. Prevalence of type2 diabetes mellitus among inland residents in China (2000–2014): a meta-analysis. J Diabetes Invest (2016.0) 7 845-52. DOI: 10.1111/jdi.12514 12. Lutfiana NC, van Boven JFM, Zubair MAM, Pena MJ, Alffenaar J-WC. Diabetes mellitus comorbidity in patients enrolled in tuberculosis drug efficacy trials around the world: a systematic review. Br J Clin Pharmacol (2019.0) 85 1407-17. DOI: 10.1111/bcp.13935 13. Scordo JM, Aguillón-Durán GP, Ayala D, Quirino-Cerrillo AP, Rodríguez-Reyna E, Mora-Guzmán F, Caso JA, Ledezma-Campos E, Schlesinger LS, Torrelles JB. A prospective cross-sectional study of tuberculosis in elderly Hispanics reveals that BCG vaccination at birth is protective whereas diabetes is not a risk factor. PLoS ONE (2021.0) 16 e0255194. DOI: 10.1371/journal.pone.0255194 14. Hsu AH, Lee JJ, Chiang CY, Li YH, Chen LK, Lin CB. Diabetes is associated with drug-resistant tuberculosis in Eastern Taiwan. Int J Tuberc Lung Dis (2013.0) 17 354-6. DOI: 10.5588/ijtld.11.0670 15. Wang C, Yang C, Chen H, Chuang S, Chong I, Hwang J, Huang M. Impact of type 2 diabetes on manifestations and treatment outcome of pulmonary tuberculosis. Epidemiol Infect (2009.0) 137 203-10. DOI: 10.1017/S0950268808000782 16. Liu Q, Li W, Xue M, Chen Y, Du X, Wang C, Han L, Tang Y, Feng Y, Tao C. Diabetes mellitus and the risk of multidrug resistant tuberculosis: a meta-analysis. Sci Rep (2017.0) 7 1-7. PMID: 28127051 17. Huangfu P, Ugarte-Gil C, Golub J, Pearson F, Critchley J. The effects of diabetes on tuberculosis treatment outcomes: an updated systematic review and meta-analysis. Int J Tuberc Lung Dis (2019.0) 23 783-96. DOI: 10.5588/ijtld.18.0433 18. 18.Wu Q, Wang M, Zhang Y, Wang W, Ye T-F, Liu K, Chen S-H. Epidemiological Characteristics and Their Influencing Factors Among Pulmonary Tuberculosis Patients With and Without Diabetes Mellitus: A Survey Study From Drug Resistance Surveillance in East China.Frontiers in public health2022,9. 19. 19.Song W-m, Li Y-f, Liu J-y, Tao N-n, Liu Y, Zhang Q-y, Xu T-t, Li S-j et al. An Q-q, Liu S-q : Drug resistance of previously treated tuberculosis patients with diabetes mellitus in Shandong, China. Respiratory Medicine 2020, 163. 20. Rarome BB, Aisah N, Setyoningrum RA, Mertaniasih NM. GeneXpert MTB/RIF and Mycobacterium tuberculosis Sputum Culture in establishing the diagnosis of Pulmonary Tuberculosis and Rifampicin Resistance in Suspected Childhood Pulmonary Tuberculosis in Soetomo Hospital. Indonesian J Trop Infect Disease (2020.0) 8 152-60. DOI: 10.20473/ijtid.v8i3.15503 21. 21.Shi J, He G, Ning H, Wu L, Wu Z, Ye X, Qiu C, Jiang X. Application of matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS) in the detection of drug resistance of Mycobacterium tuberculosis in re-treated patients.Tuberculosis2022,135. 22. Definitions W. Reporting framework for tuberculosis–2013 revision (2013.0) 23. Classification and diagnosis of diabetes: Standards of Medical Care in Diabetes-2021. Diabetes Care (2021.0) 44 15-s33 24. Kikvidze M, Mikiashvili L. Impact of diabetes mellitus on drug-resistant tuberculosis treatment outcomes in Georgia - Cohort study. Eur Respir J (2013.0) 42 P2826 25. Shi L, Gao J, Gao M, Deng P, Chen S, He M, Feng W, Yang X, Huang Y, He F. Interim effectiveness and safety comparison of Bedaquiline-Containing regimens for treatment of Diabetic Versus non-diabetic MDR/XDR-TB patients in China: a Multicenter Retrospective Cohort Study. Infect Dis Therapy (2021.0) 10 457-70. DOI: 10.1007/s40121-021-00396-9 26. Kautzky-Willer A, Harreiter J, Pacini G. Sex and gender differences in risk, pathophysiology and complications of type 2 diabetes Mellitus. Endocr Rev (2016.0) 37 278-316. DOI: 10.1210/er.2015-1137 27. Martinez MC, Latorre Mdo R. Risk factors for hypertension and diabetes mellitus in metallurgic and siderurgic company’s workers. Arq Bras Cardiol (2006.0) 87 471-9. DOI: 10.1590/S0066-782X2006001700012 28. Chaudhary F, Chaudhary S. Awareness about diabetes risk factors & complications in diabetic patients: a cross sectional study. Nishtar Med J (2010.0) 2 84-8 29. Rajendran M, Zaki RA, Aghamohammadi N. Contributing risk factors towards the prevalence of multidrug-resistant tuberculosis in Malaysia: a systematic review. Tuberculosis (Edinb) (2020.0) 122 101925. DOI: 10.1016/j.tube.2020.101925 30. Mphande-Nyasulu FA, Puengpipattrakul P, Praipruksaphan M, Keeree A, Ruanngean K. Prevalence of tuberculosis (TB), including multi-drug-resistant and extensively-drug-resistant TB, and association with occupation in adults at Sirindhorn Hospital, Bangkok. IJID Reg (2022.0) 2 141-8. DOI: 10.1016/j.ijregi.2022.01.004 31. Baluku JB, Mukasa D, Bongomin F, Stadelmann A, Nuwagira E, Haller S, Ntabadde K, Turyahabwe S. Gender differences among patients with drug resistant tuberculosis and HIV co-infection in Uganda: a countrywide retrospective cohort study. BMC Infect Dis (2021.0) 21 1-11. DOI: 10.1186/s12879-021-06801-5 32. Mor Z, Goldblatt D, Kaidar-Shwartz H, Cedar N, Rorman E, Chemtob D. Drug-resistant tuberculosis in Israel: risk factors and treatment outcomes. Int J Tuberculosis Lung Disease (2014.0) 18 1195-201. DOI: 10.5588/ijtld.14.0192 33. Pradipta IS, Van’t Boveneind-Vrubleuskaya N, Akkerman OW, Alffenaar JC, Hak E. Treatment outcomes of drug-resistant tuberculosis in the Netherlands, 2005–2015. Antimicrob Resist Infect Control (2019.0) 8 115. DOI: 10.1186/s13756-019-0561-z 34. McBryde ES, Meehan MT, Doan TN, Ragonnet R, Marais BJ, Guernier V, Trauer JM. The risk of global epidemic replacement with drug-resistant Mycobacterium tuberculosis strains. Int J Infect Dis (2017.0) 56 14-20. DOI: 10.1016/j.ijid.2017.01.031 35. Zammarchi L, Bartalesi F, Bartoloni A. Tuberculosis in tropical areas and immigrants. Mediterranean J Hematol Infect Dis (2014.0) 6 e2014043-3. DOI: 10.4084/mjhid.2014.043
--- title: Characterization and utility of two monoclonal antibodies to cholera toxin B subunit authors: - Noel Verjan Garcia - Ian Carlosalberto Santisteban Celis - Matthew Dent - Nobuyuki Matoba journal: Scientific Reports year: 2023 pmcid: PMC10016189 doi: 10.1038/s41598-023-30834-2 license: CC BY 4.0 --- # Characterization and utility of two monoclonal antibodies to cholera toxin B subunit ## Abstract Cholera toxin B subunit (CTB) is a potent immunomodulator exploitable in mucosal vaccine and immunotherapeutic development. To aid in the characterization of pleiotropic biological functions of CTB and its variants, we generated a panel of anti-CTB monoclonal antibodies (mAbs). By ELISA and surface plasmon resonance, two mAbs, 7A12B3 and 9F9C7, were analyzed for their binding affinities to cholera holotoxin (CTX), CTB, and EPICERTIN: a recombinant CTB variant possessing mucosal healing activity. Both 7A12B3 and 9F9C7 bound efficiently to CTX, CTB, and EPICERTIN with equilibrium dissociation constants at low to sub-nanomolar concentrations but bound weakly, if at all, to Escherichia coli heat-labile enterotoxin B subunit. In a cyclic adenosine monophosphate assay using Caco2 human colon epithelial cells, the 7A12B3 mAb was found to be a potent inhibitor of CTX, whereas 9F9C7 had relatively weak inhibitory activity. Meanwhile, the 9F9C7 mAb effectively detected CTB and EPICERTIN bound to the surface of Caco2 cells and mouse spleen leukocytes by flow cytometry. Using 9F9C7 in immunohistochemistry, we confirmed the preferential localization of EPICERTIN in colon crypts following oral administration of the protein in mice. Collectively, these mAbs provide valuable tools to investigate the biological functions and preclinical development of CTB variants. ## Introduction Cholera toxin B subunit (CTB, approx. 58 kDa), is a nontoxic component of the cholera toxin (CTX) released by Vibrio cholerae, a Gram-negative bacterium causing profuse secretory diarrhea1. CTB is a potent immunomodulator consisting of five identical polypeptide chains non-covalently associated in a ring-shaped pentameric structure that mediates CTX binding to the monosialoganglioside GM1 receptor2–4. CTB has been produced in various recombinant expression platforms as model vaccines5–8, as it induces a strong immune response characterized by neutralizing antibodies to the CTX via oral administration9,10. CTB has also been used as a molecular scaffold for immunization11,12 and for induction of peripheral (oral) immunological tolerance to suppress allergies and various autoimmune disorders such as experimental autoimmune encephalitis (EAE), diabetes, arthritis and uveitis in an antigen-specific manner13–17. CTB is considered a mucosal immunomodulator that induces a strong mucosal IgA response while it seems to suppress systemic T helper (TH1, TH2 and TH17) responses through the induction of interleukin-10 (IL-10) and transforming growth factor-β (TGF-β)-mediated regulatory T (Treg) cell induction and suppression of proinflammatory IL-618, although the exact mechanism behind such immunomodulation remains elusive. The pleiotropic functions of CTB have recently been expanded through the construction of CTB variants with novel biological activities, including a plant-made CTB variant modified with C-terminal hexapeptide extension containing a KDEL endoplasmic reticulum (ER) retention motif19, later designated as EPICERTIN20, which showed a significant enhancement of mucosal wound healing activity in the colon21–25. In mice, oral administration of EPICERTIN increased specific subsets of immune cell populations, such as macrophages, natural killer cells, Treg cells and TH17 cells in the colon lamina propria and upregulated wound healing pathway genes mediated by TGF-β in the colon. Mucosal healing effects of EPICERTIN were confirmed in an in vivo dextran sodium sulfate (DSS) acute colitis model, an azoxymethane/DSS model of ulcerative colitis and tumorigenicity21, and a model of chronic colitis-like colon inflammation induced by repeated doses of DSS, where the wound healing effects were not neutralized by the induction of mucosal anti-CTB antibodies24. The prolonged residence of EPICERTIN in the ER, through interaction with the KDEL receptor, appears to trigger an unfolded protein response, which leads to the activation of TGF-β signaling and transcription of wound healing pathway genes in colon epithelial cells23. However, the mucosal healing effects could also originate from modulation of mucosal immune cells. Indeed, early studies reported multiples roles of CTX and CTB in mouse and human leukocytes. For example, CTB inhibited T-cell proliferation induced by mitogens (e.g. Concanavalin A) and antigens26, and antigen-conjugated CTB promoted its presentation and lowered the threshold of antigen required for T cell activation27. Therefore, to facilitate the investigation of the mechanisms behind the biological activities of EPICERTIN and other CTB-derived molecules, we aimed at characterizing two monoclonal antibodies (mAbs) that were generated against CTB to aid in the study of those proteins. Here, we report on the isolation and characterization of two mAbs, 7A12B3 and 9F9C7, that were generated against CTB, demonstrating their unique and distinct binding profiles by means of ELISA, surface plasmon resonance (SPR)28 and cyclic adenosine monophosphate (cAMP) assays. In addition, we show the utility of these mAbs in neutralizing CTX and detecting CTB bound both to murine leukocytes by flow cytometry and to colon tissues by immunohistochemistry (IHC). Together with our previous studies that utilized these mAbs in immunofluorescence, immunoprecipitation and pharmacokinetic analyses of EPICERTIN20,23, the present study underscores that 7A12B3 and 9F9C7 will aid in investigating as-yet-uncharacterized biological mechanisms of action of CTB and its derivatives as candidate mucosal vaccines and immunotherapeutics. ## Animal care Immunization of Wistar rats for hybridoma generation was conducted by GenScript USA, Inc. (Piscataway, NJ), whereas flow cytometry and IHC experiments using C57BL/6J mice were performed at the University of Louisville. Studies were approved by each institution’s Institutional Animal Care and Use Committee and are reported in compliance with the ARRIVE guidelines. General procedures for animal care and housing in these studies were in accordance with the current Association for Assessment and Accreditation of Laboratory Animal Care recommendations, current requirements stated in the Guide for the Care and Use of Laboratory Animals (National Research Council), and current requirements as stated by the U.S. Department of Agriculture through the Animal Welfare Act and Animal Welfare regulations (July 2020). ## Reagents Cholera enterotoxin (C8052), cholera toxin B-subunit (CTB; C9903) and heat labile enterotoxin B subunit (LTB; E9656) of Escherichia coli were purchased from Sigma (Sigma Aldrich). PhenoVue Fluor 594-WGA was obtained from PerkinElmer (Perkin Elmer Health Science Inc. Boston MA, USA), and Alexa Fluor 647-conjugated rat anti mouse/human CD324 (E-cadherin) (clone DECMA-1) was obtained from Biolegend. ## Generation of anti-CTB mAbs To facilitate the generation of CTX-neutralizing mAbs, an Asn4-glycosylated CTB variant produced in Nicotiana benthamiana19,29 (gCTB) was used as an antigen for immunization. The plant-expressed gCTB was purified by metal-affinity chromatography followed by CHT hydroxyapatite chromatography, as described previously19. The purified protein was used to immunize Wistar rats to generate a panel of monoclonal antibodies by Genscript USA, Inc. using standard hybridoma technology30,31. Briefly, three Wistar rats were immunized with 50 µg of keyhole limpet hemocyanin conjugated gCTB (gCTB-KLH) and subsequently boosted with 25 µg gCTB-KLH every two weeks, for a total of three booster doses. Two weeks after the last booster immunization, the lymphocyte population from spleens were isolated and used to generate immortalized hybridomas by fusion with the Sp$\frac{2}{0}$-Ag14 myeloma cells following standard methods. Hybridomas were grown in Hypoxanthine-aminopterin-thymidine medium supplemented with IL-6, and the culture supernatants were screened for Abs reactive with CTB by antigen-capture enzyme-linked immunosorbent assay (ELISA) and GM1-capture ELISA. Two anti-CTB mAbs-producing hybridomas named 7A12B3 and 9F9C7 were selected and expanded, and mAbs were produced in hybridoma cell culture supernatants using the invitro roller bottle cell culture method followed by protein G-affinity chromatography purification. They were both determined to be of the IgG2a subclass by using the Pro-Detect™ Rapid Antibody Isotyping Kit- Rat (Thermo Fisher Scientific). Three Wistar rats were immunized with a KLH-conjugated gCTB. All rats consistently showed high anti-CTB serum antibody titers after three doses of the antigen, as analyzed by both CTB antigen-capture ELISA and GM1-capture ELISA. Given that rat #1 showed an optimal response ratio for direct vs. GM1-bound CTB at a higher dilution factor (1:243,000), indicative of a significant proportion of antibodies targeting the GM1-binding facet of CTB, this rat was selected for hybridoma generation. A total of 40 positive hybridoma clones were obtained. Hybridoma supernatants were analyzed for the presence of anti-CTB antibodies by GM1-capture and CTB antigen-capture ELISAs (Fig. 1A). At this stage, the 7A12, 8F8 and 9F9 hybridomas were initially selected based on the high binding affinity to immobilized CTB in antigen-capture ELISA and in GM1-capture ELISA. Of note, among the three hybridomas, 7A12 and 9F9 mAbs showed very distinctive CTB-binding patterns in ELISA (Fig. 1B), where the binding signal of 7A12 mAb was almost completely abolished in GM1-capture ELISA whereas 9F9 mAb showed similar binding responses regardless of the ELISA formats. Thus, we proceeded with subculturing of these two hybridomas under limiting dilution conditions to isolate single clones, and the resultant 7A12B3 and 9F9C7 hybridomas were selected for subsequent studies. The Pro-Detect™ Rapid Antibody Isotyping Kit- Rat revealed that both 7A12B3 and 9F9C7 mAbs are of the IgG2a isotype with kappa light chains. Figure 1Generation and screening of hybridoma supernatants reacting to CTB. ( A) Three Wistar rats were immunized with gCTB to generate a panel of mAbs and blood serum from each rat was diluted and analyzed for the level of anti-CTB IgG antibodies by antigen-capture and GM1-capture ELISA. ( B) Characterization of three hybridomas, 7A12, 8F8 and 9F9, which showed high anti-CTB antibody levels by ELISA. The hybridoma cell culture supernatants showed distinctive binding patterns to CTB in antigen-capture vs. GM1-capture ELISA. All three hybridoma showed similarly high binding in CTB-capture ELISA. By contrast, 7A12 showed the least and negligible binding while 9F9 showed the highest binding in GM1-capture ELISA. Bars represent means ± range of technical duplicate values. ## Enzyme-linked immunosorbent assays Binding affinities of 7A12B3 and 9F9C7 mAbs to CTX were determined by antigen (CTX, CTB, LTB)-capture ELISA, GM1-capture ELISA, and competitive GM1-capture KDEL-detection (GM1/KDEL) ELISA as described previously22 with minor modifications as follows: ## Antigen-capture ELISA Ninety-six-well ELISA plates (Nunc, MaxiSorp) were coated with 100 µL per well of 2 µg/mL CTB, CTX or LTB in phosphate-buffered saline (PBS) and incubated at 4 °C overnight. The plates were washed 3 times with PBST (PBS + $0.05\%$ Tween 20) and blocked with 150 μL/well of blocking solution ($5\%$ non-fat dry milk in PBST) for 1 h at room temperature. A total of 100 μL/well of serially diluted hybridoma supernatants starting at 1:100 dilution, or purified 7A12B3 and 9F9C7 mAbs diluted in $1\%$ PBST (PBST containing $1\%$ non-fat dry milk) were added after washing 3 times with PBST and incubated at room temperature for 2 h. After 3 washes with PBST, a horseradish peroxidase (HRP)-conjugated goat anti-rat IgG-H&L antibody (SouthernBiotech, Birmingham AL, USA) diluted at 1:5000 was added and incubated at room temperature for 1 h. The plates were washed 3 times and the enzyme substrate tetramethylbenzidine (TMB, 100 µL/well) was added and incubated for 2–3 min at room temperature before stopping the reaction with 100 µl/well of 0.6 N H2SO4 stop solution. Absorbance at 450 nm was measured using a BioTek Synergy HT microplate reader (Winooski, VT, USA). ## GM1-capture ELISA ELISA plates were coated with 100 μL per well of 2 μg/mL monoganglioside GM1 (SIGMA-Aldrich) diluted in PBS. After 2-h incubation at room temperature, the plates were washed three times with PBST and blocked with 150 μL/well of blocking solution for overnight (12–16 h) at 4 °C. The plates were washed 3 × with PBST and a fixed concentration (0.3 µg/mL) of CTB in $1\%$ PBSTM was applied to the plates and incubated at room temperature for 2 h. After washing with PBST, dilutions of hybridomas supernatants or purified mAbs in $1\%$ PBSTM were added and the bound antibodies were detected as described above. ## GM1/KDEL ELISA ELISA plates were coated with GM1 and blocked as described above. After washing the plates 3 times with PBST, 50 µL/well of mAbs at 0, 0.1, 0.3 or 1 µg/mL and 50 µL/well of 2-fold serially diluted EPICERTIN (starting from 2 µg/mL), both prepared in $1\%$ PBSTM, were simultaneously added to plates and incubated at room temperature for 2 h. After washing with PBST, 100 µL/well of mouse anti-KDEL mAb (Enzo LifeSciences; Farmingdale, NY, USA) diluted 1:1000 in $1\%$ PBSTM was added, and plates were incubated at room temperature for 1 h. Plates were washed and goat anti-mouse IgG-HRP (SouthernBiotech) diluted 1:5000 in $1\%$ PBSTM was added, followed by incubation at room temperature for 1 h. After washing 3 times with PBST, the HRP enzyme activity was measured as described above. ## Competitive GM1/KDEL ELISA ELISA plates were coated with GM1, blocked, and washed as described above. Separately, equal volumes of 2-fold serially diluted mAbs (starting from 8 µg/mL) and EPICERTIN at a fixed concentration (0.2 µg/mL), both prepared in $1\%$ PBSTM, were mixed and incubated at room temperature for 30 min in a non-binding round-bottom plate. Then, mAb-EPICERTIN mixtures were added to the GM1-coated plates (100 µL/well), followed by incubation at room temperature for 2 h. The plate-bound EPICERTIN was detected as described above. ## Surface plasmon resonance Binding affinities of 7A12B3 and 9F9C7 mAbs to CTX, commercial CTB, and EPICERTIN were also determined by surface plasmon resonance (SPR) using the Biacore Gold Seal T200 (GE Healthcare) equipped with a CM5 sensor chip as previously described19. The ligands 7A12B3 (150 kDa) and 9F9C7 (150 kDa) mAbs were immobilized on the carboxylated dextran matrix of a CM5 chip sensor surface using amine-coupling chemistry. The surfaces of flow cells were activated with a 1:1 mixture of 0.1 M NHS (N-hydroxysuccinimide) and 0.4 M EDC (3-(N,N-dimethylamino) propyl-N-ethylcarbodiimide) at a flow rate of 5 μl/min for 14 min. The ligands at a concentration of 5 μg/ml in 10 mM sodium acetate, pH 4.0, were immobilized at a density of 205–220 RU (7A12B3) and 709 (9F9C7). Flow cells 1 and 3 were left as a reference blank, while flow cells 2 and 4 were used for 7A12B3 and 9F9C7 ligands. Both surfaces were blocked with 1 M ethanolamine, pH 8.0, with a 7 min injection time. Running buffer was 10 mM HEPES, 150 mM NaCl, $0.005\%$ P20, pH 7.4. To collect steady state data and kinetic binding of analytes CTX, CTB, and EPICERTIN to immobilized 7A12B3 mAb, the analytes were diluted to 26.9 nM in running buffer. For 9F9C7 the analytes CTX, CTB, and EPICERTIN were diluted to 1670 nM in running buffer. Samples were 2-fold serially diluted and injected at a flow rate of 10 μL/min and 30 μL/min at 25 °C, respectively. The complexes with 7A12B3 mAb were allowed to associate for 120 s and dissociate for 600 s, whereas the complexes with 9F9C7 mAb were allowed to associate for 180 s and dissociate for 900 s. The surfaces were regenerated with 10 mM Glycine pH 2.0 for 30 s and 60 s for 7A12B3 and 9F9C7 mAbs, respectively. Triplicate injections (in random order) of each sample and a buffer blank were flowed over the two surfaces. Data were collected at a rate of 10 Hz. The data were fit to a Bivalent model using the global data analysis option available within Biacore Evaluation software. ## Intracellular cAMP in Caco2 cell line Caco2 cells were grown in 12-well cluster plates (Thermo Scientific Nunc Cell-Culture Treated, Roskilde, Denmark) at a density of 7 × 105 Caco2 cells/well containing EMEM (Gibco BRL) medium supplemented with $20\%$ FBS, 5 mM HEPES, 5 mM NEAA, 5 mM Sodium Pyruvate and Penicillin/Streptomycin for 24 h. The culture medium was removed, and the cells were washed twice with PBS and then incubated in serum-free EMEM medium containing 2.5 mM HEPES, $0.01\%$ bovine serum albumin, 1 mM 3-Isobutyl-1-methylxanthine (MP Biomedicals, LLC, Solon, Ohio), and the indicated concentrations of 7A12B3 and 9F9C7 mAbs, and 0.5 µg/mL (5.68 nM) of CTX (Sigma, St. Louis, Mo) or rat IgG isotype control for 30 min at 4 °C (on ice). The plates were subsequently transferred to 37 °C in a $5\%$ CO2 incubator for 2 h. The culture plates were centrifuged at 1000×g, the culture medium removed, and the cells washed twice with PBS before incubation in 200 µL of 0.1 N HCl for 20 min at room temperature to allow cell lysis and cAMP extraction32. cAMP was detected using a sensitive colorimetric ELISA-based kit (Enzo Life Sciences, Farmingdale, NY) following the manufacturer’s conditions. A standard curve was constructed with known concentrations of cAMP and the cAMP levels in the samples (pmol/7 × 105 Caco2 cells) were determined according to the equation generated by the standard curve. ## Cell isolation and flow cytometry C57BL/6J female mice were obtained from Jackson Laboratories (Bar Harbor, ME) and used between 8 and 12-week-old. The spleens of two naïve mice were collected and minced, and the cell suspension passed through a 40 µm cell strainer after treatment with ACK buffer to lyse red cells. The cells were counted, the Fc receptors were blocked with mouse γ-globulins (20 µg/mL), and the cells were subsequently incubated with EPICERTIN or an EPICERTIN variant with Gly33 → Asp mutation (EPICERTING33D; 23 5 µg/mL) for 30 min on ice. After two washes with FACS buffer, unlabeled 9F9C7 mAb was added at 5 µg/mL and the cell suspension incubated on ice for 30 min. Later a FITC-conjugated goat anti-Rat IgG (Poly4054) antibody was added followed by washing and incubation with fluorochrome-labeled antibodies to cell specific markers, including PE-conjugated rat anti-CD4 (RM4-5), PE-conjugated rat anti-IA-IE (M$\frac{5}{114.15.6}$), PE-conjugated rat anti-Ly6C (Hk1.4), PE-conjugated rat anti-Gr-1 (RB6-8C5), PE or APC-conjugated Armenian Hamster anti-CD11c (N418), APC-conjugated rat anti-CD8a (53-6.7), APC-conjugated rat anti-CD11b (M$\frac{1}{70}$), APC-conjugated rat anti-Ly6G (1A8), APC-conjugated rat anti-F$\frac{4}{80}$ (BM8) and PE or APC-conjugated Rat IgG2a κ isotype control (RTK2758) antibodies, all from Biolegend. APC-conjugated rat anti-CD19 (1D3) was from eBioscience. Flow cytometric analysis was performed on a FACSCalibur or a BD SLRFortessa (BD Biosciences) and the data were processed with FlowJo_v10.8.0_CL software. The geometric mean fluorescence intensity values generated by the goat-anti Rat IgG secondary antibody were subtracted from that of 9F9C7 anti-CTB specific mAb. The procedures with mice were approved by the Institutional Animal Care and Use Committee of University of Louisville. ## Immunohistochemistry EPICERTIN in PBS (3 μg/100µL) was administered by oral gavage to five female C57BL/6 J mice after neutralization of the gastric acid with 200 μL of sodium bicarbonate (30 mg/mL). The mice were sacrificed at 0, 3, 6, 12 or 24 h, and the colon tissues were washed with PBS and embedded in OCT compound to make 7 µm thick frozen sections. Cryosections of the colon tissue were fixed in $100\%$ Methanol at − 20 °C for 3 min, dried and blocked with $10\%$ FBS in PBS containing 20 μg/mL mouse γ-globulins for 1 h at RT. The tissue sections were stained with 5 μg/mL Pheno Vue Fluor 594-WGA (PerkinElmer Health Sciences), 2 μg/mL anti-E-cadherin, and 2 μg/mL anti-CTB (9F9C7 mAb) for 1 h at RT. The slides were washed in PBS, images were collected with a Nikon A1R Confocal laser scanning microscope using 20 × and 60 × magnification lenses with appropriate channels, and the data were processed with the NIS Elements imaging software. ## Data analysis The half-maximal effective concentration (EC50) values were determined by non-linear regression analysis using Prism v.9.1.0. One-way ANOVA with Bonferroni’s multiple comparison posttest was used to analyze absorbance values of cAMP levels, using Prism v.9.1.0 (GraphPad Software, La Jolla, CA, USA). A value of $p \leq 0.05$ was considered significant. ## The mAbs 7A12B3 and 9F9C7 bind CTX, CTB, and EPICERTIN with high affinities To determine the antigen-binding properties of 7A12B3 and 9F9C7 mAbs, we initially performed antigen-capture ELISA wherein CTB, CTX, and LTB were coated on the plates. Both 7A12B3 and 9F9C7 mAbs showed similar high binding affinities to CTB, expressed as half-maximal effective concentrations (EC50) of 0.028 ± 0.003 and 0.036 ± 0.003 µg/mL, respectively (Fig. 2, left panel). Likewise, 7A12B3 and 9F9C7 mAbs bound to CTX with high binding affinities represented by low EC50 values (0.017 ± 0.003 and 0.041 ± 0.005 µg/mL, respectively). The 9F9C7 mAb showed slightly lower affinity to CTX than 7A12B3, possibly due to marginal occlusion of the epitope by the holotoxin A subunit (Fig. 2, middle panel). Despite LTB’s high amino acid sequence similarity to CTB33,34, both 7A12B3 and 9F9C7 mAbs showed a substantially lower affinity to LTB than to CTB and CTX, with an EC50 value of 1.109 ± 0.162 and 135 ± 70 µg/mL, respectively (Fig. 2, right panel). Average EC50 values of 7A12B3 and 9F9C7 mAbs binding to all three molecules are presented in Table 1.Figure 2Analysis of 7A12B3 and 9F9C7 mAbs binding to CTX, CTB and LTB in antigen-capture ELISA. ELISA plates were coated with 2 µg/mL of CTB, CTX or the Escherichia coli heat-labile enterotoxin B subunit (LTB). Three-fold serially diluted 7A12B3 or 9F9C7 mAbs (3000—0.051 ng/mL for CTB and CTX; 100,000—1.69 ng/mL for LTB) were added to the plates and incubated, and plate-bound mAbs were detected with an anti-rat IgG secondary antibody. Representative graphs are shown. The assays were performed in triplicate, and each data point represents the mean ± SD. Data were analyzed and plotted using the GraphPad Prism 9 software and obtained from at least two independent experiments. The half-maximal effective concentrations (EC50s) were determined by nonlinear regression analysis (GraphPad Prism 9) and displayed in Table 1.Table 1Average EC50 values and standard deviation ($$n = 2$$) of 7A12B3 and 9F9C7 mAbs binding to CTX, CTB and LTB.CTXCTBLTBEC50 (μg/mL) 7A12B3 mAb0.017 ± 0.0030.028 ± 0.0031.09 ± 0.162 9F9C7 mAb0.041 ± 0.0050.036 ± 0.003 > 100 To further dissect the antigen-binding profiles of 7A12B3 and 9F9C7, SPR analysis was employed, in which each mAbs was immobilized on a CM5 sensor chip while CTX, CTB, EPICERTIN, and LTB were used as soluble analytes. Figure 3 shows representative sensorgrams. Analysis of binding kinetics revealed that 7A12B3 mAb had an average association rate constant (kon) of 1.4 × 106 (1/Ms), a dissociation rate constant (koff) of 1.8 × 10–4 (1/s), and an average equilibrium dissociation constant (KD) of 129 pM to CTX. This mAb also showed similar high binding affinity to CTB and EPICERTIN, with average KD values of 88.9 and 159 pM, respectively (Fig. 3A). On the other hand, 9F9C7 mAb showed slower association and slower dissociation for CTX and CTB compared to 7A12B3. The 9F9C7 mAb also showed slower association but similar dissociation to EPICERTIN, compared to 7A12B3 (Fig. 3B). Thus, 9F9C7 mAbs turned out to have overall lower binding affinities to the three analytes, as 9F9C7 showed an average KD of 6.1 nM to CTX, 4.4 nM to CTB, and 33.2 nM to EPICERTIN. These values correspond to approximately 50 times (CTX and CTB) and 200 times (EPICERTIN) lower affinity when compared to 7A12B3. In sharp contrast, neither mAbs showed measurable binding to LTB under the conditions used in this SPR analysis (Fig. 3A,B). Average association and dissociation rate constants and KD values are summarized in Table 2.Figure 3SPR analysis of 7A12B3 and 9F9C7 mAbs binding affinities to CTX, CTB, EPICERTIN and LTB. Each mAb was immobilized on a CM5 sensor chip. Each analyte (CTX, CTB, EPICERTIN, and LTB) was tested in a range of concentrations (0, 1.68, 3.37, 6.73, 13.47, 26.9375 nM) against immobilized 7A12B3 mAb or 9F9C7 mAb (0, 106.3, 212.5, 425, 850, 1670 nM). The collected kinetic data were blank subtracted, so the concentration at 0 nM is not shown. Representative sensorgrams for (A) 7A12B3 and (B) 9F9C7 are shown after Bivalent model fitting. Each experiment was conducted with at least three replicates. Table 2Association and dissociation constants of CTX, CTB and EPICERTIN binding to 7A12B3 and 9F9C7 mAbs generated by surface plasmon resonance.kon (1/Ms)koff (1/s)KD (pM)Chi2 (RU2)7A12B3 mAb CTX1.4 × 106 (1.3 × 106–1.4 × 106)1.8 × 10–4 (1.5 × 10–4–2.1 × 10–4)129.0 (120.0–151.0)7.6 × 10–3 (5.8 × 10–3–8.8 × 10–3) CTB2.6 × 106 (2.4 × 106–2.8 × 106)2.3 × 10–4 (2.0 × 10–4–2.6 × 10–4)88.9 (81.1–103.0)10.0 × 10–3 (7.2 × 10–3–16.0 × 10–3)EPICERTIN1.1 × 106 (1.0 × 106–1.2 × 106)1.7 × 10–4 (1.6 × 10–4–1.7 × 10–4)159.0 (145.0–173.0)6.0 × 10–3 (4.1 × 10–3–8.0 × 10–3)kon (1/Ms)koff (1/s)KD (nM)Chi2(RU2)9F9C7 mAb CTX1.5 × 104 (1.4 × 104–1.5 × 104)9.0 × 10–5 (8.4 × 10–5–10.0 × 10–5)6.1 (5.6–6.5)8.3 × 10–3 (7.3 × 10–3–9.3 × 10–3) CTB2.2 × 104 (2.1 × 104–2.3 × 104)9.9 × 10–5 (1.0 × 10–4–10.0 × 10–5)4.4 (4.3–4.5)9.6 × 10–3 (7.2 × 10–3–10.6 × 10–3) EPICERTIN4.9 × 103 (4.8 × 103–5.0 × 103)1.6 × 10–4 (1.4 × 10–4–1.9 × 10–4)33.2 (28.0–39.0)3.0 × 10–2 (2.8 × 10–2–3.1 × 10–2) ## The 7A12B3 mAb, but not 9F9C7, blocks CTB binding to its receptor GM1 A GM1-capture ELISA was initially conducted to analyze the impact of 7A12B3 and 9F9C7 mAbs on the receptor binding activity of CTB. Consistent with our observations from the culture supernatants of the parental hybridoma clones (Fig. 1B), the 7A12B3 mAb markedly blocked the binding of CTB to GM1 whereas no significant effect was observed in the presence of 9F9C7 mAb (data not shown). To analyze more rigorously the blocking properties of 7A12B3 mAb in CTB-GM1 interaction, we used a competitive GM1/KDEL ELISA, in which EPICERTIN bound to the glycosphingolipid receptor was detected using anti-KDEL mAb22. The 7A12B3 mAb at 0.1–1 µg/mL dose-dependently inhibited the binding of EPICERTIN to GM1 (Fig. 4A, left panel). In contrast, 9F9C7 mAb showed a relatively smaller effect on EPICERTIN binding to GM1 and only at a concentration of 1 µg/mL shifted the EPICERTIN binding curve to a level very similar to that observed with 0.1 µg/mL of 7A12B3 mAb (Fig. 4A, right panel). In a competitive GM1/KDEL ELISA in which varying concentrations of respective mAbs were pre-incubated with a fixed concentration of EPICERTIN at 100 ng/mL, the IC50 values for 7A12B3 and 9F9C7 mAbs on the EPICERTIN binding to GM1 were determined to be 201.2 vs. 993.7 ng/mL respectively (Fig. 4B).Figure 4The 7A12B3 mAb, but not 9F9C7, blocks EPICERTIN binding to GM1 ganglioside. EPICERTIN and GM1-capture KDEL-detection (GM1/KDEL) ELISA methods were employed to reveal the impact of 7A12B3 and 9F9C7 mAbs on CTB’s binding to the ganglioside receptor. ( A) Three concentrations of each antibody (0.1, 0.3 and 1.0 μg/mL) were preincubated with varying concentrations of EPICERTIN and applied to ELISA plates coated with GM1. Plate bound EPICERTIN was detected with an anti-KDEL mAb, as described previously22. The 7A12B3 mAb concentration-dependently blocked the interaction of EPICERTIN with GM1 at 0.1–1 µg/mL, whereas the 9F9C7 mAb had much less effects. ( B) A competitive GM1/KDEL ELISA was employed to determine the effect of mAbs on EPICERTIN–GM1 interaction. Varying concentrations of respective mAbs, including a rat IgG2a isotype control, were pre-incubated with 100 ng/mL of EPICERTIN and applied to ELISA plates coated with GM1. Plate-bound EPICERTIN was detected with an anti-KDEL mAb, as described previously22. Half maximal inhibitory concentration (IC50) of 7A12B3 (201.2 ng/mL) and 9F9C7 (993.7 ng/mL) mAbs were determined by a non-linear regression analysis using GraphPad Prism 9. Data were obtained from at least three independent experiments, and representative graph from one experiment is shown. Each data point represents the mean ± range of duplicate samples. ## The 7A12B3 mAb effectively inhibits CTX-induced cAMP in Caco2 cells The inhibitory effects of 7A12B3 and 9F9C7 mAbs on the biological functions of CTX were analyzed in the Caco2 cell line model of cAMP induced by CTX. Figure 5A shows that CTX (0.5 µg/mL) preincubated with 7A12B3 mAb (1 µg/mL) induced a significantly lower level of cytoplasmic cAMP in Caco2 cells when compared to CTX alone ($86.7\%$ inhibition; $p \leq 0.0001$), whereas 9F9C7 mAb showed significant yet less inhibitory effect ($62.6\%$ inhibition; $$p \leq 0.0032$$). The inhibitory effect of 7A12B3 mAbs was significantly different from that of 9F9C7 mAb ($$p \leq 0.0274$$) or a rat IgG2a isotype control ($$p \leq 0.0002$$). The marginal inhibition observed with a rat IgG2a isotype control antibody was not statistically significant from the PBS vehicle control ($$p \leq 0.0626$$). The inhibitory effects of 7A12B3 and 9F9C7 mAbs on CTX-induced elevation of cAMP in Caco2 cells were concentration dependent (Fig. 5B). When CTX was co-incubated with 1 µg/mL of mAbs, 7A12B3 inhibited the induction of cAMP by $88.1\%$, whereas significantly less inhibition was observed with 9F9C7 mAb ($66.8\%$). Both mAbs had minimal inhibitory effects at 0.25 µg/mL ($14\%$ vs. $12\%$, respectively), which were indistinguishable from the background effects of a Rat IgG isotype control antibody. Figure 5The 7A12B3 mAb effectively inhibits CTX-induced cAMP levels in Caco2 cells. ( A) The inhibitory effects of 7A12B3 and 9F9C7 mAbs (0.25- 1.0 µg/mL) on CTX-induced cAMP were evaluated under preincubation of the mAbs with CTX (0.5 µg/mL) for 20 min. The 7A12B3 mAb (1.0 μg/mL) strongly inhibited CTX-induced cAMP levels in Caco2 cells, whereas 9F9C7 at the same concentration had reduced inhibitory effects. ** $p \leq 0.0001$, $p \leq 0.01$, one-way measures ANOVA with Bonferroni’s multiple comparisons tests. Inset shows a standard curve of cAMP, with a non-linear regression analysis (GraphPad Prism 9) used to determine cAMP values in samples. ( B) Concentration-dependent inhibition of CTX (0.5 μg/mL)-induced cAMP levels in Caco2 cells by 7A12B3 and 9F9C9 mAbs. * $p \leq 0.05$, $p \leq 0.01$, two-way measures ANOVA with Bonferroni’s multiple comparisons tests. Data obtained from two independent experiments. Each data point represents the mean ± SD of triplicate samples. ## The 9F9C7 mAb effectively detects CTB docking on the surface of target cells Flow cytometric analysis was conducted to evaluate the utility of 7A12B3 and 9F9C7 mAbs to detect CTB and its variants bound to the surface of target cells. Although 7A12B3 was able to detect EPICERTIN on the surface of Caco2 epithelial cells, 9F9C7 showed superior detectability with increased fluorescence signal at the same concentration used (Fig. 6A, left panel). Thus, the latter mAb was used in further analysis. Figure 6The 9F9C7 mAb detects EPICERTIN but not EPICERTING33D bound to the surface of human Caco2 cell line and to mouse spleen leukocytes. ( A) Flow cytometry analysis of EPICERTIN binding to Caco2 cells detected with 2 µg/mL of 7A12B3 or 9F9C7 mAbs followed by FITC-labeled goat-anti-rat IgG (left panel). Detection of EPICERTIN, CTB and EPICERTING33D (2 µg/mL) binding to human Caco2 cell line detected with anti-CTB 9F9C9 mAb (right panel). ( B) Staining pattern and gated spleen leukocytes that were screened for EPICERTIN and EPICERTING33D binding by flow cytometry. Spleen leukocytes were incubated with 5 μg/mL of EPICERTIN or EPICERTING33D for 30 min on ice and then stained with a panel of fluorochrome-labeled antibodies and FITC-conjugated 9F9C7 anti-CTB mAb. ( C) Histograms and geometric mean fluorescence intensity of EPICERTIN and EPICERTING33D bound to the mouse spleen leukocytes gated in panel B. Using the 9F9C7 mAb, we observed strong and comparable binding of EPICERTIN and CTB to the surface of Caco2 cells, similar to our previous findings23. However, here we found that EPICERTING33D, a variant lacking GM1-binding activity, was only marginally detected on the cell surface, while the whole population of Caco2 cells bound CTB and EPT (Fig. 6A, right panel). Next, we attempted to characterize EPICERTIN’s target immune cells using 9F9C7 mAb. To this end, mouse spleen cells were incubated with EPICERTIN or EPICERTING33D, followed by staining with different combination of cell-surface marker-specific antibodies and 9F9C7, and gated to sort target cell subpopulations, as shown in Fig. 6B. The geometric mean fluorescence intensity (G-MFI) of EPICERTIN and EPICERTING33D detected with FITC-labeled goat anti-Rat IgG antibody above the background levels generated by this second antibody alone is shown in Fig. 6C. We found that EPICERTIN efficiently bound to the surface of all myeloid and lymphoid cells analyzed, with major histocompatibility complex class II (MHC II)-positive dendritic cells and macrophages being the most prominent targets. Surprisingly, EPICRTING33D appeared to recognize some of the cell types, including B cells, monocytes, macrophages, and dendritic cells, although the degrees of binding to these cells were overall much lower compared to EPICERTIN (Fig. 6C). ## The 9F9C7 mAb detects EPICERTIN by immunohistochemistry on frozen colon sections Fluorescent immunohistochemistry was conducted using FITC-conjugated 9F9C7 mAb to detect and locate EPICERTIN bound to the surface of colon epithelial cells upon oral administration of the protein in mice. EPICERTIN was detected on frozen colon tissue sections at 6, 12 and 24 h after oral administration of the protein, whereas it was not detected in untreated animal tissues or 0 and 3 h after oral administration. Representative confocal images of an EPICERTIN-treated animal tissue isolated 24 h post oral administration and a tissue from a control untreated animal are shown in Fig. 7. The tissue was stained with antibodies to the epithelial cell–cell adhesion protein E-cadherin and the lectin WGA to discriminate the apical plasma membrane of colon epithelial cells and the membrane-enclosed secretory granules of goblet cells within the crypt. Of note, the fluorescence signal was consistently detected on epithelial cells within the colonic glands while not prominent on epithelial cells facing the luminal side of the colon (Fig. 7). The staining of EPICERTIN delineated the luminal side of differentiated crypt epithelial cells (near the colon crypt opening) and less differentiated epithelial cells located at the bottom of the crypts. Additionally, the image disclosed that EPICERTIN’s fluorescence signal at the plasma membrane seem to follow closely that of the adhesion molecule E-cadherin (although not overlapping) in less differentiated and mucin-rich crypt epithelial cells. Meanwhile, FITC-9F9C7 mAb did not show any fluorescence signal in colon tissues from EPICERTIN-untreated mice, confirming the specificity of the antibody. Figure 7IHC detection of EPICERTIN bound to the surface epithelial cells of mouse colon tissue. Confocal laser scanning microcopy of colon tissues from mice treated or left untreated with EPICERTIN by oral gavage. Colon tissues were collected from EPICERTIN-dosed mice at 0, 3, 6, 12, and 24 h after oral administration or from untreated animals and embedded in OCT compound for cryosectioning. Seven microns thick tissue sections were made and stained with 5 μg/mL Pheno Vue Fluor 594-WGA, 2 μg/mL anti-E-cadherin and 2 μg/mL anti-CTB (FITC-9F9C7 mAb). Representative confocal images of an EPICERTIN-treated animal tissue isolated at 24 h and a tissue from a control untreated animal are shown. White arrows indicate distribution of EPT at the plasma membrane of individual colon crypt epithelial cells. Images were collected with a Nikon A1R Confocal laser scanning microscope using 20 × (top panel) and 60 × (lower panel) magnification lenses with appropriate channels and the data processed with the NIS Elements imaging software. Areas delineated by dotted line squares in panel A correspond to high magnification images shown in lower panel. ## Discussion Since the introduction of hybridoma technology over 45 years ago30, several anti-CTX mAbs targeting different epitopes on the A and B subunits have been generated in early studies35–40. Some of those mAbs recognized the GM1 receptor binding site of CTB or showed distinctive neutralizing CTX activity41, whereas other mAbs that were generated against CTX peptides, often resulted in the generation of mAbs with polyspecific binding properties or completely lacked CTX binding activity42,43. They aided in building the current understanding of CTX secretion, assembly44,45, endocytosis and intoxication46, and were also instrumental in understanding the potent immunogenicity of CTB and the structurally homologous LTB36,47. However, most of those anti-CTB mAbs were characterized only using outdated immunoassay-based methods, providing limited information about their binding profiles. Recently, novel recombinant CTB variants and fusion molecules have been generated, some of which were found to have unique biological functions, such as mucosal healing promoted by a CTB variant containing an ER retention motif, EPICERTIN21,23. To aid in the preclinical development of CTB-based vaccines and biotherapeutics, we attempted to isolate and characterize new anti-CTB mAbs that are suitable for mechanistic investigations and pharmacological studies. The 7A12 and 9F9 hybridoma cell culture supernatants were found to bind CTB with distinctive features in both antigen- and GM1-capture ELISAs (Fig. 1). Both mAbs bound CTB with high affinity, but only 9F9 effectively bound CTB in GM1-capture ELISA, indicating that 7A12 recognizes an epitope near or within the region of CTB responsible for GM1 interaction. On the other hand, the 9F9 hybridoma supernatant appeared to recognize a distinct epitope most probably not involved in GM1 binding. These results demonstrate that our screening procedure employed here successfully led to the isolation of two mAbs with distinct CTB-binding profiles in terms of reactivity with the antigen’s GM1-receptor binding site. The 7A12B3 and 9F9C7 mAbs were found to bind the native CTX and CTB with similar binding affinities in direct ELISA (Fig. 2, Table 1). However, SPR analysis revealed that these mAbs have distinct binding kinetics. The overall binding affinity of 7A12B3 mAb was higher than that of 9F9C7 mAb; 159 pM vs. 33.2 nM for EPICERTIN, 129 pM vs. 6.1 nM for CTX, and 88.9 pM vs. 4.4 nM for CTB, respectively (Fig. 3, Table 2). In contrast, neither 7A12B3 or 9F9C7 bound to LTB in SPR (Fig. 3), along the lines of the ELISA data that also showed substantially low affinity of these mAbs to LTB compared to CTB and CTX (Fig. 2). These results demonstrate the exquisite specificity of 7A12B3 and 9F9C7 mAbs to CTB, given that LTB has high (~ $84\%$) amino-acid sequence homology with CTB33,34. The binding affinity of 7A12B3 to EPICERTIN was slightly lower than to CTX or CTB, and those differences might be explained by the Asn4 → Ser mutation and/or the presence of C-terminal extension comprised of the hexapeptide SEKDEL sequence in EPICERTIN. Of note, 7A12B3 mAb but not 9F9C7 effectively inhibited the binding of EPICERTIN to GM1 ganglioside (Fig. 4), strengthening the idea that the former recognizes an epitope near the GM1 binding site of CTB, whereas the latter is relatively indifferent to CTB-GM1 interaction. CTX induces cAMP overproduction in the cytoplasm of target cells. Our data demonstrated that the 7A12B3 mAb has strong CTX-neutralizing effects, almost completely inhibiting the cytoplasmic accumulation of cAMP induced by CTX in Caco2 cells (Fig. 5B). Interestingly, even though 9F9C7 mAb appeared to bind to an epitope distal to the GM1-binding site, we found that the mAb was also able to inhibit the effects of CTX on the elevation of cytoplasmic cAMP in Caco2 cells, although at lower levels than 7A12B3 mAb. We speculate that 9F9C7 mAb may form complexes with CTX in solution, which in turn collaterally compromises CTX-GM1 interaction and/or entry to target cells. Based on the results from the competitive ELISA (Fig. 4) and CTX cAMP reporter assays (Fig. 5), 7A12B3 was thought to target an epitope proximal to the GM1 binding site, an area of CTB that would be occluded after engaging the cell-surface glycosphingolipid receptor. However, flow cytometry analysis revealed that the mAb is capable of detecting EPICERTIN on the surface of Caco2 epithelial cells (Fig. 6A, left panel). Nevertheless, 9F9C7, which was selected based on effective recognition of GM1-bound CTB (Fig. 1B), showed superior detectability of cell-bound EPICERTIN and thus justified the use of this mAb to explore the target cell binding profile of EPICERTIN. The flow cytometry analysis (Fig. 6) revealed that EPICERTIN and CTB equally bound to the surface of Caco2 cells, as anticipated from their similar binding affinity to GM1 ganglioside19. In sharp contrast, EPICERTING33D was only marginally detected on the cell surface (Fig. 6A, right panel), suggesting that the glycosphingolipid is the primary receptor for EPICERTIN in the colon epithelial cell line. To our surprise, however, we found inconsistent binding patterns of EPICERTIN and EPICERTING33D in mouse spleen leukocytes. For instance, EPICERTIN’s geometric mean fluorescence intensity (gMFI) ranged between 50 and 300 whereas the gMFI of EPICERTING33D ranged from 0 to 45 (Fig. 6B,C). Although GM1 ganglioside has been long considered the sole receptor for CTB binding and internalization by epithelial cells, recent findings pointed to the presence of alternative receptors, such as fucosylated glycoconjugates48–50. In addition, cycling of KDEL receptors between the Golgi and cell membrane51 could partly account for the cellular binding patterns of EPICERTIN and the G33D variant. Thus, differential expression of those receptors might explain CTB binding to leukocytes in a cell type specific manner. Nevertheless, because the degree of binding was overall substantially higher with EPICERTIN than with the non-GM1-binding counterpart, it seems reasonable to assume that EPICERTIN’s effects on immune cells are likely mediated by GM1 receptor engagement. The expression of GM1 is not limited to intestinal epithelial cells. It is expressed in a variety of other cell types, including cortical and peripheral neurons52,53 and leukocytes50, among others. A differential expression of GM1 on human monocytes suggested the presence of two monocytes subpopulations with functional differences in terms of endocytic activity and lipopolysaccharide responsiveness in peripheral blood54. CTB is known to bind to GM1 expressed on the surface of leukocytes, particularly innate immune cells such as dendritic cells, macrophages and B cells, which are the major antigen-presenting cells8. CTB binding to GM1 on B cells was associated with cAMP-independent inhibition of mitogen-stimulated B cell proliferation and enhanced expression of MHCII molecules26,55, whereas binding of CTB on T lymphocytes was found to inhibit mitogen or antigen-induced T-cell proliferation26. Of note, however, the nature of the enhanced immune responses to antigens coupled to CTB and the dampening of autoimmune responses by this protein are still largely unknown. In the case of antibody-mediated immune responses against infectious microorganisms, the increased MHC II expression on B cells induced by CTB might partially explain the immunomodulatory effect favoring this outcome8. In the case of suppression of airway allergic inflammation, CTB’s therapeutic effect appeared to reside in its capacity to reprogram dendritic cells to instruct B cells for IgA class switch56. As shown in Fig. 6, EPICERTIN highly bound to antigen-presenting cells compared to other leukocytes, particularly MHC II+ CD11clo dendritic cells, MHC II+ F480+ macrophages and CD19+ B cells. Although it remains a matter of speculation at this point, such preferential binding may suggest EPICERTIN’s distinctive effects on these cells that could have implications for the protein’s immunomodulatory effects. The specific interaction of CTB with GM1 ganglioside expressed on the surface of intestinal epithelial cells is a well-known mechanism responsible for the internalization of CTX and its virulence during V. cholerae infection57. This high affinity interaction has been exploited in vaccine development where CTB is used as an adjuvant and carrier protein. Additionally, the ability of CTB to undergo retrograde transportation in target cells may provide opportunities for the development of novel pharmaceutical products with unique biological functions, as exemplified by EPICERTIN, which was found to be retained in the ER of colon epithelial cells where it induces an unfolded protein response leading to epithelial repair activity21. However, the type of colon epithelial cell targeted/responsible for such a response remains to be identified. In the IHC analysis on cryosections of mouse colon tissue using the 9F9C7 mAb (Fig. 7), we were able to clearly detect EPICERTIN in the colon at 6 h and up to 24 h after oral administration. Interestingly, EPICERTIN was detected mainly on the surface of epithelial cells lining the openings of colonic crypts with consistent detection on less differentiated cells at the bottom of the crypts, including crypt-resident goblet cells that are densely stained with the WGA lectin58. This observation suggests that EPICERTIN might have prominent effects on the colon stem cell compartment with proliferative capability than on differentiated epithelial cells. However, this conjecture needs further verification as we cannot rule out the possibility that the detection of EPICERTIN mostly in the crypt base region might be a procedural artifact during the flushing procedure of colons before tissue embedding, which could have inadvertently removed EPICERTIN bound to the inter-crypt epithelium exposed on the luminal side. Our future study will address this issue by further IHC analysis of ex vivo-cultured mouse and human colon tissues. In conclusion, the present study demonstrated that mAbs 7A12B3 and 9F9C7 bind CTX, CTB and EPICERTIN with high affinity and specificity. The 7A12B3 mAb effectively inhibited the binding of CTB to GM1 and neutralized CTX, whereas the 9F9C7 mAb showed superior capacity to detect EPICERTIN binding to the surface of target cells. Coupled with our earlier reports showing the utility of 9F9C7 in immunofluorescence and immunoprecipitation23 and 7A12B3 in rodent pharmacokinetic analysis of EPICERTIN25, these mAbs provide valuable tools to facilitate the investigation and development of CTB variants as novel biopharmaceutical candidates. ## References 1. Harris J, LaRocque R, Qadri F, Ryan E, Calderwood S. Cholera*. Lancet* (2012) 379 2466-2476. DOI: 10.1016/S0140-6736(12)60436-X 2. Cuatrecasas P. Interaction of. Biochemistry (1973) 12 3547-3558. DOI: 10.1021/bi00742a031 3. Holmgren J, Lönnroth I, Svennerholm L. Fixation and inactivation of cholera toxin by GM1 ganglioside. Scand. J. Infect. Dis. (1973) 5 77-78. DOI: 10.3109/inf.1973.5.issue-1.15 4. Merritt EA. Crystal structure of cholera toxin B-pentamer bound to receptor GM1 pentasaccharide. Protein Sci. (1994) 3 166-175. DOI: 10.1002/pro.5560030202 5. Holmgren J, Czerkinsky C, Lycke N, Svennerholm A-M. Strategies for the induction of immune responses at mucosal surfaces making use of cholera toxin B subunit as immunogen, carrier, and adjuvant. Am. J. Trop. Med. Hyg. (1994) 50 42-54. PMID: 8203723 6. Azegami T, Itoh H, Kiyono H, Yuki Y. Novel transgenic rice-based vaccines. Arch. Immunol. Ther. Exp. (2015) 63 87-99. DOI: 10.1007/s00005-014-0303-0 7. Baldauf KJ, Royal JM, Hamorsky KT, Matoba N. Cholera toxin B: one subunit with many pharmaceutical applications. Toxins (2015) 7 974-996. DOI: 10.3390/toxins7030974 8. Stratmann T. Cholera toxin subunit B as adjuvant—An accelerator in protective immunity and a break in autoimmunity. Vaccines (2015) 3 579-596. DOI: 10.3390/vaccines3030579 9. Solbreux PM, Dive C, Vaerman J-P. Anti-cholera toxin IgA-, IgG-and IgM-secreting cells in various rat lymphoid tissues after repeated intestinal or parenteral immunizations. Immunol. Invest. (1990) 19 435-451. DOI: 10.3109/08820139009052971 10. Apter F, Lencer W, Finkelstein R, Mekalanos J, Neutra M. Monoclonal immunoglobulin A antibodies directed against cholera toxin prevent the toxin-induced chloride secretory response and block toxin binding to intestinal epithelial cells in vitro. Infect. Immun. (1993) 61 5271-5278. DOI: 10.1128/iai.61.12.5271-5278.1993 11. Sanchez J, Johansson S, Löwenadler B, Svennerholm A, Holmgren J. Recombinant cholera toxin B subunit and gene fusion proteins for oral vaccination. Res. Microbiol. (1990) 141 971-979. DOI: 10.1016/0923-2508(90)90137-F 12. Vendetti S. Polyclonal Treg cells enhance the activity of a mucosal adjuvant. Immunol. Cell Biol. (2010) 88 698-706. DOI: 10.1038/icb.2010.76 13. Arakawa T. A plant-based cholera toxin B subunit–insulin fusion protein protects against the development of autoimmune diabetes. Nat. Biotechnol. (1998) 16 934-938. DOI: 10.1038/nbt1098-934 14. Tarkowski A, Sun JB, Holmdahl R, Holmgren J, Czerkinsky C. Treatment of experimental autoimmune arthritis by nasal administration of a type II collagen–cholera toxoid conjugate vaccine. Arthritis Rheum. (1999) 42 1628-1634. DOI: 10.1002/1529-0131(199908)42:8<1628::AID-ANR10>3.0.CO;2-T 15. Rask C. Prolonged oral treatment with low doses of allergen conjugated to cholera toxin B subunit suppresses immunoglobulin E antibody responses in sensitized mice. Clin. Exp. Allergy (2000) 30 1024-1032. DOI: 10.1046/j.1365-2222.2000.00849.x 16. Phipps PA. Prevention of mucosally induced uveitis with a HSP60-derived peptide linked to cholera toxin B subunit. Eur. J. Immunol. (2003) 33 224-232. DOI: 10.1002/immu.200390025 17. Ruhlman T, Ahangari R, Devine A, Samsam M, Daniell H. Expression of cholera toxin B–proinsulin fusion protein in lettuce and tobacco chloroplasts—Oral administration protects against development of insulitis in non-obese diabetic mice. Plant Biotechnol. J. (2007) 5 495-510. DOI: 10.1111/j.1467-7652.2007.00259.x 18. Sun JB, Czerkinsky C, Holmgren J. Mucosally induced immunological tolerance, regulatory T cells and the adjuvant effect by cholera toxin B subunit. Scand. J. Immunol. (2010) 71 1-11. DOI: 10.1111/j.1365-3083.2009.02321.x 19. Hamorsky KT. Rapid and scalable plant-based production of a cholera toxin B subunit variant to aid in mass vaccination against cholera outbreaks. PLoS Negl. Trop. Dis. (2013) 7 e2046. DOI: 10.1371/journal.pntd.0002046 20. Reeves MA. Spray-dried formulation of epicertin, a recombinant cholera toxin B subunit variant that induces mucosal healing. Pharmaceutics (2021) 13 576. DOI: 10.3390/pharmaceutics13040576 21. Baldauf K. Oral administration of a recombinant cholera toxin B subunit promotes mucosal healing in the colon. Mucosal Immunol. (2017) 10 887-900. DOI: 10.1038/mi.2016.95 22. Morris DA, Reeves MA, Royal JM, Hamorsky KT, Matoba N. Isolation and detection of a KDEL-tagged recombinant cholera toxin B subunit from. Process Biochem. (2021) 101 42-49. DOI: 10.1016/j.procbio.2020.10.018 23. Royal JM. A modified cholera toxin B subunit containing an ER retention motif enhances colon epithelial repair via an unfolded protein response. FASEB J. (2019) 33 13527-13545. DOI: 10.1096/fj.201901255R 24. Royal JM, Reeves MA, Matoba N. Repeated oral administration of a KDEL-tagged recombinant cholera toxin B subunit effectively mitigates DSS colitis despite a robust immunogenic response. Toxins (2019) 11 678. DOI: 10.3390/toxins11120678 25. Tuse D. Pharmacokinetics and safety studies in rodent models support development of EPICERTIN as a novel topical wound-healing biologic for ulcerative colitis. J. Pharmacol. Exp. Ther. (2022) 380 162-170. DOI: 10.1124/jpet.121.000904 26. Woogen SD, Ealding W, Elson CO. Inhibition of murine lymphocyte proliferation by the B subunit of cholera toxin. J. Immunol. (1987) 139 3764-3770. DOI: 10.4049/jimmunol.139.11.3764 27. George-Chandy A. Cholera toxin B subunit as a carrier molecule promotes antigen presentation and increases CD40 and CD86 expression on antigen-presenting cells. Infect. Immun. (2001) 69 5716-5725. DOI: 10.1128/IAI.69.9.5716-5725.2001 28. Canziani GA, Klakamp S, Myszka DG. Kinetic screening of antibodies from crude hybridoma samples using Biacore. Anal. Biochem. (2004) 325 301-307. DOI: 10.1016/j.ab.2003.11.004 29. Matoba N, Davis KR, Palmer KE. Recombinant protein expression in. Methods Mol. Biol. (2011) 701 199-219. DOI: 10.1007/978-1-61737-957-4_11 30. Kohler G, Milstein C. Continuous cultures of fused cells secreting antibody of predefined specificity. Nature (1975) 256 495-497. DOI: 10.1038/256495a0 31. Tabll A, Abbas AT, El-Kafrawy S, Wahid A. Monoclonal antibodies: Principles and applications of immmunodiagnosis and immunotherapy for hepatitis C virus. World. J. Hepatol. (2015) 7 2369-2383. DOI: 10.4254/wjh.v7.i22.2369 32. Yu RK, Usuki S, Itokazu Y, Wu HC. Novel GM1 ganglioside-like peptide mimics prevent the association of cholera toxin to human intestinal epithelial cells in vitro. Glycobiology (2016) 26 63-73. DOI: 10.1093/glycob/cww015 33. Holmner A, Askarieh G, Okvist M, Krengel U. Blood group antigen recognition by. J. Mol. Biol. (2007) 371 754-764. DOI: 10.1016/j.jmb.2007.05.064 34. Lebens M. Synthesis of hybrid molecules between heat-labile enterotoxin and cholera toxin B subunits: Potential for use in a broad-spectrum vaccine. Infect. Immun. (1996) 64 2144-2150. DOI: 10.1128/iai.64.6.2144-2150.1996 35. Jobling MG, Holmes RK. Mutational analysis of ganglioside GM1-binding ability, pentamer formation, and epitopes of cholera toxin B (CTB) subunits and CTB/heat-labile enterotoxin B subunit chimeras. Infect. Immun. (2002) 70 1260-1271. DOI: 10.1128/IAI.70.3.1260-1271.2002 36. Holmes RK, Twiddy EM. Characterization of monoclonal antibodies that react with unique and cross-reacting determinants of cholera enterotoxin and its subunits. Infect. Immun. (1983) 42 914-923. DOI: 10.1128/iai.42.3.914-923.1983 37. Robb M, Nichols JC, Whoriskey SK, Murphy JR. Isolation of hybridoma cell lines and characterization of monoclonal antibodies against cholera enterotoxin and its subunits. Infect. Immun. (1982) 38 267-272. DOI: 10.1128/iai.38.1.267-272.1982 38. Chou SF. Production and purification of monoclonal and polyclonal antibodies against cholera toxin. Hybrid. Hybridomics (2004) 23 258-261. DOI: 10.1089/1536859041651376 39. Kenimer JG, Probst PG, Karpas AB, Burns DL, Kaslow HR. Monoclonal antibodies against the enzymatic subunit of both pertussis and cholera toxins. Dev. Biol. Stand. (1991) 73 133-141. PMID: 1778307 40. Remmers EF, Colwell RR, Goldsby RA. Production and characterization of monoclonal antibodies to cholera toxin. Infect. Immun. (1982) 37 70-76. DOI: 10.1128/iai.37.1.70-76.1982 41. Ludwig DS, Holmes RK, Schoolnik GK. Chemical and immunochemical studies on the receptor binding domain of cholera toxin B subunit. J. Biol. Chem. (1985) 260 12528-12534. DOI: 10.1016/S0021-9258(17)38903-2 42. Otte L, Knaute T, Schneider-Mergener J, Kramer A. Molecular basis for the binding polyspecificity of an anti-cholera toxin peptide 3 monoclonal antibody. J. Mol. Recognit. (2006) 19 49-59. DOI: 10.1002/jmr.757 43. Anglister J, Zilber B. Antibodies against a peptide of cholera toxin differing in cross-reactivity with the toxin differ in their specific interactions with the peptide as observed by 1H NMR spectroscopy. Biochemistry (1990) 29 921-928. DOI: 10.1021/bi00456a011 44. Tinker JK, Erbe JL, Hol WG, Holmes RK. Cholera holotoxin assembly requires a hydrophobic domain at the A-B5 interface: Mutational analysis and development of an in vitro assembly system. Infect. Immun. (2003) 71 4093-4101. DOI: 10.1128/IAI.71.7.4093-4101.2003 45. Reichow SL, Korotkov KV, Hol WG, Gonen T. Structure of the cholera toxin secretion channel in its closed state. Nat. Struct. Mol. Biol. (2010) 17 1226-1232. DOI: 10.1038/nsmb.1910 46. Torgersen ML, Skretting G, van Deurs B, Sandvig K. Internalization of cholera toxin by different endocytic mechanisms. J. Cell. Sci. (2001) 114 3737-3747. DOI: 10.1242/jcs.114.20.3737 47. Belisle BW, Twiddy EM, Holmes RK. Monoclonal antibodies with an expanded repertoire of specificities and potent neutralizing activity for. Infect. Immun. (1984) 46 759-764. DOI: 10.1128/iai.46.3.759-764.1984 48. Sethi A. Cell type and receptor identity regulate cholera toxin subunit B (CTB) internalization. Interface Focus (2019) 9 20180076. DOI: 10.1098/rsfs.2018.0076 49. Wands AM. Fucosylation and protein glycosylation create functional receptors for cholera toxin. Elife (2015) 4 e09545. DOI: 10.7554/eLife.09545 50. Cervin J. GM1 ganglioside-independent intoxication by Cholera toxin. PLOS Pathog. (2018) 14 e1006862. DOI: 10.1371/journal.ppat.1006862 51. Jia J. KDEL receptor is a cell surface receptor that cycles between the plasma membrane and the Golgi via clathrin-mediated transport carriers. Cell. Mol. Life. Sci. (2021) 78 1085-1100. DOI: 10.1007/s00018-020-03570-3 52. Kozireski-Chuback D, Wu G, Ledeen RW. Developmental appearance of nuclear GM1 in neurons of the central and peripheral nervous systems. Dev. Brain Res. (1999) 115 201-208. DOI: 10.1016/S0165-3806(99)00062-0 53. Okada E, Maeda T, Watanabe T. Immunocytochemical study on cholera toxin binding sites by monoclonal anti-cholera toxin antibody in neuronal tissue culture. Brain Res. (1982) 242 233-241. DOI: 10.1016/0006-8993(82)90305-5 54. Moreno-Altamirano MMB, Aguilar-Carmona I, Sánchez-García FJ. Expression of GM1, a marker of lipid rafts, defines two subsets of human monocytes with differential endocytic capacity and lipopolysaccharide responsiveness. Immunology (2007) 120 536-543. DOI: 10.1111/j.1365-2567.2006.02531.x 55. Francis M. Cyclic AMP-independent effects of cholera toxin on B cell activation. II. Binding of ganglioside GM1 induces B cell activation. J. Immunol. (1992) 148 1999-2005. DOI: 10.4049/jimmunol.148.7.1999 56. Smits H. Cholera toxin B suppresses allergic inflammation through induction of secretory IgA. Mucosal Immunol. (2009) 2 331-339. DOI: 10.1038/mi.2009.16 57. Holmgren J, Lönnroth I, Svennerholm L. Tissue receptor for cholera exotoxin: postulated structure from studies with GM1 ganglioside and related glycolipids. Infect. Immun. (1973) 8 208-214. DOI: 10.1128/iai.8.2.208-214.1973 58. Nystrom EEL. An intercrypt subpopulation of goblet cells is essential for colonic mucus barrier function. Science (2021) 372 eabb1590. DOI: 10.1126/science.abb1590
--- title: 'Differential Impact of Body Mass Index in Hip Arthroscopy: Obesity Does Not Impact Outcomes' authors: - Misty Suri - Arjun Verma - Mohammed Asad Khalid - Michael Nammour - Deryk Jones - Brian Godshaw journal: The Ochsner Journal year: 2023 pmcid: PMC10016209 doi: 10.31486/toj.22.0077 license: CC BY 4.0 --- # Differential Impact of Body Mass Index in Hip Arthroscopy: Obesity Does Not Impact Outcomes ## Abstract Background: Hip arthroscopy is commonly used for the treatment of hip pathologies. As population obesity rates continue to increase, elucidating the impact of body mass index (BMI) on hip arthroscopy outcomes is essential. This investigation was conducted to quantify the effects of BMI on hip arthroscopy outcomes. Methods: We conducted a retrospective medical records review of 459 patients undergoing hip arthroscopy at a single center from 2008 to 2016. The Harris Hip Score (HHS) and 2 component scores of the 12-Item Short Form Survey—the physical component score (PCS-12) and the mental component score (MCS-12)—were used to measure outcomes. Patients were stratified into 4 cohorts based on their BMI: underweight (BMI <18.5 kg/m2), normal weight (BMI 18.5 to 24.9 kg/m2), overweight (BMI 25.0 to 29.9 kg/m2), and obese (BMI ≥30.0 kg/m2). Results: At 1 and 2 years postoperatively, all cohorts experienced statistically significant improvements in the HHS and PCS-12. At 3 years postoperatively, statistically significant improvements were seen in the HHS for all cohorts; in the PCS-12 for the normal weight, overweight, and obese cohorts; and in the MCS-12 for the normal weight cohort. Intercohort differences were not statistically significant at 1, 2, or 3 years postoperatively. Conclusion: In our population, BMI did not have statistically significant effects on patient outcome scores following hip arthroscopy. All patient cohorts showed postoperative improvements, and differences between BMI cohorts were not statistically significant at any postoperative time point. ## INTRODUCTION In recent decades, global obesity rates have continued to increase. Since 1975, obesity rates have more than tripled, and as of 2016, an estimated $39\%$ of adults were overweight and $13\%$ were obese.1 If current trends continue, an estimated 65 million additional Americans will be obese by 2030, raising the obesity rate to $42\%$.2,3 The financial impact of obesity cannot be understated, as, by 2030, the estimated loss of productivity related to obesity could reach $580 billion annually in the United States.3 Annual medical costs in obese individuals are, on average, $42\%$ higher than those for healthy-weight individuals.2 Interestingly, although the financial and health-related ramifications of obesity have been widely examined, the debate continues regarding the optimal measures of obesity.4 Body mass index (BMI) is calculated using a patient's height and weight and is used to classify patients as underweight (BMI <18.5 kg/m2), normal weight (BMI 18.5 to 24.9 kg/m2), overweight (BMI 25.0 to 29.9 kg/m2), and obese (BMI ≥30.0 kg/m2).5 BMI is often used as an indicator of a patient's current health and the possibility of future health problems. Importantly, while BMI is not a perfect measurement of obesity, it has been shown to possess high specificity and to be clinically equivalent to other proposed methods for identifying obesity.6 Obesity has been shown in the orthopedic literature to be associated with worse postoperative outcomes and to increase the risk of complications. Harrison et al showed that after partial meniscectomy or arthroscopic knee debridement, obese patients had worse physical functioning as measured by the 36-Item Short Form Survey compared to nonobese patients.7 Warrender et al found that after arthroscopic rotator cuff repair, obese patients had worse functional outcomes, longer operative times, and longer hospital stay compared to nonobese patients.8 A meta-analysis by Yuan et al showed a 2-fold increased risk of surgical site infection in obese patients.9 Hip arthroscopy is commonly used to treat hip pathologies such as labral tears, femoroacetabular impingement, and loose bodies. The goals of hip arthroscopy are to alleviate symptoms, improve hip function, and delay the progression of hip osteoarthritis. The use of hip arthroscopy has continued to increase in recent years, with an 18-fold increase in hip arthroscopy cases from 1999 to 2009.10 Clohisy et al found that nearly $42\%$ of patients undergoing hip arthroscopy for femoroacetabular impingement are overweight or obese.11 As hip arthroscopy has grown in popularity and the obesity rate has continued to rise, evaluation of hip arthroscopy outcomes in the obese patient is a growing need. Although a significant amount of orthopedic literature is available on obesity and outcomes after knee and shoulder arthroscopy and arthroplasty, few studies of hip arthroscopy and obesity have been done. These studies have suggested poorer outcomes in patients receiving hip arthroscopy as BMI increases.12-16 Given the increasing popularity of this procedure and rising obesity rates, we conducted this investigation to examine and quantify the differential effect of BMI and obesity on the outcomes of patients undergoing hip arthroscopy. ## METHODS After approval from the institutional review board, we conducted a retrospective review of prospectively collected data for all patients undergoing hip arthroscopy from 2008 to 2016 at a single institution. All procedures were performed by a single, fellowship-trained surgeon (MS). Inclusion criteria included hip arthroscopy with a minimum of 1 year of clinical follow-up. Exclusion criteria included revision hip arthroscopy and clinical follow-up of <1 year. Demographic data obtained from the medical records were sex, age at time of hip arthroscopy, and BMI. We stratified patients into 4 cohorts based on their BMI at the time of surgery: underweight (BMI <18.5 kg/m2), normal weight (BMI 18.5 to 24.9 kg/m2), overweight (BMI 25.0 to 29.9 kg/m2), and obese (BMI ≥30.0 kg/m2). We assessed outcomes using 2 validated methods. We used the Harris Hip Score (HHS), a commonly used questionnaire for assessing hip dysfunction, to assess patient pain, function, activity, and various physical examination data. The HHS has a maximum score of 100, with higher values indicating better outcomes.17 We also used 2 component scores of the 12-Item Short Form Survey (SF-12)—the physical component score (PCS-12) and the mental component score (MCS-12)—to assess patients’ physical and mental health. The SF-12 and its components also have a maximum score of 100, with higher scores indicating better physical and mental health.18 Patients completed the questionnaires at their clinical appointments preoperatively and at 6 weeks, 3 months, 6 months, 1 year, 2 years, and 3 years postoperatively. SAS version 9.4 for Windows (SAS Institute, Inc.) was used for all statistical analyses. Tests were performed with a significance level of α=0.05, and any values were considered statistically significant if P<α. Analysis of variance, Wilks lambda, and solution for fixed effects were used to assess outcome measures based on BMI and between BMI cohorts. Intracohort P values were calculated using paired t tests to compare preoperative and postoperative outcome scores. ## RESULTS Of the 484 patients who underwent hip arthroscopy during the study period, 25 met the exclusion criteria. Of the remaining 459 patients with recorded preoperative BMIs, $46.4\%$ were in the normal weight cohort, and the average BMI for all patients was 25.7 kg/m2. The majority of patients were female ($59.3\%$), and the average age of all patients was 35.6 years (Table 1). Table 1. \| Variable \| Value \| \| --- \| --- \| \| Age, years, mean \| 35.6 \| \| Sex \| \| \| Female \| 272 (59.3) \| \| Male \| 187 (40.7) \| \| Body mass index, kg/m2, mean \| 25.7 \| \| Body mass index cohorts \| \| \| Underweight \| 16 (3.5) \| \| Normal weight \| 213 (46.4) \| \| Overweight \| 147 (32.0) \| \| Obese \| 83 (18.1) \| The HHS results for each BMI cohort are shown in Table 2 and Figure 1. The underweight group had the highest preoperative HHS of 58.0, and the overweight group had the lowest initial score at 50.6. In the underweight, normal weight, and overweight groups, the highest HHS value was at 6 months postoperatively and then steadily declined in the subsequent years. The obese group had its highest HHS value at 1 year postoperatively, followed by a steady decline. At the 3-year postoperative time point, the obese group had the highest overall HHS, and the underweight group had the lowest overall improvement. Compared to their preoperative scores, all BMI cohorts had statistically significant improvements in the HHS at 1, 2, and 3 years postoperatively. However, the intercohort differences in HHS between BMI cohorts were not statistically significant at any of the time points. PCS-12 results for each BMI cohort are shown in Table 2 and Figure 2. The underweight group again had the highest baseline PCS-12 at 37.1, while the obese group had the lowest score at 34.2. The PCS-12 increased at each time point until 3 years postoperatively for the normal weight and overweight groups. The underweight group exhibited similar score increases at most follow-up time points, except for a decline in the PCS-12 at 6 months postoperatively. Likewise, the PCS-12 for the obese cohort increased at most follow-up points but exhibited declines at 3 months and 6 months. At 3 years postoperatively, the obese group had the highest PCS-12 at 53.6, the largest improvement from the preoperative PCS-12, with an average 19.4-point improvement. In comparison, the overweight group had a 17.2-point improvement, the normal weight group had a 15.8-point improvement, and the underweight group had the least improvement at 11.9 points. Compared to preoperative scores, all BMI cohorts experienced statistically significant improvement in the PCS-12 at 3 months, 6 months, 1 year, and 2 years postoperatively. The normal weight, overweight, and obese groups also experienced statistically significant improvements in the PCS-12 at 3 years postoperatively compared to preoperative scores. At 3 years postoperatively, the mean score in the underweight group improved compared to preoperative values, but the difference was not statistically significant. The intercohort differences between BMI cohorts for the PCS-12 were not statistically significant at any time point. Figure 2.: Mean scores for the physical component score (PCS-12) from baseline (preoperatively) to 36 months postoperatively stratified by body mass index cohort. The PCS-12, which is the physical health subcomponent of the 12-Item Short Form Survey, has a maximum score of 100, with higher scores indicating better outcomes. MCS-12 results for each BMI cohort are shown in Table 2 and Figure 3. The overweight group had the highest initial score of 51.2, and the underweight group had the lowest initial score of 48.0. At 1 year postoperatively, all BMI cohorts experienced statistically significant improvements in the MCS-12. At 2 years postoperatively, the improvement of the underweight and normal weight groups remained statistically significant. By 3 years postoperatively, only the normal weight group achieved improvement of statistical significance. At 3 years postoperatively, the obese group had the highest MCS-12 at 59.0, the largest improvement from the preoperative MCS-12 with an average 8.1-point improvement. In comparison to baseline scores, the underweight group improved by 7.5 points, the normal weight group by 5.9 points, and the overweight group by 5.1 points. The intercohort differences in the MCS-12 were not statistically significant at any time point. Figure 3.: Mean scores for the mental component score (MCS-12) from baseline (preoperatively) to 36 months postoperatively stratified by body mass index cohort. The MCS-12, which is the mental health subcomponent of the 12-Item Short Form Survey, has a maximum score of 100, with higher scores indicating better outcomes. ## DISCUSSION This retrospective review provides the results of patient-reported outcomes from different BMI cohorts and demonstrates significant improvements in all outcome measures (HHS, PCS-12, and MCS-12) in each BMI cohort with no significant difference between groups at final follow-up. Despite the paucity of orthopedic literature discussing the relationship between obesity and outcomes after hip arthroscopy, some studies have indicated that patients with higher BMIs have poorer outcomes following hip arthroscopy. Gupta et al performed 2 studies evaluating the effect of obesity on hip arthroscopy outcomes.12,14 In a 2015 study, they compared patient-reported outcomes after hip arthroscopy from 62 obese patients and 124 controls.12 Their results showed that preoperatively, obese patients started with lower patient-reported outcome scores compared to nonobese patients, and at 2 years postarthroscopy, obese patients had significantly lower patient-reported outcome scores. However, statistically significant improvement was seen in both the obese and nonobese populations. The researchers concluded that both groups demonstrated significant improvement and that the change was similar between the 2 groups.12 In another 2015 study, Gupta et al conducted a cohort analysis of 680 patients to determine if obesity impacted postoperative outcome scores.14 They found that obese patients had lower preoperative and postoperative scores compared to nonobese patients, but both nonobese and obese patients showed substantial improvement. Bech et al performed a systematic review of 3 studies on the outcomes of hip arthroscopy in obese patients compared with a nonobese cohort.15 They found that although obese patients obtained similar improvements postoperatively, obese patients had lower patient-reported outcome scores at follow-up, were 4.7 times more likely to require re-arthroscopy, and were 2.2 times more likely to require conversion to total hip replacement than nonobese patients. Bech et al concluded that because of the lower overall outcome scores and increased reoperation rate, hip arthroscopy should be used with caution in obese patients.15 Schairer et al conducted a retrospective population-based analysis of 7,351 patients to evaluate the conversion rate to total hip arthroplasty 2 years after hip arthroscopy.19 Their results showed that conversion to total hip arthroplasty was highest in patients 60 to 69 years old ($35\%$) and that obese patients were more likely to undergo total hip arthroplasty within 2 years: $22.8\%$ of obese patients vs $11.4\%$ of non-obese patients (odds ratio 2.31, $P \leq 0.001$). Of note, patients treated at low volume hip arthroscopy centers (<10 procedures performed annually) were significantly more likely to undergo total hip arthroplasty within 2 years than patients treated at medium volume (10 to 49 procedures performed annually) and high volume (>49 procedures performed annually) centers.19 Our study, however, showed that patients with higher BMIs enjoyed outcomes similar to those of patients with lower BMIs. Our results show that all BMI cohorts experienced improved patient-reported outcome scores at 1, 2, and 3 years postoperatively. Moreover, intercohort differences in patient-reported outcome scores showed no statistically significant differences at 1, 2, or 3 years postoperatively. In fact, this study found that the obese group had the greatest magnitude of improvement in all outcome measures. The obese group also had the highest HHS, PCS-12, and MCS-12 at 3 years postoperatively (Table 2). The underweight group had the least improvement in the HHS and PCS-12. These results suggest that obese patients do not experience worse outcomes at 1, 2, or 3 years postoperatively, a finding that differs from prior studies. The senior lead author of this study (MS), based on the definitions in Schairer et al,19 is a high-volume surgeon who performs >50 hip arthroscopies annually. Thus, experience may have played a role in improved outcomes across all groups. Regardless, increased BMI did not appear to be a limiter of our patients’ long-term outcomes following hip arthroscopy as reported in other studies. One of the strengths of this study is that all patients were treated in a single institution by the same surgeon under similar conditions, helping to decrease the risk of variability that may naturally occur by considering patients treated at a variety of institutions by multiple surgeons. Further, all patients were examined preoperatively and postoperatively by the treating physician using consistent assessment and physical examination, allowing for accurate documentation of patients’ progression throughout their course of treatment. A weakness of this study is that it was conducted at a single center and is entirely retrospective. Moreover, we did not use specific indications for arthroscopy and did not delineate or assess patients’ expectations of postoperative activity levels. Another weakness is that we did not examine the rate of re-arthroscopy or conversion to total hip arthroplasty. Some of the previously cited articles discuss obesity as a risk factor for reoperation and for conversion to total hip arthroplasty.12,14,15,20 This study does not address these concerns, but based on the literature, treating hip surgeons should discuss with obese patients that they may be at increased risk for reoperation or conversion to total hip arthroplasty. ## CONCLUSION The results of this study indicate no correlation between BMI and patient outcomes following hip arthroscopy in our patient population. These results are encouraging in that patients with higher BMIs may choose hip arthroscopy to treat a variety of hip pathologies while still expecting favorable outcomes regardless of body habitus. ## References 1. World Health Organization 2. Finkelstein EA, Khavjou OA, Thompson H. Obesity and severe obesity forecasts through 2030. Am J Prev Med (2012.0) 42 563-570. DOI: 10.1016/j.amepre.2011.10.026 3. Wang YC, McPherson K, Marsh T, Gortmaker SL, Brown M. Health and economic burden of the projected obesity trends in the USA and the UK [published correction appears in. Lancet (2011.0) 378 815-825. DOI: 10.1016/S0140-6736(11)60814-3 4. Sweeting HN. Measurement and definitions of obesity in childhood and adolescence: a field guide for the uninitiated. Nutr J (2007.0) 6 32. DOI: 10.1186/1475-2891-6-32 5. National Institutes of Health [published correction appears in. Obes Res (1998.0) 6 51S-209S. PMID: 9813653 6. Sommer I, Teufer B, Szelag M. The performance of anthropometric tools to determine obesity: a systematic review and meta-analysis. Sci Rep (2020.0) 10 12699. DOI: 10.1038/s41598-020-69498-7 7. Harrison MM, Morrell J, Hopman WM. Influence of obesity on outcome after knee arthroscopy. Arthroscopy (2004.0) 20 691-695. DOI: 10.1016/j.arthro.2004.06.004 8. Warrender WJ, Brown OL, Abboud JA. Outcomes of arthroscopic rotator cuff repairs in obese patients. J Shoulder Elbow Surg (2011.0) 20 961-967. DOI: 10.1016/j.jse.2010.11.006 9. Yuan K, Chen HL. Obesity and surgical site infections risk in orthopedics: a meta-analysis. Int J Surg (2013.0) 11 383-388. DOI: 10.1016/j.ijsu.2013.02.018 10. Colvin AC, Harrast J, Harner C. Trends in hip arthroscopy. J Bone Joint Surg Am (2012.0) 94 e23. DOI: 10.2106/JBJS.J.01886 11. Clohisy JC, Baca G, Beaulé PE. Descriptive epidemiology of femoroacetabular impingement: a North American cohort of patients undergoing surgery. Am J Sports Med (2013.0) 41 1348-1356. DOI: 10.1177/0363546513488861 12. Gupta A, Redmond JM, Hammarstedt JE, Stake CE, Domb BG. Does obesity affect outcomes in hip arthroscopy? A matched-pair controlled study with minimum 2-year follow-up. Am J Sports Med (2015.0) 43 965-971. DOI: 10.1177/0363546514565089 13. Bogunovic L, Gottlieb M, Pashos G, Baca G, Clohisy JC. Why do hip arthroscopy procedures fail?. Clin Orthop Relat Res (2013.0) 471 2523-2529. DOI: 10.1007/s11999-013-3015-6 14. Gupta A, Redmond JM, Hammarstedt JE, Lindner D, Stake CE, Domb BG. Does obesity affect outcomes after hip arthroscopy? A cohort analysis. J Bone Joint Surg Am (2015.0) 97 16-23. DOI: 10.2106/JBJS.N.00625 15. Bech NH, Kodde IF, Dusseldorp F, Druyts PA, Jansen SP, Haverkamp D. Hip arthroscopy in obese, a successful combination?. J Hip Preserv Surg (2015.0) 3 37-42. DOI: 10.1093/jhps/hnv076 16. Collins JA, Beutel BG, Garofolo G, Youm T. Correlation of obesity with patient-reported outcomes and complications after hip arthroscopy. Arthroscopy (2015.0) 31 57-62. DOI: 10.1016/j.arthro.2014.07.013 17. Harris WH. Traumatic arthritis of the hip after dislocation and acetabular fractures: treatment by mold arthroplasty. An end-result study using a new method of result evaluation. J Bone Joint Surg Am (1969.0) 51 737-755. PMID: 5783851 18. Ware JE, Kosinksi M, Keller SD. SF-12: How to Score the SF-12 Physical and Mental Summary Scales (1995.0) 19. Schairer WW, Nwachukwu BU, McCormick F, Lyman S, Mayman D. Use of hip arthroscopy and risk of conversion to total hip arthroplasty: a population-based analysis. Arthroscopy (2016.0) 32 587-593. DOI: 10.1016/j.arthro.2015.10.002 20. Perets I, Rybalko D, Chaharbakhshi EO, Mu BH, Chen AW, Domb BG. Minimum five-year outcomes of hip arthroscopy for the treatment of femoroacetabular impingement and labral tears in patients with obesity: a match-controlled study. J Bone Joint Surg Am (2018.0) 100 965-973. DOI: 10.2106/JBJS.17.00892
--- title: 'Stop the Divide: Facilitators and Barriers to Uptake of Digital Health Interventions Among Socially Disadvantaged Populations' authors: - Eboni G. Price-Haywood - Connie Arnold - Jewel Harden-Barrios - Terry Davis journal: The Ochsner Journal year: 2023 pmcid: PMC10016217 doi: 10.31486/toj.22.0101 license: CC BY 4.0 --- # Stop the Divide: Facilitators and Barriers to Uptake of Digital Health Interventions Among Socially Disadvantaged Populations ## Abstract Background: The coronavirus disease 2019 pandemic ushered in rapid adoption of telehealth services. This study examines patient and provider experience and provides recommendations for facilitating the use of digital health interventions among socially disadvantaged populations. Methods: This qualitative study was conducted from May to July 2021 via semistructured interviews. Forty patients and 30 primary care providers (PCPs) in Louisiana were recruited within an integrated delivery health system and a rural health center. Technology acceptance models were used to develop a thematic coding scheme. Results: Most patients self-identified as Black ($67.5\%$) and female ($72.5\%$), had a mean age of 51 years, lived in an urban area ($76.9\%$), and had Medicaid ($57.9\%$). Most PCPs were White ($79.3\%$) and male ($51.7\%$), had a mean age of 39 years, and reported Medicaid as the predominant insurer ($58.6\%$). Patient use of smartphones for internet access to health and nonhealth activities was common. PCPs noted the need to address misinformation or misinterpretation of information on the internet. Most patients had used a patient portal ($72.5\%$) and noted the convenience of messaging. PCPs reported large increases in messaging workloads. Most patients had had telemedicine visits ($65.6\%$); however, Wi-Fi/broadband problems limited video visits. PCPs expressed concerns regarding the types of chief complaints that are appropriate for telemedicine visits and reported workflow inefficiencies when clinic sessions had mixed visit types. Patients and PCPs valued remote telemonitoring as adjuncts to care; however, limited service availability and insurance coverage were barriers. Conclusion: Infrastructure barriers (broadband, insurance) and workload imbalance temper enthusiasm for using digital health solutions. Health systems should implement complementary patient and provider user-centric strategies for facilitating uptake of technology. ## INTRODUCTION The coronavirus disease 2019 (COVID-19) pandemic ushered in rapid adoption of telehealth services in the United States1 and shed light on the ability of health systems to quickly adjust health care delivery models during crises. Telehealth, defined as the use of electronic information and telecommunications technologies to support long-distance clinical care,2 has 2 forms: [1] 2-way synchronous interactive communication via audio and visual equipment (eg, telemedicine) and [2] asynchronous interactions via various technologies (eg, patient portals, email/text messaging, mobile applications [apps], sensors/tracking devices). Notably, disparities in access to and use of telehealth among medically underserved, socially disadvantaged populations were observed during the pandemic. In a 2021 national survey of households in the United States, 1 in 4 respondents reported having used telehealth services in the prior 4 weeks.3 The highest rates of telehealth usage were among individuals who had Medicaid/Medicare insurance, those with an annual income <$25,000, and those who identified as Black. Rates were lowest among individuals who were uninsured and young adults ages 18 to 24 years. Significant disparities were seen in audio vs video telehealth use. Video telehealth rates were highest among young adults, White individuals, and those with private insurance and/or incomes of at least $100,000. The lowest rates of video telehealth were among individuals who reported less than a high school education, those who were 65 years and older, and racial/ethnic minorities. Regarding asynchronous telecommunication, the 2020 Health Information National Trends Survey (HINTS) of US adults demonstrated that rates at which individuals were offered access to their online medical record via a patient portal and subsequently accessed their record increased between 2014 and 2020 by $17\%$ and $13\%$, respectively; however, no significant increases occurred between 2019 and 2020.4 Individuals whose providers encouraged them to use patient portals used their online medical record at higher rates than those who were not encouraged. Approximately 4 in 10 portal users accessed their records through a smartphone app. Individuals who accessed their portals via smartphone and computer had higher rates of portal use compared to those who accessed their portals via only 1 method. A 2021 Pew Research Center study demonstrated that while Black and Hispanic adults were less likely to own a traditional computer or have high-speed internet at home compared to White adults, there were no differences in access to smartphones and tablets.5 Rural adults, however, were less likely to have home broadband or to own a mobile device or traditional computer compared to urban and suburban adults.6 Expansion of remote physiologic monitoring (RPM) services was made easier with updates to the Centers for Medicare and Medicaid Services policy for reimbursement.7 RPM facilitates management of acute and chronic conditions by transmission of electronically collected data (eg, blood pressure, blood glucose, weight) that are automatically uploaded via an approved device to a secure location (eg, electronic medical record) where the data are available for analysis and interpretation by a clinical provider. Nonetheless, the uptake of RPM was limited. In the 2020 HINTS study, few portal users transmitted health data to a service or app.4 Kirkland et al suggest that socioeconomic status and clinic location impact the level of engagement with data transmission.8 Fritz et al further suggest that race and neighborhood disadvantage impact patients’ choice of RPM program type (telephone-based vs patient portal app).9 Notwithstanding the numerous barriers to equitable access to virtual care options, telehealth is here to stay. Therefore, health systems should find effective ways to mitigate the digital divide. We conducted this study to gain insights from primary care clinicians and their patients who are users or nonusers of telehealth services. The main objectives of this study were to [1] compare patient and primary care provider (PCP) perceptions of facilitators and barriers to engagement in using patient portals, remote telemonitoring, and/or telemedicine in medically underserved communities; and [2] recommend strategies for effectively facilitating interactive patient-provider use of these interventions among socioeconomically disadvantaged and medically underserved populations. ## Study Setting, Population, and Design This qualitative study was conducted at Ochsner Health, Louisiana's largest nonprofit, academic, multispecialty health care delivery system, between May and July 2021 (during the COVID-19 pandemic). The study investigators targeted 8 clinics located in north and southeast Louisiana that serve a large Black/African American and/or Medicaid-/Medicare-insured population. These clinics included 5 in Shreveport and Monroe and 3 in the New Orleans metropolitan area. All clinics were located in federally designated medically underserved areas as defined by the US Department of Health and Human Services. Two clinics are part of a rural federally qualified health center in North Louisiana; all the others are primary care practices owned or managed by Ochsner Health. The study recruited PCPs (physicians and nurse practitioners) who work at the targeted clinics in internal medicine, family medicine, or medicine-pediatric specialties and who had either self-referred themselves to the study or were recommended by clinic management as key stakeholders. PCPs recruited in the New Orleans area were early adopters of incorporating telehealth into their clinical practices. In contrast, PCPs in Shreveport and Monroe were mostly new users at the time of this study. Patients receiving care at the target clinics were recruited through a variety of mechanisms to obtain a convenience sample for the study: [1] MyChart recruitment notices, [2] onsite recruitment, [3] PCP referrals, and [4] referrals from study participants. This study was approved by the Ochsner Health Institutional Review Board (IRB), with IRB acknowledgement from the Louisiana State University Health Sciences Center–Shreveport. ## Digital Health Technology In 2012, Ochsner Health implemented the Epic electronic medical record (Epic Systems Corporation) that includes the MyChart patient portal. To date, approximately 925,564 patients have activated their patient portal accounts within the entire health system across Louisiana, Mississippi, Alabama, and the Gulf South. The portal can be accessed via mobile device (eg, smartphone, tablet) or computer with internet service. In 2015, Ochsner launched digital medicine programming in the New Orleans area for outpatient chronic care management of hypertension, diabetes, hyperlipidemia, and chronic obstructive pulmonary disorder, as well as antepartum and postpartum care.10 The digital medicine program includes health coaches for lifestyle counseling and clinical pharmacists for medication management under a collaborative drug therapeutic management agreement with PCPs. These programs were launched in Ochsner clinics in North Louisiana in 2020. Participation in these programs requires activation of patient portal accounts and remote monitoring equipment capable of syncing with MyChart (blood pressure cuff, glucometer, weight scale, spirometer). More recently, Ochsner launched telemedicine virtual visits with a rapid escalation of use during the COVID-19 pandemic in 2020. Virtual visits are also conducted via audiovisual connection through MyChart. The rural health center clinics use multiple platforms for telehealth. For patient portal access, the clinics use the AthenaNet portal system that is tethered to the AthenaHealth cloud-based electronic medical record system.11 For telemedicine, the clinics used StarLeaf, a messaging, meeting, and calling platform.12 Prior to the COVID-19 pandemic, the rural health center clinics piloted the Esvyda platform (ESVYDA! Inc.) for RPM of hypertension and diabetes.13 At the time of this report, a registered nurse had been hired to formally launch the RPM program. ## Participant Survey Study participants completed online surveys or structured telephone interviews to self-report demographic data prior to the conduct of interviews or focus groups. Self-reported age, sex, and race were collected from patients and PCPs. For patients, additional survey items included urban/rural status, primary insurance, use of the patient portal within the last 12 months, and use of telemedicine (video and/or audio only) during the COVID-19 pandemic. For PCPs, survey items also included type of provider (physician or nurse practitioner), level of training for physicians (resident, teaching faculty/staff), and estimated percentages for the most common insurance type and for use of the patient portal among the PCPs’ assigned patient panels. ## Semistructured Interviews and Focus Groups All interviews and focus groups were conducted by 2 health services research faculty experienced in qualitative methods. The investigators developed a moderator's guide that was used to structure the discussion. Open-ended questions were asked to assess experience, attitudes, satisfaction, and challenges with using health technology. Twenty-eight interviews were conducted via telephone. Six focus groups were conducted in person or via Zoom (4 provider groups with 27 participants; 2 patient groups with 15 participants). All interviews and focus groups were conducted in English, audio-taped, and transcribed verbatim. In addition, an investigator took notes of all sessions, and the notes were later organized by theme. The discussions explored perceptions of patient use of smartphones, mobile devices, and computers; patient use of mobile device apps and internet for health and non-health-related activities; provider/patient use of the patient portal, digital medicine programs, and/or telemedicine; and the benefits and challenges of using these technologies from their perspectives. For patients who were nonusers of a given technology, the discussions explored reasons for nonusage and perceptions of potential facilitators or barriers to use. ## Qualitative Data Analysis The team used the technology acceptance model and the unified theory of acceptance and use of technology model as frameworks for developing a thematic coding scheme.14 Key concept domains within these models include behavioral intention (defined as motivation or willingness to use technology); attitude (defined as evaluative judgment of technology use); perceived ease of use (defined as perception of minimal effort to use); perceived usefulness (defined as perception that technology will enhance experience); social influence (defined as important social contacts, such as family, believe technology should be used); and perceived behavioral control (subdomains include self-efficacy, facilitating conditions, and controllability). Facilitating conditions are factors that facilitate or impede technology use such as skills, resources, and technical support. Controllability reflects perceptions of the amount of control one has to use technology. Emerging themes that did not clearly fit into the model concepts were broadly coded as benefits/facilitators or challenges/barriers to technology use. One member of the research team (EPH) served as the primary coder, and 2 investigators served as secondary reviewers (CA, TD). The secondary reviewers used their session notes to guide their review of the primary coding scheme. Upon consensus of coded themes, perspectives of patients were compared to those of the PCPs for similarities and differences in common themes. NVivo 12 software (QSR International) was used to organize, store, analyze, and visualize data. ## Participant Characteristics A total of 40 patients and 30 providers participated in the study (Table 1). Most patients were middle-aged, female, Black/African Americans who live in urban areas, were insured by Medicaid, and had used a patient portal and/or telemedicine services. Most PCPs were younger White male physicians who reported serving mostly publicly insured populations (Medicaid/Medicare). Among all interviews ($$n = 28$$) and focus group discussions ($$n = 6$$), the frequencies of technology acceptance themes were $51.2\%$ attitude, $44.8\%$ facilitating conditions, $34.5\%$ perceived ease of use, $27.6\%$ behavioral intention, $24.1\%$ perceived usefulness, $20.7\%$ self-efficacy, $6.9\%$ controllability, and $6.9\%$ social influence. Table 1. \| Study Group/Characteristic \| Value \| \| --- \| --- \| \| Patients, n=40 \| \| \| Age, years, mean ± SD, n=37 \| 51 ± 13.2 \| \| Female \| 29 (72.5) \| \| Race \| Race \| \| White \| 12 (30.0) \| \| Black or African American \| 27 (67.5) \| \| Other \| 1 (2.5) \| \| Lives in the city, n=39 \| 30 (76.9) \| \| Receives care at a rural federally qualified health center \| 4 (10.0) \| \| Insurance, n=38 \| Insurance, n=38 \| \| Medicaid \| 22 (57.9) \| \| Medicare \| 11 (29.0) \| \| Commercial \| 4 (10.5) \| \| \| 1 (2.6) \| \| Used the patient portal within the past 12 months \| 29 (72.5) \| \| Used telemedicine services during the pandemic, n=32 \| 21 (65.6) \| \| Primary care providers, n=29 a \| Primary care providers, n=29 a \| \| Age, years, mean ± SD \| 39 ± 11.3 \| \| Female \| 14 (48.3) \| \| Race \| \| \| White \| 23 (79.3) \| \| Black or African American \| 3 (10.3) \| \| Other \| 3 (10.3) \| \| Type of primary care provider \| Type of primary care provider \| \| Physicianb \| 26 (89.7) \| \| Nurse practitioner \| 3 (10.3) \| \| Works in a rural federally qualified health center \| 4 (13.8) \| \| Predominant insurance type served by clinic \| Predominant insurance type served by clinic \| \| Medicaid \| 17 (58.6) \| \| Medicare \| 8 (27.6) \| \| Commercial \| 2 (6.9) \| \| Other \| 2 (6.9) \| \| Estimated percentage of patients using the patient portal \| Estimated percentage of patients using the patient portal \| \| <25% \| 15 (51.7) \| \| 25% to 49% \| 10 (34.5) \| \| ≥50% \| 4 (13.8) \| ## General Use of Technology Approximately $60\%$ of interview and focus group discussions provided insights into patients’ use of the internet to search for health information such as diagnosing symptoms or looking up medication side effects. The Google search engine was the most frequently cited source of information (Figure). Common challenges that the PCPs voiced included the downstream consequence of misinformation or misinterpretation of information on the internet and the need to help educate patients about which online resources are reliable. One-third of the interview/focus group discussions revealed patients’ common use of the internet for banking, shopping, paying bills, and other activities and preferential use of their smartphones to access the internet, email, and various apps. While smartphone access was very common, knowledge of how to use various features of these mobile devices varied. Additionally, where patients lived (rural vs urban) impacted access to high-speed broadband and therefore to the use of digital technology. Figure.: Word cloud of common themes for patient use of the internet for heath information. ## Patient Portal Usage Table 2 displays similarities and differences in patient and PCP perceptions of electronic medical record–tethered portals. The benefits of using a patient portal were highlighted in $55\%$ of all interviews and focus group discussions. The most notable benefits identified were the consolidation of medical record information that can be easily accessed via mobile devices; the ease of scheduling appointments and requesting medication refills; and the ability to send messages. Descriptions of portal challenges (raised in $37.9\%$ of interviews/focus groups) mostly focused on problems with logging in, remembering passwords, and navigating website/app upgrades. The advantages and disadvantages of the portal messaging tool were highlighted in almost $50\%$ of the interviews/focus groups. Notably, patients appreciated the convenience of medical advice messaging, whereas PCPs lamented the overall large number of messages irrespective of the source of messaging (phone or portal) and care team workload inefficiencies in managing these messages. Table 2. \| Thematic Domain \| Patienta \| Primary Care Provider \| \| --- \| --- \| --- \| \| Attitude \| User: When the MyChart thing came, that was just great. I love it. Nonuser: Listen, I know how to read, so I’d rather get everything by mail. \| I’ve had more elder, more elderly tell me…This whole smartphone thing isn’t for me. I still got a flip phone. But even then, sometimes, they’ll be like I have a daughter that has a computer. I’m going to get her to sign up for it. \| \| Behavioral intention \| User: I did my routine. I called the nurse and I find out just, you know, like that. So, I don’t know if they’re going to text me or call me with information. They probably would but I never ask. \| And I have really found that you can’t judge it by age. I’ve had plenty people in their 20s say they don’t want to be on the portal. I have plenty of patients in their 80s who are on the portal who love it. So, you really can’t judge by age. \| \| Perceived behavioral control (subdomains facilitating conditions and controllability) \| User: I use MyOchsner mostly on my cellphone because of convenience. (facilitating conditions) Nonuser: Being home by myself, it's really rough for me with computers and cell phones. And with the COVID, everything is online, online, online, and it frustrates me. (controllability) \| I look at their chart to say, “Hey, I see you’re active. Are you still able to get on?” Because I don’t want to send them results or messages and then they can’t ever see it. And some of them will say, “Oh, I forgot my username, forgot my password.” Not all of them have, you know, the newest phones where you can use finger touch or face recognition. So the ones that still need to remember that, yes, I find that some of them do have trouble remembering. (facilitating conditions) \| \| Perceived ease of use \| User: It saves talking on the phone or going to the doctor and bothering somebody. You can just push a few flip flops there and wait and hear “Ding dong” when it's ready for you. \| It needs to be a lot more visually simple. There should not be a drop-down menu that takes up the entire computer screen. \| \| Perceived usefulness \| User: So the fact that they had a system that could consolidate all of that…It made it easier for me to manage all my co-pays and all that kind of stuff…I mean, I get my alerts when I have an appointment, my reminders. User: The best thing about it, like I said, is keeping in touch with my physicians….my doctor or his nurse is pretty quick on answering these questions. It benefits me a lot, yes. \| Some have used a portal in another system and it was pretty worthless. So, when I try to tell them — no, I’m telling you this is a very different much more advanced thing. I think the messaging volume is a lot no matter how they contact us. So, I think there's just a big burden of messages period. But doesn’t matter whether it comes in that way or phone and then I prefer portal versus phone. \| ## Remote Monitoring of Chronic Diseases Table 3 displays similarities and differences in patient and PCP perceptions of remote monitoring of chronic diseases. Most patients in the study were not using a remote monitoring program. In contrast, most of the PCPs in the study reported having some patients enrolled in a remote monitoring program. Nonetheless, patients and PCPs appreciated the value of such programming for chronic disease management (thematic domain of perceived usefulness). For patients, self-efficacy and perceived ease of use were prominent themes. PCPs additionally conveyed the importance of patient access to mobile devices and internet/Wi-Fi services, as well as willingness or interest in having someone review their blood pressure or glucose and adjust medications between clinic visits. PCPs also reported lack of insurance coverage for monitoring devices as a major barrier. Table 3. \| Thematic Domain \| Patienta \| Primary Care Provider \| \| --- \| --- \| --- \| \| Behavioral intention \| \| If they don’t want that accountability and they’re really not ready to do anything, then they’re just going to say, “I’m not going to do that.” But there's some that are excited about it and so they get more engaged. \| \| Perceived behavioral control (subdomain facilitating conditions) \| \| Half the people I’ve put on the hypertension or digital medicine stuff, at least 25% refused because of lack of access to internet. [The readings] won’t upload to Epic until they get connected to Wi-Fi…I think that's part of it. If they are on a plan with limited amount of data or something like that…those charges still apply. That's not free. \| \| Perceived behavioral control (subdomain self-efficacy) \| Nonuser: My husband would really need something like that. Every time he goes to the doctor, his pressure is high. But once again, we’re in our 60s, and those computer things…. \| Some of them get gifted smartphones…but I would say even within that, there's not that decision. Like, I’m not going to log in to take my blood pressure, because I don’t want [to]. It's like, I don’t know how, and I don’t have people around me that value that, and so it's more of, like, I can’t do it. \| \| Perceived ease of use \| User: It beats taking it and writing it down and putting in a log. And then they’ll add to a sheet on MyChart so they can read it. \| The hypertension is way easier. It's just a blood pressure check; it uploads. Once you get through that first technology difficulty, get them the cuff, it seems like they can do it better. The touchpoints are less, even though I have a lot of disenrollment because they’re just like, “I can’t deal with the people calling me.” But the diabetes one is, like, orders of magnitude more…. \| \| Perceived usefulness \| User: I take my blood pressure, pulse, blood sugar, temperature, and O2 sat, and I put them all into the app and it directly links to her along with the MyChart. And then if I have an abnormal blood pressure, they’ll send me a text asking me to retake it after 15 minutes and I retake it and it goes back to them. And I can put notes in about why I think my blood pressure was high or, you know, I hadn’t taken my medicine yet. \| I have some that have been loyal users. I think the ones that tend to be a little bit more anxious about it and not really as comfortable with taking their blood pressure at home on their own like the touch points. Some people prefer to have someone checking on them if their blood pressure is up or if their sugar goes up or down. \| ## Telemedicine Usage Table 4 displays similarities and differences in patient and PCP perceptions of telemedicine. The benefit of telemedicine was a major theme raised in $51\%$ of the interviews and focus group discussions. Key highlights were the convenience of not having to travel to the clinic and the reduction of unnecessary in-person visits. The challenges of conducting telemedicine virtual visits were discussed in $45\%$ of the interviews/focus groups. Major concerns were related to technical glitches with audiovisual connections; determining which types of chief complaints were appropriate for virtual vs in-person visits; patient focus, engagement, and safety during visits if conducted outside of the home (eg, in their car, at work, other places); and the need for a care team workflow redesign to reduce patient wait times for virtual visits. Table 4. \| Thematic Domain \| Patienta \| Primary Care Provider \| \| --- \| --- \| --- \| \| Attitude \| User: I like the virtual because you don’t have to leave your home. You don’t have to get dressed. \| You don’t need to drive the 20 minutes to the clinic and wait for an hour in the waiting room for me to spend three minutes in the room and say, “Yeah, your toenail is infected. I’m going to send some antibiotics and do some warm soaks. Bye.” … that's like three hours out of their day for me to tell them that. \| \| Behavioral intention \| User: But if it's going over results or something, or just going over something, then the video is fine. But, like, if I’m seeing them for the first time or they’re doing, obviously, some kind of test, or I have a concern, I’d rather be face-to-face. \| I’ve got some repeat business on telemedicine, so some of them really like it. Some of them, I offer it and they say, “No, I’d rather come in person.” So we thought that when COVID was kind of dying down and we went back to in-office visits that we would continue to have a large portion of virtual visits and it really hasn’t panned out as much as we thought it would. \| \| Perceived behavioral control (subdomain facilitating conditions) \| User: I tried that once. But the video would not connect. So I ended up talking to him over the phone. \| They all start out as audio-visual but if the connection's bad, sometimes you’ll end up just calling them for the rest of the visit. \| \| Perceived usefulness \| User: I’m able to see the doctor one-on-one. It's not physically in person, but it is. I’m able to talk to him…if I talk to him on the phone, just talking, sometimes it doesn’t do what it would do if I was talking to them through virtual. You can see when they go to talk to you. It's just better that way, it seems. \| I mean, it's definitely a useful tool when it's used for the right reasons…medication follow-ups, established problems…Now, of course, if they have an MSK problem and it's something that requires an exam, I think it's just all knowing when to use it. \| \| Social influence \| \| The one thing that's good that's coming out of the pandemic though is more of my older patients know how to video chat with their family now using their phone. Family members have even sent them phones to video chat. So, they know how to use the phone, how to text, and how to video chat. That's assisted during visits for me to be able to talk to family so that they understand things as well. \| \| Other–wait time/workflow \| User: But I found that many of them were not punctual with the time that they were supposed to be on the visit. They were always, I don’t know, I found that mine were always a little bit late. \| My biggest gripe with telehealth is it's sometimes hard to communicate with the patient, like, that I’m coming, if I’m running a little behind or something. If you can’t just do virtual or in-person, that's almost impossible unless you’re really good about staying on time. \| ## DISCUSSION Study participants (patients and PCPs) valued digital health solutions. Patients reported having mobile devices to access the technology, but they face digital literacy and infrastructure barriers to equitable utilization of the technology (eg, access to high-speed broadband). Provider endorsement appears to influence patient engagement with digital health technology. PCPs voiced concerns about increasing workloads and workflow disruptions. Providers’ experience with workload imbalance could temper their enthusiasm for incorporating digital technology in their clinical practice and limit their endorsement of it for patient use. Technology is becoming increasingly important for accessing health care, self-care tools, and health information. However, a digital divide remains between those with and without access to technology. According to Pew Research Center population survey studies, while the overall rate of internet usage has increased over time, there are age, education, and income-related gaps in who is using the internet.15,16 Moreover, there are disparities in access to broadband service at home across racial minorities and individuals with lower levels of income and education. Individuals with less than a high school education, those with lower incomes, and younger adults are more likely to rely on smartphones for online access. US adults in lower income households have lower levels of technology uptake15,16; in households with an income <$30,000 vs households with an income >$100,000, technology uptake is as follows: smartphone, $76\%$ vs $97\%$; desktop or laptop computer, $59\%$ vs $92\%$; broadband, $57\%$ vs $93\%$; and tablet computer, $41\%$ vs $68\%$, respectively. Complicating matters further are sociodemographic differences in digital readiness (ie, confidence in one's ability to use technology).17 Within this larger social context, an imperative for health systems is to implement user-centric strategies for facilitating uptake of technology while mitigating the risk of worsening the digital divide. The authors recommend systematically assessing patients’ resources for accessing technology as a social determinant of health. This assessment should include documenting whether patients have a smartphone, and if so, how they use it (eg, phone calls only, surfing the internet, managing bills). Additionally, use of other devices such as computers and tablets should be recorded, as well as patients’ preferred location for accessing the internet or Wi-Fi (eg, home, work, public library, other). Every patient should be asked whether they feel unskilled and/or need help with using digital devices or technology as this information may provide insights into their level of digital literacy. Health systems should also assess patients’ hesitancy or preference for using technology to manage their care. Ideally, the workflow for capturing this information would be integrated into procedures for assessing social determinants of health. Study findings suggest that health system operational definitions and corresponding measures for level of patient engagement in using digital health interventions are needed. Doing so permits tracking patterns of use across subpopulations to identify opportunities for targeted education and outreach to patients who appear to be less engaged. Ideally, selection of subpopulations to monitor should be data-driven, based on local trends. Nonetheless, factors known to influence patient portal utilization include age, race, ethnicity, degree of comorbidity, education level, health literacy, attitudes/preferences for using technology, and patient preferences for how to access services.18-20 These predictors of utilization substantiate the need to systematically collect such information as a standard practice. Provider education is needed about the power of messaging the positive value of technology for care delivery and self-management. For example, regarding RPM, Walker et al suggest that patients may value increasing their disease-specific knowledge, triggering earlier clinical assessments and treatment, improving self-management, and enhancing shared decision-making.21 These benefits of RPM could be incorporated into discussions when referring patients for enrollment. Regarding telemedicine (video teleconferencing), Fischer et al suggest that demographic differences in use may reflect differences in willingness to use it.22 Fear of losing interpersonal contact and increased burdens associated with learning something new, increased out-of-pocket costs, and lack of trust in technology may temper perceived benefits. Therefore, acknowledgement of concerns and reassurance about what procedures are in place to address these concerns are equally important for building trust in the value of such programs. Providers are likely to avoid messaging of any kind if the perceived net result is increased workloads and work inefficiencies. In this study, providers resoundingly emphasized the importance of care team triage for managing electronic medical record system tasks which include responding to patient portal messages. Regarding telemedicine, providers recommended designing schedule templates so that clinic sessions are devoted to only 1 type of visit, in-person clinic or telemedicine only. In 2022, the American Medical Association published a Digital Health Implementation Playbook Series that provides a tactical approach to planning, executing, and evaluating the success of telehealth programs.23 Workflow redesign that captures the entire life cycle of a visit (before, during, and after) is paramount. A good workflow makes the process easier for patients, providers, and staff. Telehealth must also be inclusive: [1] identifying community resources for patients who may have challenges accessing technology, [2] incorporating medical interpreters in the telemedicine visit, [3] providing educational and technical support to patients as needed, and [4] keeping the caregiver as part of the process. ## Limitations This qualitative study has several limitations that may limit external generalizability of the results. This study was conducted during the COVID-19 pandemic which may have influenced uptake of technology use because of concerns about overall public health safety. Study participants were recruited through a variety of methods (eg, management recommendation, convenience sampling), so patient and provider perceptions captured in the interviews may reflect response bias in favor of or against technology. Nonetheless, our study findings confirm perceptions previously reported in the literature. Most patients and providers successfully recruited were from urban areas with fewer representing rural areas. In the midst of the pandemic with resource limitations, prioritizing this research study was difficult for the rural clinics. The experience of non-English-speaking patients and individuals with limited English proficiency using technology was not captured. Future studies must target these underrepresented populations (eg, rural, limited English proficiency) for further exploration. Finally, patient access to and use of RPM for chronic disease management varied geographically based on when the services became available. ## CONCLUSION Most study participants had positive views about the use of technology to manage health. Therefore, facilitating conditions such as availability of resources for using technology (proxy user, education on how to set up and/or use, internet/Wi-Fi) is critical. Some features of patient portal functionality are convenient and desirable for patients but may inadvertently increase provider workloads. Remote monitoring technology was uniformly seen as useful, but uptake may be facilitated/hindered by insurance coverage of devices, literacy (health and technology), internet/Wi-Fi access, and aspects of programming that may be engaging or disengaging. Uptake of telemedicine is largely influenced by access to high-speed broadband which affects the quality of video teleconferencing. ## References 1. Koonin LM, Hoots B, Tsang CA. Trends in the use of telehealth during the emergence of the COVID-19 pandemic – United States, January-March 2020 [published correction appears in. MMWR Morb Mortal Wkly Rep (2020.0) 69 1595-1599. DOI: 10.15585/mmwr.mm6943a3 2. What is telehealth? U.S. Department of Health and Human Services 3. Karimi M, Lee EC, Couture SJ 4. Johnson C, Richwine C, Patel V.. Individual's access and use of patient portals and smartphone health apps, 2020 5. Atske S, Perrin A. Home broadband adoption, computer ownership vary by race, ethnicity in the U.S. Pew Research Center 6. Vogels EA. Some digital divides persist between rural, urban and suburban America. Pew Research Center 7. Final policy, payment, and quality provisions changes to the Medicare physician fee schedule for calendar year 2021 8. Kirkland E, Schumann SO, Schreiner A. Patient demographics and clinic type are associated with patient engagement within a remote monitoring program. Telemed J E Health (2021.0) 27 843-850. DOI: 10.1089/tmj.2020.0535 9. Fritz BA, Ramsey B, Taylor D. Association of race and neighborhood disadvantage with patient engagement in a home-based COVID-19 remote monitoring program. J Gen Intern Med (2022.0) 37 838-846. DOI: 10.1007/s11606-021-07207-4 10. Manage high blood pressure or Type 2 diabetes on the go. Ochsner Digital Medicine 11. 11.athenaOne technical requirements. athenahealth. Published November 16, 2022. Accessed January 3, 2023. athenahealth.com/∼/media/athenaweb/files/pdf/athenahealth_tech_requirements.pdf 12. 12.StarLeaf. Wikipedia. Updated December 15, 2022. Accessed January 23, 2023. en.wikipedia.org/wiki/StarLeaf 13. 13.esvyda eHealth anytime, anywhere. ESVYDA! Inc. Accessed May 24, 2024. esvyda.com/ 14. Holden RJ, Karsh BT. The technology acceptance model: its past and its future in health care. J Biomed Inform (2010.0) 43 159-172. DOI: 10.1016/j.jbi.2009.07.002 15. Internet/broadband fact sheet: who uses the internet 16. Vogels EA. Digital divide persists even as Americans with lower incomes make gains in tech adoption 17. Horrigan JB. Digital readiness gaps 18. Irizarry T, DeVito Dabbs A, Curran CR. Patient portals and patient engagement: a state of the science review. J Med Internet Res (2015.0) 17 e148. DOI: 10.2196/jmir.4255 19. Goldzweig CL, Orshansky G, Paige NM. Electronic patient portals: evidence on health outcomes, satisfaction, efficiency, and attitudes: a systematic review. Ann Intern Med (2013.0) 159 677-687. DOI: 10.7326/0003-4819-159-10-201311190-00006 20. Riippa I, Linna M, Rönkkö I, Kröger V. Use of an electronic patient portal among the chronically ill: an observational study. J Med Internet Res (2014.0) 16 e275. DOI: 10.2196/jmir.3722 21. Walker RC, Tong A, Howard K, Palmer SC. Patient expectations and experiences of remote monitoring for chronic diseases: systematic review and thematic synthesis of qualitative studies. Int J Med Inform (2019.0) 124 78-85. DOI: 10.1016/j.ijmedinf.2019.01.013 22. Fischer SH, Ray KN, Mehrotra A, Bloom EL. Uscher-Pines L. Prevalence and characteristics of telehealth utilization in the United States. JAMA Netw Open (2020.0) 3 e2022302. DOI: 10.1001/jamanetworkopen.2020.22302 23. Digital health implementation playbook series. American Medical Association
--- title: The role of obesity in sarcopenia and the optimal body composition to prevent against sarcopenia and obesity authors: - Chaoran Liu - Keith Yu-Kin Cheng - Xin Tong - Wing-Hoi Cheung - Simon Kwoon-Ho Chow - Sheung Wai Law - Ronald Man Yeung Wong journal: Frontiers in Endocrinology year: 2023 pmcid: PMC10016224 doi: 10.3389/fendo.2023.1077255 license: CC BY 4.0 --- # The role of obesity in sarcopenia and the optimal body composition to prevent against sarcopenia and obesity ## Abstract ### Background Elderly people with low lean and high fat mass, are diagnosed with sarcopenic obesity (SO), and often have poor clinical outcomes. This study aimed to explore the relationship between obesity and sarcopenia, and the optimal proportion of fat and muscle for old individuals. ### Methods Participants aged 60 years or above were instructed to perform bioelectrical impedance analysis to obtain the muscle and fat indicators, and handgrip strength was also performed. Sarcopenia was diagnosed according to predicted appendicular skeletal muscle mass and function. Body mass index (BMI) and body fat percentage (BF%) were used to define obesity. The association of muscle and fat indicators were analyzed by Pearson’s correlation coefficient. Pearson Chi-Square test was utilized to estimate odds ratios (OR) and $95\%$ confidence intervals (CI) on the risk of sarcopenia according to obesity status. ### Results 1637 old subjects (74.8 ± 7.8 years) participated in this study. Not only fat mass, but also muscle indicators were positively correlated to BMI and body weight ($p \leq 0.05$). Absolute muscle and fat mass in different positions had positive associations ($p \leq 0.05$). Muscle mass and strength were negatively related to appendicular fat mass percentage ($p \leq 0.05$). When defined by BMI (OR = 0.69, $95\%$ CI [0.56, 0.86]; $$p \leq 0.001$$), obesity was a protective factor for sarcopenia, whilst it was a risk factor when using BF% (OR = 1.38, $95\%$ CI [1.13, 1.69]; $$p \leq 0.002$$) as the definition. The risk of sarcopenia reduced with the increase of BMI in both genders. It was increased with raised BF% in males but displayed a U-shaped curve for females. BF% 26.0–$34.6\%$ in old females and lower than $23.9\%$ in old males are recommended for sarcopenia and obesity prevention. ### Conclusion Skeletal muscle mass had strong positive relationship with absolute fat mass but negative associations with the percentage of appendicular fat mass. Obesity was a risk factor of sarcopenia when defined by BF% instead of BMI. The management of BF% can accurately help elderly people prevent against both sarcopenia and obesity. ## Introduction The aging population has been an important challenge in public health and is posing a huge socioeconomic burden [1]. A recent cohort study indicated that increased body mass index (BMI) was associated with lower all-cause and non-cardiovascular disease mortality in Chinese old people [2]. This observation supports the “obesity paradox” again. However, gaining BMI can also have undesirable metabolic risks including excess adiposity accumulation, which leads to cardiovascular diseases and diabetes mellitus [3]. Body composition analyses have also reported that excess body fat increases all-cause and disease-cause mortality, and people with low lean mass have been found to have higher death rates [4, 5]. Therefore, the management of an optimal body composition for old people is important. It is well known that BMI only considers body mass rather than body composition, which may not be appropriate for old individuals [2], and understanding the optimal body composition to balance fat and lean mass is warranted [6]. Four main phenotypes have been classified with body adiposity and muscle mass composition, which are sarcopenia, obesity, sarcopenic obesity, and healthy status [7]. Sarcopenia is an age-related muscle disorder, and is associated with increased risk of fall, fracture, and mortality [8, 9]. The Asian Working Group for Sarcopenia (AWGS) 2019 consensus recommends using lower muscle mass with poorer grip strength or physical performance to define sarcopenia [10]. On the other hand, the European Working Group on Sarcopenia in Older People 2 (EWGSOP2) revised consensus identifies sarcopenia in older adults with low grip strength and muscle mass, and those with a combination of poor physical performance are considered to have severe sarcopenia [11]. It is known that lower BMI is commonly found in people with sarcopenia [12]. Similar to BMI, body fat mass indicators including body fat percentage (BF%), are also used to diagnose obesity and estimate the risks of obesity-related diseases in older people [13, 14]. Old individuals with both low muscle mass and high adiposity are sarcopenic obese (SO) which fail to benefit from the “obesity paradox” due to their higher risk of all-cause mortality [15]. There has been evidence from pre-clinical studies indicating that adipose tissue damages muscle homeostasis, resulting in muscle atrophy and regeneration capacity reduction [16, 17]. This finding was regarded as the pathogenic mechanism of sarcopenic obesity [17]. Since sarcopenia, obesity, and sarcopenic obesity all lead to various adverse clinical outcomes of old people, it is necessary to establish the proper body indicator cut-offs for reference to decrease relevant risks. This cross-sectional study aims to explore i) the relationship between fat and muscle indicators in Asian elderly people, ii) the role of obesity in sarcopenia and muscle maintenance based on BMI- and BF%-defined obesity, and iii) the optimal BMI and BF% to prevent against both sarcopenia and obesity in old individuals. ## Study population Elderly people were screened from the community or outpatient clinics at Prince of Wales Hospital in Hong Kong from 2019 to 2021. The inclusion criteria were 1) aged 60 years old or above, and 2) Chinese ethnicity. The exclusion criteria were: 1) severe foot deformity which is unable to acquire the BIA data, and 2) unable to communicate and understand the test instructions, e.g., severe dementia. This study was approved by The Joint Chinese University of Hong Kong – New Territories East Cluster Clinical Research Ethics Committee (Ref. CREC 2018.602). ## Assessment of muscle and fat All participants height were measured by an ultrasonic sensor (Clifford H.K. Co., Hong Kong). The whole-body skeletal muscle mass (SMM), body fat mass (BFM), arms fat mass (AFM), legs fat mass (LFM), and trunk fat mass (TFM), arms fat-free mass (AFFM), legs fat-free mass (LFFM), as well as waist-hip ratio (WHR) were assessed and directly obtained from the bioelectrical impedance analysis (BIA) system (InBody 120, Seoul, Korea). The tests were performed according to the manual instructions. In brief, subjects stood on the BIA device platform barefoot, and held the electrodes until the measurement was completed. Other body composition values were calculated as follows: fat mass index (FMI) = BFM/height2, skeletal muscle mass index (SMI) = SMM/height2, BF% = BFM/body weight, leg fat mass percentage (LFM%) = LFM/leg mass, arms fat mass percentage (AFM%) = AFM/arm mass, trunk fat mass percentage (TFM%) = TFM/trunk mass, leg fat-free mass percentage (LFFM%) = LFFM/leg mass, arm fat-free mass percentage (AFFM%) = AFFM/arm mass. We previously found that the value of muscle mass index detected by BIA (InBody 120) was 2.89 ± 0.38 kg/m2 higher than measured by dual-energy X-ray absorptiometry (DXA) (Horizon, Hologic, Marlborough, MA, USA), which was considered the gold standard [18]. Therefore, we recruited another 48 volunteers and utilized our previous method to establish a model to predict the DXA-measured appendicular skeletal muscle mass index (ASMI) based on BIA-measured SMI and demographic information via test- ($$n = 32$$) and validation ($$n = 16$$) groups [18]. Multiple regression and Bland–Altman analyses were performed. SMI, age, sex, and anthropometric parameters including height, weight, and BMI were involved as potential contributions to establish the best model [18]. The final prediction model is: ASMI (DXA) = 0.378 + 0.662 * (BIA SMI) – 0.003 * (Age) – 0.032 * (BMI); R2 = 0.862. The mean difference between predicted and actual value was 0.04 ± 0.25 kg/m2 in the validation group. Handgrip strength (HGS) was measured by the dynamometer (5030JI, JAMAR, Bolingbrook, IL, USA). Participants seated with 90° elbow flexion and executed the test 3 trials per hand. The maximal reading was recorded [10]. ## Diagnosis of sarcopenia and obesity Cut-off points according to the AWGS 2019 were used. Participants with both low muscle mass and strength was defined as sarcopenia. Male with ASMI (predicted) < 7.0 kg/m2, and HGS < 28 kg, and female with ASMI (predicted) < 5.4 kg/m2, and HGS < 18 kg were sarcopenic. Two criteria were used to diagnose obesity according to the previous studies of SO [19]. The BMI ≥ 25 kg/m2 was used to define obesity as recommended by WHO for East Asians [20]; and BF% > $27\%$ in male and $35\%$ in female, which was used in previous SO studies for classification of obesity, and was close to the 60th percentile of BF% in our cohort (21–23). ## Statistical analyses Continuous variables were presented as mean ± standard error (SD), and categorical variables were expressed as number and percentage. Pearson’s correlation coefficient was used to test the correlations between variables, including age, height, weight, muscle- and fat-related indicators. One-way ANOVA with post-hoc analysis by Bonferroni test was used to analyze the differences of body parameters between normal, only sarcopenic, only obese, and sarcopenic obese groups. The Pearson Chi-square test was performed to detect the role of obesity in sarcopenia via odds ratios (OR), as well as the proper values of BMI and BF% to prevent sarcopenia according to the fifth distributions of BMI and BF%. The age-related descent rate of muscle mass and strength in people with or without obesity, as well as the ASMI prediction model were estimated by using regression coefficient (β) from linear regression analysis. Python 3.10.1 and R 4.0.2 were utilized for the analyses. p ≤ 0.05 was regarded as statistical significance in differences. ## The associations of fat and muscle indicators 1637 old subjects (age: 74.8 ± 7.8, range: 60–98 years; $83.6\%$ female) were included without missing data (Table 1). After analyzing data from the whole cohort (both genders), the Pearson’s correlations (Figure 1A) showed that age (≥ 60 years) was not related to BMI, WHR, and fat mass in different body positions ($p \leq 0.05$). Higher fat mass percentage in the whole and partial body, fat mass index, and lower weight, height, fat-free mass (FFM) in partial body, percentage of FFM, SMM, SMI, ASMI, and handgrip strength were related to increased age ($p \leq 0.05$). The percentage of fat mass in arms and legs were inversely correlated with SMM, SMI, and ASMI ($p \leq 0.05$). TFM% was positively related to SMI ($p \leq 0.001$), but not ASMI ($p \leq 0.05$). Higher TFM% was associated with reduced SMM and HGS ($p \leq 0.05$). BF% was weakly and negatively related to SMM and ASMI, but positively related to SMI ($p \leq 0.05$). Body weight, BMI, absolute fat mass, and WHR had similar trends to be positively associated with almost all muscle and fat parameters instead of fat-free mass percentage ($p \leq 0.05$). ASMI and HGS were both negatively related to the percentage of fat mass in limbs ($p \leq 0.05$). The correlation of muscle and fat indicators in females and males was shown Supplementary Figure 1. In males, SMI and WHR reduced with advanced age, which were not significant in females. SMM in both genders was negatively related to percentage of appendicular fat mass ($p \leq 0.05$), but positively associated with BF% in females. The inverse association between TFM% and HGS was only found in males rather than females. ASMI was inversely related to AFM% but not LFM% in both genders. In females, SMI increased with higher LFM%, and HGS increased with higher WHR, which were not found in males. ## The characteristics of sarcopenia, obesity, and sarcopenic obesity in Asian old people Subjects were divided into four groups based on sarcopenia and two obesity definitions (Table 1). More SO patients were detected when obesity was defined by BF% ($25\%$ in male, $17.3\%$ in female, and $18.6\%$ in total). If BMI ≥ 25 kg/m2 was used to define obesity, the prevalence of SO was $14.2\%$ in male, $11.8\%$ in female, and $12.2\%$ in total. Fat mass percentage in the trunk was similar between individuals with sarcopenia and non-sarcopenia when compared within the people with or without obesity, respectively ($p \leq 0.05$), except for males defined with obesity by BMI. WHR was similar or higher in the healthy group compared to only sarcopenic group, as well as in only obese group compared to sarcopenic obesity group. Appendicular fat mass was comparable or lower in sarcopenic groups with matched obesity status, but significantly higher when demonstrated by percentage. The highest percentage of arm and leg fat mass was found in SO ($p \leq 0.05$). Although BFM was similar between obese status-matched sarcopenic and non-sarcopenic groups, lower SMM was shown in the former groups (Figures 1B, C). With similar ASMI, BMI-defined SO had remarkedly higher BF% and lower HGS than the normal group ($p \leq 0.05$). There were no significant differences of ASMI and HGS between the two sarcopenic groups when obesity was defined by BF% ($p \leq 0.05$). ## The role of obesity in sarcopenia and muscle maintenance The ORs with $95\%$ confidence interval (CI) showed the risk of sarcopenia in elderlies with obesity (Table 2). BMI- and BF% defined obesity had opposite roles in sarcopenia. When the population without obesity was regarded as the reference group (OR = 1.00), obesity defined by BMI was a protective factor of sarcopenia in both male and female (ORs < 1.00, $p \leq 0.05$), while BF%-defined obesity was a risk factor (ORs > 1.00, $p \leq 0.05$). We also estimated the annual rate of muscle mass and strength decline based on obesity status in the elderly females (Figures 2A–D) and males (Figures 3A–D). For females, individuals with obesity had a steeper slope of ASMI (β: -0.017 vs. -0.006) and HGS (β: -0.238 vs. -0.206) decline when defined by BMI. Similar trends were also found in BF%-defined females with obesity, with the regression coefficient (β: -0.013 vs. -0.004) in ASMI, and in HGS (β: -0.253 vs. -0.189). Faster decline of ASMI in BMI-defined male with obesity was identified (β: -0.041 vs. -0.037). Other indicators in male without obesity declined more than male with obesity. Supplementary Table 1 showed the corresponding regression equations. ## Optimal BMI and BF% in the elderly to decrease risk of sarcopenia To specify the optimal BMI and BF% that should be maintained in elderlies to prevent sarcopenia, the recommended classification of BMI (<18.5, 18.5–22.9, 23–24.9, 25–29.9, ≥30) [20], as well as the fifth distributions of BF% (<19.1, 19.1–23.8, 23.9–27.4, 27.5–31.5, >31.5 in males, <26.0, 26.0–30.9, 31.0–34.6, 34.7–38.2, >38.2 in females) were used to calculate the ORs of sarcopenic prevalence according to the intervals of BMI and BF% (Supplementary Table 2). BMI 18.5–22.9, and the lowest BF% (<19.1) were chosen as reference groups. With the increase of BMI, a trend of reduced risks of sarcopenia were found in both male and female (Figure 4A). The significant effect of sarcopenia prevention was found in BMI 25–29.5 group in male ($$p \leq 0.02$$), and BMI ≥ 30 in female ($$p \leq 0.001$$). BMI <18.5 increased the risk of sarcopenia in female ($$p \leq 0.05$$). BF% and the risk of sarcopenia displayed a U-shaped curve in female, but the OR was lineally raised in male over $23.8\%$ (Figure 4B). The significant protective effects were found in BF% 26.0–$30.9\%$ and 31.0–$34.6\%$ groups compared to the lowest BF% group in female ($p \leq 0.05$). Nevertheless, the risk of sarcopenia was comparable in the first four BF% groups, but significantly higher in the fifth group with BF% > 31.5 ($p \leq 0.01$) in male. To minimize the risk of sarcopenia, females should keep their BMI over 18.5 kg/m2, as well as BF% between $26.0\%$ and $34.6\%$. In males, higher BMI and BF% less than $23.9\%$ were recommended. Figure 4: The risk of sarcopenia in males and females with different BMI and BF%. (A) showed that BMI was classified into 5 intervals based on the recommendation from WHO, the normal BMI (18.5–22.9) was regarded the reference group with OR=1.00. Blue points as OR values and blue shade as 95% CI represented male, and red represented female. (B) showed that BF% was classified into 5 intervals by quintile, the group with the lowest value of BF% was reference group. The specific interval of male (blue) was shown on the upper horizontal axis and female (red) on the lower horizontal axis. ## Discussion Muscle and fat are two widely studied tissues that contribute to a significant portion of our bodies. Without a large change of body composition, the increase of BMI is usually accompanied with both fat and muscle mass in adults. For old people, a lower BMI has become a predictor of sarcopenia [12]. Various biomarkers for sarcopenia identification may be derived from this characteristic, such as lower triglycerides in sarcopenic patients [24]. However, the gain of weight or BMI for elderly people without monitoring body composition is inadvisable, since older people have less lipid turnover and higher risks of metabolic diseases [25]. A weak but significantly positive correlation between BF% and age was found in the elderly. This finding was also applicable in people from middle to old age [26]. Although patients with sarcopenia have similar or even lower levels of absolute fat mass compared to non-sarcopenic people, their relative fat mass increased especially in limbs. Appendicular fat mass percentage was inversely related to ASMI and HGS when analyzed the whole cohort. Therefore, the fat deposition in limbs can be a potential diagnostic indicator of sarcopenia. Central obesity was associated with the development of metabolic complications and adverse clinical outcomes [27]. We found higher TFM% was related to lower HGS in males, but to higher ASMI and SMI in females. Although WHR in non-sarcopenic individuals was also similar or higher compared to the sarcopenic ones, higher WHR in females was positively related to muscle mass and strength indicators. Previous studies also showed that females with central obesity but not males had lower prevalence of sarcopenia [28]. This finding indicated there were greater adverse effects of fat accumulation and central obesity on the muscle of males. In-vitro studies showed that the coculture of mature adipocytes and skeletal muscle progenitor cells led to a reduction of nuclei number in myosin heavy chain (MHC)-positive myotubes [29]. Fat deposition in extremities may play a role of muscle loss and dysfunction in sarcopenic patients via paracrine of adipokines and cytokines. Circulation lipid metabolites may also play roles in aggravating muscle metabolism disorders, which mainly affects the energy metabolism and muscle function [30]. There is a well-known paradox that obesity is related to a lower risk of mortality [31]. However, this finding depends on the definition of obesity by using BMI. When obesity was defined by BF%, obesity became related to higher death rate [14]. Hence, the body composition may be the missing gap. According to the body composition, old individuals can be separated into sarcopenia, obesity, SO, and healthy status. Individuals with SO had lower muscle mass, strength, and higher adiposity, as well as higher all-cause mortality and worse surgical prognosis [17, 32]. In our study, SO was more prevalent in males than females, and when obesity was defined by BF% than BMI. When defined by BMI, SO could be diagnosed dominantly by muscle function test since their muscle mass was large. Although with higher BMI and absolute muscle mass than sarcopenia alone, SO patients had lower muscle quality, high risk of physical disability, as well as more metabolic issues, which may induce poor clinical outcomes [33]. If defined by BF%, ASMI became comparable between simple sarcopenic and SO patients due to the shrunken discrepancies of BMI among groups. AFM% and LFM% were significantly higher in SO and may be biomarkers of this disease. In most cases, SO patients had different demographic features when diagnosed by different obesity definitions. A recommendation of the standard diagnostic criteria of SO should be noted in the future according to the risk of adverse events and outcomes with different definitions. When obesity was defined by BMI, we found that it was a protective factor of sarcopenia despite the various metabolic problems that can occur [34]. On the contrary, pre-clinical studies reported that obesity impaired muscle glucose tolerance, imbalanced protein synthesis and degradation, and oxidative stress which ultimately led to muscle atrophy, especially in old animals [16, 17]. This may be caused by the severe obesity and exorbitant BF% in diet-induced obese animal models [35]. In our study, we observed that obesity defined by BF% was a risk factor of sarcopenia which was consistent with pre-clinical findings. The ratio of body fat was not only associated with metabolic syndromes and adverse events but also with sarcopenia [36, 37]. BF% contains the information of lean mass, fat mass, sarcopenia, and obesity, which is better than BMI that only contains body mass for elderly people. Similar to previous findings, in the elderly female group with obesity, a faster decline of muscle mass and strength with aging was observed [38]. Although they had larger muscle storage, the muscle regeneration may be impaired [17]. Nevertheless, the muscle decline in males was not as sensitive to obesity as in females. To explore the casual relationship between obesity and sarcopenia, a prospective study is needed. The management of body composition is important, and there are several strategies. Resistance training combined with nutrient supplementation, such as protein is preferable to maintain muscle mass [39]. As for elderlies with obesity, the combination of caloric restriction (low-fat, proper high-protein diet with moderately decreased energy), as well as aerobic and resistance training have been recommended [40, 41]. In order to identify the optimal BMI and BF% to prevent sarcopenia, we divided the elderly population into 5 subgroups according to BMI and BF% distribution as previous methods [4]. The differences caused by gender was apparent. For instance, lower BMI (<18.5 kg/m2) dramatically increased the sarcopenia risk in females instead of males. In addition, the lowest interval of BF% in females also harmed muscle status. Adipose tissue is an essential endocrine organ that regulates hormonal levels. The lowest BMI and BF% resulted in low estrogen levels in menopausal female [42]. It was reported that reduction of estradiol concentrations attenuated satellite cell proliferation, and the ability to maintain muscle mass and strength [43]. The excess accumulation of fat also affects muscle phenotypes metabolically [17]. Hence, we identified a range of BF% to prevent both sarcopenia and obesity in females. Males with the highest interval of BF% had three times greater risk of sarcopenia than the lowest subgroup. In a large cohort of men, increment of fat mass was associated with mortality, which may be associated with the high prevalence of sarcopenia [4]. It is necessary to control the adiposity levels in old males due to the faster increasing trend of obesity compared to females [44]. BMI was not as sensitive as BF% to simultaneously identify metabolic and sarcopenic risks. From our results, it is recommended for females to have a BMI between 18.5 kg/m2 and 25 kg/m2, and BF% between $26.0\%$ and $34.6\%$ to prevent sarcopenia and obesity. For males, the BMI should be lower than 25 kg/m2 and BF% lower than $23.9\%$. Those with high BF% warrants early attention due to the higher potential to suffer both muscle and metabolic disorders. Since muscle disorders are associated with high risk of mortality, the reservation of muscle mass and strength is important [9]. At present, numerous home-based, economical body fat percentage analysis instruments have been utilized for general body composition supervision, which old people will greatly benefit from. We also recommend that annual health examinations can consider to include BF% in elderlies, and body composition can be maintained through regular exercise and nutrition supplements. Our study has several strengths. This study exhibited the correlation between various muscle and fat indicators comprehensively. We compared the role of obesity in sarcopenia with two different obesity definitions, and found that higher body fat percentage is related to the increased risk of sarcopenia, but higher BMI is associated with the lower risk of sarcopenia. Our findings indicate that body composition should be focused on in the elderly to observe the risks of both sarcopenia and obesity. The optimal range of BMI and BF% to resist sarcopenia for elderly individuals has also been shown in this study. There are some limitations in this study. We diagnosed sarcopenia based on the AWGS 2019 consensus with only ASMI and HGS. This is due to the fact that the EWGSOP2 consensus only requires these two parameters for diagnosis, and the addition of physical performance defines severity. We wanted to avoid confusion from readers worldwide. However, as recommended by AWGS 2019 consensus, physical performance parameters such as 6-metre walk, short physical performance battery (SPPB), or 5-time chair stand test should also be evaluated in future studies. In addition, we used a prediction model to estimate ASMI, so that an error from the true value may be present. The sample size of male participants was smaller which may cause the false-negative results. The blood samples as well as comorbidity information were not collected for further analyses. This was a cross-sectional study which only showed the relative risk instead of revealing the causal relationship between obesity and sarcopenia, and thus prospective studies are warranted. Our study revealed that muscle mass and strength elevated along with BMI and absolute fat mass increment. Obesity is a protective factor of sarcopenia when defined by BMI but is a risk factor when defined by BF%. As for the fat distribution, appendicular fat mass percentage was inversely relevant to muscle mass in both genders, and trunk fat mass percentage was negatively related to muscle strength only in males. The prevalence of SO in Chinese old people was higher if obesity was defined by BF% than BMI. In females with obesity, the annual rate of muscle mass and strength decline was faster than the non-obese group, but this finding did not present in males. The lowest incidence of sarcopenia was found in females with the BF% 26.0–$34.6\%$, and BMI over 18.5 kg/m2. A trend showed that BF% less than $23.9\%$ in males was better for sarcopenia prevention. Due to the negative effects of adipose tissue on muscle in pre-clinical studies, a longitudinal obese cohort to explore the alterations of muscle and its function with advanced age is warranted to elucidate the role of fat in muscle clinically. ## Data availability statement The datasets presented in this article are not readily available unless a valid and reasonable purpose is given. Requests to access the datasets should be directed to louischeung@cuhk.edu.hk. ## Ethics statement The studies involving human participants were reviewed and approved by The Joint Chinese University of Hong Kong – New Territories East Cluster Clinical Research Ethics Committee (Ref. CREC 2018.602). The patients/participants provided their written informed consent to participate in this study. ## Author contributions CL: writing-original draft and editing; conceptualization; methodology. KY-KC: investigation; data curation. XT: statistical analysis; data visualization. W-HC: supervision; writing-review and editing. SK-HC: supervision; writing-review and editing. SL: conceptualization; validation. RW: conceptualization; investigation; supervision; writing-review and editing. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fendo.2023.1077255/full#supplementary-material ## References 1. Michel J-P, Leonardi M, Martin M, Prina M. Who's report for the decade of healthy ageing 2021–30 sets the stage for globally comparable data on healthy ageing. Lancet Healthy Longevity (2021) 2. DOI: 10.1016/s2666-7568(21)00002-7 2. Lv Y, Mao C, Gao X, Ji JS, Kraus VB, Yin Z. The obesity paradox is mostly driven by decreased noncardiovascular disease mortality in the oldest old in China: A 20-year prospective cohort study. Nat Aging (2022) 2. DOI: 10.1038/s43587-022-00201-3 3. Piche ME, Tchernof A, Despres JP. Obesity phenotypes, diabetes, and cardiovascular diseases. Circ Res (2020) 126. DOI: 10.1161/CIRCRESAHA.120.316101 4. Lee DH, Keum N, Hu FB, Orav EJ, Rimm EB, Willett WC. Predicted lean body mass, fat mass, and all cause and cause specific mortality in men: Prospective us cohort study. BMJ (2018) 362. DOI: 10.1136/bmj.k2575 5. Winter JE, MacInnis RJ, Wattanapenpaiboon N, Nowson CA. Bmi and all-cause mortality in older adults: A meta-analysis. Am J Clin Nutr (2014) 99. DOI: 10.3945/ajcn.113.068122 6. Liu C, Wong PY, Chung YL, Chow SK, Cheung WH, Law SW. Deciphering the "Obesity paradox" in the elderly: A systematic review and meta-analysis of sarcopenic obesity. Obes Rev (2022) 24. DOI: 10.1111/obr.13534 7. Prado CM, Siervo M, Mire E, Heymsfield SB, Stephan BC, Broyles S. A population-based approach to define body-composition phenotypes. Am J Clin Nutr (2014) 99. DOI: 10.3945/ajcn.113.078576 8. Yeung SSY, Reijnierse EM, Pham VK, Trappenburg MC, Lim WK, Meskers CGM. Sarcopenia and its association with falls and fractures in older adults: A systematic review and meta-analysis. J Cachexia Sarcopenia Muscle (2019) 10 485-500. DOI: 10.1002/jcsm.12411 9. Liu P, Hao Q, Hai S, Wang H, Cao L, Dong B. Sarcopenia as a predictor of all-cause mortality among community-dwelling older people: A systematic review and meta-analysis. Maturitas (2017) 103 16-22. DOI: 10.1016/j.maturitas.2017.04.007 10. Chen LK, Woo J, Assantachai P, Auyeung TW, Chou MY, Iijima K. Asian Working group for sarcopenia: 2019 consensus update on sarcopenia diagnosis and treatment. J Am Med Dir Assoc (2020) 21 300-7.e2. DOI: 10.1016/j.jamda.2019.12.012 11. Cruz-Jentoft AJ, Bahat G, Bauer J, Boirie Y, Bruyère O, Cederholm T. Sarcopenia: Revised European consensus on definition and diagnosis. Age Ageing (2019) 48 16-31. DOI: 10.1093/ageing/afy169 12. Khongsri N, Tongsuntud S, Limampai P, Kuptniratsaikul V. The prevalence of sarcopenia and related factors in a community-dwelling elders Thai population. Osteoporos Sarcopenia (2016) 2. DOI: 10.1016/j.afos.2016.05.001 13. Macek P, Biskup M, Terek-Derszniak M, Stachura M, Krol H, Gozdz S. Optimal body fat percentage cut-off values in predicting the obesity-related cardiovascular risk factors: A cross-sectional cohort study. Diabetes Metab Syndr Obes (2020) 13. DOI: 10.2147/DMSO.S248444 14. Padwal R, Leslie WD, Lix LM, Majumdar SR. Relationship among body fat percentage, body mass index, and all-cause mortality: A cohort study. Ann Intern Med (2016) 164. DOI: 10.7326/M15-1181 15. Zhang X, Xie X, Dou Q, Liu C, Zhang W, Yang Y. Association of sarcopenic obesity with the risk of all-cause mortality among adults over a broad range of different settings: A updated meta-analysis. BMC Geriatr (2019) 19 183. DOI: 10.1186/s12877-019-1195-y 16. Messa GAM, Piasecki M, Hurst J, Hill C, Tallis J, Degens H. The impact of a high-fat diet in mice is dependent on duration and age, and differs between muscles. J Exp Biol (2020) 223 jeb217117. DOI: 10.1242/jeb.217117 17. Batsis JA, Villareal DT. Sarcopenic obesity in older adults: Aetiology, epidemiology and treatment strategies. Nat Rev Endocrinol (2018) 14. DOI: 10.1038/s41574-018-0062-9 18. Cheng KY, Chow SK, Hung VW, Wong CH, Wong RM, Tsang CS. Diagnosis of sarcopenia by evaluating skeletal muscle mass by adjusted bioimpedance analysis validated with dual-energy X-ray absorptiometry. J Cachexia Sarcopenia Muscle (2021) 12. DOI: 10.1002/jcsm.12825 19. Gao Q, Mei F, Shang Y, Hu K, Chen F, Zhao L. Global prevalence of sarcopenic obesity in older adults: A systematic review and meta-analysis. Clin Nutr (2021) 40. DOI: 10.1016/j.clnu.2021.06.009 20. 20 World Health OrganizationRegional Office for the Western P. The Asia-pacific perspective: Redefining obesity and its treatment. Sydney: Health Communications Australia (2000).. The Asia-pacific perspective: Redefining obesity and its treatment (2000) 21. Perna S, Peroni G, Faliva MA, Bartolo A, Naso M, Miccono A. Sarcopenia and sarcopenic obesity in comparison: Prevalence, metabolic profile, and key differences. a cross-sectional study in Italian hospitalized elderly. Aging Clin Exp Res (2017) 29. DOI: 10.1007/s40520-016-0701-8 22. Kemmler W, von Stengel S, Engelke K, Sieber C, Freiberger E. Prevalence of sarcopenic obesity in Germany using established definitions: Baseline data of the Formosa study. Osteoporos Int (2016) 27. DOI: 10.1007/s00198-015-3303-y 23. Du Y, Wang X, Xie H, Zheng S, Wu X, Zhu X. Sex differences in the prevalence and adverse outcomes of sarcopenia and sarcopenic obesity in community dwelling elderly in East China using the awgs criteria. BMC Endocr Disord (2019) 19 109. DOI: 10.1186/s12902-019-0432-x 24. Picca A, Coelho-Junior HJ, Calvani R, Marzetti E, Vetrano DL. Biomarkers shared by frailty and sarcopenia in older adults: A systematic review and meta-analysis. Ageing Res Rev (2021) 73. DOI: 10.1016/j.arr.2021.101530 25. Arner P, Bernard S, Appelsved L, Fu KY, Andersson DP, Salehpour M. Adipose lipid turnover and long-term changes in body weight. Nat Med (2019) 25. DOI: 10.1038/s41591-019-0565-5 26. Macek P, Terek-Derszniak M, Biskup M, Krol H, Smok-Kalwat J, Gozdz S. Assessment of age-induced changes in body fat percentage and bmi aided by Bayesian modelling: A cross-sectional cohort study in middle-aged and older adults. Clin Interv Aging (2020) 15. DOI: 10.2147/CIA.S277171 27. Sniderman AD, Bhopal R, Prabhakaran D, Sarrafzadegan N, Tchernof A. Why might south asians be so susceptible to central obesity and its atherogenic consequences? the adipose tissue overflow hypothesis. Int J Epidemiol (2007) 36. DOI: 10.1093/ije/dyl245 28. Choi S, Chon J, Lee SA, Yoo MC, Yun Y, Chung SJ. Central obesity is associated with lower prevalence of sarcopenia in older women, but not in men: A cross-sectional study. BMC Geriatr (2022) 22 406. DOI: 10.1186/s12877-022-03102-7 29. Takegahara Y, Yamanouchi K, Nakamura K, Nakano S, Nishihara M. Myotube formation is affected by adipogenic lineage cells in a cell-to-Cell contact-independent manner. Exp Cell Res (2014) 324. DOI: 10.1016/j.yexcr.2014.03.021 30. Bucci L, Yani SL, Fabbri C, Bijlsma AY, Maier AB, Meskers CG. Circulating levels of adipokines and igf-1 are associated with skeletal muscle strength of young and old healthy subjects. Biogerontology (2013) 14. DOI: 10.1007/s10522-013-9428-5 31. Donini LM, Pinto A, Giusti AM, Lenzi A, Poggiogalle E. Obesity or bmi paradox? beneath the tip of the iceberg. Front Nutr (2020) 7. DOI: 10.3389/fnut.2020.00053 32. Rossi AP, Urbani S, Fantin F, Nori N, Brandimarte P, Martini A. Worsening disability and hospitalization risk in sarcopenic obese and dynapenic abdominal obese: A 5.5 years follow-up study in elderly men and women. Front Endocrinol (Lausanne) (2020) 11. DOI: 10.3389/fendo.2020.00314 33. Roh E, Choi KM. Health consequences of sarcopenic obesity: A narrative review. Front Endocrinol (Lausanne) (2020) 11. DOI: 10.3389/fendo.2020.00332 34. Lechleitner M. Obesity and the metabolic syndrome in the elderly–a mini-review. Gerontology (2008) 54. DOI: 10.1159/000161734 35. Tekus E, Miko A, Furedi N, Rostas I, Tenk J, Kiss T. Body fat of rats of different age groups and nutritional states: Assessment by micro-ct and skinfold thickness. J Appl Physiol (1985) (2018) 124. DOI: 10.1152/japplphysiol.00884.2016 36. Goossens GH. The metabolic phenotype in obesity: Fat mass, body fat distribution, and adipose tissue function. Obes Facts (2017) 10. DOI: 10.1159/000471488 37. Batsis JA, Mackenzie TA, Bartels SJ, Sahakyan KR, Somers VK, Lopez-Jimenez F. Diagnostic accuracy of body mass index to identify obesity in older adults: Nhanes 1999-2004. Int J Obes (Lond) (2016) 40. DOI: 10.1038/ijo.2015.243 38. Koster A, Ding J, Stenholm S, Caserotti P, Houston DK, Nicklas BJ. Does the amount of fat mass predict age-related loss of lean mass, muscle strength, and muscle quality in older adults. J Gerontol A Biol Sci Med Sci (2011) 66. DOI: 10.1093/gerona/glr070 39. Chen CA, Lai MC, Huang H, Wu CE. Interventions for body composition and upper and lower extremity muscle strength in older adults in rural Taiwan: A horizontal case study. Int J Environ Res Public Health (2022) 19 7869. DOI: 10.3390/ijerph19137869 40. Lopez P, Taaffe DR, Galvao DA, Newton RU, Nonemacher ER, Wendt VM. Resistance training effectiveness on body composition and body weight outcomes in individuals with overweight and obesity across the lifespan: A systematic review and meta-analysis. Obes Rev (2022) 23. DOI: 10.1111/obr.13428 41. Rankin JW. Effective diet and exercise interventions to improve body composition in obese individuals. Am J Lifestyle Med (2013) 9 48-62. DOI: 10.1177/1559827613507879 42. Oh H, Coburn SB, Matthews CE, Falk RT, LeBlanc ES, Wactawski-Wende J. Anthropometric measures and serum estrogen metabolism in postmenopausal women: The women's health initiative observational study. Breast Cancer Res (2017) 19 28. DOI: 10.1186/s13058-017-0810-0 43. Geraci A, Calvani R, Ferri E, Marzetti E, Arosio B, Cesari M. Sarcopenia and menopause: The role of estradiol. Front Endocrinol (Lausanne) (2021) 12. DOI: 10.3389/fendo.2021.682012 44. Kim KB, Shin YA. Males with obesity and overweight. J Obes Metab Syndr (2020) 29 18-25. DOI: 10.7570/jomes20008
--- title: The Gengnianchun recipe attenuates insulin resistance-induced diminished ovarian reserve through inhibiting the senescence of granulosa cells authors: - Hongna Gao - Lingyun Gao - Yanqiu Rao - Laidi Qian - Mingqing Li - Wenjun Wang journal: Frontiers in Endocrinology year: 2023 pmcid: PMC10016225 doi: 10.3389/fendo.2023.1133280 license: CC BY 4.0 --- # The Gengnianchun recipe attenuates insulin resistance-induced diminished ovarian reserve through inhibiting the senescence of granulosa cells ## Abstract ### Introduction Insulin resistance (IR) is found in patients with polycystic ovary syndrome (PCOS), but the effects and mechanisms of IR on diminished ovarian reserve (DOR) remain unclear. This study set out to investigate the effects of IR on ovarian reserve; to explore the effects of high concentrations of insulin on the function of ovarian cells in vitro; and to validate the hypothesis that the Gengnianchun recipe (GNC) helps to attenuate DOR caused by IR through reducing the senescence of granulosa cells. ### Methods Estrus cycle, follicle count, and sex hormone levels were detected to evaluate ovarian function in mice with IR caused by feeding a high-fat diet (HFD). In addition, KGN cells (human granulosa cell line) were treated with high concentrations of insulin. The staining for senescence-associatedβ-galactosidase (SA-β-gal), cell cycle, and expression levels of mRNA and gene proteins related to cell aging were detected in KGN cells treated with high concentrations of insulin. Mice treated with an HFD were fed metformin, GNC, or saline solution for 6 weeks by oral gavage. HOMA-IR, the area under the curve (AUC) of the oral glucose tolerance test (OGTT), levels of fasting blood glucose (FBG), and fasting serum insulin (FINS) were examined to confirm the IR status. Then estrus cycle, follicle count, and sex hormone levels were detected to evaluate ovarian function. Expression levels of mRNA and gene proteins related to cell aging were detected in the ovarian tissue of mice in each group. ### Results The results demonstrated that IR reduced murine ovarian reserves, and high doses of insulin caused granulosa cells to senesce. There was a considerable improvement in HFD-induced IR status in the metformin (Met) and GNC treatment groups. In addition, the expression levels of aging-associated biomarkers were much lower in GNC mice than Met mice; and both the latter groups had considerably lower levels than the HFD group. Moreover, higher follicle counts in different stages and shorter diestrus in the Met or GNC groups compared to the HFD group indicated that ovarian aging could be largely reversed. ### Discussion This work showed that: IR impaired ovarian reserve; high concentrations of insulin induced granulosa cell aging; and GNC attenuated ovarian function through inhibiting IR-induced cell aging. ## Introduction Diminished ovarian reserve (DOR), which is defined as a decrease in the number or quality of ovarian follicles, plays a major role in female infertility and increased miscarriage rates [1, 2]. Unfortunately, there is a significant upward trend in DOR prevalence. The prevalence of DOR in infertile patients attending in vitro fertilization centers in the United States increased from $19\%$ to $26\%$, while diagnoses among patients <40 y have increased by $42\%$ from 2004 to 2011 [3]. DOR is characterized by a lower number of antral follicles, an elevated level of serum follicle-stimulating hormone (FSH), and decreased levels of serum estradiol (E2) and anti-Müllerian hormone (AMH). DOR can be caused by genetic factors, environmental pollutants, infections, and chronic stress. However, in our clinical practice, we find that DOR insidiously increases with the development of PCOS and insulin resistance (IR). IR and hyperinsulinemia are closely associated and occur concurrently [4], which supports findings that compensatory hyperinsulinemia can be an adverse side effect of insulin resistance. IR and hyperinsulinemia are associated with polycystic ovary syndrome (PCOS) [5]. Earlier studies also support positive associations between DOR and HOMA-IR in patients and mice [6, 7]. However, the adverse effects of IR on DOR and the detailed pathogenic mechanisms of IR remain elusive. Cellular senescence, one of the hallmarks of aging, leads to age-related disease and dysfunction [8]. Senescent cells, triggered by a variety of stressors including telomere dysfunction and genotoxic and oxidative stress, are characterized by a state of irreversible cell-cycle arrest, secretion of senescence-associated secretory phenotypes (SASPs), and increased senescence‐associated β‐galactosidase (SA-β-Gal) activity [9]. Moreover, hyperinsulinemia leads to cell-cycle-induced senescence, which has been demonstrated on neurons in an Alzheimer’s disease mouse model [10]. In in vitro studies and in humans, chronic hyperinsulinemia results in cell cycle exit and a premature senescence of adipocytes, and this trajectory has been reversed by the administration of metformin [11]. Several human and mouse studies emphasize the presence of senescence in human hepatocyte cells and the involvement of senescent cells in the development and progression of non-alcoholic fatty liver disease [12, 13]. The presence of cell cycle arrest in granulosa cells has been confirmed in patients with premature ovarian insufficiency in vitro and in vivo, although cell senescence in that case was caused by reactive oxygen species [14]. Gengnianchun (GNC), a traditional Chinese medicine (TCM) formula, is composed of Radix rehmanniae, Rhizoma coptidis, Radix paeoniae alba, Rhizoma anemarrhenae, Cistanche salsa, Radix morindae officinalis, Poria, Epimedium brevicornum, Cortex phellodendri amurensis, Fructus lycii, Semen cuscutae, and Carapax et plastrum testudinis [15]. According to TCM theory, GNC has a kidney/liver tonifying effect that is used to alleviate declining functions related to aging. To date, GNC has been shown to have beneficial effects on aging-related conditions such as menopause and Alzheimer’s disease [16, 17]. GNC has also been shown to improve learning and memory, delay skin aging, and enhance resistance to oxidative stress [18, 19]. Considering these findings, GNC might be able to delay the process of aging. The GNC dose selection for treatment in this study was based on our previous study, which indicated that GNC can significantly preserved the ovarian reserve of rats [20]. GNC was also shown to modulate the hypothalamus-pituitary-ovary axis, and increase estradiol receptor (ER) levels in the pituitary gland and ovaries [21], but the specific mechanism remains to be elucidated. In sum, IR reduces ovarian reserve by inducing granulosa cell senescence and GNC seems to be able to protect ovarian function. To test the hypothesis, an IR mouse model was established and the ovarian reserve was evaluated. The underlying mechanism was verified by detecting senescence-associated changes in KGN cells after treatment with high concentrations of insulin. Ovarian reserve in the IR mouse model was assessed after administering GNC by oral gavage. ## Drugs and reagents The following drugs and reagents were used in the study: DMEM/F12 media without phenol red (Gibco), fetal bovine serum (FBS, Sciencell), RNA isolation kit, Color SYBR Green qPCR Master Mix and 4× reverse transcription Mix (EZBioscience, A0012), enzyme linked immunosorbent assay (ELISA) kits (E2, Labor Diagnostika Nord, FR E-2000; FSH, Immunoway, KE1425; LH, Immunoway, KE1475; AMH, Jingmei, JM11692M1; insulin, Alpco, 80-INSMSU-E01, E10), Normal rabbit IgG (CST, 2729), Normal mouse IgG (CST, 7076), p53 antibody (Abcam ab131442), p16 antibody (Cell Signaling Technology Cat, sc1661), p21 antibody (Cell Signaling Technology, sc-6246), AGER antibody (Abcam, ab3611), Senescenceβ-Galactosidase Staining Kit (Beyotime, C0602), Cell Cycle Staining Kit (Multi Sciences, CCS021), insulin (Absin, abs42019847), metformin (Topscience, T8526), high-fat diet (Jiangsu Xietong Pharmaceutical Bio-engineering Co., Ltd., D12451). ## Animals Four-week-old C57BL/6 female mice were purchased from Shanghai Slac Laboratory Animal Ltd. The mice were housed in the SPF facility at a constant temperature (25°C) and humidity ($55\%$) with a 12-h light/dark cycle. After adaptive feeding for 1 week with water and normal feed, all mice were randomly divided into four groups: control group (Ctrl), model group (HFD), metformin group (Met), and GNC group (GNC). Mice in groups HFD, Met, and GNC were fed the high fat diet for 9 weeks according to previous reports and changes in estrus cycle were monitored [22]. After 3 weeks, mice in Ctrl and HFD were given normal saline by gavage for 6 weeks, while mice in Met and GNC were given metformin (200 mg/kg·d body weight) and GNC decoction (2.77 g/kg body weight), respectively by gavage for 6 weeks. Metformin was dissolved in sterile water. The mouse dose of GNC was converted from the rat dose according to our previous study and the GNC decoction was dissolved into hot water [20]. Oral gavage volumes were adjusted according to body weight, which was measured every 7 d [23], while mice in Ctrl were administered the same volume of solvent solution. Experiments were conducted in accordance with Medical Laboratory Criteria. Animal studies were reviewed and approved by the Animal Experimental Ethical Committee of Fudan University. ## Gengnianchun formula The GNC formula, containing 12 crude herbs, is a conventional medicine for clinical therapy. The GNC formula is composed of Radix rehmanniae, Rhizoma coptidis, Radix paeoniae alba, Rhizoma anemarrhenae, Cistanche salsa, Radix morindae officinalis, Poria, Epimedium brevicornum, Cortex phellodendri amurensis, Fructus lycii, Semen cuscutae, and Carapax et plastrum testudinis, which were purchased from Tianjiang Pharmaceutical (Jiangyin, China). ## Oral glucose tolerance test and HOMA-IR The oral glucose tolerance test (OGTT) was performed at 8–10 weeks of age. Overnight (12–14 h) fasted mice with free access to water were orally dosed with $20\%$ dextrose anhydrous. Following glucose solution administration, blood samples were collected from the tail vein and glucose levels were measured at 0, 30, 60, 90, and 120 min. IR was estimated using the homeostasis model assessment for IR (HOMA-IR: fasting serum insulin (FINS) [mU/ml] × fasting blood glucose(FBG) [mmol/l]/22.5). ## Estrous cycle examination Estrus cycles were assessed daily with a vaginal smear using normal saline. Approximately 10 ul of saline solution was drawn into the pipette, which was gently inserted into the vaginal canal. A light microscope was used to observe the vaginal fluid after it was stained with hematoxylin and eosin (H&E). The estrous cycle was classified into four stages: proestrus, estrus, metestrus, and diestrus based on the type of major cells in the vaginal smear. Proestrus is confirmed by predominantly nucleated epithelial cells; estrus is characterized by anucleated keratinized epithelial cells; metestrus shows a combination of anucleated keratinized epithelial cells and neutrophils; and diestrus is characterized by a higher number of neutrophils. ## Ovary serial sectioning and follicle counting Ovarian histological analyses and follicle counts were performed on paraffin embedded sections stained with H&E taken from the largest cross-section of each ovary. Each ovary was serially sectioned using a microtome at 5 um. To quantify the total number of follicles in each ovary, the average of five different sections of each ovary was counted throughout the entire ovary. According to the modified Oktay system, follicles were classified into four stages [24]. Counting was repeated three times by different researchers, with each replicate containing 14 mice. After summing the number of follicles obtained by different researchers, an average was then obtained. ## Determination of E2, FSH, LH, AMH, and insulin concentrations Mice were sacrificed after a 12–14-h fast at 13 weeks of age. After clotting at room temperature (RT) for 1 h, blood samples were centrifuged at 2000 g for 15 min. Serum samples were stored at 80°C until analysis was performed. Enzyme-linked immunosorbent assay (ELISA) were used to measure serum hormones following the manufacturer’s instructions. The levels of LH, FSH, AMH, estradiol (E2), and insulin were evaluated using commercial ELISA kits. There were 14 samples in each group and each experiment was performed in triplicate. Diluted samples and serially diluted standard reagents were prepared according to the manufacturer’s instructions. In 96-well plates, standard reagents and samples were added, and then enzyme conjugate was added to each well. The plate was incubated at RT and washed three times with wash buffer. Substrate solution was placed in each well of the plate and the reaction was incubated in the dark. Stop solution was used to stop the reaction. The absorbance of each well was read at 450 nm (Biotek Multisken MK3). Standard curves were used to calculate concentrations. ## Immunohistochemistry The tissue array sections were deparaffinized and rehydrated in a series of ethanol gradient, and the endogenous peroxidase activity was quenched by immersing in methanol containing $0.3\%$ hydrogen peroxide. After heating for 30 min at 100°C in saline sodium citrate for antigen retrieval, the sections were incubated overnight at 4°C with primary antibody after blocking the slices using goat serum. The tissue sections were washed three times. They were then incubated with secondary antibodies against IgG at a 1:100 dilution and then stained with DAPI. A microscope (Olympus BX53; Olympus, Tokyo, Japan) and digital camera (Olympus DP73; Olympus) were used for image collection. To estimate the density of each marker from each mouse, five slides were used, and five random images of each slide were taken to calculate the mean density value with Image Pro-Plus 6.0. Quantification of immunostaining was based on both the percentage of positive cells and the intensity of staining [25]. Independently, two senior pathologists blinded to the samples evaluated IHC staining results. ## KGN cell culture and treatment KGN cells (human granulosa-like tumor cell line) were purchased from Guangzhou Saiku Biotechnology Co., Ltd and identified using the STR method. DMEM/F12 media without phenol red containing $10\%$ FBS was used to culture KGN cells. Cells were digested and plated on six-well plates for insulin treatments. When the cells had adhered and reached a suitable density, insulin was added after replacing the culture media with DMEM/F12 and starving the cells for 12 h. The concentration of insulin in this study was chosen according to previous studies [26, 27]. ## Cell cycle analysis using flow cytometry For cell cycle distribution, cells were cultured with 0.5 or 1 ug/ml insulin for 72 h. The cells were collected, washed with cold PBS once, and stained with 500 μL DNA Staining Solution and 5 μL Permeabilization Solution in the dark at 37°C for 30 min. For each experiment, 2 × 105 cells were recorded. A flow cytometer (Beckman Coulter) was used to analyze cell cycles. ## Senescence-associated b-galactosidase staining b-Galactosidase staining was performed with a senescence-associated b-Galactosidase Staining Kit (Beyotime, China). PBS was used to wash cells three times and stationary liquid was used to fix them for 15 min at RT. Next, the cells were incubated overnight at 37°C without carbon dioxide in darkness with the working solution containing $5\%$ 5-bromo-4-chloro-3-indolyl-b-d-galactopyrano-side (X-gal). Cells were photographed using a light microscope. ## Cellular RNA extraction and real-time PCR Cells were washed in PBS after treatment with insulin for 72 h. NanoDrop (Thermo Scientific) was used to measure the concentration of RNA extracted with an RNA isolation kit. RNA was reverse transcribed into cDNA under the following conditions: one cycle at 42 °C for 15 min and one cycle at 95 °C for 30 s. RT-PCR reactions were performed under the following conditions: one cycle at 50 °C for 2 min, one cycle at 95 °C for 5 min, followed by 40 cycles at 95 °C for 10 s, and 60 °C for 30 s. Samples were normalized to beta-2 microglobulin (B2M), and the comparative CT (threshold cycle) method used to calculate gene expression levels. The primer sequences are listed in Table 1. Table 1 \| Species \| Gene \| Direction \| Sequence \| \| --- \| --- \| --- \| --- \| \| Human \| B2m \| Forward \| GAGGCTATCCAGCGTACTCCA \| \| \| \| Reverse \| CGGCAGGCATACTCATCTTTT \| \| Human \| P16 \| Forward \| GGGTTTTCGTGGTTCACATCC \| \| \| \| Reverse \| CTAGACGCTGGCTCCTCAGTA \| \| Human \| P21 \| Forward \| CGATGGAACTTCGACTTTGTCA \| \| \| \| Reverse \| GCACAAGGGTACAAGACAGTG \| \| Human \| P53 \| Forward \| ACAGCTTTGAGGTGCGTGTTT \| \| \| \| Reverse \| CCCTTTCTTGCGGAGATTCTCT \| \| Human \| AGER \| Forward \| TTTGAGTCCATCACTAACGTCA \| \| \| \| Reverse \| GGTAGATGGCATCAATGAATCG \| \| Human \| INHA \| Forward \| GACTTTGCCACTGAGTTGATTT \| \| \| \| Reverse \| CGATCAGCATTTCCAATATGCA \| \| Human \| IL6 \| Forward \| ACTCACCTCTTCAGAACGAATTG \| \| \| \| Reverse \| CCATCTTTGGAAGGTTCAGGTTG \| \| Human \| IL8 \| Forward \| ACTGAGAGTGATTGAGAGTGGAC \| \| \| \| Reverse \| AACCCTCTGCACCCAGTTTTC \| \| Human \| TNF \| Forward \| GAGGCCAAGCCCTGGTATG \| \| \| \| Reverse \| CGGGCCGATTGATCTCAGC \| \| Human \| GM-CSF \| Forward \| AGGAGGGAGATCCGGTGTC \| \| \| \| Reverse \| TTGCGAGACGTTAATCCTGAC \| \| Mouse \| B2m \| Forward \| TTCTGGTGCTTGTCTCACTGA \| \| \| \| Reverse \| CAGTATGTTCGGCTTCCCATTC \| \| Mouse \| P21 \| Forward \| CCATGAGCGCATCGCAATC \| \| \| \| Reverse \| CCATGAGCGCATCGCAATC \| \| Mouse \| P53 \| Forward \| CTCTCCCCCGCAAAAGAAAAA \| \| \| \| Reverse \| CGGAACATCTCGAAGCGTTTA \| \| Mouse \| P16 \| Forward \| CGCAGGTTCTTGGTCACTGT \| \| \| \| Reverse \| TGTTCACGAAAGCCAGAGCG \| \| Mouse \| AGER \| Forward \| CTTGCTCTATGGGGAGCTGTA \| \| \| \| Reverse \| GGAGGATTTGAGCCACGCT \| ## Western blot analysis In six-well plates, cells were washed twice in PBS, lysed on ice for 10 min with 120 ul RIPA buffer containing PMSF and protease inhibitor cocktail and then centrifuged at 4 °C and 12 000 g for 20 min. The protein concentration was determined by a BCA kit (Beyotime, P0012). A mixture of the protein solution and SDS-PAGE loading buffer was boiled at 97°C for 10 min. Protein samples (20 µg) were transferred to polyvinylidene fluoride membranes after being loaded onto the gel. After blocking with skim milk at RT for 1 h, the membranes were incubated overnight with primary antibody solutions at 4°C. The membranes were incubated with secondary antibody solutions at RT for 1 h followed by three washes with TBST. Then the membranes were washed three times with TBST for 10 min each time. An ImageQuant LAS 4000mini system was used to detect the bands with an enhanced chemiluminescent substrate kit. DAPDH was used as a housekeeping gene, relative quantitative protein expression was assessed using ImageJ. ## RNA sequencing and DEGs analysis RNA-seq was performed to compare the global gene expression between control KGN ($$n = 3$$), 0.5 ug/ml insulin-treated KGN ($$n = 3$$), and 1 ug/ml insulin-treated KGN ($$n = 3$$). Principal component analysis (PCA) was then conducted. Volcano plots of differentially expressed mRNAs (DEMs) were generated. The q-value <0.05 and \|fold change (FC)\| >2 were set as the standard for selecting differently expressed genes (DEGs). DEMs with log2 (fold change) >0.58 were labeled in red ($P \leq 0.05$); DEMs with log2 (fold change) <-0.58 were marked in green ($P \leq 0.05$). GO and KEGG analysis of DEGs were performed. ## Statistical analysis Concentrations are expressed as mean ± SD. SPSS 22.0 software was used for the statistical analyses. Each experiment included at least three independent samples and was repeated at least three times. The difference between two groups with equal variance was compared using t-tests. One-way ANOVA was used to compare differences between more than two groups. $P \leq 0.05$ was taken to indicate a significant difference. Symbols for statistical significance levels: : $P \leq 0.01$; : $P \leq 0.001$; *: $P \leq 0.0001.$ ## The insulin-resistant mouse model was successfully established To verify the status of IR, we measured the serum insulin and fasting blood glucose levels of each group of mice. As Figure 1A shows, the body weight of the HFD-fed mice (HFD) was much higher than for normal diet fed mice (Ctrl). Basal glucose and insulin concentrations were significantly increased in HFD compared with Ctrl following 12–14 h of fasting (Figures 1B, C). We also found that HFD had a higher level of significance ($$P \leq 0.0001$$) in HOMA-IR (Figure 1D) compared with Ctrl. Figure 1:** The insulin-resistant mouse model was successfully established. Control group: C57BL/6 mice were fed with a standard diet for 9 weeks; HDF group: C57BL/6 mice were fed with a high fat diet for 9 weeks, respectively. To validate that the insulin-resistant mouse model was successfully established, weight, fasting plasma glucose, and fasting insulin levels in the mice were assessed. (A) Weight (B) Fasting plasma glucose after 12 h of fasting was determined at the same time point each week, for 8 weeks. n=14 in each group; t-test. (C) Fasting insulin concentration was determined by ELISA after 12 h of fasting, n=14 with one repeat in each assay; t-test. (D) The HOMA-IR insulin resistance index was calculated, n=14 with one repeat in each assay; t-test. **:P<0.0001 compared to group “Ctrl”. ## Insulin resistance decreased the ovarian reserve We first assessed the estrus cycle using vaginal smears. As shown in Figure 2A, the estrus cycle was classified into four stages proestrus, estrus, metestrus, and diestrus. Compared with Ctrl, the percent of diestrus in HFD was much higher (Figure 2B). Figure 2C shows that the levels of FSH and FSH/LH ratio in HFD were significantly higher than those in Ctrl, while the E2 and AMH concentrations were lower in HFD. Furthermore, we tested the ovarian reserve of those mice by measuring the number of ovarian follicles. Figure 2D shows the different stages of follicle growth, including primordial follicle (single oocytes or multi-oocytes surrounded by a thin layer of flattened granulosa cells), primary follicle (oocyte and a layer cubic granulosa cells), secondary follicles (the formation of more than two layers of granulosa cells in the follicles), antral follicles (a fluid‐filled cavity is formed inside each follicle), and corpora lutea (expulsion of a mature oocyte). As shown in Figure 2E, the numbers of primordial follicles ($P \leq 0.001$), primary follicles ($P \leq 0.01$) and secondary follicles ($P \leq 0.05$) in HFD were significantly lower than those of the control group. Figure 2: HFD induced-insulin resistance decreased the ovarian reserve. (A) Different stages of the estrous cycle were assessed by collecting vaginal smears at the same time point every day. (B) Estrus cycles were assessed every day for the last 5 weeks, n=14 with one repeat in each assay. (C) Serum FSH, LH, E2, and AMH were determined by ELISA; FSH/LH was calculated, n=14 with one repeat in each assay; t-test. (D) Different stages of follicles: primordial follicles; primary follicles, secondary follicles, antral follicles, and corpora lutea. (E) Detailed counting of follicles at different stages, n=14 with one repeat in each assay; t-test. : P<0.05, : P<0.01, : P<0.001, **: P<0.0001 compared to group “Ctrl”. ## Long-time exposure of KGN cells to high dose insulin induced granulosa cell senescence As shown in Figures 3A, B, after a 72-h manipulation of 0.5 µg/ml and 1 ug/ml insulin, the percentage of G0/G1 phase cells increased significantly in the 1 ug/ml insulin-treated group compared to the control group. The mRNA expression of p21, p16, and p53 increased significantly in the 0.5– 1 ug/ml insulin-treated groups and H2O2-treated group compared to the control group (Figure 3C). Moreover, cytokines of SASP such as IL6, IL8, TNF-a, and GM-CSF mRNA expression were significantly increased in the 0.5 ug/ml and 1 ug/ml insulin groups (Figure 3D). As shown in Figures 3E, F, the protein expression of p21, p16, p53, IL1, IL6, and TNF-a significantly increased in the high dose insulin groups (0.6–1ug/ml) compared to the control group after 72 h. To further confirm our hypothesis, KGN cells treated for 72 d with 0.5 ug/ml and 1 ug/ml insulin were then fixed and subjected to senescence associated β-galactosidase (SA-β-Gal) staining. Strikingly, more than $90\%$ of the KGN cells treated with high dose insulin were positive for SA-β-Gal staining (Figure 3G). Figure 3: Long time exposure to high dose insulin induced granulosa cell senescence. KGN cells were treated with solvent, 0.5, and 1 µM insulin, respectively. (A, B) Cell cycles were detected by flow cytometry after treatment with 0.5 uM or 1 uM insulin for 72 h. (C, D) RNA was extracted from KGN cells. Relative mRNA expression of p21, p53, p16, IL6, IL8, TNF-a, and GM-CSF were examined by qPCR. n=3; t-test. (E, F) RNA was extracted from the cells. p21, p53, p16, IL1, IL6, and TNF-a protein expression were examined by western blot analysis. n=3; t-test. n=3 with one repeat in each assay; one-way ANOVA. (G) KGN cells with strong SA-β-Gal expression were detected induced by the high dose insulin. : P<0.05, : P<0.01, : P<0.001, **: P<0.0001 compared to group “0”. ## Higher levels of biomarkers of aging in HFD-fed mice Female C57B6/L mice were supplemented with a normal diet (Ctrl) or high-fat-diet (HFD) for 9 weeks. Relative mRNA and protein expression of p53, p21, and p16 were detected by RT-PCR and IHC. As shown in Figures 4A, B, the level of p53, p21, and p16 mRNA and protein expression in HFD were significantly increased compared to that of Ctrl. Figure 4: Levels of senescence-associated biomarkers in HFD-fed mice were greatly increased. (A) RNA was extracted from the ovaries of mice in Ctrl and HFD groups, n=3 at each time point and each group; unpaired t-test. (B) IHC of p16, p53, and p21 in Ctrl (n = 5) and HFD (n = 5) mouse ovaries. The error bars indicate the mean values ± SDs, unpaired t-test. : P<0.05, **: P<0.001 compared to “Ctrl”. ## RNA-seq expression analysis for the high dose insulin-treated KGN cells RNA-seq gene expression analysis in the KGN cells treated with PBS (control group) and 1 ug/ml insulin (insulin group) was assessed. According to the results in Figure 3, 1 ug/ml had a more pronounced effect in causing KGN cell senescence; therefore, we selected 1 ug/ml insulin rather than 0.5 ug/ml insulin to treat KGN cells for RNA-seq expression analysis. The raw data of RNA-seq gene expression analysis can be found in https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE223248. In comparison with the control group, 22 DEGs were up-regulated and 26 DEGs were down-regulated in the insulin group (Figure 5A). A heat map (Figure 5B) and volcano plot (Figure 5C) were plotted using the fold change and corrected P-values. As shown in Figure 5D, KEGG pathway analysis was then performed on the DEGs, and a total of 47 pathways were found to be involved; they were mainly enriched for cell growth and death, signaling molecules and interaction, and endocrine system. ( Figure 5D). Top 30 gene ontology (GO) terms demonstrated that, after a 72-h 1 ug/ml insulin treatment the DEGs were mainly enriched to the ERK1 and ERK2 cascade pathway, integral component of plasma membrane, and signaling receptor binding. ( Figure 5E). Figure 5: Differentially expressed genes (DEGs) in KGN cells treated with high dose insulin. (A) Column graph of DEGs in different treatment groups showing both up-regulated and down-regulated genes. (B) Heatmap. Green indicates repressed mRNA levels and red indicates elevated levels after insulin treatment. (log2 fold ratio ≥ 0.58; P < 0.01). (C) The volcano plot of DEGs present, in which green dots indicate down-regulated genes and red dots indicate up-regulated genes in response to insulin treatment. (D) KEGG enrichment analysis of differentially expressed genes. (E) GO terms of the top 30 enriched genes. The GO enrichment analysis grouped these differently expressed genes into functional groups. The green column represents biological processes, the blue column represents cellular components, and the red column represents molecular components. GO, Gene Ontology. ## RT-PCR was performed to verify the RNA-seq results To validate the RNA-Seq results, six genes were chosen for qRT-PCR analysis. Glycosylation end-product-specific receptor (AGER), Enoyl acyl carrier protein reductase (INHA), Kruppel-like factor 15 (KLF15), Telomerase Reverse Transcriptase (TERT), and Ankyrin repeat and kinase domain containing 1 (ANKK1) were up-regulated in KGN cells treated with 0.5 ug/ml and 1 ug/ml insulin. Additionally, protein tyrosine phosphatase non-receptor 22 (PTPN22) and tumor necrosis factor ligand superfamily member 15 (TNFSF15) were significantly down-regulated after insulin treatment. The qRT-PCR results were in accordance with the RNA-Seq results, indicating the data reliability of RNA-seq (Figure 6). Figure 6: qRT-PCR verification of RNA-Seq analysis of gene expression. : (A–G) The mRNA expression of AGER, INHA, KLF15, PTPN22, TERT, TNFSF15, and ANKK1 were detected by : P<0.05, : P<0.01, **: P<0.0001 compared to group “0”. ## Gengnianchun attenuated insulin resistance and mitigated damage to the ovarian reserve Female C57B6/L mice were fed with a saline solution (Ctrl), HFD (HFD), 200 mg/kg·d Metformin (Met), or Gengnianchun (GNC). As Figure 7A shows, the body weight of HFD was much higher than for Ctrl, Met, or GNC. Basal glucose, insulin concentrations, and HOMA-IR were significantly increased in HFD compared with the other groups following 12–14 h of fasting (Figures 7B–D). We also found that an HFD resulted in a higher level of significance ($$P \leq 0.0001$$) in area under the curve (AUC) of OGTT compared with Ctrl. Moreover, Met and GNC were able to reverse this adverse change (Figures 7E, F). As shown in Figure 7G, compared with Ctrl, the percent of diestrus mice in HFD was much higher, while the increase was reversed in Met and GNC. Moreover, the numbers of primordial follicles ($P \leq 0.001$), primary follicles ($P \leq 0.01$), and secondary follicles ($P \leq 0.05$) in HFD were significantly lower than those of Ctrl, which were also reversed in Met and GNC (Figure 7H). Figure 7I shows that the levels of FSH and the ratio of FSH/LH in HFD were significantly higher than those mice in Ctrl,Met, and GNC groups, while the concentrations of AMH and E2 in HFD were significantly lower than those mice in Ctrl,Met, and GNC groups These findings all provide evidence for the therapeutic efficacy of the Gengnianchun formula on ovarian function. Figure 7: Gengnianchun reversed both HFD induced-insulin resistance and decreased ovarian function. (A) Weight of mice in the control group (Ctrl), HFD-fed group (HFD), Metformin group (Met), and Gengnianchun-fed group (GNC). (B–F) Fasting blood glucose, HOMA-IR, OGTT, and area under the curve (AUC) of mice in each group were assessed after a 12–14-h fast. (E–G) Different stages of the estrus cycle were assessed by collecting vaginal smears at same time point every day. Estrus cycle was assessed every day for the last 3 weeks, n=14 with one repeat in each assay. (H) Detailed counting of follicles at different stages, n=14 with one repeat in each assay; t-test. (I) Serum FSH, LH, E2, AMH, and fasting insulin were determined by ELISA with a 12–14-h fast before sacrifice; FSH/LH was calculated, n=14 with one repeat in each assay; : P<0.05, : P<0.01, : P<0.001, **: P<0.0001 compared to group “Ctrl”. ## Gengnianchun reduced aging-related mRNA and protein levels As shown in Figure 8A, gene expressions of P53, P16, and P21 were significantly upregulated in HDF compared to Ctrl, and the use of metformin and GNC formula reversed the up-regulation of these proteins. The levels of protein expression of P53, P16, and P21 were tested using western blotting (WB) and IHC (Figures 8B–F) and the results showed that the relative levels of proteins in HFD were significantly higher than the Met and GNC groups, thus confirming the efficacy of metformin and GNC formula in vivo. Figure 8: Gengnianchun reduced aging-related mRNA and protein levels. (A) The relative mRNA expression of P53, P16, and P21 in ovaries of mice in each group were detected by qPCR. n=3; t-test. (B–D) The protein expression of P53, P16, and P21 in the ovaries of mice in each group. n=3; t-test. (E, F) The protein expression of p16, p53, and p21 in Ctrl (n = 5) and HFD (n = 5) mouse ovaries were determined by IHC. n=5; : P<0.05, : P<0.01, : P<0.001, **: P<0.0001. ## Discussion DOR has a significant impact on female reproductive health and pregnancy rates. It is common for genetic factors, environmental pollution, and infections to contribute to DOR in the modern world. It is well known that insulin resistance and hyperinsulinemia always exist simultaneously [28], and these are associated with PCOS. However, during routine clinical practice, we found that women with IR were also more likely to suffer from DOR, and our experience showed that the GNC formula could be effective in attenuating diminished ovarian function. In this study, we first investigated the effects of IR on ovarian reserve in mice. There is now widespread agreement that HFD feeding results in IR in C57BL/6 mice [29, 30]. Consistent with these previous findings, we established an IR mouse model by exposing mice to an HFD for 9 weeks, which resulted in increased weight, increased levels of fasting blood glucose and fasting insulin, and increased the HOMA-IR index in HFD mice compared with Ctrl mice. Furthermore, we found that HFD had lower counts in the numbers of primordial follicles, primary follicles and secondary follicles, longer diestrus, and higher levels of sex hormones compared with Ctrl, which supported the IR-diminished ovarian reserve of mice. To further explore the specific mechanism of IR on the ovarian reserve in vivo, we treated the cells with high concentrations of insulin. We found that KGN cells underwent cell senescenceafter treatment with high concentrations of insulin. SASPs, which are a prominent source of chronic inflammation in the aging microenvironment [31, 32], are associated with the increased activation of nuclear factor kappa-B (NF-kB, also known as NF-kappaB) pathway [33]. Based on the results of RNA-seq of insulin-treated KGN cells, we found that AGER, also known as RAGE, was up-regulated. AGER was found to be able to activate the NF-kB signaling pathway and ERK signaling pathway [34, 35], which may indicate a potential molecular mechanism of high concentrations of insulin leading to cell senescence. To further verify the adverse effect of IR on ovarian reserve in vitro, metformin was used to alleviate HFD-induced IR. Moreover, a comparison of ovarian function was made between the HFD and Met groups. Metformin is the drug most commonly used to alleviate IR [36], with a dose of 200 mg/kg·d based on previous study [37]. Our results showed that metformin decreased HFD-induced IR Notably, we also found that a reversal of ovarian reserve existed in Met mice compared with HFD, which supported previous findings that IR results in DOR. According to TCM theory, GNC has a kidney/liver tonifying effect that is used to alleviate declining functions related to aging. GNC has been shown to improve learning and memory, delay skin aging, and enhance resistance to oxidative stress [17, 18]. Furthermore, GNC can significantly extend lifespan and mitigate damage to the ovarian reserve according to our previous study [19, 20]. According to our results, GNC formula significantly reduced IR and increased the ovarian reserve. The expression levels of p53, p21, and p16 detected by WB in GNC were much lower than those in Met, which suggested that GNC was more effective in attenuating ovarian aging compared to metformin. However, the protein expression levels of p16, p53, and p21 detected by IHC showed no notable differences among the groups. Considering the higher accuracy of WB quantification over IHC, we conclude that GNC was more effective in attenuating ovarian aging than metformin. This suggests that other potential therapeutic targets of GNC may exist, and these require further exploration. According to our results, metformin and GNC were able to alleviate the IR-induced diminished ovarian reserve, which is consistent with research showing that metformin can alleviate aging-related diseases and improve health span [38, 39]. Furthermore, the SASP of senescent cells and accumulation of senescent cells are the major causes of excessive inflammation in age-related disorders [40], which is consistent with the elevated inflammatory factors in DOR [41, 42]. However, we must acknowledge that there were limitations in the present study. It is commonly known that oocyte-granulosa cell communication plays a crucial role in the development of follicles [43]. Thus, one limitation of our study was that the effect of the senescent granulosa cells on oocytes was not described. In clinical practice, we found that patients with IR were more vulnerable to DOR, which has not been previously reported. Despite its preliminary character, this study clearly indicated the adverse effects of IR on ovarian reserve and granulosa cells, none of which have been previously reported. One important future direction of our study is to probe the specific mechanisms of IR on senescent granulosa cells; next, we want to understand whether these cells have adverse effects on oocytes. We also assume that the SASP of senescent cells may play a key role in oocyte dysfunction. Moreover, we found that the P53 gene was up-regulated in mouse ovaries based on IHC images. This poses an additional hypothesis regarding the involvement of IR and senescent granulosa cells in oocyte aging that warrants further investigation. In conclusion, we identified that IR decreased ovarian reserve via modulating granulosa cell senescence and GNC had preservatory effects upon the ovarian reserve through regulating insulin resistance. Remarkably, these results from a mouse model and KGN cells appear to agree with our previous clinical observations. However, the detailed mechanisms of hyperinsulinemia and senescent oocyte granulosa cells need to be further investigated. ## Data availability statement The data presented in this study has been deposited and made publicly available in an acceptable repository. The raw data of RNA-seq gene expression analysis can be found in [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE223248]. ## Ethics statement The animal study was reviewed and approved by the Animal Experimental Ethical Committee of Fudan University. ## Author contributions WW and HG contributed to the conception and design of the study. LG and HG performed the majority of experiments, data acquisitions, analyzed data, and wrote the manuscript. YR assisted with animal experiments. LQ helped analyze results. WW and ML supervised the study and helped to finalize the manuscript. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The handling editor TZ declared a past co-authorship with the author ML. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## References 1. Ubaldi FM, Rienzi L, Ferrero S, Baroni E, Sapienza F, Cobellis L. Management of poor responders in IVF reprod biomed. Online (2005) 10. DOI: 10.1016/s1472-6483(10)60946-7 2. Bunnewell SJ, Honess ER, Karia AM, Keay SD, Al Wattar BH, Quenby S. Diminished ovarian reserve in recurrent pregnancy loss: a systematic review and meta-analysis. Fertil Steril (2020) 113 818-827.e3. DOI: 10.1016/j.fertnstert.2019.11.014 3. Devine K, Mumford SL, Wu M, DeCherney AH, Hill MJ, Propst A. Diminished ovarian reserve in the united states assisted reproductive technology population: Diagnostic trends among 181,536 cycles from the society for assisted reproductive technology clinic outcomes reporting system. Fertil Steril (2015) 104 612-19 e3. DOI: 10.1016/j.fertnstert.2015.05.017 4. Abdul-Ghani M, DeFronzo RA. Insulin resistance and hyperinsulinemia: the egg and the chicken. J Clin Endocrinol Metab (2021) 106. DOI: 10.1210/clinem/dgaa364 5. Nandi A, Chen Z, Patel R, Poretsky L. Polycystic ovary syndrome. Endocrinol Metab Clin North Am (2014) 43. DOI: 10.1016/j.ecl.2013.10.003 6. Verit FF, Akyol H, Sakar MN. Low anti-mullerian hormone levels may be associated with cardiovascular risk markers in women with diminished ovarian reserve. Gynecol Endocrinol (2016) 32. DOI: 10.3109/09513590.2015.1116065 7. Li XH, Wang HP, Tan J, Wu YD, Yang M, Mao CZ. Loss of pigment epithelium-derived factor leads to ovarian oxidative damage accompanied by diminished ovarian reserve in mice. Life Sci (2019) 216. DOI: 10.1016/j.lfs.2018.11.015 8. López-Otín C, Blasco MA, Partridge L, Serrano M, Kroemer G. The hallmarks of aging. Cell (2013) 153. DOI: 10.1016/j.cell.2013.05.039 9. Gorgoulis V, Adams PD, Alimonti A, Bennett DC, Bischof O, Bishop C. Cellular senescence: Defining a path forward. Cell (2019) 179. DOI: 10.1016/j.cell.2019.10.005 10. Chow HM, Shi M, Cheng A, Gao Y, Chen G, Song X. Age-related hyperinsulinemia leads to insulin resistance in neurons and cell-cycle-induced senescence. Nat Neurosci (2019) 22. DOI: 10.1038/s41593-019-0505-1 11. Li Q, Hagberg CE, Silva Cascales H, Lang S, Hyvönen MT, Salehzadeh F. Obesity and hyperinsulinemia drive adipocytes to activate a cell cycle program and senesce. Nat Med (2021) 27. DOI: 10.1038/s41591-021-01501-8 12. Baboota RK, Spinelli R, Erlandsson MC, Brandao BB, Lino M, Yang H. Chronic hyperinsulinemia promotes human hepatocyte senescence. Mol Metab (2022) 64. DOI: 10.1016/j.molmet.2022.101558 13. Ogrodnik M, Miwa S, Tchkonia T, Tiniakos D, Wilson CL, Lahat A. Cellular senescence drives age-dependent hepatic steatosis. Nat Commun (2017) 8. DOI: 10.1038/ncomms15691 14. Zhou XY, Zhang J, Li Y, Chen YX, Wu XM, Li X. Advanced oxidation protein products induce G1/G0-phase arrest in granulosa cells. Oxid Med Cell Longev (2021) 6634718. DOI: 10.1155/2021/6634718 15. Meng F, Li J, Rao Y, Wang W, Fu Y. Gengnianchun extends the lifespan of. Oxid Med Cell Longev (2018) 4740739. DOI: 10.1155/2018/4740739 16. Zhang Y, Cao Y, Wang L. The effects of a new, improved Chinese medicine, gengnianchun formula granules, on hot flushes, depression, anxiety, and sleep in Chinese peri- and postmenopausal women: a randomized placebo-controlled trial. Menopause (2020) 27 899-905. DOI: 10.1097/GME.0000000000001558 17. Chen PL, Wang WJ, Rao YQ, Li J, Cheng MJ. Serum containing gengnianchun formula suppresses amyloid β−induced inflammatory cytokines in BV−2 microglial cells by inhibiting the NF−κB and JNK signaling pathways. Mol Med Rep (2018) 17. DOI: 10.3892/mmr.2018.8524 18. Rao YQ, Li J, Wang WJ. Effects of gengnianchun on learning and memory ability, neurotransmitter, cytokines, and leptin in ovariectomized rats. Int J Clin Exp Med (2015) 8 19. Zhang L, Wang WJ. [Research advances of traditional Chinese medicine in delaying skin aging]. Zhong Xi Yi Jie He Xue Bao (2009) 7. DOI: 10.3736/jcim20090315 20. Zhao F, Wang W. Gengnianchun recipe protects ovarian reserve of rats treated by 4-vinylcyclohexene diepoxide. Int J Endocrinol (2020) 2020. DOI: 10.1155/2020/9725898 21. Bu S, Sun M, Zhang Y. Effect of Gengnianjian on up regulated estrogen receptor mRNA to substance P and beta-endorphin in hypothalamus of aging female rats. Chin J Integr Tradit West Med (1998) 18 28-31 22. Gao SX, Guo J, Fan GQ, Qiao Y, Zhao RQ, Yang XJ. ZAG alleviates HFD-induced insulin resistance accompanied with decreased lipid depot in skeletal muscle in mice. J Lipid Res (2018) 59. DOI: 10.1194/jlr.M082180 23. Liu KJ, Wang WJ, Li DJ, Jin HF, Zhou WJ. Effect of gengnianchun recipe on bone mineral density, bone biomechanical parameters and serum lipid level in ovariectomized rats. Chin J Integr Med (2006) 12. DOI: 10.1007/BF02857360 24. Oktay K, Schenken RS, Nelson JF. Proliferating cell nuclear antigen marks the initiation of follicular growth in the rat. Biol Reprod (1995) 53 295-301. DOI: 10.1095/biolreprod53.2.295 25. Hardy RG, Tselepis C, Hoyland J, Wallis Y, Pretlow TP, Talbot I. Aberrant p-cadherin expression is an early event in hyperplastic and dysplastic transformation in the colon. Gut (2002) 50. DOI: 10.1136/gut.50.4.513 26. Na HG, Kim Y-D, Bae CH, Choi YS, Jin HJ, Shin KC. High concentration of insulin induces MUC5AC expression. Am J Rhinol Allergy (2018) 32. DOI: 10.1177/1945892418782223 27. Wu N, Yang D, Wu Z, Yan M, Zhang P, Liu Y. Insulin in high concentration recede cigarette smoke extract induced cellular senescence of airway epithelial cell through autophagy pathway. Biochem Biophys Res Commun (2019) 509 498-505. DOI: 10.1016/j.bbrc.2018.12.130 28. Gulli G, Ferrannini E, Stern M, Haffner S, DeFronzo RA. The metabolic profile of NIDDM is fully established in glucose-tolerant offspring of two Mexican-American NIDDM parents. Diabetes (1992) 41. DOI: 10.2337/diab.41.12.1575 29. Xie Z, Gao G, Wang H, Li E, Yuan Y, Xu J. Dehydroabietic acid alleviates high fat diet-induced insulin resistance and hepatic steatosis through dual activation of PPAR-γ and PPAR-α. BioMed Pharmacother (2020) 127. DOI: 10.1016/j.biopha.2020.110155 30. Jorquera G, Meneses-Valdés R, Rosales-Soto G, Valladares-Ide D, Campos C, Silva-Monasterio M. High extracellular ATP levels released through pannexin-1 channels mediate inflammation and insulin resistance in skeletal muscle fibres of diet-induced obese mice. Diabetologia (2021) 64. DOI: 10.1007/s00125-021-05418-2 31. Yin Y, Chen H, Wang Y, Zhang L, Wang X. Roles of extracellular vesicles in the aging microenvironment and age-related diseases. J Extracell Vesicles (2021) 10. DOI: 10.1002/jev2.12154 32. Ohtani N. The roles and mechanisms of senescence-associated secretory phenotype (SASP): can it be controlled by senolysis. Inflammation Regener (2022) 42. DOI: 10.1186/s41232-022-00197-8 33. Malavolta M, Pierpaoli E, Giacconi R, Basso A, Cardelli M, Piacenza F. Anti-inflammatory activity of tocotrienols in age-related pathologies: A SASPected involvement of cellular senescence. Biol Proced Online (2018) 20 22. DOI: 10.1186/s12575-018-0087-4 34. Alves M, Calegari VC, Cunha DA, Saad MJ, Velloso LA, Rocha EM. Increased expression of advanced glycation end-products and their receptor, and activation of nuclear factor kappa-b in lacrimal glands of diabetic rats. Diabetologia (2005) 48. DOI: 10.1007/s00125-005-0010-9 35. Ma L, Sun P, Zhang JC, Zhang Q, Yao SL. Proinflammatory effects of S100A8/A9. Int J Mol Med (2017) 40. DOI: 10.3892/ijmm.2017.2987 36. Herman R, Kravos NA, Jensterle M, Janež A, Dolžan V. Metformin and insulin resistance: A review of the underlying mechanisms behind changes in GLUT4-mediated glucose transport. Int J Mol Sci (2022) 23. DOI: 10.3390/ijms23031264 37. Gao LH, Liu Q, Liu SN, Chen ZY, Li CN, Lei L. A refined-JinQi-JiangTang tablet ameliorates prediabetes by reducing insulin resistance and improving beta cell function in mice. J Ethnopharmacol (2014) 151. DOI: 10.1016/j.jep.2013.11.024 38. Kulkarni AS, Gubbi S, Barzilai N. Benefits of metformin in attenuating the hallmarks of aging. Cell Metab (2020) 32 15-30. DOI: 10.1016/j.cmet.2020.04.001 39. Chen S, Gan D, Lin S, Zhong Y, Chen M, Zou X. Metformin in aging and aging-related diseases: clinical applications and relevant mechanisms. Theranostics (2022) 12. DOI: 10.7150/thno.71360 40. Wang TW, Johmura Y, Suzuki N, Omori S, Migita T, Yamaguchi K. Blocking PD-L1-PD-1 improves senescence surveillance and ageing phenotypes. Nature (2022) 611. DOI: 10.1038/s41586-022-05388-4 41. Gao H, Gao L, Wang W. Advances in the cellular immunological pathogenesis and related treatment of primary ovarian insufficiency. Am J Reprod Immunol (2022) 88. DOI: 10.1111/aji.13622 42. Lliberos C, Liew SH, Zareie P, La Gruta NL, Mansell A, Hutt K. Evaluation of inflammation and follicle depletion during ovarian ageing in mice. Sci Rep (2021) 11 278. DOI: 10.1038/s41598-020-79488-4 43. Chen M, He C, Zhu K, Chen Z, Meng Z, Jiang X. Resveratrol ameliorates polycystic ovary syndrome. Theranostics (2022) 12. DOI: 10.7150/thno
--- title: 'Waist-to-height ratio and new-onset hypertension in middle-aged and older adult females from 2011 to 2015: A 4-year follow-up retrospective cohort study from the China Health and Retirement Longitudinal Study' authors: - Yang Wu - Yingmu Tong - Hai Wang - Xing Zhang - Yunxiang Long - Qinglin Li - Jie Ren - Chang Liu journal: Frontiers in Public Health year: 2023 pmcid: PMC10016226 doi: 10.3389/fpubh.2023.1122995 license: CC BY 4.0 --- # Waist-to-height ratio and new-onset hypertension in middle-aged and older adult females from 2011 to 2015: A 4-year follow-up retrospective cohort study from the China Health and Retirement Longitudinal Study ## Abstract ### Background Central obesity was closely associated with hypertension. Middle-aged and older adult females, defined as those aged 45 and above, were more likely to suffer from central obesity. For waist-to-height ratio (WHtR) was used as central obesity assessment, the object of this study was to illustrate the relationship between WHtR and the incidence of hypertension in middle-aged and older adult females in China. ### Methods Data used in this prospective cohort study was derived from the China Health and Retirement Longitudinal Study (CHARLS) in a baseline survey from 2011 to 2012 with a follow-up duration of 4 years. The waist-to-height ratio was calculated as waist circumstance divided by height, and the cohort was divided into different groups based on WHtR level. The outcome variable was new-onset hypertension. ### Results Of the 2,438 participants included in the study, 1,821 ($74.7\%$) had high WHtR levels (WHtR ≥ 0.5). As WHtR was closely related to new-onset hypertension in a multivariable logistics regression mode [OR: 7.89 ($95\%$ CI: 2.10–29.67)], individuals with high WHtR were also more likely to suffer from hypertension compared with low WHtR levels [OR: 1.34 ($95\%$ CI: 1.06–1.69)]. ### Conclusion WHtR is positively related to the risk of hypertension incidents among middle-aged and older adult females. Individuals with WHtR ≥ 0.5 were more likely to suffer from hypertension. ## Background In recent years, the problem of global aging has continued to intensify. As individuals get older, organ function and metabolism levels decreased significantly, both of which led to metabolic-related disease. Several studies illustrated the positive relationship between aging and hypertension incidence [1]. Besides these, physiological changes during menopause made a great role in regulating blood pressure [2, 3]. Middle-aged and older adult females, defined as females aged 45 and above, were at high risk of suffering from hypertension. Central obesity, manifesting as extra fat collected in the abdomen and stomach, raised attention worldwide for its rapidly increased incidence. However, the growth of age was also closely related to central obesity [4, 5]. For a higher proportion of body fat and sex hormones difference, the incidence of central obesity was higher among females than males (6–8). As general obesity showed little relationship, higher relevance between central obesity and different metabolic-related diseases was illustrated (9–11). Moreover, the positive relationship between central obesity and hypertension was also revealed [12]. As middle-aged and older adult females were at high risk of suffering from hypertension, which often led to a bad outcome. There was an urgent need for risk evaluation. Waist-to-height ratio (WHtR), a proxy index for central obesity assessment, has been widely accepted as a valuable tool for a health assessment with a cut-off point of 0.5 [13]. Though there were some investigations that revealed the relationship between WHtR and the incidence of metabolism-related disease (14–16), few studies explored the association between WHtR and the incidence of hypertension among middle-aged and older adult females in China. Therefore, in this study, we aimed to explore the relationship between WHtR and the incidence of hypertension and testing the usefulness of the cut-off points for health assessment in WHtR. ## Study design and population The cohort of this study originated from the China Health and Retirement Longitudinal Study (CHARLS) from 2011 to 2015, which is in charge of the National Development Institute of Peking University. CHARLS is an ongoing representative survey targeting individuals aged 45 and above from 450 villages and 150 counties or districts within 28 provinces in mainland China. The baseline wave was conducted between June 2011 and March 2012 and 17,708 individuals were involved. Among all the participants, 13,013 provided venous blood. All the participants were followed every 2 years. Previous research papers [17] have shown information about this. Individuals: [1] with complete information of venous blood sample in wave 1; [2] followed up at least once in wave 2, 3; [3] with complete information on WHtR met inclusion criteria. Individuals: [1] who were male; [2] combined with hypertension in wave 1; [3] with missing data on age or age <45; [4] who were not interviewed in 2015 were excluded from the study. A total of 2,438 individuals were enrolled in the study. As WHtR = 0.5 was used as a cut-off point for health assessment in the previous study, the cohort was divided into two groups based on WHtR level (Figure 1). Figure 1: Flow chart of the study. ## Follow-up duration and new-onset hypertension As all the baseline characteristic was collected from 2011 to 2012 in wave 1, all the participants followed two waves every 2 years (wave 2 and 3) until 2015. During the follow-up in waves 2 and 3, new-onset hypertension was assessed by the following criteria: [1] an SBP higher than 140 mm Hg or a DBP higher than 90 mmHg; [2] self-report of a doctor diagnosis; and [3] self-report of antihypertensive treatment. ## Other covariates The interviewers trained by CHARLS collected information on demographic background, health status, and biomarkers according to the questionnaire. Demographic background including age, gender (male/female), education level (illiteracy, primary school, middle school, high school, and above), residence (urban/rural), marital status (married/single) were recorded, Health status consisting of 14 comorbidities (hypertension, diabetes, dyslipidemia, cancer, kidney disease, stroke, heart problem, liver disease, chronic lung disease, digestive disease, nervous problem, memory-related diseases, arthritis, and asthma) and comorbidity-related treatment taken by respondents. The options used to assess the history of alcohol drinking during the interview included: [1] I never had a drink; [2] I used to drink less than once a month; and [3] I used to drink more than once a month. In our study, option 2 and option 3 were regarded as a history of alcohol drinking. Biomarkers, including weight, height, waist circumstance, systolic pressure, and diastolic pressure, were all tested standardly by the interviewer. The blood collection, transported at 4°C temperature and sent to the local laboratory, was executed by the staff of the Chinese Center for Disease Control and Prevention (China CDC) during baseline survey. Then the plasma and buff coat were both frozen at −20°C, transported to Beijing within 2 weeks, and they would be placed in a deep freezer and stored at −80°C until assay before all the serum markers were assayed. eGFR was calculated using the CKD-EPI creatinine formula. WHtR was calculated as waist circumstance divided by height. Height was measured by the height measuring instrument vertically. Training surveyors circled a soft tape at the navel level to measure waist circumstance. Diabetes was diagnosed as one of the following criteria: [1] self-report of a diagnosis by a doctor; [2] HbA1c ≥ $6.5\%$; [3] plasma glucose ≥ 11.1 mmol/L (casual) or plasma glucose ≥ 7.0 mmol/L (fasting); [4] self-report of the diabetes-related treatment. Dyslipidemia was diagnosed as one of the following criteria: [1] self-report of a diagnosis by a doctor; [2] total cholesterol (TC) ≥ 240 mg/dl; [3] high-density lipoprotein cholesterol (HDL) ≤ 40 mg/dl; [4] low-density lipoprotein cholesterol (LDL) ≥ 160 mg/dl; [5] triglycerides (TG) ≥ 150 mg/dl; [6] self-report of the anti-dyslipidemia treatment. Kidney disease was diagnosed as one of the following criteria: [1] self-report of a diagnosis by a doctor; [2] self-report of the kidney disease-related treatment; [3] Estimated glomerular filtration rate (eGFR) <60 ml/min/1.73 m2. ## Statistical analysis All variables were shown as follows: continuous variables with median (IQR) and counts percentages for categorical variables. Mann-Whitney U and Chi-squared tests were used to compare baseline characteristics among cohorts with different levels of WHtR. Univariable and multivariable logistics regression was used to estimate the relationship between WHtR and new-onset hypertension. Four models were constructed, including model 1 (crude), model 2 (adjusted for age), model 3 (adjusted by age, SBP, DBP, residence, education level, digestive disease, smoking) and model 4 (adjusted by age, SBP, DBP, residence, education level, marital status, diabetes, dyslipidemia, kidney disease, cancer, chronic lung disease, liver disease, heart problem, stroke, digestive disease, nervous problems, memory-related disease, arthritis, asthma, smoking, and alcohol drinking). The interaction of different variables on new-onset hypertension was also calculated in model 4. Restrict cubic spline (RCS) functions and smooth curve fitting (penalized spline method) were used to assess the dose-response relationship and the potential non-linear relationship between WHtR and new-onset hypertension. Receiver Operating Characteristic (ROC) analyses were used to compare the effectiveness of new-onset hypertension prediction between WHtR and BMI. As the age of 60 was regarded as criterion for older people, so cut-off of age at 60 was chosen to assess the relationship between WHtR and new onset hypertension in different age groups. Both sensitivity and subgroup analysis were used to test the robustness of our findings. Statistical analyses were performed using the R package (version 4.2.1), and $p \leq 0.05$ was considered statisticallysignificant. ## Baseline characteristics of study participants There were 2,438 individuals included in the final cohort. As the baseline characteristics were shown in Table 1, the median age was 55.5 years old. Individuals with WHtR ≥ 0.5 accounted for 1,821 ($74.7\%$) of the cohort. Compared to individuals with WHtR <0.5, individuals with WHtR ≥ 0.5 were older (56.0 vs. 55.0, $$p \leq 0.022$$), had a higher level in both systolic pressures (121.0 vs. 118.0 mmHg; $p \leq 0.001$) and diastolic pressure (72.0 vs. 70.0 mmHg; $p \leq 0.001$) at baseline. Besides these, individuals with low WHtR had a significantly lower prevalence of diabetes ($7.1\%$ vs. $13.0\%$, $p \leq 0.001$) and dyslipidemia ($30.8\%$ vs. $47.2\%$, $p \leq 0.001$). Table 1 \| Unnamed: 0 \| Overall \| WHtR <0.5 \| WHtR ≥0.5 \| p \| \| --- \| --- \| --- \| --- \| --- \| \| \| n = 2,438 \| n = 617 \| n = 1,821 \| \| \| Age \| 55.50 (49.00, 61.00) \| 55.00 (49.00, 60.00) \| 56.00 (49.00, 62.00) \| 0.022 \| \| WHtR \| 0.54 (0.50, 0.59) \| 0.47 (0.45, 0.49) \| 0.56 (0.53, 0.60) \| <0.001 \| \| eGFR (ml/min/1.73 m2) \| 97.71 (87.15, 104.60) \| 97.91 (87.72, 105.08) \| 97.66 (87.07, 104.45) \| 0.466 \| \| SBP (mmHg) \| 120.00 (112.00, 129.00) \| 118.00 (109.00, 126.75) \| 121.00 (112.00, 129.00) \| <0.001 \| \| DBP (mmHg) \| 72.00 (66.00, 78.00) \| 70.00 (64.00, 76.00) \| 72.00 (66.00, 78.00) \| <0.001 \| \| BMI (kg/m2) \| 23.08 (20.88, 25.42) \| 20.11 (18.47, 21.57) \| 24.11 (22.13, 26.09) \| <0.001 \| \| Residence, n (%) \| \| \| \| 0.027 \| \| Urban \| 314 (12.9) \| 63 (10.3) \| 251 (13.8) \| \| \| Rural \| 2,113 (87.1) \| 551 (89.7) \| 1,562 (86.2) \| \| \| Education level, n (%) \| \| \| \| 0.04 \| \| Illiteracy \| 1,422 (58.3) \| 374 (60.6) \| 1,048 (57.6) \| \| \| Primary school \| 443 (18.2) \| 89 (14.4) \| 354 (19.4) \| \| \| Middle school \| 395 (16.2) \| 103 (16.7) \| 292 (16.0) \| \| \| High school and above \| 178 (7.3) \| 51 (8.3) \| 127 (7.0) \| \| \| Marital status, n (%) \| \| \| \| 0.925 \| \| Alone \| 406 (16.7) \| 104 (16.9) \| 302 (16.6) \| \| \| Married \| 2032 (83.3) \| 513 (83.1) \| 1,519 (83.4) \| \| \| Dyslipidemia, n (%) \| 1,049 (43.0) \| 190 (30.8) \| 859 (47.2) \| <0.001 \| \| Diabetes, n (%) \| 280 (11.5) \| 44 (7.1) \| 236 (13.0) \| <0.001 \| \| Cancer, n (%) \| 31 (1.3) \| 8 (1.3) \| 23 (1.3) \| 1 \| \| Chronic lung disease, n (%) \| 203 (8.4) \| 64 (10.5) \| 139 (7.7) \| 0.037 \| \| Liver disease, n (%) \| 98 (4.0) \| 22 (3.6) \| 76 (4.2) \| 0.588 \| \| Heart problem, n (%) \| 217 (8.9) \| 49 (8.0) \| 168 (9.3) \| 0.381 \| \| Stroke, n (%) \| 27 (1.1) \| 7 (1.1) \| 20 (1.1) \| 1 \| \| Kidney disease, n (%) \| 186 (7.6) \| 45 (7.3) \| 141 (7.7) \| 0.783 \| \| Digestive disease, n (%) \| 684 (28.2) \| 195 (31.9) \| 489 (27.0) \| 0.021 \| \| Nervous problems, n (%) \| 41 (1.7) \| 14 (2.3) \| 27 (1.5) \| 0.259 \| \| Memory related disease, n (%) \| 13 (0.5) \| 2 (0.3) \| 11 (0.6) \| 0.613 \| \| Arthritis, n (%) \| 936 (38.5) \| 214 (34.8) \| 722 (39.7) \| 0.034 \| \| Asthma, n (%) \| 72 (3.0) \| 20 (3.3) \| 52 (2.9) \| 0.728 \| \| Smoking, n (%) \| 195 (8.0) \| 57 (9.3) \| 138 (7.6) \| 0.217 \| \| Alcohol drinking, n (%) \| 174 (7.7) \| 47 (8.4) \| 127 (7.5) \| 0.572 \| \| Glu (mg/dl) \| 100.62 (93.60, 109.26) \| 98.82 (92.16, 107.19) \| 101.34 (93.96, 110.34) \| <0.001 \| \| Creatinine (mg/dl) \| 0.67 (0.60, 0.76) \| 0.67 (0.60, 0.76) \| 0.67 (0.60, 0.76) \| 0.847 \| \| Total cholesterol (mg/dl) \| 193.30 (169.33, 217.27) \| 188.66 (165.66, 211.86) \| 195.23 (170.88, 219.59) \| <0.001 \| \| Triglycerides (mg/dl) \| 104.43 (74.34, 146.02) \| 86.73 (65.49, 127.44) \| 108.86 (78.76, 153.10) \| <0.001 \| \| HDL (mg/dl) \| 104.43 (74.34, 146.02) \| 86.73 (65.49, 127.44) \| 108.86 (78.76, 153.10) \| <0.001 \| \| LDL (mg/dl) \| 117.53 (95.49, 139.95) \| 112.11 (93.56, 133.76) \| 119.07 (95.88, 141.50) \| <0.001 \| \| Hb1Ac (%) \| 5.10 (4.90, 5.40) \| 5.10 (4.80, 5.30) \| 5.10 (4.90, 5.47) \| <0.001 \| ## Relationship between WHtR and new-onset hypertension The relationship between WHtR and new-onset hypertension was assessed in logistics regression. As the result showed, WHtR showed a positive relationship with new-onset hypertension [OR: 21.34 ($95\%$ CI: 6.49–72.93)] (Table 2) in logistics regression. The restricted cubic spline model showed a U-shape relationship between WHtR and new-onset hypertension (Figure 2) with the lowest relationship of hypertension at WHtR = 0.49. As WHtR was used as a marker of health assessment with the cut-off point at 0.5, the relationship between different level of WHtR and new-onset hypertension were further analyzed. Different models were also used to assess the relationship between WHtR and new-onset hypertension. Individuals with WHtR ≥ 0.5 were 1.34 times higher in suffering from hypertension [OR: 1.34 ($95\%$ CI: 1.06–1.69)] (Table 3). Table 3 \| Unnamed: 0 \| Unnamed: 1 \| Model 1 \| Model 1.1 \| Model 2 \| Model 2.1 \| Model 3 \| Model 3.1 \| Model 4 \| Model 4.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| \| OR (95% CI) \| p \| OR (95% CI) \| p \| OR (95% CI) \| p \| OR (95% CI) \| p \| \| WHtR as continuous \| \| 21.34 (6.37–71.51) \| <0.001 \| 16.06 (4.82–53.51) \| <0.001 \| 6.92 (2.01–23.81) \| 0.002 \| 7.89 (2.1–29.67) \| 0.002 \| \| WHtR as categorical \| <0.5 \| 1.00 (Ref.) \| \| 1.00 (Ref.) \| \| 1.00 (Ref.) \| \| 1.00 (Ref.) \| \| \| \| ≥0.5 \| 1.55 (1.26–1.9) \| <0.001 \| 1.51 (1.23–1.86) \| <0.001 \| 1.34 (1.08–1.67) \| 0.009 \| 1.34 (1.06–1.69) \| 0.014 \| ## Subgroup and sensitivity analysis As is shown, participants with WHtR ≥ 0.5 were more likely to suffer from hypertension when age ≥60 [OR: 1.64 ($95\%$ CI: 1.09–2.47)], living in rural [OR: 1.32 ($95\%$ CI: 1.03–1.69)], not combined with diabetes [OR: 1.33 ($95\%$ CI: 1.04–1.70)], combined with dyslipidemia [OR: 1.61 ($95\%$ CI: 1.08–2.39)] and not combined with kidney disease [OR: 1.30 ($95\%$ CI: 1.03–1.66)] (Figure 3). WHtR was also more positively related to new-onset hypertension (Supplementary Figure 1). Sensitivity analysis was in accordance with the results (Supplementary Tables 1–3). Figure 3: Subgroup analysis of relationship between WHtR and new-onset hypertension. WHtR as categorical. All model was adjusted by age, SBP, DBP, residence, education level, marital status, diabetes, dyslipidemia, kidney disease, cancer, chronic lung disease, liver disease, heart problem, stroke, digestive disease, nervous problems, memory related disease, arthritis, asthma, smoking and drinking unless the variable was used as a subgroup variable. ## Discussion We investigated the relationship between WHtR and new-onset hypertension among middle-aged and older women in China. As the results showed, WHtR showed a positive relationship with new-onset hypertension. Besides these, the cut-off point at 0.5 was practical for health assessment. Individuals with WHtR more than 0.5 had a significantly higher incidence of hypertension when compared to others. Though the relationship between obesity and the risk of hypertension was well-established, the mechanism of this relation was quite complex. Several mechanisms were contributing to hypertension development. As adipose tissue accumulated, the renin–angiotensin–aldosterone system (RAAS) was highly promoted, leading to high sodium and water retention [18, 19]. Besides these, changes in endocrine level also played an important role [20]. Decreased adiponectin secretion could also lead to insulin resistance. A high level of leptin could also result in the inflammatory response upregulating. Moreover, fatty acid accumulation was typical among obesity combined with dyslipidemia. All of these endocrine transformations could result in increased blood vessel stiffness, which is the early histology change in hypertension. Decreased estrogen levels in middle-aged and older women also played an important role in hypertension development. The level of ANP and Ang II were elevated for the reason estrogen decreased, both of which could increase the activity of RAAS [21]. However, a low level of estrogen could also lead to a reduction in lipid clearance, accelerating dyslipidemia formation [22]. Decreased levels of estrogen and endocrine dysfunction contributed significantly to hypertension development. As global aging has been accelerating in recent years, age-related diseases raised more attention from all around the world. For the low level of metabolism, obesity and overweight were one of the most common comorbidities threatening the quality of life among the aged. Different obesity subtypes were further studied in recent years. The relationship between abdominal fat accumulation and increased endocrine dysfunction led to a higher incidence of cardiovascular risk among central obesity [23, 24]. Besides these, central obesity was also related to a reduction in quality of life and an increment in health expenses. As the incidence of central obesity raised rapidly during recent years [7, 25], several studies indicated a higher proportion of central obesity among females than males [6, 8], which could be attributed to fertility and decreased estrogen levels. The relationship between obesity and hypertension was well-studied. Yuri et al. [ 26] revealed the relationship between obesity and hypertension among women in Indonesia. In a cross-section study held by Wang et al. [ 27], a synergistic effect of BMI and waist circumstance on the incidence of HBP (defined as SBP ≥ 140 mmHg/or DBP ≥ 90 mmHg or use of antihypertensive medication within 2 weeks) was confirmed in the aged. Though BMI was recognized as an obesity-related marker for a long time, some investigators argued its limitation on not considering the adverse effect of intra-abdominal fat [28]. WHtR was highly recommended for central obesity assessment for its easy measurement, elimination of the impact of height, and universality among different gender, and races. The effectiveness of metabolism-related disease prediction, including metabolic syndrome, hypertension, diabetes, dyslipidemia, and cardiovascular diseases, was compared among different obesity markers (29–31). According to a multicenter cross-section study held by Akbari et al. [ 30], WHtR performed better in hypertension prediction than WHR and BMI. Lee et al. compared the influence of different anthropometric indices on metabolic risk. WHtR was more strongly associated with hypertension in females [32]. Our study used the ROC curve to assess the predicate ability and WHtR showed a higher predictive ability than BMI (Supplementary Figure 2), BMI was associated with a lower increment in new-onset hypertension when compared with WHtR (Supplementary Table 4). Besides these, our study also showed a significantly lower tendency of suffering from hypertension in WHtR <0.5 groups, which supports the usage of WHtR = 0.5 as a cut-off point for health assessment. Obesity was closely related to the incidence of hypertension, which brought a heavy burden to public health. Our study showed a close relationship between WHtR and new-onset hypertension. Though WHtR showed a better ability for hypertension prediction than other markers among middle-aged and older females, new markers or formulas were urgently needed for much more precise prediction. Besides this, how to prevent hypertension among middle-aged and older females was also an essential factor that needs further research. One significant advantage of our study was that this was the first study to analyze the relationship between WHtR and new-onset hypertension among middle-aged and older adult females in China. However, there were still some limitations that should be noticed. First, all the health information was collected according to self-report by participants. However, some participants might be unaware of their diseases, which could lead to bias in baseline information and the outcome variable. Second, some participants had no information about WHtR, leading to being excluded from the final cohort. These might make an impact on results. ## Conclusion This study explored the relationship between WHtR and new-onset hypertension among middle-aged and older adult females in China. As the result shows, WHtR was positively related to hypertension. More attention should be paid to individuals with high WHtR. ## Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. ## Ethics statement The studies involving human participants were reviewed and approved by Biomedical Ethics Review Committee of Peking University (IRB00001052-11015). The patients/participants provided their written informed consent to participate in this study. ## Author contributions YW, HW, and YT: methodology, writing, and revision. XZ, YL, QL, and JR: data curation and investigation. CL: supervision, reviewing, and editing the manuscript. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher's note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpubh.2023.1122995/full#supplementary-material ## References 1. Buford TW. Hypertension and aging. Ageing Res Rev. (2016) 26 96-111. DOI: 10.1016/j.arr.2016.01.007 2. Prevention D. Evaluation, and management of high blood pressure in adults: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Hypertension. (2018) 71 e13-e115. DOI: 10.1161/HYP.0000000000000065 3. Mozaffarian D, Benjamin EJ, Go AS, Arnett DK, Blaha MJ, Cushman M. Heart disease and stroke statistics-−2015 update: a report from the American Heart Association. Circulation. (2015) 131 e29-322. DOI: 10.1161/CIR.0000000000000152 4. Moriarty JP, Branda ME, Olsen KD, Shah ND, Borah BJ, Wagie AE. The effects of incremental costs of smoking and obesity on health care costs among adults: a 7-year longitudinal study. J Occup Environ Med. (2012) 54 286-91. DOI: 10.1097/JOM.0b013e318246f1f4 5. Misganaw A, Mariam DH, Araya T, Aneneh A. Validity of verbal autopsy method to determine causes of death among adults in the urban setting of Ethiopia. BMC Med Res Methodol. (2012) 12 130. DOI: 10.1186/1471-2288-12-130 6. Feng WY, Li XD, Li J, Shen Y, Li Q. Prevalence and risk factors of central obesity among adults with normal BMI in Shaanxi, China: a cross-sectional study. Int J Environ Res Public Health. (2021) 18 11439. DOI: 10.3390/ijerph182111439 7. Wong MCS, Huang J, Wang J, Chan PSF, Lok V, Chen X. Global, regional and time-trend prevalence of central obesity: a systematic review and meta-analysis of 13.2 million subjects. Eur J Epidemiol. (2020) 35 673-83. DOI: 10.1007/s10654-020-00650-3 8. Israel E, Hassen K, Markos M, Wolde K, Hawulte B. Central obesity and associated factors among urban adults in Dire Dawa Administrative City, Eastern Ethiopia. Diab Metab Syndr Obes. (2022) 15 601-14. DOI: 10.2147/DMSO.S348098 9. Goh VHH, Hart WG. Excess fat in the abdomen but not general obesity is associated with poorer metabolic and cardiovascular health in premenopausal and postmenopausal Asian women. Maturitas. (2018) 107 33-8. DOI: 10.1016/j.maturitas.2017.10.002 10. Cameron AJ, Magliano DJ, Shaw JE, Zimmet PZ, Carstensen B, Alberti KG. The influence of hip circumference on the relationship between abdominal obesity and mortality. Int J Epidemiol. (2012) 41 484-94. DOI: 10.1093/ije/dyr198 11. Piché ME, Poirier P, Lemieux I, Després JP. Overview of epidemiology and contribution of obesity and body fat distribution to cardiovascular disease: an update. Prog Cardiovasc Dis. (2018) 61 103-13. DOI: 10.1016/j.pcad.2018.06.004 12. Niu J, Seo DC. Central obesity and hypertension in Chinese adults: a 12-year longitudinal examination. Prev Med. (2014) 62 113-8. DOI: 10.1016/j.ypmed.2014.02.012 13. Browning LM, Hsieh SD, Ashwell M. systematic review of waist-to-height ratio as a screening tool for the prediction of cardiovascular disease and diabetes: 0.5 could be a suitable global boundary value. Nutr Res Rev. (2010) 23 247-69. DOI: 10.1017/S0954422410000144 14. Choi JR, Koh SB, Choi E. Waist-to-height ratio index for predicting incidences of hypertension: the ARIRANG study. BMC Public Health. (2018) 18 767. DOI: 10.1186/s12889-018-5662-8 15. Hou X, Chen S, Hu G, Chen P, Wu J, Ma X. Stronger associations of waist circumference and waist-to-height ratio with diabetes than BMI in Chinese adults. Diabetes Res Clin Pract. (2019) 147 9-18. DOI: 10.1016/j.diabres.2018.07.029 16. Shen S, Lu Y, Qi H, Li F, Shen Z, Wu L. Waist-to-height ratio is an effective indicator for comprehensive cardiovascular health. Sci Rep. (2017) 7 43046. DOI: 10.1038/srep43046 17. Zhao Y, Hu Y, Smith JP, Strauss J, Yang G. Cohort profile: the China Health and Retirement Longitudinal Study (CHARLS). Int J Epidemiol. (2014) 43 61-8. DOI: 10.1093/ije/dys203 18. Mende CW. Obesity and hypertension: a common coexistence. J Clin Hypertens (Greenwich). (2012) 14 137-8. DOI: 10.1111/j.1751-7176.2011.00578.x 19. Wofford MR, Hall JE. Pathophysiology and treatment of obesity hypertension. Curr Pharm Des. (2004) 10 3621-37. DOI: 10.2174/1381612043382855 20. Gnatiuc L, Alegre-Díaz J, Halsey J, Herrington WG, López-Cervantes M, Lewington S. Adiposity and blood pressure in 110 000 Mexican adults. Hypertension. (2017) 69 608-14. DOI: 10.1161/HYPERTENSIONAHA.116.08791 21. O'Donnell E, Floras JS, Harvey PJ. Estrogen status and the renin angiotensin aldosterone system. Am J Physiol Regul Integr Comp Physiol. (2014) 307 R498-500. DOI: 10.1152/ajpregu.00182.2014 22. Jiao L, Machuki JO, Wu Q, Shi M, Fu L, Adekunle AO. Estrogen and calcium handling proteins: new discoveries and mechanisms in cardiovascular diseases. Am J Physiol Heart Circ Physiol. (2020) 318 H820-h829. DOI: 10.1152/ajpheart.00734.2019 23. Després JP. Body fat distribution and risk of cardiovascular disease: an update. Circulation. (2012) 126 1301-13. DOI: 10.1161/CIRCULATIONAHA.111.067264 24. Chuang SY, Hsu YY, Chen RC, Liu WL, Pan WH. Abdominal obesity and low skeletal muscle mass jointly predict total mortality and cardiovascular mortality in an elderly Asian population. J Gerontol A Biol Sci Med Sci. (2016) 71 1049-55. DOI: 10.1093/gerona/glv192 25. Li X, Niu H, Bai X, Wang Y, Wang W. Association of obesity and hypertension: a cohort study in China. Int J Hypertens. (2021) 2021 1607475. DOI: 10.1155/2021/1607475 26. Nurdiantami Y, Watanabe K, Tanaka E, Pradono J, Anme T. Association of general and central obesity with hypertension. Clin Nutr. (2018) 37 1259-63. DOI: 10.1016/j.clnu.2017.05.012 27. Zhang W, He K, Zhao H, Hu X, Yin C, Zhao X. Association of body mass index and waist circumference with high blood pressure in older adults. BMC Geriatr. (2021) 21 260. DOI: 10.1186/s12877-021-02154-5 28. Kopelman PG. Obesity as a medical problem. Nature. (2000) 404 635-43. DOI: 10.1038/35007508 29. Ma YL, Jin CH, Zhao CC, Ke JF, Wang JW, Wang YJ. Waist-to-height ratio is a simple and practical alternative to waist circumference to diagnose metabolic syndrome in type 2 diabetes. Front Nutr. (2022) 9 986090. DOI: 10.3389/fnut.2022.986090 30. Akbari-Khezrabadi A, Zibaeenezhad MJ, Shojaeefard E, Naseri A, Mousavi S, Sarejloo S. Can anthropometric indices predict the chance of hypertension? A multicentre cross-sectional study in Iran. BMJ Open. (2022) 12 e062328. DOI: 10.1136/bmjopen-2022-062328 31. Ashwell M, Gunn P, Gibson S. Waist-to-height ratio is a better screening tool than waist circumference and BMI for adult cardiometabolic risk factors: systematic review and meta-analysis. Obes Rev. (2012) 13 275-86. DOI: 10.1111/j.1467-789X.2011.00952.x 32. Lee BJ, Yim MH. Comparison of anthropometric and body composition indices in the identification of metabolic risk factors. Sci Rep. (2021) 11 9931. DOI: 10.1038/s41598-021-89422-x
--- title: 'Mechanisms influencing the factors of urban built environments and coronavirus disease 2019 at macroscopic and microscopic scales: The role of cities' authors: - Longhao Zhang - Xin Han - Jun Wu - Lei Wang journal: Frontiers in Public Health year: 2023 pmcid: PMC10016229 doi: 10.3389/fpubh.2023.1137489 license: CC BY 4.0 --- # Mechanisms influencing the factors of urban built environments and coronavirus disease 2019 at macroscopic and microscopic scales: The role of cities ## Abstract In late 2019, the coronavirus disease 2019 (COVID-19) pandemic soundlessly slinked in and swept the world, exerting a tremendous impact on lifestyles. This study investigated changes in the infection rates of COVID-19 and the urban built environment in 45 areas in Manhattan, New York, and the relationship between the factors of the urban built environment and COVID-19. COVID-19 was used as the outcome variable, which represents the situation under normal conditions vs. non-pharmacological intervention (NPI), to analyze the macroscopic (macro) and microscopic (micro) factors of the urban built environment. Computer vision was introduced to quantify the material space of urban places from street-level panoramic images of the urban streetscape. The study then extracted the microscopic factors of the urban built environment. The micro factors were composed of two parts. The first was the urban level, which was composed of urban buildings, Panoramic View Green View Index, roads, the sky, and buildings (walls). The second was the streets' green structure, which consisted of macrophanerophyte, bush, and grass. The macro factors comprised population density, traffic, and points of interest. This study analyzed correlations from multiple levels using linear regression models. It also effectively explored the relationship between the urban built environment and COVID-19 transmission and the mechanism of its influence from multiple perspectives. ## 1. Introduction Novel pneumonia caused by coronavirus 2019 (COVID-19), which leads to severe acute respiratory syndrome, has been raging worldwide for nearly 3 years since December 2019 [1]. The speed of transmission of the virus, its infectiousness, and the number of mutations have been the most unprecedented in human medical history. Large cities and metropolitan areas have been the areas most affected by the spread of the virus, exacerbated by the areas' dense population distribution [2]. To strictly control the rate of COVID-19 transmission and to reduce the rates of infection and deaths, countries have adopted non-pharmaceutical interventions (NPIs) including urban lockdown, home isolation, controlled social distancing, and travel restrictions [3, 4]. Ever since it reared its ugly head, COVID-19 has attracted substantial attention from the global community, and various studies on COVID-19 have emerged accordingly. A majority of the studies have focused on the related factors of sociodemographics and the urban built environment. The results vary from two different research perspectives. From the sociodemographic perspective, the risk of COVID-19 infection is much higher for the elderly and children than it is for young and middle-aged adults (5–7). In addition, the degree of economic development across regions may exert an impact on the transmission rate of COVID-19 [8, 9]. For example, the availability of health insurance has been highly correlated with the spread of COVID-19 [10]. Low-income areas, especially older communities with low levels of income, have been more susceptible to COVID-19 infection [11, 12], all of which have been related to regional economic development. Using logistic regression models, several studies have demonstrated that levels of regional literacy are also associated with the prevalence of COVID-19 [13]. Other factors have also been correlated with COVID-19 transmission, such as blood type, respiratory disease, and chronic diseases. Additionally, personal habits have been associated with COVID-19 transmission [14, 15]. The strict implementation of NPI against COVID-19 has proven effective in mitigating the spread of the virus [16, 17]. In the post-pandemic era, the patterns of behavior in daily life have changed due to reliable NPIs implemented by governments [18]. Under the influence of NPIs, the range of activities of urban residents has been significantly reduced, thereby rendering them increasingly dependent largely on the surroundings of their homes, the natural urban environment, and the built urban environment [19]. The surroundings of urban homes and the built environment have exerted a direct effect on the physical and mental health of urban residents [20]. Simultaneously, low-density neighborhoods, large homes, developed urban residential surroundings and infrastructure, rich urban greenery, and large urban green spaces can greatly enhance the life satisfaction and wellbeing of residents under COVID-19 NPIs [21]. From the perspective of the urban built environment, the impact of the urban built environment on COVID-19 is extremely important in addition to socio-demographic factors, which has been confirmed by many studies. The relationship between COVID-19 transmission and population density is relatively controversial. Previous studies demonstrate that the incidence and transmission of COVID-19 are higher in densely populated areas with high population contact [22]. A comparison of the results of linear regression models from 182 countries points to a positive association between population density and COVID-19 transmission [23]. In contrast, the results of structural equation modeling at the city level illustrate that population density is negatively associated with COVID-19 transmission in Tehran [24]. This result is interesting, where a few studies argue that urban population density is non-significantly correlated with the spread of COVID-19 [25]. The relationship between urban population density and the transmission rate of COVID-19 is complex. Thus, the various responses of governments and urban residents to the pandemic across nations may lead to different results, which are reasonably explained by the findings of Hamidi et al. [ 26] in the United States, Lin et al. [ 27] in China, and Boterman [28] in the Netherlands. Meanwhile, the relationship between urban building density and COVID-19 lacks elucidation and is subject to a certain degree of controversy [29]. A few studies demonstrate that no correlation exists between building density and COVID-19 after omitting certain confounding factors [28]. During the COVID-19 pandemic and under government NPIs, the wellbeing of residents living in high-density areas was negatively correlated with living density due to changes in the scope of life and lifestyle behaviors. However, this compact urban form leads to relatively easy access to urban healthcare resources, which could improve the health status of residents [18, 30]. From the perspective of the urban built environment alone, different factors in the urban built environment may have an impact on the spread and transmission of COVID-19 during an epidemic pandemic. For example, public transportation [31] and points of interest (POI) [32], among others, are generally considered to exhibit a positive association with COVID-19 transmission. When the outbreak was in its emergent stage, public transportation was considered the main method of COVID-19 transmission. Therefore, many governments advised urban residents to avoid public transportation as much as possible while introducing corresponding NPIs and limiting the range of activities of residents. In addition, they frequently urged urban residents to use multiple modes of transportation, such as self-driving, walking, and cycling [33]. The results of analyses using multiscale geographically weighted regression suggested that the high availability of medical resources around a community could effectively inhibit the spread of COVID-19 [7]. Through structural equation modeling and categorical regression modeling, other analyses demonstrated high-quality housing and high-quality green space as being negatively associated with the spread of COVID-19 [10, 34]. Green spaces around large residential areas exerted an inhibitory effect on the deterioration of urban health and wellbeing due to COVID-19 [30]. Notably, the risk of infection and transmission rates were high for neighborhoods with high levels of community convenience [35]. Integrating all patients with COVID-19 into high-grade urban hospitals is unrealistic because hospital capacity is far from adequate for treating such a large number of patients at the burgeoning stage of a pandemic such as COVID-19. Moreover, the risk of collapse of urban public healthcare is prevalent as demonstrated by the collapse of public healthcare to varying degrees in various countries during the COVID-19 outbreak. Therefore, a community-level system for identifying and isolating individuals with infection is essential to the response to COVID-19 [36]. In summary, several questions can be elicited from the influence of the urban built environment on the spread of COVID-19: (I) how the macroscopic urban built environment and the microscopic urban built environment have an impact on the spread of COVID-19 in the urban built environment; and (II) what is the impact of the macroscopic urban built environment and the microscopic built environment on the incidence and lethality of COVID-19. However, most of the community-level studies at this stage have used administrative boundaries to delineate the selection of variables, and the disadvantage of this method of variable selection is that it does not reflect the actual activities of residents. In this regard, Li et al. used structural equation modeling to reveal the relationship between commercial vitality and transportation infrastructure on the increase in the number of confirmed cases, and innovatively used buffer zones to extract urban built environment factors around confirmed cases [37]. Wang et al. used walking circles at different times to investigate the correlation between urban built environment and community level spatial distribution [38]. By extracting the established environmental factors in both spatial dimensions and examining the correlation between these factors and the prevalence of COVID-19, the issue of the transmission mechanism of COVID-19 before and after the implementation of community-level NPI measures was then analyzed. Studies at this stage ignore the lack of multi-level studies on the mechanisms of the urban micro-built environment influencing the spread of COVID-19. Whether the urban street green environment and urban street spatial quality have an impact on the spread of COVID-19 has not been explored, and the impact of urban built environment on the long-term trend and overall trend of COVID-19 has not been considered comprehensively. In this study, based on the study of the influence mechanism between the macroscopic built environment and COVID-19, the influence mechanism between the microscopic built environment and COVID-19 was considered at multiple levels using Google Street View panoramic street view images. The impact of urban built environment on the long-term trend and overall trend of COVID-19 is investigated using multiple variables, and the influence mechanism of urban built environment on COVID-19 is examined at multiple levels (macro level and micro level) and multiple dimensions (time dimension). The results of the study can provide a basis and reference for governmental decision makers to formulate more reasonable NPI policies to slow down the spread of COVID-19 during pandemic periods. The results of the study may provide a reference solution to control the spread and spread of the virus, and the results may provide effective recommendations to contain potential respiratory disease outbreaks. ## 2.1. Research region New York City is considered the first epicenter of the COVID-19 outbreak in the United States. It has a population of ~8.51 million (as of 2017) and an area of ~1,214 km2 (including the sea). With an average of 28 people per square mile, New York City is the main international maritime, airport, and financial metropolis of the United States and has five boroughs under its jurisdiction, namely, Brooklyn, Queens, Manhattan, the Bronx, and Staten Island. Manhattan is the most densely populated and smallest of the five boroughs of New York City, which translates to a very high population and housing density when compared with those of other boroughs in New York City. Manhattan is described as the economic and cultural center of the United States and is home to New York's central business district, which houses the headquarters of most Fortune 500 companies and the headquarters of the United Nations. Thus, the nearly 50 million tourists who visit New York City each year significantly contribute to the risk of COVID-19 transmission and routes of transmission. Figure 1 describes the study area and its road network. As the center of the metropolitan area, a major outbreak of COVID-19 is likely to spread rapidly to other areas of the metropolis and continue to expand outward. Thus, understanding the relationship between the spread of COVID-19 and the factors of the urban built environment is an important aspect for urban decision-makers in mitigating the spread of the disease and in developing openness measures. Figure 1: Study area: (A) Map of the United States; (B) New York County; and (C) Manhattan road network. ## 2.2.1. Google street view images The study obtained urban street panorama images from Google Maps to reflect the physical characteristics of the urban environment. Factors related to the urban environment were extracted from these images as evaluation indexes of the urban environment. To improve the representativeness of the physical features and environmental factors of the urban environment, the study created a collection point for every 100 m on all urban roads in the study area. A total of 67,025 collection points were set up to collect images with each collection point having one image based on a 90° view. Moreover, the study collected four images for each collection point to synthesize the panoramic streetscape images, which reached 268,100 images. The images were cleaned according to the availability of data, and all images were collected from Google Street View (GSV) to analyze the physical characteristics of the city and extract the factors of the urban environment. By appropriately establishing the parameters for image retrieval, the images captured both sides and frontal images of the street. This image acquisition covered all roads in Manhattan. Figure 2 provides a demonstration of the acquisition of the GSV images. Figure 2: Demonstration of the acquisition of GSV images. ## 2.2.2. Dataset for training the neural network model The study used datasets from Cityscape, ADE20K, and S-S-G-S to train the neural network model, a dataset open to researchers at the Mercedes-Benz R&D Center and Darmstadt University of Technology and published in the 2016 Clean Vehicle Rebate Project. The dataset was collected from 50 cities in Germany and nearby countries, including street scenes in spring, summer, and autumn. Different annotators with 96 and $98\%$ pixel consistencies repeatedly annotated the 30 selected data after omitting categories that could be annotated as unclear. The drawback was that the segmentation dataset contained 33 classes, whereas the validation dataset was composed of only 19 semantic segmentation classes because the data volume of a few classes was very sparse. The ADE20K dataset is intended for Scene Understanding, which was opened by the Massachusetts Institute of Technology (MIT) in 2016 and can be used, for instance, in semantics and part segmentation. Using image information for Scene Understanding and parsing, the dataset consists of 27,000 images from Scene Understanding (an open dataset released by Princeton University in 2010) and Places (an open dataset by MIT released in 2014). The ADE20K contains more than 3,000 object classes, which greatly compensates for the shortcomings of the Cityscape dataset. The S-S-G-S dataset was constructed by Zhang et al. [ 39] in 2022 and is mainly used for the analysis of urban vegetation communities. A neural network model trained using this dataset can classify and visualize the structure of urban street vegetation communities. S-S-G-S differs from the Cityscape and ADE20K datasets in that it is directed toward the analysis of urban greenery. The study mainly used the trained DeepLabV3+ neural network model to extract urban features at the micro level. ## 2.3. COVID-19 dataset The first confirmed case of COVID-19 was reported in Manhattan on March 1, 2020. At the time, the number of confirmed cases of COVID-19 in the entire United States was only 76. However, as of March 25, 2020, the number of confirmed COVID-19 cases in the United States spiked to 69,008, and the number of deaths reached 1,045, such that COVID-19 rampaged through the country at a rate of 10,000 per day for three consecutive days. However, according to the Centers for Disease Control and Prevention, nearly $50\%$ of all confirmed cases in the United States as of March 25, 2020, are in New York State, which establishes it as the epicenter of the outbreak. Out of the 33,006 cases diagnosed in New York State, 20,011 were derived from New York City. Over time and with the introduction of various restrictive policies and concerted national efforts to combat the outbreak, the spread of COVID-19 decelerated. Moreover, the outbreak appeared to be moving in a positive direction with the advent of COVID-19 vaccines. Moving forward to late December 2021, a variant of COVID-19 (omicron) is once again ravaging New York State with a record-breaking 21,908 cases detected in New York State on December 18, 2021, alone. Moreover, an alarming spike in cases was noted in several highly vaccinated neighborhoods in Manhattan. With 7-day positivity rates exceeding $10\%$ in more than 10 areas of New York City from December 10 to 16, 2021, Manhattan, once again, clearly became a hotbed of COVID-19 transmission. A total of 790.87 cases were identified per 100,000 people, and an extremely alarming rate was noted in specific Manhattan neighborhoods as of December 24, 2021. Greenwich Village and SoHo reported 2,850 confirmed cases per 100,000 people, and Chelsea reached 2,400 confirmed cases per 100,000 people. Nevertheless, no pandemic hotspot in the nation could compare to the dire outbreak in Greenwich Village. The COVID-19 case data in the study were derived from the publication by NYC Health (https://www1.nyc.gov/site/doh/index.page), which included cumulative totals since the COVID-19 outbreak in New York City. The Department of Health (DOH) defined the first case of COVID-19 as the one confirmed on February 29, 2020. In addition, the DOH recommended the avoidance of interpreting the daily changes in these files as 1-day data due to the discrepancy between the date of the event and the date of reporting. The internal division of the study area was divided according to the Modified ZIP Code Tabulation Areas (MODZCTA). NYC Health uses MODZCTA to report information according to geographic location. However, several issues emerge when mapping data reported based on ZIP codes because they do not designate a single area but a collection of points that compose the route of mail delivery. Moreover, a few buildings and non-residential areas were frequently assigned unique ZIP codes. To address these issues, the DOH uses ZIP Code Tabulation Areas (ZCTA) to convert ZIP codes into area units. The United States Census Bureau developed ZCTA geography to map data reported according to ZIP codes using ZCTA. MODZCTA geography combines census blocks with small populations to provide stable estimates of population size for rate calculations. The visualization is available on the website of NYC Health, which also open-sources the case data (https://github.com/nychealth/coronavirus-data#geography-zip-codes-and-zctas). In this manner, accessing appropriate data is easy for researchers. This study used Manhattan, New York in the United States as the study area and created a fishing network according to the 68 zones of MODZCTA to compare the mechanisms between the factors of the urban built environment and COVID-19 transmission in different zones and to investigate the reason Manhattan became the center of the pandemic many times during the outbreak. ## 3.1. Outcome variable: COVID-19 The outcome variables of the study were the number of confirmed and suspected cases of COVID-19 [COVID_CASE_COUNT (CCC)] and the incidence of confirmed and suspected cases of COVID-19 per 100,000 people [COVID_CASE_RATE (CCR)] in Manhattan, New York, United States. The difference between CCC and CCR is that CCR is a longer-term trend than CCC, and the relationship between the factors of urban built environment and COVID-19 under NPIs can be determined by comparing with CCC. According to MODZCTA, the Manhattan area of New York City, United States, was divided into 68 areas. After data filtering, the study identified 45 valid areas, and a fishing net was generated within the study area for a total of 1,551 grids. These grids will be used for analysis and spatial cells. The study calculated the CCC and CCR of the 45 independent areas and averaged them according to the fishing nets to reflect the overall number of cases in Manhattan. In addition, by calculating and visualizing the average of the number of valid POIs and environmental factors of the urban streetscape within the grids, the study intends to better establish the relationship of the urban built environment at the macro- and micro-levels to COVID-19. Figure 3 presents the visualization results of CCC and CCR in the study area. Figure 3: Visualization of CCC and CCR in the study area. (A) CCC Visualization in Manhattan, New York. (B) CCR Visualization in Manhattan, New York. ## 3.2. Macro-scale: Factors of urban built environment The study selected only three aspects, namely, density, diversity, and traffic, from the 5D's model framework [40, 41] for the evaluation of the factors of the built urban environment and the physical characteristics of the city at the macro level. In terms of density, the study used urban population density as an evaluation indicator. In the evaluation index of diversity, the study selected the data on POIs to measure the diversity of the urban environment. A POI consists of fine-grained data that comprehensively reflect accurate information on urban land use. The POI data used in the study were downloaded from OpenStreetMap (OSM) and reclassified according to the basic functions of the city after the data were screened, which included the omission of irrelevant, duplicate, and empty data. The study obtained 16,003 valid entity POIs for Manhattan, which were classified using C·M·E·P·R (Table 1). The C·M·E·P·R classification, as a method of classifying urban POIs on the basis of built-up characteristics, categorizes urban POIs according to urban functions such as commerce, healthcare, education, public services, and entertainment. Moreover, the POIs were classified according to C·M·E·P·R. The valid POIs were mapped to the fishnet grid of the study area, and the entropy score of the POI data per grid was calculated to determine diversity [42], which is calculated as follows: The formula pi is the proportion of the ith type of POI, and n is the total number of all POI types in the fishing grid. In turn, it better reflects the influence relationship between the urban built environment and COVID-19. Figure 4 provides the visualization results of macro factors of the urban built environment. ## 3.3. Micro-scale factors of urban built environment In this study, micro-level urban built environment specifically refers to the direct perception of the features of the urban landscape by pedestrians. Many studies demonstrate that computer vision combined with panoramic urban streetscape images can extract the features of the urban built environment and evaluate the urban built environment at the street level. This tendency proves that computer vision has gradually entered the scope of urban research. The current study selects the network model open-sourced by Chen et al. [ 43] in 2018, which pertains to a semantic segmentation network based on the DeepLabV3+ neural network model. The study made this selection for two reasons. The first is that the DeepLabV3+ neural network model is the latest version in the DeepLab series, which modifies VGG16 to introduce null convolution in DeepLabV1. The Atrous Spatial Pyramid Pooling (ASPP) model is designed in DeepLabV2; DeepLabV3 combines. The model proved its accuracy by outperforming mainstream deep learning algorithms (such as SegNet and PSPNet) in performance evaluation competitions such as the PascalVOC and Cityscapes benchmark tests in 2012. The second is that DeepLabV3+ features a better recognition effect compared with other mainstream deep learning models in the interpretation of urban scenes. The reason is that the model is designed for analyzing urban scenes, such that it exhibits certain advantages compared with those of other models when recognizing green structures in urban streets. The third is that the model uses DeepLabV3 as an encoder to generate the features of arbitrary dimensions using Atrous Convolution and adopts the ASPP strategy to use multiple effective sites with upsampling to achieve multiscale feature extraction. Moreover, it uses a cascade decoder to recover boundary detail information. Depthwise Separable Convolution is also used to reduce the number of parameters to further improve the accuracy and speed of the segmentation algorithm. This study selects the panoramic green view rate, the green structure of urban streets, buildings, roads, walls, and sky visibility to represent the micro-scale features of the urban built environment (Figure 5). Figure 5: Visualization of micro-scale factors of urban built environment. (A) PVGVI, (B) bus stop, (C) road, (D) macrophanerophytes, (E) bush, and (F) grass. ## 3.4. Statistical analysis The study conducted a four-step statistical analysis, namely: Where μ is the mean and σ is the standard deviation. Where R2 denotes goodness-of-fit or the determination coefficient of linear regression and describes the percentage of explanatory variables in the regression equation. The results indicate the absence of covariance for all independent variables, VIF values are <5, and all factors can be included in the linear regression model. Table 2 \| Variables (unit) \| Min \| Max \| Mean \| SD \| VIF (Z-score) \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Dependent variable \| Dependent variable \| Dependent variable \| Dependent variable \| Dependent variable \| Dependent variable \| \| COVID Case Count (CCC) (N) \| 0 \| 27410 \| 13814.951 \| 6340.799 \| \| \| COVID Case Rate (CCR) (N) \| 0 \| 54364.28 \| 30825.563 \| 7509.91 \| \| \| Independent variables \| Independent variables \| Independent variables \| Independent variables \| Independent variables \| Independent variables \| \| Macro-scale built environment \| Macro-scale built environment \| Macro-scale built environment \| Macro-scale built environment \| Macro-scale built environment \| Macro-scale built environment \| \| Public service (N) \| 0 \| 97 \| 3.437 \| 6.911 \| 1.296 \| \| Education (N) \| 0 \| 5 \| 0.23 \| 0.576 \| 1.029 \| \| Commercial (N) \| 0 \| 63 \| 5.801 \| 8.736 \| 1.417 \| \| Medical (N) \| 0 \| 18 \| 0.406 \| 1.055 \| 1.287 \| \| Recreation (N) \| 0 \| 14 \| 0.444 \| 1.016 \| 1.244 \| \| Airports (N) \| 0 \| 1 \| 0.001 \| 0.025 \| 1.001 \| \| Bus station (N) \| 0 \| 2 \| 0.008 \| 0.098 \| 1.041 \| \| Bus stop (N) \| 0 \| 8 \| 0.932 \| 1.275 \| 1.175 \| \| Ferry (N) \| 0 \| 2 \| 0.004 \| 0.072 \| 1.005 \| \| Railway (N) \| 0 \| 3 \| 0.102 \| 0.348 \| 1.144 \| \| Taxi (N) \| 0 \| 2 \| 0.006 \| 0.088 \| 1.043 \| \| POP (N) \| 0 \| 5722.99 \| 1042.108 \| 1033.205 \| 1.167 \| \| Micro-scale built environment \| Micro-scale built environment \| Micro-scale built environment \| Micro-scale built environment \| Micro-scale built environment \| Micro-scale built environment \| \| Sky View Factors (SVF) (%) \| 0 \| 0.279 \| 0.038 \| 0.042 \| 1.224 \| \| Building (%) \| 0 \| 0.45 \| 0.186 \| 0.109 \| 2.245 \| \| Road (%) \| 0 \| 0.504 \| 0.253 \| 0.131 \| 1.999 \| \| PVGVI (%) \| 0 \| 0.529 \| 0.111 \| 0.104 \| 1.791 \| \| Wall (%) \| 0 \| 0.459 \| 0.058 \| 0.077 \| 1.256 \| \| Street greening structure \| \| \| \| \| \| \| Macrophanerophytes (%) \| 0 \| 0.59 \| 0.114 \| 0.098 \| 1.667 \| \| Bush (%) \| 0 \| 0.399 \| 0.016 \| 0.032 \| 1.518 \| \| Grass (%) \| 0 \| 0.234 \| 0.012 \| 0.029 \| 1.496 \| Lastly, data at different levels with various dependent variables were included in the ordinary least squares (OLS) model. Furthermore, the study employed the White and BP tests to verify whether or not heteroscedasticity exists in the data, to test the original hypothesis that there was no heteroscedasticity in the model, to confirm whether or not the results rejected the original hypothesis, and to determine if there was a rejection of the original hypothesis that there was heteroscedasticity. To address these concerns, the study employed the robust regression method. ## 4.1. Pearson's correlation analysis This study used Pearson correlation analysis to examine the correlations between CCC and CCR and 12 macro-level urban built environment (Public, Education, Commercial, Medical, Recreation, Airports, Bus Station, Bus Stop, Ferry, Railway, Taxi, and POP) and 8 micro-level urban built environment (i.e., Sky, Building, Road, Wall, Macrophanerophyte, Grass, Bush, and PVGVI) in Manhattan, New York, USA, respectively, using Pearson's correlation coefficient (PCC) to indicate the strength of the correlations. Figures 6A, 7A depict the relationship between CCC and CCR and macro-level factors, where CCC presents a significant negative correlation with Public (PCC = −0.12, $p \leq 0.001$), Recreation (PCC = −0.16, $p \leq 0.001$), and Ferry (PCC = −0.077, $p \leq 0.01$). Moreover, the study observes a significant negative correlation between Public and Recreation. Both correlations indicate significance at the 0.001 level. Education (PCC = 0.11, $p \leq 0.001$), Bus Stop (PCC = 0.13, $p \leq 0.001$), and POP (PCC = 0.30, $p \leq 0.001$) displayed significant positive correlations with CCC at the 0.001 level of significance. Although the study noted no correlation among Commercial, Medical, Airports, Bus Station, Railway, Tax, and CCC, their PCC values are close to 0, and all p-values are >0.05. Figure 4C illustrates that Public, Education, Commercial, Medical, Bus Stop, Railway, Taxi, and POP have significant positive correlations with CCR, where Commercial (PCC = 0.26, $p \leq 0.001$), Medical (PCC = 0.11, $p \leq 0.001$), Bus Stop (PCC = 0.16, $p \leq 0.001$), Railway (PCC = 0.10, $p \leq 0.001$), and POP (PCC = 0.092, $p \leq 0.001$) demonstrated showed significance at the 0.001 level, which indicate a significant positive correlation with CCC. Lastly, the study found no correlation among Recreation, Airports, Ferry, and CCR. Figure 6: Correlation coefficient between urban built environment and CCC Pearson. (A) Correlation coefficient between macro urban built environment and CCC Pearson. (B) Correlation coefficient between micro urban built environment and CCC Pearson. p < 0.05, p < 0.01, p < 0.001. Figure 7: Correlation coefficient between urban built environment and CCR Pearson. (A) Correlation coefficient between macro urban built environment and CCR Pearson. (B) Correlation coefficient between micro urban built environment and CCR Pearson. p < 0.05, p < 0.01, p < 0.001. Figures 6B, 7B present the relationship of CCC and CCR to micro-level factors, where positive correlations were noted among Building (PCC = 0.15, $p \leq 0.001$), Road (PCC = 0.26, $p \leq 0.001$), and CCC, and all of them show. The study found negative correlations among Wall, Grass, Bush, PVGVI, and CCC, where Grass (PCC = −0.18, $p \leq 0.001$) and Bush (PCC = −0.15, $p \leq 0.001$) at the 0.001 level of significance, which indicates a significant negative correlation with CCC, whereas no correlation was found between Sky and Macrophanerophytes to CCC. Figure 4D points to a positive correlation among Building, Road, Wall, and CCR at the 0.001 level of significance, which indicates a significant negative correlation with CCR. The correlation between Macrophanerophytes, Grass, Bush, PVGVI, and CCR was all negative at the 0.001 level of significance, among which the PCC value of Macrophanerophytes was −0.5, which extremely exceeded the other variables and indicates a significant negative correlation with CCR. ## 4.2. Robust regression model The OLS linear regression of CCC and CCR as outcome variables resulted in four models. Macro- and micro-level factors of the urban built environment were separately included as variables in the models to determine the relationship of CCC and CCR to the independent variables at different levels. The equation for OLS linear regression is as follows: Where Y is the dependent variable, X denotes the matrix of explanatory variables, β represents the vector of coefficients, and ε is the vector of random error terms. The variables were included in the OLS model for the White and BP tests. Table 3 presents the results. In the case of heteroscedasticity, the study conducted the White and BP tests to verify the original hypothesis, that is, no heteroscedasticity exists in the model. Table 3 illustrates that both tests reject the original hypothesis at $p \leq 0.05$, which indicates that heteroscedasticity exists in the model. Table 3 \| White heteroscedasticity test \| White heteroscedasticity test.1 \| BP heteroscedasticity test \| BP heteroscedasticity test.1 \| \| --- \| --- \| --- \| --- \| \| X 2 \| P \| X 2 \| P \| \| White test and BP test results of CCC and macro urban built environment \| White test and BP test results of CCC and macro urban built environment \| White test and BP test results of CCC and macro urban built environment \| White test and BP test results of CCC and macro urban built environment \| \| 91.844 \| 0.034 \| 24.380 \| 0.018 \| \| White test and BP test results of CCC and micro-level urban built environment \| White test and BP test results of CCC and micro-level urban built environment \| White test and BP test results of CCC and micro-level urban built environment \| White test and BP test results of CCC and micro-level urban built environment \| \| 182.923 \| 0.000 \| 120.011 \| 0.000 \| \| Results of white test and BP test of CCR and macro urban built environment \| Results of white test and BP test of CCR and macro urban built environment \| Results of white test and BP test of CCR and macro urban built environment \| Results of white test and BP test of CCR and macro urban built environment \| \| 182.923 \| 0.034 \| 120.011 \| 0.000 \| \| Results of white test and BP test of CCR and micro-level urban built environment \| Results of white test and BP test of CCR and micro-level urban built environment \| Results of white test and BP test of CCR and micro-level urban built environment \| Results of white test and BP test of CCR and micro-level urban built environment \| \| 435.371 \| 0.000 \| 273.668 \| 0.000 \| Table 3 suggests that heteroscedasticity exists in the regression data, and the conclusions obtained by the commonly used OLS regression estimation method may be biased because it considers the minimized residual sum of squares as a criterion. Therefore, it also considers anomalous data. In this case in the model regression considered for robust regression analysis (M-estimation), the study uses the Huber robust method with the following formula: where a real function ρ defined in a one-dimensional Euclidean space R is selected for the independent identically distributed equal precision model, such that aiT denotes the row vector of the design matrix; X is the extreme value solution; and P represents the weight of the corresponding observation or observation error. Tables 4, 5 depict Models 1 and 2, respectively. The difference between the models is the use of CCC and CCR as the dependent variables, respectively. CCR can be used to illustrate the long-term trend of COVID-19, which could help in analyzing the impact of NPIs on the relationship between the urban built environment and COVID-19. Alternatively, CCC can be used for analyzing the relationship between the impact of a pure urban built environment and COVID-19. Table 4 depicts the correlation between Model 1 with CCC as the dependent variable and 20 factors of the urban built environment as the independent variables. It uses robust regression analysis (M-estimation) to construct the correlation between the variables of urban built environment and COVID-19. The study finds that the macro-level factors, Education, Commercial, POP, and Bus Stop, exert a significant positive influence on the relationship between the urban built environment and COVID-19. The correlation coefficient of POP was 0.297, which exceeded all other variables. In particular, Commercial is the only factor that exerts a significant positive effect on CCC and CCR as the dependent variables for both regression models. The regression coefficient of Public is −0.255 with a p-value of 0.004, which is more significant than the other variables. The micro-level factors that displayed significant negative effects in the micro-urban built environment were significantly higher; Building, Wall, Grass, Bush, and PVGVI exerted significant negative effects on CCC, where Grass obtained a regression coefficient of −0.357 and a p-value of 0.000, which were higher than those of the other variables in the same model in terms of significance and regression coefficient. PVGVI and Grass exhibited a significant negative effect relationship for Models 1 and 2. The regression coefficient for Grass was higher in Model 1 than that in Model 2; however, the significance of both Models is the same (p-values = 0.000). Road and Macrophanerophytes exerted a significant positive effect on CCC; both p-values were 0.000, which is higher than the other variables in terms of significance, except for Grass, which is equal. Table 5 presents the results of the robust regression analysis for Model 2 with CCR as the dependent variable and the 20 urban built environment factors as the independent variables. The finding indicates that Public and Commercial show a significant positive relationship with CCR at the macro level, whereas POP indicates a significant negative relationship with CCR. Public and POP produced the opposite results for both models (Model 1: Public: regression coefficient = −0.255, POP: regression coefficient = 0.297; Model 2: Public: regression coefficient = 0.07, POP: regression coefficient = −0.088). The micro-level factors Sky, Building, and Wall presented a significant positive relationship with CCR, whereas Macrophanerophytes, Grass, and PVGVI pointed to a significant negative relationship with CCR. PVGVI is more significant in Model 2 than it was in Model 1 (Model 1: PVGVIp = 0.023; Model 2: PVGVIp = 0.000). Macrophanerophytes present opposite results in Models 1 and 2 (Model 1: regression coefficient = 0.213; Model 2: regression coefficient = −0.047). ## 5. Discussion This study investigated the relationship between the factors of the urban built environment and COVID-19 using robust regression analysis (M-estimation) based on solving the heteroscedasticity of the OLS regression model. The study categorized urban built environment into two dimensions, namely, macro and micro, in two urban spatial dimensions, where macro-level factors include variables related to urban traffic, and micro-level factors pertain to urban green structures. ## 5.1. COVID-19 and urban built environment The study used the relationship between the number (CCC) and incidence (CCR) of confirmed and suspected cases of COVID-19 per 100,000 people in Manhattan, United States, as an entry point for the factors of the urban built environment. However, in the regression analysis with CCR as the dependent variable, POP exhibited a significant negative effect on CCR. In analyzing this entirely contradictory result, the study considered the effect of Commercial, which exerted a significant positive effect on CCC and CCR but with different factors at 0.116 and 0.041, respectively. In other words, residents can obtain necessities in a small area after the outbreak of a potential pandemic, and NPIs are better compared with those in areas with low population density. Residents in these areas must travel long distances to obtain essential resources, where long-distance travel implies increased chances of contact with strangers and COVID-19 infection. In summary: (i) high population density increases the likelihood of human contact, which facilitates the spread of the virus. However, with the implementation of NPIs, residents could only move within a small area; thus, the virus could not spread among areas. ( ii) Areas with high-density populations typically have relatively well-developed infrastructure to provide convenient and timely treatment for residents, which, thereby, inhibits the spread of NPI [26, 29]. In particular, under strict NPIs, the outdoor activities of residents are restricted, which effectively inhibits the spread of the virus in high-density areas [24]. However, at the CCC level, the government for areas with high population density and high commercial activities performs better in terms of pandemic control and detection than did areas with low population density with a higher detection rate than that of areas with low population density. This finding results in a higher number of confirmed and suspected cases compared with those of areas with low population density. At the same time, control and control efforts are correspondingly lower due to the lower population density, which results in an increased number of cases without data. The situation of no data collection. The number of bus stops tends to be proportional to population density; the higher the population density, the higher the number of bus stops. Essentially, bus stops are places where urban residents are most likely to come into contact with strangers. A high frequency of contact with strangers implies an increased chance of infection. In general, public transportation infrastructure that increases population contact is considered a key factor in the spread of infectious diseases [25]. Thus, a range of effective measures should be taken to limit the spread of disease in public transport, including limiting passenger density, increasing the frequency of services, and reserving tickets. Other low-carbon and environmentally friendly active transportation modes, such as walking and bicycling, should also be encouraged. Similarly, at the public level, the results of CCC and CCR indicate a clear contradiction. From the CCR level, the higher the use of public facilities, the higher the probability of exposure to unfamiliar environments. Thus, the virus is likely to spread through public facilities before the introduction of corresponding NPIs. The number of public facilities in areas with high population density far exceeds that in areas with low population density, such that corresponding transmission rates and probability of transmission are also higher. On the contrary, at the CCC level, the frequency of the use of public facilities is suppressed due to NPIs, and residents will voluntarily reduce their frequency of use of public facilities when they are aware of a potential pandemic. This scenario indirectly leads to a negative correlation between CCC and the Public with a regression index of −0.235 over other factors. Schools tend to be places where pedestrian traffic is high, no less than in commercial areas, and the interaction between students and teachers may accelerate the spread of the virus. When students are infected with the virus at school, NCPI can easily infect family members through parent–child interaction, and the spread of the virus within colleges and universities is typically difficult to reasonably control. As the implementation of NPIs led to school closure, the correlation between schools and CCR became negligible; instead, many schools were requisitioned for the isolation of patients, which exerted a positive effect on the control of the outbreak after lockdowns. Alfano et al. [ 44] demonstrated that the premature opening of schools increased the number of COVID-19 cases in Italy. This result suggests that during an outbreak, the government should implement strict NPIs in schools while ensuring equity in education. The higher the PVGVI, the farther away from the city center, the lower the population density, and the less space and medium for virus transmission and corresponding inhibitory effects on virus transmission. Macrophanerophytes exerted a significant positive effect on CCC, whereas Bush and Grass exerted a significant negative effect on CCC when analyzed from the perspective of the green structure of urban streets. The reason for this phenomenon may be that in densely populated areas with developed commercial activities, the green structure is relatively homogeneous and shows a single-tree state. Conversely, areas with a rich green structure have correspondingly low population density and more homogeneous commercial activities, which can be analyzed in combination with macro-level POP and Commercial. ## 5.2. Research values A series of recommendations for the results of the study have the following applications: (i) they can be applied at the level of prevention of widespread spread of COVID-19 in cities to minimize the risk of infection and the rate of virus transmission among urban residents by exploring the mechanisms of influence of the built environment and COVID-19. Effective control of virus transmission was achieved at the early stage of the outbreak. ( ii) Based on the results of the study, government officials and policy makers can better formulate more reasonable NPI policies to prevent widespread infection and cross-infection and reduce the risk of infection among urban residents, while ensuring the wellbeing, health and comfort of urban residents. ( iii) The study uses Google Street View panoramic street view images to extract and quantify urban micro built environment factors from the macro built environment and the micro built environment, respectively, to explore the impact of COVID-19 at the urban street level, and the results provide a data base for future urban renewal. This enables cities to play a more important role in facing the trend of COVID-19 epidemic normalization. ## 5.3. Research limitations This study has its limitations. First, the data published by NYC Health are divided according to the MODZCTA, where individual buildings are designated unique zip codes in several instances. This tendency can exert a confounding effect on the data, and although the study screened a few of the confounding factors at certain levels, this data-level confounding continues to exist. Moreover, although the study sample was expanded according to fishnet divisions, the original sample only comprises 45 areas, which is not representative of all areas in the United States. Second, other demographic data for Manhattan were not available or could not be specifically mapped within each study area, such as household income structure, demographics, gender, underlying disease status, occupation, and ethnic composition, which have been noted in previous studies to be associated with 2019 coronavirus disease transmission. At the same time, within the time point of the COVID-19 pandemic, the lives of residents were frequently restricted by various NPIs, which resulted in extremely complex and confusing life activities and social relationships. Thus, the study selected only 20 variables, which indicates the exclusion of other potential variables such as the density of foot traffic in the region. Previous studies demonstrated that individual behaviors exerted an effect on the spread of COVID-19; however, such variables are statistically unavailable, relatively difficult to obtain, and more difficult to collect in the field due to various policy restrictions imposed by NPIs. The absence of such variables may have led to certain anomalies in the results of the study. Moreover, the effect of spatial autocorrelation cannot be avoided despite the multilevel and multidimensional considerations. Thus, future studies should consider additional aspects and potential variables to explore the relationship between the factors of the urban built environment and COVID-19. This, data on COVID-19 published by NYC Health provided substantial support to various urban studies on COVID-19. However, the published information on the number of cases is, in fact, incomplete due to the lack of statistical data on the number of cases due to the current pandemic policy implemented in the United States. Thus, certain individuals contracted COVID-19 but displayed no symptoms (asymptomatic) due to the lack of assurance of the detection rates of COVID-19 in the population. Moreover, individuals with the infection were not sampled for nucleic acids; thus, they remained unaware of their COVID-19 infection, which rendered their network and range of activities and transmission of the virus virtually uncontrollable and unavoidable. Possible non-linear effects of the variables in this study. The starting point of the robust regression is still based on the processing method of linear data, but the principle of adopting the method should be considered when processing the experimental data, and if the data have non-linear effects, the experimental data can be made permutation substitution so that they are transformed into a linear functional relationship for the test. In future studies, we will apply multiscale geographically weighted regression models with added potential factors to calibrate the existing models for further accuracy in the analysis given that more data are available at the city level. ## 6. Conclusion The study draws preliminary conclusions on the relationship between the urban built environment and COVID-19 transmission, which focused on the relationship between CCC and CCR as the independent variables and the influence of the urban built environment. The correlation between the urban built environment and COVID-19 transmission was determined using Robust regression analysis (M-estimation). The major findings are summarized as follows: The current COVID-19 situation remains severe, and predicting the direction of the pandemic is difficult. To cope with more severe pandemic situations, this study provides several recommendations for urban built environments in the context of its results. First, the government should provide easy access to essential resources for urban residents within a controlled range, reduce the frequency of long-distance travel, save travel costs, reduce unnecessary human contact, and control the medium of transmission to reduce the speed and efficiency of the virus transmission. Second, for areas with high population density and commercial activities, strict NPIs should be implemented, such that if a potential outbreak occurs, then the area can quickly and adequately mobilize favorable resources to effectively control the outbreak. Third, the frequency of use of public facilities should be controlled. Although urban public transportation is an important part of the future low-carbon city, it continues to play an important role in the spread of the virus at this stage. In addition, the number of passengers should be controlled, their health status should be strictly tested, and safe social distancing should be observed to effectively control the spread of the virus. The government can promote and introduce incentives to encourage residents to use other modes of travel. Fourth, schools or educational settings were found to be at risk during outbreaks of COVID-19 due to their dense population and foot traffic; thus, a series of strong measures should be taken such as distance teaching or a limited number of people in schools. The recommendations may serve as a reference for solutions for other cities at the level of controlling the transmission and spread of the virus. Meanwhile, the findings may provide valid suggestions for curbing potential outbreaks of respiratory diseases. However, the applicability of the variables is limited and does not reflect the regional economic level, demographic, and other sociodemographic characteristics of the city due to the limitations of this study and the data sources. Therefore, the generalizability of the results should be carefully considered. ## Author's note The S-G-S-S datasets used in this study can be downloaded and used from our GitHub site (https://github.com/muteisdope/S-G-S-S-Dataset.git), allowing users to modify, upload, and optimize the datasets. We will continue to upload new datasets and optimize the datasets in the future. Our research team based on Python language, Pytorch deep learning framework, DeepLabV3+ neural network used in our research, the code can be downloaded from our GitHub website (https://github.com/muteisdope/Model.git). ## Data availability statement The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/supplementary material. ## Author contributions LZ and XH: conceptualization and writing—original draft. LZ and JW: resources. LW and JW: supervision. LW: validation. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher's note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## References 1. Guan W, Ni Z, Hu Y, Liang W, Ou C, He J. Clinical characteristics of coronavirus disease 2019 in China. N Engl J Med. (2020) 382 1708-20. DOI: 10.1056/NEJMoa2002032 2. Hamidi S, Ewing R, Sabouri S. Longitudinal analyses of the relationship between development density and the COVID-19 morbidity and mortality rates: early evidence from 1,165 metropolitan counties in the United States. Health Place. (2020) 64 102378. DOI: 10.1016/j.healthplace.2020.102378 3. Sharifi A, Khavarian-Garmsir AR. The COVID-19 pandemic: impacts on cities and major lessons for urban planning, design, and management. Sci Total Environ. (2020) 749 142391. DOI: 10.1016/j.scitotenv.2020.142391 4. Lee W, Kim H, Choi HM, Heo S, Fong KC, Yang J. Urban environments and COVID-19 in three Eastern states of the United States. Sci Total Environ. (2021) 779 146334. DOI: 10.1016/j.scitotenv.2021.146334 5. Gao Z, Wang S, Gu J, Gu C, Liu R. A community-level study on COVID-19 transmission and policy interventions in Wuhan, China. Cities. (2022) 127 103745. DOI: 10.1016/j.cities.2022.103745 6. Wang D, Hu B, Hu C, Zhu F, Liu X, Zhang J. Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus–infected pneumonia in Wuhan, China. JAMA. (2020) 323 1061. DOI: 10.1001/jama.2020.1585 7. Mollalo A, Vahedi B, Rivera KM. GIS-based spatial modeling of COVID-19 incidence rate in the continental United States. Sci Total Environ. (2020) 728 138884. DOI: 10.1016/j.scitotenv.2020.138884 8. Aycock L, Chen X. Levels of economic development and the spread of coronavirus disease 2019 (COVID-19) in 50 US states and territories and 28 European countries: an association analysis of aggregated data. Glob Health J. (2021) 5 24-30. DOI: 10.1016/j.glohj.2021.02.006 9. Oshakbayev K, Zhankalova Z, Gazaliyeva M, Mustafin K, Bedelbayeva G, Dukenbayeva B. Association between COVID-19 morbidity, mortality, and gross domestic product, overweight/obesity, non-communicable diseases, vaccination rate: a cross-sectional study. J Infect Public Health. (2022) 15 255-60. DOI: 10.1016/j.jiph.2022.01.009 10. Wang L, Zhang S, Yang Z, Zhao Z, Moudon AV, Feng H. What county-level factors influence COVID-19 incidence in the United States? Findings from the first wave of the pandemic. Cities. (2021) 118 103396. DOI: 10.1016/j.cities.2021.103396 11. Das A, Ghosh S, Das K, Basu T, Dutta I, Das M. Living environment matters: unravelling the spatial clustering of COVID-19 hotspots in Kolkata megacity, India. Sust Cities Soc. (2021) 65 102577. DOI: 10.1016/j.scs.2020.102577 12. Ahmed F, Ahmed N, Pissarides C, Stiglitz J. Why inequality could spread COVID-19. Lancet Public Health. (2020) 5 e240. DOI: 10.1016/S2468-2667(20)30085-2 13. DuPre NC, Karimi S, Zhang CH, Blair L, Gupta A, Alharbi LM. Sci Total Environ. (2021). DOI: 10.1016/j.scitotenv.2021.147495 14. Onder G, Rezza G, Brusaferro S. Case-fatality rate and characteristics of patients dying in relation to COVID-19 in Italy. JAMA. (2020) 323 1775-6. DOI: 10.1001/jama.2020.4683 15. Zhou F, Yu T, Du R, Fan G, Liu Y, Liu Z. Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study. Lancet. (2020) 395 1054-62. DOI: 10.1016/S0140-6736(20)30566-3 16. Anderson RM, Heesterbeek H, Klinkenberg D, Hollingsworth TD. How will country-based mitigation measures influence the course of the COVID-19 pandemic?. Lancet. (2020) 395 931-4. DOI: 10.1016/S0140-6736(20)30567-5 17. Ficetola GF, Rubolini D. Containment measures limit environmental effects on COVID-19 early outbreak dynamics. Sci Total Environ. (2021) 761 144432. DOI: 10.1016/j.scitotenv.2020.144432 18. Paköz MZ, Is M. Rethinking urban density, vitality and healthy environment in the post-pandemic city: the case of Istanbul. Cities. (2022) 124 103598. DOI: 10.1016/j.cities.2022.103598 19. Allam Z, Jones D. Pandemic stricken cities on lockdown. Where are our planning and design professionals [now, then and into the future]?. Land Use Policy. (2020) 97 104805. DOI: 10.1016/j.landusepol.2020.104805 20. Mouratidis K, Yiannakou A. COVID-19 and urban planning: built environment, health, and well-being in Greek cities before and during the pandemic. Cities. (2022) 121 103491. DOI: 10.1016/j.cities.2021.103491 21. Mouratidis K. COVID-19 and the compact city: implications for well-being and sustainable urban planning. Sci Total Environ. (2022) 811 152332. DOI: 10.1016/j.scitotenv.2021.152332 22. Cordes J, Castro MC. Spatial analysis of COVID-19 clusters and contextual factors in New York City. Spatial Spatio Temp Epidemiol. (2020) 34 100355. DOI: 10.1016/j.sste.2020.100355 23. Nguimkeu P, Tadadjeu S. Why is the number of COVID-19 cases lower than expected in Sub-Saharan Africa? A cross-sectional analysis of the role of demographic and geographic factors. World Dev. (2021) 138 105251. DOI: 10.1016/j.worlddev.2020.105251 24. Khavarian-Garmsir AR, Sharifi A, Moradpour N. Are high-density districts more vulnerable to the COVID-19 pandemic?. Sust Cities Soc. (2021) 70 102911. DOI: 10.1016/j.scs.2021.102911 25. AbouKorin SAA, Han H, Mahran MGN. Role of urban planning characteristics in forming pandemic resilient cities – case study of Covid-19 impacts on European cities within England, Germany and Italy. Cities. (2021) 118 103324. DOI: 10.1016/j.cities.2021.103324 26. Hamidi S, Sabouri S, Ewing R. Does density aggravate the COVID-19 pandemic?. J Am Plann Assoc. (2020) 86 495-509. DOI: 10.1080/01944363.2020.1777891 27. Lin C, Lau AKH, Fung JCH, Guo C, Chan JWM, Yeung DW. A mechanism-based parameterisation scheme to investigate the association between transmission rate of COVID-19 and meteorological factors on plains in China. Sci Total Environ. (2020) 737 140348. DOI: 10.1016/j.scitotenv.2020.140348 28. Boterman WR. Urban-rural polarisation in times of the corona outbreak? The early demographic and geographic patterns of the SARS-CoV-2 pandemic in the Netherlands. Tijd voor Econ Soc Geog. (2020) 111 513-29. DOI: 10.1111/tesg.12437 29. Liu L. Emerging study on the transmission of the Novel Coronavirus (COVID-19) from urban perspective: evidence from China. Cities. (2020) 103 102759. DOI: 10.1016/j.cities.2020.102759 30. Mouratidis K. How COVID-19 reshaped quality of life in cities: a synthesis and implications for urban planning. Land Use Policy. (2021) 111 105772. DOI: 10.1016/j.landusepol.2021.105772 31. Figueroa JF, Wadhera RK, Mehtsun WT, Riley K, Phelan J, Jha AK. Association of race, ethnicity, and community-level factors with COVID-19 cases and deaths across U.S. counties. Healthcare. (2021) 9 100495. DOI: 10.1016/j.hjdsi.2020.100495 32. Liu C, Liu Z, Guan C. The impacts of the built environment on the incidence rate of COVID-19: a case study of King County, Washington. Sust Cities Soc. (2021) 74 103144. DOI: 10.1016/j.scs.2021.103144 33. Guida C, Carpentieri G. Quality of life in the urban environment and primary health services for the elderly during the Covid-19 pandemic: an application to the city of Milan (Italy). Cities. (2021) 110 103038. DOI: 10.1016/j.cities.2020.103038 34. Lu Y, Chen L, Liu X, Yang Y, Sullivan WC, Xu W. Green spaces mitigate racial disparity of health: a higher ratio of green spaces indicates a lower racial disparity in SARS-CoV-2 infection rates in the USA. Environ Int. (2021) 152 106465. DOI: 10.1016/j.envint.2021.106465 35. Zhang Y, Chen N, Du W, Li Y, Zheng X. Multi-source sensor based urban habitat and resident health sensing: a case study of Wuhan, China. Build Environ. (2021) 198 107883. DOI: 10.1016/j.buildenv.2021.107883 36. Wang J. Vision of China's future urban construction reform: in the perspective of comprehensive prevention and control for multi disasters. Sust Cities Soc. (2021) 64 102511. DOI: 10.1016/j.scs.2020.102511 37. Li B, Peng Y, He H, Wang M, Feng T. Built environment and early infection of COVID-19 in urban districts: a case study of Huangzhou. Sust Cities Soc. (2021) 66 102685. DOI: 10.1016/j.scs.2020.102685 38. Wang J, Zeng F, Tang H, Wang J, Xing L. Correlations between the urban built environmental factors and the spatial distribution at the community level in the reported COVID-19 samples: a case study of Wuhan. Cities. (2022) 129 103932. DOI: 10.1016/j.cities.2022.103932 39. Zhang L, Wang L, Wu J, Li P, Dong J, Wang T. Decoding urban green spaces: deep learning and google street view measure green structures. SSRN J. (2022). DOI: 10.2139/ssrn.4180331 40. Kang C-D. The S + 5Ds: spatial access to pedestrian environments and walking in Seoul, Korea. Cities. (2018) 77 130-41. DOI: 10.1016/j.cities.2018.01.019 41. Lu Y, Chen L, Yang Y, Gou Z. The association of built environment and physical activity in older adults: using a citywide public housing scheme to reduce residential self-selection bias. IJERPH. (2018) 15 1973. DOI: 10.3390/ijerph15091973 42. Chen L, Lu Y, Ye Y, Xiao Y, Yang L. Examining the association between the built environment and pedestrian volume using street view images. Cities. (2022) 127 103734. DOI: 10.1016/j.cities.2022.103734 43. Chen L-C, Zhu, Y, Papandreou, G, Schroff, F, Adam, H,. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation.. (2018) 44. Alfano V, Ercolano S, Cicatiello L. School openings and the COVID-19 outbreak in Italy. A provincial-level analysis using the synthetic control method. Health Policy. (2021) 125 1200-7. DOI: 10.1016/j.healthpol.2021.06.010
--- title: 'Identifying the prevalence and correlates of multimorbidity in middle-aged men and women: a cross-sectional population-based study in four African countries' authors: - Lisa K Micklesfield - Richard Munthali - Godfred Agongo - Gershim Asiki - Palwende Boua - Solomon SR Choma - Nigel J Crowther - June Fabian - Francesc Xavier Gómez-Olivé - Chodziwadziwa Kabudula - Eric Maimela - Shukri F Mohamed - Engelbert A Nonterah - Frederick J Raal - Hermann Sorgho - Furahini D Tluway - Alisha N Wade - Shane A Norris - Michele Ramsay journal: BMJ Open year: 2023 pmcid: PMC10016250 doi: 10.1136/bmjopen-2022-067788 license: CC BY 4.0 --- # Identifying the prevalence and correlates of multimorbidity in middle-aged men and women: a cross-sectional population-based study in four African countries ## Abstract ### Objectives To determine the prevalence of multimorbidity, to identify which chronic conditions cluster together and to identify factors associated with a greater risk for multimorbidity in sub-Saharan Africa (SSA). ### Design Cross-sectional, multicentre, population-based study. ### Setting Six urban and rural communities in four sub-Saharan African countries. ### Participants Men ($$n = 4808$$) and women ($$n = 5892$$) between the ages of 40 and 60 years from the AWI-Gen study. ### Measures Sociodemographic and anthropometric data, and multimorbidity as defined by the presence of two or more of the following conditions: HIV infection, cardiovascular disease, chronic kidney disease, asthma, diabetes, dyslipidaemia, hypertension. ### Results Multimorbidity prevalence was higher in women compared with men ($47.2\%$ vs $35\%$), and higher in South African men and women compared with their East and West African counterparts. The most common disease combination at all sites was dyslipidaemia and hypertension, with this combination being more prevalent in South African women than any single disease ($25\%$ vs $21.6\%$). Age and body mass index were associated with a higher risk of multimorbidity in men and women; however, lifestyle correlates such as smoking and physical activity were different between the sexes. ### Conclusions The high prevalence of multimorbidity in middle-aged adults in SSA is of concern, with women currently at higher risk. This prevalence is expected to increase in men, as well as in the East and West African region with the ongoing epidemiological transition. Identifying common disease clusters and correlates of multimorbidity is critical to providing effective interventions. ## Introduction Sub-Saharan Africa (SSA) is experiencing the highest rate of urbanisation globally and together with an increase in life expectancy must prepare for the prevalence of non-communicable diseases (NCDs) to increase further.1 Multimorbidity, the co-occurrence of two or more chronic diseases in one individual, is common, challenging the affected individual, their attending healthcare professionals and an overstretched health system. A recent systematic review highlighted a paucity of multimorbidity costing studies from low-income and middle-income countries (LMICs).2 When describing global patterns of multimorbidity, Afshar et al reported a positive but non-linear association between country GDP and multimorbidity prevalence, and also identified an inverse association between multimorbidity and socioeconomic status (SES) in countries with the highest GDP, with the gradient sometimes reversed in countries with lower GDP.3 Understanding the social and structural forces that drive the clustering of diseases, thereby exacerbating their impact on disease outcomes, is defined as syndemics4 and is critical to our understanding of multimorbidity. Recent data from LMICs, including South Africa, have been reported in a scoping review of NCD multimorbidity, which ranged in prevalence from $3.2\%$ to $90.5\%$, and reported that $95.3\%$ of the studies found female sex to be a risk factor for multimorbidity.5 The studies included in this review used various chronic diseases comprising the multimorbidity profile, and different diagnostic criteria for each chronic disease, highlighting the need for a consensus on the definition of multimorbidity and the core conditions that it is comprised of. Further, the inclusion of chronic infectious diseases such as HIV and tuberculosis should also be considered when quantifying the burden of multimorbidity but may be more relevant in some settings than others. There is a dearth of multimorbidity prevalence data from Africa, with studies reporting data from single countries such as South Africa,6 7 Ghana,8 Burkina Faso9 and Kenya.10 Although the influence of age and SES on multimorbidity prevalence is well recognised globally,3 5 11 12 several African studies have also identified associations between lifestyle factors, mental health and multimorbidity.7 10 12 Comparing and contrasting different African settings and identifying which diseases cluster together and the factors associated with this clustering will assist in the design of effective interventions. The double burden of non-communicable and infectious diseases as well as the impact of urbanisation on disease risk highlights the need to undertake multimorbidity research in Africa, and the heterogeneity in the epidemiological and nutrition transition, and disease burden of different African countries, provides an opportunity to compare and contrast the correlates and prevalence of multimorbidity across different African settings. The aim of this study was to determine the prevalence of multimorbidity, to identify which chronic conditions cluster together and to identify factors associated with a greater risk for multimorbidity in six rural and urban communities in four SSA countries. This study will contribute to the evidence base on multimorbidity by providing prevalence data across different African settings and identifying common disease clusters to potentially understand the pathogenesis of different diseases. This will inform the design of more effective interventions and provide formative data on multimorbidity for policy-makers in the planning and implementation of more effective health systems with particular relevance to other LMICs. ## Study population Data included in this study are from the Africa Wits-INDEPTH (University of the Witwatersrand, Johannesburg, and the International Network for the Demographic Evaluation of Populations and Their Health) partnership for Genomic Studies (AWI-Gen),13 which is a National Institutes of Health-funded Collaborative Centre of the Human Heredity and Health in Africa (H3Africa) Consortium. AWI-Gen is a population-based cross-sectional study of adults and includes six participating urban and rural centres in four SSA countries. West Africa included two countries, Ghana (Navrongo) and Burkina Faso (Nanoro); East Africa included Kenya (Nairobi); and South Africa had three study sites (Soweto, Agincourt and Dikgale). The details of the recruitment strategy as well as other data collected are described in Ali et al.14 For this study, we only included participants from 40 to 60 years ($$n = 10$$ 700). ## Sociodemographic and anthropometric data Standard structured AWI-Gen questionnaires were administered by trained field staff with some country and site-specific modifications to suit their context.14 Data on self-reported sociodemographics (age, highest level of education attained, partnership status, employment status) were collected. Each study participant was assigned to an SES quintile, which was determined for each study site by categorising factor scores that were predicted from a principal component analysis of the number of household assets. Alcohol consumption was categorised as never, current problematic, current non-problematic or former consumer,15 but was not available for the Soweto women. Smoking of tobacco products was categorised as never, current or previous smoker. Total moderate–vigorous intensity physical activity (MVPA) was calculated as minutes per week from the accumulation of occupation, walking for travel and leisure time activity collected by self-report using the Global Physical Activity Questionnaire.16 Weight and height were used to calculate body mass index (BMI: weight (kg)/height (m2)). ## Multimorbidity We defined multimorbidity as the presence of two or more of the following seven conditions (HIV infection, cardiovascular disease (CVD), chronic kidney disease (CKD), asthma, diabetes, dyslipidaemia and hypertension) for all sites except for women from Soweto, where data was only available for four of the conditions (HIV infection, diabetes, dyslipidaemia and hypertension). Obesity was not included as one of the conditions as in the context of this paper it is considered an NCD risk factor rather than a disease. For this reason, BMI was included as an independent variable in the multinomial regression. HIV status was self-reported although participants in South Africa and Kenya were offered a voluntary government-approved rapid HIV test. Due to the low prevalence of HIV in Burkina Faso and Ghana, participants were not offered HIV tests. Hypertension was defined as systolic blood pressure≥140 mm Hg and/or diastolic blood pressure≥90 mm Hg, in line with the seventh report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure, or if the participant was taking hypertension medication.17 Diabetes was defined using the WHO criteria, which are the presence of a previous diagnosis of diabetes by a healthcare professional or fasting blood glucose≥7 mmol/L or random glucose≥11.1 mmol/L, or on diabetes medication at the time of recruitment.18 Dyslipidaemia was defined as the presence of one of the following: total cholesterol (TC)≥5.0 mmol/L, or high-density lipoprotein cholesterol (HDL-C)<1.0 mmol/L for men and <1.3 mmol/L for women, or low-density lipoprotein cholesterol (LDL-C) ≥3. 0 mmol/L, or triglycerides (TG)≥1.7 mmol/L, or on lipid-modifying medication,19 or self-report of ever being diagnosed by a health professional with high cholesterol. The Randox Daytona Plus (Randox Laboratories, UK) autoanalyzer was used to analyse HDL-C, TG and TC on fasting venous blood samples. LDL-C was calculated using the Friedewald equation.20 CKD was defined as estimated glomerular filtration rate (eGFR)<60 mL/min per 1.73 m2 (calculated using the Chronic Kidney Disease Epidemiology (CKD-EPI) (creatinine) equation 2009, without adjustment for African American ethnicity), presence of albuminuria (urine albumin creatinine ratio>3 mg/mmol) or both.21 As the study was cross-sectional, low eGFR and albuminuria were not confirmed with follow-up testing. At each partner site, blood and urine specimens were processed and stored at −112°F. After completion of the study, each partner site transported samples on dry ice to a central laboratory in Johannesburg, South Africa. At the central laboratory, all analyses were batched and performed according to good laboratory practice with external monitoring for quality control. Serum and urine creatinine (mg/dL) were measured using Jaffe’s kinetic method calibrated to an isotope dilution mass spectrometry-traceable standard. Urinary albumin concentration was measured with immunoturbidimetry. The urine albumin to urine creatinine ratio (mg/mmol) was calculated from these measurements. CVD was defined as present if the participant reported having had a heart attack or stroke or transient ischaemic attack. Participants previously diagnosed with congestive heart failure or angina were also classified as having CVD. No data were available for angina or heart attack prevalence for the Soweto men, so CVD prevalence was calculated using the remaining data. Asthma was determined by self-report or use of medication for the condition. ## Statistical analysis Data were summarised using means and standard deviations (±SD) for continuous parametric data, and median (IQR) for non-parametric data. A Student’s t-test was used to test for differences between men and women on parametric continuous variables within each site, while Kruskal-Wallis test was used for non-parametric data. Chi-square test was used for categorical data. Due to significant sex differences within sites (online supplemental tables 1–4), all further analyses were stratified by sex. Since the outcome had three categories, that is, zero conditions, one condition and at least two conditions (multimorbidity) of the seven conditions under study, multinomial logistic regression was used to explore the factors associated with either one condition or at least two conditions, with none as the reference group, in men and women separately. A p value<0.05 was considered statistically significant in all tests carried out using STATA V.14.1 SE. To understand the different multimorbidity patterns between men and women, within and between countries, and between regions, that is, South (South African sites), East (Kenyan site) and West (Ghana and Burkina Faso) Africa, different diseases were plotted using UpSetR, an R package.22 Separate analyses were done for men and women, separated by site and then geographic region. ## Patient and public involvement This study did not involve any patients and/or the public. ## Sociodemographic and lifestyle factors Only participants between the ages of 40 and 60 years were selected for these analyses ($$n = 10$$ 700), and the median age ranged from 48 years in Nairobi to 51 years in Agincourt, Dikgale and Navrongo (online supplemental tables 1 and 2). The only site where there was a significant sex difference in age was Navrongo where women were older than men. Sex differences in education level attained were significant at all sites, except Dikgale where the majority of participants had completed secondary school ($55.2\%$), while in Navrongo and Nanoro the majority of the participants did not have any formal education. Employment was above $90\%$ in both Nairobi and Nanoro, $62.6\%$ and $60\%$ in Navrongo and Soweto, respectively, and only $36.1\%$ in Agincourt and $37.7\%$ in Dikgale. There were significant sex differences in SES at all six sites. Smoking and alcohol consumption were significantly different between the sexes at all sites except Soweto (alcohol status was not known in the women), as previously reported.23 ## Multimorbidity prevalence For the total study population as well as when the sites were combined by geographical area, multimorbidity prevalence was higher in women compared with men, with nearly $50\%$ of the women in the total sample presenting with multimorbidity compared with $35\%$ of the men (figure 1). Multimorbidity prevalence was highest in the South African men and women ($51.7\%$ and $64.9\%$, respectively), followed by East Africa ($31.3\%$ and $48.4\%$, respectively) and then West Africa ($20.2\%$ and $24.1\%$, respectively). Overall, the site with the highest multimorbidity prevalence was Agincourt with $66.6\%$, and the site with the lowest prevalence was Nanoro with $21.2\%$ (online supplemental tables 3 and 4). Figure 1: Prevalence of 0, 1, 2, 3 and 4 or more chronic conditions (including HIV infection, cardiovascular disease, chronic kidney disease, asthma, diabetes, dyslipidaemia and hypertension) in men and women from South, East and West Africa, and the full AWI-Gen cohort. ## Multimorbidity clusters Multimorbidity clustering for the total study population, and for the three geographical regions, stratified by sex, are presented in figures 2 and 3, respectively. Figure 2: Multimorbidity clustering for the total AWI-Gen cohort in (A) women and (B) men. Figure 3: Multimorbidity clustering in (A) South African women, (B) South African men, (C) East African women, (D) East African men, (E) West African women and (F) West African men. ## Study population, stratified by sex The majority of the total study population of men ($42.9\%$) and women ($44.8\%$) had 1 disease only, and in those with only 2 diseases, the most common combination was dyslipidaemia and hypertension (men $12\%$; women $18\%$) (figure 2A, B). While in men, the most common combination of three diseases was dyslipidaemia, hypertension and CKD ($2\%$ of the total study population), in women the most common combination was dyslipidaemia, hypertension and HIV infection ($3.2\%$). The prevalence of 4 or more diseases was marginally higher in women compared with men ($3.3\%$ vs $2.1\%$) with the most common cluster of 4 diseases being dyslipidaemia, hypertension, HIV and CKD in both sexes. ## Geographical regions, stratified by sex The majority of men ($51.7\%$) and women ($64.9\%$) living in South Africa presented with multimorbidity (online supplemental table 2, figure 1), with two diseases (hypertension and dyslipidaemia) (figure 3A, B) as the most common combination in the men ($33.4\%$) and women ($41.3\%$). In contrast the majority of men ($51.1\%$) and women ($45.6\%$) from East Africa reported only one disease, although the pattern in those with two diseases was similar to South Africa as dyslipidaemia and hypertension were the most common cluster (figure 3C, D), and this was the same in West Africa (figure 3E, F). It is in the clustering of three or more diseases that there are differences between the sites as the most common clustering of three diseases in South African men and women was dyslipidaemia, hypertension and HIV, while in East and West Africa dyslipidaemia and hypertension were most commonly clustered with CKD. ## Sites, stratified by sex At all sites except Nanoro, the prevalence of multimorbidity was higher in women than in men, with the prevalence being highest in Agincourt women ($74.1\%$) and lowest in men from Navongo ($18\%$) (online supplemental tables 3 and 4). In all the sites, the most common disease combination in men and women with two diseases only was dyslipidaemia and hypertension (online supplemental figure 1–12), with the lowest prevalence in Nanoro women ($7.6\%$) and the highest prevalence in Soweto women ($30.5\%$). There were differences between the sites in the most common disease clusters in men and women with three diseases only, with dyslipidaemia and hypertension clustered with CKD in Soweto men, Nairobi men and women, Navrongo men and women and Nanoro women. In men from Dikgale, and Agincourt men and women, the most common disease cluster in those with only three diseases was dyslipidaemia, hypertension and HIV, while in Dikgale and Soweto women, and Nanoro men, it was dyslipidaemia, hypertension and diabetes. The highest prevalence with 4 or more diseases was reported in women from the South African sites: $8\%$ of the Agincourt women and $7.6\%$ of the Dikgale women (data was not available for Soweto women). Except for Nairobi women, <$1\%$ of the men and women from the East and West African sites presented with 4 or more diseases. ## Multinomial logistic regression, stratified by sex Risk factors associated with having one condition and with having multimorbidity, compared with no condition, were determined using multinomial regression and are presented for women (table 1) and men (table 2). Due to the lower prevalence of multimorbidity in women from East and West Africa compared with South Africa, women living in Nairobi and Nanoro were more likely to have one condition compared with Agincourt women, and conversely, together with women from Navrongo, they were at a lower relative risk of multimorbidity compared with women from Agincourt. In women, being older was associated with a higher relative risk of multimorbidity, and a 1 kg/m2 higher BMI was associated with a $7\%$ higher risk of having 1 condition and an $11\%$ higher risk of multimorbidity. Further, being a former consumer of alcohol compared with never consuming alcohol was associated with a higher risk of having one disease, and with multimorbidity. None of the other measures of SES or lifestyle behaviours were associated with multimorbidity risk in women. When compared with the Agincourt men, the relative risk of multimorbidity was lower at all sites, except Soweto. Similar to the women, older age and BMI were associated with a higher risk of multimorbidity with every year of age being associated with a $3\%$ higher risk of multimorbidity, and a 1 kg/m2 higher BMI was associated with a $9\%$ higher risk of having 1 condition and a $14\%$ higher risk of multimorbidity. While smoking was not associated with disease risk, the current consumption of alcohol, whether non-problematic or problematic, was associated with a lower risk of having one disease, and current non-problematic consumption of alcohol was associated with a $32\%$ lower risk of multimorbidity. Another lifestyle behaviour that was associated with disease risk was time spent in MVPA, with more time spent in MVPA associated with a lower risk of having one condition, and with multimorbidity. Living with a partner either presently or in the past was associated with a higher risk of one condition and an even higher risk of multimorbidity when compared with not living with a partner. The attainment of primary school level education was associated with a $30\%$ higher risk of multimorbidity compared with no formal education, and being employed was associated with a $33\%$ lower risk of having multimorbidity compared with being unemployed; however, there was no association between SES quintile and risk for one disease or multimorbidity. ## Discussion In this study of middle-aged men and women living in South, East and West Africa we have reported sex differences in the prevalence of multimorbidity, and identified socio-demographic, socio-economic and lifestyle factors associated with multimorbidity risk in men and women. While the prevalence of multimorbidity was more than $50\%$ in South Africa, it was only just over $20\%$ in West Africa, with East Africa reporting a prevalence of $31.3\%$ in men and $48.4\%$ in women. What was consistent across all sites was the higher prevalence of multimorbidity in women compared with men, and in both men and women, age and BMI were independently associated with a higher risk of multimorbidity. These findings highlight the need for longitudinal studies of ageing African populations to examine the effect of multimorbidity on mortality, as studies from high-income countries have shown that this is influenced by disease type, number and certain combinations that comprise multimorbidity.24 25 Our study has also shown that the most commonly occurring disease cluster at all SSA sites was dyslipidaemia with hypertension. A recent systematic review of prevalence studies from South Africa reported a $3\%$–$23\%$ prevalence of multimorbidity in studies with a wide range of age groups and $30\%$–$71\%$ in older adults.26 The higher prevalence of multimorbidity in the South African sites in the current study ($54\%$–$66\%$) may be attributed to the advanced epidemiological transition and increasing urbanisation in South Africa.27 This coincides with a higher BMI and prevalence of obesity in the three South African sites compared with the West and East African sites.28 Despite being considered a rural site, Agincourt reported the highest prevalence of multimorbidity at $66.6\%$. This prevalence is similar to those reported from the same study site but including larger sample sizes of participants older than 40 years (n>3000) and different chronic diseases in the definition of multimorbidity.29 30 Kabudula et al 31 have described a ‘protracted’ epidemiological transition in Agincourt between 1993 and 2013 as a result of the coexistence of infectious and NCDs, as well as the influence of social changes, and their results highlight the different transition experience in LMICs compared with high-income countries.27 Although the West African sites, Ghana and Burkina Faso, reported the lowest prevalence of multimorbidity and a similar prevalence between men and women, it is expected that this will increase, following a similar trajectory to the South African sites. Mohamed et al 10 have reported a lower multimorbidity prevalence of $28.7\%$ ($31.5\%$ in women vs $21.4\%$ in men) in the same AWI-Gen study participants from Nairobi, Kenya. Reasons for the different prevalence compared with the current study is that they included 16 chronic conditions, and all measurements, except for hypertension and obesity, were via self-report. This may have resulted in an underestimation of multimorbidity prevalence as it is well recognised that many of these chronic diseases are undiagnosed in Africa.32–34 The higher prevalence of multimorbidity in women compared with men has been well described,35 with the studies from South Africa showing less consistent results.26 Dyslipidaemia was the most commonly occurring disease across all the sites, with the prevalence ranging from $42.7\%$ in men from Navrongo to $87.4\%$ in women from Dikgale. These results are similar to other African studies,36 37 with a recent study by Masilela et al,38 reporting a prevalence of $76.7\%$ in South African adults receiving care for diabetes and hypertension, and Reiger et al 39 reporting a prevalence of $67.3\%$ in over 4000 adults aged 40 years and older from Agincourt, South Africa. Many of these African studies have reported that low HDL-C is the main driver of the high prevalence of dyslipidaemia, and it is important to debate whether the current international cut-offs for HDL-C, triglyceride and other cholesterol measures are relevant to the African population. A low HDL-C is frequently seen with obesity and the metabolic syndrome and is associated with hypertriglyceridaemia, particularly post prandially.40 The main driver of atherosclerotic CVD is LDL-C so this may be a more important marker of risk for CVD.41 If this is the main criterion for defining dyslipidaemia and low HDL-C were excluded from the definition, then the prevalence of dyslipidaemia would be considered to have a lesser effect in the context of multimorbidity. Data from Agincourt has reported that while low HDL (<1.19 mmol/L) was prevalent in $26.5\%$ of the adults aged 40 years and older, high LDL (>4.1 mmol/L) was measured in only $3.7\%$.42 In the current study, the most common cluster of two diseases at all sites was dyslipidaemia with hypertension, with this prevalence being higher than the prevalence of any single disease in Agincourt women, and Soweto men and women. This is in contrast to a recent systematic review from South Africa which identified the common disease clusters as hypertension and diabetes, hypertension and HIV, and tuberculosis (TB) and HIV.26 Irrespective of the difference in findings, these studies highlight the high prevalence of hypertension in these large samples of middle-aged men and women from East, West and South Africa as previously reported by Gómez-Olivé et al.34 A recent systematic review and meta-analysis assessing CKD in African participants with hypertension provides strong evidence for the devastating outcomes of untreated hypertension as it reports a $17.8\%$ pooled prevalence of CKD in people with hypertension.43 In the total sample of men in the current study the most prevalent combination of three diseases was dyslipidaemia, hypertension and kidney disease ($2\%$ of the total sample) while in the women this was dyslipidaemia, hypertension and HIV ($3\%$ of the total sample). The co-occurrence of kidney disease and hypertension was more common than kidney disease and diabetes in both men and women, a finding which has previously been reported in Africa.44 Several multimorbidity patterns including a ‘cardiorespiratory’ pattern and a ‘metabolic’ pattern were identified across countries in the Study on global AGEing and adult health (SAGE) reporting data on 12 chronic conditions, although the factors associated with multimorbidity differed between countries and could be explained by their diverse development status.11 Reporting on the South African data only, Chidumwa et al, identified three groups using latent class analysis.7 While $88\%$ of the sample were classified as minimal multimorbidity risk, $11\%$ were classified as concordant (hypertension and diabetes) multimorbidity and $6\%$ as discordant (angina, asthma, chronic lung disease, arthritis and depression) multimorbidity.7 Identifying factors associated with multimorbidity clustering is important in designing country or region-specific strategies to manage multimorbidity. Results from this study as well as global and African studies have reported an increased risk of multimorbidity with increasing age.10 30 Although this association is well accepted and understood, in the study by Garin et al,11 multimorbidity prevalence decreased with age in the South African cohort, which they suggested may be due to the decrease in HIV prevalence with age. The current study also reported an $11\%$–$14\%$ higher risk of multimorbidity with each 1 kg/m2 increase in BMI in both men and women. Overweight and obesity are well known risk factors for NCDs such as diabetes, hypertension and dyslipidaemia, particularly a low HDL-C, and obesity is a targeted risk factor by the WHO in their Global Action Plan for the prevention and control of NCDs.45 Although age and BMI were associated with multimorbidity prevalence in both men and women, lifestyle correlates displayed sex differences. In the women being a former consumer of alcohol was associated with nearly two times the risk of having multimorbidity, while men who were current, non-problematic consumers of alcohol were at lower risk of multimorbidity. Data from the same SSA cohort has reported that current alcohol consumption is lower in women compared with men at all AWI-Gen sites, with the characteristics of alcohol consumption such as type and frequency of consumption being significantly different between the sexes.23 This may explain the different associations with multimorbidity risk in the current study, and together with the finding that time spent in MVPA was inversely associated with multimorbidity risk in men only, suggests that further intervention studies to reduce multimorbidity risk may need to be stratified by sex. The influence of socioeconomic patterns on disease risk is well recognised, and may differ between countries at different stages of the epidemiological transition.27 Employment was associated with a reduced risk of multimorbidity in this study and supports the inverse association between SES and disease risk reported by others in LMICs.46 We have shown that when compared with men who have never married or been in a union, men who are and men who have previously been in a union (divorced, separated, partner deceased) were at a significantly higher risk of multimorbidity. In their study of adults living in four SSA countries, Ajayi et al 47 reported similar associations with BMI but in only two of the sites, rural and periurban Uganda, with no associations in adults from Tanzania and South Africa. We have previously shown in the AWI-Gen cohort that married men had a higher BMI compared with their unmarried counterparts, which consisted of those who had never been married or were no longer married,48 while results from the Prospective Urban and Rural Epidemiological (PURE) study have shown that people who have experienced marital loss are at higher CVD risk.49 Marital status may represent social position, SES as well as cultural beliefs around body size and marriage, and these results highlight the complexity of the association with disease risk. It is acknowledged that these data are cross-sectional and single screen testing without follow-up may result in the overestimation of diseases such as CKD. A further limitation of the study is that data were only available for four of the seven conditions for the Soweto women; however, description of the disease profile of this subsample still makes a contribution to the limited literature from LMICs. Further, only prevalence and correlates of multimorbidity could be identified; however, the sample size is large and represents four SSA countries to help understand differences and similarities between populations at different stages of the epidemiological transition. These prevalence figures make an important contribution to the increasing knowledge around multimorbidity and may be particularly useful in developing indices that can be used as part of machine learning approaches in predicting multimorbidity in the future. In conclusion, this study has shown that nearly half of the population-based sample of women and more than a third of the men between the ages of 40 and 60 years from South, East and West Africa are living with two or more chronic diseases. Common disease clusters have been identified and future efforts should focus on managing multiple commonly occurring diseases rather than single diseases. Although the prevalence of multimorbidity is higher in women living in South Africa, it is expected that as the obesity epidemic continues to increase that the prevalence will increase in men as well as East and West Africa. Identifying correlates of multimorbidity is critical to providing focused and effective interventions. ## Data availability statement Data are available upon reasonable request. The data have been submitted to the European Genome-Phenome Archive (EGA), accession number EGA00001002482. The Human Heredity and Health in Africa (H3Africa) Data and Biospecimen Access Committee (DBAC) will review requests for the AWI-Gen phenotype dataset. Related documents including study protocol and statistical analysis plan will be available upon request from the corresponding author. ## Patient consent for publication Not applicable. ## Ethics approval This study involves human participants and was approved by Human Research Ethics Committee (Medical) of the University of the Witwatersrand (certificate numbers M121029 and M170880). Each study centre obtained local ethics approval. Participants gave informed consent to participate in the study before taking part. ## References 1. Gouda HN, Charlson F, Sorsdahl K. Burden of non-communicable diseases in sub-saharan africa, 1990-2017: results from the global burden of disease study 2017. Lancet Glob Health (2019) 7 e1375-87. DOI: 10.1016/S2214-109X(19)30374-2 2. Tran PB, Kazibwe J, Nikolaidis GF. Costs of multimorbidity: a systematic review and meta-analyses. BMC Med (2022) 20 234. DOI: 10.1186/s12916-022-02427-9 3. Afshar S, Roderick PJ, Kowal P. Multimorbidity and the inequalities of global ageing: a cross-sectional study of 28 countries using the world health surveys. BMC Public Health (2015) 15. DOI: 10.1186/s12889-015-2008-7 4. Mendenhall E. Syndemics: a new path for global health research. Lancet (2017) 389 889-91. DOI: 10.1016/S0140-6736(17)30602-5 5. Abebe F, Schneider M, Asrat B. Multimorbidity of chronic non-communicable diseases in low- and middle-income countries: a scoping review. J Comorb (2020) 10. DOI: 10.1177/2235042X20961919 6. Lalkhen H, Mash R. Multimorbidity in non-communicable diseases in South African primary healthcare. S Afr Med J (2015) 105 134-8. DOI: 10.7196/samj.8696 7. Chidumwa G, Maposa I, Corso B. Identifying co-occurrence and clustering of chronic diseases using latent class analysis: cross-sectional findings from sage South Africa wave 2. BMJ Open (2021) 11. DOI: 10.1136/bmjopen-2020-041604 8. Nimako BA, Baiden F, Sackey SO. Multimorbidity of chronic diseases among adult patients presenting to an inner-city clinic in Ghana. Global Health (2013) 9. DOI: 10.1186/1744-8603-9-61 9. Hien H, Berthé A, Drabo MK. Prevalence and patterns of multimorbidity among the elderly in Burkina Faso: cross-sectional study. Trop Med Int Health (2014) 19 1328-33. DOI: 10.1111/tmi.12377 10. Mohamed SF, Haregu TN, Uthman OA. Multimorbidity from chronic conditions among adults in urban slums: the awi-gen nairobi site study findings. Glob Heart (2021) 16 6. DOI: 10.5334/gh.771 11. Garin N, Koyanagi A, Chatterji S. Global multimorbidity patterns: a cross-sectional, population-based, multi-country study. J Gerontol A Biol Sci Med Sci (2016) 71 205-14. DOI: 10.1093/gerona/glv128 12. Alaba O, Chola L. The social determinants of multimorbidity in South Africa. Int J Equity Health (2013) 12. DOI: 10.1186/1475-9276-12-63 13. Ramsay M, Crowther N, Tambo E. H3Africa AWI-gen collaborative centre: a resource to study the interplay between genomic and environmental risk factors for cardiometabolic diseases in four sub-Saharan African countries. Glob Health Epidemiol Genom (2016) 1. DOI: 10.1017/gheg.2016.17 14. Ali SA, Soo C, Agongo G. Genomic and environmental risk factors for cardiometabolic diseases in africa: methods used for phase 1 of the AWI-gen population cross-sectional study. Glob Health Action (2018) 11 1507133. DOI: 10.1080/16549716.2018.1507133 15. Ewing JA. Detecting alcoholism. The CAGE questionnaire. JAMA (1984) 252 1905-7. DOI: 10.1001/jama.252.14.1905 16. Bull FC, Maslin TS, Armstrong T. Global physical activity questionnaire (GPAQ): nine country reliability and validity study. J Phys Act Health (2009) 6 790-804. DOI: 10.1123/jpah.6.6.790 17. Chobanian AV, Bakris GL, Black HR. Seventh report of the joint National Committee on prevention, detection, evaluation, and treatment of high blood pressure. Hypertension (2003) 42 1206-52. DOI: 10.1161/01.HYP.0000107251.49515.c2 18. Classification and diagnosis of diabetes: standards of medical care in diabetes-2021. Diabetes Care (2021) 44 S15-33. DOI: 10.2337/dc21-S002 19. Klug E, Raal FJ, Marais AD. South African dyslipidaemia guideline consensus statement: 2018 update a joint statement from the south african heart association (sa heart) and the lipid and atherosclerosis Society of Southern Africa (lassa). S Afr Med J (2018) 108 973-1000. DOI: 10.7196/SAMJ.2018.v108i11.13383 20. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem (1972) 18 499-502. DOI: 10.1093/clinchem/18.6.499 21. Stevens PE, Levin A. Evaluation and management of chronic kidney disease: synopsis of the kidney disease: improving global outcomes 2012 clinical practice guideline. Ann Intern Med (2013) 158 825-30. DOI: 10.7326/0003-4819-158-11-201306040-00007 22. Conway JR, Lex A, Gehlenborg N. UpSetR: an R package for the visualization of intersecting sets and their properties. Bioinformatics (2017) 33 2938-40. DOI: 10.1093/bioinformatics/btx364 23. Boua PR, Soo CC, Debpuur C. Prevalence and socio-demographic correlates of tobacco and alcohol use in four sub-Saharan African countries: a cross-sectional study of middle-aged adults. BMC Public Health (2021) 21. DOI: 10.1186/s12889-021-11084-1 24. Schäfer I, Kaduszkiewicz H, Nguyen TS. Multimorbidity patterns and 5-year overall mortality: results from a claims data-based observational study. J Comorb (2018) 8. DOI: 10.1177/2235042X18816588 25. Willadsen TG, Siersma V, Nicolaisdóttir DR. Multimorbidity and mortality: a 15-year longitudinal registry-based nationwide danish population study. J Comorb (2018) 8 2235042X18804063. DOI: 10.1177/2235042X18804063 26. Roomaney RA, van Wyk B, Turawa EB. Multimorbidity in south africa: a systematic review of prevalence studies. BMJ Open (2021) 11. DOI: 10.1136/bmjopen-2021-048676 27. Stringhini S, Forrester TE, Plange-Rhule J. The social patterning of risk factors for noncommunicable diseases in five countries: evidence from the modeling the epidemiologic transition study (Mets). BMC Public Health (2016) 16. DOI: 10.1186/s12889-016-3589-5 28. Ramsay M, Crowther NJ, Agongo G. Regional and sex-specific variation in BMI distribution in four sub-saharan african countries: the h3Africa AWI-gen study. Glob Health Action (2018) 11 1556561. DOI: 10.1080/16549716.2018.1556561 29. Chang AY, Gómez-Olivé FX, Payne C. Chronic multimorbidity among older adults in rural South Africa. BMJ Glob Health (2019) 4. DOI: 10.1136/bmjgh-2018-001386 30. Wade AN, Payne CF, Berkman L. Multimorbidity and mortality in an older, rural black South African population cohort with high prevalence of HIV findings from the HAALSI study. BMJ Open (2021) 11. DOI: 10.1136/bmjopen-2020-047777 31. Kabudula CW, Houle B, Collinson MA. Progression of the epidemiological transition in a rural South African setting: findings from population surveillance in agincourt, 1993-2013. BMC Public Health (2017) 17. DOI: 10.1186/s12889-017-4312-x 32. Hoffman RM, Chibwana F, Kahn D. High rates of uncontrolled blood pressure in malawian adults living with HIV and hypertension. Glob Heart (2021) 16 81. DOI: 10.5334/gh.1081 33. Stokes A, Berry KM, Mchiza Z. Prevalence and unmet need for diabetes care across the care continuum in a national sample of South African adults: evidence from the SANHANES-1, 2011-2012. PLoS One (2017) 12. DOI: 10.1371/journal.pone.0184264 34. Gómez-Olivé FX, Ali SA, Made F. Regional and sex differences in the prevalence and awareness of hypertension: an h3africa AWI-gen study across 6 sites in sub-saharan africa. Glob Heart (2017) 12 81-90. DOI: 10.1016/j.gheart.2017.01.007 35. Violan C, Foguet-Boreu Q, Flores-Mateo G. Prevalence, determinants and patterns of multimorbidity in primary care: a systematic review of observational studies. PLoS One (2014) 9. DOI: 10.1371/journal.pone.0102149 36. Chori B, Danladi B, Nwakile P. Prevalence, patterns and predictors of dyslipidaemia in Nigeria: a report from the REMAH study. Cardiovasc J Afr (2022) 33 52-9. DOI: 10.5830/CVJA-2021-037 37. Fiseha T, Alemu W, Dereje H. Prevalence of dyslipidaemia among HIV-infected patients receiving combination antiretroviral therapy in North shewa, Ethiopia. PLoS One (2021) 16. DOI: 10.1371/journal.pone.0250328 38. Masilela C, Adeniyi OV, Benjeddou M. Prevalence, patterns and determinants of dyslipidaemia among south african adults with comorbidities. Sci Rep (2022) 12 337. DOI: 10.1038/s41598-021-04150-6 39. Reiger S, Jardim TV, Abrahams-Gessel S. Awareness, treatment, and control of dyslipidemia in rural South Africa: the HAALSI (health and aging in Africa: a longitudinal study of an indepth community in South Africa) study. PLoS One (2017) 12. DOI: 10.1371/journal.pone.0187347 40. Klop B, Elte JWF, Cabezas MC. Dyslipidemia in obesity: mechanisms and potential targets. Nutrients (2013) 5 1218-40. DOI: 10.3390/nu5041218 41. Ference BA, Ginsberg HN, Graham I. Low-density lipoproteins cause atherosclerotic cardiovascular disease. 1. Evidence from genetic, epidemiologic, and clinical studies. A consensus statement from the European Atherosclerosis Society consensus panel. Eur Heart J (2017) 38 2459-72. DOI: 10.1093/eurheartj/ehx144 42. Gaziano TA, Abrahams-Gessel S, Gomez-Olive FX. Cardiometabolic risk in a population of older adults with multiple co-morbidities in rural south africa: the HAALSI (health and aging in africa: longitudinal studies of indepth communities) study. BMC Public Health (2017) 17. DOI: 10.1186/s12889-017-4117-y 43. Ajayi SO, Ekrikpo UE, Ekanem AM. Prevalence of chronic kidney disease as a marker of hypertension target organ damage in Africa: a systematic review and meta-analysis. Int J Hypertens (2021) 2021. DOI: 10.1155/2021/7243523 44. Bikbov B, Purcell CA, Levey AS. Global, regional, and national burden of chronic kidney disease, 1990-2017: a systematic analysis for the global burden of disease study 2017. Lancet (2020) 395 709-33. DOI: 10.1016/S0140-6736(20)30045-3 45. WHO global NCD action plan 2013-2020. (2013) 46. Hosseinpoor AR, Bergen N, Kunst A. Socioeconomic inequalities in risk factors for non communicable diseases in low-income and middle-income countries: results from the world health survey. BMC Public Health (2012) 12. DOI: 10.1186/1471-2458-12-912 47. Ajayi IO, Adebamowo C, Adami H-O. Urban-rural and geographic differences in overweight and obesity in four sub-saharan african adult populations: a multi-country cross-sectional study. BMC Public Health (2016) 16. DOI: 10.1186/s12889-016-3789-z 48. Micklesfield LK, Kagura J, Munthali R. Demographic, socio-economic and behavioural correlates of BMI in middle-aged black men and women from urban johannesburg, south africa. Glob Health Action (2018) 11 1448250. DOI: 10.1080/16549716.2018.1448250 49. Egbujie BA, Igumbor EU, Puoane T. A cross-sectional study of socioeconomic status and cardiovascular disease risk among participants in the prospective urban rural epidemiological (pure) study. S Afr Med J (2016) 106 900-6. DOI: 10.7196/SAMJ.2016.v106i9.10456
--- title: Do socioeconomic inequities arise during school-based physical activity interventions? An exploratory case study of the GoActive trial authors: - Olivia Alliott - Hannah Fairbrother - Kirsten Corder - Paul Wilkinson - Esther van Sluijs journal: BMJ Open year: 2023 pmcid: PMC10016273 doi: 10.1136/bmjopen-2022-065953 license: CC BY 4.0 --- # Do socioeconomic inequities arise during school-based physical activity interventions? An exploratory case study of the GoActive trial ## Abstract ### Objective To investigate socioeconomic inequities in the intervention and evaluation process of the GoActive school-based physical activity intervention and demonstrate a novel approach to evaluating intervention-related inequalities. ### Design Exploratory post-hoc secondary data analysis of trial data. ### Setting The GoActive trial was run in secondary schools across Cambridgeshire and Essex (UK), between September 2016 and July 2018. ### Participants 13–14 years old adolescents ($$n = 2838$$, 16 schools). ### Methods Socioeconomic inequities across six stages in the intervention and evaluation process were evaluated: [1] provision of and access to resources; [2] intervention uptake; [3] intervention effectiveness (accelerometer-assessed moderate-to-vigorous physical activity (MVPA)); [4] long-term compliance; [5] response in evaluation; and [6] impact on health. Data from self-report and objective measures were analysed by individual-level and school-level socioeconomic position (SEP) using a combination of classical hypothesis tests and multilevel regression modelling. ### Results Stage: [1] There was no difference in the provision of physical activity resources by school-level SEP (eg, quality of facilities (0–3), low=2.6 (0.5); high=2.5 (0.4). [ 2] Students of low-SEP engaged significantly less with the intervention (eg, website access: low=$37.2\%$; middle=$45.4\%$; high=$47.0\%$; $$p \leq 0.001$$). [ 3] There was a positive intervention effect on MVPA in adolescents of low-SEP (3.13 min/day, $95\%$ CI −1.27 to 7.54, but not middle/high (−1.49; $95\%$ CI −6.54 to 3.57). [ 4] At 10 months post-intervention, this difference increased (low SEP: 4.90; $95\%$ CI 0.09 to 9.70; middle/high SEP: −2.76; $95\%$ CI −6.78 to 1.26). [ 5] There was greater non-compliance to evaluation measures among adolescents of low-SEP (eg, % accelerometer compliance (low vs high): baseline: 88.4 vs 92.5; post-intervention: 61.6 vs 69.2; follow-up: 54.5 vs 70.2. [ 6] The intervention effect on body mass index (BMI) z-score was more favourable in adolescents of low-SEP (low SEP: −0.10; $95\%$ CI −0.19 to 0.00; middle/high: 0.03; $95\%$ CI −0.05 to 0.12). ### Conclusions These analyses suggest the GoActive intervention had a more favourable positive effect on MVPA and BMI in adolescents of low-SEP, despite lower intervention engagement. However, differential response to evaluation measures may have biassed these conclusions. We demonstrate a novel way of evaluating inequities within young people’s physical activity intervention evaluations. ### Trial registration number ISRCTN31583496. ## Introduction The health benefits of physical activity are well-established1 and physical inactivity has been identified as a major public health concern.2 Active adolescents experience better present and long-term health and are more likely to become active and healthy adults.3–5 However, globally over $80\%$ of students aged 11–17 years are insufficiently active to accrue the benefits.6 Similar to other health behaviours, disparities in physical activity during adolescence may contribute to inequities in current and future health.7 Recent review-level evidence highlights the importance of promoting and enabling physical activity among adolescents living in the context of socioeconomic deprivation, who report experiencing more barriers to physical activity when compared with other socioeconomic groups.8 Despite regularly collecting relevant information at baseline, most controlled trials of physical activity interventions in young people do not analyse differences in intervention effect across socioeconomic groups.9 This has led to a scarcity of evidence regarding the differential impact of intervention across socioeconomic groups.9 Public health literature suggests the extent to which inequities are perpetuated or reduced can depend on the nature of the intervention.10 ‘High-risk strategies’ target individuals with a higher risk of developing the disease, whereas population strategies attempt to lower the risk of the entire population by shifting the distribution of underlying risk factors, such as physical inactivity.11 As a consequence of compulsory education in many countries, the potential for schools to deliver wide-reaching and equitable physical activity interventions has been well documented.12 13 Taking a population approach, school-based interventions have been studied and deemed successful if average physical activity levels increase.14 However, population strategies have the potential to inadvertently exacerbate health inequities within a population.15 Researchers have begun to consider the potential for interventions to have a differential effect across individuals, commonly named ‘intervention generated inequities’.10 However, across young people’s physical activity literature these studies have tended to focus on differential effects by gender.9 Limited evidence from individual evaluations of physical activity and school-based interventions document socioeconomic inequities negatively impacting those of a low-socioeconomic position (SEP) in the provision of, and access to, interventions and resources,16 17 intervention uptake,18 intervention efficacy,17 19 long-term compliance20 and differential response in evaluations.9 21 22 These previous studies offer examples of various points in the research and intervention process where inequities might emerge. Going forward we propose a broader approach is needed, looking at intervention generated inequities throughout the whole research and intervention process of a single intervention. Based on the work of White et al,10 Love identifies key stages throughout a physical activity intervention where inequities can be introduced.22 Understanding how inequities might emerge at each of these stages is essential for the development of equitable school-based physical activity interventions, as while inequities at each of these stages could be small, together they may lead to significant inequities in final outcomes.10 The aim of this paper is to take a case-study approach to investigate if and how socioeconomic inequities arise during the intervention and evaluation process of the GoActive school-based physical activity intervention. In doing so, we demonstrate a novel way of studying inequities across the intervention implementation and evaluation process that could be applied more broadly. ## Methods This paper describes exploratory secondary analyses of the GoActive trial data. These analyses were not detailed in the statistical analysis plan for the main trial analyses, but were guided by a prespecific statistical analysis plan. The GoActive trial was run between September 2016 and July 2018. Ethical approval for the GoActive trial was obtained from the University of Cambridge Psychology Research Ethics Committee (PRE.2015.126). The trial was prospectively registered (ISRCTN31583496). ## Participants and randomisation Sixteen state-run secondary schools in Cambridgeshire and Essex agreed to participate. All Year 9 students (age 13–14 years) and their parents/carers received written study information. Students provided written assent and parents provided passive informed consent (opt-out consent).23 School-level randomisation, stratified by the percentage of students eligible for pupil premium funding at each school (below or above county-specific median) and county (Cambridgeshire or Essex), occurred after baseline measurement.23 Pupil premium funding aims to reduce the effects of deprivation on educational attainment and is used here as a proxy measure for school-level deprivation.24 ## GoActive intervention GoActive was a theory-based intervention developed following an evidence-based iterative approach.23 The primary aim of GoActive was to increase students’ moderate-to-vigorous intensity physical activity (MVPA) across the week.23 GoActive was delivered over 12 weeks to all students in the intervention schools irrespective of whether they participated in study measurements. The control schools followed normal practice. GoActive was implemented using a tiered-leadership system led by mentors (older students within the school) and supported by peer-elected Year 9 leaders.23 During the intervention, Year 9 tutor groups chose 2 activities per week from a selection of 20. These activities required little or no equipment and were designed to appeal to a variety of students (including Ultimate Frisbee, Zumba and Hula Hoop). Schools had access to the GoActive intervention website where they could find activity instructions cards which included an overview of each activity, suggested adaptations, safety tips, ‘factoids’ and a short video.23 Mentors remained with the class throughout the intervention, whereas peer-leaders changed each week. During the first 6 weeks, additional leadership was provided by a local authority-funded intervention facilitator (health trainers employed by local councils) who continued to provide remote support thereafter.23 Teachers were encouraged to dedicate one tutor time per week to do one of the chosen activities as a class. Students could gain points for trying these new activities at any time in or out of school, irrespective of intensity or duration.23 There was no expectation of time spent in the activities, points were rewarded for taking part. Individual points remained private and students could enter their points at any time on the GoActive website with an individual password and login details. Students were encouraged to regularly log these points to unlock rewards such as a sports bag, t-shirt or hoodie. While remaining private these points were entered into between-class competitions.23 The results of the main GoActive trail analysis reported no overall intervention effect on average daily MVPA.25 Subgroup analyses conducted as part of the trial evaluation reported a suggestion of a positive intervention effect among students of a low/middle-SEP. Across all MVPA outcomes, those of high-SEP appeared to benefit least when compared with low/middle-SEP students. Full details of the trial methods have been published elsewhere.23 ## Methodological approach of the current study As outlined above, we take a case study approach to demonstrate how socioeconomic inequities can be explored throughout the research and intervention process, using the GoActive intervention as an example. As this is an exploratory post-hoc analysis, we operationalised the model proposed by Love to include research questions based on the available GoActive data collected as part of the main GoActive trial (figure 1).22 For the remainder of this paper, we refer to the stages outlined in figure 1 when describing our research and findings. Figure 1: Intervention stages and accompanying research questions explored throughout the study. Based on the model developed by Love.22 MVPA, moderate-to-vigorous physical activity; NCDs, non-communicable diseases; SEP, socioeconomic position. This paper focuses on socioeconomic inequities, therefore all of the research questions presented in figure 1 consider SEP. We use individual-level and area-level SEP, as using these different levels are important when evaluating the full contribution of socioeconomic conditions.26 The relevance of different indicators of SEP is dependent on on the research focus, health outcome and stage in the life course.26 Taking this approach, we used pupil premium funding (see description in section 2.1) as a school-level indicator of SEP during stages 1 and 2, where the object of analysis is the school, not the individual. Schools were categorised as low-SEP if the percentage of students eligible for pupil premium was below the county-specific mean and high-SEP if the percentage was above. For the remaining stages, we use an individual-level indicator of SEP derived from the Family Affluence Scale (FAS).27 In response to the recent review evidence highlighted in the introduction8 (published after the main GoActive trial) and because of our focus on socioeconomic inequities we compare students of low-SEP to students of middle/high-SEP during stages 3, 4 and 5. This is a different approach to that of the main trail which grouped students of low-SEP and middle-SEP together. All measures are described in further detail below. ## Measures Study measurements were taken at four time points during the GoActive trial: Baseline (BL), mid-intervention (T2; 6 weeks after intervention start), post-intervention (T3; 14–16 weeks after intervention start) and 10-month follow-up (T4; 10 months after intervention end.25 A summary of demographic measures and the measures specific to each stage are described below, the best available measures from the trial data were used to address the research questions under each stage. For conciseness, the following shorter titles are applied to each stage: stage 1—provision and access, stage 2—intervention uptake, stage 3—intervention effect, stage 4—long-term compliance, stage 5—evaluation participation and stage 6—health outcomes. ## Demographic measures Participant descriptive characteristics were self-reported at baseline.23 Participants reported gender from three response options (male, female and prefer not to say).25 Individual-level SEP was reported using the FAS, which is composed of six items relating to: [1] family car ownership, [2] holidays, [3] computers, [4] availability of bathrooms, [5] dishwasher ownership and [6] having their own bedroom. These were used as a proxy measure of individual-level socioeconomic position by summing the answers (possible range 0–13), and dividing into predefined affluence groups (low=0–6, middle=7–9 and high=10–13).25 Ethnicity was self-reported by participants, who were given 20 response options and an additional free-text option.25 The reported options were recoded into five categories in accordance with published recommendations: [1] ‘white’, [2] ‘mixed ethnicity’ (ie, identifying with multiple ethnicities), [3] ‘Asian’ (including South Asian and Chinese), [4] ‘African and/or Caribbean’ and [5] ‘other’.25 28 ## Stage 1: provision and access Data on the school physical activity policy and social and physical environment were self-reported at baseline by contact teachers (often Physical Education or Year 9 lead) at all schools.23 The data were used to highlight the potential for socioeconomic differences in the provision of physical activity opportunities and access to resources at baseline, which may have impacted the delivery of GoActive. These data were collected using a questionnaire previously used in the Year 9 data collection of the Sport Physical Activity and Eating Behaviour, Environmental Determinants in Young People (SPEEDY) study.29 A list of 16 physical activity facilities available at each school were given a quality rating (0=facility not present, 1=low quality facility, 2=middle quality facility and 3=high quality facility). Ratings were summed and divided by the number of available facilities to give an average quality rating. An average rating was also used to indicate the suitability of the school grounds for sport, informal games and general play across three measures (1=not at all suitable, 2=somewhat suitable and 3=very suitable). The provision of physical activity opportunities was assessed using the extracurricular opportunities on offer at each school derived from a list of 24 (including space to add ‘other’ activities; one activity=one point, eg, Rounders=1) and weekly hours of PE, measured using an open-ended question where teachers rounded to the nearest half-hour. The suitability of the area around the school for physical activity was assessed on a scale of 1–5 (1=strongly disagree to 5=strongly agree) across three measures, shielding from hedges/trees/fences, maintenance of the grounds and the presence of vandalism. Finally, the school’s attitude towards physical activity was assessed using the same 1–5 agreement scale across five measures which included encouraging physical activity at school and outside school, educating about the risks of physical activity and how to practice safe physical activity and encouraging active travel. Pupil premium was used as a school-level indicator of SEP, which was reported by teachers in the school environment questionnaire.25 Table 2 shows that regardless of school-level SEP, teachers reported their schools to be suitable for physical activity at baseline. Differences between the provision of and access to physical activity facilities by school-level SEP were tested, but none were identified as statistically significant with p values >0.05. Table 2 \| Unnamed: 0 \| Schools of low-SEP (N=8) \| Schools of high-SEP (N=8) \| \| --- \| --- \| --- \| \| \| Mean (SD) \| Mean (SD) \| \| Physical activity environment \| Physical activity environment \| Physical activity environment \| \| School level measure (possible range) \| \| \| \| Quality of school physical activity facilities (0–3) \| 2.6 (0.5) \| 2.5 (0.4) \| \| Suitability of school grounds for physical activity (3–9) \| 8.3 (1.5) \| 8.0 (1.4) \| \| Extra-curricular opportunities for physical activity (0–25) \| 11.0 (2.2) \| 12.5 (3.7) \| \| Weekly hours of physical education (0+) \| 2.0 (0.0) \| 2.2 (.4) \| \| Area around school suitable for physical activity (3–15) \| 11.9 (2.2) \| 12.8 (1.2) \| \| School attitude towards physical activity (5–25) \| 18 (6.0) \| 19.3 (6.3) \| \| Recruitment rates \| Recruitment rates \| Recruitment rates \| \| Number of Year 9 students at baseline (N) \| 1648 \| 1759 \| \| Recruited at baseline N (%) \| 1369 (83.1) \| 1469 (83.5) \| \| Students from each family affluence group by school-level SEP \| N (%) \| N (%) \| \| Low individual-SEP \| 266 (19.4) \| 132 (9.0) \| \| Middle individual-SEP \| 598 (43.7) \| 608 (41.4) \| \| High individual-SEP \| 505 (36.9) \| 729 (49.6) \| ## Stage 2: intervention uptake Under stage 2, research questions explore engagement with the GoActive intervention. Recruitment data were used to assess the initial uptake of the intervention by school-level SEP. Evaluation uptake was measured as whether participants provided baseline questionnaire data, which was a requirement for participating in GoActive.25 Trained measurement staff checked the questionnaires on completion and helped students complete missing sections.23 Intervention uptake was assessed using data on students’ engagement with the GoActive website as this was the primary method for tracking the activities participants engaged in both in and out of school. This included whether students accessed the GoActive website at any time during the intervention period and was recorded as a categorical variable (accessed vs not). Of the students who did access the website, the number of times they visited and the number of points they logged throughout the intervention were recorded. Table 2 provides a breakdown of recruitment by school-level SEP, suggesting that a lower proportion of students from low-SEP were recruited into the GoActive trial, particularly in high-SEP schools. Table 3 presents the uptake of the GoActive intervention by individual-level SEP using website engagement. The results show that significantly fewer students of low-SEP than middle-SEP and high-SEP accessed the GoActive intervention website. There was no difference in engagement found for those who did access the website. Table 3 \| Unnamed: 0 \| Low-SEPN=235 \| Middle-SEPN=670 \| High-SEPN=606 \| X2 \| df \| P value (adjusted for ties) \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Accessed the website N (%) \| 94 (40.0) \| 304 (45.4) \| 315 (52.0) \| 16.52 \| 2 \| 0.0 \| \| Mean website points (SD) \| 49.8 (123.1) \| 53.2 (85.1) \| 55.0 (87.8) \| 0.53 \| 2 \| 0.77 \| \| Mean website visits (SD) \| 14.2 (28.1) \| 15.5 (21.0) \| 16.0 (22.8) \| 0.74 \| 2 \| 0.69 \| ## Stage 3: intervention effect During stage 3, differential intervention efficacy was explored for the GoActive primary outcome, daily accelerometer assessed MVPA at 14–16 weeks post-intervention.23 Participants were asked to wear a wrist-worn activity monitor (Axivity) assessing acceleration (continuous waveform data) continuously (24 hours a day) for 7 days.23 The Axivity monitor has been validated to assess energy expenditure and to have increased wear time adherence and acceptability than hip-worn monitors in adolescents.23 30–32 Monitor output was processed to provide minutes spent in MVPA to be equivalent to ≥2000 ActiGraph counts per minute23; further details on accelerometer data processing can be found elsewhere.25 Table 4 shows the moderating effect of SEP on the effectiveness of the GoActive intervention on average daily minutes of MVPA. The results of the interaction analysis suggest at the post-intervention measurement the intervention effect in students of middle/high-SEP was 4.56 ($95\%$ CI −9.56 to 0.41) min/day less MVPA than students of low-SEP. However, subgroup analyses did not show statistically significant effects in either group. Table 4 \| Unnamed: 0 \| B \| 95% CI \| P value \| Model N \| \| --- \| --- \| --- \| --- \| --- \| \| 14–16 weeks post intervention \| 14–16 weeks post intervention \| 14–16 weeks post intervention \| 14–16 weeks post intervention \| 14–16 weeks post intervention \| \| MVPA \| \| \| \| \| \| Interaction term \| \| \| \| \| \| Intervention×SEP \| −4.56 \| −9.56 to 0.41 \| 0.069 \| 1878 \| \| Stratified analysis \| \| \| \| \| \| Low-SEP \| 3.13 \| −1.27 to 7.54 \| 0.150 \| 241 \| \| Middle/high-SEP \| −1.49 \| −6.54 to 3.57 \| 0.540 \| 1637 \| \| 10 months post intervention \| 10 months post intervention \| 10 months post intervention \| 10 months post intervention \| 10 months post intervention \| \| MVPA \| \| \| \| \| \| Interaction term \| \| \| \| \| \| Intervention×SEP \| −7.53 \| −12.89 to −2.17 \| 0.009 \| 1785 \| \| Stratified analysis \| \| \| \| \| \| Low-SEP \| 4.90 \| 0.09 to 9.70 \| 0.046 \| 203 \| \| Middle/high-SEP \| −2.76 \| −6.78 to 1.26 \| 0.164 \| 1582 \| \| BMI z-score \| \| \| \| \| \| Interaction effect \| \| \| \| \| \| Intervention×SEP \| 0.12 \| −0.02 to 0.26 \| 0.096 \| 2070 \| \| Stratified analysis \| \| \| \| \| \| Low-SEP \| −0.10 \| −0.19 to 0.0 \| 0.055 \| 247 \| \| Middle/high-SEP \| 0.03 \| −0.05 to 0.12 \| 0.413 \| 1823 \| \| Body fat (%) \| \| \| \| \| \| Interaction term \| \| \| \| \| \| Intervention×SEP \| 1.09 \| −0.63 to 2.81 \| 0.198 \| 1873 \| \| Stratified analysis \| \| \| \| \| \| Low-SEP \| −0.69 \| −3.17 to 1.78 \| 0.560 \| 216 \| \| Middle/high-SEP \| 0.41 \| −0.75 to 1.57 \| 0.464 \| 1657 \| \| Waist circumference (cm) \| \| \| \| \| \| Interaction term \| \| \| \| \| \| Intervention×SEP \| 0.73 \| −0.68 to 2.15 \| 0.287 \| 2089 \| \| Stratified analysis \| \| \| \| \| \| Low-SEP \| −0.71 \| −1.64 to 1.30 \| 0.808 \| 249 \| \| Middle/high-SEP \| 0.56 \| −0.17 to 1.30 \| 0.124 \| 1840 \| ## Stage 4: long-term compliance Stage 4 used accelerometer measurements taken at 10 months post-intervention to reflect long-term compliance to the intervention by exploring compliance to the primary outcome after the intervention period. Average daily minutes of MVPA was used as described above. At 10 months post intervention, the difference in intervention effect increased to −7.53 ($95\%$ CI −12.89 to −2.17) min/day MVPA in favour of participants of low-SEP (table 4). Subsequent stratified analyses showed a positive intervention effect in participants of a low-SEP (4.90; $95\%$ CI 0.09 to 9.70) but not those of middle/high-SEP. ## Stage 5: evaluation participation During stage 5, differential participation in evaluation measures was assessed using compliance with questionnaire and accelerometer measures. Questionnaire compliance was defined as whether participants provided questionnaire data at each measurement occasion. Research staff working on the GoActive study recorded whether a questionnaire for each participant had been completed and checked at each measurement point. Accelerometer compliance was determined as whether participants provided valid accelerometer data at each measurement point. In line with the main GoActive trial analysis, participants were required to provide 6 hours of wear time from a possible 42 hours in each daytime quadrant: morning (06:00 to 12:00), afternoon (12:00 to 18:00), evening (20:00 to 24:00) and night (24:00 to 06:00).25 Figure 2A shows that questionnaire compliance decreased throughout the intervention across all socioeconomic groups. The figure also shows an association between individual-level SEP and questionnaire compliance (lower compliance among students of low-SEP). Differences in compliance increased with time from T2 to T3 to T4 (T2: X2 = 23.45, $$p \leq 0.00$$; T3: X2 = 15.25, $$p \leq 0.00$$; T4: X2 = 43.88, $$p \leq 0.00$$). Figure 2B shows this trend was also observed for accelerometer compliance (BL: X2 = 8.90, $$p \leq 0.02$$; T3: X2 = 8.12, $$p \leq 0.02$$; T4: X2 = 33.65, $$p \leq 0.00$$). Figure 2: Compliance to study evaluation measures throughout the GoActive trial by individual level-SEP, indicated by (A) percentage of students proving questionnaire data; and (B) percentage of students providing accelerometer data. BL, baseline. ## Stage 6: health outcomes Related health outcomes were explored during stage 6 using anthropometric measures. During a school site visit, trained measurement staff conducted the following measures according to standardised operation procedures: height (m), weight (kg), waist circumference (cm) and bioimpedance to assess body fat percentage (%).23 Body mass index (BMI) SD scores were calculated from height and weight data (i.e. weight/height2 (kg/m2)) and categorised according to age and gender standardised International Obesity Task Force thresholds.25 Table 4 shows an indication ($$p \leq 0.09$$) of a more favourable intervention effect on the BMI z-score in participants of low-SEP, however the interaction term was not statistically significant. No interaction effects were observed for waist circumference or body fat. Subsequent stratified analyses suggest a favourable intervention effect on BMI z-score among adolescents of low-SEP when compared with the control condition (low SEP: −0.10; $95\%$ CI −0.19 to 0.00), but not for those of middle/high-SEP (middle/high: 0.03; $95\%$ CI −0.05 to 0.12). See online supplemental table 1 for mean physical activity and anthropometric outcomes by SEP and randomisation group at each measurement point. ## Analysis Characteristics of the sample were described using mean, SD and frequency values. Data from all measurement points were included across the analyses described below and were stratified by either individual-level or school-level SEP. All included analyses were exploratory, but guided by an analysis plan developed prior to release of the data. Research questions under stages 1 (provision and access) and 2 (intervention uptake) used self-reported data from the school environment and student questionnaires. Data were explored using simple tabulations, graphical techniques and basic summary statistics and analysed using Kruskal-Wallis one-way analysis of variance by school-level SEP. This test was selected due to the skewness of the data. P values were adjusted for ties as the same values occurred in more than one sample. For further analyses under stage 2, website access by the intervention group was also explored using Pearson’s χ2. Of those who accessed the website, differences in the number of visits and points logged by individual-level SEP were analysed using the Kruskal-Wallis test as described above. Research questions under stages 3 (intervention effect) and 4 (long-term compliance) were explored using accelerometer assessed MVPA, interaction analyses were run to examine if the effect of the independent variable (intervention vs control) on the dependent variable (daily average MVPA) differed by individual-level SEP, following statistical procedures from the main GoActive analyses.23 For MVPA at T3 and T4 (ie, the primary outcome), the intervention effect, representing the baseline-adjusted difference in change from baseline between the intervention and control groups, was estimated from a linear regression model, including randomisation group, baseline value of the outcome (i.e. analysis of covariance), the randomisation stratifiers (ie, pupil premium funding and county) and an interaction between individual-level SEP and group allocation. Models were also run separately for low and middle/high socioeconomic groups to assess intervention effects within subgroups. Robust SEs were calculated to allow for the non-independence of individuals within schools. Under stage 5 (evaluation participation), we examine differential response to evaluation measures by individual-level SEP. We examined accelerometer compliance and self-report compliance (eg, questionnaire completion vs no completion) using Pearson’s χ2. Stage 6 (health outcomes) was explored using anthropometric outcomes. Interaction analyses were used to examine if the effect of the independent variable (intervention vs control) on the dependent variable (BMI, waist circumference or body fat) differed by individual-level SEP, separate analyses were run for each anthropometric variable following the same analytical approach as stages 3 and 4. All analyses were conducted using Stata V.15.1 software. ## Patient and public involvement None for the purpose of this secondary data analysis. ## Sample description A total of 2838 students provided baseline questionnaire data. Table 1 provides an overview of baseline characteristics by individual-level SEP. Overall, mean age was 13.3 (SD 0.4) years, just over half of the participants were men ($51.4\%$) and the majority of the participants were of white British ethnicity ($84.7\%$). Fewer participants were of a low-SEP ($14.0\%$), than of middle-SEP ($42.5\%$) and high-SEP ($43.5\%$). Table 1 \| Unnamed: 0 \| Low-SEP \| Middle-SEP \| High-SEP \| \| --- \| --- \| --- \| --- \| \| N (%) \| N (%) \| N (%) \| N (%) \| \| Participant \| 398 (14.0) \| 1206 (42.5) \| 1234 (43.5) \| \| Gender \| Gender \| Gender \| Gender \| \| Male \| 196 (6.9) \| 598 (21.1) \| 684 (24.1) \| \| Female \| 202 (7.1) \| 608 (21.4) \| 550 (19.4) \| \| Ethnic group \| Ethnic group \| Ethnic group \| Ethnic group \| \| White \| 319 (11.3) \| 1032 (36.5) \| 1071 (37.8) \| \| Mixed/multiple ethnic background \| 32 (1.1) \| 73 (2.6) \| 76 (2.7) \| \| Asian or Asian British \| 20 (0.7) \| 52 (1.8) \| 36 (1.3) \| \| Black or black British \| 16 (0.6) \| 30 (1.1) \| 24 (0.8) \| \| Other ethnic group \| 10 (0.4) \| 16 (0.6) \| 22 (0.7) \| \| Mean (SD) \| Mean (SD) \| Mean (SD) \| Mean (SD) \| \| Age \| 13.2 (0.4) \| 13.3 (0.4) \| 13.3 (0.5) \| \| BMI \| 21.1 (4.3) \| 20.5 (3.7) \| 20.1 (3.5) \| \| Body fat % \| 22.1 (10.2) \| 21.3 (10.1) \| 19.9 (9.7) \| \| Waist circumference \| 71.7 (11.1) \| 70.5 (9.7) \| 69.0 (8.9) \| ## Discussion Taking a case-study approach we investigated if and how socioeconomic inequities arose during the intervention and evaluation process of a school-based physical activity intervention called GoActive. In doing so, we present a novel approach to analysing young people’s physical activity interventions from an equality lens. The findings described below demonstrate the benefit of taking this approach to intervention evaluation, providing insight beyond the main trial analysis. We discuss our main findings in relation to three key elements: intervention context and engagement (stages 1 and 2), intervention effectiveness (stages 3, 4 and 6) and intervention recruitment and evaluation (stages 2 and 5). ## Intervention context and engagement Our finding that school-level SEP did not appear to influence the school physical activity environment at baseline, contrasts with previous research highlighting socioeconomic inequities in school physical activity provision and resources.8 33 34 While it is likely this could be the result of the small sample size of included schools ($$n = 16$$) and resultant limited power to show significant differences, this could also be due to the UK context of GoActive, where national and local policy, such as the Schools Premises Regulations, impose minimum standards for school sports grounds and facilities.35 36 In addition to extra funding available for low-SEP schools, such a pupil premium funding24 which could be spent on the provision of physical activity resources and opportunities. In relation to engagement, significantly fewer adolescents of low-SEP accessed the GoActive website. Of those who did, a graded effect was observed with adolescents of low-SEP engaging the least. One explanation for this, as highlighted in previous research, is that students living in the context of socioeconomic deprivation interact differently with the school environment (eg, the use of equipment, fostering of autonomy, competence and relatedness, update of extracurricular opportunities) potentially impacting their engagement with GoActive.37 38 Furthermore, review evidence reports that adolescents of low-SEP experience multiple barriers to engaging in physical activity interventions, including digital exclusion.14 ## Intervention effectiveness Despite apparently lower engagement, our exploratory analyses suggest that participants of a low-SEP responded more favourably to GoActive, with a difference in effect of 7.53 min/day at 10 months post-intervention to participants of a middle/high-SEP. It may be that students of a low-SEP had a lower engagement with the website but were more engaged with other elements of the intervention that we have no data on. The observed intervention effect of ~5 min of MVPA per day may be important for health,39 and was the targeted effect in the main GoActive trial. A similar pattern of effect was also observed for BMI z-score.23 Overall, these findings support the potential for school-based interventions to reduce inequities in physical activity and obesity. It is possible more deprived students particularly benefitted from the chance to try the variety of new activities offered during the GoActive intervention.8 25 This is especially promising given the stark socioeconomically patterned inequities in overweight and obesity in the UK and other high-income countries.8 40 Our choice to treat adolescents of low-SEP as an independent group was based on recent review evidence that their experiences of physical activity notably differ to those of middle-SEP and high-SEP, highlighting the value of looking at them as a separate group.8 By doing so, our findings add to the main trial moderation analyses where participants of low-SEP and middle-SEP were grouped, suggesting the intervention effect was primarily experienced among low-SEP adolescents. While the approach initially taken was prespecified and common among existing literature,23 mainly due to the small sample size of low-SEP groups, these exploratory analyses suggest that important differences in effect may be overlooked when taking this approach. ## Intervention recruitment and evaluation Recruitment data showed that $14\%$ of those participating in the GoActive trial were of a low-SEP and the majority of these students attended low-SEP schools. In the East of England, data from the Family Resources Survey (2016–2019) reports $19.5\%$ of young people were living in poverty at the time GoActive was delivered.41 It is possible that this is due to the small sample of 16 schools that are unlikely to be representative of the county. Furthermore, while ‘living in poverty’ is a different indicator of SEP than family affluence, measures of SEP are shown to be highly correlated.26 42 It is therefore worth considering, when comparing these percentages, that adolescents of low-SEP might be under-represented in the overall GoActive sample, aligning with evidence that socioeconomically disadvantaged groups are ‘hard to reach’ and recruit into research.43 Of those recruited into GoActive, inequities in study evaluation measures were observed. These results are consistent with previously reported socioeconomic patterns in response to survey evaluation measures.10 44 45 Higher accelerometer non-response has also been reported among socioeconomically-deprived children,46 47 however, there is a lack of research looking at socioeconomic patterning in accelerometer compliance among adolescent populations. Based on these findings, it is important to acknowledge that our analyses were conducted using a small subset of students of low-SEP which may result in bias in our conclusions. It is possible that differential engagement and response to evaluation measures resulted in a subset of students of low-SEP who were not reflective of the group more broadly, impacting the generalisability of our results. Furthermore, it is possible this may have impacted the results of our analyses under stages 3 and 4, where those who remained in GoActive are more likely to be those who got most out of it. ## Strengths and limitations Previous research has begun to look at differential effectiveness using the primary outcome of the intervention.9 To our knowledge, this is the first paper to provide an example of how inequities can be explored throughout the intervention and research process of young people’s physical activity interventions. Taking a stage-based approach we highlight differential engagement in specific components of the GoActive intervention, including accessing the GoActive website and in response to evaluation measures. We further highlight the potential of school-based interventions to reduce inequities in MVPA and obesity. Further strengths include the diversity of data collected during the GoActive trial which allowed us to build a more holistic picture of inequities during the trial and the use of device-measured MVPA, which aligns with public health research recommendations for the objective and comprehensive evaluation of health promotion programmes.25 48 While presenting our result as exploratory, rather than confirmatory, we acknowledge the small sample size of the low-SEP group and school-level data raises problems with regards to statistical power.49 The subjective quantification of school environment features may have given rise to self-report biases.37 It is possible that teachers’ reported acceptability of physical activity resources was relative to school-level deprivation, with teachers at high-SEP schools expecting a higher standard of resources and facilities. It is also suggested that some activity types (eg, biking, stair walking) and intensities can be misclassified by wrist-worn accelerometers.50 If these behaviours are also socioeconomically patterned, this may have led to an underestimation or overestimation of the difference in effect between subgroups. It is also possible that differential access to computers outside of school hours may have impacted engagement with the GoActive website.14 Further limitations include the relative lack of participants of a low-SEP and of non-white ethnicity.25 It is acknowledged that this is an exploratory post-hoc study of the main trial data collected for GoActive. It is therefore presented as an example of one approach to exploring intervention-generated inequities throughout the intervention and evaluation process. With this in mind, the operationalisation of each stage (figure 1) and the resultant analyses were based on the best available data from the GoActive trial and not what would ideally be the most appropriate data to address each stage of Love’s model. Data were not available to address other relevant questions, such as whether schools has access to facilities specifically needed to run GoActive (rather than general facilities) (to address stage 1), the SEP of schools that agreed to participate versus those who did not (to address stage 2), the role the intervention development process could have played in the uptake of and engagement with the intervention (to address stage 5) or the cumulative effects of inequities across the stages of Love’s model. A further limitation was not being able to use the focus group and interview data collected as part of the GoActive process evaluation, as information was not available on the SEP of the participants involved. To properly address each stage of Love’s model, the stages need to be considered and embedded in the research design. ## Recommendation for future research and practice As highlighted above, this paper presents a case-study example of how to analyse young people’s physical activity interventions with an equity lens. Drawing on the stages developed by Love, a framework for future studies to apply, adapt and develop is provided.22 While the paper focuses on young people’s physical activity interventions, the application of this approach more broadly is encouraged. The data required for such a comprehensive analysis should be considered during the design stage of future interventions and trials. This will help prevent the development and implementation of unequitable interventions, making better use of public resources.51 The financial and resource requirement for running sufficiently large trials to detect a main intervention effect and differential effects between subgroups are acknowledged.9 To tackle this, Love et al have previously recommended encouraging coordinated efforts towards fewer, high-quality, large trials, adequately powered to address questions of differential effectiveness.9 This study echoes this statement, continuing to solely add evidence on overall effectiveness will continue to limit the evidence-base and our understanding from progressing. Mobilising the approach presented in this project for existing intervention strategies will further help develop our understanding of why current interventions appear to be ineffective in tackling physical inactivity during adolescence.12 In addition to developing our understanding of the most appropriate data to address each stage of the model, going forward it would be useful to apply this approach to a range of trials to provide researchers and public health professionals with further examples of how to assess inequities at each stage, generating ideas within the research community and continuing to develop this approach. The results of this stage-based analysis show the potential for universal school-based physical activity interventions to positively impact socioeconomically deprived students (who remained participating in the trial), reducing inequities. Importantly this contradicts the common assumption that interventions generate or exacerbate inequities.9 The results also demonstrate how intervention components that require individual agency, such as accessing the GoActive website, can exacerbate inequities.52 53 This should be considered in the development and implementation of school policy, especially in schools with a high proportion of students of a low-SEP. Due to the exploratory nature of this study, it would be beneficial for future research to further study the potential benefit of school-based physical activity interventions for students of low-SEP. It may be useful to explore the application of easily accessible interventions, such as the Daily Mile, to a secondary school setting.54 Recruiting and retaining participants of a low-SEP can be challenging, which means they are often under-represented in research.14 To increase the reach of interventions and to be able to conduct statistically powered subgroup analyses, the development of active and targeted recruitment of adolescents living in the context of socioeconomic deprivation is an important area for future research. The lack of representation of low-SEP groups in intervention development is an opportunity for growth within school-based interventions and an important area to be considered in the development and evaluation of interventions. Strategies are also needed to better engage these adolescents in the research process, for example, involving them in the design and research process through patient and public involvement.55 ## Conclusion This was an exploratory study exploring whether and how socioeconomic inequities might arise throughout a school-based physical activity intervention. We demonstrate how the GoActive trial positively affected the physical activity and BMI of low-SEP students. However, differential engagement in the intervention and response to evaluation measures may have biassed these conclusions. The continued development and evaluation of school-based interventions from an equity lens is essential as we move out of the COVID-19 pandemic, where disparities in school-based physical activity were exacerbated.56 ## Data availability statement All data relevant to the study are included in the article or uploaded as supplementary information. The authors do not have the authority to share the data that support the findings of this study and the data are not openly available because of ethical and legal considerations. Non-identifiable data can be made available to bona fide researchers on submission of a reasonable request to datasharing@mrc-epid.cam.ac.uk. The principles and processes for accessing and sharing data are outlined in the MRC Epidemiology Unit Data Access and Data Sharing Policy. ## Patient consent for publication Not applicable. ## Ethics approval Not applicable. ## References 1. Pratt M, Ramirez Varela A, Salvo D. Attacking the pandemic of physical inactivity: what is holding us back?. Br J Sports Med (2020) 54 760-2. DOI: 10.1136/bjsports-2019-101392 2. Blair SN. Physical inactivity: the biggest public health problem of the 21st century. Br J Sports Med (2009) 43 1-2. PMID: 19136507 3. Corder K, Winpenny E, Love R. Change in physical activity from adolescence to early adulthood: a systematic review and meta-analysis of longitudinal cohort studies. Br J Sports Med (2019) 53 496-503. DOI: 10.1136/bjsports-2016-097330 4. Koivusilta L, Rimpelä A, Rimpelä M. Health related lifestyle in adolescence predicts adult educational level: a longitudinal study from Finland. J Epidemiol Community Health (1998) 52 794-801. DOI: 10.1136/jech.52.12.794 5. Koivusilta LK, West P, Saaristo VMA. From childhood socio-economic position to adult educational level-do health behaviours in adolescence matter? A longitudinal study. BMC Public Health (2013) 13 1-9. DOI: 10.1186/1471-2458-13-711 6. Guthold R, Stevens GA, Riley LM. Global trends in insufficient physical activity among adolescents: a pooled analysis of 298 population-based surveys with 1·6 million participants. Lancet Child Adolesc Health (2020) 4 23-35. DOI: 10.1016/S2352-4642(19)30323-2 7. Vander Ploeg KA, Maximova K, McGavock J. Do school-based physical activity interventions increase or reduce inequalities in health?. Soc Sci Med (2014) 112 80-7. DOI: 10.1016/j.socscimed.2014.04.032 8. Alliott O, Ryan M, Fairbrother H. Do adolescents’ experiences of the barriers to and facilitators of physical activity differ by socioeconomic position? A systematic review of qualitative evidence. Obes Rev (2022) 23. DOI: 10.1111/obr.13374 9. Love RE, Adams J, van Sluijs EMF. Equity effects of children’s physical activity interventions: a systematic scoping review. Int J Behav Nutr Phys Act (2017) 14. DOI: 10.1186/s12966-017-0586-8 10. White M, Adams J, Heywood P. How and why do interventions that increase health overall widen inequalities within populations? health, inequality and society. Social Inequality and Public Health (2009). DOI: 10.1332/policypress/9781847423207.001.0001 11. Rose G. Sick individuals and sick populations. Int J Epidemiol (1985) 14 32-8. DOI: 10.1093/ije/14.1.32 12. Neil-Sztramko SE, Caldwell H, Dobbins M. School-Based physical activity programs for promoting physical activity and fitness in children and adolescents aged 6 to 18. Cochrane Database Syst Rev (2021) 9. DOI: 10.1002/14651858.CD007651.pub3 13. Venkatraman T, Honeyford K, Costelloe CE. Sociodemographic profiles, educational attainment and physical activity associated with the daily mileregistration in primary schools in england: a national cross-sectional linkage study. J Epidemiol Community Health (2021) 75 137-44. DOI: 10.1136/jech-2020-214203 14. Craike M, Wiesner G, Hilland TA. Interventions to improve physical activity among socioeconomically disadvantaged groups: an umbrella review. Int J Behav Nutr Phys Act (2018) 15 43. DOI: 10.1186/s12966-018-0676-2 15. Frohlich KL, Potvin L. Transcending the known in public health practice: the inequality paradox: the population approach and vulnerable populations. Am J Public Health (2008) 98 216-21. DOI: 10.2105/AJPH.2007.114777 16. Fernandes M, Sturm R. Facility provision in elementary schools: correlates with physical education, recess, and obesity. Prev Med (2010) 50 Suppl 1 S30-5. DOI: 10.1016/j.ypmed.2009.09.022 17. Sallis JF, Conway TL, Prochaska JJ. The association of school environments with youth physical activity. Am J Public Health (2001) 91 618-20. DOI: 10.2105/ajph.91.4.618 18. Beauchamp A, Backholer K, Magliano D. The effect of obesity prevention interventions according to socioeconomic position: a systematic review. Obes Rev (2014) 15 541-54. DOI: 10.1111/obr.12161 19. Rush E, Reed P, McLennan S. A school-based obesity control programme: project energize. two-year outcomes. Br J Nutr (2012) 107 581-7. DOI: 10.1017/S0007114511003151 20. Williams NA, Coday M, Somes G. Risk factors for poor attendance in a family-based pediatric obesity intervention program for young children. J Dev Behav Pediatr (2010) 31 705-12. DOI: 10.1097/DBP.0b013e3181f17b1c 21. Craig CL, Bauman A, Gauvin L. ParticipACTION: a mass media campaign targeting parents of inactive children; knowledge, saliency, and trialing behaviours. Int J Behav Nutr Phys Act (2009) 6 88. DOI: 10.1186/1479-5868-6-88 22. Love R. Inequalities in children’s physical activity and interventions: centre for diet and physical activity research (2019) 23. Brown HE, Whittle F, Jong ST. A cluster randomised controlled trial to evaluate the effectiveness and cost-effectiveness of the goactive intervention to increase physical activity among adolescents aged 13–14 years. BMJ Open (2017) 7. DOI: 10.1136/bmjopen-2016-014419 24. 24Department of Education GU. Guidance using pupil premium: guidance for school leaders. 2019.. Guidance using pupil premium: guidance for school leaders (2019) 25. Corder KL, Brown HE, Croxson CH. A school-based, peer-led programme to increase physical activity among 13- to 14-year-old adolescents: the goactive cluster RCT. Public Health Res (2021) 9 1-134. DOI: 10.3310/phr09060 26. Galobardes B, Lynch J, Smith GD. Measuring socioeconomic position in health research. Br Med Bull (2007) 81–82 21-37. DOI: 10.1093/bmb/ldm001 27. Pardo-Crespo MR, Narla NP, Williams AR. Comparison of individual-level versus area-level socioeconomic measures in assessing health outcomes of children in Olmsted County, Minnesota. J Epidemiol Community Health (2013) 67 305-10. DOI: 10.1136/jech-2012-201742 28. Johnson MRD, Bhopal RS, Ingleby JD. A glossary for the first world congress on migration. Ethnicity, Race and Health (2019) 172 85-8. DOI: 10.1016/j.puhe.2019.05.001 29. Harrison F, van Sluijs EMF, Corder K. School grounds and physical activity: associations at secondary schools, and over the transition from primary to secondary schools. Health Place (2016) 39 34-42. DOI: 10.1016/j.healthplace.2016.02.004 30. Rowlands AV, Olds TS, Hillsdon M. Assessing sedentary behavior with the geneactiv: introducing the sedentary sphere. Med Sci Sports Exerc (2014) 46 1235-47. DOI: 10.1249/MSS.0000000000000224 31. Phillips LRS, Parfitt G, Rowlands AV. Calibration of the GENEA accelerometer for assessment of physical activity intensity in children. J Sci Med Sport (2013) 16 124-8. DOI: 10.1016/j.jsams.2012.05.013 32. Schaefer CA, Nigg CR, Hill JO. Establishing and evaluating wrist cutpoints for the geneactiv accelerometer in youth. Med Sci Sports Exerc (2014) 46 826-33. DOI: 10.1249/MSS.0000000000000150 33. Hollis JL, Williams AJ, Sutherland R. A systematic review and meta-analysis of moderate-to-vigorous physical activity levels in elementary school physical education lessons. Prev Med (2016) 86 34-54. DOI: 10.1016/j.ypmed.2015.11.018 34. Carlson JA, Mignano AM, Norman GJ. Socioeconomic disparities in elementary school practices and children’s physical activity during school. Am J Health Promot (2014) 28 S47-53. DOI: 10.4278/ajhp.130430-QUAN-206 35. The school premises (england) regulations 2012. (2012) 36. Area guidelines for mainstream schools. (2014) 37. Foubister C, van Sluijs EMF, Vignoles A. The school policy, social, and physical environment and change in adolescent physical activity: an exploratory analysis using the LASSO. PLoS One (2021) 16. DOI: 10.1371/journal.pone.0249328 38. Moore GF, Littlecott HJ, Evans R. School composition, school culture and socioeconomic inequalities in young people’s health: multi-level analysis of the health behaviour in school-aged children (HBSC) survey in Wales. Br Educ Res J (2017) 43 310-29. DOI: 10.1002/berj.3265 39. Ekelund U, Luan J, Sherar LB. Moderate to vigorous physical activity and sedentary time and cardiometabolic risk factors in children and adolescents. JAMA (2012) 307 704-12. DOI: 10.1001/jama.2012.156 40. Mayor S. Socioeconomic disadvantage is linked to obesity across generations, UK study finds. BMJ (2017) 356. DOI: 10.1136/bmj.j163 41. Agrawal S, Phillips D. Catching up or falling behind? geographical inequalities in the UK and how they have changed in recent years. The Institute for Fiscal Studies (2020) 2 42. Galobardes B, Shaw M, Lawlor DA. Indicators of socioeconomic position (Part 1). J Epidemiol Community Health (2006) 60 7-12. DOI: 10.1136/jech.2004.023531 43. Bonevski B, Randell M, Paul C. Reaching the hard-to-reach: a systematic review of strategies for improving health and medical research with socially disadvantaged groups. BMC Med Res Methodol (2014) 14 1-29. DOI: 10.1186/1471-2288-14-42 44. Turrell G, Patterson C, Oldenburg B. The socio-economic patterning of survey participation and non-response error in a multilevel study of food purchasing behaviour: area- and individual-level characteristics. Public Health Nutr (2003) 6 181-9. DOI: 10.1079/PHN2002415 45. Ligthart KAM, Buitendijk L, Koes BW. The association between ethnicity, socioeconomic status and compliance to pediatric weight-management interventions-a systematic review. Obes Res Clin Pract (2017) 11 1-51. DOI: 10.1016/j.orcp.2016.04.001 46. Rich C, Cortina-Borja M, Dezateux C. Predictors of non-response in a UK-wide cohort study of children’s accelerometer-determined physical activity using postal methods. BMJ Open (2013) 3. DOI: 10.1136/bmjopen-2012-002290 47. Love R, Adams J, Atkin A. Socioeconomic and ethnic differences in children’s vigorous intensity physical activity: a cross-sectional analysis of the UK millennium cohort study. BMJ Open (2019) 9. DOI: 10.1136/bmjopen-2018-027627 48. 48All-Party Commission on Physical A. Tackling physical inactivity – A coordinated approach. 2014.. Tackling physical inactivity – A coordinated approach (2014) 49. Haas JP. Sample size and power. Am J Infect Control (2012) 40 766-7. DOI: 10.1016/j.ajic.2012.05.020 50. Arvidsson D, Fridolfsson J, Börjesson M. Measurement of physical activity in clinical practice using accelerometers. J Intern Med (2019) 286 137-53. DOI: 10.1111/joim.12908 51. Wight D, Wimbush E, Jepson R. Six steps in quality intervention development (6squid). J Epidemiol Community Health (2016) 70 520-5. DOI: 10.1136/jech-2015-205952 52. Coggon J, Adams J. “Let them choose not to eat cake…”: public health ethics, effectiveness and equity in government obesity strategy. Future Healthc J (2021) 8 49-52. DOI: 10.7861/fhj.2020-0246 53. Adams J, Mytton O, White M. Why are some population interventions for diet and obesity more equitable and effective than others? the role of individual agency. PLoS Med (2016) 13. DOI: 10.1371/journal.pmed.1001990 54. Marchant E, Todd C, Stratton G. The daily mile: whole-school recommendations for implementation and sustainability. A mixed-methods study. PLoS One (2020) 15. DOI: 10.1371/journal.pone.0228149 55. McDonagh JE, Bateman B. “ nothing about us without us ”: considerations for research involving young people. Arch Dis Child Educ Pract Ed (2012) 97 55-60. DOI: 10.1136/adc.2010.197947 56. Ng K, Cooper J, McHale F. Barriers and facilitators to changes in adolescent physical activity during COVID-19. BMJ Open Sport Exerc Med (2020) 6. DOI: 10.1136/bmjsem-2020-000919
--- title: Chitosan oligosaccharide improves ovarian granulosa cells inflammation and oxidative stress in patients with polycystic ovary syndrome authors: - Qi Xie - Wenli Hong - Yuan Li - Shuyi Ling - Ziqiong Zhou - Yuqing Dai - Wenbo Wu - Ruoxin Weng - Zhisheng Zhong - Jun Tan - Yuehui Zheng journal: Frontiers in Immunology year: 2023 pmcid: PMC10016348 doi: 10.3389/fimmu.2023.1086232 license: CC BY 4.0 --- # Chitosan oligosaccharide improves ovarian granulosa cells inflammation and oxidative stress in patients with polycystic ovary syndrome ## Abstract ### Introduction Polycystic Ovary Syndrome (PCOS) is the most common reproductive endocrine disorder among women of reproductive age, which is one of the main causes of anovulatory infertility. Even though the rapidly developed assisted reproductive technology (ART) could effectively solve fertility problems, some PCOS patients still have not obtained satisfactory clinical outcomes. The poor quality of oocytes caused by the abnormal follicular development of PCOS may directly contribute to the failure of ART treatment. Ovarian granulosa cells (GCs) are the most closely related cells to oocytes, and changes in their functional status have a direct impact on oocyte formation. Previous studies have shown that changes in the ovarian microenvironment, like oxidative stress and inflammation, may cause PCOS-related aberrant follicular development by impairing the physiological state of the GCs. Therefore, optimizing the ovarian microenvironment is a feasible method for enhancing the development potential of PCOS oocytes. ### Methods In this study, we first detected the expression of inflammatory-related factors (TGF-β1, IL-10, TNFα, IL-6) and oxidative stress-related factors (HIF-1α and VEGFA), as well as the proliferation ability and apoptosis level of GCs, which were collected from control patients (non-PCOS) and PCOS patients, respectively. Subsequently, human ovarian granulosa cell line (KGN) cells were used to verify the anti-inflammatory and anti-oxidative stress effects of chitosan oligosaccharide (COS) on GCs, as well as to investigate the optimal culture time and concentration of COS. The optimal culture conditions were then used to culture GCs from PCOS patients and control patients. ### Results The results showed that GCs from PCOS patients exhibited obvious inflammation and oxidative stress and significantly reduced proliferation and increased apoptosis. Furthermore, COS can increase the expression of anti-inflammatory factors (TGF-β1 and IL-10) and decrease the expression of pro-inflammatory factors (TNFα and IL-6), as well as promote the proliferation of GCs. Moreover, we found that COS can reduce the level of reactive oxygen species in GCs under oxidative stress by inhibiting the expression of HIF-1α and VEGFA and by suppressing the apoptosis of GCs induced by oxidative stress. ### Conclusion We find that inflammation and oxidative stress exist in the GCs of PCOS patients, and COS can reduce these factors, thereby improving the function of GCs. ## Introduction Polycystic ovary syndrome (PCOS) is the most common reproductive endocrine and metabolic disorder leading to infertility in women of childbearing age. It is mainly manifested as hyperandrogenemia (clinical and biochemical), ovarian dysfunction, and polycystic ovary morphology, usually associated with insulin resistance (IR) and obesity. Recently, PCOS was found to be accompanied by oxidative stress and chronic low-grade inflammation. Ovarian dysfunction remains the main feature of the syndrome and the leading cause of anovulatory infertility. The prevalence of the condition is approximately contributing to 10-$13\%$ of women of reproductive age and up to $30\%$-$60\%$ of patients with ovulatory dysfunctional infertility [1]. In recent years, assisted reproductive technology (ART) has developed rapidly and has become an important way to solve infertility problems in PCOS patients. However, some PCOS patients still do not achieve satisfactory clinical outcomes after ART treatment, which is mainly attributed to the pathological changes of abnormal follicular development [2]. However, the related molecular regulatory mechanisms are still not fully understood. Therefore, exploring the molecular mechanisms of abnormal follicle development in PCOS and finding ways to improve follicle quality in PCOS patients are key nodes and potential applications for improving clinical pregnancy outcomes in ART. During the normal follicular development process, granulosa cells (GCs) are the most important cells, which directly affect the development of the oocyte. Several studies have shown that the GCs dysfunction can result in abnormal follicular development in women with PCOS [3]. It is worth noting that aberrant changes in the ovarian microenvironment, such as oxidative stress and inflammation, can impair the physiological state of GCs, which may also be the main cause of subsequent PCOS follicular development disorders [4, 5]. Previous studies have found significantly higher levels of inflammation factors and reactive oxygen species (ROS) in PCOS patients’ ovaries when compared with normal women [6, 7], suggesting that the ovarian microenvironment in PCOS patients is in a low-grade chronic inflammation [8]. Further studies have confirmed that the increased level of inflammation in follicular fluid can cause GCs dysfunction, which disturb the normal development of oocytes [9]. In addition, the increased oxidative stress also induces the apoptosis of GCs [10]. A high level of ROS can damage the biological function of GCs, which results in the poor follicle quality [11]. All the above studies suggest that inflammation and oxidative stress in GCs are likely to be the crucial reasons causing abnormal follicle development in PCOS and that alleviating the inflammation and oxidative stress in GCs may improve follicular development. Chitosan oligosaccharide (COS) is mainly derived from crustaceans and also exists in fungi, insects, cell membranes of algae and cell walls of higher plants. Various biological effects of COS have been reported, such as immunomodulatory [12], anti-tumor [13], antibacterial, antifungal [14, 15], antioxidant [16] and anti-inflammatory [17]. Several studies have reported that COS can act as a potent free radical scavenger to balance biomolecules in oxidative and antioxidant processes in cellular systems and inhibit intracellular ROS formation [18, 19]. Additionally, COS inhibits the inflammatory response of macrophages induced by lipopolysaccharide (LPS) and reduces the expression of pro-inflammatory cytokines TNF-α and IL-6 [13]. More importantly, our team found that COS has a protective effect on hydrogen peroxide- (H2O2-) stimulated oxidative damage in human ovarian granulosa cell line (KGN) [20], and COS can promote the proliferation of ovarian germ stem cells and reshape the ovarian function by improving the ovarian microenvironment and stimulating the secretion of immune related factors [21]. In this study, we firstly compared the clinical data of PCOS patients and normal patients, then collected GCs from both groups to explore the correlation between GCs and inflammation and oxidative stress in PCOS patients. Because of the number of GCs obtained after collection and purification were limited, and they had a limited lifespan in vitro under the stimulation of supraphysiological doses of FSH and HCG in vivo, therefore, we used KGN cells to investigate the ameliorative effect of COS on inflammation and oxidative stress. KGN is a tumor cell derived from human granulosa cells. It not only has the ability of infinite proliferation, but also has the normal biological function and biological activity contained in normal GCs. KGN as a tool to study granulosa cells has been proved by many studies, and many people use KGN as a tool to study the function and role of GCs [22, 23]. Finally, GCs from PCOS patients and control patients were cultured with COS to verify its ameliorating effect on inflammation and oxidative stress in PCOS related GCs. ## Clinical specimens Follicular fluid was obtained from patients receiving in vitro fertilization or intracytoplasmic sperm injection (ICSI) in Reproductive Medicine Center of Jiangxi Provincial Maternal and Child Health Hospital. After the follicles were fully developed, human chorionic gonadotropin (HCG) was given to promote ovulation and eggs were retrieved 36h later. All samples were collected with written informed consent. Patients with PCOS were diagnosed according to Rotterdam criteria (60 cases). patients with infertility due to fallopian tube factors or male factors were assigned to the normal group (60 cases). ## Isolation, extraction, and culture of GCs The follicular fluid collected after oocyte pick-up (OPU) was centrifuged to discard the supernatant, and the remaining cell precipitates were suspended in the same volume DMEM/F12 culture medium. The cell suspension was transferred to the surface of $50\%$ Percoll (GE) separation solution at a volume ratio of 2:3, and GCs were obtained after centrifugation. Erythrocytes are removed with erythrocyte lysate. Then, the cells are used for culture, protein or RNA extraction, or frozen at -80°C. For culture, the cells were suspended in DMEM/F12 medium containing $10\%$ FBS in a 12-well plate and placed in cell incubator (37°C, $5\%$ CO2). After 24 hours, the culture medium was replaced with a fresh medium. ## EDU cell proliferation detection The isolated GCs are collected in EP tubes and the PBS are removed by centrifugation. 100 μL of 50 μM EDU medium are added to each tube and incubated for 2 hours. The mediums are discarded, and the cells are washed once or twice with PBS for 5 mins each time, then are stained after incubation with EDU and incubated with PBS containing $4\%$ paraformaldehyde for 30 mins at room temperature. The cells are washed with PBS for 5 mins and the PBS was discarded. 100 μL of osmolyte was added and incubated for 10 min, the cells are washed once with PBS for 5 mins. 100 μL of 1X Apollo® staining reaction solution are added and incubated for 30 mins at room temperature in a shaker protected from light. DNA staining: add 1X Hoechst33342 reaction solution, incubates for 30 mins at room temperature in a light-proof shaker, discards the staining reaction solution, then washes 1~3 times with PBS, and immediately observe and takes pictures with a fluorescent microscope. ## Detection of apoptosis by TUNEL The naturally dried GCs were incubated with cell fixative for 15 mins at room temperature and then the fixative was removed, incubated with deionized water for 5 mins, and then the deionized water was removed. Permeant was added and incubated for 10 mins at room temperature, then permeant was removed and washed twice with deionized water for 5 mins each. Add 50 μL of 1X TdT Buffer per sample to cover the cells, leave at room temperature for 10 mins and discard, then add 50 μL of TdT enzyme incubation solution and incubate at 37°C for 2 hours. Add 100 μL of 2X SSC and leave at room temperature for 15 mins to terminate the reaction, discard the SSC. Add an appropriate amount of PBS and wash the samples twice for 5 mins each time. Add 100 μL of 1X DAPI reaction solution per sample and incubate for 5 mins at room temperature. Add 100 μL 1X DAPI reaction solution, incubate for 30 mins at room temperature avoiding light in a shaker and discard the staining reaction solution, wash the samples three times with PBS for 5 mins each time, followed immediately by fluorescence microscopy for observation and counting of photographs. ## Protein extraction and immunoblotting RIPA lysate was used to extract GCs proteins, and the extracted proteins were added to the loading buffer and boiled at 95°C for 5min. After that, the extracted proteins were subjected to immunoprotein blotting according to the instructions. The main antibodies used in this study included TGF-β1 (Abcam, AB92486), IL-10 (Abcam, AB34843), TNFα (Abcam, AB6671), IL-6 (Abcam, Ab6672), HIF-1α (Cell Signaling, #3716), VEGFA (Proteintech, 66828-1-LG), and GAPDH (Proteintech, 60004-1-LG), and all secondary antibodies were purchased from Immunoway. ## RNA extraction and RT-PCR The total RNA of GCs was extracted by Trizol method, and the RNA was reverse-transcribed according to the instructions of TaKaRa kit. The cDNA obtained from reverse transcription was amplified to obtain △CT value. The results were calculated and compared using 2-[A-B]-[C-D] (A: average CT value of target genes in the experimental group, B: average CT value of reference genes in the experimental group, C: Average CT value of target genes in the control group, D: average CT value of reference genes in the control group). The primer serial numbers are as follows in Table 1. Table 1 \| Gene \| Forward primer (5’-3’) \| Reverse primer (5’-3’) \| \| --- \| --- \| --- \| \| GAPDH \| ACATCGCTCAGACACCATG \| TGTAGTTGAGGTCAATGAAGGG \| \| TGF-β1 \| CTAATGGTGGAAACCCACAACG \| TATCGCCAGGAATTGTTGCTG \| \| TNF-α \| CCTCTCTCTAATCAGCCCTCTG \| GAGGACCTGGGAGTAGATGAG \| \| IL-10 \| GACTTTAAGGGTTACCTGGGTTG \| TCACATGCGCCTTGATGTCTG \| \| IL-6 \| ACTCACCTCTTCAGAACGAATTG \| CCATCTTTGGAAGGTTCAGGTTG \| \| HIF-1α \| ATCCATGTGACCATGAGGAAATG \| TCGGCTAGTTAGGGTACACTTC \| \| VEGA \| AGGGCAGAATCATCACGAAGT \| AGGGTCTCGATTGGATGGCA \| ## Cell proliferation assay After the cells were digested and centrifuged, the cells were repeatedly suspended with an appropriate amount of culture medium, and then the total number of cells was calculated after sample counting under a microscope, and then the cell suspension was diluted to a density of 3×104/mL. The cell suspension was inoculated into 96-well plates (100 μL/well), and the wells were divided into blank group (no cells but only medium), negative control group (same volume of medium without COS), and experimental group (COS concentration of 100, 200, 300 μg/mL medium) according to the experimental design, and five replicate wells were set up for each group. The plates were incubated (37°C, $5\%$ CO2) for a period of time according to the experimental design, serum-free medium containing $10\%$ CCK8 was prepared before each assay, and 100μL of the configured medium was added to each well, and the whole process was carried out under light-proof conditions. The plates were incubated for 2 hours under light-proof conditions. After incubation, the OD value of the samples at 450 nm was measured by a microplate reader immediately. The measured OD value was subtracted from the blank group as the final measured value, and the cell proliferation curve was drawn by statistical software. ## Flow cytometry The cultured GCs were digested with trypsin without EDTA, and the cells were centrifuged and washed twice with pre-cooled PBS at 300 g at 2-8 °C for 5 min. Cells were suspended in 400 μL 1× Annexin V binding solution at a concentration of approximately 1×106 cells/mL. 5 μL Annexin V-FITC staining solution was added to the cell suspension, and then the cells were gently mixed and incubated at 2-8 °C under dark conditions for 15 mins. Add 5-10μL PI staining solution, mix it gently, and incubate for 5 mins at 2-8 °C under dark conditions. Then, the solution was detected by flow cytometry immediately. ## Reactive oxygen species detection DCFH-DA was diluted with a serum-free medium at 1:1000 to reach a final concentration of 10μM. Remove the cell culture medium and add an appropriate volume of diluted DCFH-DA, which should be sufficient to cover the cells. A ROSUP control stimulus was added to the positive control group at a ratio of 1:1000. The cells were placed in 37°C culture incubator and incubated for 20 mins. Cells were washed three times with serum-free cell culture medium to adequately remove free DCFH-DA, followed by observation and analysis by fluorescence microscopy. ## Statistical analysis SPSS 26.0 software was used to analyze the data. Independent T test was used to analyze the clinical data conforming to the normal distribution, and non-parametric test was used to analyze the clinical data conforming to the normal distribution ($25\%$ and $75\%$ percentile were used to represent the indicators not conforming to the normal distribution). $P \leq 0.05$ was statistically significant. All experiments were repeated at least three times for each group. Data were statistically analyzed by GraphPad Prism6.0 software, and analysis method was ANOVA and T test. $P \leq 0.05$, $P \leq 0.01$ and $P \leq 0.001$ were statistically significant. ## Clinical information analysis of PCOS patients The clinical characteristics of patients were shown in Table 2. BMI, age, AMH, basal LH, basal P, basal T, number of antral follicles in left and right ovary, total number of harvested eggs, MIIcleavage rate and high-quality embryo rate were significantly different between the control group and PCOS group ($P \leq 0.05$). The MIIcleavage rate and excellent embryo rate of MII in control group were significantly higher than those in PCOS group. There were no significant differences in basal FSH and basal E2 between the control and PCOS groups ($P \leq 0.05$). Table 2 \| Unnamed: 0 \| Control group \| PCOS group \| Z/t \| p \| \| --- \| --- \| --- \| --- \| --- \| \| Number of cases \| 60 \| 60 \| \| \| \| BMI \| 22.29 ± 3.24 \| 23.64 ± 3.49 \| -2.303 \| 0.023 \| \| FSH(IU/L) \| 4.96 ± 1.23 \| 4.82 ± 1.40 \| 0.581 \| 0.562 \| \| Age \| (28,33) \| (24,30.25) \| -4.216 \| 0.0 \| \| AMH \| (2.2,3.67) \| (4.98,12.98) \| -6.719 \| 0.0 \| \| LH(mIU/ml) \| (2.285,4.645) \| (3.93,11.235) \| -5.063 \| 0.0 \| \| E2(pg/ml) \| (28.5,54.75) \| (31.075,46.0) \| -0.588 \| 0.557 \| \| P(ng/dl) \| (0.2,0.3) \| (0.1,0.3) \| -2.672 \| 0.008 \| \| T(ng/dl) \| (23.415,34.97) \| (33.43,58.15) \| -5.374 \| 0.0 \| \| Sinus follicles(L) \| (6,10) \| (10,12) \| -6.57 \| 0.0 \| \| Sinus follicles(R) \| (6,10) \| (10,12) \| -6.566 \| 0.0 \| \| Number of oocytes retrieved \| (10,17) \| (12,23.25) \| -3.422 \| 0.001 \| \| Cleavage rate \| (0.721,1.0) \| (0.713,0.780) \| -2.418 \| 0.016 \| \| Normal fertilization rate \| (0.667,0.90) \| (0.634,0.877) \| -1.136 \| 0.256 \| \| Rate of good quality embryo at day 3 \| (0.143,0.369) \| (0,0.280) \| -2.983 \| 0.003 \| ## Expression of factors associated with inflammation and oxidative stress in GCs from PCOS and control group We found that the expression of anti-inflammatory factors TGF-β1 and IL-10 in GCs of PCOS patients was significantly lower than those in control patients. In contrast, PCOS patients showed a higher expression of pro-inflammatory factors TNF-α and IL-6 mRNA and protein when compared to control patients. Furthermore, the expressions of oxidative stress-related factors HIF-1α and VEGFA were also found to be significantly higher in GCs of PCOS patients than those in control patients, indicating that PCOS related GCs were accompanied with inflammation and oxidative stress (Figure 1). Figure 1: Expression of factors associated with inflammation and oxidative stress in GCs from PCOS (P) and control (C) group. Compared to the control group, the expression of IL10 and TGF-β1, which were related to anti-inflammatory, was significantly decreased in GCs from PCOS group (n=25 to 50 in each group). Meanwhile, the expression of pro-inflammatory cytokines (TNFα and IL-6) and oxidative stress-related factors (HIF-1α and VEGFA) was remarkably increased in GCs from PCOS group when compared with control group (n=25 to 46 in each group). (P<0.05, P<0.01, P<0.001). ## Proliferation and apoptosis of GCs from PCOS and control group We investigated the proliferation and apoptosis of GCs isolated from two groups. As shown in Figure 2, GCs in PCOS group represented lower proliferation(A-B) and higher apoptosis levels compared to control group(C-F), revealing that the growth ability of GCs was remarkably lower in PCOS group. Figure 2: Proliferation and apoptosis of GCs from PCOS and control group. The proliferation ability of granulosa cells in PCOS group was significantly lower than that in control group (A, B), n=5 in each group. The apoptosis level of granulosa cells in PCOS group was significantly higher than that in control group (C–F), n=5 in each group. P<0.05, *P<0.01 vs Control group. ## COS improves inflammatory response and oxidative stress in KGN cells We treated KGN cells with COS to evaluate the effect of COS on inflammatory and oxidative stress. The experiment of cell culture medium containing COS was divided into four groups. The Control group was the KGN cell culture group alone, 100 μg/mL group, 200 μg/mL group and 300 μg/ml group were the KGN cells cultured in COS medium containing 100 μg/mL, 200μg/mL and 300μg/mL concentrations in the medium. After being cultured at 24 hours and 36 hours, proteins were extracted to detect the protein expression of inflammatory factors. We found that COS promoted the expression of anti-inflammatory cytokines TGF-β1 and IL-10 in KGN cells in a dose-aging relationship, and the most obvious effect was at 300 μg/mL for 36 hours (Figures 3A, B). Meanwhile, COS inhibited the expression of pro-inflammatory cytokines TNF-α and IL-6 in KGN cells in a dose-aging relationship. The most pronounced effect was observed at a concentration of 300 μg/mL for 36 hours (Figures 3C, D). Figure 3: COS improves inflammatory and oxidative stress in KGN cells. COS promotes the expression of anti-inflammatory factors TGF-β1 and IL-10 protein in KGN cells in a dose and time-effect relationship. (Protein expression of TGF- β1 and IL-10 after COS culture for 24 hours (A); Protein expression of TGF- β1 and IL-10 after COS culture for 36 hours (B). COS inhibits the expression of pro-inflammatory factors TNF-α and IL-6 in KGN cells in a dose and time-dependent relationship. (Protein expression of TNF-α and IL-6 after COS culture for 24 hours (C); Protein expression of TNF-α and IL-6 after COS culture for 36 hours (D). Determination of IC50 Value of Hydrogen Peroxide Damage to KGN Cells (E). COS reduces H2O2-induced reactive oxygen generation in KGN cells (F). COS inhibits the protein expression of HIF-1α and VEGFA in KGN cells in a dose-dependent manner (G). P<0.05, P<0.01, P<0.001 vs Control group; # P<0.05, ## P<0.01 vs Control group; # P<0.05, ## P<0.01 vs H2O2 group. In order to explore the ameliorative effect of COS on GCs oxidative stress, we established an H2O2-induced GCs oxidative stress model to simulate the GCs oxidative stress environment of PCOS patients. In this study, KGN cells were induced by different concentrations of H2O2 (100, 200, 300, 400, 500, 600μmol/L). CCK-8 cell viability experiment showed that the survival rate of KGN cells decreased logarithmically with the increase of H2O2 concentration. Survival curves were drawn according to the survival rates of KGN cells with different concentrations of H2O2, and the IC50 value was 300μmol/L. The oxidative damage model of KGN cells was established using an IC50 concentration of H2O2 = 300 μmol/L to start subsequent grouping experiments (Figure 3E). Based on the established oxidative stress injury model, the protein expression of HIF-1α and VEGFA in KGN cells was detected after 24 hours culture with COS (100 μg/mL, 200 μg/mL, 300 μg/mL). The results showed that the expression of HIF-1α and VEGFA in the injury model group was significantly increased compared with that in the Control group, and the expression of HIF-1α and VEGFA decreased in a concentration-dependent manner after the addition of COS (Figure 3F). Based on the established oxidative stress injury model, KGN cells were cultured with COS (100 μg/mL, 200 μg/mL, 300 μg/mL) for 24 hours, and then the ROS Assay Kit was used to detect the level of ROS in KGN cells. The results showed that the fluorescence intensity of the damage model group was significantly higher than that of the control group, and the fluorescence intensity decreased with the addition of COS, especially when the COS was 200 μg/mL and 300 μg/mL (Figure 3G). These results indicate that COS can reduce the production of reactive oxygen species induced by H2O2 in KGN cells. ## COS promotes KGN cells proliferation and inhibits apoptosis Cells from the above four groups were subjected to cell proliferation assay with OD at 450 λ, and analysis of variance by repeated measurements showed that cell proliferation was influenced by treatment duration and COS concentration (Table 3). Compared with the Control group, COS (100 μg/mL, 200 μg/mL, 300 μg/mL) promoted the proliferation of GCs in a dose-dependent manner. The OD values of 200 μg/mL and 300 μg/mL groups were significantly different from those of Control and 100μg/mL groups at 48 hours points (Figure 4A). On the basis of the established model of oxidative stress injury, KGN cells were cultured with COS (100 μg/mL, 200 μg/mL, 300 μg/mL) for 24 hours. The results showed that H2O2 significantly increased the apoptosis rate of KGN cells, 200 μg/mL and 300 μg/mL COS significantly reduced the apoptosis rate of KGN cells. This suggests that COS can reverse cell damage caused by oxidative stress (Figure 4B). ## COS improves ovarian GCs inflammation and oxidative stress in PCOS patients The optimal COS concentration (300 μg/mL) and optimal culture time (36 hours) were used to culture the extracted GCs. We found that COS promoted the expression of anti-inflammatory factors TGF-β1 and IL-10, and inhibited the expression of pro-inflammatory factors TNFα and IL-6 in GCs of PCOS patients (Figure 5A). Meanwhile, the expression of oxidative stress-related factors HIF-1α and VEGFA was also decreased in GCs of PCOS patients (Figure 5B). Using ROS Assay Kit to detect their reactive oxygen levels, the fluorescence intensity of GCs in PCOS patients was significantly higher than that in control patients, while the fluorescence intensity of GCs in PCOS patients decreased after the addition of COS (Figure 5C), indicating that COS can reduce the ROS level of GCs in PCOS patients. Figure 5: COS improves ovarian GCs inflammation and oxidative stress in PCOS patients. Effect of COS on the expression of inflammatory factors in ovarian GCs in patients with PCOS (A). COS inhibits the expression of oxidative stress related factors in ovarian GCs of patients with PCOS (B). COS decreases reactive oxygen species in GCs of patients with PCOS (C). P<0.05, P<0.01, P<0.001 vs Control group; # P<0.05, ## P<0.01 vs PCOS+300 μg/mL group. ## Discussion Polycystic ovary syndrome (PCOS) is the most common reproductive endocrine and metabolic disorder leading to infertility in women of reproductive age and is also one of the leading causes of anovulatory infertility. In this study, after analyzing clinical data collected from PCOS and control groups, we found that although the number of ovarian antrum follicles and a total number of eggs obtained were higher in PCOS patients than in the normal group, the cleavage rate was significantly lower than that in the normal group. This result suggests that there is abnormal follicular development in PCOS patients. However, the pathological causes are not fully understood. It is well known that GCs are essential for oocyte development, and GCs delivers amino acids and other nutrients to oocytes through gap junctions [24], which play a key role in important biological events such as oocyte meiosis, oocyte fertilization, and later embryonic dev elopement. Several studies have shown that the dysfunction of GCs will directly lead to abnormal follicular development in women with PCOS [3]. For example, apoptosis of GCs not only promotes follicular atresia [3], but also leads to the generation of low-quality oocytes [25, 26]. Previous studies have confirmed that chronic low-degree inflammation and oxidative stress are closely associated with the occurrence and development of abnormal PCOS follicles [27], and inflammation and oxidative stress can lead to increased apoptosis of GCs. In addition, oocyte was closely surrounded by GCs, implying important role of GCs in oocyte development. The increased apoptosis of GCs was closely related to the abnormal follicular development of PCOS. We hypothesize that the effects of inflammation and oxidative stress on follicular development in PCOS are likely to be mediated through the action of GCs. Therefore, to investigate in depth the causes of inflammation and oxidative stress leading to abnormal follicular development in PCOS, we focused on the relationship between GCs and inflammation and oxidative stress (Figure 6). Figure 6: The microenvironment of GCs in PCOS patients is characterized by inflammation and oxidative stress. COS improves the inflammation and oxidative stress of GCs and thus enhances the proliferative capacity of GCs, reduces the apoptotic level of GCs, and ultimately improves abnormal follicular development in PCOS patients. We found abnormal expression of inflammatory factors and oxidative stress-related factors in GCs of PCOS patients through the results of GCs extraction, and we found that the proliferation ability of GCs in PCOS group was weakened and the apoptosis level was increased. We found that the expression of TGF-β1 protein level was significantly lower than that of the control group, but the expression of mRNA was higher than that of the normal group. We hypothesized that the inconsistent expression of TGF-β1 gene and protein may be a consequence of the effect of ovulation induction drugs on GCs, or the inhibition of the translation of anti-inflammatory factor mRNA into protein in PCOS patients. The exact mechanisms involved need to be further explored. As key pro-inflammatory factors, abnormally elevated levels of TNF-α and IL-6 have been shown to inhibit the proliferation, differentiation, and maturation of PCOS follicles [28, 29]. It has been reported that IL10 and TGF-β had anti-inflammatory properties and played a key role in disease prevention and autoimmunity [30, 31]. Reduced expression of these two factors can result in inflammatory response, which leads to abnormal steroid synthesis and delayed follicular maturation and ovarian dysfunction. Conversely, TNF-α and IL-6, which belongs to pro-inflammatory factors, have negatively effects on follicular maturation, fertilization, embryonic implantation [29, 32]. Increased expression of TNF-α induces apoptosis of antral follicular GCs, which leads to increased follicular membrane thickness and decreased granular layer thickness [32]. In addition, IL-6 can reduce the activity of aromatase in the follicle, resulting in decreased concentration of estradiol in the follicle, fertility and fertilization ability [29]. In our study, we find a significantly lower expression of IL10 and TGF-β and a obviously higher expression of TNF-α and IL-6 in PCOS patients, which suggests that one of the reason causing the abnormal biological functions of GCs in PCOS patients are probably associated with these abnormal inflammatory related factors. Hypoxia-inducible factor-1α (HIF-1α) is a functional subunit of hypoxia-inducible factor-1 (HIF-1) that degrades rapidly under normoxic pressure. Increased ROS levels can inhibit HIF-1α ubiquitination by inhibiting the activity of the proline hydroxylase family (PHDs), thus increasing HIF-1α levels in vivo [33]. During follicular growth and development, low oxygen partial pressure in the follicular microenvironment can stimulate the expression of HIF-1α in ovarian GCs [34], which indicated an elevated ROS levels in GCs of PCOS patients. It was shown that HIF-1α plays a key role in follicular development and ovulation in mammals. HIF-1α is specifically expressed in ovarian cells and mainly in GCs, suggesting that it may be directly involved in the regulation of mammalian ovarian physiological function. In addition, this study also found that elevated gene expression of HIF-1α was detected in the ovaries of PCOS rats [35, 36]. HIF-1α is also a central regulator of hypoxia stress response, and this transcription factor primarily regulates hypoxia-induced genes, including vascular endothelial growth factor A (VEGFA) (the most dominant and important member of VEGF) used in the angiogenic response [37]. In the ovary, VEGF is mainly expressed in GCs and follicular membrane cells, but rarely expressed in mesenchymal cells [38]. VEGF production is usually present in hypoxic cells and hypoxia is one of the most potent triggers of VEGF expression, involved in the transcription, stabilization, translation and release of VEGF [39]. VEGF is a factor involved in normal reproductive function and follicular development, and it has been proved that VEGF levels in serum of PCOS patients are significantly increased [40, 41]. As mentioned previously, increased expression of TNF-α in PCOS patients induces apoptosis in sinus follicle GCs [42], and the increase of oxidative stress can also lead to the increase of the apoptosis level of GCs. From this, we inferred that the abnormal follicular development in PCOS patients is likely due to inflammation and oxidative stress affecting the proliferation and apoptosis of GCs. In conclusion, we found that inflammation and oxidative stress exist in the microenvironment of GCs in PCOS patients, and the resulting decreased proliferation ability or increased apoptosis level of GCs is likely to be one of the important reasons for abnormal follicular development of PCOS. Thus, improving inflammation and oxidative stress in GCs is crucial for follicular development of PCOS patients. KGN cells were cultured with different COS concentrations to evaluate the effect of COS on cells inflammation. We found that COS improved the proliferation and secretory function of cells in a dose-dependent manner, and this effect was most obvious at a COS concentration of 300 μg/mL for 36 hours. We assessed the ameliorative effect of COS on oxidative stress in cells by pretreatment with different concentrations of COS based on oxidative stress modeling. We found that COS increased the antioxidant and anti-apoptotic activity of cells in a dose-dependent manner. At the same time, our study showed that the COS group at 300 μg/mL still did not recover the level of the control group, indicating that COS did not completely reverse the damage of oxidative stress on cells and the reasons for this need to be further studied. It was found that COS could promote the expression of anti-inflammatory factors TGF-β1 and IL-10 in ovarian GCs of PCOS patients by culturing GCs from PCOS patients and control patients at the optimal concentration and time, Inhibited the expression of pro-inflammatory factors TNFα, IL-6, oxidative stress related factors HIF-1α and VEGFA, and decreased the level of reactive oxygen species. ## Conclusion In summary, we found inflammation and oxidative stress in the microenvironment of GCs from PCOS patients and found that GCs from PCOS patients had diminished proliferative capacity and increased levels of apoptosis. By using COS in vitro cell culture experiments, we found that COS can increase the expression of anti-inflammatory factors TGF-β1 and IL-10 and decrease the expression of pro-inflammatory factors TNFα and IL-6, as well as promote the proliferation of GCs. COS can reduce the level of reactive oxygen species in GCs under oxidative stress by inhibiting the expression of HIF-1α and VEGFA and suppressing the apoptosis of GCs induced by oxidative stress. Finding COS new pharmacological application in infertility treatment is expected to provide a new therapeutic treatment for clinical PCOS patients by enhancing the GCs proliferative capacity and decreasing their apoptosis level via improving the inflammation and oxidative stress of GCs into a suitable intensity. ## Data availability statement The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s. ## Ethics statement The studies involving human participants were reviewed and approved by Medical Ethics Committee of Jiangxi Maternal and Child Health Hospital. The patients/participants provided their written informed consent to participate in this study. ## Author contributions YZ, JT, and ZSZ developed the concept and design. QX and WH performed most of the experiments and wrote the manuscript draft. YL participated in the manuscript preparation and editing. SL, ZQZ, YD, WW, and RW performed some of the experiments and provided critical discussion of the manuscript. All authors contributed to the article and approved the submitted version. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## References 1. Bozdag G, Mumusoglu S, Zengin D, Karabulut E, Yildiz BO. The prevalence and phenotypic features of polycystic ovary syndrome: A systematic review and meta-analysis. Hum Reprod (Oxford England) (2016) 31. DOI: 10.1093/humrep/dew218 2. De Leo V, Musacchio MC, Cappelli V, Massaro MG, Morgante G, Petraglia F. Genetic, hormonal and metabolic aspects of pcos: An update. Reprod Biol endocrinology: RB&E (2016) 14 38. DOI: 10.1186/s12958-016-0173-x 3. Jakimiuk AJ, Weitsman SR, Navab A, Magoffin DA. Luteinizing hormone receptor, steroidogenesis acute regulatory protein, and steroidogenic enzyme messenger ribonucleic acids are overexpressed in thecal and granulosa cells from polycystic ovaries. J Clin Endocrinol Metab (2001) 86. DOI: 10.1210/jcem.86.3.7318 4. Nehir Aytan A, Bastu E, Demiral I, Bulut H, Dogan M, Buyru F. Relationship between hyperandrogenism, obesity, inflammation and polycystic ovary syndrome. Gynecological Endocrinol (2016) 32. DOI: 10.3109/09513590.2016.1155208 5. Lai Q, Xiang W, Li Q, Zhang H, Li Y, Zhu G. Oxidative stress in granulosa cells contributes to poor oocyte quality and ivf-et outcomes in women with polycystic ovary syndrome. Front Med (2018) 12. DOI: 10.1007/s11684-017-0575-y 6. Xiong Y, Liang X, Yang X, Li Y, Wei L. Low-grade chronic inflammation in the peripheral blood and ovaries of women with polycystic ovarian syndrome. Eur J Obstetrics Gynecology Reprod Biol (2011) 159. DOI: 10.1016/j.ejogrb.2011.07.012 7. Sak S, Uyanikoglu H, Incebiyik A, Incebiyik H, Hilali NG, Sabuncu T. Associations of serum fetuin-a and oxidative stress parameters with polycystic ovary syndrome. Clin Exp Reprod Med (2018) 45. DOI: 10.5653/cerm.2018.45.3.116 8. Yang J, Zhong T, Xiao G, Chen Y, Liu J, Xia C. Polymorphisms and haplotypes of the tgf-B1 gene are associated with risk of polycystic ovary syndrome in Chinese han women. Eur J Obstetrics Gynecology Reprod Biol (2015) 186 1-7. DOI: 10.1016/j.ejogrb.2014.11.004 9. Su Y, Sugiura K, Eppig JJ. Mouse oocyte control of granulosa cell development and function: Paracrine regulation of cumulus cell metabolism. Semin Reprod Med (2009) 27 32-42. DOI: 10.1055/s-0028-1108008 10. Wang Y, Li C, Ali I, Li L, Wang G. N-acetylcysteine modulates non-esterified fatty acid-induced pyroptosis and inflammation in granulosa cells. Mol Immunol (2020) 127. DOI: 10.1016/j.molimm.2020.09.011 11. Qiao J, Wang Z, Feng H, Miao Y, Wang Q, Yu Y. The root of reduced fertility in aged women and possible therapentic options: Current status and future perspects. Mol Aspects Med (2014) 38 54-85. DOI: 10.1016/j.mam.2013.06.001 12. Tsvetkov YE, Paulovičová E, Paulovičová L, Farkaš P, Nifantiev NE. Synthesis of biotin-tagged chitosan oligosaccharides and assessment of their immunomodulatory activity. Front Chem (2020) 8. DOI: 10.3389/fchem.2020.554732 13. Dario Rafael OH, Luis Fernándo ZG, Abraham PT, Pedro Alberto VL, Guadalupe GS, Pablo PJ. Production of chitosan-oligosaccharides by the chitin-hydrolytic system of trichoderma harzianum and their antimicrobial and anticancer effects. Carbohydr Res (2019) 486. DOI: 10.1016/j.carres.2019.107836 14. Mukhtar Ahmed KB, Khan MMA, Siddiqui H, Jahan A. Chitosan and its oligosaccharides, a promising option for sustainable crop production- a review. Carbohydr Polymers (2020) 227. DOI: 10.1016/j.carbpol.2019.115331 15. Yang T, Chou C, Li C. Antibacterial activity of n-alkylated disaccharide chitosan derivatives. Int J Food Microbiol (2005) 97. DOI: 10.1016/S0168-1605(03)00083-7 16. Lan R, Li Y, Chang Q, Zhao Z. Dietary chitosan oligosaccharides alleviate heat stress-induced intestinal oxidative stress and inflammatory response in yellow-feather broilers. Poultry Sci (2020) 99. DOI: 10.1016/j.psj.2020.09.050 17. Hsu S, Yang C, Tsai H, Lin C, Fang Y, Shieh C. Chitosan oligosaccharides suppress nuclear factor-kappa b activation and ameliorate experimental autoimmune uveoretinitis in mice. Int J Mol Sci (2020) 21. DOI: 10.3390/ijms21218326 18. Xu W, Huang H, Lin C, Jiang Z. Chitooligosaccharides protect rat cortical neurons against copper induced damage by attenuating intracellular level of reactive oxygen species. Bioorganic Medicinal Chem Lett (2010) 20. DOI: 10.1016/j.bmcl.2010.03.105 19. Yi Z, Luo X, Zhao L. Research advances in chitosan oligosaccharides: From multiple biological activities to clinical applications. Curr Medicinal Chem (2020) 27. DOI: 10.2174/0929867326666190712180147 20. Yang Z, Hong W, Zheng K, Feng J, Hu C, Tan J. Chitosan oligosaccharides alleviate H2o2-stimulated granulosa cell damage. Oxid Med Cell Longevity (2022) 2022. DOI: 10.1155/2022/4247042 21. Huang Y, Ye H, Zhu F, Hu C, Zheng Y. The role of chito-oligosaccharide in regulating ovarian germ stem cells function and restoring ovarian function in chemotherapy mice. Reprod Biol endocrinology: RB&E (2021) 19 14. DOI: 10.1186/s12958-021-00699-z 22. Yi S, Zheng B, Zhu Y, Cai Y, Sun H, Zhou J. Melatonin ameliorates excessive PINK1/Parkin-mediated mitophagy by enhancing SIRT1 expression in granulosa cells of PCOS. Am J Physiol Endocrinol Metab (2020) 319 E91-E101. DOI: 10.1152/ajpendo.00006.2020 23. Liu Y, Liu H, Li Z, Fan H, Yan X, Liu X. The release of peripheral immune inflammatory cytokines promote an inflammatory cascade in PCOS patients via altering the follicular microenvironment. Front Immunol (2021) 12. DOI: 10.3389/fimmu.2021.685724 24. Sugiura K, Pendola FL, Eppig JJ. Oocyte control of metabolic cooperativity between oocytes and companion granulosa cells: Energy metabolism. Dev Biol (2005) 279 20-30. DOI: 10.1016/j.ydbio.2004.11.027 25. Zhang J, Xu Y, Liu H, Pan Z. Micrornas in ovarian follicular atresia and granulosa cell apoptosis. Reprod Biol endocrinology: RB&E (2019) 17. DOI: 10.1186/s12958-018-0450-y 26. Fan Y, Chang Y, Wei L, Chen J, Li J, Goldsmith S. Apoptosis of mural granulosa cells is increased in women with diminished ovarian reserve. J Assisted Reprod Genet (2019) 36. DOI: 10.1007/s10815-019-01446-5 27. Alanbay I, Ercan CM, Sakinci M, Coksuer H, Ozturk M, Tapan S. A macrophage activation marker chitotriosidase in women with pcos: Does low-grade chronic inflammation in pcos relate to pcos itself or obesity. Arch Gynecology Obstetrics (2012) 286. DOI: 10.1007/s00404-012-2425-0 28. Artimani T, Karimi J, Mehdizadeh M, Yavangi M, Khanlarzadeh E, Ghorbani M. Evaluation of pro-Oxidant-Antioxidant balance (Pab) and its association with inflammatory cytokines in polycystic ovary syndrome (Pcos). Gynecological Endocrinol (2018) 34. DOI: 10.1080/09513590.2017.1371691 29. Nuñez-Calonge R, Cortés S, Gutierrez Gonzalez LM, Kireev R, Vara E, Ortega L. Oxidative stress in follicular fluid of young women with low response compared with fertile oocyte donors. Reprod Biomedicine Online (2016) 32. DOI: 10.1016/j.rbmo.2015.12.010 30. Saraiva M, O'Garra A. The regulation of il-10 production by immune cells. Nat Rev Immunol (2010) 10. DOI: 10.1038/nri2711 31. Vural P, Değirmencioğlu S, Saral NY, Akgül C. Tumor necrosis factor alpha (-308), interleukin-6 (-174) and interleukin-10 (-1082) gene polymorphisms in polycystic ovary syndrome. Eur J Obstetrics Gynecology Reprod Biol (2010) 150. DOI: 10.1016/j.ejogrb.2010.02.010 32. Mohammadi S, Kayedpoor P, Karimzadeh-Bardei L, Nabiuni M. The effect of curcumin on tnf-A, il-6 and crp expression in a model of polycystic ovary syndrome as an inflammation state. J Reprod Infertility (2017) 18 33. Jaakkola P, Mole DR, Tian YM, Wilson MI, Gielbert J, Gaskell SJ. Targeting of hif-alpha to the Von hippel-lindau ubiquitylation complex by O2-regulated prolyl hydroxylation. Sci (New York NY) (2001) 292. DOI: 10.1126/science.1059796 34. Nishimura R, Okuda K. Hypoxia is important for establishing vascularization during corpus luteum formation in cattle. J Reprod Dev (2010) 56. DOI: 10.1262/jrd.09-162e 35. Wang F, Zhang Z, Wang Z, Xiao K, Wang Q, Su J. Expression and clinical significance of the hif-1a/Et-2 signaling pathway during the development and treatment of polycystic ovary syndrome. J Mol Histol (2015) 46. DOI: 10.1007/s10735-015-9609-4 36. Zhang J, Zhang Z, Wu Y, Chen L, Luo Q, Chen J. Regulatory effect of hypoxia-inducible factor-1α on hcg-stimulated endothelin-2 expression in granulosa cells from the pmsg-treated rat ovary. J Reprod Dev (2012) 58. DOI: 10.1262/jrd.2012-089 37. Giovanni Artini P, Monteleone P, Parisen Toldin MR, Matteucci C, Ruggiero M, Cela V. Growth factors and folliculogenesis in polycystic ovary patients. Expert Rev Endocrinol Metab (2007) 2. DOI: 10.1586/17446651.2.2.215 38. Artini PG, Ruggiero M, Parisen Toldin MR, Monteleone P, Monti M, Cela V. Vascular endothelial growth factor and its soluble receptor in patients with polycystic ovary syndrome undergoing ivf. Hum Fertility (Cambridge England) (2009) 12. DOI: 10.1080/14647270802621358 39. Pagès G, Pouysségur J. Transcriptional regulation of the vascular endothelial growth factor gene–a concert of activating factors. Cardiovasc Res (2005) 65. DOI: 10.1016/j.cardiores.2004.09.032 40. Kaczmarek MM, Schams D, Ziecik AJ. Role of vascular endothelial growth factor in ovarian physiology - an overview. Reprod Biol (2005) 5 41. Peitsidis P, Agrawal R. Role of vascular endothelial growth factor in women with pco and pcos: A systematic review. Reprod Biomedicine Online (2010) 20. DOI: 10.1016/j.rbmo.2010.01.007 42. Qiao J, Feng HL. Extra- and intra-ovarian factors in polycystic ovary syndrome: Impact on oocyte maturation and embryo developmental competence. Hum Reprod Update (2011) 17 17-33. DOI: 10.1093/humupd/dmq032
--- title: 'Spinal disinhibition: evidence for a hyperpathia phenotype in painful diabetic neuropathy' authors: - Anne Marshall - Alise Kalteniece - Maryam Ferdousi - Shazli Azmi - Edward B Jude - Clare Adamson - Luca D’Onofrio - Shaishav Dhage - Handrean Soran - Jackie Campbell - Corinne A Lee-Kubli - Shaheen Hamdy - Rayaz A Malik - Nigel A Calcutt - Andrew G Marshall journal: Brain Communications year: 2023 pmcid: PMC10016414 doi: 10.1093/braincomms/fcad051 license: CC BY 4.0 --- # Spinal disinhibition: evidence for a hyperpathia phenotype in painful diabetic neuropathy ## Abstract The dominant sensory phenotype in patients with diabetic polyneuropathy and neuropathic pain is a loss of function. This raises questions as to which mechanisms underlie pain generation in the face of potentially reduced afferent input. One potential mechanism is spinal disinhibition, whereby a loss of spinal inhibition leads to increased ascending nociceptive drive due to amplification of, or a failure to suppress, incoming signals from the periphery. We aimed to explore whether a putative biomarker of spinal disinhibition, impaired rate-dependent depression of the Hoffmann reflex, is associated with a mechanistically appropriate and distinct pain phenotype in patients with painful diabetic neuropathy. In this cross-sectional study, 93 patients with diabetic neuropathy underwent testing of Hoffmann reflex rate-dependent depression and detailed clinical and sensory phenotyping, including quantitative sensory testing. Compared to neuropathic patients without pain, patients with painful diabetic neuropathy had impaired Hoffmann reflex rate-dependent depression at 1, 2 and 3 Hz (P ≤ 0.001). Patients with painful diabetic neuropathy exhibited an overall loss of function profile on quantitative sensory testing. However, within the painful diabetic neuropathy group, cluster analysis showed evidence of greater spinal disinhibition associated with greater mechanical pain sensitivity, relative heat hyperalgesia and higher ratings of spontaneous burning pain. These findings support spinal disinhibition as an important centrally mediated pain amplification mechanism in painful diabetic neuropathy. Furthermore, our analysis indicates an association between spinal disinhibition and a distinct phenotype, arguably akin to hyperpathia, with combined loss and relative gain of function leading to increasing nociceptive drive. Marshall et al. report that patients with painful diabetic neuropathy and evidence of spinal disinhibition, demonstrated by impairment of Hoffman reflex rate-dependent depression, display a mechanistically relevant pain phenotype. This phenotype, akin to hyperpathia, exhibits a combined loss and relative gain of function leading to increased nociceptive drive. ## Graphical Abstract Graphical Abstract ## Introduction Diabetic peripheral neuropathy (DPN) is characterized by an array of ‘negative’ and ‘positive’ sensory symptoms in which numbness may paradoxically coexist with prickling, stabbing, burning or aching pain.1 These painful sensations may occur in response to normally innocuous stimuli (allodynia), because of increased sensitivity to painful stimuli (hyperalgesia) or arise spontaneously. The symptoms that dominate can vary dramatically between patients, and it is not yet established whether signs and symptoms change with progression of neuropathy.2 The recent systematic application of sensory phenotyping using quantitative sensory testing (QST) has enabled stratification of patients with DPN into clusters of characteristics and has been proposed to potentially segregate patients based on underlying physiological mechanisms.3-6 It has been widely argued that ‘dying back’ nerve fibre degeneration and/or nerve fibre regeneration may contribute to the development of pain in diabetic and other forms of neuropathic pain by causing nociceptors to become abnormally spontaneously active or to become sensitized (i.e. respond more vigorously to a given stimulus or to a lower strength of stimulus).7-10 These features could contribute to spontaneous pain, such as burning pain, as well as hyperalgesia and allodynia. However, preclinical studies have reported diminished release of excitatory neurotransmitters in the spinal cord of diabetic rats during periods of stimulus-evoked behavioral hyperalgesia,11,12 and QST studies have reported that only a small proportion of patients with painful DPN fit into the ‘irritable nociceptor’ sensory phenotype.3,4,13 Indeed, loss of function consistent with a ‘deafferentation’ phenotype is the most commonly demonstrated QST profile in patients with DPN.3,4,13 Whilst variability exists within and across these broad sensory phenotypes, suggesting diversity of pain generation or modulation mechanisms, it remains unclear as to how a deafferentation phenotype, with apparent loss of function of nociceptive pathways, leads to pain. It is increasingly accepted that pro-nociceptive pathophysiological changes occur within the spinal cord in diabetes that could generate or maintain pain. Temporal summation of pain (wind-up)14 or alterations in descending pain modulation15-17 have been investigated. Whilst imaging studies suggest enhanced descending facilitation,17 evidence of these mechanisms in pain generation in patients with DPN has been inconsistent.4,18,19 A further potential mechanism of interest is spinal disinhibition, whereby inappropriate amplification of, or failure to suppress, incoming signals from the periphery leads to facilitation of ascending nociceptive drive—a process that could generate pain despite peripheral loss of function. In diabetic rodents, spinal disinhibition results from a brain-derived neurotrophic factor-dependent reduction in the expression of potassium chloride co-transporter 2 in the dorsal horn of the spinal cord, leading to a shift in the function of ionotropic GABA-A receptors from inhibitory towards excitatory.20 A biomarker for spinal disinhibition in diabetic rodents is impaired Hoffmann reflex rate-dependent depression (HRDD).21 Importantly, interventions targeting the underlying mechanisms of spinal disinhibition in rats both normalize impairments in HRDD and ameliorate behavioral manifestations of pain.22 We have recently translated these experimental findings to the clinical setting by demonstrating impairment of HRDD in subjects with painful DPN,22,23 indicating that spinal disinhibition may be a dominant pain mechanism in a proportion of these patients. It is currently unknown whether patients with painful DPN and impaired HRDD have a distinct pain phenotype reflecting spinal disinhibition or whether impairment of HRDD is associated with other mechanisms facilitating ascending spinal nociceptive information such as wind-up or impaired descending pain modulation. To further investigate the relationship between spinal disinhibition and pain phenotype, we have explored QST somatosensory profiles and conditioned pain modulation in conjunction with HRDD in a cohort of patients with diabetes, with and without neuropathic pain. ## Materials and methods This was an observational cross-sectional study. Research Ethics Committee approval was granted (East Midlands—Leicester South Research Ethics Committee reference 17/EM/0076), and written informed consent was obtained from each participant. Study conduct adhered to the tenets of the Declaration of Helsinki. Consecutive patients attending secondary care diabetes clinics at Manchester University NHS Foundation Trust and Tameside and Glossop Integrated Care NHS Foundation Trust between November 2017 and February 2020 were invited to take part in the study. Participants underwent assessment during a single research visit. ## Study participants Ninety-three patients with type 1 or type 2 diabetes were recruited into the study. The majority ($$n = 90$$/93) were recruited from a previously reported cohort.23,24 Detailed demographic data including age, gender and ethnicity along with type and duration of diabetes, co-morbidities, medication, height, weight, blood pressure, HbA1c, lipids and renal function were documented. Participants found to have other common causes of neuropathy based on a family history as well as testing for serum B12, folate, immunoglobulins, electrophoresis and anti-nuclear antibody were excluded from the study. ## Neuropathy and pain questionnaires Participants completed five questionnaires. The Neuropathy Symptom Profile (NSP), a yes or no questionnaire that documents sensory, autonomic and motor symptoms, including weakness, has been validated in patients with DPN25 and found to be particularly useful in recognizing patterns of symptoms. The Small Fibre Neuropathy and Symptom Inventory Questionnaire (SFN-SIQ) was used to assess the presence of sensory and autonomic symptoms including changes in sweating patterns, diarrhoea, constipation, urinary tract problems, dry eyes, dry mouth, dizziness, hot flushes, palpitations, sensitive leg skin, restless legs, burning feet and sheet intolerance. The Diabetic Neuropathy Symptom Score (DNS) was used as a diabetes-specific simplified scoring system assessing pain, numbness, tingling and ataxia26 with any score above 0 representing an abnormality. The Neuropathy Pain Scale (NPS), a 0-10 pain rating scale, was used to define the severity of symptoms based on patient responses to questions about pain intensity and pain descriptors, for example, sharp, dull, hot, cold, skin sensitivity and itch. Participants were asked to mark on three visual analogue scales (VAS) current pain, average pain over the past 24 h and worst pain over the past 24 h. ## Nerve conduction and H-reflex studies Nerve conduction and H-reflex studies were performed using a DANTEC Keypoint system (Dantec Dynamics Ltd., Bristol, UK). Participants were semi-recumbent at 45° with limb temperature maintained between 32° and 35°. Sural sensory amplitude and conduction velocity along with peroneal motor nerve amplitude and conduction velocity were recorded. For H-reflex studies, tibial nerve stimulation was performed using 1-ms square wave monophasic pulses delivered using surface silver–silver chloride electrodes, to the popliteal fossa. Surface silver–silver chloride recording electrodes with a diameter of 9 mm were placed on the long axis of soleus (Fig. 1). H-reflex recruitment curves were obtained to determine peak–peak H-reflex maximal amplitude by incrementing stimulation current by 1 mA (1-ms duration). A random inter-stimulation interval with a minimum of 10 s was observed. For HRDD, a submaximal stimulus strength (to achieve a response of $75\%$ of maximum H-reflex on the rising phase of the recruitment curve) was used. H-wave responses were recorded in trains of ten stimuli delivered at 1–3 Hz. HRDD was calculated as the mean H-reflex amplitude of responses 2–5 of a stimulus train, expressed as a percentage of the amplitude of the first recorded H-reflex in the train. Therefore, a higher value of HRDD indicates a smaller degree of depression than a lower value and vice versa. The average of stimulus responses 2–5 was used as this has been shown to be the optimal value to discriminate between patients with painful and painless DPN.24 Figure 1: A schematic representation for eliciting and recording the H-reflex (Created with BioRender.com). ## Corneal confocal microscopy Corneal confocal microscopy (CCM) was used to quantify corneal small nerve fibre pathology and has been validated against the current gold standard of intraepidermal nerve fibre density.27 Images of the corneal sub-basal nerve plexus were captured using the Heidelberg Retina Tomograph 3 with Rostock Cornea Module (Heidelberg Eye Explorer, Heidelberg Engineering GmBH, Heidelberg, Germany) following an established protocol.28 For image analysis, six representative images (three per eye) were selected by M.F. and A.K., who were blinded to participant status. Corneal nerve fibre density [total number of main nerves per square millimetre (no./mm2)], corneal nerve fibre length [total length of main nerves and nerve branches per square millimetre (mm/mm2)] and corneal nerve branch density [total number of branches per square millimetre (no./mm2)] were quantified. ## Quantitative sensory testing A full QST battery, representing seven tests assessing 13 parameters, was performed on all patients with diabetes using the standardized DFNS testing protocol.29 The investigators (A.M. and A.G.M.) underwent formal training at the University of Mannheim prior to commencing this study. Tests for thermal sensation were performed at the beginning of the testing paradigm, prior to mechanical assessments. The thermal sensory testing device (TSA-II NeuroSensory Analyser Medoc, Ltd., Ramat-Yishai, Israel) was positioned on the skin on the dorsum of the right foot. Cold detection threshold (CDT) and warm detection threshold (WDT) along with cold pain threshold (CPT) and heat pain threshold (HPT) were recorded. The threshold was determined as the arithmetic mean of three results using the difference between measured threshold and baseline temperature (32°C) for CDT and WDT and absolute temperature for CPT and HPT. Testing of the thermal sensory limen was also performed and calculated subtracting the arithmetic mean of the CDT from the arithmetic mean of the WDT. Paradoxical heat sensations were recorded. Mechanical detection threshold was assessed using standardized Von Frey hairs (0.25, 0.5, 1, 2, 4, 8, 16, 32, 64, 128, 256 and 512 mN Opti-hair2-Set, Marstock Nervtest, Germany) and calculated using a modified method of limits (geometric mean of five supra and subthreshold stimulus responses). Mechanical pain threshold (MPT), mechanical pain sensitivity (MPS) and wind-up ratio were all assessed using a set of seven pinprick stimulators with standardized intensities (8, 16, 32, 64, 128, 256 and 512 mN). MPT was calculated using a modified method of limits (geometric mean of five supra and subthreshold stimulus responses). The degree of MPS was calculated using the geometric mean of pain ratings for pinprick stimuli, and wind-up ratio was calculated as the arithmetic mean of the pain intensity rating for the series of stimuli divided by the arithmetic mean of the pain intensity rating for the single stimulus. Dynamic mechanical allodynia, the degree of pain sensitivity to innocuous stimuli, was assessed on the dorsum of the right foot using a cotton wisp (exerting a force of 3 mN), a Q-tip (exerting a force of 100 mN) and a soft brush (exerting a force of between 200 and 400 nM), applied in a balanced order and pain ratings recorded. Dynamic mechanical allodynia was calculated as geometric mean of pain ratings. Vibration detection thresholds were recorded using a tuning fork (Rydel Seiffer 64 Hz with fixed weights) over the medial malleolus with the threshold determined by the arithmetic mean of the three values. Pressure pain thresholds were recorded using a pressure algometer (FDN200, Wagner Instruments, USA) with a blunt contact area of 1 cm² placed on the skin above the abductor hallucis muscle. The threshold was determined as the arithmetic mean of the three recordings. The raw QST data from each test were log transformed and converted into z-scores (with exception of paradoxical heat sensations and dynamic mechanical allodynia) to normalize the data for age, sex and body site tested. This transformation enables comparison between cohorts and DFNS reference data29 and allows for the identification of specific QST profiles. Positive z-score values denote a gain in function, and negative z-scores denote a loss of function in each of the parameters. We calculated a value for mechanical pain differential as (z-score for MPS) − (z-score for MPT). This pain differential (MPS-MPT) gave us a value that represents the ‘relative’ gain and loss of function for mechanical pinprick. A high pain differential score represents patients who have a high MPS ‘relative’ to MPT. We also calculated a value for thermal pain differentials: (z-score for CPT) − (z-score for CDT) termed CPT-CDT and (z-score for HPT) − (z-score for WDT) termed HPT-WDT. ## Conditioned pain modulation Conditioned pain modulation requires intact descending pathways and has been shown to be attenuated in patients with chronic pain.15,30 Pressure pain threshold on the right abductor pollicis brevis was used as the test stimulus. A pressure algometer (FDN200, Wagner Instruments, USA) with a blunt contact area of 1 cm² was placed on the skin above the abductor hallucis muscle on the right hand. Pressure was applied with increasing intensity at a rate of 0.5 kg (50 kPa)/s. The patient was asked to indicate as soon as the sensation of pressure changed to an additional painful ‘burning’, ‘stinging’ or ‘aching’ sensation and the value on the algometer recorded. The test was repeated three times with a break of 10 s in between and mean value recorded. A conditioning stimulus using noxious cold was then administered. The left hand of the patient was immersed up to the wrist in a water bath of melting ice water for up to 180 s. The patient was asked to rate how painful this was (0–100) every 15 s. When the patient could no longer tolerate it, their hand was removed from the water bath and the time noted. The test stimuli were then repeated on the right hand (non-submerged) as detailed above and the level of pain intensity rated again. The conditioned pain modulation effect was calculated as the difference (post conditioning stimulus minus pre) in pressure pain thresholds. A positive value indicates efficient conditioned pain modulation. ## Statistical methods Statistical analyses were performed using Prism 9 statistical software (GraphPad Software Inc, La Jolla, CA, USA) and IBM SPSS 29 (cluster analysis). Data were tested for normality with the Shapiro–Wilk test of normality. Categorical data were analysed using chi-square Fisher’s exact test of association. Parametric data were analysed using unpaired t-test to compare means between two groups. Results were reported as mean ± standard deviation. Non-parametric data were analysed using Mann–Whitney test between two groups. Results were reported as median with interquartile range. A P-value of <0.05 considered significant. Correlations were performed using Spearman’s rank test and expressed as a coefficient (r) with P-values. A Bonferroni correction was applied to account for multiple comparisons resulting in a significant P-value of 0.0016. A k-means clustering algorithm (SPSS) was used to further investigate associations within the pain cohort, grouping the data set based on HRDD, QST parameters and pain descriptors from the NPS questionnaire. ## Results A total of 93 patients with DPN, 37 with neuropathic pain and (VAS > 0) and 56 without neuropathic pain (VAS = 0) were recruited. Within the painful DPN group, seven patients were taking medication to treat neuropathic pain (2 × duloxetine, 3 × gapapentinoids and 2 × tricyclics). An additional five patients were taking selective serotonin reuptake inhibitors. Coincidentally, within the painless DPN group, two patients were taking tricyclics. Current, average and maximum pain scores did not differ significantly between patients with type 1 or type 2 diabetes. There was no significant difference for age, gender, ethnicity, body mass index (BMI) and type or duration of diabetes between the pain and no pain cohorts (Table 1). The HbA1c was significantly ($$P \leq 0.018$$) higher in patients with painless DPN compared to patients with painful DPN. Table 1 \| Unnamed: 0 \| DPN with pain (n = 37) \| DPN without pain (n = 56) \| \| --- \| --- \| --- \| \| Type of diabetes (1/2) \| 11/26 \| 21/35 \| \| Gender (female/male) \| 17/20 \| 19/37 \| \| Ethnicity (White/Asian/Black) \| 28/7/2 \| 41/12/3 \| \| \| Median (interquartile range) \| Median (interquartile range) \| \| Age (years) \| 62 (53–71.5) \| 65 (52.5–71) \| \| Duration (years) \| 15 (9–22) \| 16 (10–23) \| \| HbA1c (mmol/mol) \| 53.5 (46.6–57.3)* \| 58.0 (34.0–69.0) \| \| BMI (kg/m²) \| 29.2 (25.5–31.8) \| 27.7 (24.6–32) \| \| NSP \| 5.0 (3.25–10)* \| 2.0 (0.5–3.5) \| \| SNAP (µV) \| 7.5 (3–15.5) \| 6.8 (4–11.8) \| \| SNCV (m/s) \| 41.2 (38.9–47.9) \| 43.1 (40–46.7) \| \| PMNAP (mV) \| 3.6 (2.3–5.6) \| 3.5 (2.5–5.4) \| \| PMNCV (m/s) \| 41.2 (37.4–43.5) \| 40.9 (38.6–44.1) \| \| CNFD (no./mm²) \| 24.48 (18.49–28.39) \| 26.04 (18.75–30.21) \| \| CNFL (mm/mm²) \| 17.75 (13.58–21.25) \| 16.93 (13.33–21.25) \| \| CNBD (no./mm²) \| 49.48 (31.51–87.5) \| 40.62 (23.96–58.85) \| \| VAS pain current (0–100) \| 14.0 (4.25–30.75) \| 0 \| \| VAS pain av past 24 h \| 35.5 (15.5–65.5) \| 0 \| \| VAS pain max past 24 h \| 51.0 (26.5–76.5) \| 0 \| \| \| Mean ± SD \| Mean ± SD \| \| HRDD mean H2–5 @ 1 Hz \| 64.83 ± 22.60* \| 36.42 ± 16.69 \| \| HRDD mean H2-5 @ 2 Hz \| 52.51 ± 28.28* \| 30.41 ± 15.31 \| \| HRDD mean H2-5 @ 3 Hz \| 53.00 ± 26.19* \| 28.22 ± 14.32 \| ## Neuropathy assessments and questionnaires There were no significant differences in data obtained from nerve conduction studies and corneal confocal microscopy between the cohorts of patients with and without neuropathic pain. Of all subjects studied, 22 patients (10 with neuropathic pain and 12 without neuropathic pain) had both nerve conduction parameters within local normative values and CCM parameters within the previously published normative range.31Figure 2A and B show the distribution of descriptor ratings on the NPS and VAS pain scores reported by patients with painful DPN. Figure 2C shows the number of patients, with DPN with and without pain, reporting neuropathy symptoms on the DNS. The NSP, DNS and SFN-SIQ were significantly (all $P \leq 0.001$) higher in patients with neuropathic pain compared to those without pain. Symptoms related to pain and hypersensitivity represented the highest proportion of divergent scores between the two groups, with dry eyes, dry mouth, changes in sweating and dizziness on standing increased in the pain cohort (Fig. 2D). Of note, a small number of patients in the group without neuropathic pain reported burning sensation, sensitive skin, sheet intolerance and restless legs (Fig. 2D) that were not described as painful by these patients. Figure 2: Pain scales and descriptors. (A) NPS in patients with painful DPN. The coloured bars represent the distribution of patient responses for each score (0–10). (B) VAS pain score in patients with painful DPN for current, average and maximum pain during the past 24 h. The coloured segments represent the proportion of patient responses within each category. (C) Diabetic Neuropathy Symptom Score in patients with DPN with (magenta bars) and without (blue bars) neuropathic pain. (D) SFN-SIQ in patients with DPN with (magenta bars) and without (blue bars) neuropathic pain. Statistically significant P-values shown (Mann–Whitney U test). ## Painful and painless DPN is associated with a loss of function sensory profile Individual z-scores for QST parameters are summarized in Fig. 3A and B. Whilst the mean z-score for all parameters fell within the normative range of DFNS control data,29,32 there was evidence of a loss of function for innocuous and noxious thermal and mechanical detection thresholds as well as for MPS. The z-scores for wind-up ratio and pressure pain threshold were within the normative range. Figure 4 shows the proportion of patients with z-scores outside the DFNS normative range. Approximately $20\%$ of patients in both the painful and painless DPN cohorts exhibited abnormal loss of function (z-score > −1.96) in mechanical detection and MPT. A smaller number (∼$10\%$) from both cohorts showed abnormal loss of function in thermal detection thresholds. A small minority of patients from both cohorts exhibited a gain in function (z-score > +1.96) in CPT, HPT, MPS, wind-up ratio and pressure pain threshold. The painful DPN cohort showed a significantly greater loss of function of CDT ($$P \leq 0.047$$), mechanical detection threshold ($$P \leq 0.037$$) and MPT ($$P \leq 0.013$$) compared to patients with painless DPN indicating greater cold and mechanical hypoesthesia. Dynamic mechanical allodynia was present to a greater extent in patients with neuropathic pain but did not reach a level of significance (Fig. 3C). There was no significant difference in the presence of paradoxical heat sensations between the two groups (Fig. 3C). The level of conditioned pain modulation did not differ significantly between the two groups. Figure 3: Sensory profile. (A) Scatter plot and mean ± 95% confidence interval (CI) of z-scores for thermal quantitative sensory testing parameters in patients with DPN with (magenta dots) and without (blue dots) neuropathic pain. (B) Scatter plot and mean ± 95% CI of z-scores for mechanical quantitative sensory testing parameters in patients with DPN with (magenta dots) and without (blue dots) neuropathic pain. (C) Dynamic mechanical allodynia and paradoxical heat sensations in patients with DPN with (magenta dot) and without (blue dot) neuropathic pain. Statistically significant P-values shown (Mann–Whitney U test). CDT, cold detection threshold; CPT, cold pain threshold; DMA, dynamic mechanical allodynia; HPT, heat pain threshold; MDT, mechanical detection threshold; MPS, mechanical pain sensitivity; MPT, mechanical pain threshold; PHS, paradoxical heat sensation; PPT, pressure pain threshold; TSL, thermal sensory limen; VDT, vibration detection threshold; WDT, warm detection threshold; WUR, wind-up ratio. Figure 4: Loss and gain of sensory function. Comparison of patients with DPN with (magenta) and without (blue) neuropathic pain who have QST values outside the 95% confidence interval of the German research network of neuropathic pain reference database. Statistically significant P-values shown (Fisher’s exact test). CDT, cold detection threshold; CPT, cold pain threshold; DMA, dynamic mechanical allodynia; HPT, heat pain threshold; MDT, mechanical detection threshold; MPS, mechanical pain sensitivity; MPT, mechanical pain threshold; PHS, paradoxical heat sensation; PPT, pressure pain threshold; TSL, thermal sensory limen; VDT, vibration detection threshold; WDT, warm detection threshold; WUR, wind-up ratio. Across the whole cohort, age was negatively correlated with thermal, mechanical and vibration detection z-scores, even after adjustment for the duration of diabetes, indicating a loss of function with increasing age in patients with diabetes that was over and above the z-score transformation to account for age. A greater loss of function of thermal and mechanical detection parameters was associated with increasing large and small fibre neuropathy. ## Patients with painful DPN show impaired HRDD HRDD recordings were available from 82 patients (Supplementary Fig. 1) as 11 patient recordings were incomplete or had technically compromised stimulus response trains. Between group analysis of HRDD (Fig. 5) was consistent with our previously reported findings.23 Thus, HRDD was significantly impaired in patients with painful DPN when compared to patients with painless DPN at 1, 2 and 3 Hz (all P ≤ 0.001) (Fig. 5). Amongst all patients with DPN ($$n = 82$$), there were no significant correlations between HRDD and individual QST z-scores or with conditioned pain modulation (Supplementary Table 1). Figure 5: H-reflex rate-dependent depression. HRDD at 1, 2 and 3 Hz in patients with DPN with (magenta) and without (blue) pain. Statistically significant P-values shown (unpaired t-test). ## Increasing impairment of HRDD is associated with relative thermal and mechanical hyperalgesia Patients with painful DPN demonstrated significantly ($$P \leq 0.013$$) greater loss of function in MPT compared to patients with painless DPN, indicating that patients with painful DPN required a stronger stimulus to feel pinprick as painful compared to patients with painless DPN. MPS did not differ significantly between the painful and painless DPN groups. However, patients with the most impaired HRDD amongst those with painful DPN also showed the most gain, or least loss, of function in mechanical pain reporting (Supplementary Table 2). To further investigate differences in sensory phenotypes within the pain cohort that could reflect spinal disinhibition as a dominant pain mechanism, we divided patients according to their HRDD status: those with HRDD above 2 SD of the mean of patients with DPN and no pain ($$n = 11$$) and an equivalent number of patients with painful DPN and the most efficient HRDD. This approach was used as no patients with painful DPN demonstrated efficient HRDD outside 2 SD of patients with painless DPN. Figure 6A shows the QST profiles of these two cohorts of patients (see also Supplementary Table 3). Thermal detection and thermal pain thresholds for patients with the most efficient HRDD showed a similar degree of loss of function (Fig. 6A orange boxes). In contrast, patients with impaired HRDD demonstrated relatively less loss of function in thermal pain thresholds (Fig. 6A, aqua boxes). Therefore, patients with impaired HRDD require a greater temperature change to initially detect heat/cold, but once perceived, it rapidly becomes painful. The mean z-score for MPT was comparable in both groups; most patients exhibited loss of function with reduced ability in detecting a sharp sensation. Patients in the pain group with relatively unimpaired HRDD also demonstrated a large loss of function in MPS. However, in patients in the pain cohort with the most impaired HRDD, MPS was relatively preserved with a less pronounced loss of function, but without a gain of function (Fig. 6A, yellow box). In a separate analysis, patients amongst all groups were also divided into two cohorts: HRDD above mean +2 SD of the control group ($$n = 14$$) and HRDD below mean −2 SD of the control group ($$n = 11$$), using our previously published normative data from healthy control participants.24 This resulted in similar findings (Supplementary Fig. 2A). Figure 6: Hyperpathia profile in patients with painful DPN and impaired HRDD. (A) QST profile for patients with painful DPN and impaired HRDD (red circles/line) and patients with painful DPN and intact HRDD (blue circles/line). Orange boxes highlight z-scores for thermal detection and pain thresholds in patients with painful DPN and intact HRDD. Aqua boxes highlight z-scores for thermal detection and pain thresholds in patients with painful DPN and impaired HRDD. Yellow box highlights z-scores for mechanical pain threshold and sensitivity in both patients with painful DPN with intact and impaired HRDD. Purple box highlights z-scores for vibration detection thresholds in both patients with painful DPN with intact and impaired HRDD. (B) Mechanical and thermal pain differentials in patients with painful DPN and impaired HRDD (red) and intact HRDD (blue). P-values are shown (unpaired t-test). CDT, cold detection threshold; CPT, cold pain threshold; HPT, heat pain threshold; MDT, mechanical detection threshold; MPS, mechanical pain sensitivity; MPT, mechanical pain threshold; PPT, pressure pain threshold; TSL, thermal sensory limen; VDT, vibration detection threshold; WDT, warm detection threshold; WUR, wind-up ratio. Although not statistically significant, these findings led us to explore the difference between the first sensation of sharpness/pain and, once felt, the pain scores attributed to this sharpness as well as the difference in thermal pain and detection thresholds for cold and heat. ## Mechanical pain differential = mechanical pain sensitivity − mechanical pain threshold (MPS-MPT) This pain differential gives a value that represents the ‘relative’ gain and loss of function for mechanical pinprick. A high pain differential score represents patients who have a high MPS ‘relative’ to their MPT. There was no significant difference in MPS-MPT between the pain and no-pain groups. However, MPS-MPT was significantly ($$P \leq 0.023$$) higher in patients displaying the most impaired HRDD compared to those with the most efficient HRDD (Supplementary Fig. 2B). Amongst the pain group, MPS-MPT was also higher in patients with the most impaired HRDD compared to those with relatively unimpaired HRDD, although this was not significant ($$P \leq 0.0589$$) (Fig. 6B). Both within the pain cohort and across the whole patient group, increasing MPS-MPT values were associated with increasing impairment of HRDD [pain cohort: MPS-MPT and HRDD at 3 Hz (rs = 0.414, $$P \leq 0.028$$) (Fig. 7A); all patients: 1 Hz (rs = 0.232, $$P \leq 0.039$$) and 3 Hz (rs = 0.231, $$P \leq 0.047$$) (Supplementary Table 1)]. Although not significant following Bonferroni correction, a greater impairment of HRDD is associated with higher ratings for pinprick-evoked pain relative to detection thresholds for pinprick pain. Figure 7: Correlation graphs. Scatterplots demonstrating hyperpathia (A–D) phenotype in patients with DPN. (A) Mechanical pain differential (MPS-MPT) and H-reflex rate-dependent depression [HRDD (H2–5)] at 3 Hz. Red dots: Cluster 1. Orange dots: Cluster 2. X in highlighted circles indicates centroids of each cluster. (B) Mechanical pain differential (MPS-MPT) and NPS 3 (burning pain). (C) NPS 3 (burning pain) and H-reflex rate-dependent depression [HRDD (H2-5)] at 3 Hz. (D) Heat pain differential (HPT-WDT) and H-reflex rate-dependent depression [HRDD (H2-5)] at 3 Hz. ## Cold pain differential = cold pain threshold − cold detection threshold There was no significant difference in CPT-CDT between the pain and no pain groups. Amongst the pain group, there was no significant difference in CPT-CDT between patients with the most impaired HRDD compared to those with unimpaired HRDD (Fig. 6B). There was no correlation between CPT-CDT and HRDD at any frequency. However, within the pain group, a gain of function in CPT was associated with increasing impairment of HRDD: 1 Hz (rs = 0.377, $$P \leq 0.048$$); 2 Hz (rs = 0.423, $$P \leq 0.031$$); however, this was not significant following Bonferroni correction (Supplementary Table 2). ## Heat pain differential = heat pain threshold − warm detection threshold There was no significant difference in HPT-WDT between the pain and no-pain groups. Amongst the pain group HPT-WDT was significantly higher in patients with the most impaired HRDD compared to those with relatively unimpaired HRDD ($$P \leq 0.0226$$) (Fig. 6B). Within the pain cohort, increasing HPT-WDT values were associated with increasing impairment of HRDD (HPT-WDT and HRDD at 2 Hz (rs = 0.552, $$P \leq 0.003$$) and 3 Hz (rs = 0.524, $$P \leq 0.007$$) (Fig. 7D), although not significant following Bonferroni correction. Therefore, amongst patients with painful DPN, a greater impairment of HRDD was associated with heat stimuli being felt as painful at lower temperatures relative to innocuous heat detection. ## Impairment of HRDD and accompanying relative mechanical hyperalgesia are associated with higher ratings of symptomatic burning pain Within the cohort of patients with painful DPN, increasing impairment of HRDD and increasing values of MPS-MPT were associated with increasing reported scores for burning pain on the NPS questionnaire: HRDD at 3 Hz (rs = 0.389, $$P \leq 0.037$$) (Fig. 7C); MPS-MPT (rs = 0.484, $$P \leq 0.002$$) (Fig. 7B). Although not significant following Bonferroni correction, these associations, along with the earlier described associations between HRDD and MPS-MPT, lend support to the presence of a mechanistically relevant relationship between these parameters. Furthermore, MPS-MPT showed a significant correlation with ratings for intensity of pain on the NPS (rs = 0.547, $P \leq 0.001$) (Supplementary Table 2). ## Cluster analysis The large number of parameters in the correlation analysis ($$n = 38$$), resulted in a Bonferroni corrected significance value $P \leq 0.001$, to account for multiple comparisons. This highly conservative correction runs the risk of a type 2 error and overlooking potential associations. To further explore the phenotypic manifestations of spinal disinhibition, we used a k-means cluster analysis technique that separates data sets for maximal similarity within clusters and dissimilarity between clusters, using 26 parameters (HRDD/QST/NPS pain descriptors) for segregation into two clusters. Five parameters significantly discriminated between the two clusters: HRDD at 1, 2 and 3 Hz, MPS-MPT and HPT-WDT. NPS3 (burning pain) was close to being significant (Supplementary Table 4A). Cluster 1 ($$n = 11$$) exhibited impaired HRDD at all three frequencies, along with a relatively high value for both MPS-MPT (high MPS relative to their MPT) and HPT-WDT (heat hyperalgesia relative to innocuous heat detection) and higher reported scores for burning pain (NPS3). Cluster 2 ($$n = 21$$) exhibited intact HRDD at all three frequencies, along with MPS-MPT and HPT-WDT values close to 0, indicating a lack of relative mechanical and thermal hyperalgesia respectively (Fig. 7A). The mean values for each parameter within the two clusters can be seen in the final cluster centres table (Supplementary Table 4B). ## Discussion A current major goal of research in clinical pain is to develop individualized treatment strategies based on presumptive or identifiable mechanisms of pain to increase efficacy and reduce side effects. We have previously shown that cohorts of patients with painful DPN have impaired HRDD, a biomarker of spinal disinhibition, when compared to patients with painless DPN and control subjects.22-24 However, individual patients with painful DPN exhibit HRDD values that vary markedly. A major objective of the current study was to explore whether impairment in HRDD was associated with a distinct pain phenotype. By deep phenotyping of patients using neuropathy symptom and pain questionnaires combined with QST, we demonstrated that patients with painful DPN exhibiting the most impaired HRDD (and hence greater spinal disinhibition) showed (i) greater MPS, especially when compared to mechanical pain detection, (ii) relative heat hyperalgesia when compared to innocuous warm detection, and (iii) higher ratings of spontaneous burning pain. These initial findings raise the intriguing possibilities that not only this is impaired HRDD in painful DPN associated with a distinct pain phenotype but also this phenotype is mechanistically appropriate for spinal disinhibition. In line with previous studies,4,13 the dominant QST profile for our patients with DPN was that of loss of function. Whilst only a small minority of patients had thermal threshold parameters outside the normal range, approximately one in five demonstrated loss of function for tests of mechanical sensation. Nerve conduction parameters significantly correlated with the multiple of QST z-scores indicating a relationship between large fibre neuropathy and loss of function. In addition, greater corneal nerve fibre loss was associated with thermal hypoesthesia and mechanical hypoalgesia. We have shown a less marked loss of function than previously demonstrated in large multi-centre sensory phenotyping studies in DPN,4,13 which most likely reflects the less severe neuropathy in our cohort of patients. Indeed, ∼$50\%$ of patients in the Pain in Neuropathy Study (PiNS) had absent sural sensory nerve action potentials. As we were investigating the role of HRDD, our recruitment deliberately targeted patients most likely to have an adequate H-reflex and hence less severe neuropathy. However, of note, the severity of neuropathy was comparable between painful and painless cohorts, allowing valid phenotype comparison between the groups. A predicted impact of spinal disinhibition (impaired HRDD) is that a given peripheral input to the dorsal horn of the spinal cord will be less suppressed than in the normal state or even facilitated. Within the painful DPN cohort, we found positive correlations between HRDD, MPS and differential scores for both mechanical and heat thresholds. Whilst correlation coefficients were moderate, this exploratory analysis reveals a consistent relationship suggesting that spinal disinhibition may be associated with a combined sensory detection loss and hyperalgesia profile. In support of these findings, cluster analysis revealed that the presence of a spinal disinhibition sub-group in which impaired HRDD is clustered with mechanical and heat differentials and burning pain. Patients with painful DPN required greater pinprick force to detect a stimulus as painful. Accordingly, in the absence of central facilitation/lack of suppression, one might expect lower pain scores during the stimulus response function. However, in patients with the most impaired HRDD, all of whom had painful DPN, mechanical pain differentials were higher than in patients with the most efficient HRDD. These findings indicate that when patients with impaired HRDD detect a painful punctate mechanical stimulus, they rate that stimulus as more painful than patients do with preserved/relatively preserved spinal inhibition. It is important to note that ratings of pain intensity relative to stimulus detection were only tested in the mechanical domain and the equivalent tests for cold and heat pain were not performed. In this respect, future hypothesis-driven studies that compare thermal pain threshold and pain ratings as well as the mechanical differential will be of interest. However, despite having equivalent impairment in detection thresholds for innocuous heat, patients with painful DPN who had greater impairment of HRDD showed a relative gain in function for heat pain detection (i.e. a relative heat hyperalgesia) compared to patients with painful DPN and efficient HRDD. These initial findings, which are arguably akin to hyperpathia, indicate that patients with spinal disinhibition may have a pain phenotype consistent with spinal amplification/reduced suppression. Furthermore, these exploratory findings provide a potential mechanism by which patients with DPN and an apparent deafferented phenotype develop neuropathic pain that can be detected psychophysically. Defined as ‘a painful syndrome characterized by an abnormally painful reaction to a stimulus, especially a repetitive stimulus, as well as an increased threshold’ (IASP), hyperpathia is not easily captured with QST and therefore may be underestimated.33 Classification of patients based on their sensory profile or phenotype has previously defined three clusters: those with sensory loss, thermal hyperalgesia and mechanical hyperalgesia.5 However, it is likely that these groups are not distinct, with additional nuanced phenotypes to be further defined.5,34 Indeed, our cluster analysis did not discriminate between the two clusters (with and without impaired HRDD) based on any individual QST parameters. However, the mechanical and heat differential scores were significant segregators. Therefore, the current findings do reveal a consistent theme linking spinal disinhibition to a combined sensory detection loss and hyperalgesia profile. However, future larger scale studies that enable hierarchical cluster analysis of QST sensory profiles will be needed to test this hypothesis.35 Neuropathic pain and particular phenotypic profiles are likely to relate to a complex interplay in the balance between peripheral inputs and central processing. Impairment of HRDD was not associated with a sensory profile suggestive of an irritable nociceptor phenotype. Both within the painful DPN group and across all patients with DPN, there was no significant relationship between HRDD and magnitude of wind-up or between HRDD and conditioned pain modulation. Whilst this suggests the mechanisms underlying spinal disinhibition are not directly related to other spinal processes that putatively result in central pain amplification (wind-up/temporal facilitation or dysfunctional descending pain modulation), it does not exclude an interaction or competition between these mechanisms. For example, an appropriate descending inhibitory control signal acting on a disinhibited spinal cord dorsal horn could be rendered ineffective or even facilitate ascending nociceptive drive. Interestingly, in the current study, the level of conditioned pain modulation measured with a pressure algometer did not differ significantly between patients with DPN with and without pain. This is consistent with recent findings obtained utilizing mechanical test stimuli.19 Indeed, the latter study also showed that conditioned pain modulation was unexpectedly more efficient in patients with painful DPN when noxious heat was used as a test stimulus.19 Further work incorporating different measures of the descending pain modulation system will be required to explore these potential interactions. The circuitry and pharmacology of HRDD exhibits considerable complexity and can be modified by a number of factors.36 It is also possible that other neurophysiological processes within the spinal cord might act concomitantly to modify spinal disinhibition, HRDD and nociceptive signalling in the dorsal horn of the spinal cord. However, there is currently no evidence that potential candidates, such as loss of segmental inhibition due to alterations of primary afferent depolarization, are implicated in animal models of diabetic neuropathy, and recent evidence suggests any effect of primary afferent depolarization on HRDD to be minimal.37 Limitations of this study include the collection of pain ratings by a one-time assessment rather than in a diary. Furthermore, patients continued to take their anti-neuropathic pain medication that would be expected to impact on the pain ratings. Treatments with particular anti-neuropathic pain drugs could also differentially alter HRDD that could increase variability or, by normalizing HRDD, have a tendency to underestimate the initial level of spinal disinhibition.38 Moreover, we enrolled patients with mild/moderate DPN as patients with severe polyneuropathy are likely to have an absent or inadequate H-reflex preventing an assessment of HRDD.39,40 Finally, the study is cross-sectional in nature. Longitudinal studies will be needed to assess the role of spinal disinhibition in evolving pain phenotypes or in the transition to chronic neuropathic pain as well as for the systematic evaluation of the effects of anti-neuropathic pain medications on HRDD, pain and sensory phenotype. Both HRDD and QST are non-invasive and potentially widely applicable and broadly applicable in a clinical setting. HRDD and the distinct QST profile could be utilized to identify patients with painful DPN in whom disinhibition is a primary mechanism. This could direct mechanism-led therapeutics and drug discovery. For example, pharmacological intervention studies in diabetic rodents using duloxetine normalize HRDD and diminish behavioural indices of pain.22,41 Furthermore, an initial study indicates that the degree of normalization of HRDD predicts a therapeutic response to gabapentin in patients with painful DPN.38 ## Conclusion In conclusion, we have demonstrated that patients with painful DPN have impairment of HRDD and therefore evidence of spinal disinhibition. Furthermore, our initial findings using detailed pain profiling have revealed that greater impairment of HRDD is associated with higher patient ratings for burning pain and a ‘hyperpathia’ type profile, characterized by a loss of function in mechanical and thermal detection but with relatively high pain sensitivity. Further investigations to confirm and expand these intriguing findings are needed including an exploration of the therapeutic implications of identifying impaired HRDD and the interactions of spinal disinhibition with other peripheral and centrally mediated mechanisms of pain. ## Supplementary material Supplementary material is available at Brain Communications online ## Funding This research was funded by grant 1-17ICTS-062 from the American Diabetes Association to NAC and AGM. ## Competing interests The authors report no competing interests. ## Data availability Data sets are available from the corresponding author upon request. ## References 1. Rosenberger DC, Blechschmidt V, Timmerman H, Wolff A, Treede RD. Challenges of neuropathic pain: Focus on diabetic neuropathy. J Neural Transm (Vienna) (2020.0) 127 589-624. PMID: 32036431 2. Kopf S, Groener JB, Kender Z. Deep phenotyping neuropathy: An underestimated complication in patients with pre-diabetes and type 2 diabetes associated with albuminuria. Diabetes Res Clin Pract (2018.0) 146 191-201. PMID: 30389624 3. Maier C, Baron R, Tolle TR. Quantitative sensory testing in the German research network on neuropathic pain (DFNS): Somatosensory abnormalities in 1236 patients with different neuropathic pain syndromes. Pain (2010.0) 150 439-450. PMID: 20627413 4. Themistocleous AC, Ramirez JD, Shillo PR. The Pain in Neuropathy Study (PiNS): A cross-sectional observational study determining the somatosensory phenotype of painful and painless diabetic neuropathy. Pain (2016.0) 157 1132-1145. PMID: 27088890 5. Baron R, Maier C, Attal N. Peripheral neuropathic pain: A mechanism-related organizing principle based on sensory profiles. Pain (2017.0) 158 261-272. PMID: 27893485 6. Themistocleous AC, Crombez G, Baskozos G, Bennett DL. Using stratified medicine to understand, diagnose, and treat neuropathic pain. Pain (2018.0) 159 S31-S42. PMID: 30113945 7. Sopacua M, Hoeijmakers JGJ, Merkies ISJ, Lauria G, Waxman SG, Faber CG. Small-fiber neuropathy: Expanding the clinical pain universe. J Peripher Nerv Syst (2019.0) 24 19-33. PMID: 30569495 8. Kleggetveit IP, Namer B, Schmidt R. High spontaneous activity of C-nociceptors in painful polyneuropathy. Pain (2012.0) 153 2040-2047. PMID: 22986070 9. Serra J, Duan WR, Locke C, Solà R, Liu W, Nothaft W. Effects of a T-type calcium channel blocker, ABT-639, on spontaneous activity in C-nociceptors in patients with painful diabetic neuropathy: A randomized controlled trial. Pain (2015.0) 156 2175-2183. PMID: 26035253 10. Bennett DL, Clark AJ, Huang J, Waxman SG, Dib-Hajj SD. The role of voltage-gated sodium channels in pain signaling. Physiol Rev (2019.0) 99 1079-1151. PMID: 30672368 11. Calcutt NA, Stiller C, Gustafsson H, Malmberg AB. Elevated substance-P-like immunoreactivity levels in spinal dialysates during the formalin test in normal and diabetic rats. Brain Res (2000.0) 856 20-27. PMID: 10677607 12. Malmberg AB, O'Connor WT, Glennon JC, Ceseña R, Calcutt NA. Impaired formalin-evoked changes of spinal amino acid levels in diabetic rats. Brain Res (2006.0) 1115 48-53. PMID: 16920081 13. Raputova J, Srotova I, Vlckova E. Sensory phenotype and risk factors for painful diabetic neuropathy: A cross-sectional observational study. Pain (2017.0) 158 2340-2353. PMID: 28858986 14. Herrero JF, Laird JM, López-García JA. Wind-up of spinal cord neurones and pain sensation: Much ado about something?. Prog Neurobiol (2000.0) 61 169-203. PMID: 10704997 15. Yarnitsky D, Granot M, Nahman-Averbuch H, Khamaisi M, Granovsky Y. Conditioned pain modulation predicts duloxetine efficacy in painful diabetic neuropathy. Pain (2012.0) 153 1193-1198. PMID: 22480803 16. Granovsky Y, Nahman-Averbuch H, Khamaisi M, Granot M. Efficient conditioned pain modulation despite pain persistence in painful diabetic neuropathy. Pain Rep (2017.0) 2 e592. PMID: 29392208 17. Segerdahl AR, Themistocleous AC, Fido D, Bennett DL, Tracey I. A brain-based pain facilitation mechanism contributes to painful diabetic polyneuropathy. Brain (2018.0) 141 357-364. PMID: 29346515 18. Sierra-Silvestre E, Somerville M, Bisset L, Coppieters MW. Altered pain processing in patients with type 1 and 2 diabetes: Systematic review and meta-analysis of pain detection thresholds and pain modulation mechanisms. BMJ Open Diabetes Res Care (2020.0) 8 e001566 19. Granovsky Y, Topaz LS, Laycock H. Conditioned pain modulation is more efficient in painful than in non-painful diabetic polyneuropathy patients. Pain (2021.0) 163 827-833 20. Jolivalt CG, Lee CA, Ramos KM, Calcutt NA. Allodynia and hyperalgesia in diabetic rats are mediated by GABA and depletion of spinal potassium-chloride co-transporters. Pain (2008.0) 140 48-57. PMID: 18755547 21. Lee-Kubli CA, Calcutt NA. Altered rate-dependent depression of the spinal H-reflex as an indicator of spinal disinhibition in models of neuropathic pain. Pain (2014.0) 155 250-260. PMID: 24103402 22. Marshall AG, Lee-Kubli C, Azmi S. Spinal disinhibition in experimental and clinical painful diabetic neuropathy. Diabetes (2017.0) 66 1380-1390. PMID: 28202580 23. Worthington A, Kalteniece A, Ferdousi M. Spinal inhibitory dysfunction in patients with painful or painless diabetic neuropathy. Diabetes Care (2021.0) 44 1835-1841. PMID: 34385346 24. Worthington A, Kalteniece A, Ferdousi M. Optimal utility of H-reflex RDD as a biomarker of spinal disinhibition in painful and painless diabetic neuropathy. Diagnostics (Basel) (2021.0) 11 1247. PMID: 34359330 25. Dyck PJ, Karnes J, O'Brien PC, Swanson CJ. Neuropathy Symptom Profile in health, motor neuron disease, diabetic neuropathy, and amyloidosis. Neurology (1986.0) 36 1300-1308. PMID: 3762934 26. Meijer JW, Smit AJ, Sonderen EV, Groothoff JW, Eisma WH, Links TP. Symptom scoring systems to diagnose distal polyneuropathy in diabetes: The Diabetic Neuropathy Symptom Score. Diabet Med (2002.0) 19 962-965. PMID: 12421436 27. Quattrini C, Tavakoli M, Jeziorska M. Surrogate markers of small fiber damage in human diabetic neuropathy. Diabetes (2007.0) 56 2148-2154. PMID: 17513704 28. Tavakoli M, Malik RA. Corneal confocal microscopy: A novel non-invasive technique to quantify small fibre pathology in peripheral neuropathies. J Vis Exp (2011.0) 47 2194 29. Rolke R, Baron R, Maier C. Quantitative sensory testing in the German Research Network on Neuropathic Pain (DFNS): Standardized protocol and reference values. Pain (2006.0) 123 231-243. PMID: 16697110 30. Bannister K, Dickenson AH. The plasticity of descending controls in pain: Translational probing. J Physiol (2017.0) 595 4159-4166. PMID: 28387936 31. Tavakoli M, Ferdousi M, Petropoulos IN. Normative values for corneal nerve morphology assessed using corneal confocal microscopy: A multinational normative data set. Diabetes Care (2015.0) 38 838-843. PMID: 25633665 32. Magerl W, Krumova EK, Baron R, Tölle T, Treede RD, Maier C. Reference data for quantitative sensory testing (QST): Refined stratification for age and a novel method for statistical comparison of group data. Pain (2010.0) 151 598-605. PMID: 20965658 33. Helme RD, Finnerup NB, Jensen TS. Hyperpathia: “to be or not to be: That is the question”. Pain (2018.0) 159 1005-1009. PMID: 29771244 34. Freeman R, Baron R, Bouhassira D, Cabrera J, Emir B. Sensory profiles of patients with neuropathic pain based on the neuropathic pain symptoms and signs. Pain (2014.0) 155 367-376. PMID: 24472518 35. Vollert J, Attal N, Baron R. Quantitative sensory testing using DFNS protocol in Europe: An evaluation of heterogeneity across multiple centers in patients with peripheral neuropathic pain and healthy subjects. Pain (2016.0) 157 750-758. PMID: 26630440 36. Lee-Kubli C, Marshall AG, Malik RA, Calcutt NA. The H-reflex as a biomarker for spinal disinhibition in painful diabetic neuropathy. Curr Diab Rep (2018.0) 18 1. PMID: 29362940 37. Metz K, Concha-Matos I, Hari K 38. Zhou X, Zhu Y, Wang Z. Rate-dependent depression: A predictor of the therapeutic efficacy in treating painful diabetic peripheral neuropathy. Diabetes (2022.0) 71 1272-1281. PMID: 35234842 39. Trujillo-Hernandez B, Huerta M, Trujillo X, Vasquez C, Perez-Vargas D, Millan-Guerrero RO. F-wave and H-reflex alterations in recently diagnosed diabetic patients. J Clin Neurosci (2005.0) 12 763-766. PMID: 16054365 40. Millan-Guerrero R, Trujillo-Hernandez B, Isais-Millan S. H-reflex and clinical examination in the diagnosis of diabetic polyneuropathy. J Int Med Res (2012.0) 40 694-700. PMID: 22613432 41. Mixcoatl-Zecuatl T, Jolivalt CG. A spinal mechanism of action for duloxetine in a rat model of painful diabetic neuropathy. Br J Pharmacol (2011.0) 164 159-169. PMID: 21410686
--- title: Central obesity as assessed by conicity index and a-body shape index associates with cardiovascular risk factors and mortality in kidney failure patients authors: - Kakei Ryu - Mohamed E. Suliman - Abdul Rashid Qureshi - Zhimin Chen - Carla Maria Avesani - Torkel B. Brismar - Jonaz Ripsweden - Peter Barany - Olof Heimbürger - Peter Stenvinkel - Bengt Lindholm journal: Frontiers in Nutrition year: 2023 pmcid: PMC10016612 doi: 10.3389/fnut.2023.1035343 license: CC BY 4.0 --- # Central obesity as assessed by conicity index and a-body shape index associates with cardiovascular risk factors and mortality in kidney failure patients ## Abstract ### Background Anthropometric indices of central obesity, waist circumference (WC), conicity index (CI), and a-body shape index (ABSI), are prognostic indicators of cardiovascular (CV) risk. The association of CI and ABSI with other CV risk indices, markers of nutritional status and inflammation, and clinical outcomes in chronic kidney disease (CKD) stage 5 (CKD5) patients was investigated. ### Methods In a cross-sectional study with longitudinal follow up of 203 clinically stable patients with CKD5 (median age 56 years; $68\%$ males, $17\%$ diabetics, $22\%$ with CV disease, and $39\%$ malnourished), we investigated CI and ABSI and their associations with atherogenic index of plasma (AIP), Framingham CV risk score (FRS), Agatston scoring of coronary artery calcium (CAC) and aortic valve calcium (AVC), handgrip strength (HGS), high sensitivity C-reactive protein (hsCRP) and interleukin-6 (IL-6). CV events (CVE) and all-cause mortality during up to 10-years follow up were analyzed by multivariate survival analysis of restricted mean survival time (RMST). ### Results Chronic kidney disease patients with middle and highest CI and ABSI tertiles (indicating greater abdominal fat deposition), compared to those with the lowest CI and ABSI tertiles, tended to be older, more often men and diabetic, had significantly higher levels of hsCRP, IL-6, AIP, FRS, CAC and AVC scores. CI and ABSI were positively correlated with CAC, FRS, AIP, hsCRP and IL-6. Both CI and ABSI were negatively correlated with HGS. In age-weighted survival analysis, higher CI and ABSI were associated with higher risk of CVE (Wald test = 4.92, $$p \leq 0.027$$; Wald test = 4.95, $$p \leq 0.026$$, respectively) and all-cause mortality (Wald test = 5.24, $$p \leq 0.022$$; Wald test = 5.19, $$p \leq 0.023$$, respectively). In RMST analysis, low vs. high and middle tertiles of CI and ABSI associated with prolonged CVE-free time and death-free time, and these differences between groups increased over time. ### Conclusion Abdominal fat deposit indices, CI and ABSI, predicted CV outcomes and all-cause mortality, and were significantly associated with the inflammatory status in CKD patients. ## Introduction Obesity is a growing public health problem affecting a substantial and rapidly increasing proportion of the population across different ages and racial/ethnic groups [1]. Adipose tissue is a metabolically active endocrine organ [2] and an increase in fat mass has a significant impact on metabolic risk factors [3]. Obesity is accompanied by a wide range of complications, including kidney damage [1, 4] and the prevalence of obesity is rising among patients with chronic kidney disease (CKD) [5]. Obesity may cause kidney damage because of closely linked comorbidities, such as type 2 diabetes, hypertension, dyslipidemia, and accelerated atherosclerosis, and through direct effects on the kidneys mediated through hemodynamic and hormonal influences, insulin resistance, low-grade inflammation, adipokines, oxidative stress, protein glycation, and endothelial dysfunction [6, 7]. Fat mass and its distribution can be assessed by imaging techniques such as computed tomography and magnetic resonance as well as by ultrasound and dual energy X-ray absorptiometry (DXA). However, the use of these methods in clinical practice and in research may be limited by restricted availability of expensive instruments, high costs of maintenance, and need of expert operators (8–10). An alternative option to assess adiposity is to use anthropometric measurements, which are cost-effective, non-invasive, readily available, and affordable methods to assess obesity, but not always reliable for precise diagnose (8–10). Historically, the most common method for defining obesity is body mass index (BMI), which can be considered as a crude marker of general obesity [11]. However, BMI does not differentiate fat mass from lean body mass, cannot precisely diagnose adiposity [12] and it does not account for body fat distribution in different body regions [11, 12]. Moreover, BMI may be influenced by volume overload, which is common in CKD. Therefore, BMI does not convince as a reliable measure of body fat content in CKD patients [13]. In recent years, it has become clear that different regional adipose tissue locations have different metabolic implications and matters more than total adipose tissue mass [3] and the focus has been directed to the importance of the regional distribution of body fat, specifically central obesity, which cannot be assessed by BMI. Waist circumference (WC), a measure of central obesity, has been suggested as a better substitute for BMI [14] as abdominal obesity is more closely related to morbidity and mortality than BMI [15]. In both renal and non-renal populations, abdominal fat has been more significantly associated with increased mortality than total or peripheral fat (16–20), suggesting that the location of adipose tissue, rather than the total fat mass, is a main determinant of metabolic and inflammatory consequences of obesity in CKD [21]. Specific anthropometric indices of central obesity, which are derived from WC, namely conicity index (CI) and a-body shape index (ABSI), have been suggested as possibly better prognostic indicators than BMI. The CI is an index of central obesity using WC, height, and weight to assess fat distribution [22] while ABSI is an index based on WC that is nearly independent of height, weight, and BMI [23]. The current study was undertaken to investigate the association of CI and ABSI with cardiovascular risk indices, nutritional and inflammatory markers, and the clinical outcome in CKD stage 5 patients. ## Patient and methods In this cross-sectional observational study with analyses of longitudinal data, we investigated CI and ABSI in 203 clinically stable patients with CKD stage 5 (CKD5) including 104 non-dialyzed (CKD5-ND) patients close to the initiation of dialysis therapy and 99 dialyzed (CKD5-D) patients treated by peritoneal dialysis (PD, $$n = 67$$) or hemodialysis (HD, $$n = 32$$). The patients were recruited from three cohort studies, Kärltx [24] including 80 patients, MIA [25] including 73 patients, and MIMICK2 [26] including 50 patients, performed at Department of Renal Medicine, Karolinska University Hospital, Stockholm, Sweden. Inclusion criteria were CKD5, age > 18 years, and available data on WC and coronary artery calcification (CAC) score. Exclusion criteria were signs of overt clinical infection, unwillingness to participate and no measurements of WC or CAC. The study was conducted in adherence to the Declaration of Helsinki and authorized by the Swedish Ethical Review Authority. A written informed consent was gained from each patient. Waist circumference (WC) was measured at a level midway between the inferior margin of the last rib and the uppermost lateral iliac crest in standing position. Body mass index (BMI) was calculated as the subject’s body weight in kilograms divided by the square of the subject’s height in meters (kg/m2). Conicity index (CI) was calculated according to the equation defined by Valdez et al. [ 22], as follows: A Body Shape Index (ABSI) was calculated using the following formula [23]: Atherogenic index of plasma (AIP) was calculated as the logarithmically transformed ratio of serum triglycerides (TG) to high-density lipoprotein cholesterol (HDL-C) as in this formula [27]: Framingham cardiovascular disease (CVD) risk score (FRS), an estimate of 10-year risk of developing CVD, was calculated from age and sex stratified tables with scores for diabetes, systolic blood pressure (SBP), anti-hypertensive medication, total cholesterol, HDL-C and smoking habit [28]. Coronary artery calcium (CAC) score and aortic valve calcium (AVC) score were measured using the scoring system described by Agatston et al. [ 29] as briefly defined elsewhere [30]. Subjective global assessment (SGA) of nutritional status was evaluated using questionnaire and physical examination [31]. Based on this assessment, each patient received a nutritional status score: [1] normal nutritional status, [2] mild malnutrition, [3] moderate malnutrition or [4] severe malnutrition. In this study, malnutrition was defined as an SGA score > 1. Handgrip strength (HGS) was determined in both hands by using a Harpenden Handgrip Dynamometer (Yamar, Jackson, MI, United States). Each measurement was repeated three times for each arm, and the highest value for each arm listed. For HD patients, we measured the right arm handgrip strength, because fistulas were usually located in the left arm. Circulating biomarkers reflecting cardiovascular risk, nutritional status, and inflammation: Plasma and serum were separated and kept frozen at −70°C if not analyzed immediately. High-sensitivity C-reactive protein (hsCRP) was measured by the nephelometry method. Interleukin-6 (IL-6) was measured with ELISA commercial kits (Roche Diagnostics GmbH, Penzberg, Germany). Serum cholesterol, HDL-C and TG were analyzed by standard enzymatic procedures (Boehringer Mannheim, Mannheim, Germany). Low-density lipoprotein (LDL) cholesterol was calculated according to the Friedewald et al. [ 32] formula. The remaining biochemical analyses were done using routine methods at Karolinska University Hospital at Huddinge. Primary outcome was cardiovascular events (CVE) and all-cause mortality. Follow-up time was up to 10 years (median 5.7 years, interquartile range, IQR, 2.9–8.9 years). Clinical and outcome data was retrieved from patient records. CVE was defined as occurrence after inclusion of one or more of the following events: myocardial infarct (non-ST-elevation myocardial infarction, NSTEMI, acute myocardial infarction, AMI), onset of ischemic heart disease requiring percutaneous coronary intervention (PCI) and coronary artery bypass grafting (CABG), transient ischemic attack, stroke, peripheral vascular ischemia, or severe aortic valve stenosis requiring surgery. ## Statistical analyses Data are expressed as median (IQR, interquartile range), median ($95\%$ confidence interval) or number (percentage), as appropriate. Statistical significance was set at the level of $p \leq 0.05.$ Comparisons between two groups were assessed with the non-parametric Wilcoxon test for continuous variables and Chi-square test for nominal variables. Non-parametric Spearman rank correlation analysis was used to determine associations between variables and a multivariate linear regression analysis to determine the independent associated variables with CI and ABSI. Survival analyses and CVE were made with multivariate survival curve, using the lowest tertile of the CI or ABSI as reference in age-weighted analysis. We analyzed event-free time of CV events and all-cause mortality using multivariate analysis of restricted mean survival time (RMST) which is a novel alternative to Cox proportional hazards model that can be applied also when the proportional hazards assumption is not fulfilled [33]. The average length of event-free time until CVE or death occurred was calculated using RMST from baseline to a particular time point during 10 years of follow up. The time difference of RMST (ΔRMST) representing the difference between low tertile versus high and middle tertiles of CI or ABSI of mean event-free time was calculated by subtracting the RMST for the low tertile of CI or ABSI from the RMST for the high and middle tertile. Statistical analyses were performed using Statistical Software Stata 17.0 (Stata Corporation, College Station, TX, USA), Statistical Analysis Systems (SAS) version 9.4 level 1 7 M (SAS Campus Drive, Cary, NC, United States). ## Results Conicity index and ABSI were measured in 203 CKD patients with median age 56 years; $68\%$ of the patients were males, $17\%$ were diabetic, $22\%$ had CVD, and $39\%$ were malnourished (Tables 1, 2). Baseline clinical and biochemical characteristics for men and women, respectively, are shown in Supplementary Table 1. Males were taller, heavier, and had a longer waist circumference, and CI and ABSI were slightly higher in males while cholesterol was significantly higher in females. Patients were divided into two groups according to the distribution of lower tertile vs. the middle and higher tertiles of CI (Table 1) and ABSI (Table 2), respectively. Table 1 shows that patients with middle and higher CI (indicating greater abdominal fat deposition), compared to those with the lowest CI tertile, tended to be older, more often male, and diabetic, had lower concentrations of total cholesterol and lower levels of HDL-C, whereas concentrations of TG did not differ between the two patient groups. BMI, FBMI, and LBMI were significantly higher in patients with higher CI, whereas HGS tended to be weaker in patients with high CI (Table 1). The patients with higher CI had significantly higher levels of inflammation markers, such as hsCRP and IL-6 and significantly higher AIP, CAC and AVC scores. There were no significant differences in the systolic and diastolic blood pressure or presence of CVD between the two patient groups. Table 2 shows that the patients with higher ABSI were older and included more males and diabetic patients. The patients with higher ABSI, compared to lower ABSI tertile, had lower concentrations of total cholesterol and lower levels of HDL-C, whereas TG did not differ between the two patient groups. Moreover, BMI, FBMI, and LBMI were not significantly different between the two patient groups, whereas HGS tended to be lower in patients with higher ABSI (Table 2). The patients with higher ABSI had significantly higher concentrations of hsCRP and IL-6. Table 2 shows that the patients with higher ABSI had significantly higher AIP, CAC and AVC scores. There were no significant differences in the systolic and diastolic blood pressure or presence of CVD between the two patient groups. Table 3 displays a Spearman rank correlation matrix of associations between adiposity indices CI and ABSI and cardiovascular indices FRS and AIP. Both CI and ABSI were significantly associated with CAC and AVC scores, FRS and AIP. Moreover, CI but not ABSI, was significantly correlated with AVC. As expected, CI showed strong correlation (rho = 0.92) with ABSI. BMI was associated with CI but did not correlate with ABSI. Table 3 \| Unnamed: 0 \| BMI \| CI \| ABSI \| CAC score \| AVC score \| FRS% \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| CI \| 0.39c \| \| \| \| \| \| \| ABSI \| 0.04 \| 0.92c \| \| \| \| \| \| CAC score \| 0.20b \| 0.35c \| 0.29c \| \| \| \| \| AVC score \| 0.21b \| 0.30c \| 0.22b \| 0.45c \| \| \| \| FRS% \| 0.27c \| 0.43c \| 0.37c \| 0.69c \| 0.50c \| \| \| AIP \| 0.30c \| 0.27c \| 0.16a \| 0.20b \| 0.13 \| 0.28c \| Univariate correlations of CI and ABSI with inflammatory markers, nutritional and anthropometric indices as well as lipids are shown in Supplementary Table 2. CI and ABSI were positively and significantly correlated with hsCRP and IL-6. CI was negatively and weakly associated with SGA, whereas ABSI did not show a significant correlation with SGA. Both CI and ABSI were negatively correlated with HGS. Serum albumin was not correlated with CI and ABSI. Among lipid parameters, HDL-C was negatively associated with CI and ABSI, whereas TG was positively associated with CI but did not significantly correlate with ABSI. Total cholesterol did not show a significant correlation with CI and ABSI. Moreover, in a linear regression analysis in a model including FRS, hsCRP, CAC and SGA as shown in Table 4, we found that CI was independently associated with FRS, CAC and SGA. However, ABSI was independently associated only with FRS. Table 4 \| Unnamed: 0 \| CI (adj r2 = 0.18) \| CI (adj r2 = 0.18).1 \| ABSI (adj r2 = 0.10) \| ABSI (adj r2 = 0.10).1 \| \| --- \| --- \| --- \| --- \| --- \| \| \| β \| p \| β \| p \| \| FRS \| 0.31 \| <0.001 \| 0.230 \| 0.003 \| \| hsCRP \| 0.38 \| 0.573 \| 0.064 \| 0.355 \| \| CAC score \| 0.12 \| 0.036 \| 0.144 \| 0.057 \| \| SGA \| −0.21 \| 0.002 \| −0.135 \| 0.052 \| ## Follow-up data for cardiovascular events During the follow-up period of up to 10 years (median 5.7 years, IQR 2.4–9.8 years), the patients experienced 59 CVE. Compared to those free of CVE, the patients who faced CVE had a higher proportion of diabetics (33 vs. $10\%$; $p \leq 0.001$) and had higher hsCRP concentration (3.8 vs. 1.3 mg/l; $p \leq 0.001$), AIP (0.41 vs. 0.11; $p \leq 0.001$), CI (1.37 vs. 1.32; $p \leq 0.001$), ABSI (0.86 vs. 0.84; $$p \leq 0.005$$) and BMI (25.2 vs. 24.2 kg/m2; $p \leq 0.05$). However, there was no significant difference in nutritional status, as reflected by SGA, between the two groups. In multivariate age-weighted analysis, adjusted for gender, DM, CVD, and total cholesterol, the cumulative incidence curve showed that patients with middle and higher tertiles of CI and ABSI, respectively, had higher risk of CVE compared to those with lower CI (Wald test: 4.92, $$p \leq 0.027$$) and ABSI (Wald test: 4.95, $$p \leq 0.026$$) tertiles, respectively. For estimating the CV event-free time for low vs. middle and high tertiles of CI and ABSI, respectively, for the whole follow up period, or the mean event-free time during a prespecified period, we calculated multivariate ΔRMST representing the mean absolute difference of event-free time between the CI or ABSI tertile groups during the follow-up period. At 10 years of follow-up, the ΔRMST (Table 5), showing the difference between low vs. middle and high tertiles was 0.92 [$95\%$ CI, −0.03–1.87] years for CI and 1.12 [$95\%$ CI, 0.15–2.09] years for ABSI, indicating that the patients with the low tertiles had longer time free of CVE than the patients with the middle and high CI or ABSI tertiles, respectively. Moreover, in further analysis, Figures 1A,B show point estimates of RMST at 3, 6, and 9 years. As shown in Figures 1A,B, ΔRMST, i.e., the incremental benefit in event-free time of low tertile over high and middle tertiles of CI, was 0.14 years at 3 years, 0.54 years at 6 years, and 0.68 years at 9 years. All differences at 6 years but not at 3, 9, and 10 years were statistically significant. The corresponding incremental benefit (i.e., ΔRMST) of low tertile over high and middle tertiles of ABSI was 0.08, 0.48 and 0.81 years at 3, 6 and 9 years, respectively. Differences at 6 and 10 years but not at 3 years and 9 years were statistically significant. The same analyses were also performed separately for females and males (Supplementary Tables 3, 4; Supplementary Figures 1, 2). In men, low tertiles associated with a significant benefit in RMST compared to high and middle tertiles, but, for women, there was no clear RMST benefit, perhaps because the numbers are insufficient for analysis. ## Follow-up data for survival analysis During a follow-up period of up to 10 years, there were 69 deaths ($34\%$). Compared to alive patients during the follow-up period, the group of patients who died had a higher prevalence of diabetics (29 vs. $11\%$; $p \leq 0.001$), CVD (37 vs. $14\%$; $p \leq 0.001$), higher levels of CRP (4.2 vs. 1.1 mg/l; $p \leq 0.001$), and higher AIP (0.35 vs. 0.12; $p \leq 0.001$), CI (1.37 vs. 1.32; $p \leq 0.001$) and ABSI (0.86 vs. 0.84; $p \leq 0.01$). There were no significant differences according to SGA and BMI between the patients who died and those who lived. Age-weighted survival analysis adjusted by gender, DM, CVD, and total cholesterol showed that the low CI tertile and low ABSI tertiles were associated with better survival than the middle and higher CI tertiles (Wald test = 5.24, $$p \leq 0.022$$) and middle and higher ABSI tertiles (Wald test = 5.19, $$p \leq 0.023$$), respectively. For estimating all-cause mortality associated with low vs. middle and high tertiles of CI or ABSI, we applied RMST analysis and calculated ΔRMST. Figures 2A,B show point estimates of RMST at 3, 6 and 9 years for all-cause mortality. The ΔRMST (Table 6) shows that the difference in survival time for low tertile vs. high and middle tertiles of CI was 0.05, 0.46, 0.90 and 1.19 years at 3, 6, 9 and 10 years, respectively. The corresponding figures for low tertile vs. high and middle tertiles of ABSI was 0.05, 0.49, 0.86 and 1.09 years at 3, 6, 9 and 10 years, respectively. The incremental benefit of low CI and low ABSI at 6, 9 and 10 years was statistically significant while this was not the case at 3 years. Figure 2: Cumulative incidence of all-cause mortality for 10 years follow-up of CKD5 patients with low versus high and middle tertiles of CI (A) and ABSI (B) respectively. The mean difference between the two tertile groups of restricted mean survival time truncated at 3, 6, and 9 years, ΔRMST with $95\%$ confidence interval ($95\%$CI), is noted. TABLE_PLACEHOLDER:Table 6 ## Discussion There is a consensus that adipose tissue location, and especially central obesity, rather than total adipose tissue mass, is of importance for the metabolic and inflammatory consequences of obesity in CKD, and WC has been shown to be a better predictor of outcomes than BMI in CKD patients [34]. The CI and ABSI are two proposed indices to assess central obesity with the shared characteristic laying in the fact that they both are based on WC, with adjustment for height and weight. Accordingly, an elevated CI or ABSI indicate that WC is higher than expected for a given height and weight suggesting accumulation of adipose tissue around the abdominal region. In this study, we applied these anthropometric indices as estimates of central obesity in CKD stage 5 patients. The current study demonstrated four major findings. First, higher CI and ABSI were associated with scores indicating increased cardiovascular risk, such as CAC and AVC scores, FRS and AIP. In addition, patients who had a history of CVD had higher values of CI and ABSI. Second, CI and ABSI were significantly associated with inflammatory status and negatively with HGS%. Third, CI and ABSI were associated with the risk of CVE during the follow-up period. Lastly, CI and ABSI were associated with all-cause mortality. These results show that the abdominal fat deposit indices, CI and ABSI, in CKD patients are significantly associated with several cardiovascular risk indices including inflammatory status and suggest that they could be of value for predicting cardiovascular outcomes and mortality among these patients. It is generally acknowledged that CKD patients suffer from accelerated atherosclerosis and that CVD represents the leading cause of death in these patients [35]. Adiposity is a known precursor of atherosclerosis and the increasing prevalence of obesity has implications for the risk of diabetes, CVD and also for development of CKD [6]. Moreover, the prevalence of obesity has increased substantially among patients with CKD [36]. Adiposity can be widely divided into general adiposity, which is usually assessed by BMI, and into central adiposity, that is reflected by WC. Central obesity is believed to be more pathogenic and more important as a predictor of cardiovascular metabolic disease compared to general obesity [37, 38] and may have a greater association with metabolic health risks. Indeed, the central obesity indices as CI and ABSI, which are calculated from WC, are associated with higher risk of CVD in the general population [23, 39, 40]. Like previous reports [21, 41], in the current study, CKD patients with higher CI or ABSI were older, more often male, had diabetes and had signs of dyslipidemia, other cardiovascular risk indices and inflammatory state. Moreover, our study showed that the patients with high CI tertiles had higher BMI, fat body mass and lean body mass, whereas the tertiles of ABSI did not show differences in BMI, fat body mass and lean body mass. Although nutritional status, as assessed by SGA, did not differ between the tertiles of CI and ABSI likely since both indices are related with adiposity and can mask malnutrition when assessed by SGA, handgrip strength (HGS) showed strong reverse associations with both CI and ABSI. In a previous study, Cordeiro et al. [ 21] reported that HD patients with an increased CI tended to be more malnourished and had weaker HGS. HGS is a useful tool for the systematic assessment of muscle strength related to nutritional status. Reduced HGS is a common finding among CKD patients and strongly associated with morbidity and mortality [42]. Similar to our findings of reverse associations of HGS with ABSI and CI, Krakauer and Krakauer [43] reported that HGS was inversely associated with ABSI in the National Health and Nutrition Examination Survey (NHANES) 2011–2014 data including 9,803 adults in the United States population. This association may also sign for low muscle function, a finding that can occur concomitantly with higher adiposity [44]. Our results demonstrated that central obesity, as reflected by CI and ABSI, was positively associated with two measures of cardiac calcification, CAC as well as AVC. The association of CI with CAC and AVC was stronger than that of ABSI with CAC and AVC. This was supported by the findings of multivariate regression analysis that CI was independently associated with FRS and CAC and SGA in a model including FRS, hsCRP, CAC and SGA, whereas in an alternative analysis including the same variables ABSI was independently associated with FRS, but not with CAC and other variables (Table 4). The association of central obesity with risk of coronary atherosclerosis was supported by our findings that AIP correlated with CI and ABSI. However, these findings are not in accordance with a previous study showing that CI was not correlated with AIP in relatively lean maintenance hemodialysis patients [45]. The AIP is a marker of atherogenicity, and it is considered as an independent predictor of rapid progression of coronary atherosclerosis [46]. Therefore, the association of AIP with the central obesity indices in the current study may underline the relationship of central obesity with cardiovascular risk. Indeed, one of the interesting findings in this study is that FRS was strongly correlated with CI and ABSI. The relationship between obesity and cardiovascular diseases is well known and is predominantly related to the visceral accumulation of fat depots. Obesity, in particular visceral obesity, is a well-known risk factor of CVD and CKD patients are subjected to accelerated atherosclerosis and frequently suffer from vascular calcification. Excessive accumulation of visceral fat is associated with inflammation and linked to atherosclerotic events [47]. In CKD patients, abdominal fat has been reported to be associated with inflammation [21, 48]. In agreement with another study in HD patients [49], we found that high CI and ABSI were correlated with increased concentrations of CRP and IL-6, suggesting that abdominal fat deposition could be a significant contributor to increased CRP production in HD patients. Altogether these findings are adding further support to the concept of abdominal obesity being a promotor for inflammation and a risk factor for CVD in CKD patients. This concept is supported by our findings that high CI and ABSI in our patients were associated with a higher risk for CVE and higher all-cause mortality risk. This is consistent with a previous study that showed a relationship between CI and total mortality in prevalent HD patients [21]. Moreover, in the current study, using RMST, the patients with low CI or ABSI had incremental benefits of increased CV event-free time and prolonged survival over the patients with high CI or high ABSI. Notably, these benefits, which were small before 3 years, increased steadily and showed a substantial improvement during the 10 years follow-up period. The strengths of this study include detailed phenotyping of patients using anthropometric, imaging and laboratory measurements with few missing values and no patient being lost to follow up. The study also has some limitations that should be considered when interpreting the results. Firstly, as in any observational study, causality cannot be inferred. Secondly, we performed age-weighted analysis and considered several potential confounding factors such as gender, DM, CVD, and total cholesterol use but acknowledge the existence of residual and unknown confounding, and that the relatively small number of CKD patients may not have provided enough statistical power. Secondly, inclusion of both dialyzed and non-dialyzed patients may limit the interpretation. Thirdly, we rely on anthropometric indices for estimation of abdominal fat based only on WC. Nevertheless, WC has been validated against assessment of fat mass by computed tomography in CKD patients [50], and associations of anthropometric indices with visceral fat and metabolic risk indicators are in general as strong as those obtained by magnetic resonance imaging for measuring adipose tissue stores [51]. Finally, because body weight and anthropometrics can be influenced by the hydration state, fluid status may have influenced the anthropometric indices. In conclusion, the present study shows that abdominal body fat indices, in particular CI and ABSI, associate with cardiovascular risk indices, poor CV outcome and inflammatory status in CKD5 patients. It indicates that central obesity is a factor of importance when predicting CV outcomes and suggests that it may represent a mortality risk factor. Further studies on a larger scale are needed to confirm these findings. ## Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. ## Ethics statement The studies involving human participants were reviewed and approved by The Ethics Committee of the Karolinska Institute (EPN) at the Karolinska University Hospital Huddinge, Stockholm, Sweden. The patients/participants provided their written informed consent to participate in this study. ## Author contributions KR, MS, ZC, AQ, BL, and PS: conceptualization. KR, AQ, BL, OH, PB, and PS: data curation. OH, PB, TB, JR, CA, PS, AQ, and BL: investigation. BL, PS, AQ, and MS: project administration. BL and PS: supervision. MS, KR, BL, and AQ: writing—original draft. All authors contributed to the article and approved the submitted version. ## Funding Baxter Novum is the result of a grant from Baxter Healthcare Corporation to Karolinska Institutet. ## Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ## Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. ## Supplementary material The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnut.2023.1035343/full#supplementary-material ## References 1. Hruby A, Hu FB. The epidemiology of obesity: A big picture. PharmacoEconomics (2015) 33 673-89. DOI: 10.1007/s40273-014-0243-x 2. Coelho M, Oliveira T, Fernandes R. Biochemistry of adipose tissue: an endocrine organ. Arch Med Sci (2013) 9 191-200. DOI: 10.5114/aoms.2013.33181 3. Gómez-Hernández A, Beneit N, Díaz-Castroverde S, Escribano Ó. Differential role of adipose tissues in obesity and related metabolic and vascular complications. Int J Endocrinol (2016) 2016 1216783-15. DOI: 10.1155/2016/1216783 4. Stenvinkel P, Zoccali C, Ikizler TA. Obesity in CKD—what should nephrologists know?. J Am Soc Nephrol (2013) 24 1727-36. DOI: 10.1681/ASN.2013040330 5. Kramer HJ, Saranathan A, Luke A, Durazo-Arvizu RA, Guichan C, Hou S. Increasing body mass index and obesity in the incident ESRD population. J Am Soc Nephrol (2006) 17 1453-9. DOI: 10.1681/ASN.2005111241 6. Kovesdy CP, Furth SL, Zoccali C. Electronic address mwo, world kidney day steering C. obesity and kidney disease: hidden consequences of the epidemic. J Ren Nutr (2017) 27 75-7. DOI: 10.1053/j.jrn.2017.01.001 7. Wickman C, Kramer H. Obesity and kidney disease: potential mechanisms. Semin Nephrol (2013) 33 14-22. DOI: 10.1016/j.semnephrol.2012.12.006 8. Prado CM, Cushen SJ, Orsso CE, Ryan AM. Sarcopenia and cachexia in the era of obesity: clinical and nutritional impact. Proc Nutr Soc (2016) 75 188-98. DOI: 10.1017/S0029665115004279 9. Ribeiro Filho FF, Mariosa LS, Ferreira SR, Zanella MT. Visceral fat and metabolic syndrome: more than a simple association. Arq Bras Endocrinol Metabol (2006) 50 230-8. DOI: 10.1590/S0004-27302006000200009 10. Vogt BP, Caramori JCT. Recognition of visceral obesity beyond body fat: assessment of cardiovascular risk in chronic kidney disease using anthropometry. Nutrire (2017) 42 19. DOI: 10.1186/s41110-017-0041-2 11. Nuttall FQ. Body mass index: obesity, BMI, and health: A critical review. Nutr Today (2015) 50 117-28. DOI: 10.1097/NT.0000000000000092 12. Rodrigues J, Santin F, Barbosa Brito FS, Carrero JJ, Lindholm B, Cuppari L. Sensitivity and specificity of body mass index as a marker of obesity in elderly patients on hemodialysis. J Ren Nutr (2016) 26 65-71. DOI: 10.1053/j.jrn.2015.09.001 13. Agarwal R, Bills JE, Light RP. Diagnosing obesity by body mass index in chronic kidney disease: an explanation for the "obesity paradox?". Hypertension (2010) 56 893-900. DOI: 10.1161/HYPERTENSIONAHA.110.160747 14. Ross R, Neeland IJ, Yamashita S, Shai I, Seidell J, Magni P. Waist circumference as a vital sign in clinical practice: a consensus statement from the IAS and ICCR working Group on visceral obesity. Nat Rev Endocrinol (2020) 16 177-89. DOI: 10.1038/s41574-019-0310-7 15. Jayedi A, Soltani S, Zargar MS, Khan TA, Shab-Bidar S. Central fatness and risk of all cause mortality: systematic review and dose-response meta-analysis of 72 prospective cohort studies. BMJ (2020) 370 m 3324. DOI: 10.1136/bmj.m3324 16. Evaluation, and treatment of overweight and obesity in adults: executive summary. Expert panel on the identification, evaluation, and treatment of overweight in adults. Am J Clin Nutr (1998) 68 899-917. PMID: 9771869 17. Pischon T, Boeing H, Hoffmann K, Bergmann M, Schulze MB, Overvad K. General and abdominal adiposity and risk of death in Europe. N Engl J Med (2008) 359 2105-20. DOI: 10.1056/NEJMoa0801891 18. Elsayed EF, Sarnak MJ, Tighiouart H, Griffith JL, Kurth T, Salem DN. Waist-to-hip ratio, body mass index, and subsequent kidney disease and death. Am J Kidney Dis (2008) 52 29-38. DOI: 10.1053/j.ajkd.2008.02.363 19. Elsayed EF, Tighiouart H, Weiner DE, Griffith J, Salem D, Levey AS. Waist-to-hip ratio and body mass index as risk factors for cardiovascular events in CKD. Am J Kidney Dis (2008) 52 49-57. DOI: 10.1053/j.ajkd.2008.04.002 20. Postorino M, Marino C, Tripepi G, Zoccali C, Group CW. Abdominal obesity and all-cause and cardiovascular mortality in end-stage renal disease. J Am Coll Cardiol (2009) 53 1265-72. DOI: 10.1016/j.jacc.2008.12.040 21. Cordeiro AC, Qureshi AR, Stenvinkel P, Heimburger O, Axelsson J, Barany P. Abdominal fat deposition is associated with increased inflammation, protein-energy wasting and worse outcome in patients undergoing haemodialysis. Nephrol Dial Transplant (2010) 25 562-8. DOI: 10.1093/ndt/gfp492 22. Valdez R, Seidell JC, Ahn YI, Weiss KM. A new index of abdominal adiposity as an indicator of risk for cardiovascular disease. A cross-population study. Int J Obes Relat Metab Disord (1993) 17 77-82. PMID: 8384168 23. Krakauer NY, Krakauer JC. A new body shape index predicts mortality hazard independently of body mass index. PLoS One (2012) 7 e39504. DOI: 10.1371/journal.pone.0039504 24. Qureshi AR, Olauson H, Witasp A, Haarhaus M, Brandenburg V, Wernerson A. Increased circulating sclerostin levels in end-stage renal disease predict biopsy-verified vascular medial calcification and coronary artery calcification. Kidney Int (2015) 88 1356-64. DOI: 10.1038/ki.2015.194 25. Stenvinkel P, Heimbürger O, Paultre F, Diczfalusy U, Wang T, Berglund L. Strong association between malnutrition, inflammation, and atherosclerosis in chronic renal failure. Kidney Int (1999) 55 1899-911. DOI: 10.1046/j.1523-1755.1999.00422.x 26. Xu H, Cabezas-Rodriguez I, Qureshi AR, Heimburger O, Barany P, Snaedal S. Increased levels of modified advanced oxidation protein products are associated with central and peripheral blood pressure in peritoneal dialysis patients. Perit Dial Int (2015) 35 460-70. DOI: 10.3747/pdi.2013.00064 27. Dobiasova M, Frohlich J. The new atherogenic plasma index reflects the triglyceride and HDL-cholesterol ratio, the lipoprotein particle size and the cholesterol esterification rate: changes during lipanor therapy. Vnitr Lek (2000) 46 152-6. PMID: 11048517 28. D'Agostino RB, Vasan RS, Pencina MJ, Wolf PA, Cobain M, Massaro JM. General cardiovascular risk profile for use in primary care: the Framingham heart study. Circulation (2008) 117 743-53. DOI: 10.1161/CIRCULATIONAHA.107.699579 29. Agatston AS, Janowitz WR, Hildner FJ, Zusmer NR, Viamonte M, Detrano R. Quantification of coronary artery calcium using ultrafast computed tomography. J Am Coll Cardiol (1990) 15 827-32. DOI: 10.1016/0735-1097(90)90282-T 30. Mukai H, Dai L, Chen Z, Lindholm B, Ripsweden J, Brismar TB. Inverse J-shaped relation between coronary arterial calcium density and mortality in advanced chronic kidney disease. Nephrol Dial Transplant (2020) 35 1202-11. DOI: 10.1093/ndt/gfy352 31. Qureshi AR, Alvestrand A, Danielsson A, Divino-Filho JC, Gutierrez A, Lindholm B. Factors predicting malnutrition in hemodialysis patients: a cross-sectional study. Kidney Int (1998) 53 773-82. DOI: 10.1046/j.1523-1755.1998.00812.x 32. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem (1972) 18 499-502. DOI: 10.1093/clinchem/18.6.499 33. Han K, Jung I. Restricted mean survival time for survival analysis: A quick guide for clinical researchers. Korean J Radiol (2022) 23 495-9. DOI: 10.3348/kjr.2022.0061 34. Fitzpatrick J, Sozio SM, Jaar BG, McAdams-DeMarco MA, Estrella MM, Tereshchenko LG. Association of Abdominal Adiposity with cardiovascular mortality in incident hemodialysis. Am J Nephrol (2018) 48 406-14. DOI: 10.1159/000494281 35. Tonelli M, Wiebe N, Culleton B, House A, Rabbat C, Fok M. Chronic kidney disease and mortality risk: a systematic review. J Am Soc Nephrol (2006) 17 2034-47. DOI: 10.1681/ASN.2005101085 36. Evangelista LS, Cho WK, Kim Y. Obesity and chronic kidney disease: A population-based study among south Koreans. PLoS One (2018) 13 e0193559. DOI: 10.1371/journal.pone.0193559 37. Jia Z, Zhou Y, Liu X, Wang Y, Zhao X, Wang Y. Comparison of different anthropometric measures as predictors of diabetes incidence in a Chinese population. Diabetes Res Clin Pract (2011) 92 265-71. DOI: 10.1016/j.diabres.2011.01.021 38. Li G, Chen X, Jang Y, Wang J, Xing X, Yang W. Obesity, coronary heart disease risk factors and diabetes in Chinese: an approach to the criteria of obesity in the Chinese population. Obes Rev (2002) 3 167-72. DOI: 10.1046/j.1467-789X.2002.00067.x 39. Malara M, Keska A, Tkaczyk J, Lutoslawska G. Body shape index versus body mass index as correlates of health risk in young healthy sedentary men. J Transl Med (2015) 13 75. DOI: 10.1186/s12967-015-0426-z 40. Motamed N, Perumal D, Zamani F, Ashrafi H, Haghjoo M, Saeedian FS. Conicity index and waist-to-hip ratio are superior obesity indices in predicting 10-year cardiovascular risk among men and women. Clin Cardiol (2015) 38 527-34. DOI: 10.1002/clc.22437 41. El Said HW, Mohamed OM, El Said TW, El Serwi AB. Central obesity and risks of cardiovascular events and mortality in prevalent hemodialysis patients. Int Urol Nephrol (2017) 49 1251-60. DOI: 10.1007/s11255-017-1568-0 42. Oliveira MC, Bufarah MNB, Balbi AL. Handgrip strength in end stage of renal disease—a narrative review. Forum Nutr (2018) 43 1-8. DOI: 10.1186/s41110-018-0073-2 43. Krakauer NY, Krakauer JC. Association of Body Shape Index (ABSI) with hand grip strength. Int J Environ Res Public Health (2020) 17 6797. DOI: 10.3390/ijerph17186797 44. Bellafronte NT, de Queiros Mattoso Ono A, Chiarello PG. Sarcopenic obesity in chronic kidney disease: challenges in diagnosis using different diagnostic criteria. Med Princ Pract (2021) 30 477-86. DOI: 10.1159/000517597 45. Zhou C, Peng H, Yuan J, Lin X, Zha Y. Visceral, general, abdominal adiposity and atherogenic index of plasma in relatively lean hemodialysis patients. BMC Nephrol (2018) 19 206. DOI: 10.1186/s12882-018-0996-0 46. Won KB, Heo R, Park HB, Lee BK, Lin FY, Hadamitzky M. Atherogenic index of plasma and the risk of rapid progression of coronary atherosclerosis beyond traditional risk factors. Atherosclerosis (2021) 324 46-51. DOI: 10.1016/j.atherosclerosis.2021.03.009 47. Alexopoulos N, Katritsis D, Raggi P. Visceral adipose tissue as a source of inflammation and promoter of atherosclerosis. Atherosclerosis (2014) 233 104-12. DOI: 10.1016/j.atherosclerosis.2013.12.023 48. Cordeiro AC, Qureshi AR, Lindholm B, Amparo FC, Tito-Paladino-Filho A, Perini M. Visceral fat and coronary artery calcification in patients with chronic kidney disease. Nephrol Dial Transplant (2013) 28 iv152. DOI: 10.1093/ndt/gft250 49. Ruperto M, Barril G, Sanchez-Muniz FJ. Conicity index as a contributor marker of inflammation in haemodialysis patients. Nutr Hosp (2013) 28 1688-95. DOI: 10.3305/nh.2013.28.5.6626 50. Sanches FM, Avesani CM, Kamimura MA, Lemos MM, Axelsson J, Vasselai P. Waist circumference and visceral fat in CKD: a cross-sectional study. Am J Kidney Dis (2008) 52 66-73. DOI: 10.1053/j.ajkd.2008.02.004 51. Scherzer R, Shen W, Bacchetti P, Kotler D, Lewis CE, Shlipak MG. Simple anthropometric measures correlate with metabolic risk indicators as strongly as magnetic resonance imaging-measured adipose tissue depots in both HIV-infected and control subjects. Am J Clin Nutr (2008) 87 1809-17. DOI: 10.1093/ajcn/87.6.1809
--- title: Comparison of growth in neutered Domestic Shorthair kittens with growth in sexually-intact cats authors: - Carina Salt - Richard F. Butterwick - Kristin S. Henzel - Alexander J. German journal: PLOS ONE year: 2023 pmcid: PMC10016642 doi: 10.1371/journal.pone.0283016 license: CC BY 4.0 --- # Comparison of growth in neutered Domestic Shorthair kittens with growth in sexually-intact cats ## Abstract The first aim of these studies was to compare growth patterns of healthy kittens neutered during growth with growth standards created for sexually-intact kittens. A second aim was to clarify the impact of neutering in kittens on body composition and body shape. Study 1 was a retrospective observational study comparing median growth trajectories of healthy, client-owned domestic shorthair (DSH) kittens in optimal body condition and neutered at different ages, with previously-created growth standards from a similar, sexually-intact, population. The neuter groups contained between 3.0k and 9.3k cats. For all neuter groups in both sexes, the median growth trajectory inclined upwards after the procedure, with this being more marked in female than in male kittens. This upwards inclination was less marked for kittens neutered later during growth in both sexes, with the effect being least in kittens neutered after 28–29 weeks. Study 2 was an analysis of new body composition and zoometric data from a previously-published randomised study, comparing growth-related measures between 11 pairs of sexually-intact and neutered (at 19 weeks age) female DSH cats in a research population. Before neutering, the growth pattern in neutered kittens and sexually-intact kittens was similar, but neutered kittens were heavier by 52 weeks (mean difference in fold change vs. 10 weeks 1.34, 95-CI: 1.07–1.72), had a greater fat mass (mean difference in fold change vs. 10 weeks 1.91, 95-CI 1.09–3.21) and greater lean mass (mean difference in fold change vs. 10 weeks 1.23, 95-CI: 1.03–1.48). Abdominal girth (mean difference in fold change vs. 10 weeks 1.20, 95-CI: 1.04–1.39) and rib cage length (mean difference in fold change vs. 10 weeks 1.18, 95-CI: 1.02–1.36) were also greater, but there were no differences in other zoometric measurements. Veterinarians should consider the potential impact that neutering has on gain of adipose tissue, especially early neutering in female kittens. Bodyweight should be monitored closely during growth and especially after neutering to prevent inappropriate weight gain. ## Introduction The owners of most pet cats in the developed world choose to have them neutered by surgical gonadectomy (castration or ovariohysterectomy for male and female cats, respectively) [1], which has perceived benefits including controlling the stray and pet cat populations, reducing undesirable behaviour (especially in male cats), and decreasing the risk of neoplasia of both the mammary gland and urogenital tract [2,3]. However, neutering can have adverse effects on health in cats including increased risk of feline lower urinary tract disease and increased risk of diabetes mellitus [4,5], as well as being a known risk factor for the development of obesity [6–9] characterised by an expansion of white adipose tissue mass that can lead to adverse health effects [10]. This obesity risk is thought to be associated with increased food intake, which peaks at 8–10 weeks after neutering, but declines to the intake seen in sexually-intact kittens by 18 weeks after neutering [8,9]. The effect is that neutered kittens fed ad libitum grow significantly faster and end their growth phase heavier than sexually-intact cats, both in terms of bodyweight and fat mass [8,9]. As a result, it is recommended that food intake be controlled after neutering to prevent the development of obesity [7]. Given the known effects of neutering on food intake and weight gain in cats, their impact on patterns of growth is pertinent, not least because most cats are neutered before 1 year of age [11,12]. Optimal growth is key priority for all species, especially those that act as companion animals including cats. Providing complete and balanced nutrition is important, but it is also important to ensure that the correct amount be fed so that the pattern of growth is optimal. Malnutrition at this critical stage can cause growth retardation [13,14], but this is uncommon. In reality, overnutrition is a greater concern, leading to overly rapid growth and increasing the risk of developing obesity [14–16]. Recently, evidence-based growth standards have been developed for male and female domestic shorthair cats [16] and show promise for use as a tool for monitoring the growth, not least given that body condition scoring [17] has not been validated for assessing body composition in kittens. The recently developed growth charts utilised data from cats that remained sexually-intact during growth, but the impact of neutering on such patterns was not studied [16]. In dogs, neutering had only a minor impact on growth trajectory, within one centile of the median line [18] and, as a consequence, standards for sexually-intact animals were appropriate for neutered animals. That said, the impact of neutering differed depending upon its timing [14]: a slight upward shift in growth trajectory was observed when neutering occurred before 37 weeks, whilst a slight downward shift in growth trajectory was observed in dogs neutered after 37 weeks. Therefore, in determining the effect of neutering on the pattern of growth in cats, the impact of timing is also likely to be important, and this is of interest because of recent shifts towards early-age neutering, with a study from Australia demonstrating $60\%$ of cats being neutered before 6 months’ age and $22\%$ before 4 months’ age [19]. Early-age neutering performed before puberty (5–9 month’s age) can effectively prevent unwanted litters in cats, thereby helping to control both stray and pet cat populations [20]. However, the physiological impact of neutering at such a stage of immaturity has not been explored in detail. In ome previous study, pre-pubertal neutering was not identified as a risk factor for overweight or obesity, compared with obesity at a traditional age (7 months) [20–22]. However, in a second study, early-age neutering resulted in a marked increase in food intake, increased bodyweight and also body fat mass, with the effects being most notable in female, compared with male, kittens [9]. The aim of the study was to inform recommendations around using the growth standards in neutered kittens. Given previous findings in dogs [16], we hypothesised that the growth pattern of neutered cats would not differ from that of sexually-intact kittens. Therefore, we investigated the growth patterns of healthy kittens of normal body condition score, neutered during growth, with those of the growth standards. Given that any increase in bodyweight could be due to a change in body composition or body size, we also explored the impact of neutering on body shape and body composition, by analysing new data gathered from a previous study assessing the effects of neutering on food intake, bodyweight and body composition in growing female kittens [9]. ## Study design This paper reports two analytic studies: study 1 was a retrospective observational study to compare median weight trajectories of healthy, neutered, client-owned domestic shorthair (DSH) kittens in optimal body condition with available growth standards for sexually intact kittens [16]; study 2 was an analysis of new data, from a previously-published randomised trial [9], comparing growth-related measures at four time points between pairs of sexually-intact and neutered female siblings from litters of DSH cats in a research population. The objective of study 1 was to determine whether neutering changes patterns of healthy weight gain during growth, whilst the objective of study 2 was to gain further insight into the causes of any neutering-associated changes in weight gain. ## Study populations Data for Study 1 were derived from the clinical records of client-owned DSH kittens attending Banfield® Pet Hospital (BANFIELD), a network of over 900 primary care veterinary hospitals located mainly in the USA. Data were collected between April 1994 and November 2016, with $77\%$ of data points relating to 2004 onwards. Bodyweight is routinely measured during consultations, whilst birth date and breed data are supplied by owners when the pet is first registered, but not independently verified. Data for Study 2 were derived from a previous trial involving pairs of 11-week-old female DSH littermates housed at the Waltham Petcare Science Institute (WALTHAM) in the UK. The original study was reviewed and approved by the Waltham Ethical Review Committee and complied with UK Home Office regulations. Full details of the study population and experimental design have been reported previously [9]; briefly, 11 littermate pairs were randomly assigned to either a neutered (neutered at 19 weeks old) or a sexually-intact (kept sexually intact) groups, and were offered free access to a dry diet until the age of 1 year. The kittens were group-housed for socialisation during the day and individually housed overnight. Kittens were fed a nutritionally-complete, commercial dry diet formulated to meet the nutritional requirements for gestation, lactation and growth and confirmed by full nutritional analysis (moisture 7·5 g, protein 33·1 g, fat 20·8 g, ash 7·35 g, non-free extract 31·4 g, predicted metabolisable energy (ME) 1628·4 kJ/100 g as fed). ## Data extraction and eligibility criteria For Study 1, the BANFIELD medical records database was searched for weight measurements from DSH cats under 130 weeks (2.5 years) of age, calculated from measurement date and date of birth. For any individual recorded as having been neutered, the date for the procedure was required. Given that neuter dates were only available for cats where castration or ovariohysterectomy was performed at a BANFIELD hospital, those where procedures were performed elsewhere were excluded from the dataset. Additional eligibility criteria were applied at this stage, consistent with those used when creating the growth standards, as described elsewhere [16]. Data were only included from cats attending routine preventative healthcare visits (these were determined by Banfield’s visit type classification and included reasons such as vaccination, deworming and check-ups), or visits where the diagnosis was recorded as ‘healthy pet’. Measurements were excluded from the dataset if the recorded bodyweight had been rolled over from a previous visit (which occasionally happens when weigh scales were unavailable at the time the kitten is checked in for their appointment), or if the medical records indicated that the cat had been weighed whilst still in a pet carrier. Also excluded, were all observations from cats where the recorded sex was unclear (e.g., a cat recorded as male, but who had undergone an ovariohysterectomy), and data from cats diagnosed with a health condition before 4y of age associated that might either have affected their growth or might have caused weight loss or gain after skeletal maturity. Examples included diabetes mellitus and dwarfism. Finally, only data points were included from cats that had received a body condition rating of ‘normal’ or ‘ideal’ at one or more visits between the ages of 78 weeks and 130 weeks (1.5 to 2.5 years), which was taken as an indicator of having optimal body condition in young adulthood, and had never received an abnormal body condition rating (e.g., very thin, thin, heavy, overweight, markedly obese) at any point up to the age of 208 weeks (4 years). The restriction to healthy kittens of good body condition score ensured that the study was examining patterns of healthy growth, on the same basis as the kitten growth curves were constructed. The dataset was then restricted to kittens with at least one recorded bodyweight remaining between the ages of 5 weeks and 91 weeks. All available data on the 11 cat pairs in Study 2 was used, meaning that no special extraction or eligibility criteria were needed. ## Generation and recording of clinical data The methods for recording signalment data (date of birth, breed and sex), bodyweight, neuter status, neuter date and clinical diagnoses in the BANFIELD clinical records used in Study 1 have been described previously [15]; the methods for assessing and recording body condition score (BCS) at BANFIELD, and how this was treated when preparing the dataset, have also been described for dogs [18]; exactly the same approach was taken for handling BCS in cats [16]. Briefly, BCS was mapped onto a 3-category scale (‘thin’, ‘normal’, or ‘heavy’) for the purposes of the data extraction and analysis. If BCS was unavailable, but had been recorded at a previous or subsequent consultation where the bodyweight was within ±$5\%$ of the weight recorded at that visit, then that BCS replaced the unknown one [18]. For the randomised trial data used in Study 2, body composition, morphological measurements and fasting blood samples were taken at 11, 18, 30 and 52 weeks of age, with neutering taking place at 19 weeks of age. Bodyweight was measured weekly, but only the measurements from 11, 18, 30 and 52 weeks were utilised for the current study. As described previously [9], lean mass and fat mass were determined using dual-energy X-ray absorptiometry (DEXA), whereby kittens were sedated, placed in lateral recumbency and body composition was analysed using a Hologic QDR-1000 W pencil beam dual-energy X-ray absorptiometer (Hologic, Inc., Waltham, MA, USA). Zoometric parameters were measured in fasted, conscious and non-sedated cats. Measurements were taken from their left-hand side, whilst they were standing with legs perpendicular to the ground and with their head in an upright position and looking forward. Height was measured at the points of the scapula; length from manubrium of the sternum to a parallel point below the anus; girth around the narrowest point of the waist; ribcage at the deepest part of the thorax (approximately the point of the sternum and eighth rib); chest depth at mid-thorax, from spine to the deepest part; elbow width across the condyle of the humerus; forelimb length from olecranon to carpus; and hindlimb length from patella to tarsus. Length was measured with a purpose-adapted measuring stick; girth and ribcage were measured with a graduated measuring tape; whilst chest depth, forelimb and hindlimb were measured with digital callipers. Fat mass, lean mass and bodyweight were recorded in kg, whilst all zoometric measurements were measured in cm. ## Sample size For Study 1, the aim was to include as many cats as possible that met the eligibility criteria, so a formal sample size calculation was not performed. However, the resulting dataset was of comparable size to that used to create the growth standards [16]. The data used in Study 2 were from a historic trial, and the method used to decide sample size was not reported; however, the fact that statistical differences were found with some of the endpoints [9], suggests that the study was not universally underpowered. ## Data cleaning For the Study 1 dataset, several data subsets were constructed for male and female cats, according to age of neutering. The neutering age splits used corresponded to the lower quartile, median and upper quartile of the neutering ages observed for that sex across all castration and ovariohysterectomy procedures of DSH cats between April 1994 and November 2016 (regardless of whether the cat was eligible for the study population). Each subset comprised all data from sexually-intact cats of the relevant sex up to the lower end of the group’s neuter age range, then subsequently only cats neutered after the lower end but before the upper end of the group’s neuter age range. This resulted in 8 data subsets, comprising male and female subsets for each of the 4 neutering age ranges. The only data cleaning prior to this step was the removal of weights which had been rounded to the nearest whole unit (pounds), as previously described [9,16]. Subsequently, each of the data subsets was cleaned independently, again using a similar approach to that used for the dataset of the growth standards [16]. Briefly, extreme outliers (i.e., bodyweight entries >3 times the median value for cats >9mo age) were removed, as were population outliers, identified as data points outside of loess-smoothed curves fitted through the outlier limits of bodyweight for each of 50 equally-sized age groups. The outlier limits were defined as $175\%$ of the upper and lower whiskers of a box-and-whisker plot. Additional data cleaning measures were then implemented for cats with repeat visit data, to remove weights that were deemed to be implausible given the remaining weight trajectory for that individual. When considering the data for Study 2, it was assumed that for height, chest depth, length, elbow width, forelimb and hindlimb, a large drop (>$10\%$) over successive time points was unlikely, since it would be expected that these measurements should normally increase due to growth over the time intervals used in the trial. Pairs of data points displaying such a drop were identified and, for each pair, the data point with the largest z-score (calculated over all instances of that measurement at that time point) was removed. Weight, girth, ribcage, fat and lean were not cleaned in this way because these measures could be influenced by changes in adiposity, meaning they could not necessarily be assumed to be increasing. Instead, they were examined visually for any obviously erroneous data. ## Statistical analysis—Study 1 For the BANFIELD client-owned cats, the effect of neutering on bodyweight (analysis 1) was firstly assessed by modelling the growth trajectories separately in each of the data subsets, and then comparing the average growth trajectories with growth standards for sexually-intact cats [16]. The same modelling of growth trajectories was used as for the growth standards [16]; briefly, the trajectories were modelled using generalised modelling for location, shape and scale (GAMLSS), specifically the BCCG (Box-Cox Cole-Green) model, which is a semi-parametric technique allowing the central tendency, spread, skewness and kurtosis of the data to be estimated as smooth functions of the predictor variable(s). The variables used were age (raised to the power of 0.1 which initial data exploration suggested improved the model fit, as it did for the growth standards [16]) and bodyweight. The median and interquartile ($25\%$ and $75\%$) centiles were then extracted from each model and converted to z-scores according to the growth standards. The z-score transformation was used because it makes interpretation of different growth patterns more straightforward, since the standard centiles of the growth standards become equally spaced horizonal lines, with the median centile line lying along the x-axis. Further, the transformation converts a trajectory fully following the growth standards to a horizonal line, whilst positive and negative gradients depict faster and slower growth, respectively. These z-score transformations were displayed for each data subset, alongside the known neuter age-ranges, to enable visual inspection of any disturbances to average growth after neutering. A secondary analysis (analysis 2), compared z-scores before and after neutering within individuals. For each cat in the dataset, z-scores at two visits were identified: the last visit prior to neutering and the last visit in the data post-neutering. Kittens that did not have both such visits, or where those visits were less than 3 months apart, were excluded from the analysis. This subset was analysed using a linear model, which included the post-neutering z-score as the dependent variable and pre-neutering z-score, sex and neuter group as the predictor variables. For each combination of sex and neuter group, the estimated mean difference in z-score from pre- to post-neutering was calculated for a cat at median weight (i.e., z-score equal to zero) before neutering, along with P-values and $95\%$ confidence intervals. As we were making multiple comparisons, a Tukey HSD correction [23] was used to control the familywise error rate. $P \leq 0.05$ was considered to denote significance. The aim of this part of the analysis was to compare the patterns of growth (as measured by bodyweight) in kittens, neutered at four different age periods, with previously calculated growth standards based on sexually intact kittens. In analysis 1, the median growth trajectories for each neuter group were calculated and expressed as z-scores of the growth standards. These are shown in Figs 1 and 2, for male and female kittens, respectively; these same median growth trajectories are also depicted relative to the centile standards in the (S1 and S2 Figs). In all neuter groups of both sexes, median growth trajectory was close to the zero z-score line (representing the median growth standard) before neutering, but there was an upwards inclination in trajectory afterwards: for male kittens, the upwards inclination in median growth was 0.34, 0.41, 0.27 and 0.14, in neuter groups 1, 2, 3 and 4, respectively, by the end of the growth period; for female kittens, the upwards inclination in median growth was 1.27, 1.21, 1.39 and 0.51, in neuter groups 1, 2, 3 and 4, respectively, by the end of the growth period. For all ages, the upwards inclination was more marked in female compared with male kittens. However, in both sexes, the inclination was less steep in kittens neutered later. Fig 1: Median male growth trajectory by neuter group on z-score scale.Neuter Groups 1–4 represent, respectively, neutering ages of up to 20 weeks (0.7k cats), 20–23 weeks (2.1k cats), 23–28 weeks (2.5k cats) and >28 weeks (3.0k cats). Groups calculated from the lower quartile, median and upper quartile of ages at all neutering procedures performed on DSH cats between April 1994 and November 2016. The solid blue line represents the mean trajectory, whilst the blue-shaded ribbon represents the interquartile range. The grey shaded area represents the neutering age range for the group, and the solid grey vertical line shows the median observed neutering age. Grid lines represent the standard growth centiles, such that a growth trajectory following the centile curves would be horizonal in these plots. In all groups, there was an upwards inclination in growth trajectory, which was most marked in neuter groups 1 and 2. Fig 2: Median female growth trajectory by neuter group on z-score scale.Neuter Groups 1–4 represent, respectively, neutering ages of up to 21 weeks (1.0k cats), 21–25 weeks (2.7k cats), 25–29 weeks (3.0k cats) and >29 weeks (3.8k cats). Groups were calculated from the lower quartile, median and upper quartile of ages at all ovariohysterectomy procedures performed on DSH cats between April 1994 and November 2016. The solid blue line represents the mean trajectory, whilst the blue-shaded ribbon represents the interquartile range. The grey shaded area represents the neutering age range for the group, and the solid grey vertical line shows the median observed neutering age. Grid lines represent the standard growth centiles, such that a growth trajectory following the centile curves would be horizonal in these plots. In all groups, there was an upwards inclination in growth trajectory, which was marked in neuter groups 1–3, but only modest in neuter group 4. The purpose of analysis 2 was to compare the difference in z-score before and after neutering within individuals. Fig 3 shows the estimated change in z-score post-neutering for a cat at median weight, for different combinations of sex and neuter group, together with $95\%$ confidence intervals and Tukey post-hoc comparison groups. The number of cats in each of these combinations varied from 364 (Male, Neuter Group 4) to 653 (Female, Neuter Group 2). Fig 3: Estimated change in z-score post-neutering for a cat at median weight.Shown for different combinations of sex (female: Red; male: Blue) and neuter group, with 95% confidence intervals and Tukey post-hoc comparison groups. Neuter Groups 1–4 represent, respectively, neutering ages representing quartiles in the general neutered population–up to 20 weeks (448 cats), 20–23 weeks (451 cats), 23–28 weeks (507 cats) and >28 weeks (364 cats) for males, and up to 21 weeks (628 cats), 21–25 weeks (653 cats), 25–29 weeks (515 cats) and >29 weeks (497 cats) for females. The lower horizontal dotted line shows the point of zero change (meaning the cat remained on its initial centile line) and the upper horizontal dotted line represents an increase equal to one standard growth centile interval. The annotations give the Tukey post-hoc comparison groups, which indicate whether differences exist between estimated means for different combinations of neuter group and sex. Only comparisons between neuter groups of the same sex, and between sexes within the same neuter group were tested, and the groups within the relevant neuter group and relevant sex are as indicated in the annotations; within sets of means sharing the same neuter group or sex, those with different letters/numbers for that sex/group are significantly different at the 5% significance level. In all groups, there was a significant increase in z-score post neutering, and this was significantly larger in female than male. On average, there was a significant increase in z-score after neutering for all neuter groups and sexes (all P-values < 0.001), with this increase being larger for females than for males. It was also apparent that the latest neutered groups tended to show a directionally smaller increase than the earlier neutered groups, as noted in the population level analysis, although this was only significant for female neuter group 4 compared to earlier neutered groups. ## Statistical analysis—Study 2 The 11 morphological measurements were simultaneously analysed using a Bayesian mixed model with a multivariate endpoint. This approach was preferred to separate univariate models because it allows for correlations amongst measures and renders separate adjustments for multiple endpoints unnecessary. The dependent variables were log-transformed to improve normality and then standardised to similar scales (which can improve numeric efficiency) using a z-transformation. The model contained identical terms for all measures. Group (neutered or sexually-intact), week and their interaction were included as fixed factors, whilst litter ID and cat ID (nested in litter ID) were included as random factors. Correlation between measures was allowed at the level of both cat and litter. A normal distribution, with a mean of 0 and standard deviation of 4, was used as the prior for every endpoint; this distribution easily covered the entire set of raw data and, therefore, was not considered particularly informative; however, it still enabled the model to converge, which was not the case with a formal non-informative prior. Missing values in the dependent variables were imputed during model fitting, which ensured that other measures for that individual and timepoint could still be used. Posterior values were sampled using Markov chain Monte Carlo (MCMC); 4 chains of 10,000 values were simulated, with the first 5,000 values used as warm-up and the remaining values thinned to 1 in 5, giving a final set of 4,000 posterior samples. To assess model fit, a density trace comparing 40 posterior samples with the actual data was created for each endpoint for each week and residual plots were reviewed. Finally, estimated means and simultaneous $95\%$ credible intervals were calculated for each measure for each group and timepoint, and for the difference between the groups in the fold-change with respect to each measure from 11 weeks (when both groups were still sexually-intact) to each subsequent timepoint. Simultaneous credible intervals are essentially multivariate intervals which are calculated to have a given probability of containing the true value of the multivariate quantity under consideration and, therefore, control for type I error. The purpose of this second study was to compare growth-related measurements between pairs of sexually-intact and neutered female DSH littermates, in order to gain insights into the nature of growth changes associated with neutering; this was achieved with a Bayesian mixed model fitted to 11 morphological measurements taken at 4 time points. Overall, the fit of this model was deemed to be satisfactory, with the results of the posterior predictive checking of the model shown in S3 Fig. Posterior density traces coincided well with the observed traces for most of the measure-timepoint combinations; the main exceptions were chest depth and elbow width, at 10wk, and length at 18wk, where there were relatively small divergences. Fig 4 shows estimated means and simultaneous $95\%$ credible intervals for each measure, in each group (sexually-intact and neutered) and at each timepoint; Fig 5 shows the difference between the groups in the fold-change from 11wk (when both groups were still sexually intact) to each subsequent timepoint; and Fig 6 shows the proportional change in body composition between week 10 and week 52. Full details of means and $95\%$ credible intervals for the values at each timepoint, the fold change from 10wk-old and the difference in the fold change between groups for all endpoints are given in (S1 Table). By 52wk, neutered kittens were heavier (mean difference 1.34, 95-CI: 1.07–1.72), had more fat mass (mean difference 1.91, 95-CI 1.09–3.21) and more lean mass (mean difference 1.23, 95-CI: 1.03–1.48) compared with sexually-intact kittens. Abdominal girth (mean difference 1.20, 95-CI: 1.04–1.39) and rib cage length (mean difference 1.18, 95-CI: 1.02–1.36) were also greater, but there were no differences in other zoometric measurements. Fig 4: Estimated means and simultaneous 95% credible intervals for each measure, for each group and timepoint, shown on the original scales.The 11 sexually-intact cats are shown in red, whilst the 11 paired cats that were neutered are shown in blue; the grey dashed line indicates age of neutering. Panel A shows mass measurements, whilst and panels B and C show zoometric measurements. Weight, fat mass and lean mass measured by dual-energy X-ray absorptiometry (Hologic QDR-1000 W; Hologic, Inc., Waltham, MA, USA); height measured at the points of the scapula; ribcage measured at the deepest part of the thorax; length measured from manubrium of the sternum to a parallel point below the anus; girth measured around the narrowest point of the waist; forelimb bone measured from olecranon to carpus; elbow width, measured across the humeral condyles; hindlimb bone measured from patella to tarsus; chest depth measured mid-thorax from spine to the deepest part. Fig 5: The estimated mean and 95% simultaneous credible intervals (difference between sexually-intact and neutered kittens (11 cats per group), in the fold-change for each measure, between 11 wk (when both groups were still sexually intact) and each subsequent timepoint.Black dots represent the mean values, whilst error bars represent the 95% simultaneous credible intervals; the vertical grey dashed line indicates age of neutering. Differences are shown as the fold change of the neutered group divided by the fold change of the sexually-intact group, so that values >1 (shown by the horizontal dashed line) indicate that the neutered group had a greater difference in fold change. Panel A shows mass measurements, whilst and panels B and C show zoometric measurements. For details of the measurements see the legend for Fig 4. Fig 6: Estimated mean fat mass (green) and lean mass (lilac) at the initial and final measurement points for the sexually-intact and neutered kittens (11 cats per group).The intervals shown are simultaneous 95% credible intervals, whilst the green and lilac portions of each column represent fat and lean mass, respectively. ## Software All analyses were performed with an open-access online statistical software environment (R, version 3.6.1) [24], using the packages gamlss [25] for the GAMLSS modelling, the multcomp package [23] for multiple comparisons and brms [26] for the Bayesian modelling. Graphics were produced using ggplot2 [27]. ## Sample population, data extraction and cleaning Before creating subsets according to age at neutering, the dataset for Study 1 comprised 43.5k bodyweights (19.1k male; 24.4k female) across 11.4k cats (5.00k male; 6.41k female); this decreased to 36.7k bodyweights (16.1k male; 20.6k female) across 10.2k cats (4.45k male; 5.71k female) after removing weights that had been rounded to the nearest whole unit. The neutering age splits were calculated from a dataset of 188k castration and 181k ovariohysterectomy operations, comprising all such operations between April 1994 and November 2016 (regardless of whether the cat was eligible for the study population), as explained previously. For male kittens, the resulting groups were <0.39y, 0.39–0.44y, 0.44–0.54y and >0.54y; for female kittens, the resulting groups were <0.40, 0.40–0.48y, 0.48–0.56y and >0.56y. For convenience, the groups are subsequently referred to as neuter groups 1 to 4 for each sex, respectively. Table 1 shows the number of rows of data and individual cats in each neuter group at each stage in the subsequent data cleaning process, from subset creation to final datasets. Since each neuter group included data for all eligible sexually-intact kittens up to the lower end of the neuter age range, the datasets were not of equal size; the first group was notably smaller overall than the other three groups in both sexes, due to differences in the numbers of eligible cats and the distribution of visits with age (for example, some kittens did not have any visits prior to the lower boundary of the first neuter age range and, therefore, could not be used for this group). There was less variation in the number of cats remaining in each group after neutering, however, ranging from 692 to 945 in male kittens and 717 to 1,246 in female kittens. Table 1 \| Sex \| Neuter Groupa \| Cleaning Stage \| Cleaning Stage.1 \| Cleaning Stage.2 \| Cleaning Stage.3 \| Cleaning Stage.4 \| Cleaning Stage.5 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Sex \| Neuter Groupa \| Dataset Creation \| Removal of Extreme Outliers \| Population Level Cleaning \| Removal of Duplicate Observations \| Within-Cat Cleaning \| FINAL DATA \| \| M \| 1(up to 0.39y) \| 3,541 rows775 cats \| 3,541 rows775 cats \| 3,492 rows774 cats \| 3,490 rows774 cats \| 3,039 rows712 cats \| 3,039 rows712 cats \| \| \| 2(0.39–0.44y) \| 7,875 rows2,641 cats \| 7,876 rows2,641 cats \| 7,695 rows2,605 cats \| 7,689 rows2,605 cats \| 5,961 rows2,141 cats \| 5,961 rows2,141 cats \| \| \| 3(0.44–0.54y) \| 9,064 rows3,038 cats \| 9,063 rows3,038 cats \| 8,862 rows2,995 cats \| 8,855 rows2,995 cats \| 6,863 rows2,453 cats \| 6,863 rows2,453 cats \| \| \| 4(over 0.54y) \| 10,149 rows3,888 cats \| 10,148 rows3,888 cats \| 9,925 rows3,825 cats \| 9,916 rows3,825 cats \| 7,198 rows3,007 cats \| 7,198 rows3,007 cats \| \| F \| 1(up to 0.40y) \| 4,780 rows1,032 cats \| 4,779 rows1,032 cats \| 4,692 rows1,029 cats \| 4,687 rows1,029 cats \| 4,061 rows952 cats \| 4,061 rows952 cats \| \| \| 2(0.40–0.48y) \| 10,625 rows3,319 cats \| 10,622 rows3,319 cats \| 10,324 rows3,267 cats \| 10,310 rows3,267 cats \| 7,987 rows2,662 cats \| 7,987 rows2,662 cats \| \| \| 3(0.48–0.56y) \| 11,091 rows3,789 cats \| 11,090 rows3,789 cats \| 10,730 rows3,712 cats \| 10,772 rows3,712 cats \| 8,061 rows3,010 cats \| 8,061 rows3,010 cats \| \| \| 4(over 0.56y) \| 13,123 rows4,870 cats \| 13,122 rows4,870 cats \| 12,729 rows4,782 cats \| 12,719 rows4,782 cats \| 9,281 rows3,785 cats \| 9,281 rows3,785 cats \| The original raw data for Study 2 had 3 missing values (fat mass, 2; lean mass, 1), for unknown reasons, whilst a further 6 observations (chest depth, 3; height 2; hindlimb, 1) were removed during data cleaning. ## Discussion In the current study, we have compared patterns of growth in neutered healthy domestic shorthair kittens with growth depicted by standards created using data from sexually-intact kittens. Our ultimate aim was to inform guidelines to support the usage of the standards in neutered kittens. Neutering was associated with an upwards deviation in growth, which ranged from 0.14 of a centile unit in late-neutered male kittens to 1.4 centile units in early-neutered female kittens. Thus, the physiological changes resulting from neutering provoke more rapid weight gain such that cats are likely to be heavier when they reach skeletal maturity. It is noteworthy that these changes are more marked than was seen in a recent study in dogs [18]; although early neutering in dogs led to a modest increase in growth rate, and later neutering a modest decrease in growth rate, these changes were slight and within a single centile unit. The reasons for this difference between species are not known and require further study. As the purpose of the study was to examine weight changes post-neutering for cats showing apparently healthy growth, cats eligible for inclusion in the first part of the study had never received an abnormal body condition rating (including heavy, overweight or markedly obese) before the age of 2.5y. Therefore, the impact of neutering, and especially early neutering, on bodyweight calculated here is likely to be an under-estimate for the population as a whole. The results of the second part of this study suggest that the weight gain after neutering mainly (but not exclusively) comprised adipose tissue, which is consistent with previously-reported associations between neutering and obesity [7,8]. On the face of it, the results from these two studies contradict one another given that cats were reportedly in ideal condition, based on body condition assessments. However, this difference is likely to be due to the inaccuracy of body condition assessments rather than the fact that there was no gain in adipose tissue. Prior to 2010, veterinarians scored body condition subjectively using a 3-category system (underweight, normal, overweight) whilst, after this time, body condition was scored using a 5-point BCS [18]. Although BCS correlates moderately well with body fat mass measured by DEXA, it is only semi-quantitative, insensitive and only been validated in adult cats [17,28]. Further, scores are known to be affected by operator expertise, whereby BCS from veterinarians correlate better than for veterinary technicians and observers without training [29]. In previous studies, one unit on the 9-point system equates to a difference of approximately $10\%$ (but up to $15\%$) bodyweight [17,28], whilst half a unit on the 5-unit system equates with an approximately $7\%$ change in body fat percentage [29]; it is likely that the both the 3-category system and 5-point BCS used in the current study are even less sensitive (given fewer scores with which to differentiate cats). Therefore, slight increases in adipose tissue might well have been missed with both systems. Increases in adiposity might also have been obscured by other concurrent changes in body composition, given that a smaller, but nonetheless significant, increase in lean mass was also seen after neutering in the second study whilst some changes in body shape were also observed (e.g., increased abdominal girth and rib cage length). Alternatively, the bulk of the increase in adipose tissue could have been in an intra-abdominal, rather than subcutaneous, position. It is possible that both such changes (accumulation of abdominal fat, concurrent increase in lean mass) made an obvious change in BCS was less evident. The changes in lean mass observed after neutering in the current study contrast with those of other studies, where no change in lean mass was seen [30]. Increased lean mass was also not reported in the original paper about this study; only percentage lean was included [9]. This difference may be explained by the fact that percentage fat mass is negatively correlated with percentage lean mass such that as body fat percentage increases, lean tissue percentage decreases. Given that the difference in the fat mass increase between 10 and 52 weeks of age was of a much greater magnitude in the neutered kittens than the difference in the lean mass increase (1.91 vs. 1.23 times greater in the neutered kittens), the increase in lean mass was obscured. This effect would only be exacerbated if studies were insufficiently powered. The nature of this lean mass increase is unclear, not least because the lean mass measurement detected by DEXA not only includes muscle but other soft tissues such as abdominal organs [31]. Therefore, it is possible that the increase in lean mass might have included increased abdominal organ size; it has been found previously that the growth rate of pigs is positively correlated with abdominal organ mass [32]. Other body composition techniques, such as computed tomography, would be required to differentiate changes in muscle mass from organ mass. The finding of increased fat and lean mass is consistent with the effects of overnutrition in other species. For example, children with obesity are often tall for age, and there is an increase in lean mass as well as fat mass [33]. Further, an increased plane of nutrition in production animals also results in increases in stature, body fat and lean mass [33]. Such overnutrition is likely to be the direct result of physiological changes that occur after neutering, most notably increases in food intake which, as previously reported [8,9], result from the direct result of decreased sex hormone production, not least given that the effect can be reversed by administration of oestradiol [30]. A further consideration is the complex interaction between sex, hormones, growth and adiposity; as mentioned above, children with obesity are often taller than their peers in pre-puberty but have an earlier onset of puberty with less of a ‘growth spurt’ at this stage [34]. This is the result of early growth plate maturation, which is triggered by the hormonal changes that occur at puberty. Therefore, it is plausible that prepubertal neutering in kittens might delay epiphyseal growth plate closure, given an absence of the usual hormonal changes at this time, meaning an increase in long bone length and a taller stature compared with sexually-intact cats or those neutered after puberty. However, the results of the current study are not consistent with such a hypothesis because there were no significant differences in zoometric measurements pertaining to length (elbow width, forelimb, height and length). The findings of the current study, whereby neutering in cats was associated with weight gain and increased adipose tissue mass, is similar to the findings of other studies [7–9]. Previous studies have suggested that these changes are the result of increased voluntary food intake [7–9], possibly, with a concurrent decrease in physical activity and energy expenditure [8]; the latter is supported by findings from a recent study, examining growth patterns in male and female kittens, whereby neutering was associated with decreased energy requirements for growth in both male and female kittens [35]. Neutering had a greater impact on the growth trajectory of female compared with male kittens, and this has also been reported previously [6]. Although the reason for the difference between the sexes is not known, it might be due to effects of the absence of female sex hormones since oestradiol administration both to male and female cats abrogates the increased food intake that occurs in both male and female cats after neutering [30]. Experimental studies in rodents have suggested that this effect might be mediated through modulation of cholecystokinin (CCK)-dependent satiety signals [36,37]. However, further studies would be required to confirm whether the same effect occurs in cats. In recent, years the pre-pubertal neutering of cats has been promoted within the veterinary profession [19] mainly because of perceived benefits such as population control, as well as reducing development of unwanted male characteristics and behaviour (unwanted urination aggression, roaming and marking) and a decreased risk of some types of neoplasia [19,20,22]. Perceived barriers to early neutering, such as surgical and anaesthetic risk, have also been addressed with improvements in anaesthetic and surgical techniques, and recovery has been shown to be quicker [3]. Further, concerns over increased risk of developmental (orthopaedic or urogenital) disorders have not proved to be a major concern, at least in cats [2]. Thus, on a risk-benefit analysis, pre-pubertal neutering has largely been promoted as positive both ethically and on a welfare basis. However, in these discussions, relatively little consideration has been given to the possible effect of early neutering on increased adiposity and obesity. The results of the current study, which demonstrate that early neutering leads to a more profound increase in weight than neutering later on, especially in female kittens, should prompt a reconsideration of the risk-benefit analysis. Although it could be argued that any possible increased risk is easily addressed by recommending that owners limit food intake after neutering, this might be more difficult to implement in clinical practice. Obesity is a complex, multifactorial disease, associated with comorbidities, shortened lifespan and poorer quality of life [38]. Despite increasing awareness, prevalence has still increased dramatically recently, and strategies to manage it are imperfect [38]. Therefore, if early neutering is to be promoted by veterinarians, much better guidance on management of weight post-operatively is essential. The authors recommend regular and frequent weight checks after neutering, ideally at least every month, until the impact on weight status is established. The recently-developed growth charts [16] could be used to identify changes in growth trajectory, and adjustments to food intake can then be made, for example if centiles are crossed in an upwards direction. Such an approach would increase the likelihood of a cat reaching skeletal maturity still at a healthy weight, but it might not be sufficient to prevent obesity later in life, not least because weight is gained progressively throughout early-to-mid adult life [39]. A lifelong programme of weight monitoring would be needed to address this, as outlined previously [40]. Finally, close attention should be placed on nutritional management post-neutering, as recently reviewed [41]; recommendations would include avoiding ad libitum feeding, accurately calculating energy requirements (to 314 kJ [75 kcal] per kg0.67 per day) and feeding diet formulated for neutered cats [41]. Given the use of electronic patient records in study 1, it was not possible to determine whether any such strategies had been adopted in any of the cats that participated, and any impact it might have had. It is also not possible to determine whether such nutritional strategies post-neutering would have an effect on adipose tissue deposition given that, in study 2, all cats were offered free access to dry food during their growth phase. The study has some limitations that should be acknowledged, many of which have already been reported previously [16,18]. First, the population used for study 1 was restricted to cats in optimal body condition, so as to fulfil the first aim of the study of informing recommendations around using the growth standards in neutered kittens. Given that the magnitude of weight gain in kittens developing obesity during growth would be even more than seen in the kittens we studied, it is likely that we have under-estimated the effect of neutering on body fat mass in the population as a whole. Second, data from pet cats were collected retrospectively, over an extended period, from a network of veterinary hospitals across North America. This might have introduced variability, because of changes to practice protocols and data collection, as discussed elsewhere [16,18], and the results from these cats might not be directly comparable with the British research colony cats used in the second study. Finally, only female kittens neutered at 19 weeks were assessed in study 2 meaning that we were unable to assess the effects in male kittens and also differences in neutering at different ages. Further studies would be required both to confirm and extend the current findings. ## Conclusion Neutering in pet kittens leads to an upwards inclination in growth, with the effect being greater in female than male kittens and more profound in kittens neutered earlier in life. Changes in weight after neutering are mainly the result of increased adipose tissue, although there is a lesser gain in lean mass. Veterinarians should be aware of the potential impact that neutering, especially when undertaken early in life, has on gain of adipose tissue. As well as considering this in any risk-benefit analysis about age of neutering, close-monitoring of bodyweight should be undertaken post-surgery, to identify and rectify unwanted weight gain. ## References 1. Sánchez-Vizcaíno F, Noble PM, Jones PH, Menacere T, Buchan I, Reynolds S. Demographics of dogs, cats, and rabbits attending veterinary practices in Great Britain as recorded in their electronic health records. BMC Vet Res (2017.0) 13 218. DOI: 10.1186/s12917-017-1138-9 2. Reichler IM. Gonadectomy in cats and dogs: a review of risks and benefits. Reprod Domest Anim (2009.0) 44 29-35. DOI: 10.1111/j.1439-0531.2009.01437.x 3. Root Kustritz MV. Pros, cons, and techniques of pediatric neutering. Vet Clin North Am Small Anim Pract (2014.0) 44 221-233. DOI: 10.1016/j.cvsm.2013.10.002 4. Panciera DL, Thomas CB, Eicker SW, Atkins CE. Epizootiologic patterns of diabetes mellitus in cats: 333 cases (1980–1986). J Am Vet Med Assoc (1990.0) 197 1504-1508. PMID: 2272886 5. Prahl A, Guptill L, Glickman NW, Tetrick M, Glickman LT. Time trends and risk factors for diabetes mellitus in cats presented to veterinary teaching hospitals. J Feline Med Surg (2007.0) 9 351-358. DOI: 10.1016/j.jfms.2007.02.004 6. Fettman MJ, Stanton CA, Banks LL, Hamar DW, Johnson DE, Hegstad RL. Effects of neutering on bodyweight, metabolic rate and glucose tolerance of domestic cats. Res Vet Sci (1997.0) 62 131-136. DOI: 10.1016/s0034-5288(97)90134-x 7. Nguyen PG, Dumon HJ, Siliart BS, Martin LJ, Sergheraert R, Biourge VC. Effects of dietary fat and energy on body weight and composition after gonadectomy in cats. Am J Vet Res (2004.0) 65 1708-1713. DOI: 10.2460/ajvr.2004.65.1708 8. Vester BM, Sutter SM, Keel TL, Graves TK, Swanson KS. Ovariohysterectomy alters body composition and adipose and skeletal muscle gene expression in cats fed a high-protein or moderate-protein diet. Animal (2009.0) 3 1287-1298. DOI: 10.1017/S1751731109004868 9. Alexander LG, Salt C, Thomas G, Butterwick R. Effects of neutering on food intake, body weight and body composition in growing female kittens. Br J Nutr (2011.0) 106 S19-S23. DOI: 10.1017/S0007114511001851 10. German AJ, Ryan VH, German AC, Wood IS, Trayhurn P. Obesity, its associated disorders and the role of inflammatory adipokines in companion animals. Vet J (2010.0) 185 4-9. DOI: 10.1016/j.tvjl.2010.04.004 11. Murray JK, Roberts MA, Whitmarsh A, Gruffydd-Jones TJ. Survey of the characteristics of cats owned by households in the UK and factors affecting their neutered status. Vet Rec (2009.0) 164 137-141. DOI: 10.1136/vr.164.5.137 12. Chu K, Anderson WM, Rieser MY. Population characteristics and neuter status of cats living in households in the United States. J Am Vet Med Assoc (2009.0) 234 1023-1030. DOI: 10.2460/javma.234.8.1023 13. Clark L, Seawright AA, Gartner RJW. Longbone abnormalities in kittens following vitamin a administration. J Comp Pathol (1970.0) 80 113-121. DOI: 10.1016/0021-9975(70)90038-1 14. Salt C, Morris PJ, Butterwick RF, Lund EM, Cole TJ, German AJ. Comparison of growth patterns in healthy dogs and dogs in abnormal body condition using growth standards. PLoS One (2020.0) 15 e0238521. DOI: 10.1371/journal.pone.0238521 15. Serisier S, Feugier A, Venet C, Biourge V, German AJ. Faster growth rate in ad libitum-fed cats: a risk factor predicting the likelihood of becoming overweight during adulthood. J Nutr Sci (2013.0) 2 e11. DOI: 10.1017/jns.2013.10 16. Salt C, German AJ, Henzel KS, Butterwick RF. Growth standard charts for monitoring bodyweight in intact domestic shorthair kittens from the USA. PLoS ONE 2022 (2022.0) 17 e0277531. DOI: 10.1371/journal.pone.0277531 17. Laflamme D. Development and validation of a body condition score system for cats: a clinical tool. Feline pract (1997.0) 25 13-18 18. Salt C, Morris PJ, German AJ, Wilson D, Lund EM, Cole TJ. Growth standard charts for monitoring bodyweight in dogs of different sizes. PLoS One (2017.0) 12 e0182064. DOI: 10.1371/journal.pone.0182064 19. Mazeau L, Wylie C, Boland L, Beatty JA. A shift towards early-age desexing of cats under veterinary care in Australia. Sci Rep (2021.0) 11 811. DOI: 10.1038/s41598-020-79513-6 20. Spain CV, Scarlett JM, Houpt KA. Long-term risks and benefits of early-age gonadectomy in dogs. J Am Vet Med Assoc (2004.0) 224 380-387. DOI: 10.2460/javma.2004.224.380 21. Root MV. Early spay–neuter in the cat: effect on development of obesity and metabolic rate. Vet Clin Nutr (1995.0) 2 132-134 22. Stubbs WP, Bloomberg MS, Scruggs SL, Shille VM, Lane TJ. Effects of prepubertal gonadectomy on physical and behavioral development in cats. J Am Vet Med Assoc (1996.0) 209 1864-1871. PMID: 8944799 23. Hothorn T, Bretz F, Westfall P. Simultaneous inference in general parametric models. Biom J (2008.0) 50 346-363. DOI: 10.1002/bimj.200810425 24. 24R Core Team. R: A language and environment for statistical computing. [cited 2021 February 2]. R Foundation for Statistical Computing, Vienna, Austria. http://www.r-project.org/. 25. Rigby RA, Stasinopoulos DM. Generalized additive models for location, scale and shape. J R Stat Soc: Ser C Appl. Stat (2005.0) 54 507-554. DOI: 10.1111/j.1467-9876.2005.00510.x 26. Bürkner PC. brms: An R package for Bayesian multilevel models using Stan. J Stat Soft (2017.0) 80 1-28. DOI: 10.18637/jss.v080.i01 27. Wickham H.. ggplot2: Elegant Graphics for Data Analysis (2016.0) 28. German AJ, Holden SL, Moxham GL, Holmes KL, Hackett RM, Rawlings JM. A simple, reliable tool for owners to assess the body condition of their dog or cat. J Nutr (2006.0) 136 2031S-2033S. DOI: 10.1093/jn/136.7.2031S 29. Shoveller AK, DiGennaro J, Lanman C, Spangler D. Trained vs untrained evaluator assessment of body condition score as a predictor of percent body fat in adult cats. J Feline Med Surg (2014.0) 16 957-65. DOI: 10.1177/1098612X14527472 30. Cave NJ, Backus RC, Marks SL, Klasing KC. Oestradiol, but not genistein, inhibits the rise in food intake following gonadectomy in cats, but genistein is associated with an increase in lean body mass. J Anim Physiol Anim Nutr (Berl) (2007.0) 91 400-410. DOI: 10.1111/j.1439-0396.2006.00667.x 31. Buckinx F, Landi F, Cesari M, Fielding RA, Visser M, Engelke K. Pitfalls in the measurement of muscle mass: a need for a reference standard. J Cachexia Sarcopenia Muscle (2018.0) 9 269-278. DOI: 10.1002/jcsm.12268 32. Koong LJ, Nienaber JA, Pekas JC, Yen JT. Effects of plane of nutrition on organ size and fasting heat production in pigs. J Nutr (1982.0) 112 1638-42. DOI: 10.1093/jn/112.8.1638 33. Forbes GB. Lean body mass and fat in obese children. Pediatrics (1964.0) 34 308-314. PMID: 14211097 34. Chung S. Growth and Puberty in Obese Children and Implications of Body Composition. J Obes Metab Syndr (2017.0) 26 243-250. DOI: 10.7570/jomes.2017.26.4.243 35. Merenda MEZ, Sato J, Scheibel S, Uemoto AT, Rossoni DF, Dos Santos MP. Growth Curve and Energy Intake in Male and Female Cats. Top Companion Anim Med (2021.0) 44 100518. DOI: 10.1016/j.tcam.2021.100518 36. Geary N, Trace D, McEwen B, Smith GP. Cyclic estradiol replacement increases the satiety effect of CCK-8 in ovariectomized rats. Physiol Behav (1994.0) 56 281-9. DOI: 10.1016/0031-9384(94)90196-1 37. Geary N, Trace D, McEwen B, Smith GP. Cyclic estradiol replacement increases the satiety effect of CCK-8 in ovariectomized rats. Physiol Behav (1994.0) 56 281-9. DOI: 10.1016/0031-9384(94)90196-1 38. German AJ. Weight management in obese pets: the tailoring concept and how it can improve results. Acta Vet Scand (2016.0) 58 57. DOI: 10.1186/s13028-016-0238-z 39. Serisier S, Feugier A, Venet C, Biourge V, German AJ. Faster growth rate in ad libitum-fed cats: a risk factor predicting the likelihood of becoming overweight during adulthood. J Nutr Sci (2013.0) 2 e11. DOI: 10.1017/jns.2013.10 40. German AJ. Obesity Prevention and Weight Maintenance After Loss. Vet Clin North Am Small Anim Pract (2016.0) 46 913-29. DOI: 10.1016/j.cvsm.2016.04.011 41. Vendramini THA, Amaral AR, Pedrinelli V, Zafalon RVA, Rodrigues RBA. Neutering in dogs and cats: current scientific evidence and importance of adequate nutritional management. Nutr Res Rev (2020.0) 33 134-144. DOI: 10.1017/S0954422419000271
--- title: 'Metabolic, inflammatory and adipokine differences on overweight/obese children with and without metabolic syndrome: A cross-sectional study' authors: - Idalia Cura–Esquivel - Marlene Marisol Perales-Quintana - Liliana Torres-González - Katia Guzmán-Avilán - Linda Muñoz-Espinosa - Paula Cordero-Pérez journal: PLOS ONE year: 2023 pmcid: PMC10016645 doi: 10.1371/journal.pone.0281381 license: CC BY 4.0 --- # Metabolic, inflammatory and adipokine differences on overweight/obese children with and without metabolic syndrome: A cross-sectional study ## Abstract ### Background Obesity is associated with low-grade inflammation and metabolic syndrome (MetS) in both children and adults. Our aim was to describe metabolic, inflammatory and adipokine differences on overweight/obese children with and without MetS. ### Methods This was an observational study. A total of 107 children and adolescents aged 6–18 years were included. Among this sample, $$n = 21$$ had normal body weight, $$n = 22$$ had overweight/obesity without MetS, and $$n = 64$$ had overweight/obesity with MetS. Anthropometric data and biochemical, adipokine, and inflammatory markers were measured. Different ratios were then assessed for estimate the probability of MetS. ROC analysis was used to estimate the diagnostic accuracy and optimal cutoff points for ratios. ### Results Serum CRP levels were higher among children with overweight/obesity with MetS. Adipokines like PAI-1 and leptin were significantly lower in children with normal body weight. The Adipo/Lep ratio was highest in the group with normal body weight. TG/HDL-C and TC/HDL-C ratios were significantly correlated with BMI, DBP, PCR, and PAI-1. TC/HDL-C ratio was significantly correlated with SBP and resistin. TGL/HDL-C ratio was significantly correlated with waist and hip circumferences, fasting glucose, and MCP-1. The AUC for TG/HDL-C at the optimal cutoff of 2.39 showed $85.71\%$ sensitivity and $71.43\%$ specificity. CT/HDL-C at the optimal cutoff of 3.70 showed $65.08\%$ sensitivity and $81.82\%$ specificity. Levels of both ratios increased significantly as additional MetS criteria were fulfilled. ### Conclusion Low-grade inflammation is correlated with MetS in children with overweight/obesity. TGL, HDL-C and TGL/HDL-C ratio, obtainable from routine lab tests, allows identification of MetS in children with overweight or obesity. ## Introduction Childhood obesity is a global public health problem. Obesity, a state of chronic low-grade inflammation, results from accumulation of visceral fat, which leads to complications such as metabolic syndrome (MetS). There is currently no clear consensus on the definition of pediatric MetS [1]; however, the term refers to a set of metabolic risk factors that include obesity, dyslipidemia, hypertension, and type 2 diabetes mellitus [2]. The prevalence of MetS among children in Mexico has been reported to be as high as $54.6\%$ [3]. Its increase in recent decades has also raised the prevalence of associated comorbidities and, since it is also considered a predictor of cardiometabolic diseases in adulthood, identification and early therapeutic intervention are crucial. It has been postulated that peripheral insulin resistance (IR) and abdominal obesity are the main factors contributing to MetS, and that its metabolic changes affect lipid metabolism due to increased low-density lipoprotein cholesterol (LDL-C), decreased high-density lipoprotein cholesterol (HDL-C), increased triglycerides (TGL), and increased fatty acids [4]. Obesity is also linked to changes in serum lipoproteins, which are in turn associated with the development of atherosclerosis. Evidence suggests that the atherosclerotic process begins in childhood. The prevalence of atherogenic dyslipidemia is increasing among children and adolescents with obesity, and is characterized by hypertriglyceridemia, increased very-low-density lipoprotein cholesterol (VLDL-C), and reduced HDL-C; its association with MetS also increases cardiovascular disease risk [5]. In adults, the relation between lipids such as TGL and HDL-C and the ratio of total cholesterol (TC) and HDL-C are widely used to assess MetS. These ratios indicate balance between all atherogenic cholesterols (including VLDL-C and HDL-C) and are thus important determinants of cardiovascular risk. Obesity is related to both classic and novel risk factors, including prothrombotic factors (fibrinogen, plasminogen activator inhibitor-1 [PAI-1], homocysteine), inflammatory factors (interleukin 10 [IL-10], interleukin 6 [IL-6], tumor necrosis factor alpha [TNF-α], monocyte chemoattractant protein 1 (MCP-1), C-reactive protein [CRP]), and some adipocytokines (leptin, adiponectin) [6, 7]. Inflammation arising from adipose tissue has been identified as an important source of systemic inflammation and may be associated with IR. The complex MetS pathophysiology is also associated with hormone (adipokines) changes and inflammatory markers [8]. Adiponectin plays a protective role against IR and cardiovascular diseases (CVD) [9]. In contrast, leptin has proinflammatory effects; high levels are associated with development of IR and CVD. Hypoadiponectinemia and hyperleptinemia are observed in both adults and children with obesity, and the adiponectin/leptin (Adipo/Lep) ratio has been proposed as a sensitive MetS marker in children and adolescents [10]. The state of low-grade inflammation in obesity is exacerbated in individuals with MetS. Specifically, increased levels of inflammatory markers, including CRP, have been detected in children, adolescents, and adults with obesity and MetS [11]. Thus, recent attention has focused on the relations between inflammation and hormonal dysfunction (adipokines), and their relations with MetS. As such, the objective herein was to describe metabolic, inflammatory and adipokine differences on overweight/obese children with and without MetS. ## Design of the study This was a cross-sectional analytical study of children and adolescents attending pediatric consultation at the University Hospital “Dr. José Eleuterio González” of the Autonomous University of Nuevo León in Monterrey, N.L., Mexico conducted between January 2017 and December 2019. This public hospital, with 500 beds, is the largest in Northeast Mexico with patients coming principally from the State of Nuevo Leon and surrounding states in Northern Mexico (Coahuila, Tamaulipas, and San Luis Potosi). The institutional ethics committee approved the study (PE17-00010). A detailed letter explaining the study aims was provided to all parents or guardians and informed consent was obtained. ## Study population Three groups were included: (I) Normal weight children: healthy children with adequate weight and height for age; (II) Obese / overweight children without metabolic syndrome; (III) Obese / overweight children without metabolic syndrome. The inclusion criteria for group (II) and (III) were: younger than age 18 years; and body mass index (BMI) ≥85th percentile according to the Centers for Disease Control and Prevention (CDC). The exclusion criteria were: congenital malformation; previous diagnosis with endocrinological, kidney, or hepatic disorder; use of any medication affecting serum lipid concentration; and refusal to participate in the study. ## Definitions Overweight and obesity were defined according to the criteria established by the CDC. Overweight was considered a BMI between the 85th and 95th percentiles. Obesity was considered a BMI ≥95th percentile. MetS was defined according to the de Ferranti criteria [12] and was considered present when the patient met three or more of the following criteria: (I) abdominal obesity defined as a waist circumference (WC) >75th percentile; (II) hypertension defined as a blood pressure >90th percentile; (III) TGL >100 mg/dL; (IV) HDL-C <50 mg/dL; and (V) fasting glucose >100 mg/dL. The Adipo/Lep ratio was obtained as (Serum adiponectin levels) / (Serum leptin levels). The TC/HDL-C was calculated as (Total cholesterol) / (High-density lipoprotein cholesterol). The TGL/HDL-C was obtained as (Triglycerides) / (High-density lipoprotein cholesterol). While the RCP/HDL-C was obtained as (C-reactive protein) / (High-density lipoprotein cholesterol). ## Data collection At the hospital visit, sex and age were recorded and anthropometrics (height, weight and waist and hip circumferences) were measured. BMI was calculated as body weight (kg)/height2 (m). Blood pressure was measured using a sphygmomanometer while the child was seated. ## Biochemical and inflammatory parameters Blood samples were taken to measure biochemical and inflammatory parameters, and adipokines. TC, HDL-C, TGL, and glucose levels were determined using an ILAB-Aries self-analyzer spectrophotometer and diagnostic kits (Instrumentation Laboratory, Bedford, MA, USA) according to the supplier’s specifications. Cytokine (IL-6, TNF-α, MCP-1) concentrations were measured using a commercially available enzyme-linked immunoassays (Human IL-6 Immunoassay, Quantikine ELISA Kit; Human TNF-α Quantikine ELISA Kit; and Human CCL2/MCP-1 Immunoassay, respectively, Bio-Techne, Minneapolis, MN, USA) and are reported in pg/mL. Inflammatory marker CRP was measured by human CRP ELISA kit (Bio-Techne, Minneapolis, MN, USA) and is reported in mg/L. Adipokine (adiponectin, leptin), resistin, and PAI-1 levels were measured using an enzyme-linked immunoassay kit. Serum leptin level was measured by human leptin ELISA, Clinical Range kit and is reported in ng/mL (BioVendor Research and Diagnostic products, Karasek, Czech Republic). Adiponectin was measured by a Human Adiponectin/Acrp30 DuoSet ELISA kit and is reported in mg/mL and resistin was measured by a Human Resistin Quantikine ELISA Kit (both from R&D Systems, Minneapolis, MN, USA) and are reported in ng/mL. PAI-1 was measured by a PAI1 Human ELISA Kit and is reported in ng/mL (Thermo Fisher Scientific, Waltham, MA, USA). Ratios previously described within populations of patients who with overweight and obesity were also evaluated: Adipo/Lep, TC/HDL-C, TGL/HDL-C. ## Statistical analysis Analyses were performed using GraphPad Prism software (v. 6.0; GraphPad, San Diego, CA, USA) or SPSS software (v.22.0; Chicago, Ill., USA) and MedCalc Statistical Software version 20.009 (MedCalc Software bvba, Ostend, Belgium). Normally distributed variables are presented as means and standard deviations and were analyzed by ANOVA-tests. Non-normally distributed variables are presented as medians and interquartile ranges and were compared by Kruskal-Wallis tests. Bivariate and multivariate logistic regression analyses were conducted to determine factors associated with MetS, variables with $p \leq 0.05$ in bivariate analysis were included in multivariate analysis. Receiver operating characteristic (ROC) analysis was performed to determine the area under the curve (AUC) to assess the precision of the TGL/HDL and TC/HDL ratios for identify children with overweight/obesity, with and without MetS. To determine the optimal cutoff point, the Younden index was used. Sensitivity and specificity of the cutoff points were calculated. A correlation study for TGL/HDL-C and TC/HDL-C was carried out using the Spearman correlation. For all analyses, $p \leq 0.05$ was considered statistically significant. ## Sample characteristics The total sample was 107 patients, among whom 63 were male ($58.80\%$) and 44 were female ($41.10\%$); their mean age was 10.52(1.76) years. Among the total sample, 21 children ($19.60\%$) had normal body weight and 86 ($80.40\%$) had overweight/obesity. ## MetS diagnosis Among the study sample, 64 ($59.81\%$) had obesity or overweight and met the MetS diagnostic criteria of three of the five Ferranti criteria. In this subgroup, $76.64\%$ had abdominal obesity, $33.64\%$ presented arterial hypertension, $60.75\%$ had elevated TGL levels, $81.31\%$ had low HDL-C levels, and $4.67\%$ presented with hyperglycemia (Fig 1). Fig 1: Frequency of the Ferranti criteria used for the diagnosis of metabolic syndrome. ## Anthropometric, biochemical, adipokine, and cytokine characteristics Sample anthropometric, biochemical, adipokine, and cytokine characteristics are described in Table 1. Compared with children with normal body weight, those with overweight/obesity and MetS had significantly higher BMI, systolic blood pressure (SBP), diastolic blood pressure (DBP), TGL, and MCP-1, and significantly lower HDL-C and TGL/HDL-C ratio (4.51(3.15); $p \leq 0.0001$). There were no significant differences on indices between the children with overweight/obesity without MetS and those with normal body weight ($p \leq 0.05$) (Table 1). Table 1 \| Unnamed: 0 \| Normal weight children (n = 21) \| Obese/overweight children without metabolic syndrome (n = 22) \| Obese/overweight children with metabolic syndrome (n = 64) \| p \| \| --- \| --- \| --- \| --- \| --- \| \| Age in years, mean (age range) \| 8.70 (6–11) \| 10.00 (6–15) \| 11.00 (8–15) \| \| \| Sex Male/Female, n (%) \| 12/9 (57.14/42.86) \| 14/8 (63.64/36.36) \| 37/27 (57.81 / 42.19) \| \| \| Obese / Overweight \| \| 17/5 (77.27 / 22.73) \| 54/10 (84.38 / 15.63) \| \| \| Anthropometric variables, mean (SD) \| Anthropometric variables, mean (SD) \| Anthropometric variables, mean (SD) \| Anthropometric variables, mean (SD) \| Anthropometric variables, mean (SD) \| \| Height (cm) \| 134.00 (9.30) \| 149.00 (11.00) \| 147.00 (8.70) \| < 0.001 † ‡ \| \| Weight (kg) \| 28.00 (4.80) \| 61.00 (17.00) \| 64.00 (16.00) \| < 0.001 † ‡ \| \| BMI (kg/m2) \| 16.00 (2.10) \| 27.00 (4.50) \| 29.00 (5.30) \| < 0.001 † ‡ \| \| Waist circunference (cm) \| 57.00 (6.40) \| 91.00 (12.00) \| 89.00 (9.20) \| < 0.001 † ‡ \| \| Hip circunference(cm) \| 67.00 (7.40) \| 97.00 (12.00) \| 99.00 (11.00) \| < 0.001 † ‡ \| \| Blood pressure, mean (SD) \| Blood pressure, mean (SD) \| Blood pressure, mean (SD) \| Blood pressure, mean (SD) \| Blood pressure, mean (SD) \| \| SBP (mmHg) \| 98.00 (7.90) \| 108.00 (11.00) \| 111.00 (16.00) \| 0.001 † ‡ \| \| DBP (mmHg) \| 60.00 (8.00) \| 59.00 (10.00) \| 69.00 (10.00) \| < 0.001 ‡ § \| \| Biochemical variables, median (IQR) \| Biochemical variables, median (IQR) \| Biochemical variables, median (IQR) \| Biochemical variables, median (IQR) \| Biochemical variables, median (IQR) \| \| Total cholesterol (mg/dL) \| 150.0 (130.0–173.0) \| 141.00 (116.0–160.0) \| 157.00 (134.0–184.0) \| 0.075 \| \| HDL cholesterol (mg/dL) \| 49.00 (55.00–41.00) \| 46.00 (34.00–51.00) \| 39.00 (34.00–43.00) \| < 0.001 ‡ § \| \| Triglycerides (mg/dL) \| 76.00 (57.00–99.00) \| 75.00 (51.00–101.00) \| 183.00 (123.00–223.00) \| < 0.001 ‡ § \| \| Fasting glucose (mg/dL) \| 80.00 (78.00–85.00) \| 78.00 (74.00–85.00) \| 83.00 (77.00–87.00) \| 0.052 \| \| CRP (mg/L) \| 0.10 (0.10–0.38) \| 0.50 (0.15–0.89) \| 1.10 (0.33–3.50) \| < 0.001 † ‡ \| \| Adipokines, median (IQR) \| Adipokines, median (IQR) \| Adipokines, median (IQR) \| Adipokines, median (IQR) \| Adipokines, median (IQR) \| \| Adiponectin (mg/mL) \| 31.13 (12.46–36.90) \| 25.42 (17.90–36.83) \| 23.82 (15.56–32.87) \| 0.258 \| \| Resistin (ng/mL) \| 17.43 (12.05–19.60) \| 19.68 (16.55–29.30) \| 21.35 (18.31–28.61) \| 0.020 ‡ \| \| PAI-1 (ng/mL) \| 19.18 (15.31–22.58) \| 27.61 (23.20–37.38) \| 27.37 (17.89–41.26) \| <0.001 † ‡ \| \| Leptin (ng/mL) \| 2.31 (1.47–4.90) \| 18.17 (11.74–29.17) \| 16.89 (2.52–23.62) \| < 0.001 † ‡ \| \| Cytokines, median (IQR) \| Cytokines, median (IQR) \| Cytokines, median (IQR) \| Cytokines, median (IQR) \| Cytokines, median (IQR) \| \| IL-6 (pg/mL) \| 16.76 (16.66–17.16) \| 17.16 (16.71–18.01) \| 17.60 (16.76–18.75) \| 0.007 ‡ \| \| TNF-α (pg/mL) \| 17.65 (13.51–20.31) \| 17.65 (16.02–21.45) \| 18.68 (16.30–127.30) \| 0.018 ‡ \| \| MCP-1 (pg/mL) \| 331.6 (182.7–405.3) \| 279.8 (395.4–231.3) \| 372.9 (272.2–770.60) \| 0.033 ‡ § \| \| Ratio, mean (SD) \| \| \| \| \| \| Adipo/Lep ratio \| 6.68 (7.89) \| 1.08 (1.62) \| 1.08 (0.85) \| < 0.001 † ‡ \| \| RCP/HDL-C ratio \| 0.002 (0.001) \| 0.01 (0.01) \| 0.02 (0.05) \| < 0.001 † ‡ \| \| CT/HDL-C ratio \| 3.08 (0.98) \| 3.28 (0.53) \| 4.24 (1.16) \| < 0.001 ‡ § \| \| TGL/HDL-C ratio \| 1.46 (0.91) \| 1.78 (1.84) \| 4.51 (3.15) \| < 0.001 ‡ § \| After bivariate analysis, we found that children with high levels of HDL-C have lower probability of having MetS (OR = 0.88, $95\%$ CI = 0.81–0.95, $$p \leq 0.002$$). In addition, children with high levels of TGL have lower probability of having MetS (OR = 1.022, $95\%$ CI = 1.01–1.033, $p \leq 0.001$). Serum leptin, resistin, PAI-1, and CRP levels differed significantly between children with normal body weight and those with overweight/obesity, and between those with and without MetS (all $p \leq 0.05$), being higher in children with overweight/obesity and MetS. The highest adiponectin levels were in children with normal body weight (31.13; IQR 12.46–36.90 mg/mL) and the lowest were in children with overweight/obesity with MetS (23.82; IQR 15.56–32.87 mg/mL). In contrast, the highest leptin levels were in children with overweight/obesity without MetS (18.17; IQR 11.74–29.17 ng/mL) and the lowest were in those with normal body weight (2.31; IQR 1.47–4.90 ng/mL) (Table 1). Consequently, the Adipo/Lep ratio was highest in children with normal body weight (6.68(7.89)), lower in those with overweight/obesity, and not significantly different between those without and with MetS (1.08(1.62) vs 1.08(0.85), respectively) (Table 1). Serum CRP levels in children with normal body weight (0.10; IQR 0.10–0.38 mg/L) were significantly lower than in those with obesity and MetS (1.10; IQR 0.33–3.50 mg/L, $p \leq 0.0001$). Among the ratios evaluated (Table 1), only TC/HDL-C and TGL/HDL-C differed significantly between children with overweight/obesity with and without MetS ($p \leq 0.0001$). TC/HDL-C and TGL/HDL-C ROC curve analyses differentiated between children with overweight/obesity with MetS, as shown in Fig 2 and Table 2. TGL/HDL-C had the highest probability for MetS with an AUC of 0.85 and a cutoff value >2.39. Correlations between TG/HDL-C and TC/HDL-C and important variables are shown in Table 3. Both ratios were correlated with BMI, DBP, PCR, and PAI-1. Only the TC/HDL-C ratio was significantly correlated with SBP and resistin. Only the TGL/HDL-C ratio was significantly correlated with waist and hip circumferences, fasting glucose, and MCP-1. Fig 2: ROC curves for TGL/HDL-C and TC/HDL-C ratios in predicting metabolic syndrome in children with overweigh or obesity. TABLE_PLACEHOLDER:Table 2 TABLE_PLACEHOLDER:Table 3 Fig 3: Levels of TGL/HDL-C and TC/HD-C ratios in relation to the number of fulfilled Ferranti criteria.‡ showed significant difference vs 0 Ferranti criteria fulfilled; § showed significant difference vs 1 Ferranti criteria fulfilled; # showed significant difference vs 2 Ferranti criteria fulfilled; & showed significant difference vs 0 Ferranti criteria fulfilled. ## Discussion Obesity-related diseases and complications were previously considered exclusive to the adult population, so evaluating them in childhood or adolescence was not routine practice. Nevertheless, many studies have now shown that childhood obesity tends to perpetuate into adulthood; this favors early development of metabolic disease and increases risk for CVD and diabetes, which in turn decreases life expectancy [13, 14]. Herein, we evaluated children with overweight/obesity and identified that in addition to abnormal weight, they typically present with at least one metabolic alteration, of which the most prevalent are dyslipidemia ($60.75\%$) and arterial hypertension ($33.64\%$). The prevalence of pediatric MetS is variable and depends on the diagnostic criteria and component cutoff values [1]. We found a MetS prevalence of $58.81\%$ using the Ferranti criteria, consistent with the meta-analysis by Bitew et al. showing a general prevalence of $56.32\%$ in studies using the same criteria [15]. The most prevalent MetS criterion herein were low HDL-C ($81.31\%$), central obesity ($76.64\%$), and high total TGL ($60.75\%$). Obesity is a state of chronic low-grade inflammation, in which nutrient overload, increased metabolic demands, and lipotoxicity at the adipose level contribute to production of inflammatory mediators. Some of these markers are synthesized by adipocytes, including acute phase proteins such as CRP, haptoglobin, PAI-1, TNF-α, resistin, and cytokines (IL-1b, IL-6, IL-8, IL-10) [16, 17]. MetS is characterized by multiple cardiovascular risk factors; the endothelial dysfunction from this prothrombotic, inflammatory state is caused by the expression of inflammatory cytokines and cell adhesion molecules [17, 18]. As the main inhibitor of fibrinolysis, high levels of PAI-1 can increase coronary heart disease risk. Increased PAI-1 is involved with control of insulin signaling in adipocytes and can be considered a component of MetS [19]. IR states have been associated with elevated PAI-1 levels and altered plasma lipids, which helps explain the characteristic prothrombotic state of these pathologies [19, 20]. Children in our sample with overweight or obesity showed elevated PAI-1 levels compared with those with normal body weight, and similarly elevated TGL and blood pressure. In contrast, PAI-1 levels did not differ between those with and without MetS, providing further evidence of its association with plasma lipids but not necessarily MetS. However, MCP-1 levels were considerably elevated in children with obesity and MetS compared with those with normal body weight. MCP-1 levels were also correlated with waist and hip circumferences, BMI, and DBP. Kim et al. found similar correlations with BMI and WC in young Koreans [21]. Resistin, a protein suspected to be related to obesity and IR, is reportedly increased in children with central obesity [22]. Herein, resistin was elevated in children with overweight/obesity, both with and without MetS; however, the only significant difference was between children with MetS and those with normal body weight. That no difference was found between the groups with obese/overweight with and without MetS suggests that plasma resistin may be a weak biochemical marker of metabolic dysfunction. This supports the notion that only a small proportion of variance in resistin can be explained by MetS-related factors. Although central obesity assessed with WC is considered a better marker of metabolic risk than high BMI in adults, pediatric results have been contradictory [23, 24]. Our subsample with overweight/obesity showed significant differences on various inflammatory markers, and those with abdominal obesity had higher CRP levels compared with those without. Findings were consistent for PAI-1 and resistin, but not MCP-1, suggesting that WC may be correlated with inflammation and metabolic risk regardless of MetS status. Other adipokines evaluated herein were adiponectin and leptin. While leptin acts primarily in the hypothalamus to control food intake, satiety, and energy expenditure, adiponectin is associated with reduced total body fat mass and promotes insulin sensitivity [25, 26]. Obesity and MetS are characterized by decreased serum adiponectin in parallel with increased concentrations of circulating leptin. Consequently, the Adipo/Lep ratio is associated with BMI and MetS status [26, 27]. The results herein show a negative correlation between Adipo/Lep ratio and BMI, meaning that Adipo/Lep ratio is significantly lower in children and adolescents with obesity, with or without MetS, compared with children with normal body weight. This biomarker decreases with increasing metabolic risk factors, which is why it has been proposed as predictive of MetS [25, 26]. Herein, adiponectin concentrations were lower in children with overweight/obesity with MetS compared with those without MetS, providing further evidence that adiponectin decreases in the presence of previously identified MetS parameters [28]. Dyslipidemia, particularly TC and TGL levels, has been described as an important risk factor for CVD, based on various indices [29, 30]. TC and TGL reflect the concentrations of the lipoproteins that transport them. HDL-C has antiatherogenic activity. Together, TC/HDL-C and TGL/HDL-C ratios may reflect the balance between these lipoproteins and could serve as a useful marker for cardiovascular risk. These relations have been evaluated in adults and children with and without obesity [31–33]. In an otherwise adult healthy Mexican sample, TGL/HDL-C ratio was associated with low insulin sensitivity and MetS, suggesting that it may serve as a reference index for MetS [32]. Herein, we found that TGL/HDL-C ratio was higher in children with overweight/obesity compared with children without overweight/obesity. When evaluating this in children with obesity with and without MetS, the association was stronger in the presence of MetS. This index also rose with increased numbers of fulfilled MetS criteria. In bivariate analysis and multivariate logistic regression, only HDL-C and TGL showed a significant correlation with MetS. This confirms that both biochemical markers are relevant in the pathophysiology of this syndrome, thus contributing to the usefulness of TC/HDL-C and TGL/HDL-C ratios for MetS screening in obese children, such as has previously been reported [34]. The TGL/HDL-C ratio was predictive of MetS with an AUC of 0.848 ($95\%$ confidence interval [CI]: 0.753–0.917). The optimal cutoff value was >2.390. Our TGL/HDL-C cutoff value for MetS identification was higher than the values of 1.25 reported for Chinese children with obesity [34] and 2.0 for Korean children with overweight [35]. These differences may be attributable to population-based ethnic and genetic variations. Herein, TGL/HDL-C ratio was correlated with BMI, WC, fasting glucose, and the inflammatory parameters CRP and PAI-1, suggesting its value for identifying MetS. Elevated TC/HDL-C has been associated with a proinflammatory state in adults and adolescents, as it is strongly related to elevated CRP levels [33]. Herein, TC/HDL-C was evaluated as an indicator of cardiovascular risk. Consistent with other studies, it was significantly correlated with BMI, hypertension, and systemic inflammatory parameters: CRP, resistin, and PAI-1. This evidence confirms TC/HDL-C as a significant low-grade inflammation parameter. As such, it can be used to predict cardiovascular risk, as it reflects an imbalance between cholesterol transported by atherogenic lipoproteins and protective lipoproteins. It is widely accepted that obesity induces lipid biochemistry alterations in the development of atherogenic dyslipidemia, a critical factor in cardiovascular events among adults. The presence of atherosclerotic plaques has been reported in autopsies of children as young as age two years in the Bogalusa Heart Study [36]. Early identification is a crucial step toward reducing related morbidity and mortality. Herein, metabolic risk factors like obesity, atherogenic dyslipidemia (high TGL, low HDL-C), and high blood pressure were the most common MetS parameters (Fig 1). RCP levels differed significantly between children with and without MetS, and were positively correlated with metabolic risk factors such as TGL/HDL-C and TC/HDL-C. It is known that childhood dyslipidemia can trigger low-grade inflammation even in the absence of obesity, since these parameters are increased even in children who without overweight. However, in children with obesity, these parameters are considerably increased in the presence of MetS. In sum, the main risk factors correlated with metabolic disease and cardiovascular risk were WC, hypertension, atherogenic dyslipidemia (elevated TC, low HDL-C), HDL-C, TGL and inflammatory parameters (CRP, PAI-1). Of note, HDL-C, TGL, TGL/HDL-C ratio and TC/HDL-C ratio are available from routine lab tests, simplifying surveillance for cardiovascular risk among children with overweight or obesity. To date, few studies in Mexico have evaluated the cardiovascular risk indices in children. Therefore, some limitations of the present study must be acknowledged. This study was based on a population of children and adolescents who attended an obesity consultation motivated by themselves or their parents to receive treatment, consequently, the results cannot be generalized since the sample consisted mainly of children of a medium-low socioeconomic level who were more predisposed to the development of metabolic disorders. Moreover, a weakness of this study was the comparison of our group of children with populations of different ethnic groups, making it difficult to compare the results obtained with those of other authors, especially considering the population differences and various methodologies used. Despite these limitations, we must highlight that Mexico is one of the first places in childhood obesity; therefore, this study is extremely important for the recognition of risk factors since childhood that influence the appearance of chronic-degenerative diseases in adulthood. In conclusion in the population evaluated HDL-C, TGL, TGL/HDL-C ratio and TC/HDL-C ratio shown major alteration in overweight and obese children with MetS, which can be explained because the lipid parameters are part of the MetS diagnostic criteria; however, the inflammatory parameters and adipokines evaluated did not shown a difference in overweight or obese children with/without MetS, confirming the chronic inflammation state that has been previously described in patients under these conditions. ## References 1. Peña-Espinoza BI, Granados-Silvestre MLÁ, Sánchez-Pozos K, Ortiz-López MG, Menjivar M. Metabolic syndrome in Mexican children: Low effectiveness of diagnostic definitions.. Endocrinol Diabetes Nutr (2017) 64 369-376. DOI: 10.1016/j.endinu.2017.04.004 2. Magge SN, Goodman E, Armstrong SC. Committee on Nutrition. The Metabolic Syndrome in Children and Adolescents: Shifting the Focus to Cardiometabolic Risk Factor Clustering. Pediatrics (2017) 140 e20171603. DOI: 10.1542/peds.2017-1603 3. Ávila-Curiel A, Galindo-Gómez C, Juárez-Martínez L, Osorio-Victoria ML. Síndrome metabólico en niños de 6 a 12 años con obesidad, en escuelas públicas de siete municipios del Estado de México. Salud Publica Mex (2018) 60 395-403. DOI: 10.21149/8470 4. Koyama S, Ichikawa G, Kojima M, Shimura N, Sairenchi T, Arisaka O. Adiposity rebound and the development of metabolic syndrome. Pediatrics (2014) 133 e114-9. DOI: 10.1542/peds.2013-0966 5. Daniels SR, Greer FR. Committee on Nutrition. Lipid screening and cardiovascular health in childhood. Pediatrics (2008) 122 198-208. DOI: 10.1542/peds.2008-1349 6. Stroescu RF, Mărginean O, Bizerea T, Gafencu M, Voicu A, Doroș G. Adiponectin, leptin and high sensitivity C-reactive protein values in obese children—important markers for metabolic syndrome. J Pediatr Endocrinol Metab (2019) 32 27-31. DOI: 10.1515/jpem-2018-0378 7. Srikanthan K, Feyh A, Visweshwar H, Shapiro JI, Sodhi K. Systematic Review of Metabolic Syndrome Biomarkers: A Panel for Early Detection, Management, and Risk Stratification in the West Virginian Population. Int J Med Sci (2016) 13 25-38. DOI: 10.7150/ijms.13800 8. Kumari R, Kumar S, Kant R. An update on metabolic syndrome: Metabolic risk markers and adipokines in the development of metabolic syndrome.. Diabetes Metab Syndr (2019) 13 2409-2417. DOI: 10.1016/j.dsx.2019.06.005 9. Frithioff-Bøjsøe C, Lund MAV, Lausten-Thomsen U, Hedley PL, Pedersen O, Christiansen M. Leptin, adiponectin, and their ratio as markers of insulin resistance and cardiometabolic risk in childhood obesity. Pediatr Diabetes (2020) 21 194-202. DOI: 10.1111/pedi.12964 10. Li G, Xu L, Zhao Y, Li L, Fu J, Zhang Q. Leptin-adiponectin imbalance as a marker of metabolic syndrome among Chinese children and adolescents: The BCAMS study. PLoS One (2017) 12 e0186222. DOI: 10.1371/journal.pone.0186222 11. Mărginean CO, Meliţ LE, Ghiga DV, Mărginean MO. Early Inflammatory Status Related to Pediatric Obesity. Front Pediatr. (2019). DOI: 10.3389/fped.2019.00241 12. Ferranti SD, Gauvreau K, Ludwig DS, Neufeld EJ, Newburger JW, Rifai N. Prevalence of the metabolic syndrome in American adolescents: findings from the Third National Health and Nutrition Examination Survey. Circulation (2004) 110 2494-7. DOI: 10.1161/01.CIR.0000145117.40114.C7 13. Twig G, Yaniv G, Levine H, Leiba A, Goldberger N, Derazne E. Body-Mass Index in 2.3 Million Adolescents and Cardiovascular Death in Adulthood. N Engl J Med (2016) 374 2430-40. DOI: 10.1056/NEJMoa1503840 14. Franks PW, Hanson RL, Knowler WC, Sievers ML, Bennett PH, Looker HC. Childhood obesity, other cardiovascular risk factors, and premature death. N Engl J Med (2010) 362 485-93. DOI: 10.1056/NEJMoa0904130 15. Bitew ZW, Alemu A, Ayele EG, Tenaw Z, Alebel A, Worku T. Metabolic syndrome among children and adolescents in low and middle income countries: a systematic review and meta-analysis. Diabetol Metab Syndr (2020) 12 93. DOI: 10.1186/s13098-020-00601-8 16. Rochlani Y, Pothineni NV, Kovelamudi S, Mehta JL. Metabolic syndrome: pathophysiology, management, and modulation by natural compounds. Ther Adv Cardiovasc Dis (2017) 11 215-225. DOI: 10.1177/1753944717711379 17. Kwaifa IK, Bahari H, Yong YK, Noor SM. Endothelial Dysfunction in Obesity-Induced Inflammation: Molecular Mechanisms and Clinical Implications.. Biomolecules (2020) 10 291. DOI: 10.3390/biom10020291 18. Esser N, Legrand-Poels S, Piette J, Scheen AJ, Paquot N. Inflammation as a link between obesity, metabolic syndrome and type 2 diabetes. Diabetes Res Clin Pract (2014) 105 141-50. DOI: 10.1016/j.diabres.2014.04.006 19. Alessi MC, Juhan-Vague I. PAI-1 and the metabolic syndrome: links, causes, and consequences. Arterioscler Thromb Vasc Biol (2006) 26 2200-7. DOI: 10.1161/01.ATV.0000242905.41404.68 20. Kodaman N, Aldrich MC, Sobota R, Asselbergs FW, Brown NJ, Moore JH. Plasminogen Activator Inhibitor-1 and Diagnosis of the Metabolic Syndrome in a West African Population.. J Am Heart Assoc (2016) 5 e003867. DOI: 10.1161/JAHA.116.003867 21. Kim CS, Park HS, Kawada T, Kim JH, Lim D, Hubbard NE. Circulating levels of MCP-1 and IL-8 are elevated in human obese subjects and associated with obesity-related parameters.. Int J Obes (Lond). (2006) 30 1347-55. DOI: 10.1038/sj.ijo.0803259 22. Li M, Fisette A, Zhao XY, Deng JY, Mi J, Cianflone K. Serum resistin correlates with central obesity but weakly with insulin resistance in Chinese children and adolescents. Int J Obes (Lond). (2009) 33 424-39. DOI: 10.1038/ijo.2009.44 23. Janssen I, Katzmarzyk PT, Ross R. Waist circumference and not body mass index explains obesity-related health risk. Am J Clin Nutr (2004) 79 379-84. DOI: 10.1093/ajcn/79.3.379 24. Anderson LN, Lebovic G, Hamilton J, Hanley AJ, McCrindle BW, Maguire JL. Body Mass Index, Waist Circumference, and the Clustering of Cardiometabolic Risk Factors in Early Childhood.. Paediatr Perinat Epidemiol (2016) 30 160-70. DOI: 10.1111/ppe.12268 25. Yosaee S, Khodadost M, Esteghamati A, Speakman JR, Djafarian K, Bitarafan V. Adiponectin: An Indicator for Metabolic Syndrome. Iran J Public Health (2019) 48 1106-1115. PMID: 31341853 26. Frühbeck G, Catalán V, Rodríguez A, Ramírez B, Becerril S, Salvador J. Adiponectin-leptin Ratio is a Functional Biomarker of Adipose Tissue Inflammation.. Nutrients (2019) 11 454. DOI: 10.3390/nu11020454 27. Frühbeck G, Catalán V, Rodríguez A, Gómez-Ambrosi J. Adiponectin-leptin ratio: A promising index to estimate adipose tissue dysfunction. Relation with obesity-associated cardiometabolic risk. Adipocyte (2018) 7 57-62. DOI: 10.1080/21623945.2017.1402151 28. Klünder-Klünder M, Flores-Huerta S, García-Macedo R, Peralta-Romero J, Cruz M. Adiponectin in eutrophic and obese children as a biomarker to predict metabolic syndrome and each of its components. BMC Public Health (2013) 13 88. DOI: 10.1186/1471-2458-13-88 29. Xie H, Min M, Guo S, Xian Y, Yang F, Wang X. Impact of Vitamin D and Vitamin D Receptor on Risk of Cardiovascular Diseases in Children and Adolescents with Obesity in Sichuan, China: A Cross-Sectional Study. Ann Nutr Metab (2020) 76 396-404. DOI: 10.1159/000513287 30. Millán J, Pintó X, Muñoz A, Zúñiga M, Rubiés-Prat J, Pallardo LF. Lipoprotein ratios: Physiological significance and clinical usefulness in cardiovascular prevention.. Vasc Health Risk Manag (2009) 5 757-65. PMID: 19774217 31. Strufaldi MW, Souza FI, Puccini RF, Franco Mdo C. Family history of cardiovascular disease and non-HDL cholesterol in prepubescent non-obese children. Rev Assoc Med Bras (2016) 62 347-52. DOI: 10.1590/1806-9282.62.04.347 32. Baez-Duarte BG, Zamora-Gínez I, González-Duarte R, Torres-Rasgado E, Ruiz-Vivanco G, Pérez-Fuentes R. Triglyceride/high-density lipoprotein cholesterol (TG/HDL-C) index as a reference criterion of risk for metabolic syndrome (MetS) and low insulin sensitivity in apparently healthy subjects. Gac Med Mex (2017) 153 152-158. PMID: 28474700 33. Agirbasli M, Tanrikulu A, Acar Sevim B, Azizy M, Bekiroglu N. Total cholesterol-to-high-density lipoprotein cholesterol ratio predicts high-sensitivity C-reactive protein levels in Turkish children. J Clin Lipidol (2015) 9 195-200. DOI: 10.1016/j.jacl.2014.12.010 34. Liang J, Fu J, Jiang Y, Dong G, Wang X, Wu W. TriGlycerides and high-density lipoprotein cholesterol ratio compared with homeostasis model assessment insulin resistance indexes in screening for metabolic syndrome in the chinese obese children: a cross section study. BMC Pediatr (2015) 15 138. DOI: 10.1186/s12887-015-0456-y 35. Yoo DY, Kang YS, Kwon EB, Yoo EG. The triglyceride-to-high density lipoprotein cholesterol ratio in overweight Korean children and adolescents. Ann Pediatr Endocrinol Metab (2017) 22 158-163. DOI: 10.6065/apem.2017.22.3.158 36. Newman WP, Freedman DS, Voors AW, Gard PD, Srinivasan SR, Cresanta JL. Relation of serum lipoprotein levels and systolic blood pressure to early atherosclerosis. The Bogalusa Heart Study. N Engl J Med (1986) 314 138-44. DOI: 10.1056/NEJM198601163140302
--- title: Relative importance of potential risk factors for dementia in patients with hypertension authors: - Mi-Hyang Jung - Kwang-Il Kim - Jun Hyeok Lee - Ki-Chul Sung journal: PLOS ONE year: 2023 pmcid: PMC10016665 doi: 10.1371/journal.pone.0281532 license: CC BY 4.0 --- # Relative importance of potential risk factors for dementia in patients with hypertension ## Abstract Patients with hypertension are at higher risk for dementia than the general population. We sought to understand the relative importance of various risk factors in the development of dementia among patients with hypertension. This population-based cohort study used data from the Korean National Insurance Service database. Using the Cox proportional hazard model, R2 values for each potential risk factor were calculated to test the relative importance of risk factors for the development of dementia. Eligible individuals were adults 40 to 79 years of age with hypertension and without a history of stroke and dementia between 2007 and 2009. A total of 650,476 individuals (mean age, 60 ± 11 years) with hypertension were included in the analyses. During a mean follow-up of 9.5 years (±2.8 years), 57,112 cases of dementia were observed. The three strongest predictors of dementia were age, comorbidity burden (assessed using the Charlson Comorbidity Index), and female sex (R2 values, 0.0504, 0.0023, and 0.0022, respectively). The next strongest risk factors were physical inactivity, smoking, alcohol consumption, and obesity (R2 values, 0.00070, 0.00024, 0.00021, and 0.00020, respectively). Across all age groups, physical inactivity was an important risk factor for dementia occurrence. In summary, controlling and preventing comorbidities are of utmost importance to prevent dementia in patients with hypertension. More efforts should be taken to encourage physical activity among patients with hypertension across all age groups. Furthermore, smoking cessation, avoiding and limiting alcohol consumption, and maintaining an appropriate body weight are urged to prevent dementia. ## Introduction Dementia is a clinical syndrome that leads to the deterioration of cognitive function beyond the usual consequences of aging, ultimately limiting the ability to perform daily activities. Currently, more than 55 million individuals globally have dementia, and this number is expected to increase to 78 million by 2030 [1]. Dementia is a major cause of disability, dependency, and death for older adults [2]. In Korea, dementia is also expected to become an important health issue since Korea is one of the most rapidly aging countries worldwide. The prevalence of dementia in Korea is estimated to be $10.3\%$ among those over 65 years in 2020 and $15.9\%$ by 2050 (3.02 million people) [3]. Patients with hypertension are at higher risk for incident dementia than the general population [4]. This is partly attributable to the fact that both hypertension and dementia are the effects of aging, and they share common risk factors [5]. Furthermore, recent studies have shown that mid-life hypertension is associated with incident dementia [6–9]. Intriguingly, one study showed that the pattern of association between blood pressure (BP) and dementia development was different depending on whether antihypertensive drugs were taken or not [9]. Because of the considerable disease burden and lack of effective therapy for dementia, identifying and prioritizing the risk factors for dementia are essential components of the primary prevention of hypertension, which is still a growing chronic disease burden in Korea that affected 10.1 million people as of 2019 [10]. A previous study reported that approximately one-third of Alzheimer’s disease cases might be attributable to potentially modifiable risk factors, such as physical inactivity, smoking, hypertension, obesity, and diabetes [9]. Although age is usually regarded as the most powerful but non-modifiable risk factor, dementia is neither an inevitable consequence of aging nor a disease exclusive to older adults [2]. The onset of dementia at a young age (younger than 65 years) has recently received more attention as another important but underappreciated disease entity [11,12]. Therefore, we sought to understand the relative importance of various risk factors for dementia for patients with hypertension, who are at higher risk for developing incident dementia. Additionally, we investigated the age-specific relevance of these risk factors. We anticipate that the data acquired during this study could serve as basic epidemiologic data for establishing effective public health policies and applying individualized (subgroup-specific) preventive strategies. ## Data source We extracted data from the National Health Insurance Service (NHIS) National Health examinee database. The NHIS is the only insurance provider in Korea, covering $97\%$ of the population. National health screening (blood pressure and anthropometric measurements, questionnaires about health status, laboratory tests, and chest X-ray) was conducted every 2 years for the Korean population 40 years or older. The NHIS National Health examinee database includes information about demographic characteristics, diagnoses based on the International Classification of Diseases, tenth revision (ICD-10) codes, admissions, prescriptions, health screening data, and death. Detailed information about the NHIS database can be found elsewhere [13,14]. ## Study population Using the NHIS National Health examinee cohort, we identified 1,406,333 individuals (age 40–79 years) with hypertension between 2007 and 2009 (index period). Among these, we excluded 83,135 subjects with a history of stroke or dementia at baseline using the 2002 to 2006 database information. We further excluded those who did not undergo a health screening examination during the index period ($$n = 672$$,667) and those with missing data ($$n = 55$$). Ultimately, 650,476 individuals comprised the final study population and were included in the analysis. The study protocol was approved by the institutional review board of Kangbuk Samsung Hospital (KBSMC 2021-08-022). The review board waived the requirement for informed consent because anonymized and de-identified information was provided to the researchers under the strict confidentiality protocol of the NHIS. ## Data collection Trained staff members measured BP at least twice, with the patient in the sitting position after ≥5 min of rest. The body mass index (BMI) was calculated as weight divided by the height squared (kg/m2). Information regarding physical activity, smoking history, and alcohol consumption was collected using self-report questionnaires. Blood samples (fasting glucose and lipid profiles) were collected after ≥8 h of fasting. Income level was classified according to an individual’s insurance premium status. The Charlson comorbidity index was calculated to explore the role of comorbidities in the development of dementia [15]. Hypertension was defined as the presence of a claim for the prescription of antihypertensive medications with relevant ICD-10 codes for hypertension (I10-I13 and I15). Diabetes was defined as the prescription of antidiabetic medications with ICD-10 codes for diabetes (E08-E13). Furthermore, stroke was defined by ICD-10 codes for stroke (I60-64) with hospitalization for ≥2 days. ## Study outcomes and follow-up The primary study outcome was a new diagnosis of dementia. Dementia was defined when the following two conditions were simultaneously satisfied: corresponding ICD-10 codes for dementia (F00 or G30 for Alzheimer’s disease; F01 for vascular dementia; and F02, F03, or G31 for other dementia) and the prescription of anti-dementia medications (rivastigmine, galantamine, memantine, or donepezil). The accuracy of the codes for dementia according to the NHIS data has been previously tested using the Mini Mental State Examination; the positive predictive value was $94.7\%$ [16]. The study population was followed-up until the development of dementia, death, or the end of the study (December 2017). ## Statistical analysis To explore the relative importance of the risk factors in the development of dementia, we calculated the estimated explained relative risk (R2) using the Cox proportional hazards model. R2 values provide an estimate of how important each risk factor is for predicting the outcome [17]. The estimated explained relative risk model has been previously tested during large cohort studies [17–19]. Potential risk factors are as follows: age (40–59, 60–69, and 70–79 years); sex; comorbidity burden (Charlson comorbidity index ≥75th percentile of the study population or <75th percentile of the study); physical inactivity (≥1 time/week or never); smoking habit (current smoker or others); alcohol consumption (≥10 g/day or <10 g/day); obesity (BMI ≥27.5 kg/m2 or <27.5 kg/m2); use of statins; elevated fasting glucose level (≥140 mg/dL or <140 mg/dL); BP control status (systolic BP ≥130 mmHg or <130 mmHg); hypertension duration (≥5 years or <5 years); elevated total cholesterol level (≥200 mg/dL or <200 mg/dL); and low-income level (lowest quartile or others). Among many drugs, we sought to identify the role of statin use on the development of dementia for two reasons. First, statin use might reflect the atherosclerotic burden of vessels, which might affect vascular dementia. Second, apolipoprotein E (APOE) gene polymorphism might be related to Alzheimer’s disease and lipid profile [20]. Thus, we identified the role of statin in the occurrence of dementia of Alzheimer’s or vascular-type dementia. The analyses were applied to the overall hypertensive population and age subgroups. All analyses were performed using SAS version 9.4 (SAS Institute, Inc., Cary, NC, USA). ## Baseline characteristics A total of 650,476 individuals with hypertension (mean age, 60.2 ± 11.2 years; men, $48.9\%$) were included in the analyses. The mean systolic BP and diastolic BP were 134.1 ± 17.0 and 82.1 ± 11.0 mmHg, respectively; $50.9\%$ of the population had hypertension for a duration ≥5 years. A total of $23.1\%$ of the population had diabetes, and $51.5\%$, $28.6\%$, and $19.9\%$ of the population had BMI <25, 25–27.4, and ≥27.5 kg/m2, respectively. The other baseline characteristics are listed in the S1 Table. During a mean of 9.5 ± 2.8 years of follow-up, 57,112 cases of dementia developed. ## Relative importance of risk factors for dementia in the overall population Age showed the highest R2 value (0.0504) for dementia (Fig 1). After age, comorbidity burden and female sex (R2 values, 0.0023 and 0.0022, respectively) had the next highest R2 values for dementia. These were followed by physical inactivity, smoking, and alcohol consumption, which are lifestyle factors (R2 values, 0.00070, 0.00024, and 0.00021, respectively). Obesity, statin use, elevated fasting glucose levels, and BP control status were followed as other risk factors for dementia. Fig 1: The relative importance of risk factors for the development of dementia among patients with hypertension. ## Subgroup analyses by age Across all three age groups, physical inactivity was included among the three most important risk factors, with the highest R2 value for dementia (Fig 2). The three most important risk factors for middle-aged adults (age 40–59 years) were comorbidity, obesity, and physical inactivity. The three most important risk factors for older adults (age 60–69 years) were obesity, physical inactivity, and alcohol consumption. Finally, for the oldest adults (age 70–79 years), the three most important risk factors were physical inactivity, alcohol consumption, and obesity. Fig 2: Age-specific relative importance of risk factors for the development of dementia among patients with hypertension: Subgroup analyses by age. ## Discussion In this nationwide cohort of 650,476 individuals with hypertension, we found that age, female sex, and comorbidity burden are fundamental risk factors for the development of dementia. Physical inactivity, smoking, alcohol consumption, and obesity are also important risk factors for dementia. Among these lifestyle-related factors, physical inactivity is the most powerful risk factor for dementia. Our results revealed possible targets for prevention and their relative importance for patients with hypertension who are at higher risk for dementia. Furthermore, to the best of our knowledge, our study is the first to determine the relative importance of various risk factors in the occurrence of dementia among patients with treated hypertension. The prevalence of dementia is expected to increase in parallel with the rapidly aging population. As demonstrated during the current study, age is the most important risk factor for dementia and is regarded as an unmodifiable risk factor. However, it should be pointed out that aging is not an inevitable consequence of age [21]. Recently, abundant data suggested that biological age is more important than chronological age. It is crucial to focus more attention on lifestyle and environmental risk factors that could affect biological age, such as comorbidity burden, physical activity, current smoking, obesity (BMI ≥27.5 kg/m2), and alcohol consumption (≥10 g/day), which were all high-ranking risk factors for dementia during the current analyses. Among various lifestyle factors, physical inactivity was the most powerful risk factor for dementia in this study. Evidence supports that aerobic exercise may attenuate cognitive impairment and reduce dementia risk [22–24]. Exercise and regular physical activity may benefit cognitive function by regulating amyloid ß turnover, reducing inflammation, promoting neurogenesis through neurotrophin production, and improving cerebral blood flow [22–24]. It is important to note that the salutary effects of regular physical activity and exercise apply to all ages, including older adults. Several randomized clinical trials have demonstrated better cortical connectivity and larger hippocampal volume in older adults after aerobic exercise [25,26]. Moreover, the benefit of physical activity could be applied to cognitive function and vascular health, which might be particularly essential in patients with hypertension. The brain of a hypertensive patient shows increased microvascular rarefaction, decreased cerebral flow, and disrupted brain-blood barrier integrity. However, physical activity could ameliorate or reverse those changes by increasing vascular endothelial growth factor or enhancing brain-derived natriuretic factor (a molecule related to neuronal survival and synapse formation) [27]. A meta-analysis involving 19 prospective studies found that for older adults (mean age, 74 years), current smoking was associated with a $79\%$ increased risk of Alzheimer’s disease, $78\%$ increased risk of vascular dementia, and $27\%$ increased risk of any type of dementia compared to never smoking [28]. This study clearly showed that individuals (including older adults) should quit smoking to prevent dementia. In terms of alcohol consumption, most data support the risk of heavy alcohol consumption for patients with dementia. However, there is some controversy regarding the amount of alcohol allowed [29–31]. In the current study, alcohol consumption of any amount (≥10 g/day [1 unit/day]) was a risk factor, particularly for older hypertensive patients (age 60 years or older). Previous meta-analyses demonstrated that even low amounts of alcohol consumption are associated with the risk of hypertension for Asian men [32]. Avoiding and limiting alcohol consumption, especially for Asian individuals, might improve cognitive health. However, additional studies are warranted. Regarding obesity, there is a discrepancy regarding appropriate BMI levels, often referred to as the obesity paradox (lower incidence of dementia among obese individuals or lower mortality among obese individuals with dementia) [33–35]. To accurately assess this paradox, separate evaluations before and after the onset of dementia are warranted. When the scope is limited to risk factors for dementia development, obesity appears to be a risk factor for dementia, especially for middle-aged and older adults. Some studies have shown that obesity prevents the development of dementia among the oldest adults [34]. However, several studies have demonstrated that obesity increases the risk of future dementia in the long term (>10 years from baseline) and has a paradoxical association in the short term [35]. Weight loss often precedes the onset of dementia by 10 years (preclinical stage of dementia). This may be a confounding factor, especially for the oldest adults with a shorter follow-up period. In the current study, obesity (BMI ≥27.5 kg/m2) was a significant risk factor for dementia development across all age groups, which could serve as evidence supporting weight control. However, this needs to be cautiously interpreted because we did not evaluate the risk of being underweight. In the current analyses, BP control status or hypertension duration was not a high-ranking risk factor, except in younger adults. This might partly be attributable to the fact that our cohort comprised treated hypertensive patients who were on antihypertensive medications, and their BP was relatively well controlled (the prevalence of individuals with systolic BP < 130 mmHg was almost half). A prior study has shown that the relationship between BP and the risk of dementia might differ based on taking antihypertensive medications [7]. Other BP-related parameters (e.g., BP variability or white-coat phenomenon) might be more important in incident dementia [36]. Age-specific differences in the priority of risk factors also need to be discussed. Generally, the overall trend (i.e., physical inactivity and obesity as important risk factors) was similar across all age groups. One of the intriguing findings in the current study was the relative importance of comorbidity or hypertension duration in the younger adults treated with hypertension. Dementia in older adults (age 75 years or greater) is mainly because of Alzheimer’s disease, which originates from the long-term accumulation of amyloid deposits. Epigenetic factors, such as physical activity or alcohol consumption, can play a more important role in modifying the genetics of Alzheimer’s disease rather than hypertension duration in older adults [37]. Conversely, dementia in young and middle-aged adults (40–74 years) is mainly because of genetic factors (early-onset Alzheimer’s disease, frontotemporal dementia) or vascular factors (vascular dementia). Indeed, comorbidity burden and hypertension duration, which might be associated with vascular dementia, were identified as important risk factors in this study in younger adults aged 40–59. Similar to our results, Jung et al. previously reported that increased BP was associated with an increased risk of dementia in younger individuals (< 70 years old) but not in an older population (>70 years) [8]. Further studies are needed to clarify the age-specific etiologies in developing dementia. Several strengths of the current study are as follows. First, the potential risk factors and their relative importance for dementia were evaluated among a large population of hypertensive patients treated with antihypertensive medication. To the best of our knowledge, this is the first study that explicitly prioritized and determined the relative importance of various risk factors for the occurrence of dementia, an important public health issue. This study also had some limitations that should be discussed. First, causal inferences might have been limited because our study was based on a retrospective cohort study. Second, we lack information on a comparator group of an age and sex-matched general population without hypertension. Third, there were no data on other important risk factors for dementia, such as family history, genetic factors, and medication histories other than statin use, which might have resulted in the extremely low R2 values. Fourth, we did not evaluate the influence of each component of comorbidity on the occurrence of dementia, which is also a relevant study theme that warrants additional studies. Fifth, we defined dementias, which incorporate Alzheimer’s disease, vascular dementia, and other forms of dementia. Therefore, this study could not differentiate the importance of various risk factors for each type of dementia. In this context, our results should be cautiously interpreted in the context of Korean patients with hypertension generated based on the data collected herein. ## Conclusions Age and sex are fundamental risk factors for the development of dementia in patients with hypertension. Furthermore, comorbidity burden, physical inactivity, smoking, alcohol consumption, and obesity are other important risk factors for dementia. Therefore, controlling and preventing comorbidities are of utmost importance in preventing dementia in patients with hypertension. More efforts should be taken to encourage physical activity among patients with hypertension across all age groups. Furthermore, smoking cessation, avoidance of and limiting alcohol consumption, and maintaining an appropriate body weight are urged for the primary prevention of dementia. ## References 1. 1World Health Organization. 2021 [cited 28 December 2022]. In: Fact sheets of dementia [Internet]. Geneva, Switzerland: World Health Organization. https://www.who.int/news-room/fact-sheets/detail/dementia.. Fact sheets of dementia (2021) 2. Livingston G, Huntley J, Sommerlad A, Ames D, Ballard C, Banerjee S. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. Lancet (2020) 396 413-446. DOI: 10.1016/S0140-6736(20)30367-6 3. Shon C, Yoon H. Health-economic burden of dementia in South Korea. BMC Geriatr (2021) 21 549. DOI: 10.1186/s12877-021-02526-x 4. Iadecola C, Gottesman RF. Neurovascular and cognitive dysfunction in hypertension. Circ Res (2019) 124 1025-1044. DOI: 10.1161/CIRCRESAHA.118.313260 5. Novak V, Hajjar I. The relationship between blood pressure and cognitive function. Nat Rev Cardiol (2010) 7 686-698. DOI: 10.1038/nrcardio.2010.161 6. Gottesman RF, Albert MS, Alonso A, Coker LH, Coresh J, Davis SM. Associations between midlife vascular risk factors and 25-year incident dementia in the atherosclerosis risk in communities (ARIC) cohort. JAMA Neurol (2017) 74 1246-1254. DOI: 10.1001/jamaneurol.2017.1658 7. Lee CJ, Lee JY, Han K, Kim DH, Cho H, Kim KJ. Blood Pressure levels and risks of dementia: a nationwide study of 4.5 million people. Hypertension (2022) 79 218-229. DOI: 10.1161/HYPERTENSIONAHA.121.17283 8. Jung H, Yang PS, Kim D, Jang E, Yu HT, Kim TH. Associations of hypertension burden on subsequent dementia: a population-based cohort study. Sci Rep (2021) 11 12291. DOI: 10.1038/s41598-021-91923-8 9. Norton S, Matthews FE, Barnes DE, Yaffe K, Brayne C. Potential for primary prevention of Alzheimer’s disease: an analysis of population-based data. Lancet Neurol (2014) 13 788-794. DOI: 10.1016/S1474-4422(14)70136-X 10. Kim HC, Lee H, Lee HH, Seo E, Kim E, Han J. Korea hypertension fact sheet 2021: analysis of nationwide population-based data with special focus on hypertension in women. Clin Hypertens (2022) 28 1. DOI: 10.1186/s40885-021-00188-w 11. Hendriks S, Peetoom K, Bakker C, van der Flier WM, Papma JM, Koopmans R. Global prevalence of young-onset dementia: A systematic review and meta-analysis. JAMA Neurol (2021) 78 1080-1090. DOI: 10.1001/jamaneurol.2021.2161 12. Ikejima C, Yasuno F, Mizukami K, Sasaki M, Tanimukai S, Asada T. Prevalence and causes of early-onset dementia in Japan: a population-based study. Stroke (2009) 40 2709-2714. DOI: 10.1161/STROKEAHA.108.542308 13. Seong SC, Kim YY, Park SK, Khang YH, Kim HC, Park JH. Cohort profile: the National Health Insurance Service-National Health Screening Cohort (NHIS-HEALS) in Korea. BMJ Open (2017) 7 e016640. DOI: 10.1136/bmjopen-2017-016640 14. Shin JH, Jung MH, Kwon CH, Lee CJ, Kim DH, Kim HL. Disparities in mortality and cardiovascular events by income and blood pressure levels among patients with hypertension in South Korea. J Am Heart Assoc (2021) 10 e018446. DOI: 10.1161/JAHA.120.018446 15. Quan H, Li B, Couris CM, Fushimi K, Graham P, Hider P. Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. Am J Epidemiol (2011) 173 676-682. DOI: 10.1093/aje/kwq433 16. Kim D, Yang PS, Yu HT, Kim TH, Jang E, Sung JH. Risk of dementia in stroke-free patients diagnosed with atrial fibrillation: data from a population-based cohort. Eur Heart J (2019) 40 2313-2323. DOI: 10.1093/eurheartj/ehz386 17. Heller G. A measure of explained risk in the proportional hazards model. Biostatistics (2012) 13 315-325. DOI: 10.1093/biostatistics/kxr047 18. Rawshani A, Rawshani A, Sattar N, Franzén S, McGuire DK, Eliasson B. Relative prognostic importance and optimal levels of risk factors for mortality and cardiovascular outcomes in type 1 diabetes mellitus. Circulation (2019) 139 1900-1912. DOI: 10.1161/CIRCULATIONAHA.118.037454 19. Rawshani A, Rawshani A, Franzén S, Sattar N, Eliasson B, Svensson AM. Risk factors, mortality, and cardiovascular outcomes in patients with Type 2 diabetes. N Engl J Med (2018) 379 633-644. DOI: 10.1056/NEJMoa1800256 20. Martínez-Magaña JJ, Genis-Mendoza AD, Tovilla-Zarate CA, González-Castro TB, Juárez-Rojop IE, Hernández-Díaz Y. Association between APOE polymorphisms and lipid profile in Mexican Amerindian population. Mol Genet Genomic Med (2019) 7 e958. DOI: 10.1002/mgg3.958 21. Hamczyk MR, Nevado RM, Barettino A, Fuster V, Andrés V. Biological versus chronological aging: JACC focus seminar. J Am Coll Cardiol (2020) 75 919-930. DOI: 10.1016/j.jacc.2019.11.062 22. Valenzuela PL, Castillo-García A, Morales JS, de la Villa P, Hampel H, Emanuele E. Exercise benefits on Alzheimer’s disease: state-of-the-science. Ageing Res Rev (2020) 62 101108. DOI: 10.1016/j.arr.2020.101108 23. De la Rosa A, Olaso-Gonzalez G, Arc-Chagnaud C, Millan F, Salvador-Pascual A, García-Lucerga C. Physical exercise in the prevention and treatment of Alzheimer’s disease. J Sport Health Sci (2020) 9 394-404. DOI: 10.1016/j.jshs.2020.01.004 24. Ahlskog JE, Geda YE, Graff-Radford NR, Petersen RC. Physical exercise as a preventive or disease-modifying treatment of dementia and brain aging. Mayo Clin Proc (2011) 86 876-884. DOI: 10.4065/mcp.2011.0252 25. Colcombe SJ, Kramer AF, Erickson KI, Scalf P, McAuley E, Cohen NJ. Cardiovascular fitness, cortical plasticity, and aging. Proc Natl Acad Sci U S A (2004) 101 3316-3321. DOI: 10.1073/pnas.0400266101 26. Ruscheweyh R, Willemer C, Krüger K, Duning T, Warnecke T, Sommer J. Physical activity and memory functions: an interventional study. Neurobiol Aging (2011) 32 1304-1319. DOI: 10.1016/j.neurobiolaging.2009.08.001 27. Rêgo ML, Cabral DA, Costa EC, Fontes EB. Physical exercise for individuals with hypertension: it is time to emphasize its benefits on the brain and cognition. Clin Med Insights Cardiol (2019) 13 1179546819839411. DOI: 10.1177/1179546819839411 28. Anstey KJ, von Sanden C, Salim A, O’Kearney R. Smoking as a risk factor for dementia and cognitive decline: a meta-analysis of prospective studies. Am J Epidemiol (2007) 166 367-378. DOI: 10.1093/aje/kwm116 29. Cervilla JA, Prince M, Joels S, Lovestone S, Mann A. Long-term predictors of cognitive outcome in a cohort of older people with hypertension. Br J Psychiatry (2000) 177 66-71. DOI: 10.1192/bjp.177.1.66 30. Chosy EJ, Edland S, Launer L, White LR. Midlife alcohol consumption and later life cognitive impairment: light drinking is not protective and APOE genotype does not change this relationship. PLOS ONE (2022) 17 e0264575. DOI: 10.1371/journal.pone.0264575 31. Ruitenberg A, van Swieten JC, Witteman JC, Mehta KM, van Duijn CM, Hofman A. Alcohol consumption and risk of dementia: the Rotterdam Study. Lancet (2002) 359 281-286. DOI: 10.1016/S0140-6736(02)07493-7 32. Jung MH, Shin ES, Ihm SH, Jung JG, Lee HY, Kim CH. The effect of alcohol dose on the development of hypertension in Asian and Western men: systematic review and meta-analysis. Korean J Intern Med (2020) 35 906-916. DOI: 10.3904/kjim.2019.016 33. Ma Y, Ajnakina O, Steptoe A, Cadar D. Higher risk of dementia in English older individuals who are overweight or obese. Int J Epidemiol (2020) 49 1353-1365. DOI: 10.1093/ije/dyaa099 34. Fitzpatrick AL, Kuller LH, Lopez OL, Diehr P, O’Meara ES, Longstreth WT. Midlife and late-life obesity and the risk of dementia: cardiovascular health study. Arch Neurol (2009) 66 336-342. DOI: 10.1001/archneurol.2008.582 35. Bowman K, Thambisetty M, Kuchel GA, Ferrucci L, Melzer D. Obesity and longer term risks of dementia in 65–74 year olds. Age Ageing (2019) 48 367-373. DOI: 10.1093/ageing/afz002 36. Yoo JE, Shin DW, Han K, Kim D, Lee SP, Jeong SM. Blood pressure variability and the risk of dementia: A nationwide cohort study. Hypertension (2020) 75 982-990. DOI: 10.1161/HYPERTENSIONAHA.119.14033 37. Reitz C, Rogaeva E, Beecham GW. Late-onset vs nonmendelian early-onset Alzheimer disease: A distinction without a difference?. Neurol Genet (2020) 6 e512. DOI: 10.1212/NXG.0000000000000512
--- title: 'Changes in clinical markers observed from pharmacist-managed cardiovascular risk reduction clinics in federally qualified health centers: A retrospective cohort study' authors: - Jasmine D. Gonzalvo - Ashley H. Meredith - Sonak D. Pastakia - Michael Peters - Madilyn Eberle - Andrew N. Schmelz - Lauren Pence - Jessica S. Triboletti - Todd A. Walroth journal: PLOS ONE year: 2023 pmcid: PMC10016666 doi: 10.1371/journal.pone.0282940 license: CC BY 4.0 --- # Changes in clinical markers observed from pharmacist-managed cardiovascular risk reduction clinics in federally qualified health centers: A retrospective cohort study ## Abstract ### Background Reductions in hemoglobin A1c (HbA1C) have been associated with improved cardiovascular outcomes and savings in medical expenditures. One public health approach has involved pharmacists within primary care settings. The objective was to assess change in HbA1C from baseline after 3–5 months of follow up in pharmacist-managed cardiovascular risk reduction (CVRR) clinics. ### Methods This retrospective cohort chart review occurred in eight pharmacist-managed CVRR federally qualified health clinics (FQHC) in Indiana, United States. Data were collected from patients seen by a CVRR pharmacist within the timeframe of January 1, 2015 through February 28, 2020. Data collected include: demographic characteristics and clinical markers between baseline and follow-up. HbA1C from baseline after 3 to 5 months was assessed with pared t-tests analysis. Other clinical variables were assessed and additional analysis were performed at 6–8 months. Additional results are reported between 9 months and 36 months of follow up. ### Results The primary outcome evaluation included 445 patients. Over 36 months of evaluation, 3,803 encounters were described. Compared to baseline, HbA1C was reduced by $1.6\%$ ($95\%$CI -1.8, -1.4, $p \leq 0.01$) after 3–5 months of CVRR care. Reductions in HbA1C persisted at 6–8 months with a reduction of $1.8\%$ ([$95\%$CI -2.0, -1.5] $p \leq 0.01$). The follow-up losses were $29.5\%$ at 3–5 months and $93.2\%$ at 33–36 months. ### Conclusions Our study augments the existing literature by demonstrating the health improvement of pharmacist-managed CVRR clinics. The great proportion of loss to follow-up is a limitation of this study to be considered. Additional studies exploring the expansion of similar models may amplify the public health impact of pharmacist-managed CVRR services in primary care sites. ## Introduction Globally, the number one cause of death is cardiovascular disease, which includes coronary artery disease, cerebrovascular disease, rheumatic heart disease, and additional conditions of the heart and blood vessels [1, 2]. In the United States alone, one in four deaths are attributed to heart disease [2]. From 2014 to 2015, healthcare expenditures related to heart disease cost the United States approximately $219 billion [2]. Cardiovascular diseases create a large financial, emotional, and physical burden for patients, health care workers, and systems. Risk factors, including medical conditions and lifestyle choices, have been identified to prevent the risk of developing cardiovascular disease, as studies have established a strong association between elevated blood glucose and cardiovascular complications [3]. By reducing the hemoglobin A1C (HbA1C) by one percentage point, epidemiologic research demonstrated an $18\%$ reduction in combined fatal and nonfatal myocardial infarction [3]. Furthermore, each percentage point reduction in HbA1C (e.g., from $10\%$ to $9\%$) results in an estimated savings associated with medical expenditures of $685-$950 per patient, per year [4]. One approach to addressing the high burden of chronic diseases to public health, including diabetes, has been to involve pharmacists in the care of patients within primary care settings. Pharmacists have become an integral component of cardiovascular disease state management in many Federally Qualified Health Centers (FQHC) [5–7]. FQHCs are primary care clinics that receive federal funding to provide healthcare to underserved communities [8]. Similar health centers provide care to almost $10\%$ of Americans [9]. Common responsibilities of the pharmacist within the FQHC may include medication therapy management, transitions of care, cost containment, and medication access. Pharmacist-managed cardiovascular risk reduction (CVRR) clinics are a key factor in reducing health care costs and cardiovascular disease burden. These clinics address chronic disease states such as hypertension, diabetes, dyslipidemia, and smoking cessation [10–12]. Some evidence has established the impact of pharmacists in improving several cardiovascular clinical markers, such as reductions in HbA1C levels, systolic and diastolic blood pressure, and lipid levels [10–12]. These studies typically encompass six months to two years and do not focus on populations who are underserved. The result is a lack of literature surrounding the long-term cardiovascular and cost-effective benefits of pharmacist-management of cardiovascular-related diseases in populations who are under resourced. Evidence of sustained benefit from CVRR services would substantially enhance health outcomes and primary care service delivery. The primary objective of this study was to assess the change in HbA1C from baseline after 3 to 5 months of follow up in the pharmacist-managed CVRR clinics. Secondary objectives included assessing the change in systolic blood pressure (SBP), diastolic blood pressure (DBP), low-density lipoprotein (LDL) cholesterol, and non-high-density lipoprotein (non-HDL) cholesterol at 3 to 5 months. Additional secondary objectives included demonstrating the persistence of changes at 6–8 for the same clinical markers. Outcome measurements captured between 9 months and 36 months of follow-up were also described. ## Methods This retrospective cohort chart review of the electronic medical record was deemed exempt by the Indiana University IRB, and informed written or verbal consent was not required. The review included eight pharmacist-managed CVRR clinics that are part of the county health system of a metropolitan area in Indiana, United States. This safety-net health system provides care to a primarily urban population of patients who are publicly insured, underserved, underinsured, and/or uninsured through a network of FQHCs. Pharmacists managing the CVRR clinics work under a collaborative practice agreement (CPA) for medication management of diabetes, hypertension, hyperlipidemia, and tobacco use. The first CVRR clinic was implemented in 2007, with seven CVRR clinics currently operating. An eighth CVRR clinic operated for two years before the pharmacist was moved to another clinic. A CPA is a legal document between a prescriber and a pharmacist, which grants additional privileges to the pharmacist [13]. The CPA is based on current clinical disease management guidelines (American Diabetes Association’s Standards of Medical Care in Diabetes) and is reviewed and updated annually. A suggested treatment algorithm is included; however, the pharmacist may make clinical decisions based on the individual needs of the patient. The CPA allows CVRR pharmacists to initiate, change, or discontinue medications for all relevant conditions. Commonly used medication classes include: diabetes–metformin, sodium glucose cotransporter– 2 inhibitors, glucagon-like peptide-1 agonists, dipeptidyl peptidase 4 inhibitors, and insulin; hypertension–angiotensin converting enzyme inhibitors, diuretics, and calcium channel blockers; dyslipidemia–statins; and other medication classes to manage cardiovascular risk. Patients are referred to the CVRR clinic if they are currently seen or have been seen at the FQHC in the past three years, an FQHC clinician refers them to CVRR services, or they have participated in group diabetes self-management education and support (DSMES) at the FQHC. The CVRR clinic does not require specific clinical criteria for eligible patients. Typically, CVRR patients will initially have an elevated HbA1C and may also have uncontrolled blood pressure and an elevated atherosclerotic cardiovascular disease (ASCVD) risk or lipid abnormalities. CVRR pharmacists usually meet with patients for a one-on-one, 30-minute appointment every four to six weeks. During this appointment, pharmacists review pertinent disease related lab values and monitoring parameters, provide education on disease management strategies, adjust cardiovascular pharmacologic therapies, and address barriers to disease state management optimization (i.e. prior authorization facilitation, referrals to other services, and other care coordination responsibilities). The frequency may be more or less often based on the needs of the patient. Throughout the time of working with the CVRR pharmacist, the patient will continue to meet with their primary care provider (PCP) for management of all medical conditions, as deemed necessary by the PCP. A patient will be discharged from the CVRR clinic once they maintain their individualized clinical goals (e.g.; HbA1C ≤ $7\%$) for three to six months, or if they no-show three scheduled appointments in a row despite attempts to contact or re-schedule them. Data from eight CVRR clinics were included. The retrospective data were collected within the timeframe of January 1, 2015 through February 28, 2020. The process for CVRR data and outcome collection became standardized through the use of REDCap, a secure web-application database, across clinic sites on January 1, 2015. To allow for variability in time to first follow-up HbA1C after the initial CVRR visit, a window of three to five months was defined for primary outcome evaluation. Inclusion dates for each site were based on the date the clinical pharmacy service was established and continued through six months prior to either the study end date (February 28, 2020) or six months prior to the clinical pharmacy service termination, to allow for sufficient time to appropriately capture data related to the primary outcome. Given the transition to telehealth-based care in late March 2020, REDCap measures were no longer consistently captured. All patients seen by a CVRR pharmacist between January 1, 2015 and February 28, 2020 were evaluated for inclusion in the cohort. Patients were included in the study if they had two or more completed visits with the pharmacist at one of the eight clinic sites during the study period and the initial visit was on or before October 28, 2019, had a diagnosis of Type 2 diabetes mellitus (T2DM), and an initial HbA1C ≥ $8\%$. Patients were excluded if their initial HbA1C was < $8\%$ or if they did not have a diagnosis of T2DM. ## Variables and statistical analysis Data collected from the existing REDCap reports included CVRR visit information (clinic site, date of first CVRR visit, referring provider), basic patient demographics (age, gender, race, smoking status, initial statin therapy) and clinical markers (HbA1C, SBP, DBP, LDL, non-LDL, 10-year ASCVD risk). Because of the somewhat unpredictable nature of routine patient follow-up and clinical marker assessment, clinical marker data was captured according to how long after enrollment the clinical marker was assessed. The clinical marker data was assigned to three-month windows with data being allocated to whichever time period was closest to the end date of the clinical marker in question. For patients with two readings within the same window, clinical outcomes were averaged to reflect the level of control within that time period. In the event of a tie between two-time windows, data was allocated to the earlier time window. The primary outcome of interest for this analysis was the change in HbA1C from baseline to the 3–5 month visit with the CVRR pharmacist. In order to determine if there was a clinically significant reduction of 0.5 points in the HbA1C between these paired time points with $80\%$ power, allowable alpha error of 0.05, and a highly conservative standard deviation of 3.5, a sample size of 387 would be necessary [14]. With the more realistic expected standard deviation of 2, based on previous similar assessments within a population with more variability in their diabetes control, a sample size of 128 would be necessary to adequately test this hypothesis [15]. A minimum sample size of 400 was set to ensure that this investigation would have adequate power to identify significant differences in the primary outcomes. The secondary outcomes included the change in SBP, DBP, and LDL between baseline and 3–5 months of follow-up. BP was classified by the highest SBP or DBP reading into the following categories: <$\frac{120}{80}$, 120-$\frac{139}{80}$-89, 140-$\frac{159}{90}$-99, and 160-$\frac{179}{100}$-109, and ≥$\frac{180}{110.}$ In order to assess the persistence of these changes, for the patients with evaluable results the HbA1C, SBP, DBP, and LDL measurements at 6–8 months were also compared to the baseline results. A paired t-test was utilized for all primary and secondary outcome comparisons with a $p \leq 0.05$ being deemed to be statistically significant. The mean of the differences between baseline clinical markers at 3–5 months and 6–8 months were also calculated and presented with the associated $95\%$ confidence intervals. The mean difference was calculated by matching each baseline value to the same patient’s subsequent outcome marker for each timepoint whenever follow-up data was available. Outcome measurements captured between 9 months and 36 months of follow-up were also visually illustrated, however, additional statistical analyses were not performed as the study was not adequately powered to assess those differences and the potential for selection bias after 9 months limits the utility of additional analyses. Patients remaining in the cohort for longer durations were also subject to selection bias as those who were not lost to follow-up might be more likely to experience improvements in their care. Demographic characteristics were described using descriptive statistics and Stata 16® (College Station, TX) was utilized for all statistical analyses. ## Results As seen in the study diagram in Fig 1, a total of 1,270 patients were assessed for eligibility with 631 ultimately being included in the analysis. A total of 445 patients were evaluated to test the primary outcome. Over the 36 months of evaluation, a total of 3,803 encounters were described in this analysis. At 36 months, $9.7\%$ ($$n = 43$$) of the initial study population were eligible for evaluation. Fig 1: Study diagram.Patient characteristics. As seen in Table 1, the mean (SD) age of participants was 54 [11] years. There was a slight predominance of female participants ($56\%$) and African American patients were the main race ($45\%$) within this cohort. Most patients had poorly controlled diabetes at baseline with a mean HbA1C of $10.9\%$ [$95\%$CI 10.7,11.0]. The other clinical markers of SBP, DBP, and LDL were mixed, with many patients achieving the desired targets for these markers prior to enrollment. Table 1 \| Characteristic \| Results (n = 631) \| \| --- \| --- \| \| Age, years, mean [SD] \| 54.0 [11.0] \| \| Age, years, n (%) \| \| \| 11–15 \| 0 (0%) \| \| 16–20 \| 2 (< 1%) \| \| 21–25 \| 2 (< 1%) \| \| 26–30 \| 13 (2%) \| \| 31–40 \| 60 (10%) \| \| 41–50 \| 150 (24%) \| \| 51–60 \| 244 (39%) \| \| 61–70 \| 121 (19%) \| \| 71–80 \| 33 (5%) \| \| >81 \| 6 (< 1%) \| \| Gender, n (%) \| \| \| Male \| 275 (44%) \| \| Female \| 356 (56%) \| \| Race/Ethnicity, n (%) \| \| \| Black or African American \| 286 (45%) \| \| White \| 162 (26%) \| \| Hispanic \| 156 (25%) \| \| Other \| 27 (4%) \| \| HbA1C, mean [SD] \| 10.9% [1.9] \| \| HbA1C, n (%) \| \| \| 8–10% \| 268 (42%) \| \| > 10% \| 363 (58%) \| \| Blood pressure, mmHg, mean [SD] \| SBP 131.7 [18.1] / DBP 79.6 [10.8] \| \| Blood pressure, mmHg, n (%) \| \| \| <120/80 \| 129 (20%) \| \| 120-139/80-89 \| 299 (47%) \| \| 140-159/90-99 \| 145 (23%) \| \| 160-179/100-109 \| 48 (8%) \| \| ≥180/110 \| 10 (2%) \| \| LDL Cholesterol, mg/dL, mean [SD] \| 79.6 [10.8] \| \| LDL Cholesterol, mg/dL, n (%) \| \| \| <70 mg/dL \| 179 (28%) \| \| 70–100 mg/dL \| 185 (29%) \| \| 100–129 mg/dL \| 135 (21%) \| \| ≥130 mg/dL \| 132 (21%) \| \| Clinic Location, n (%) \| \| \| Forest Manor \| 47 (7%) \| \| Grassy Creek \| 31 (5%) \| \| Midtown \| 63 (10%) \| \| North Arlington \| 106 (17%) \| \| Outpatient Care Center \| 29 (5%) \| \| Pecar \| 52 (8%) \| \| West 38th Street \| 124 (20%) \| \| West Side \| 179 (28%) \| ## Primary outcome Compared to baseline, the HbA1C at three to five months was statistically significantly reduced to 9.3 ($95\%$CI [9.1, 9.5]) with a mean difference of 1.6 percent ([$95\%$CI -1.8, -1.4], $p \leq 0.01$) after three to five months of CVRR care (Fig 2). Fig 2: Change in HbA1c from baseline to 36 months. ## Secondary outcomes Reductions in HbA1C persisted at six to eight months (9.1 $95\%$CI [8.9, 9.4], $p \leq 0.01$) with a mean reduction of 1.8 percent ([$95\%$CI -2.0, -1.5],). Patients who remained under the care of the CVRR pharmacist continued to contribute clinical marker data which demonstrated sustained reductions in HbA1C throughout the time period of evaluation (Fig 2). There was a statistically significant drop from the mean baseline SBP of 131.7 ($95\%$CI 130.3, 133.1) to a mean of 130.2 ($95\%$CI 128.6–131.8, $p \leq 0.05$) with a mean difference of -2.3 mmHg SBP at the three to five-month evaluation ([$95\%$CI, -4.1, -0.4]). This statistically significant reduction was not noted at the 6 to 8 month evaluation of SBP (129.7, $95\%$ CI 127.6–131.7, $$p \leq 0.15$$) (Fig 3). Inversely, the diastolic blood pressure at the three to five month time point (78.8, $95\%$CI [77.8, 79.7], $$p \leq 0.05$$, mean difference -1.2 $95\%$CI [-2.3, -0.1]) was not statistically significantly reduced, while the six to eight month diastolic blood pressure was (77.7, $95\%$CI [76.4, 78.9], $p \leq 0.05$, mean difference -2.1, $95\%$CI[-3.4,-0.7]). The analysis of LDL demonstrated statistically significant reductions at both the three to five month time point (-4.3 mg/dL, [$95\%$CI -6.6, -1.1], $p \leq 0.05$) and six to eight month time point (-7.75 mg/dL, [$95\%$CI, -11.3,-4.2], $p \leq 0.05$) (Fig 4). All clinical markers remained reduced or slightly elevated throughout the remaining period of evaluation for patients who continued to contribute results (Figs 2–4). Fig 3: Change in systolic and diastolic blood pressure from baseline to 36 months. Fig 4: Change in LDL from baseline to 36 months. ## Main findings of the study The results of this study confirm that patients who consistently received care from pharmacist-managed CVRR services in FQHCs demonstrate sustained improvements in clinical outcomes related to cardiovascular risk for a minimum of six months. In the UKPDS study, patients who attained a lower HbA1C in the 10-year follow up had significant improvements in cardiovascular disease outcomes [3]. Patients who achieve and sustain glycemic control targets soon after diagnosis of diabetes are more likely to experience a reduction in cardiovascular risk [16]. The improvements in HbA1C demonstrated by the CVRR clinics herein exceed what would be expected from most pharmacologic diabetes treatments [17]. Despite significant improvements, the HbA1C of CVRR patients still remained above the expected goal of < $7\%$ for most individuals in our study. However, given the reductions in HbA1C, LDL, and blood pressure, our study demonstrates clinically significant decreases in cardiovascular risk in high risk populations traditionally associated with grave health disparities [18]. ## What is already known on this topic In 2017, the leading cause of death in Indiana, United States was heart disease, with rates high enough to be ranked 13th worst in the United States [19]. Patients who participate in the evaluated CVRR service represent some of the most vulnerable people in the country, in one of the states ranking worst in cardiovascular health. Populations served at FQHCs often face barriers due to social determinants of health, leading to disparities in care and poorer health outcomes [18, 20]. Given the challenges of this setting, the results of our study are particularly noteworthy for demonstrating statistically and clinically significant improvements in cardiovascular risk. The number of participants exceeded the minimum number to meet power. Our data represent the quality of care provided by six pharmacists covering eight unique FQHC locations with over a decade of clinical service provision. ## What the study adds Given the reduction in risk markers, such as HbA1C which have been observed in the CVRR clinic, there is strong potential for the reduction of major adverse cardiovascular events (MACE) for the high-risk patients seen in these clinics who maintain follow-up over time. Research consistently affirms the positive relationship between reducing HbA1C and lowering the risk of MACE, illuminating the potential for our work to significantly impact MACE outcomes [21]. Our data reflect one period of time across the patient’s continuum of diabetes care. While HbA1C values during the study time period are above $7\%$, the overall trending decrease in HbA1C is clinically valuable. The UKPDS secondary analysis found that a $1\%$ decrease in HbA1C is associated with a $35\%$ reduction in microvascular outcomes, $18\%$ reduction in myocardial infarction, and $17\%$ reduction in all-cause mortality [3]. Additionally, the hope is that patients who are discharged from CVRR services due to achievement of clinical goals are able to sustain healthy outcomes due to the knowledge gained during their interactions with the CVRR pharmacist. In 2017, diabetes expenditures in the United States were an estimated $327 billion, with a higher per person annual cost of $16,750 for those with diabetes compared to those without [22]. Each percentage reduction in HbA1C has been associated with a $685–950 per year savings in healthcare costs [4, 23]. While estimating cost savings was not an objective of this study, the observed reduction of HbA1C seen amongst patients receiving clinical pharmacist care in our clinics could result in considerable savings to the health system. Considering the vast financial impact of cardiovascular disease to society, it follows that the reduction in cardiovascular risk seen in this study is also commensurate with a significant reduction of cost. Given that the majority of patients enrolled in the CVRR service within our health-system are publicly insured, underserved, underinsured, and/or uninsured, any savings related to the care of these patients are also savings to the state and federal governments and taxpayers. Even accounting for the cost of providing the service, clinical pharmacists have demonstrated a positive return-on-investment [1]. In addition to direct revenue and cost avoidance, pharmacists also contribute to improvement in public health initiatives through their impact on quality measures. Goals for chronic disease management measures, especially those related to diabetes and hypertension, are routinely incentivized in value-based care programs by payors in order to assess the quality, safety, effectiveness, and efficiency of care provided [24]. Even in the FQHC setting, a large portion of patients covered by federally funded insurance programs, such as Medicare and Medicaid in the United States, are eligible to be included in value-based models. The improvements in clinical measures from pharmacist interventions within FQHCs can contribute to revenue through incentive payments and/or increases in per-member per-month reimbursement. Given the multiple accrediting agencies and variability in quality measures, healthcare organizations may prioritize different quality measures to evaluate performance. For example, in the United States, Healthy People 2030 and the Health Resources and Services Administration (HRSA) Uniform Data System Clinical Quality Measures specifically focus on a variety of objectives related directly to cardiovascular disease [25, 26]. Clinical pharmacists practicing in underserved areas must be aware of their specific community’s prioritized quality measures and goals. Our data highlight the need for comprehensive pharmacist integration within public health initiatives to optimize patient care and systemic efficiency. Although not formally evaluated in this analysis, there are several potential explanations for success in diabetes management models involving pharmacists. In the United States, the majority of models using referral to a pharmacist service allow for additional one-on-one time dedicated to a focused set of related disease states, as opposed to managing both chronic and acute concerns in a shorter primary care visit [4]. Pharmacist CVRR visits are typically 30 minutes in length and are focused on CVRR-related conditions. In addition, the design of the pharmacist service typically allows for more frequent follow-up visits. At our institution the pharmacist schedules visits every four to six weeks, as compared to PCP visits which may occur every three to five months. More frequent follow-up offers opportunity to reduce therapeutic inertia [7]. Patients can follow with the CVRR service for a brief limited number of appointments or indefinitely based on individual characteristics and preferences, such as continued barriers impacting the ability to manage diabetes well or finding value in continued motivational touch points to maintain clinical goals. Also, given pharmacist training and expertise related to the medication distribution process, their skillset is optimal for ensuring medication access despite common barriers such as financial burden, insurance coverage changes, and manufacturer shortages or recalls. Due to the more frequent touchpoints and this background knowledge, the CVRR pharmacist team is able to prevent or promptly address access issues. Lastly, our pharmacist team possesses up-to-date, evidence-based expertise in clinical practice guidelines, pharmacotherapy algorithms, medication adverse effects, potential medication-related interactions, and other patient-specific considerations. ## Limitations of this study Though the results of this study are robust, there are limitations. The time period used for the analysis did not account for temporary interruptions in the service. In instances of clinical pharmacist absence from the site, data may not have been collected. A number of patients were excluded from analysis due to only having one encounter with the clinical pharmacist. It is possible that there was benefit to these patients realized at the following PCP visit, however this data is not captured and was not evaluated. These patients likely have underlying differences from the patients who engaged with the clinical pharmacists, but due to the lack of detailed records, these differences are unevaluable. Possible reasons for this could include lack of engagement in their personal chronic disease state management, transportation challenges or other financial burden, and lack of perceived value in extra appointments above standard visits with their PCP. Selection bias is also a major potential limitation of this analysis as we could only analyze data from patients who had regular follow-up appointments. It is likely that these patients may have a stronger commitment to engage in healthier behaviors or perhaps an improved awareness about their disease process which may have contributed to the improvement in their clinical markers. It is also difficult to evaluate the independent impact of pharmacist managed services, because patients also maintain concurrent appointments with their primary care providers. In addition, clinical pharmacist care in these clinics is set up to allow patients to graduate out of care and return to routine care without continued follow-up by the clinical pharmacist. These transitions are not routinely documented within the electronic medical record used for this analysis making it difficult to differentiate which patients have been lost to follow-up versus those which appropriately do not require additional clinical pharmacist follow-up. Identifying individual reasons for lack of follow-up was beyond the capability of the electronic medical record report. While all included patients provided data from at least one follow-up visit, the variability in the enrollment date of patients considerably limited the availability of follow-up data after 12 months as seen in Figs 2–4. While fasting and random blood glucose values would have added beneficial insight into overall diabetes management, including the incidence of hypoglycemia, these data were not included in this analysis due to the inconsistency in documentation and reporting of this information in the electronic medical record. HbA1C was identified as the best, standardized measure to report for the purposes of this study. However, underlying conditions that could potentially affect the accuracy of HbA1C values, such as anemia or thyroid conditions, were not collected. Furthermore, specific medication changes and dietary patterns were beyond the scope of this analysis and therefore not included. However, the pharmacists managing medications in this study follow evidence-based recommendations, prioritize first-line medications where appropriate, and limit barriers such as clinical inertia [7, 27]. Additional limitations of the study include retrospective design, lack of control group and evaluation within a single health care system. ## Conclusions Our study augments the existing literature by demonstrating the health improvements associated with pharmacist-managed CVRR clinics within FQHCs in the United States. Our study identified the potential for service improvement to a network of FQHCs serving vulnerable populations, which warrants further evaluation. Additional studies exploring the augmentation or expansion of similar models may amplify the impact of pharmacist-managed CVRR services in primary care sites. ## References 1. 1World Health Organization. Cardiovascular Diseases. Accessed March 12, 2021. Available at: https://www.who.int/health-topics/cardiovascular-diseases/#tab=tab_1. 2. 2Centers for Disease Control and Prevention. Heart Disease Facts. Accessed February 18, 2023. Available at: https://www.cdc.gov/heartdisease/facts.htm. 3. 3American Diabetes Association. Implications of the United Kingdom Prospective Diabetes Study. Accessed February 18, 2023. Available at: https://care.diabetesjournals.org/content/25/suppl_1/s28. 4. Hirsch JD, Bounthavong M, Arjmand A. Estimated Cost-Effectiveness, Cost Benefit, and Risk Reduction Associated with an Endocrinologist-Pharmacist Diabetes Intense Medical Management “Tune-Up” Clinic. Journal of Managed Care & Specialty Pharmacy (2017.0) 23 318-326. PMID: 28230459 5. Rodis JL, Sevin A, Awad MH. Improving Chronic Disease Outcomes Through Medication Therapy Management in Federally Qualified Health Centers. J Prim Care Community Health (2017.0) 8 324-331. DOI: 10.1177/2150131917701797 6. Rodis JL, Capesius TR, Rainey JT. Pharmacists in Federally Qualified Health Centers: Models of Care to Improve Chronic Disease. Prev Chronic Dis (2019.0) 16 E153. DOI: 10.5888/pcd16.190163 7. Meredith AH, Buatois EM, Krenz JM. Assessment of Clinical Inertia in People with Diabetes Within Primary Care. Journal of Evaluation in Clinical Practice (2021.0) 27 365-370. DOI: 10.1111/jep.13429 8. 8Federally Qualified Health Centers. Health Resources and Services Administration. Available at: https://www.hrsa.gov/opa/eligibility-and-registration/health-centers/fqhc/index.html. Accessed February 18, 2023. 9. 9About the Health Center Program. Federally Qualified Health Centers. Available at: https://bphc.hrsa.gov/about/index.html. Accessed February 18, 2023. 10. Di Palo KE, Patel K, Kish T. Risk Reduction to Disease Management: Clinical Pharmacists as Cardiovascular Care Providers. Curr Probl Cardiol (2019.0) 44 276-293. DOI: 10.1016/j.cpcardiol.2018.07.003 11. Cohen LB, Taveira TH, Khatana SA. Pharmacist-led shared medical appointments for multiple cardiovascular risk reduction in patients with type 2 diabetes. Diabetes Educ (2011.0) 37 801-812. DOI: 10.1177/0145721711423980 12. Weber ZA, Skelley J, Sachdev G. Integration of Pharmacists into Team-Based Ambulatory Care Practice Models. American Journal of Health-System Pharmacy (2015.0) 72 745-751. DOI: 10.2146/ajhp140576 13. 13Collaborative Practice Agreements and Pharmacists’ Patient Care Services. National Center for Chronic Disease Prevention and Health Promotion. https://www.cdc.gov/dhdsp/pubs/docs/translational_tools_pharmacists.pdf Accessed February 18, 2023. 14. de Barra M, Scott CL, Scott NW, Johnston M, de Bruin M, Nkansah N. Pharmacist services for non-hospitalised patients. Cochrane Database of Systematic Reviews (2018.0). DOI: 10.1002/14651858.CD013102 15. Pastakia SD, Nuche-Berenguer B, Pekny CR. Retrospective assessment of the quality of diabetes care in a rural diabetes clinic in Western Kenya. BMC Endocrine Disorders (2018.0) 18 16. Skylar JS, Bergenstal R, Bonow RO. Intensive glycemic control and the prevention of cardiovascular events: implications of the ACCORD, ADVANCE, and VA diabetes trials. Diabetes Care (2009.0) 32 187-192. PMID: 19092168 17. Chaudhury A, Duvoor C, Reddy Dendi VS. Clinical Review of Antidiabetic Drugs: Implications for Type 2 Diabetes Mellitus Management. Front Endocrinol (Lausanne) (2017.0) 8 6. DOI: 10.3389/fendo.2017.00006 18. Graham G.. Disparities in cardiovascular disease risk in the United States. Curr Cardiol Rev (2015.0) 11 238-245. DOI: 10.2174/1573403x11666141122220003 19. 19Centers for Disease Control and Prevention. Stats of the State of Indiana. Accessed February 18, 2023. Available at: https://www.cdc.gov/nchs/pressroom/states/indiana/indiana.htm 20. Nath JB, Costigan S, Hsia RY. Changes in Demographics of Patients Seen at Federally Qualified Health Centers, 2005–2014. JAMA Intern Med (2016.0) 176 712-714. DOI: 10.1001/jamainternmed.2016.0705 21. Giugliano D., Chiodini P., Maiorino M.I.. Cardiovascular outcome trials and major cardiovascular events: does glucose matter? A systematic review with meta-analysis. J Endocrinol Invest (2019.0) 42 1165-1169. DOI: 10.1007/s40618-019-01047-0 22. 22Centers for Disease Control and Prevention. Return on Investment. Accessed February 18, 2023. Available at: https://www.cdc.gov/diabetes/dsmes-toolkit/business-case/roi.html. 23. Wagner EH, Sandhu N, Newton KM. Effect of improved glycemic control on health care costs and utilization. JAMA (2001.0) 285 182-189. DOI: 10.1001/jama.285.2.182 24. Andrawis M. Recommended quality measures for health-system pharmacy: 2019 update from the Pharmacy Accountability Measures Work Group. Am J Health Syst Pharm (2019.0) 76 874-887. DOI: 10.1093/ajhp/zxz069 25. 25Health Resources and Services Administration. Uniform Data System (UDS) Clinical Quality Measures and Related Healthy People 2020 Goals. Accessed February 18, 2023. Available at: https://bphc.hrsa.gov/program-opportunities/sac/uds-measures-and-hp-goals. 26. 26U.S. Department of Health and Human Services. Healthy People 2030. Accessed February 18, 2023. Available at: https://health.gov/healthypeople. 27. Meredith AH, Starks S, Etelemaki C. Evaluation of Sodium-Glucose Cotransporter-2 Inhibitor Use in an Underserved Ambulatory Care Population With Type 2 Diabetes. ADCES In Practice (2022.0) 10 16-24
--- title: 'Healthcare seeking patterns for TB symptoms: Findings from the first national TB prevalence survey of South Africa, 2017–2019' authors: - Sizulu Moyo - Farzana Ismail - Nkateko Mkhondo - Martie van der Walt - Sicelo S. Dlamini - Thuli Mthiyane - Inbarani Naidoo - Khangelani Zuma - Marina Tadolini - Irwin Law - Lindiwe Mvusi journal: PLOS ONE year: 2023 pmcid: PMC10016667 doi: 10.1371/journal.pone.0282125 license: CC BY 4.0 --- # Healthcare seeking patterns for TB symptoms: Findings from the first national TB prevalence survey of South Africa, 2017–2019 ## Abstract ### Background Although tuberculosis (TB) symptoms have limited sensitivity they remain an important entry point into the TB care cascade. ### Objectives To investigate self-reported healthcare seeking for TB symptoms in participants in a community-based survey. ### Methods We compared reasons for not seeking care in participants reporting ≥1 of four TB screening symptoms (cough, weight loss, night sweats, fever) in the first South African national TB prevalence survey (2017–2019). We used logistic regression analyses to identify sociodemographic and clinical characteristics associated with healthcare seeking. ### Results 5,$\frac{168}{35}$,191 ($14.7\%$) survey participants reported TB symptoms and 3,$\frac{442}{5168}$ had not sought healthcare. 2,$\frac{064}{3}$,442($60.0\%$) participants intended to seek care, 912 ($26.5\%$) regarded symptoms as benign, 399 ($11.6\%$) reported access barriers(distance and cost), 36 ($1.0\%$) took other medications and 20($0.6\%$) reported health system barriers. Of the $\frac{57}{98}$ symptomatic participants diagnosed with bacteriologically confirmed TB who had not sought care: 38($66.7\%$) intended to do so, 8($14.0\%$) regarded symptoms as benign, and 6($10.5\%$) reported access barriers. Among these 98, those with unknown HIV status(OR 0.16 $95\%$ CI 0.03–0.82), $$p \leq 0.03$$ and those who smoked tobacco products(OR 0.39, $95\%$ CI 0.17–0.89, $$p \leq 0.03$$) were significantly less likely to seek care. ### Conclusions People with TB symptoms delayed seeking healthcare, many regarded symptoms as benign while others faced access barriers. Those with unknown HIV status were significantly less likely to seek care. Strengthening community-based TB awareness and screening programmes together with self-screening models could increase awareness of the significance of TB symptoms and contribute to improving healthcare seeking and enable many people with TB to enter the TB care cascade. ## Background Over the past two decades many countries have made significant commitments to end tuberculosis (TB) through the adoption of the World Health Organization (WHO) End TB Strategy, and the United Nations (UN) Sustainable Development Goals (SDGs) [1, 2] among other national commitments. It is against this background that significant progress has been made in TB diagnostics and treatment [3–5]. However, in high TB burden settings many people with TB remain undetected, undiagnosed and untreated [6, 7]. Additionally, in the past three years, COVID-19 regressed the work and gains of health programmes including TB programmes [6, 8]. COVID-19 related disruptions drastically reduced TB testing and notifications and disrupted the supply and provision of TB medication and clinical care for people with TB resulting in an increase in global TB mortality in 2020 [6, 8]. Recovering, improving, and sustaining progress to eliminate TB requires interventions throughout the TB care cascade. This includes addressing barriers to entry into the care cascade in combination with availability and accessibility of rapid and highly sensitive diagnostic tools, and short treatment regimens with minimal adverse effects. Although the four common TB symptoms alone have poor sensitivity in screening for TB [9],they remain a significant entry point into the TB care cascade for many people who have TB. In this paper we analyze data on self-reported healthcare seeking in individuals with TB symptoms who participated in the first national TB prevalence survey undertaken in South Africa in 2017–2019. As South Africa and other countries seek to find and treat more people with TB, such analyses can inform efforts to improve and support screening, by identifying the main barriers preventing healthcare seeking among those with symptoms suggestive of TB and provide evidence for interventions to offer them testing and treatment services more efficiently. ## Methods We used data from the first South African national TB prevalence survey undertaken in 2017–2019 [10]. The detailed survey methodology has been described elsewhere [10]. In summary, the survey enrolled eligible participants aged 15 years and older in 110 clusters sampled by probability proportional to size across all the 9 provinces of the country. Those enrolled were screened for TB using four TB screening symptoms [9]: persistent cough of any duration (we also asked about the duration of the cough), unexplained weight loss, unexplained fever for ≥2 weeks and drenching night sweats for ≥2 weeks, and by digital chest X-ray (CXR). Symptoms were elicited by trained interviewers using a structured questionnaire. The questionnaire also collected data on collected data on selected behaviours, health status and reasons for not seeking care for TB symptoms among those reporting symptoms. Participants could select from the following reasons a) Distance- facility is far from where I live, b) Money- I had no money for transport to the health centre, c) Importance/relevance of symptoms- I did not consider it to be important, d) Still planning to seek care, and e) Other reasons-(open ended to capture all other reasons). Participants could select and report multiple reasons. Optional data on self-reported HIV status was also collected. Those who screened positive (symptomatic and/or CXR changes suggestive of active TB) submitted two sputum samples for testing by Xpert MTB/RIF Ultra assay (Cepheid, Sunnyvale, CA, USA) and liquid culture (Bactec MGIT 960 Becton Dickinson, Franklin Lakes, NJ, USA) and dried blood spot(DBS) samples for HIV testing if they consented to HIV testing. Sputum samples were processed at the Centre for Tuberculosis at the National Institute for Communicable Diseases (NICD). HIV testing was performed by the Centre for HIV and STIs (CHIVSTI) at the NICD. Participants who had Mycobacterium tuberculosis complex culture-positive sputum samples were considered positive for bacteriologically confirmed pulmonary TB. In the absence of a positive culture (i.e., negative, contaminated, or not done), participants were considered positive for bacteriologically confirmed pulmonary TB in the presence of an Xpert MTB/RIF Ultra-positive sample, together with an abnormal CXR suggestive of active TB (determined by central panel CXR readers), and no history of past TB. The final HIV status was based on the DBS result where available, and on self-reported status otherwise. HIV status was unknown where participants declined to disclose their status and refused DBS testing. ## Data analysis We analysed data on self-reported healthcare seeking for survey participants who reported one or more of the four screening symptoms. We used STATA v15.0 and summarized the data using frequencies, proportions, medians and IQRs as appropriate, and used the Pearson’s Chi Square test to compare categorical variables, with $p \leq 0.05$ indicating statistical significance. We assessed factors associated with seeking care using bivariate and multivariate logistic regression analysis. Variables that were significant in bivariate analyses were included in a multivariate regression analysis with variables simultaneously adjusted for each other. We report odds ratios (ORs) and adjusted odds ratios (aORs) with $95\%$ confidence intervals (CIs) and p values, and $p \leq 0.05$ indicates statistical significance. We initially grouped the data on reasons for not seeking healthcare into four categories as i) those who were planning to seek care (Still planning to seek care), ii) those facing access barriers (cost and distance;-Distance- facility is far from where I live, and Money- I had no money for transport to the health centre), iii) those who regarded their symptoms as benign (Importance/relevance of symptoms- I did not consider it to be important) and iv) Other reasons. We further examined the Other reasons category and allocated these to categories i) to iii) above as appropriate and also created two additional categories that captured health facility factors (e.g., the clinic congestion and poor staff attitudes) and use of other medication (use of over-the-counter medication, or traditional medication). ## Ethics The survey protocol was approved by the South African Medical Research Council Research Ethics Committee (EC001 $\frac{2}{2012}$). Written informed consent or assent and parent or guardian consent for participants younger than 18 years old was obtained before enrolment into the survey. ## Results Of the 35,191 participants enrolled in the survey 5,168($14.7\%$), median age of 47 years (IQR 32–61), and mostly women (3,064, ($59.3\%$)), reported at least one of the four TB symptoms. Overall, 2,968 participants reported one symptom only and this was most frequently cough ($$n = 1$$,400 ($47.2\%$)) followed by night sweats ($$n = 660$$ ($22.2\%$)), weight loss ($$n = 641$$ ($21.6\%$)), and fever($$n = 267$$ ($9.0\%$)). 3,442 /5,168 ($66.6\%$) participants reporting symptoms had not sought care for them. The majority, 2,064 ($60.0\%$) reported that they intended to seek care, 912($26.5\%$) regarded their symptoms as benign, and 399 ($11.6\%$) reported access barriers to healthcare seeking (distance—healthcare facilities far from where they lived and cost—unable to afford travel costs). There were 36 ($1.0\%$) participants who reported that they were taking other medications for their symptoms, while 20($0.6\%$) people did not seek care because of health system related factors (facility congestion, poor staff attitudes), and 4($0.1\%$) who were afraid of a possible diagnosis of TB. In a further 7 participants no reasons were given (S1 Table). There was no statistically significant difference in the proportions of those who reported that they intended to seek care by sex, diabetes, and past history of TB, (Table 1). More participants aged 25–49 years old ($62.6\%$ 25-49years, $57.5\%$ 15-24years, $58.1\%$ ≥ 50years), living in urban areas($64.9\%$ urban, $55.2\%$ rural, $$p \leq 0.000$$), those with Grade 1–12 education ($53.9\%$ no education, $61.4\%$ Grade 1–12, $56.6\%$ tertiary level, $$p \leq 0.003$$), those with known HIV status ($62.9\%$ HIV+ve, $61.7\%$ HIV-ve, $52.3\%$ HIV status unknown, p< = 0.000), those who reported smoking tobacco products ($62.7\%$ smokers, $58.3\%$ nonsmokers, $$p \leq 0.01$$), and those who reported no alcohol intake(not taking alcohol $61.6\%$, 57.8 taking alcohol, $$p \leq 0.03$$) intended to seek care for their symptoms (Table 1). Among the 558 symptomatic participants aged 15–24 years old who had not sought care, $31.7\%$ regarded their symptoms as benign and this proportion was significantly greater than in the older age groups ($26.4\%$ among those 25–49 years old and $24.5\%$ among those ≥ 50 years old, $$p \leq 0.01$$). A significantly greater proportion of those with tertiary education ($31.9\%$ tertiary level, $27.3\%$ Grade 1–12 and $21.3\%$ no education, $$p \leq 0.01$$), those with unknown HIV status ($30.9\%$ HIV status unknown, $23.7\%$ HIV+ve, $25.9\%$ HIV-ve, $$p \leq 0.01$$), those with no history of past TB ($27.2\%$ no past history of TB, $21.1\%$ with history of past TB, $$p \leq 0.01$$), and those who consumed alcohol (consume alcohol $30.2\%$, $23.7\%$ no g alcohol, $$p \leq 0.000$$) regarded their symptoms as benign (Table 1). Access barriers were most frequently reported by females ($9.9\%$ males, $12.9\%$ females, $$p \leq 0.01$$), those ≥50 years ($15.9\%$ in those ≥50 years old, $9.3\%$ in those 25–49 years old, $7.34\%$ 15–24 years old, $$p \leq 0.000$$), those with no education ($23.8\%$ no education, $9.4\%$ Grade 1–12 and $3.5\%$ tertiary, $$p \leq 0.000$$), and those living in rural areas ($17.1\%$ rural, $5.8\%$ urban, $$p \leq 0.000$$), those with unknown HIV status (HIV status unknown $15.2\%$, HIV -ve $10.3\%$, HIV +ve $12.1\%$, $$p \leq 0.002$$), those who did not consume alcohol (no alcohol $12.6\%$, $10.1\%$ taking alcohol, $$p \leq 0.02$$) and those who did not smoke cigarettes ($13.1\%$ nonsmokers, $9.1\%$ smokers, $$p \leq 0.000$$) (Table 1). Table 1 \| Unnamed: 0 \| Unnamed: 1 \| Reason for not seeking care \| Reason for not seeking care.1 \| Reason for not seeking care.2 \| Reason for not seeking care.3 \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Variable \| Did not seek care n \| Still planning to seek care n (%) \| Symptoms regarded as benign n (%) \| Access barriers n (%) \| Other reasons \| \| Sex \| \| p = 0.55 \| p = 0.06 \| p = 0.01 \| \| \| Male \| 1500 \| 891 (59.4) \| 422 (28.1) \| 149 (9.9) \| 38 \| \| Female \| 1942 \| 1,173 (60.4) \| 490 (24.5) \| 250 (12.9) \| 29 \| \| Age group (years) \| \| p = 0.02 \| p = 0.01 \| p = 0.001 \| \| \| 15–24 \| 558 \| 321 (57.5) \| 177 (31.7) \| 43 (7.7) \| 34 \| \| 25–49 \| 1522 \| 952 (62.6) \| 402 (26.4) \| 140 (9.2) \| 28 \| \| ≥50 \| 1362 \| 791 (58.1) \| 333 (24.5) \| 216 (15.9) \| 22 \| \| Locality \| \| p = 0.000 \| p = 0.33 \| p = 0.000 \| \| \| Urban \| 1677 \| 1089 (64.9) \| 457 (27.3) \| 98 (5.4) \| 33 \| \| Rural \| 1765 \| 975 (55.2) \| 455 (25.8) \| 301 (17.1) \| 76 \| \| Highest education level achieved \| \| p = 0.002 \| p = 0.01 \| p = 0.000 \| \| \| \| 572 \| 308(53.9) \| 122 (21.3) \| 136 (23.8) \| 6 \| \| Grade 1–12 \| 2755 \| 1,692 (61.4) \| 752 (27.3) \| 259 (9.4) \| 52 \| \| Tertiary \| 113 \| 64 (56.6) \| 36 (31.9) \| 4 (3.5) \| 9 \| \| Missing \| 2 \| \| \| \| \| \| # HIV status \| \| p = 0.000 \| p = 0.01 \| p = 0.002 \| \| \| HIV positive \| 574 \| 361 (62.9) \| 136 (23.7) \| 70 (12.2) \| 7 \| \| HIV negative \| 2164 \| 1,335 (61.7) \| 560 (25.9) \| 222 (10.3) \| 47 \| \| HIV status unknown \| 704 \| 368 (52.3) \| 216 (30.9) \| 107 (15.2) \| 13 \| \| Diabetes (self-report) \| \| p = 0.66 \| p = 0.47 \| P = 0.15 \| \| \| No \| 3124 \| 1,881 (60.2) \| 832 (26.6) \| 352 (11.3) \| 59 \| \| Yes \| 207 \| 118 (57.0) \| 55 (26.6) \| 30 (14.5) \| 4 \| \| Don’t know \| 108 \| 65 (60.2) \| 23 (21.3) \| 17 (15.7) \| 3 \| \| Missing \| 3 \| \| \| \| \| \| History of past TB \| \| p = 0.39 \| p = 0.01 \| p = 0.11 \| \| \| Yes \| 445 \| 276 (62.0) \| 94 (21.1) \| 61 (13.7) \| 14 \| \| No \| 2997 \| 1,782 (59.9) \| 811 (27.2) \| 331 (11.1) \| 73 \| \| unknown \| 20 \| \| \| \| \| \| Smoke tobacco products \| \| p = 0.01 \| p = 0.70 \| p = 0.000 \| \| \| No \| 1996 \| 1,163 (58.3) \| 521 (26.1) \| 268 (13.4) \| 44 \| \| Yes \| 1435 \| 899 (62.7) \| 383 (26.7) \| 130 (9.1) \| 23 \| \| Missing \| 11 \| \| \| \| \| \| Consume alcohol \| \| p = 0.03 \| p = 0.000 \| p = 0.02 \| \| \| No \| 2006 \| 1,236 (61.6) \| 476 (23.7) \| 253 (12.6) \| 41 \| \| Yes \| 1423 \| 823 (57.8) \| 430 (30.2)) \| 143 (10.1) \| 27 \| \| Missing \| 13 \| \| \| \| \| When the analyses excluded participants reporting cough of less than 2 weeks, there was no statistically significant difference in the proportions of those who reported that they intended to seek care by sex, age group, past history of TB, and diabetes (S2 Table). More participants living in urban areas, those with Grade 1–12 education and those who smoked tobacco products intended to seek care, while significantly lower proportions of those with unknown HIV status and those who consumed alcohol intended to seek care. Findings were also similar for those who regarded their symptoms as benign. Significantly more young people 15–24 years old, those with a tertiary level of education those with unknown HIV status, and those who consumed alcohol regarded their symptoms as benign. Women, the elderly, those living in rural areas, those with unknown HIV status, those who did not smoke, and those who did not consume alcohol most frequently reported access barriers. ( S2 Table). Reasons for not seeking healthcare were heterogenous across all of the four screening symptoms, across different combinations of symptoms and number of concurrent symptoms (S3 Table). Nearly a third of those reporting cough only, and a third of those who reported all four symptoms were planning to seek care. When stratified by HIV status, $72.7\%$ of those reporting all four symptoms and were HIV+ve were planning to seek care, compared to $61.0\%$ of those who were HIV-ve and $62.5\%$ of those with unknown HIV status. Among those who reported cough only, $66.7\%$, $66.5\%$ and $55.6\%$ were planning to seek care for the cough among those living with HIV (LWH), those without HIV and those with unknown HIV status respectively. These proportions were largely unchanged when analysis was restricted to cough of 2 weeks or longer, at $71.5\%$ for those HIV-positive, $60.0\%$ among those HIV-ve and $60.5\%$ among those with unknown status. ## Factors associated with self-reported care seeking for TB symptoms Overall, older people were more likely to seek care for their symptoms, with those ≥50 years old 3 times (aOR 3.14, $95\%$ CI 2.50–3.96) more likely to seek care compared to those 15–24 years old. Participants living in rural areas were also more likely to seek care than those living in urban areas (aOR 1.71, $95\%$ CI 1.03–1.33), as were those with a history of past TB (aOR 1.61, $95\%$ CI 1.37–1.90), and those with diabetes (aOR 1.48, $95\%$ CI 1.18–1.84) (Table 2). Participants who were HIV-ve (aOR 0.57, $95\%$ CI 0.49–0.67) and those with unknown HIV status (aOR 0.45,$95\%$ CI 0.37–0.56) were less likely to seek care for their symptoms as were those who smoked cigarettes (aOR 0.64, $95\%$ CI 0.54–0.74) and those who consumed alcohol (aOR 0.81, $95\%$ CI 0.70–0.94) (Table 2). Table 2 \| Characteristic \| Number reporting symptoms \| Sought care n(%) \| OR (95% CI) \| p value \| aOR (95% CI) \| p value.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Sex \| \| \| \| \| \| \| \| Male \| 2104 \| 604 (28.7) \| ref \| \| ref \| \| \| Female \| 3, 064 \| 1,122 (36.6) \| 1.44 (1.27–1.62) \| 0.0 \| 1.03 (0.89–1.19) \| 0.68 \| \| Age group (years) \| \| \| \| \| \| \| \| 15–24 \| 678 \| 120 (17.7) \| ref \| \| ref \| \| \| 25–49 \| 2121 \| 599 (28.2) \| 1.83 (1.47–2.28) \| 0.0 \| 1.66 (1.32–2.09) \| <0.001 \| \| ≥50 \| 2369 \| 1007 (42.5) \| 3.43 (2.78–4.26) \| 0.0 \| 3.14 (2.50–3.96) \| <0.001 \| \| Locality \| \| \| \| \| \| \| \| Urban \| 2381 \| 704 (29.6) \| ref \| \| ref \| \| \| Rural \| 2787 \| 1,022 (36.7) \| 1.38 (1.23–1.55) \| 0.0 \| 1.17 (1.03–1.33) \| 0.02 \| \| Highest education level achieved \| \| \| \| \| \| \| \| \| 998 \| 426 (42.69) \| \| \| ref \| \| \| Grade 1–12 \| 4009 \| 1,254 (31.28) \| 0.61 (0.53–0.70) \| 0.0 \| 0.89 (0.75–1.04) \| 0.15 \| \| Tertiary \| 157 \| 44 (28.03) \| 0.52 (0.36–0.76 \| 0.0 \| 0.98 (0.66–1.46) \| 0.93 \| \| Missing \| 4 \| \| \| \| \| \| \| # HIV status # \| \| \| \| \| \| \| \| HIV-positive \| 1017 \| 443 (43.6) \| \| \| ref \| \| \| HIV-negative \| 3156 \| 992 (31.4) \| 059 (0.51–0.69) \| 0.0 \| 0.57 (0.49–0.67) \| 0.000 \| \| HIV status unknown \| 995 \| 291 (29.3) \| 0.54 (0.45–0.64) \| 0.0 \| 0.45 (0.37–0.56 \| 0.000 \| \| Diabetes (self-report) \| \| \| \| \| \| \| \| No \| 4628 \| 1,504 (32.5) \| ref \| \| ref \| \| \| Yes \| 392 \| 185 (47.2) \| 1.86 (1.51–2.28) \| \| 1.48 (1.18–1.84) \| 0.001 \| \| Don’t know \| 139 \| 31 (22.3) \| 0.60 (0.40–0.89) \| 0.01 \| 0.62 (0.41–0.94) \| 0.02 \| \| Missing \| 9 \| \| \| \| \| \| \| History of past TB \| \| \| \| \| \| \| \| No \| 4327 \| 1,350 (31.2) \| ref \| \| ref \| \| \| Yes \| 810 \| 365 (45.1) \| 1.81(1.55–2.11) \| 0.001 \| 1.61 (1.37–1.90) \| 0.000 \| \| Unknown \| 31 \| \| \| \| \| \| \| Smoke tobacco products \| \| \| \| \| \| \| \| No \| 3252 \| 1,256 (38.6) \| ref \| \| ref \| \| \| Yes \| 1901 \| 466 (24.5) \| 0.52 (0.45–0.59) \| 0.001 \| 0.64 (0.54–0.74) \| 0.000 \| \| Missing \| 15 \| \| \| \| \| \| \| Consume alcohol \| \| \| \| \| \| \| \| No \| 3214 \| 1,208 (37.6) \| ref \| \| ref \| \| \| Yes \| 1940 \| 517 (26.7) \| 0.60 (0.53–0.68) \| 0.001 \| 0.81 (0.70–0.94) \| 0.01 \| \| Missing \| 14 \| \| \| \| \| \| In the regression analysis that excludes participants reporting cough of less than 2 weeks duration, the factors associated care seeking were generally similar to those identified when cough of any duration was used (S4 Table). Older people (aOR 1.61, $95\%$ CI1.37–1.90), those with a history of TB (aOR 1.57, $95\%$ CI 1.32–1.88), and those with diabetes (aOR 1.48, $95\%$ CI 1.18–1.84) were significantly more likely to seek care. Although those living in rural areas were also more likely to seek care than those living in urban areas, this did not reach statistical significance (1.15, $95\%$CI 1.00–1.32, $$p \leq 0.05$$). Unknown and negative HIV status, smoking tobacco products and consuming alcohol were significant predictors of not seeking care, in this subgroup. In this survey 234 people had bacteriologically confirmed TB, and among them 98 reported symptoms with 41 having sought care for their symptoms, as previously reported [10]. Among these 98 participants we found no statistically significant differences in healthcare seeking by sociodemographic characteristics and history of TB. Those with unknown HIV status OR 0.16 $95\%$ CI 0.03–0.82), $$p \leq 0.03$$, and those who smoked cigarettes (OR 0.39 (0.17–0.89), $$p \leq 0.03$$) were significantly less likely to seek care for their symptoms (Table 3). Among $\frac{57}{98}$ ($58.2\%$) participants with bacteriologically confirmed TB who had not sought care for symptoms $\frac{38}{57}$ ($66.7\%$) intended to do so, $\frac{8}{57}$ ($14.0\%$) regarded their symptoms as benign, and $\frac{6}{57}$ ($10.5\%$) reported access barriers (S1 Table). Table 3 \| Characteristic \| Number reporting symptoms \| Sought care \| OR 95% CI \| p value \| \| --- \| --- \| --- \| --- \| --- \| \| Sex \| \| \| \| \| \| Male \| 47.0 \| 19 (40.4) \| ref \| \| \| Female \| 51.0 \| 22 (43.1) \| 1.12 (0.50–2.50) \| 0.79 \| \| Age group (years) \| \| \| \| \| \| 15–24 \| 8.0 \| 2 (25.0) \| ref \| \| \| 25–49 \| 50.0 \| 21 (42.0) \| 2.17 (0.40–11.84) \| 0.37 \| \| ≥50 \| 40.0 \| 18 (45.0) \| 2.45 (0.44–13.67) \| 0.31 \| \| Locality \| \| \| \| \| \| Urban \| 56.0 \| 23 (41.2) \| ref \| \| \| Rural \| 42.0 \| 18 (42.9) \| 1.08 (0.48–2.42) \| 0.86 \| \| # HIV status \| \| \| \| \| \| HIV positive \| 31.0 \| 16 (51.6) \| ref \| \| \| HIV negative \| 53.0 \| 23 (43.4) \| 0.72 (0.30–1.75) \| 0.48 \| \| HIV status unknown \| 14.0 \| 2 (14.3) \| 0.16 (0.03–0.82) \| 0.03 \| \| Diabetes(self-report) \| \| \| \| \| \| No \| 88.0 \| 38(43.2) \| ref \| \| \| Yes \| 7.0 \| 2 (28.6) \| 0.53(0.10–2.86) \| 0.46 \| \| Don’t know \| 3.0 \| 1 (33.3) \| - \| \| \| History of past TB \| \| \| \| \| \| No \| 71.0 \| 28 (39.4) \| ref \| \| \| Yes \| 27.0 \| 13 (48.2) \| 1.43(0.58–3.48) \| 0.44 \| \| Highest education level achieved \| \| \| \| \| \| \| 14.0 \| 5(35.7) \| ref \| \| \| Grade 1–12 \| 82.0 \| 36(43.9) \| 1.41(0.43–4.57) \| 0.59 \| \| Tertiary \| 2.0 \| 0(0) \| \| \| \| Consume alcohol \| \| \| \| \| \| No \| 57.0 \| 28(49.1) \| ref \| \| \| Yes \| 41.0 \| 13 (31.7) \| 0.48 (0.21–1.11) \| 0.09 \| \| Smoke tobacco products \| \| \| \| \| \| No \| 49.0 \| 26(53.1) \| ref \| \| \| Yes \| 49.0 \| 15 (30.6) \| 0.39 (0.17–0.89) \| 0.03 \| ## Discussion In this community TB survey, overall, $66.6\%$ of symptomatic participants had not sought care for their symptoms. Among those with symptoms who were diagnosed with bacteriologically confirmed TB more than half had not sought care for their symptoms. The majority of participants delayed seeking care reporting that they were intending to do so. Also concerning is that more than $14\%$ of participants with bacteriologically confirmed TB despite presenting TB suggestive symptoms did not consider their symptoms serious enough to seek care. Timely care seeking and screening for TB enables early detection and initiation of treatment for active TB thus curbing transmission, and also providing opportunity for screening and management of close contacts [11]. Although the four WHO TB screening symptoms have low sensitivity [9], in this survey we previously reported that over $40\%$ of people with bacteriologically confirmed TB reported symptoms($6.8\%$ were symptom screen positive only and $35.0\%$ screen positive on symptoms and CXR) indicating the role of symptoms in finding people with TB [10]. Studies have shown that people with TB symptoms delay seeking care for at least one month [12, 13] which indicates that they contribute to sustained community transmission of TB. In this analysis the survey criteria for symptoms were a duration of at least two weeks for night sweats, weight loss, and fever, and although $60\%$ of participants reported intention to seek care, our findings are nonetheless suggestive of delayed healthcare seeking consistent with the literature [12, 13]. Care seeking was delayed even among those who reported cough of 2 weeks or longer (S2 and S4 Tables). Data from population surveys in South Africa has shown high levels of overall knowledge about TB and its transmission [14, 15] indicating a need to better understand and address the factors that impede prompt action on this knowledge. In this survey many people did not regard TB symptoms as serious enough to seek care. TB symptoms may be mild and may also overlap with those of other respiratory infections, and this finding is not novel. However, given the spectrum of TB disease presentation, propensity for transmission and the need to accelerate action to address the high TB burden in the country, this finding underscores a need to radically change the approach to TB symptoms. The COVID-19 pandemic has elevated the value and feasibility of various models of remote patient interaction and management, with wide use of SMS/WhatsApp messaging for screening and communicating test results. Self-screening on the WhatsApp and SMS-based TB Healthcheck App [16] which aims to increase TB screening and early case detection in South Africa is one such example, which can prompt early action on symptoms. The success of this screening intervention will also depend on its popularization and efficient management of those who self-screen positive and present for additional in-facility screening and testing. Efficient management will include fast tracking of these people to reduce patient waiting times and adherence to infection control policies to manage potential TB transmission in facilities [17]. Other barriers to seeking healthcare for TB symptoms that we found are access barriers that have previously been reported by studies on healthcare access and utilization in South Africa beyond just for TB services, especially in rural areas where distances traveled to health facilities and the related costs are a factor [18, 19]. In this analysis $10\%$ of those with bacteriologically confirmed TB reported access barriers related to distance and travel costs. Interventions such as community-based programmes undertaken by Community Healthcare Workers in the Primary Healthcare model can address some of these distance and travel cost barriers [20]. Community Healthcare Workers also provide opportunity to increase awareness about the significance of and the urgency to act on TB symptoms and to link those at risk to social support programmes that can support people to better access healthcare facilities and throughout the TB diagnosis and treatment journey. People in urban areas likely face other access barriers including the possibility of losing jobs and other opportunities while attending healthcare facilities that may be congested and have long and unpredictable patient wait times [21–23]. In this survey a small proportion of participants reported health system barriers. The 2015 South African Department of Health’s National policy on management of patient waiting times in outpatient departments proposed use of the Central Chronic Medicine Dispensing and Distribution (CCMDD) system, working with Ward-based Primary Healthcare Outreach Teams (WBPHCOTs) that include Community Healthcare Workers in the Primary Healthcare model, appointment systems, SMS/WhatsApp reminders, and patient education to minimize avoidable visits as measures to reduce patient wait times and decongest healthcare facilities [24]. Some of these recommendations including the deployment and work of Community Healthcare Workers in some communities and sending TB test results to patients by SMS have now been implemented [25]. South Africa also has TB workplace polices whose strengthening could support early detection and appropriate management of TB in the workplace [26] and remove delayed care seeking related to fears of job losses and stigma. South Africa is also implementing the Multisectoral Accountability Framework on TB which brings together multiple sectors essential for a successful TB response and this includes other government departments with responsibility for policies such as the TB workplace polices [27]. Access barriers may be exacerbated by stigma, especially given the intertwining of HIV and TB [28, 29], causing people to seek healthcare far from where they live or work. This can exacerbate healthcare seeking costs even though TB services are provided at no cost to patients. Therefore, measures to find people with TB should also address the upstream determinants including stigma and the dual TB and HIV stigma. Participants with TB who had unknown HIV status were less likely to seek care, and unknown HIV status is sometimes linked to fear and stigma that can limit testing for HIV and healthcare seeking for other symptoms [30, 31]. Females, the elderly, and those living in rural areas, were significantly more likely to seek care, whereas those without HIV, those with unknown HIV status and those without previous TB, were less likely to seek care. Men’s lower likelihood to seek care for TB symptoms when compared to women and the possible reasons for this are well documented in the literature [32–34]. Efforts to address these gaps should continue including implementing specific interventions that target men [35] and encouraging men to use self-screening modalities such as the TB Healthcheck App. As expected, participants with the comorbidities of HIV and diabetes investigated in this study were more likely to seek care probably driven by structured appointment-based consultations for these comorbid conditions. In 2020, HIV status was known for $66\%$ of notified TB cases in South Africa and among these $71\%$ were LWH with $89\%$ on ART [6]. Although a proportion of these would have initiated ART at TB diagnosis, with an estimated $62.3\%$ people living with HIV (PLWH) on ART in South Africa 2017 [36], the HIV programme plays a significant role in reaching PLWH who may have TB. Although, the sample was small, in this analysis, among those diagnosed with TB, unknown HIV status, and cigarette smoking were predictors of delayed healthcare seeking for TB symptoms. Therefore, while focus should remain on people LWH given their significantly higher TB risk, attention is also needed on those without HIV, those with unknown status and those with other risk factors including smoking cigarettes, given conditions that enable TB transmission and acquisition in many communities [6]. Healthcare workers managing other health conditions should also routinely screen their patients for TB. People who had with TB In the past were more likely to seek care and this was also previously shown in Zambia [37]. This is a notable finding since individuals with a prior history of TB have a greater risk of developing active TB, and their healthcare seeking behavior could reflect better understanding of the significance of TB symptoms. It is important that such individuals receive prompt and appropriate testing and care. This survey was undertaken before the COVID-19 pandemic which had an impact on healthcare seeking for TB. The COVID-19 pandemic led to reduced TB testing because of various reasons including travel restrictions, restrictions on building occupancy including healthcare facilities, disruption of routine activities, and people’s fear of contracting COVID-19 while attending healthcare facilities [6, 8]. Therefore, healthcare seeking patterns post the peak of the COVID-19 period could be different from those observed at the time of the survey. Desirability bias may also affect the reasons reported for not seeking healthcare in particular since the majority of participants reported that they intended to seek care. We did not collect data on the exact duration of symptoms to determine the extent of delays and further examine the intention to seek care. However, our findings are consistent with the literature indicating that people delay seeking healthcare for TB symptoms [11, 13, 38]. In a Kenyan, survey similar to ours, among survey participants with at least one symptom who did not seek healthcare and provided a reason for this, $82\%$ felt their symptoms were not serious to warrant care [38]. Desirability bias could have also affected reporting of other barriers namely health system barriers and use of other medications. Our data on alcohol intake and cigarette smoking was limited, although our findings are consistent with the literature [39–41]. Given the potential complex and causal pathways and confounding factors from symptoms to care seeking (that we did not investigate), the adjusted odd ratios cannot necessarily be interpreted in the same manner [42]. ## Conclusion People with TB symptoms delayed seeking healthcare for them: many regarded the TB symptoms as benign while others faced travel and cost access barriers. Those with unknown HIV status were significantly less likely to seek care. Strengthening community-based TB awareness and screening programmes together TB self-screening models could increase awareness of the significance of TB symptoms, minimize travel and cost access barriers, and prompt early healthcare seeking and thus enable many people with TB in communities to enter the TB care cascade. These programmes and models should be strengthened together with clear and efficient patient management pathways for those who screen positive and require further testing management in healthcare facilities. ## References 1. 1World Health Organisation The END TB strategy. https://apps.who.int/iris/bitstream/handle/10665/331326/WHO-HTM-TB-2015.19-eng.pdf?sequence=1&isAllowed=y Accessed 30 January 2023. 2. 2United Nations. Transforming our world: the 2030 Agenda for Sustainable Development. https://www.un.org/sustainabledevelopment/ Accessed 30 January 2023.. Transforming our world: the 2030 Agenda for Sustainable Development 3. 3WHO consolidated guidelines on tuberculosis. Module 3: diagnosis—rapid diagnostics for tuberculosis detection, 2021 update. Geneva: World Health Organization; 2021. Licence: CC BY-NC-SA 3.0 IGO.. Module 3: diagnosis—rapid diagnostics for tuberculosis detection, 2021 update (2021.0) 4. 4WHO consolidated guidelines on tuberculosis. Module 4: treatment—drug-susceptible tuberculosis treatment. Geneva: World Health Organization; 2022. Licence: CC BY-NC-SA 3.0 IGO.. Module 4: treatment—drug-susceptible tuberculosis treatment (2022.0) 5. 5World Health Organization. Rapid communication: key changes to the treatment of drug resistant tuberculosis. Geneva, Switzerland: WHO, 2022.. Rapid communication: key changes to the treatment of drug resistant tuberculosis (2022.0) 6. 6World Health Organization, Global tuberculosis report 2021. World Health Organization Geneva; 2021. Licence: CCBY-NC-SA3.0IGO. 7. Naidoo P, Theron G, Rangaka MX, Chihota VN, Vaughan L, Brey ZO. The South African tuberculosis care cascade: estimated losses and methodological challenges. J Infect Dis (2017.0) 216 S702-13. DOI: 10.1093/infdis/jix335 8. Ismail N, Moultrie H. Impact of COVID-19 intervention on TB testing in South Africa.. National Institute for Communicable Diseases, Centre for Tuberculosis 9. 9WHO consolidated guidelines on tuberculosis. Module 2: screening–systematic screening for tuberculosis disease. Geneva: World Health Organization; 2021. Licence: CCBY-NC-SA3.0 IGO. Module 2: screening–systematic screening for tuberculosis disease (2021.0) 10. Moyo S, Ismail F, Van der Walt M, Ismail N, Mkhondo S, Dlamini SS. Prevalence of bacteriologically confirmed pulmonary tuberculosis in South Africa, 2017–19: a multistage, cluster-based, cross-sectional survey.. Lancet Infect Dis 22 1172-1180. DOI: 10.1016/S1473-3099(22)00149-9 11. 11WHO consolidated guidelines on tuberculosis. Module 1: prevention–tuberculosis preventive treatment. Geneva: World Health Organization; 2020. Licence: CC BY-NC-SA3. 0IGO.. Module 1: prevention–tuberculosis preventive treatment (2020.0) 12. Makgopa S, Madiba S. Tuberculosis Knowledge and Delayed Health Care Seeking Among New Diagnosed Tuberculosis Patients in Primary Health Facilities in an Urban District, South Africa. Health Serv Insights (2021.0) 14 11786329211054035. DOI: 10.1177/11786329211054035 13. Getnet F, Demissie M, Assefa N, Mengistie B, Worku A. Delay in diagnosis of pulmonary tuberculosis in low-and middle-income settings: systematic review and meta-analysis. BMC Pulm Med (2017.0) 17 202. DOI: 10.1186/s12890-017-0551-y 14. Naidoo P, Simbayi L, Labadarios D, Ntsepe Y, Bikitsha N, Khan G. Predictors of knowledge about tuberculosis: results from SANHANES I, a national, cross-sectional household survey in South Africa.. BMC Public Health (2016.0) 16 276. DOI: 10.1186/s12889-016-2951-y 15. Simbayi LC, Zuma K, Zungu N, Moyo S, Marinda S. South African National HIV Prevalence, Incidence. Behaviour and Communication Survey (2019.0) 16. 16National Department of Health South Africa. https://sacoronavirus.co.za/2021/02/05/mkhize-highlights-the-way-forward-for-sas-tb-response/ 17. McCreesh N, Karat AS, Govender I, Baisley K, Diaconu K, Yates TA. Estimating the contribution of transmission in primary healthcare clinics to community-wide TB disease incidence, and the impact of infection prevention and control interventions in KwaZulu-Natal, South Africa.. BMJ Glob Health (2022.0) 7 e007136. DOI: 10.1136/bmjgh-2021-007136 18. Harris B, Goudge J, Ataguba JE, McIntyre D, Nxumalo N, Jikwana S. Inequities in access to health care in South Africa.. J Public Health Policy (2011.0) 32 S102-23. DOI: 10.1057/jphp.2011.35 19. Chinyakata R, Roman NV, Msiza FB. Stakeholders’ Perspectives on the Barriers to Accessing Health Care Services in Rural Settings: A Human Capabilities Approach. The Open Public Health Journal (2021.0) 14 336-44. DOI: 10.2174/1874944502114010336 20. 20National Department of Health South Africa. 2020. Policy Framework and Strategy for Ward-based Primary Healthcare Outreach Teams 2018/19–2023/24. https://www.health.gov.za/wp-content/uploads/2020/11/policy-wbphcot-4-april-2018_final-copy.pdfhttps://hsrcacza-my.sharepoint.com/personal/smoyo_hsrc_ac_za/Documents/updateSep2022/TBsurvey/2022/papers/Careseekingreponsetocomments/30Jan/Final/Checksforfinalsubmission/f Accessed 25 November 2022.. Policy Framework and Strategy for Ward-based Primary Healthcare Outreach Teams 2018/19–2023/24 (2020.0) 21. Goudge J, Gilson L, Russell S, Gumede T, Mills A. The household costs of health care in rural South Africa with free public primary care and hospital exemptions for the poor. Trop Med Int Health (2009.0) 14 458-467. DOI: 10.1111/j.1365-3156.2009.02256.x 22. McLaren ZM, Ardington C, Leibbrandt M. Distance decay and persistent health care disparities in South Africa.. BMC Health Serv Res (2014.0) 14 541. DOI: 10.1186/s12913-014-0541-1 23. Musakwa NO, Bor J, Nattey C, Lönnermark E, Nyasulu P, Long L. Perceived barriers to the uptake of health services among first-year university students in Johannesburg, South Africa.. PLoS ONE (2021.0) 16 e0245427. DOI: 10.1371/journal.pone.0245427 24. 24National Department of Health, South Africa. National Policy on Management of Patient Waiting Time in Out Patient Departments. https://www.knowledgehub.org.za/system/files/elibdownloads/2019-07/Patient%2520Waiting%2520time%2520Policy%252014%2520November%25202016%2520PDF.pdf Accessed 3 October 2022.. National Policy on Management of Patient Waiting Time in Out Patient Departments 25. 25National Department of Health South Africa. Central Chronic Medicines Dispensing and Distribution (CCMDD) https://www.health.gov.za/ccmdd/ Accessed 3 October 2022. 26. 26South African National Department of Health (2011) HIV & AIDS, STI AND TB Management policy for the public service. https://www.knowledgehub.org.za/system/files/elibdownloads/2020-04/HIV%26AIDS%20AND%20TB%20MANAGEMENT%20POLICY.pdf Accessed 30 January 2023. 27. 27World Health Organization 2019, Multisectoral Accountability Framework to accelerate progress to end TB by 2030. https://apps.who.int/iris/bitstream/handle/10665/331934/WHO-CDS-TB-2019.10-eng.pdf Accessed 25 November 2022. 28. Bresenham D, KippAM Medina-Marino A. Quantification and correlates of tuberculosis stigma along the tuberculosis testing and treatment cascades in South Africa: a cross-sectional study. Infect Dis Poverty (2020.0) 9 145. DOI: 10.1186/s40249-020-00762-8 29. Bajema KL, Kubiak RW, Guthrie BL, Graham SM, Govere S, Thulare H. Tuberculosis-related stigma among adults presenting for HIV testing in KwaZulu-Natal, South Africa.. BMC Public Health. (2020.0) 20 1338. DOI: 10.1186/s12889-020-09383-0 30. Sullivan MC, Rosen AO, Allen A, Benbella D, Camacho G, Cortopassi AC. Falling short of the first 90: HIV stigma and HIV testing research in the 90–90–90 Era. AIDS Behav (2020.0) 24 357-362. DOI: 10.1007/s10461-019-02771-7 31. Madiba S, Ralebona E, Lowane M. Perceived stigma as a contextual barrier to early uptake of HIV testing, treatment initiation, and disclosure; the case of patients admitted with AIDS-Related Illness in a rural hospital in South Africa.. Healthcare (2021.0) 9 962. DOI: 10.3390/healthcare9080962 32. van den Hof S., Antillon Najlis C., Bloss E., Straetemans M.. A systematic review on the role of gender in tuberculosis control. Challenge TB. The Hague (2010.0) 33. Onifade DA, Bayer AM, Mantoya R, Haro M, Alva J, Franco J. Gender factors influencing tuberculosis control in shanty towns: a qualitative study. BMC Public Health (2020.0) 10 381. DOI: 10.1186/1471-2458-10-381 34. Chikovore J, Hart G, Kumwenda M, Chipungu GA, Corbett L. ’For a mere cough, men must just chew Conjex, gain strength, and continue working’: the provider construction and tuberculosis care-seeking implications in Blantyre, Malawi.. Glob Health Action (2015.0) 8 26292. DOI: 10.3402/gha.v8.26292 35. Daniels J, Medina-Marino A, Glockner K, Grew E, Ngcelwane N, Kipp A. Masculinity, resources, and retention in care: South African men’s behaviors and experiences while engaged in TB care and treatment.. Soc Sci Med. (2021.0) 270 113639. DOI: 10.1016/j.socscimed.2020.113639 36. Marinda E, Simbayi L, Zuma K, Zungu N, Moyo S, Kondlo L. Towards achieving the 90-90-90 HIV targets: results from the South African 2017 national HIV survey.. BMC Public Health (2020.0) 20 1375. DOI: 10.1186/s12889-020-09457-z 37. Wademan DT, Mainga T, Gondwe M, Ayles H, Shanaube K, Mureithi L. ’TB is a disease which hides in the body’: Qualitative data on conceptualisations of tuberculosis recurrence among patients in Zambia and South Africa.. Glob Public Health (2022.0) 17 1713-1727. DOI: 10.1080/17441692.2021.1940235 38. Enos M, Sitienei J, Ong’ang’o J, Mungai B, Kamene M, Wambugu J. Kenya tuberculosis prevalence survey 2016: Challenges and opportunities of ending TB in Kenya.. PLoS One. (2018.0) 13 e0209098. DOI: 10.1371/journal.pone.0209098 39. 39US Department of Health and Human Services. The health consequences of smoking-50 years of progress: a report of the Surgeon General. Atlanta, GA: USA. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health, 2014. https://www.ncbi.nlm.nih.gov/books/NBK179276/ Accessed on 30 January 2023.. The health consequences of smoking-50 years of progress: a report of the Surgeon General (2014.0) 40. Lönnroth K, Williams BG, Stadlin S, Jaramillo E, Dye C. Alcohol use as a risk factor for tuberculosis–a systematic review. BMC Public Health (2008.0) 8 289. DOI: 10.1186/1471-2458-8-289 41. 41World Health Organization Global Tuberculosis Report 2022. Geneva: World Health Organization; 2022. Licence: CC BY-NC-SA3.0IGO. 42. Westreich D, Greenland S. The Table 2 Fallacy: Presenting and Interpreting Confounder and Modifier Coefficients. American Journal of Epidemiology (2013.0) 4 292-298. DOI: 10.1093/aje/kws412
--- title: Better sleep, better life? testing the role of sleep on quality of life authors: - Michaela Kudrnáčová - Aleš Kudrnáč journal: PLOS ONE year: 2023 pmcid: PMC10016705 doi: 10.1371/journal.pone.0282085 license: CC BY 4.0 --- # Better sleep, better life? testing the role of sleep on quality of life ## Abstract Previous research has shown that sleep deprivation, low quality sleep or inconvenient sleeping times are associated with lower quality of life. However, research of the longitudinal effects of sleep on quality of life is scarce. Hence, we know very little about the long-term effect of changes in sleep duration, sleep quality and the time when individuals sleep on quality of life. Using longitudinal data from three waves of the Czech Household Panel Study (2018–2020) containing responses from up to 4,523 respondents in up to 2,155 households, the study examines the effect of changes in sleep duration, sleep quality and social jetlag on satisfaction with life, happiness, work stress, subjective health and wellbeing. Although sleep duration and timing are important, panel analyses reveal that sleep quality is the strongest predictor of all sleep variables in explaining both within-person and between-person differences in quality of life indicators. ## Introduction Previous research has shown that sleeping patterns are related to quality of life (QoL) and that key aspects are the time when individuals sleep, sleep duration and sleep quality. People who obtain sufficient high-quality sleep at proper times were found to have better general health [1] and overall quality of life [2]. By contrast, individuals who sleep too much [2] or sleep poorly [3] exhibit diminished quality of life. Despite previous research on QoL and sleep being substantial, they often lack in depth and scope and we know little about the effects of these three aspects of sleep on QoL and the development of their influence over time, which are significant considerations. Using three waves of the Czech Household Panel Study data, the present study contributes to the literature by examining the effect of sleep duration, sleep quality and social jetlag on five QoL indicators and exploring the trends in time. ## Quality of life definition Originally, high QoL was perceived as a lack of stress, but the idea evolved into a multidimensional concept which emphasizes the subjectivity of experience, function and wellbeing and encompasses the physical, psychological and social domains of life [4]. QoL is an interplay between the perception of an internal state, such as the experience of happiness or feeling of good health or satisfaction, and external events in the surrounding environment, which may include family and career [5]. The model in the present study was built according to the theoretical model of QoL by Ventegodt et al. [ 6]. The model comprises various parameters grouped into three complementary categories, each being concerned with an aspect of good life: subjective, existential and objective. The above-mentioned authors incorporated notions of QoL into an integrative quality-of-life (IQOL) theory. We base our analysis on the subjective component of this all-embracing theory, which includes the following parameters: wellbeing, satisfaction with life, happiness and meaning in life (Fig 1). Fig 1: Subjective quality of life according to integrative quality-of-life (IQOL) theory.Note: Modified model of Søren Ventegodt et al. (2003:1032) integrative quality-of-life theory. The indicators for the five dimensions of quality of life refer to the indicators used in the Czech Household Panel Survey. These IQOL parameters are intertwined and crucial factors in describing QoL [4]. For instance, subjective wellbeing might be characterized as an emotional response and evaluation of satisfaction with life [7] which includes both cognitive judgments and affective reactions [4]. Since wellbeing captures a person’s emotional state and touches on their mental state, our interpretation regards these states as complementary to subjective health, which more straightforwardly encompasses physical aspects. While happiness could be described as a person’s current positive emotional condition [8], satisfaction with life represents a stable assessment of general feelings about life and indicates a long-term attitude [8]. Work also forms an important part of life, contributing to its meaning [6]. Although work can be exciting and satisfying, it may also be a cause of stress. Work stress refers to a negative psychological state which may involve numerous conditions in the working environment and consists of an interplay of cognitive, affective and physiological reactions functioning as stressors [9]. Stress causes the anatomic nervous system to release the hormone cortisol, which commonly aids in regulating sleep cycles. At elevated levels, however, cortisol results in sleep disturbances and insomnia [10]. Insufficient, excess, poor or otherwise impaired sleep, especially in the long-term, is concerning since it may result in severe physical, mental and social consequences in quality of life. ## Quality of life and its relationship to sleep According to Repair and Restoration theory (RRT), sufficient sleep rewards us with restoration and repair that no other physiological process is able to achieve [11]. After a good night’s sleep, individuals feel mentally sharp and rested. Research on body functioning also suggests that muscle repair, tissue growth and many other essential processes occur primarily during sleep [12], thereby affecting wellbeing and QoL. By contrast, insufficient sleep and accumulated sleep debt impairs mental function [13] and leads to health problems, including depression [14], obesity [15], diabetes and cardiovascular disease [16], increases the risk of cancer and reduces life expectancy [17]. IQOL and RRT theories and strong empirical evidence indicate that sleep affects QoL. Not only that sleep, in theory, restores the body and elevates the mind, studies have confirmed that sleep predicts quality of life, not the opposite [18, 19]. Previous research suggests three aspects of sleep are related to QoL: sleep duration, sleep quality and social jetlag. ## Sleep duration Sleep duration is a reliable predictor of wellbeing [18] and affects QoL. A systematic review and meta-analysis by Cappuccio et al. [ 20] found that both too short and too long periods of sleep lead to elevated mortality. There is, however, no agreement in the literature on what is normal, short or long sleep duration, each study used different cut-off points. This is also a reason why our study relates only to relative time spent sleeping (less or more hours in comparison to other respondents). A longitudinal study of 1,601 Swiss and Norwegian adolescents concluded that longer sleep duration is associated with higher levels of wellbeing [18]. In another study of adolescents ($$n = 4$$,582), shorter sleep duration was related to a lower level of happiness [21]. Ness and Saksvik-Lehouillier [22] surveyed 474 Norwegian university students and concluded that longer average sleep duration is associated with greater life satisfaction. However, some studies, such as a two-decade old experiment involving 75 university students who maintained sleep logs for three seven-day periods over three months and subsequently took part in a survey [23], claim that sleep quantity does not contribute to wellbeing. Two recent studies drawing on the German Socio-Economic Panel separately investigated sleep duration on workdays and weekends: Pagan [24] observed a sample of 105,340 individuals with disabilities for six years (2008–2013) and concluded that longer sleep duration on workdays increases life satisfaction. Piper [25] explored a sample of 68,782 individuals from the same panel data (2008–2012) and found that life satisfaction increases with longer sleep duration during workdays but not on weekends. In a study of 547 university students, Önder [26] found no correlation between sleep duration and happiness. However, the reliability of these conclusions is debatable since they were both based on small student samples, and the Turkish study involved mainly women ($80.4\%$). Similarly, a longitudinal two-year study of 1139 Chinese university students indicated that sleep duration does not predict QoL [27]. Besides sleep duration, sleep quality is also related to wellbeing [1, 22, 23] and overall QoL [28–30]. ## Sleep quality Although sleep quality is often considered affecting QoL more than sleep duration, they are not usually investigated together, the focus being solely on sleep quality. One notable example used a representative Austrian sample of 1,049 people and showed the significance of the relationship between sleep quality and QoL [30]. Research based on representative samples is scarce, and studies have principally involved student samples or patients. Poorer sleep was found to be associated with adverse effects and significantly lower levels of happiness [21] and life satisfaction among Norwegian [22] and Korean students [19]. The above-mentioned small-scale experiment by [23] on college students in the US revealed no effect of sleep quantity on QoL but found sleep quality to be a strong and consistent long-term predictor of QoL. In an experiment on the interaction of sleep with campus residence and its effect on wellbeing, the authors of a Chinese study of university students concluded that overall sleep quality deteriorated over time and that sleep had no significant effect on QoL [27]. Students are often used in experiments for their accessibility, but the general applicability of the results of studies on these samples is limited. Students are young, do not work in full-time employment, and their physiological and life characteristics differ from the general population. Other studies often use specific populations such as patients, the elderly or workers in certain heavy industries. In a study of a specific adult and mostly male population of 145 patients diagnosed with schizophrenia, the conclusions resembled other reports in that poor sleepers tend to report lower QoL [29]. A longitudinal two-year Australian study of a sample of 93 adults with autism similarly concluded that poor sleep quality predicted poor QoL [31]. Jean-Louis et al. [ 1] collected sleep data on 273 adult San Diego residents (aged 40–64 years); their investigation revealed that self-perceived sleep quality is associated with wellbeing. Another cross-sectional study researched 435 female shift-working nurses in Taiwan and also concluded that poor sleep quality in the sample resulted in poorer life quality [28]. Disrupted sleep and therefore low-quality sleep, was also found to decrease QoL and increase work stress in a sample of 35,932 Korean workers [10]. ## Social jetlag Previous studies have shown that sleep duration and sleep quality are crucial variables in predicting QoL. However, the time when individuals sleep is often overlooked. People must adjust the time when they sleep to social arrangements which do not often agree with their intrinsic preferences. This misalignment between our social and internal biological rhythms leads to social jetlag, which has previously been found to relate to QoL [32, 33]. The relationship between social jetlag and QoL is understudied, and the results of studies are inconsistent. Only two small-scale studies have been conducted on student samples, finding no link between social jetlag and QoL [26, 34]. Other studies have reported a negative correlation between social jetlag and QoL [35]. ## Summary of previous research With the exception of some studies which used longitudinal data [18, 23–25, 31, 34], the majority of studies are cross-sectional [e.g., 22, 33, 36] and hence, a deficit in longitudinal panel studies exists. Only two studies exploring the effect of sleep variables on the quality of life are nationally representative [25, 30], while the remainder of studies were conducted on either a few dozen [23] or few hundred [e.g., 1, 22, 37] individuals and mainly examined specific populations, such as adolescents [19, 21, 34], university students [e.g., 19, 21, 22, 37], people with disabilities [24], people with autism [31] or patients diagnosed with schizophrenia [29]. Although Lau et al. [ 27] concluded that social jetlag predicted QoL, caution is required in interpreting their results. Their claim that social jetlag is reflected in perceived poorer sleep and impaired wellbeing is problematic, and their results are therefore debatable. The only accepted method of measuring social jetlag is the computation model developed by Roenneberg et al. [ 38]. Even though some studies have explored two aspects of sleep, for example, sleep duration and sleep quality [e.g., 21–23], or sleep duration and social jetlag [26], none have incorporated all three aspects (sleep duration, sleep quality, social jetlag), and hence, we have insufficient knowledge of the relative importance of the three most important sleep characteristics on QoL. Based on the IQOL and RRT theories and the previous literature and considering the analytical methods allowing us to observe relative in-between and within differences, we formulated the following hypotheses on the role of sleep in QoL: ## Study design and participants The analyses used data from the Czech Household Panel Survey (CHPS) which focuses on mapping the living conditions and describing the dynamics of change among both Czech households and individuals in the long-term perspective [39]. These data were collected annually from 2015 until 2020, typically between the end of June and the end of October. A two-stage stratified random sampling method was applied and the design effects were further mitigated by the use of a large number of small primary sampling units. The original sampling frame from the very first data collection consisted of the Register of Census Districts and Buildings which had been transformed into an address database. Since the target population was the non-institutionalized population of the Czech Republic, all members of the sampled households were interviewed. In each of the following waves, the same members of the same households participating in the preceding wave were approached (e.g., in wave three in 2017, only participants from wave two were approached). The data are nationally representative of the adult population in CR. The retention rate of households between the first and sixth waves of data collection was $21.6\%$ on average, and the retention rate of individuals was $20.6\%$. All information regarding data collection including survey design is available in the Czech Social Science Data Archive [39]. A total of 5,132 paper-and-pencil self-administered questionnaires (SAQ) incorporating the key variables were collected from Czech adults in 2018, 2,046 in 2019, and 2,161 questionnaires in 2020. The final dataset contained responses from up to 4,523 respondents in up to 2,155 households. The significant drop in the sample between 2018 and 2019 was caused by the blood draw requirement. Sleep variables were included into the questionnaires during the waves 4–6 (2018–2020) due to the collaboration between Institute of Sociology and Institute of Physiology of the Czech Academy of Sciences at that time. They were measured according to the Munich Chronotype Questionnaire (MCTQ): some were measured, and some were computed (for more information on used variables, see the section Measures down below). Written informed consent was obtained from all respondents. The study followed the principles of the Declaration of Helsinki and was approved by Ethics Committee of the Institute for Clinical and Experimental Medicine and Thomayer Hospital in Prague (study number G-16–05–02). The data from the CHPS are widely used by researchers for secondary data analysis: for instance, studies are focusing on certain aspects of sleep, specifically chronotype assessment [40] and social jetlag in the work-family context [41], other studies explore the division of housework and relative resources [42] partnership trajectories [43], mechanisms of the reproduction of homeownership [44], voter turnout [45]. ## Measures We investigated the effect of sleep on the five dependent variables which describe QoL: life satisfaction, wellbeing, happiness, subjective health and work stress. At all points in time, life satisfaction was measured with responses to the question “All things considered, how satisfied are you with your life as a whole?” The response options were scaled from zero to ten, zero indicating “extremely dissatisfied” and ten indicating “extremely satisfied”. Many other studies have used the same items to measure life satisfaction [e.g., 46, 47]. Wellbeing was calculated as an average of three items to measure how often over the last two weeks respondents “have been cheerful and in good spirits”; “have felt calm and relaxed”; “have been active and vigorous”. The six response options with scores from one to six were “at no time”, “some of the time”, “less than half of the time”, “more than half of the time”, “most of the time”, “all of the time”. The resultant reliability estimates are acceptable (αt1 =.811; αt2 =.828 αt3 =.830; αt4 =.841 αt5 =.825). The scale was computed as a sum of means also ranging from one to six. The same items were measured during two out of three analysed years of data collection in 2018 and 2019 and have also been used to measure wellbeing in other studies [e.g., 48]. Perceived happiness was measured with the question “Taking all things together, how happy would you say you are?”. The respondents were asked to answer on a scale of zero to ten, zero indicating “extremely unhappy” and ten indicating “extremely happy”. The same items were measured during two out of three analysed years of data collection in 2018 and 2019 and have also been used to measure happiness in other studies [e.g., 47, 49]. Respondents rated their subjective health according to the question “In general, would you say your health is…?” on a five-point scale of “poor”, “fair”, “good”, “very good” and “excellent”. The same items were measured during two out of three analysed years of data collection in 2018 and 2019 and have also been used to measure subjective health in other studies [e.g., 47, 49]. The respondents’ perceived work stress was calculated according to the proportion of affirmative answers to the question “Have the following circumstances in your current job caused you excess worry or stress in the past 12 months?” according to the following items: “threat of layoffs or losing the job”; “workplace safety, accidents, or injuries on the job”; too many demands or too many working hours at work.” The response options were “yes” or “no”. The same items were measured during one wave [2018] during the reference period. The questions are proxies inspired by the European Working Conditions Surveys (EWCS). In addition to the dependent variables, three facets of sleep were measured. Specifically, we assessed the average sleep duration, perceived sleep quality, and social jetlag. Sleep duration was calculated as the average of answers to questions regarding the time when respondents usually fell asleep and woke up on free days and when they usually fell asleep and woke up on workdays. The same items were measured during the complete analyzed period: wave 4 [2018], wave 5 [2019] and wave 6 [2020] of data collection and have also been used to measure sleep quality in other studies [e.g., 33, 40]. Perceived quality of sleep was measured with the question “How would you rate the quality of your sleep?” according to a four-point Likert scale for the response options “very bad”, “bad”, “good” and “very good”. The same items were measured during the complete analyzed period: wave 4 [2018], wave 5 [2019] and wave 6 [2020] of data collection and have also been used to measure sleep quality in other studies [e.g., 50, 51]. Social jetlag was calculated according to a MCTQ [52] as the difference between the mid-sleep time on free days and workdays. The resultant values were converted into numeric variables which represented the hours. The results were interpreted as follows: zero indicated no sleep debt during workdays or free days, and any values above zero indicated an accumulation of sleep debt. The same items were available during the complete analyzed period: wave 4 [2018], wave 5 [2019] and wave 6 [2020] of data collection and have also been used to measure social jetlag in other studies [e.g., 14, 33, 34]. Data on age, gender, highest level of education attained (basic and secondary vocational, secondary with maturita, tertiary education), net household income (Net household income is stated in Czech Crowns (CZK). For illustration, according to the European Union–Statistics on Income and Living Conditions (EU-SILC), the average monthly net income of a Czech household reached CZK 17.5 thousand per person in 2019 [53]. The net household income categories used in this article can be roughly converted to USD as it follows: 1 = up to 918 USD, 2 = 918 to 1,197 USD, 3 = 1,198 to 1,396 USD, 4 = 1,397 to 1,596 USD, 5 = 1,597 to 2,993 USD, 6 = more than 2,994 USD.) ( 1 = up to CZK 22,999, 2 = CZK 23,000 to 29,999, 3 = CZK 30,000 to 34,999, 4 = CZK 35,000 to 39,999, 5 = CZK 50,000 to 74,999, 6 = more than CZK 75,000), number of children below the age of five in the household, and economic status were also collected and controlled for ($45.80\%$ employed, $6.20\%$ self-employed, $2.90\%$ unemployed, $8.88\%$ students, $33.07\%$ retired, and $3.14\%$ on maternity leave). The descriptive statistics for all variables used in our analyses is reported in Table 1. Table 1 \| Unnamed: 0 \| Number of respondents \| Mean \| Std. Dev. \| Min \| Max \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Gender (2018, 2019, 2020) \| 4523 \| 1.58 \| 0.49 \| 1.0 \| 2.0 \| \| Education (2018, 2019, 2020) \| 4523 \| 1.95 \| 0.76 \| 1.0 \| 3.0 \| \| Age (2018, 2019, 2020) \| 4523 \| 51.93 \| 16.766 \| 18.0 \| 96.0 \| \| Household income (2018, 2019, 2020) \| 4523 \| 3.79 \| 1.78 \| 1.0 \| 6.0 \| \| Economic status (2018, 2019, 2020) \| 4523 \| 1.79 \| 1.92 \| 0.0 \| 5.0 \| \| Social jetlag (2018, 2019, 2020) \| 4523 \| 0.87 \| 0.87 \| 0.0 \| 5.83 \| \| Sleep duration (2018, 2019, 2020) \| 4523 \| 7.48 \| 1.12 \| 3.5 \| 13.48 \| \| Quality of sleep (2018, 2019, 2020) \| 4523 \| 3.0 \| 0.68 \| 1.0 \| 4.0 \| \| Children below the age of 5 (2018,2020) \| 4523 \| 0.18 \| 0.47 \| 0.0 \| 2.0 \| \| Life satisfaction (2018, 2019, 2020) \| 4523 \| 7.47 \| 1.79 \| 0.0 \| 10.0 \| \| Wellbeing (2018, 2019) \| 3850 \| 4.08 \| 0.92 \| 1.0 \| 6.0 \| \| Subjective health (2018, 2019) \| 3867 \| 3.12 \| 1.0 \| 1.0 \| 5.0 \| \| Working stress (2018, 2020) \| 2097 \| 0.19 \| 0.24 \| 0.0 \| 1.0 \| \| Happiness (2018, 2019) \| 3857 \| 7.34 \| 1.77 \| 0.0 \| 10.0 \| ## Statistical analysis strategy To test our hypotheses on the effects of the three measured aspects of sleep on life satisfaction, wellbeing, happiness, subjective health and work stress, we analysed the CHPS data according to mixed, multilevel repeated measurement models with random intercepts for individuals, households and a random slope for time. To examine whether sleep quality, sleep duration and social jetlag would predict between-person and within-person changes in the dependent variables, we constructed three-level hierarchical models with time nested within both individuals and households. The variables at the within-person level were person-mean-centred and constituted a measurement of the degree to which an individual’s characteristics changed over time. The variables at the between-person level were grand-mean-centred and tested whether and how much individuals differed from each other. We started with null models which incorporated the dependent variables without predictors to capture the variance of the dependent variables (S1 Table). Next, we examined the longitudinal effects of the three tested facets of sleeping hygiene on the five measures of QoL by adding sleep duration, sleep quality and social jetlag variables and interaction terms for time and sleeping variables (Models 1A–5A). In the final step, Models 1B–5B decomposed the effects of sleeping on within-person and between-person effects. We then evaluated the model fits according to the general principle that models with lower deviance and AIC values than the null model are considered better fitting models [54]. ## Results Initially, we built models without predictors to examine the variance in all five of the measured aspects of quality of life. These null models (S1 Table) showed $47\%$ variance in life satisfaction between individuals and $23\%$ variance between households, $56\%$ variance in wellbeing between individuals and $20\%$ variance between households, $74\%$ variance in subjective health between individuals and $26\%$ variance between households, $51\%$ variance in working stress between individuals and $12\%$ variance between households, $56\%$ variance in happiness between individuals and $23\%$ variance between households. ## Do changes in sleep affect the quality of life over time? To test the effect of sleep over time, we added sleep duration, sleep quality, social jetlag, control variables, the fixed effect of time and interaction term for time, and each of the three variables which capture sleeping (Table 2: Models 1A–5A). The variables improved model fit in all models (life satisfaction: Δ-2LL = 247.68 [16], $p \leq .001$; ΔAIC = 215.68; wellbeing: Δ-2LL = 404.70 [16], $p \leq .001$; ΔAIC = 372.70; health: Δ-2LL = 1307.65 [16], $p \leq .001$; ΔAIC = 1275.65; work stress: Δ-2LL = 106.62 [16], $p \leq .001$; ΔAIC = 74.62; happiness: Δ-2LL = 296.54 [16], $p \leq .001$; ΔAIC = 262.54). Table 2 \| Unnamed: 0 \| Unnamed: 1 \| Model 1A \| Model 1A.1 \| Model 2A \| Model 2A.1 \| Model 3A \| Model 3A.1 \| Model 4A \| Model 4A.1 \| Model 5A \| Model 5A.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| \| Life satisfaction \| Life satisfaction \| Wellbeing \| \| Subjective health \| Subjective health \| Work stress \| Work stress \| Happiness \| Happiness \| \| Interaction terms \| \| \| \| \| \| \| \| \| \| \| \| \| \| Sleep durationtime \| −.02 \| (−.07 - .04) \| .09 \| (.04 - .14) \| .06 \| (.02 - .10) \| < .01 \| (−.02 - .01) \| .15*** \| (.06 - .24) \| \| \| Sleep qualitytime \| .05 \| (−.04 - .14) \| −.02 \| (−.09 - .06) \| −.03 \| (−.10 - .04) \| .01 \| (−.01 - .02) \| −.11 \| (−.26 - .03) \| \| \| Social jetlagtime \| −.02 \| (−.09 - .06) \| −.06* \| (−.12 - −.01) \| −.04 \| (−.09 - .01) \| < .01 \| (−.02 - .01) \| .02 \| (−.09 - .13) \| \| Sleep variables \| \| \| \| \| \| \| \| \| \| \| \| \| \| Sleep duration \| .06 \| (−.14 - .26) \| −.30* \| (−.46 - −.15) \| −.23 \| (−.37 - −.09) \| < .01 \| (−.04 - .05) \| −.56* \| (−.85 - −.27) \| \| \| Sleep quality \| .32 \| (−.01 - .65) \| .45* \| (.19 - .70) \| .47* \| (.24 - .69) \| −.06 \| (−.12 - .01) \| .99* \| (.52–1.47) \| \| \| Social jetlag \| .02 \| (−.25 - .30) \| .20 \| (−.01 - .40) \| .13 \| (−.05 - .30) \| .03 \| (−.02 - .09) \| −.12 \| (−.49 - .26) \| \| Socio-demographic variables \| \| \| \| \| \| \| \| \| \| \| \| \| \| Time \| .01 \| (−.46 - .47) \| −.53** \| (−.93 - −.13) \| −.33 \| (−.68 - .03) \| < .01 \| (−.10 - .10) \| −.69 \| (-1.43 - .05) \| \| \| Gender (ref. cat. male) \| .08 \| (−.02 - .19) \| .07* \| (.02 - .13) \| .09** \| (.03 - .14) \| −.01 \| (−.03 - .01) \| .14* \| (.03 - .25) \| \| \| Education–secondary with maturita \| .10 \| (−.03 - .23) \| −.03 \| (−.10 - .04) \| .14* \| (.07 - .21) \| −.03 \| (−.05 - .00) \| .08 \| (−.05 - .22) \| \| \| Education–tertiary \| .09 \| (−.06 - .25) \| −.02 \| (−.10 - .06) \| .26* \| (.18 - .34) \| −.04 \| (−.07 - −.01) \| .15 \| (−.01 - .31) \| \| \| Age \| .01 \| (.00 - .02) \| < .01 \| (−.01 - .00) \| −.02* \| (−.02 - −.02) \| <−.01 \| (−.01 - −.01) \| .01 \| (.00 - .01) \| \| \| Household income \| .14* \| (.10 - .17) \| .01 \| (−.01 - .03) \| .04* \| (.02 - .06) \| < .01 \| (−.01 - .00) \| .12* \| (.08 - .16) \| \| \| Economic status (ref. cat. employed) \| Economic status (ref. cat. employed) \| \| \| \| \| \| \| \| \| \| \| \| Self-employed \| .06 \| (−.17 - .29) \| −.09 \| (−.21 - .03) \| .05 \| (−.07 - .17) \| .02 \| (−.02 - .06) \| .02 \| (−.22 - .26) \| \| \| Unemployed \| −.31 \| (−.65 - .04) \| −.22* \| (−.40 - −.04) \| −.27 \| (−.44 - −.09) \| −.04 \| (−.12 - .03) \| −.12 \| (−.47 - .23) \| \| \| Student \| .58* \| (.30 - .87) \| .17* \| (.02 - .32) \| .24 \| (.10 - .38) \| −.10* \| (−.16 - −.04) \| .49 \| (.19 - .78) \| \| \| Retired \| .08 \| (−.12 - .29) \| < .01 \| (−.11 - .11) \| −.20* \| (−.30 - −.09) \| −.10*** \| (−.15 - −.05) \| .22* \| (.00 - .43) \| \| \| Maternity leave \| .27 \| (−.05 - .58) \| −.04 \| (−.20 - .13) \| −.03 \| (−.19 - .14) \| −.10 \| (−.17 - −.03) \| .02 \| (−.31 - .34) \| \| \| Number of children below the age of 5 \| Number of children below the age of 5 \| \| \| \| \| \| \| \| \| \| \| \| One child \| .18 \| (−.03 - .39) \| .02 \| (−.09 - .12) \| .08 \| (−.02 - .17) \| < .01 \| (−.02 - .04) \| .38* \| (.16 - .59) \| \| \| Two or more children \| .41* \| (.07 - .76) \| .06 \| (−.12 - .24) \| .16 \| (−.00 - .33) \| <-0.01 \| (−.07 - .05) \| .38* \| (.03 - .74) \| \| Constant \| \| 4.82* \| (3.06–6.57) \| 4.76* \| (3.41–6.11) \| 3.94*** \| (2.74–5.14) \| .45* \| (.09 - .82) \| 7.15* \| (4.63–9.67) \| \| \| Observations \| 4523 \| \| 3850 \| \| 3867 \| \| 2097 \| \| 3857 \| \| \| \| Households \| 2155 \| \| 2100 \| \| 2105 \| \| 1305 \| \| 2101 \| \| \| \| AIC \| 17502 \| \| 9551 \| \| 9013 \| \| -60 \| \| 14623 \| \| \| \| BIC \| 17662 \| \| 9701 \| \| 9169 \| \| 81 \| \| 14779 \| \| \| \| ICC households \| 15% \| \| 5% \| \| 7% \| \| 3% \| \| 5% \| \| \| \| ICC individuals \| 69% \| \| 94% \| \| 82% \| \| 95% \| \| 96% \| \| \| \| ll \| -8726 \| \| -4752 \| \| -4482 \| \| 55.31 \| \| -7287 \| \| The interaction of sleep duration and the time variable was a positive and statistically significant predictor of wellbeing ($B = .092$, $p \leq .001$), subjective health ($B = .060$, $$p \leq .005$$), and happiness ($B = .148$, $$p \leq .001$$). The effect of the interaction term was not a statistically predictor in the model for life satisfaction (B = −.019, $$p \leq .497$$) or work stress (B = −.003, $$p \leq .575$$). The interaction of sleep quality and the time variable was not a statistically significant predictor of any of the tested dependent variables (subjective health: B = −.029, $$p \leq .391$$; happiness: B = −.115, $$p \leq .110$$; life satisfaction: $B = .050$, $$p \leq .268$$; wellbeing: B = −.017, $$p \leq .652$$; work stress: $B = .006$, $$p \leq .562$$). The interaction of social jetlag and the time variable was a negative and statistically significant predictor of wellbeing (B = −.062, $$p \leq .042$$), but not a statistically significant predictor in the models for subjective health (B = −.041, $$p \leq .136$$), happiness ($B = .022$, $$p \leq .700$$), life satisfaction (B = −.016, $$p \leq .678$$) or work stress (B = −.005, $$p \leq .536$$). A graphical representation of the calculated marginal effects highlighted the differences in QoL between individuals who slept fewer or more hours on average (Fig 2), perceived their sleep to be worse or better quality (Fig 3), and suffered from less or more social jetlag (Fig 4), whereas other variables remained at their mean values. Fig 2:** Sleep duration and quality of life at the individual level in time.Note: 95% confidence intervals. Fig 3: Sleep quality and quality of life at the individual level in time.Note: 95% confidence intervals. Fig 4: Social jetlag and quality of life at the individual level in time.Note: 95% confidence intervals. ## Does sleep predict within-person and between-person changes in quality of life? Further examination of the longitudinal effect of sleep on quality of life in Models 1B–5B (Table 3) distinguish the discussed between-person and within-person effects. Separation of the between-person and within-person effects improved model fit in the models for predicting life satisfaction (Δ-2LL = 37.44 [1], $p \leq .001$; ΔAIC = 37.44), wellbeing (Δ-2LL = 15.25 [41], $p \leq .001$; ΔAIC = 13.25), health (Δ-2LL = 50.66 [1], $p \leq .001$; ΔAIC = 50.66) and happiness (Δ-2LL = 16.69 [1], $p \leq .001$; ΔAIC = 18.69) but did not show any statistically significant improvement in model fit for work stress (Δ-2LL = 3.184 [1], $p \leq .074$; ΔAIC = 3.18) over Models 1A–5A. Table 3 \| Unnamed: 0 \| Unnamed: 1 \| Model 1B \| Model 1B.1 \| Model 2B \| Model 2B.1 \| Model 3B \| Model 3B.1 \| Model 4B \| Model 4B.1 \| Model 5B \| Model 5B.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| \| Life satisfaction \| Life satisfaction \| Wellbeing \| Wellbeing \| Subjective health \| Subjective health \| Work stress \| Work stress \| Happiness \| Happiness \| \| Sleep variables—between person level \| \| \| \| \| \| \| \| \| \| \| \| \| \| Sleep duration \| −.02 \| (−.08 - .03) \| −.01 \| (−.04 - .02) \| −.04 \| (−.07 - −.02) \| < .01 \| (−.02 - .01) \| −.08 \| (−.14 - −.03) \| \| \| Sleep quality \| .65* \| (.56 - .74) \| .46* \| (.42 - .51) \| .47* \| (.42 - .51) \| −.04* \| (−.06 - −.03) \| .74*** \| (.65 - .84) \| \| \| Social jetlag \| −.09* \| (−.17 - −.00) \| −.02 \| (−.06 - .02) \| −.03 \| (−.07 - .01) \| .02* \| (.00 - .03) \| −.08 \| (−.17 - .00) \| \| Sleep variables—within person level \| \| \| \| \| \| \| \| \| \| \| \| \| \| Sleep duration \| −.01 \| (−.11 - .09) \| .03 \| (−.03 - .08) \| −.02 \| (−.07 - .03) \| −.02 \| (−.05 - .01) \| −.04 \| (−.15 - .06) \| \| \| Sleep quality \| .15* \| (.01 - .29) \| .18* \| (.10 - .26) \| .14* \| (.07 - .21) \| −.01 \| (−.04 - .03) \| .28*** \| (.14 - .43) \| \| \| Social jetlag \| .11 \| (−.03 - .24) \| .03 \| (−.05 - .11) \| .06 \| (−.01 - .13) \| .02 \| (−.01 - .05) \| .07 \| (−.08 - .21) \| \| Socio-demographic variables \| \| \| \| \| \| \| \| \| \| \| \| \| \| Time \| .03 \| (−.03 - .09) \| .05* \| (.00 - .10) \| < .01 \| (−.05 - .04) \| −.01* \| (−.02 - −.00) \| .09* \| (.00 - .18) \| \| \| Gender (ref. cat. male) \| .10 \| (−.00 - .21) \| .08 \| (.03 - .14) \| .10* \| (.04 - .15) \| −.01 \| (−.03 - .01) \| .15 \| (.04 - .26) \| \| \| Education–secondary with maturita \| .09 \| (−.04 - .22) \| −.02 \| (−.09 - .04) \| .14* \| (.07 - .20) \| −.03 \| (−.05 - .00) \| .08 \| (−.06 - .22) \| \| \| Education–tertiary \| .08 \| (−.08 - .23) \| −.02 \| (−.11 - .06) \| .25*** \| (.17 - .33) \| −.04* \| (−.07 - −.01) \| .14 \| (−.02 - .30) \| \| \| Age \| .01 \| (.00 - .02) \| < .01 \| (−.01 - .00) \| −.02* \| (−.02 - −.02) \| <−.01** \| (−.01 - −.01) \| .01* \| (.00 - .01) \| \| \| Household income \| .13* \| (.10 - .17) \| .01 \| (−.01 - .03) \| .04* \| (.02 - .05) \| < .01 \| (−.01 - .00) \| .12*** \| (.08 - .16) \| \| \| Economic status (ref. cat. employed) \| Economic status (ref. cat. employed) \| \| \| \| \| \| \| \| \| \| \| \| Self-employed \| .05 \| (−.18 - .28) \| −.09 \| (−.21 - .03) \| .05 \| (−.07 - .16) \| .02 \| (−.02 - .05) \| .01 \| (−.23 - .25) \| \| \| Unemployed \| −.31 \| (−.65 - .03) \| −.21* \| (−.39 - −.03) \| −.27 \| (−.44 - −.09) \| −.05 \| (−.12 - .03) \| −.12 \| (−.47 - .23) \| \| \| Student \| .55* \| (.26 - .83) \| .16* \| (.01 - .31) \| .22 \| (.08 - .37) \| −.10* \| (−.16 - −.04) \| .45 \| (.16 - .75) \| \| \| Retired \| .06 \| (−.15 - .26) \| < .01 \| (−.11 - .11) \| −.20* \| (−.31 - −.10) \| −.10* \| (−.15 - −.05) \| .20 \| (−.01 - .42) \| \| \| Maternity leave \| .25 \| (−.07 - .57) \| −.03 \| (−.20 - .13) \| −.03 \| (−.20 - .13) \| −.10 \| (−.17 - −.03) \| .01 \| (−.32 - .34) \| \| \| Number of children below the age of 5 \| Number of children below the age of 5 \| \| \| \| \| \| \| \| \| \| \| \| One child \| .18 \| (−.03 - .40) \| .02 \| (−.09 - .13) \| .08 \| (−.02 - .18) \| .01 \| (−.02 - .04) \| .38*** \| (.17 - .59) \| \| \| Two or more children \| .41* \| (.07 - .75) \| .08 \| (−.11 - .26) \| .18* \| (.01 - .34) \| −.01 \| (−.07 - .05) \| .40* \| (.05 - .76) \| \| Constant \| \| 4.47* \| (3.82–5.12) \| 2.74* \| (2.39–3.10) \| 2.71* \| (2.38–3.04) \| .50* \| (.37 - .63) \| 4.36* \| (3.67–5.05) \| \| \| Observations \| 4523 \| \| 3850 \| \| 3867 \| \| 2097 \| \| 3857 \| \| \| \| Households \| 2155 \| \| 2100 \| \| 2105 \| \| 1305 \| \| 2101 \| \| \| \| AIC \| 17464 \| \| 9538 \| \| 8962 \| \| -64 \| \| 14604 \| \| \| \| BIC \| 17625 \| \| 9694 \| \| 9118 \| \| 77 \| \| 14754 \| \| \| \| ICC households \| 14% \| \| 7% \| \| 5% \| \| 3% \| \| 10% \| \| \| \| ICC individuals \| 70% \| \| 87% \| \| 85% \| \| 95% \| \| 88% \| \| \| \| ll \| -8707 \| \| -4744 \| \| -4456 \| \| 56.91 \| \| -7278 \| \| The effects of sleep duration on subjective health (B = −.045, $$p \leq .001$$) and happiness (B = −.084, $$p \leq .003$$) were statistically significant at the between-person level. Sleep duration was not a statistically significant predictor of life satisfaction (B = −.021, $$p \leq .436$$), wellbeing (B = −.013, $$p \leq .364$$) or work stress (B = −.005, $$p \leq .424$$) at the between-person level. At the within-person level, the effects of sleep duration were not a statistically significant predictor of happiness (B = −.044, $$p \leq .405$$), wellbeing ($B = .028$, $$p \leq .335$$), subjective health (B = −.020, $$p \leq .412$$), work stress (B = −.022, $$p \leq .110$$) or life satisfaction (B = −.007, $$p \leq .890$$). The effects of sleep quality on life satisfaction ($B = .653$, $p \leq .001$), wellbeing ($B = .463$, $p \leq .001$), work stress (B = −.043, $p \leq .001$) subjective health ($B = .468$, $p \leq .001$) and happiness ($B = .742$, $p \leq .001$) were statistically significant at the between-person level. At the within-person level, the effects of sleep quality were a statistically significant predictor of life satisfaction ($B = .149$, $$p \leq .036$$), wellbeing ($B = .183$, $p \leq .001$), subjective health ($B = .143$, $p \leq .001$) and happiness ($B = .283$, $p \leq .001$), but not of work stress (B = −.009, $$p \leq .612$$). The effects of social jetlag on life satisfaction (B = −.086, $$p \leq .040$$) and work stress ($B = .017$, $$p \leq .020$$) were statistically significant at the between-person level. Social jetlag was not a statistically significant predictor of happiness (B = −.083, $$p \leq .052$$), wellbeing (B = −.019, $$p \leq .378$$) or health (B = −.032, $$p \leq .125$$) at the between-person level. At within-person level, the effects of social jetlag on life satisfaction ($B = .105$, $$p \leq .136$$), wellbeing ($B = .032$, $$p \leq .433$$), work stress ($B = .017$, $$p \leq .296$$), happiness ($B = .068$, $$p \leq .352$$) and health ($B = .059$, $$p \leq .087$$) were not statistically significant. ## Discussion The Czech Republic (CR) is comparable to other European countries in standard of living. The CR is on average commensurable with other European countries in life expectancy and economic activity [55] and self-perceived health [56]. While the life satisfaction score in the CR is very close to the European average, Czechs are slightly less happy, their happiness score being comparable to European countries such as Portugal, Italy and Greece [57]. The average sleep duration in the CR is 7.5 hours (see Data and methods section), which is similar to other European countries such as Belgium, France, Hungary, the Netherlands and the United Kingdom [58]. The proportion of Czechs ($31\%$) with social jetlag is also comparable to the European average [15]. However, although Czechs report around 49 minutes of social jetlag (see Data and methods section), Spaniards and Germans report longer times [59]. The source of this difference is unclear, but it is probably because the samples are non-representative. It may be also the result of distinct cultural and environmental contexts or locations. In summary, the CR represents a case study of a population with living standards, QoL and sleep patterns are comparable to other European countries. The findings of the present study can therefore be reasonably generalized to other countries. Building on IQOL theory and previous studies, the present study expands on the relationship between QoL and sleep. It contributes to the existing literature by examining the main areas of life and sleep from representative panel data to form a better understanding of how sleep and QoL are intertwined and the development of their relationship over time. The results of this study do not support the hypothesis (H1B) that QoL increases when people change their sleeping habits to spend more time sleeping. However, the results agree with previous studies which report a relationship between sleep duration and QoL [18, 21] from results which show differences between people in their perceived health and happiness according to the number of hours they spend sleeping (H1A). Individuals who spent more time sleeping also reported worse subjective health and lower levels of happiness. The negative association between subjective health and sleep duration may be a result of long-term stress or mental illness which have affected their sleeping habits since previous studies have shown that individuals with poor mental health and depressive symptoms report sleeping issues and also longer sleep duration [60]. The negative association between sleep duration and QoL agrees with previous findings [2, 18, 21]. In accordance with our hypotheses (H2A and H2B) and previous studies, sleep quality was found to be a robust and reliable predictor of QoL [1, 29, 30]. Our analyses show individuals who experience higher quality sleep also have greater satisfaction with life, more wellbeing, feel healthier, perceive less work stress and are happier (H2B). With changes over time, a positive association between improvement in quality of sleep and increase in life satisfaction, wellbeing, subjective health, and happiness is evident (H2A). The overall positive effect of change in sleep quality on QoL agrees with previous research [1, 10, 28, 30]. The only indicator not associated with a change in sleep quality is work stress, perhaps due to the complexity of the link between these indicators. A mediator variable which also captures emotional aspects, such as workplace relationships, might be missing [21]. These results also contribute to the ongoing debate regarding the ambiguous consequences of social jetlag on our lives. Our results agree with Jankowski [34] and Önder [26] are contrary to Chang and Jang [35]. Our hypothesis (H3A) that individuals with a higher level of social jetlag are less satisfied with life and experience a higher level of work stress than others was only partially confirmed. Our findings do not suggest any association between social jetlag and wellbeing, subjective health or happiness. Furthermore, a change in social jetlag has no effect on any measured QoL aspect (H2B). This may be due to social jetlag being relatively stable, as it is likely to change only as a consequence of a relatively major life change (new job, birth of a child) which results in a new sleep schedule. Therefore, individuals with less sleep debt experience a minor increase in various aspects of QoL, but individuals with more social jetlag stagnate, apart from experiencing a decrease in work stress. Since these changes are not very frequent, social jetlag has a low variation over time, leading to the absence of a longitudinal effect, except in work stress, which is most likely related to changes in employment arrangements. The results of the present study are consistent with previous studies [1, 22, 28] and suggest a strong relationship between sleep quality and QoL and a rather limited effect of sleep duration or social jetlag on QoL. A comparison of the respondents’ sleep quality indicated a slight improvement in happiness in those who experienced poorer sleep during the last wave [2020] of data collection. This may have been caused by an overall increase in sleep quality triggered by social lockdowns designed to suppress Covid-19. Poor sleepers also indicated a small decline in work stress, perhaps because of more flexible working arrangements experienced early during the Covid-19 pandemic. Longitudinal effects nonetheless remained stable over the previous three years, as we presumed. The results of this longitudinal study provide an important insight into people’s lifestyles. Despite people having different sleep requirements, the results suggest that both average sleep duration and social jetlag remain moderately stable over time. Sleep quality is also a valuable subjective measure related to other factors which encompass several important areas of life, such as mental and physical health, emotional wellbeing, cognitive functioning and feeling of safety. ## Limitations The strengths of our study are longitudinal design, differentiation of between-person and within-person effects and the advantage of a representative dataset which enabled the incorporation of all three aspects of sleep (quantity, quality, social jetlag) into a single model. This is also the first study which has tested the longitudinal effect of social jetlag on QoL. Admittedly, the study also has limitations. First, the period of measurement is relatively too brief to allow stronger claims regarding the longitudinal effect of sleep. Second, all the results are correlational. Using panel data does not qualify for asserting causal claims, and therefore it is not possible to state, for example, whether people feel less healthy because of low quality sleep or whether low-quality sleep leads to poorer health. Third, even though the CR is comparable to other European countries in living standard and sleeping habits, this is a case study of a single country. Having the opportunity to test our findings in other countries would be a great venue for future research. Fourth, the sleep indicators are self-reported and therefore have limitations despite self-reported measures being similarly reliable predictors [61]. Ideally, the measures would be collected in a medical lab or via mobile devices to aid in cross validating our results with more objective methods of measurement. Fifth, even though data were collected on regular days, the final wave partially captured the experience of the pandemic in the spring of 2020, and this study, therefore, might not be representative of the behavior under normal circumstances. However, data collection occurred during periods of eased restrictions and likely did not affect the generalizability of the results. ## Conclusion The present study delivers a comprehensive analysis built on previous studies to extend knowledge on the role of sleep in life. In measuring three distinct facets of sleep in a single longitudinal model, sleep quality was found to be the most influential factor affecting the five aspects of QoL (wellbeing, life satisfaction, subjective health, work stress and happiness). Individuals who experienced more quality sleep also reported better QoL. Improvement of sleep quality over time is also related to improvements in QoL. Sleep duration and social jetlag are also somewhat related to QoL, but in contrast to sleep quality, these factors do not appear significant. The study suggests, with the exception of extremes, that sleep duration alongside the differences in sleep habits on workdays and free days is not as important to QoL as what is considered a good night’s sleep. Sleep is vital to our functioning. Changes in lifestyle and psychological challenges which have either emerged or been amplified under the currently ongoing pandemic have undoubtedly affected sleeping habits. That topic, preferably in a study involving multiple points over time for a long-term comparison and sleep at non-standard times such as Covid-19 pandemic, will be the focus of future research. ## References 1. Jean-Louis G, Kripke DF, Ancoli-Israel S. Sleep and quality of well-being*. Sleep* (2000.0) 23 1115-1121. DOI: 10.1093/sleep/23.8.1k 2. Groeger JA, Zijlstra FRH, Dijk DJ. Sleep quantity, sleep difficulties and their perceived consequences in a representative sample of some 2000 British adults. J Sleep Res (2004.0) 13 359-371. DOI: 10.1111/j.1365-2869.2004.00418.x 3. Paunio T, Korhonen T, Hublin C, Partinen M, Kivimäki M, Koskenvuo M. Longitudinal study on poor sleep and life dissatisfaction in a nationwide cohort of twins. Am J Epidemiol (2009.0) 169 206-213. DOI: 10.1093/aje/kwn305 4. Diener E.. Subjective well-being. Psychol Bull (1984.0) 95 542-575. DOI: 10.1037/0033-2909.95.3.542 5. 5Mumtaz A, Maon SN, Aziz NISA. The Relationship Between Job Stress and Quality of Life Among Working Adults. In: Noordin F, Othman AK, Kassim ES, editors. Proceedings of the 2nd Advances in Business Research International Conference. Singapore: Springer; 2018. p. 389. doi: 10.1007/978-981-10-6053-3. DOI: 10.1007/978-981-10-6053-3 6. Ventegodt S, Merrick J, Andersen NJ. Quality of life theory I. The IQOL theory: an integrative theory of the global quality of life concept. ScientificWorldJournal (2003.0) 3 1030-1040. DOI: 10.1100/tsw.2003.82 7. Diener E, Suh EM, Lucas RE, Smith HL. Subjective Well-Being: Three Decades of Progress. Psychol Bull (1999.0) 125 276-302 8. Huppert FA, Baylis N, Keverne B. The Science of Well-Being. The Royal Society (2005.0). DOI: 10.1093/acprof:oso/9780198567523.003.0012 9. Mosadeghrad AM, Ferlie E, Rosenberg D. A study of relationship between job stress, quality of working life and turnover intention among hospital employees. Health Serv Manage Res (2011.0) 24 170-181. DOI: 10.1258/hsmr.2011.011009 10. Kim HC, Kim BK, Min KB, Min JY, Hwang SH, Park SG. Association between job stress and insomnia in Korean workers. J Occup Health (2011.0) 53 164-174. DOI: 10.1539/joh.10-0032-oa 11. Ezenwanne E.. Current concepts in the neurophysiologic basis of sleep; a review. Ann Med Health Sci Res (2011.0) 1 173-9. PMID: 23209972 12. Luyster FS, Strollo PJ, Zee PC, Walsh JK. Sleep: A health imperative. Sleep (2012.0) 35 727-734. DOI: 10.5665/sleep.1846 13. Adam K.. Sleep as a Restorative Process and a Theory to Explain Why. Prog Brain Res (1980.0) 53 289-305. DOI: 10.1016/S0079-6123(08)60070-9 14. Levandovski R, Dantas G, Fernandes LC, Caumo W, Torres I, Roenneberg T. Depression scores associate with chronotype and social jetlag in a rural population. Chronobiol Int (2011.0) 28 771-778. DOI: 10.3109/07420528.2011.602445 15. Roenneberg T, Allebrandt K V, Merrow M, Vetter C. Social jetlag and obesity. Current Biology (2012.0) 22 939-943. DOI: 10.1016/j.cub.2012.03.038 16. Parsons M, Moffitt T, Gregory A, Goldman-Mellor S, Nolan P, Poulton R. Social jetlag, obesity and metabolic disorder: investigation in a cohort study (2015.0) 39 842-848. DOI: 10.1038/ijo.2014.201 17. Borisenkov MF. Latitude of residence and position in time zone are predictors of cancer incidence, cancer mortality, and life expectancy at birth. Chronobiol Int (2011.0) 28 155-162. DOI: 10.3109/07420528.2010.541312 18. Kalak N, Lemola S, Brand S, Holsboer-Trachsler E, Grob A. Sleep duration and subjective psychological well-being in adolescence: A longitudinal study in Switzerland and Norway. Neuropsychiatr Dis Treat (2014.0) 10 1199-1207. DOI: 10.2147/NDT.S62533 19. Shin JE, Kim JK. How a good sleep predicts life satisfaction: The role of zero-sum beliefs about happiness. Front Psychol (2018.0) 9 1-4. DOI: 10.3389/fpsyg.2018.01589 20. Cappuccio FP, D’Elia L, Strazzullo P, Miller MA. Sleep duration and all-cause mortality: A systematic review and meta-analysis of prospective studies. Sleep (2010.0) 33 585-592. DOI: 10.1093/sleep/33.5.585 21. Shen L, van Schie J, Ditchburn G, Brook L, Bei B. Positive and Negative Emotions: Differential Associations with Sleep Duration and Quality in Adolescents. J Youth Adolesc (2018.0) 47 2584-2595. DOI: 10.1007/s10964-018-0899-1 22. Ness TEB, Saksvik-Lehouillier I. The Relationships between Life Satisfaction and Sleep Quality, Sleep Duration and Variability of Sleep in University Students. Journal of European Psychology Students (2018.0) 9 28-39. DOI: 10.5334/jeps.434 23. Pilcher JJ, Ott ES. The relationships between sleep and measures of health and well-being in college students: A repeated measures approach. Behavioral Medicine (1998.0) 23 170-178. DOI: 10.1080/08964289809596373 24. Pagan R.. Sleep duration, life satisfaction and disability. Disabil Health J (2016.0) 10 334-343. DOI: 10.1016/j.dhjo.2016.10.005 25. Piper AT. Sleep duration and life satisfaction. Int Rev Econ (2015.0) 305-325 26. Önder İ.. Association of happiness with morningness—eveningness preference, sleep-related variables and academic performance in university students. Biol Rhythm Res (2020.0). DOI: 10.1080/09291016.2020.1848266 27. Lau EYY, Wong ML, Ng ECW, Hui CCH, Cheung SF, Mok DSY. “Social Jetlag” in morning-type college students living on campus: Implications for physical and psychological well-being. Chronobiol Int (2013.0) 30 910-918. DOI: 10.3109/07420528.2013.789895 28. Shao MF, Chou YC, Yeh MY, Tzeng WC. Sleep quality and quality of life in female shift-working nurses. J Adv Nurs (2010.0) 66 1565-1572. DOI: 10.1111/j.1365-2648.2010.05300.x 29. Ritsner M, Kurs R, Ponizovsky A, Hadjez J. Perceived quality of life in schizophrenia: Relationships to sleep quality. Quality of Life Research (2004.0) 783-791. DOI: 10.1023/B:QURE.0000021687.18783.d6 30. Zeitlhofer J., Schmeiser-Rieder A., Tribl G., Rosenberger A., Bolitschek J., Kapfhammer G., Saletu B.. Sleep and quality of life in the Austrian population. Acta Neurol Scand (2000.0) 102 249-257. DOI: 10.1034/j.1600-0404.2000.102004249.x 31. Lawson LP, Richdale AL, Haschek A, Flower RL, Vartuli J, Arnold SRC. Cross-sectional and longitudinal predictors of quality of life in autistic individuals from adolescence to adulthood: The role of mental health and sleep quality. Autism (2020.0) 24 954-967. DOI: 10.1177/1362361320908107 32. Roenneberg T.. What is chronotype?. Sleep Biol Rhythms (2012.0) 10 75-76. DOI: 10.1111/j.1479-8425.2012.00541.x 33. Wittmann M, Dinich J, Merrow M, Roenneberg T. Social jetlag: Misalignment of biological and social time. Chronobiol Int (2006.0) 23 497-509. DOI: 10.1080/07420520500545979 34. Jankowski KS. Is the shift in chronotype associated with an alteration in well-being?. Biol Rhythm Res (2014.0) 46 237-248. DOI: 10.1080/09291016.2014.985000 35. Chang SJ, Jang SJ. Social jetlag and quality of life among nursing students: A cross-sectional study. J Adv Nurs (2019.0) 75 1418-1426. DOI: 10.1111/jan.13857 36. Jankowski KS, Vollmer C, Linke M, Randler C. Differences in sun time within the same time zone affect sleep–wake and social rhythms, but not morningness preference: Findings from a Polish–German comparison study. Time Soc (2014.0) 23 258-276. DOI: 10.1177/0961463X14535911 37. Önder I, Beşoluk Ş, Iskender M, Masal E, Demirhan E. Circadian Preferences, Sleep Quality and Sleep Patterns, Personality, Academic Motivation and Academic Achievement of university students. Learn Individ Differ (2014.0) 32 184-192. DOI: 10.1016/j.lindif.2014.02.003 38. Roenneberg T, Wirz-Justice A, Merrow M. Life between clocks: Daily temporal patterns of human chronotypes. J Biol Rhythms (2003.0) 18 80-90. DOI: 10.1177/0748730402239679 39. 39Kudrnáčová M. Czech Household Panel Survey. Data Documentation. 2019 [cited 10 Jan 2021] p. 82. Available: http://dspace.soc.cas.cz:8080/xmlui/handle/123456789/3780 40. Sládek M, Kudrnáčová Röschová M, Adámková V, Hamplová D, Sumová A. Chronotype assessment via a large scale socio-demographic survey favours yearlong Standard time over Daylight Saving Time in central Europe. Sci Rep (2020.0) 10 1-18. DOI: 10.1038/s41598-020-58413-9 41. Kudrnáčová M, Hamplová D. Social Jetlag in the Context of Work and Family. Sociológia (2022.0) 54 295-324. DOI: 10.31577/sociologia.2022.54.4.11 42. Hamplová D, Chaloupková JK, Topinková R. More Money, Less Housework? Relative Resources and Housework in the Czech Republic. J Fam Issues (2019.0) 40 2823-2848. DOI: 10.1177/0192513X19864988 43. Dudová R, Hašková H, Klímová Chaloupková J. Disentangling the link between having one child and partnership trajectories: a mixed-methods life-course research. J Fam Stud (2020.0) 0 1-22. DOI: 10.1080/13229400.2020.1839534 44. Lux M, Sunega P, Kážmér L. Intergenerational financial transfers and indirect reciprocity: determinants of the reproduction of homeownership in the post-socialist Czech Republic. Hous Stud (2021.0) 36 1294-1317. DOI: 10.1080/02673037.2018.1541441 45. Kudrnáč A.. A study of the effects of obesity and poor health on the relationship between distance to the polling station and the probability to vote. Party Politics (2021.0) 27 540-551. DOI: 10.1177/1354068819867414 46. Walsh E, Murphy A. Life satisfaction amongst working parents: examining the case of mothers and fathers in Ireland. Int J Soc Econ (2021.0) 48 622-639. DOI: 10.1108/IJSE-05-2020-0295 47. Yaya S, Ghosh S, Ghose B. Subjective happiness, health and quality of life and their sociocultural correlates among younger population in Malawi. Soc Sci (2019.0) 8. DOI: 10.3390/socsci8020055 48. Nordenmark M, Vinberg S, Strandh M. Job control and demands, work-life balance and wellbeing among self-employed men and women in Europe. Vulnerable Groups & Inclusion (2012.0) 3 18896. DOI: 10.3402/vgi.v3i0.18896 49. Vinson T, Ericson M. The social dimensions of happiness and life satisfaction of Australians: Evidence from the World Values Survey. Int J Soc Welf (2014.0) 23 240-253. DOI: 10.1111/ijsw.12062 50. Carney CE, Buysse DJ, Ancoli-Israel S, Edinger JD, Krystal AD, Lichstein KL. The consensus sleep diary: Standardizing prospective sleep self-monitoring. Sleep (2012.0) 287-302. DOI: 10.5665/sleep.1642 51. Foley D, Ancoli-Israel S, Britz P, Walsh J. Sleep disturbances and chronic disease in older adults: Results of the 2003 National Sleep Foundation Sleep in America Survey. J Psychosom Res (2004.0) 56 497-502. DOI: 10.1016/j.jpsychores.2004.02.010 52. 52WEP. MCTQ. 2020 [cited 24 Jan 2020]. Available: https://www.thewep.org/documentations/mctq 53. 53Czech Statistical Office. The sample survey within the programme of the European Union–Statistics on Income and Living Conditions (EU-SILC). 2022 Feb. 54. Kreft IG, Leeuw J de. Introducing Multilevel Modeling (1998.0). DOI: 10.4135/9781849209366 55. 55Information Services Department. Czech Republic in International Comparison (Selected Indicators). 2021. 56. 56Eurostat. Living conditions in Europe 2018 edition. Meglio E Di, editor. Luxembourg; 2018. doi: 10.2785/39876. DOI: 10.2785/39876 57. 57Eurostat. Quality of life indicators—overall experience of life. 2020 [cited 24 Mar 2021]. Available: https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Quality_of_life_indicators_-_overall_experience_of_life#Life_satisfaction_by_income_and_age_groups 58. Lakerveld J, Mackenbach JD, Horvath E, Rutters F, Compernolle S, Bárdos H. The relation between sleep duration and sedentary behaviours in European adults. Obesity Reviews (2016.0) 17 62-67. DOI: 10.1111/obr.12381 59. Randler C, Díaz-Morales JF, Jankowski KS. Synchrony in chronotype and social jetlag between dogs and humans across Europe. Time Soc (2018.0) 27 223-238. DOI: 10.1177/0961463X15596705 60. Chen X, Wang S Bin, Li XL, Huang ZH, Tan WY, Lin HC. Relationship between sleep duration and sociodemographic characteristics, mental health and chronic diseases in individuals aged from 18 to 85 years old in Guangdong province in China: A population-based cross-sectional study. BMC Psychiatry (2020.0) 20 1-10. DOI: 10.1186/s12888-020-02866-9 61. Wunsch K, Nigg CR, Weyland S, Jekauc D, Niessner C, Burchartz A. The relationship of self-reported and device-based measures of physical activity and health-related quality of life in adolescents. Health Qual Life Outcomes (2021.0) 19 1-10. DOI: 10.1186/s12955-021-01682-3
--- title: Association of lymphopenia and RDW elevation with risk of mortality in acute aortic dissection authors: - Dan Yu - Peng Chen - Xueyan Zhang - Hongjie Wang - Menaka Dhuromsingh - Jinxiu Wu - Bingyu Qin - Suping Guo - Baoquan Zhang - Chunwen Li - Hesong Zeng journal: PLOS ONE year: 2023 pmcid: PMC10016706 doi: 10.1371/journal.pone.0283008 license: CC BY 4.0 --- # Association of lymphopenia and RDW elevation with risk of mortality in acute aortic dissection ## Abstract ### Objective The study aimed to investigate whether lymphopenia and red blood cell distribution width (RDW) elevation are associated with an increased risk of mortality in acute aortic dissection (AAD). ### Methods This multicenter retrospective cohort study enrolled patients diagnosed with AAD by aortic computed tomographic angiography (CTA) from 2010 to 2021 in five teaching hospitals in central-western China. Cox proportional hazards regression and Kaplan-Meier curves were used in univariable and multivariable models. Clinical outcomes were defined as all-cause in-hospital mortality, while associations were evaluated between lymphopenia, accompanied by an elevated RDW, and risk of mortality. ### Results Of 1903 participants, the median age was 53 (interquartile range [IQR], 46–62) years, and females accounted for $21.9\%$. Adjusted increased risk of mortality was linearly related to the decreasing lymphocyte percentage (P-non-linearity = 0.942) and increasing RDW (P-non-linearity = 0.612), and per standard deviation (SD) of increment lymphocyte percentage and RDW was associated with the $26\%$ (0.74, 0.64–0.84) decrement and $5\%$ (1.05, 0.95–1.15) increment in hazard ratios (HRs) and $95\%$ confidence intervals (CIs) of mortality, respectively. Importantly, lymphopenia and elevation of RDW exhibited a significant interaction with increasing the risk of AAD mortality (P-value for interaction = 0.037). ### Conclusions Lymphopenia accompanied by the elevation of RDW, which may reflect the immune dysregulation of AAD patients, is associated with an increased risk of mortality. Assessment of immunological biomarkers derived from routine tests may provide novel perspectives for identifying the risk of mortality. ## Introduction Acute aortic dissection (AAD) is a potentially lethal condition with high morbidity and mortality [1, 2]. Despite remarkable progress in diagnostic and therapeutic techniques, the global burden of AAD remains high [3]. Assessing clinical indicators associated with the risk of mortality is greatly important for patient management, especially when the indicators can be targeted. Various indicators have been associated with the outcome of AAD, including but not limited to clinical signs, anatomy, hemodynamics, and a series of biomarkers, such as D-dimer and fibrin degradation products [4]. However, an in-depth investigation of the indicators based on the underlying pathogenesis and mechanisms will be more valuable for improving disease outcomes than such studies. AAD mainly occurs either spontaneously (sporadic) or in association with a genetic condition and trauma. The main mechanism that contributes to sporadic AAD is inflammation [3, 5–8]. Theoretically, inflammation involving the innate and adaptive immune systems is a normal response to injury [9, 10]. However, immune dysregulation manifested as a disordered immune response finally results in widespread inflammation and multiorgan damage [11]. The prognostic implications of immune dysregulation for cardiovascular disease have been extensively studied and well elucidated [12–15], although little attention has been paid to the association between dysregulation of immunologic function and AAD mortality. It is generally believed that lymphopenia is one of the typical phenotypes of immune dysregulation and is among the strongest risk factors for outcomes in cardiovascular and noncardiovascular disorders [16–21]. However, the relationship between lymphopenia and AAD mortality has rarely been studied. Furthermore, immune pathways often influence multiple variables, some of which are measured routinely on admission and are substantially neglected by clinicians. For instance, evidence has indicated that an elevated red blood cell distribution width (RDW) is associated with cardiovascular and noncardiovascular disease and death [22, 23]. However, to the best of our current knowledge, the extent to which lymphopenia is associated with the risk of mortality in AAD and whether RDW is an additive risk factor beyond lymphopenia have not yet been studied. In the present study, we characterized the associations among lymphopenia, elevated RDW, and risk of mortality in AAD. We aimed to test the hypothesis that lymphopenia of AAD patients enhances the risk of mortality, and those with concomitantly abnormal elevation of RDW have further reduced survival. ## Study population and data collection We designed a hospital-based multicenter retrospective cohort study by extracting data from electronic medical records (EMRs) of patients admitted to Tongji Hospital Tongji Medical College of Huazhong University of Science and Technology, People’s Hospital of Zhengzhou University, Central China Fuwai Hospital of Zhengzhou University, the Third Affiliated Hospital of Xinxiang Medical University, and the Second Affiliated Hospital of Chongqing Medical University, respectively, which are five teaching hospitals distributed in central-western China. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. According to the actual situation of EMRs in different hospitals, we enrolled all available patients who underwent aortic computed tomographic angiography (CTA) between August 2010 and October 2021. Participants who were diagnosed with intramural hematoma (IMH) and penetrating atherosclerotic ulcer (PAU) were excluded, along with participants with a diagnosis of aortic aneurysms, participants with diagnoses including postoperative review of aortic dissection, and participants without a diagnosis of aortic diseases. After applying the first two-round study exclusion criteria, enrolled patients were screened out for further extraction of study variables. The variables belonging to the following categories were extracted separately over the admission period: demographics, initial vital signs, medical history, routine blood tests, coagulation function, biochemical tests, CTA-based anatomical classification, and the status of all-cause in-hospital death. Furthermore, patients aged < 18 years, pregnant aortic dissection patients, non-Han patients, patients with incomplete routine blood tests, patients with disease onset ≥ 14 days, and those with vague records of onset time were further excluded after the collection of variables. The study was conducted between January 2022 and August 2022, after approval from the Research Ethics Commissions of Tongji Hospital Tongji Medical College of Huazhong University of Science and Technology (TJ- IRB20211102), People’s Hospital of Zhengzhou University (2021–190), Central China Fuwai Hospital of Zhengzhou University (2021–38), the Third Affiliated Hospital of Xinxiang Medical University (K2021-039-01), and the Second Affiliated Hospital of Chongqing Medical University (2022–15), respectively, with waived informed consent by the Ethics Commissions mentioned above. The authors had access to information that could identify individual participants during or after the data collection. ## Study variables The primary outcome of this study was all-cause in-hospital mortality associated with study baseline variables. Immune dysregulation participants were defined as AAD patients characterized by lymphopenia and elevated RDW. The primary exposure was the percentage of lymphocytes, and the secondary exposure was the RDW level. The variables of demographics, initial vital signs, medical history, and the status of all-cause in-hospital death were identified based on the EMRs in different hospitals. All analysis results of routine blood tests, coagulation function, and biochemical tests were obtained from the laboratory department of each hospital and all the blood test samplings were completed within 2 h of admission. The anatomical classification was independently judged by a skilled clinician based on aortic CTA, while the DeBakey system [6] and the category of isolated abdominal AAD [24, 25] were used for identification. The etiology of AAD was defined as genetic, traumatic, congenital disorder, vascular inflammation, infectious disease, and sporadic based on whether the patients had a history of Marfan syndrome, trauma, bicuspid aortic valve, Takayasu arteritis and syphilis. ## Statistical analysis Baseline characteristics of the analytic sample were summarized across all-cause in-hospital death status as continuous variables and were represented as the median (interquartile range [IQR]), and categorical variables were presented as a percentage. Baseline characteristics were compared using the Chi-squared test for categorical variables and the Mann–Whitney U test for continuous variables. Multivariable Cox proportional hazard regression models were used to estimate hazard ratios (HRs) and $95\%$ confidence intervals (CIs) for the associations of lymphocyte percentage and RDW level with all-cause in-hospital mortality. Referring to previous researches [2], all models were successively adjusted for age (continuous), sex (female, male), smoking history (yes, no), hypertension history (yes, no), diabetes history (yes, no), aortic valve replacement history (yes, no), anatomical classification (DeBakey Ⅰ, DeBakey Ⅱ, DeBakey Ⅲa, DeBakey Ⅲb, or isolated abdominal AAD), etiology (genetic, traumatic, congenital disorder, vascular inflammation, infectious disease, or sporadic), aorta diameter (≥ 5.5 cm, < 5.5 cm), onset time (< 24 h, 1–7 d, 8–14 d), and hospital centers (Tongji Hospital, People’s Hospital of Zhengzhou University, Central China Fuwai Hospital of Zhengzhou University, the Third Affiliated Hospital of Xinxiang Medical University, or the Second Affiliated Hospital of Chongqing Medical University). The dose-response curves presenting the hazard of lymphocyte percentage and RDW level were fitted by using the restricted cubic spline model with four knots (rms, hmisc, lattice, and survival packages in R software). Kaplan-Meier methods were performed for survival curve plotting. We further accessed the interaction between lymphopenia and RDW and their synergistic effects on the risk of AAD mortality, and the statistical significance of the interaction was examined by the joint test. Several secondary analyses were conducted to examine the robustness of our results. Stratified analyses were performed across ages, sexes, smoking history, hypertension history, diabetes history, aortic valve replacement history, anatomical classifications, etiologies, aorta diameter, onset time and hospital centers. We calculated the p-value for interaction to examine the consistency of patterns in the main results. Considering the influence of acute kidney injury, procedure of operation, transfusions, stroke or coma, and limb ischemia on AAD mortality, we analyzed the outcomes in prespecified subgroups, divided according to the option of the above-mentioned parameters. Given unavailable records of hyperlipidemia history, we further adjusted for random lipid levels on admission. Considering the potential effects on Cox proportional hazard model fit, propensity score matching (PSM) was performed to adjust for differences in baseline characteristics between in-hospital alive and in-hospital dead groups. Cox proportional hazard regression models were re-fitted in the matched population to test the stability of our results. Missing values of covariates were treated as dummy variables. SAS version 9.4 (SAS Institute, USA) and R software (the R Foundation, http://www.r-project.org, version 4.0.2) were utilized for analyses and plotting with a two-sided significance threshold of $P \leq 0.05.$ ## Study participants From all five teaching hospital centers, we identified 35,260 patients who underwent aortic CTA examination between August 2010 and October 2021. In the first round of study participant exclusion, 5214 patients were collected as those diagnosed with first-onset AAD from all five hospitals. Moreover, 2242 AAD inpatients with available EMRs were summarized after the second-round exclusion. In total, 1903 patients were confirmed eligible for statistical analysis at the third-round exclusion. A flowchart of the study, including the details of participants’ screening, is shown in Fig 1. Fig 1: Flowchart of the study.CTA, computed tomographic angiography; IMH, intramural hematoma; PAU, aortic pseudoaneurysm; EMRs, electronic medical records. ## Baseline characteristics A total of 1903 patients were included in the present analysis in which the median age was 53 (IQR 46–62) years, and females accounted for $21.9\%$. The baseline characteristics of the participants are provided in Table 1and S1 Table. Among these patients, there were 1430 ($75.1\%$) in-hospital alive and 473 ($24.9\%$) in-hospital dead patients. Generally, those in-hospital dead participants were more likely to be classified as DeBakey Ⅰ, be a smoker, undergo a surgical operation or had no procedure of operation, onset AAD within 24 h, and had acute kidney injury, stroke or coma, and had limb ischemia. However, there were no significant differences between surviving and deceased patients in age, sex, history of hypertension, diabetes and aortic valve replacement, etiology and transfusions. Table 1 \| Variables \| Total patients \| In-hospital alive \| In-hospital dead \| P-value \| \| --- \| --- \| --- \| --- \| --- \| \| N \| 1903 \| 1430 \| 473 \| \| \| Female, n (%) \| 416 (21.9%) \| 307 (21.5%) \| 109 (23.0%) \| 0.472 \| \| Age, mean median (IQR), years \| 53 (46–62) \| 53 (45–62) \| 54 (47–62) \| 0.056 \| \| Anatomical classification \| Anatomical classification \| Anatomical classification \| Anatomical classification \| Anatomical classification \| \| DeBakey Ⅰ, n (%) \| 1021 (53.7%) \| 638 (44.6%) \| 383 (81.0%) \| <0.001 \| \| DeBakey Ⅱ, n (%) \| 125 (6.6%) \| 100 (7.0%) \| 25 (5.3%) \| <0.001 \| \| DeBakey Ⅲa, n (%) \| 59 (3.1%) \| 57 (4.0%) \| 2 (0.4%) \| <0.001 \| \| DeBakey Ⅲb, n (%) \| 606 (31.8%) \| 549 (38.4%) \| 57 (12.1%) \| <0.001 \| \| Isolated abdominal AAD, n (%) \| 92 (4.8%) \| 86 (6.0%) \| 6 (1.3%) \| <0.001 \| \| Etiology \| Etiology \| Etiology \| Etiology \| Etiology \| \| Genetic (MFS), n (%) \| 21 (1.1%) \| 14 (1.0%) \| 7 (1.5%) \| 0.366 \| \| Traumatic, n (%) \| 23 (1.2%) \| 20 (1.4%) \| 3 (0.6%) \| 0.187 \| \| Congenital disorder (BAV), n (%) \| 3 (0.2%) \| 2 (0.1%) \| 1 (0.2%) \| 0.576 \| \| Vascular inflammation (Takayasu arteritis), n (%) \| 3 (0.2%) \| 3 (0.2%) \| 0 (0.0%) \| >0.99 \| \| Infectious disease (Syphilis), n (%) \| 17 (0.9%) \| 14 (1.0%) \| 3 (0.6%) \| 0.49 \| \| Sporadic, n (%) \| 1837 (96.5%) \| 1378 (96.4%) \| 459 (97.0%) \| 0.486 \| \| History \| History \| History \| History \| History \| \| Smoking, n (%) \| 653 (34.3%) \| 509 (35.6%) \| 144 (30.4%) \| 0.041 \| \| Hypertension, n (%) \| 1172 (61.6%) \| 897 (62.7%) \| 275 (58.1%) \| 0.075 \| \| Diabetes, n (%) \| 54 (2.8%) \| 44 (3.1%) \| 10 (2.1%) \| 0.274 \| \| Aortic valve replacement, n (%) \| 15 (0.8%) \| 11 (0.8%) \| 4 (0.8%) \| 0.871 \| \| Procedure of operation \| Procedure of operation \| Procedure of operation \| Procedure of operation \| Procedure of operation \| \| None, n (%) \| 573 (30.1%) \| 316 (22.1%) \| 257 (54.3%) \| <0.001 \| \| Endovascular management, n (%) \| 648 (34.1%) \| 616 (43.1%) \| 32 (6.8%) \| <0.001 \| \| Surgical operation, n (%) \| 469 (24.6%) \| 316 (22.1%) \| 153 (32.3%) \| <0.001 \| \| Surgical operation and endovascular management, n (%) \| 213 (11.2%) \| 182 (12.7%) \| 31 (6.6%) \| <0.001 \| \| Onset time \| Onset time \| Onset time \| Onset time \| Onset time \| \| < 24h, n (%) \| 1118 (58.7%) \| 802 (56.1%) \| 316 (66.8%) \| <0.001 \| \| 1-7d, n (%) \| 692 (36.4%) \| 551 (38.5%) \| 141 (29.8%) \| <0.001 \| \| 8-14d, n (%) \| 93 (4.9%) \| 77 (5.4%) \| 16 (3.4%) \| <0.001 \| \| Aorta diameter \| Aorta diameter \| Aorta diameter \| Aorta diameter \| Aorta diameter \| \| ≥ 5.5 cm, n (%) \| 1720 (96.6%) \| 1298 (97.2%) \| 422 (94.8%) \| 0.015 \| \| < 5.5 cm, n (%) \| 60 (3.4%) \| 37 (2.8%) \| 23 (5.2%) \| 0.015 \| \| Acute kidney injury, n (%) \| 275 (14.5%) \| 147 (10.3%) \| 128 (27.1%) \| <0.001 \| \| Stroke or coma, n (%) \| 119 (6.3%) \| 53 (3.7%) \| 66 (14.0%) \| <0.001 \| \| Transfusions, n (%) \| 723 (38.0%) \| 537 (37.6%) \| 186 (39.3%) \| 0.492 \| \| Limb ischemia, n (%) \| 163 (8.6%) \| 96 (6.7%) \| 67 (14.2%) \| <0.001 \| \| Hospital centers, n (%) \| Hospital centers, n (%) \| Hospital centers, n (%) \| Hospital centers, n (%) \| Hospital centers, n (%) \| \| Tongji Hospital \| 1345 (70.7%) \| 948 (66.3%) \| 397 (83.9%) \| <0.001 \| \| People’s Hospital of Zhengzhou University \| 253 (13.3%) \| 212 (14.8%) \| 41 (8.7%) \| <0.001 \| \| Central China Fuwai Hospital of Zhengzhou University \| 115 (6.0%) \| 99 (6.9%) \| 16 (3.4%) \| <0.001 \| \| Third Affiliated Hospital of Xinxiang Medical University \| 67 (3.5%) \| 56 (3.9%) \| 11 (2.3%) \| <0.001 \| \| Second Affiliated Hospital of Chongqing Medical University \| 123 (6.5%) \| 115 (8.0%) \| 8 (1.7%) \| <0.001 \| \| Lymphocyte percentage, median (IQR), % \| 8.4 (5.5–13.4) \| 8.9 (5.9–14.5) \| 7.0 (4.9–10.3) \| <0.001 \| \| RDW-SD, median (IQR), fL \| 43.4 (41.1–46.0) \| 43.1 (41.0–45.7) \| 44.0 (41.8–46.6) \| <0.001 \| \| Total cholesterol, median (IQR), mmol/L \| 3.9 (3.4–4.5) \| 3.9 (3.4–4.5) \| 3.8 (3.2–4.3) \| <0.001 \| \| LDL-C, median (IQR), mmol/L \| 2.3 (1.7–2.8) \| 2.3 (1.7–2.8) \| 2.2 (1.7–2.7) \| 0.385 \| ## Association of lymphopenia and elevation of RDW with risk of AAD all-cause in-hospital mortality In Cox regression analyses, lymphocyte percentage and RDW were firstly applied as continuous variables to fitted smoothing splines to present the dose-response relationship between these two variables and the risk of mortality. The adjusted risk of all-cause in-hospital mortality was positively associated with decreasing lymphocyte percentage and increasing RDW (Fig 2). The test of non-linearity all-cause in-hospital mortality was not statistically significant for lymphocyte percentage (P-non-linearity = 0.942) and RDW (P-non-linearity = 0.612). We converted lymphocyte percentage and RDW into categorical variables based on quintiles of the distribution. The unadjusted Kaplan–Meier survival curve demonstrated a significant variance among the groups of patients with the different quintiles categories of lymphocyte percentage and RDW level ($P \leq 0.0001$, log-rank test) (Fig 3). Fig 2: Dose-response curves for lymphocyte percentage and RDW level with risk of AAD in-hospital mortality.Hazard ratios (blue lines) and 95% confidence intervals (light blue shade) were adjusted for age, sex, smoking history, hypertension history, diabetes history, aortic valve replacement history, anatomical classification, etiology, aorta diameter, onset time and hospital centers. AAD, acute aortic dissection; RDW, red blood cell distribution width. Fig 3: Kaplan–Meier survival curve of AAD in-hospital mortality showing that outcomes significantly varied among the groups of patients with the different quintiles categories of lymphocyte percentage and RDW level (P < 0.0001 log-rank tests).AAD, acute aortic dissection; RDW, red blood cell distribution width. Stepwise Cox regression analyses were conducted to build adjusted models while adequately considering possible confounders to further explore the risk of AAD all-cause in-hospital mortality according to lymphocyte percentage and RDW level categories. Of note, these associations remained robust after stepwise adjustment for confounders (Table 2). In the crude model, the HRs and $95\%$ CIs from the lowest to the highest lymphocyte percentage categories (≤ 5.0, 5.1–7.1, 7.2–9.7, 9.8–14.8, and ≥ $14.9\%$) were 1.00 (reference), 0.86 (0.66, 1.10), 0.72 (0.55, 0.93), 0.60 (0.46, 0.79), and 0.32 (0.23, 0.44), respectively, for all-cause in-hospital mortality; while the HRs and $95\%$ CIs for the same categories were 1.00 (reference), 0.89 (0.69, 1.15), 0.79 (0.61, 1.03), 0.75 (0.57, 0.99), and 0.50 (0.35, 0.72), respectively, in the fully adjusted model. Additionally, the HRs and $95\%$ CIs from the lowest to the highest RDW categories (≤ 40.7, 40.8–42.4, 42.5–44.0, 44.1–46.7, and ≥ 46.8 fL) were 1.00 (reference), 1.33 (0.96, 1.83), 1.37(1.00, 1.88), 1.96 (1.45, 2.64), and 1.89 (1.40, 2.56), respectively, for all-cause in-hospital mortality in the crude model; while the HRs and $95\%$ CIs for the same categories were 1.00 (reference), 1.16 (0.84, 1.61), 1.17 (0.84, 1.62), 1.51 (1.11, 2.06), and 1.45 (1.06, 1.98), respectively, in the fully adjusted model. When the lymphocyte percentage and RDW were considered a continuous variable, per standard deviation (SD) of increment lymphocyte percentage and RDW was associated with the $26\%$ (0.74, 0.64–0.84) decrement and $5\%$ (1.05, 0.95–1.15) increment in HRs and $95\%$ CIs of AAD all-cause in-hospital mortality, respectively. Table 2 \| Unnamed: 0 \| Quintiles of the exposure \| Quintiles of the exposure.1 \| Quintiles of the exposure.2 \| Quintiles of the exposure.3 \| Quintiles of the exposure.4 \| Unnamed: 6 \| Unnamed: 7 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| Q1 \| Q2 \| Q3 \| Q4 \| Q5 \| P trend \| Per SD increment \| \| Lymphocyte percentage, % \| ≤ 5.0 \| 5.1–7.1 \| 7.2–9.7 \| 9.8–14.8 \| ≥ 14.9 \| \| \| \| Deaths/N \| 133/380 \| 111/377 \| 99/379 \| 83/379 \| 47/388 \| \| \| \| Crude model \| 1 (reference) \| 0.86 (0.66, 1.10) \| 0.72 (0.55, 0.93) \| 0.60 (0.46, 0.79) \| 0.32 (0.23, 0.44) \| <0.0001 \| 0.62 (0.54, 0.71) \| \| Model 1 \| 1 (reference) \| 0.85 (0.66, 1.10) \| 0.73 (0.57, 0.95) \| 0.60 (0.46, 0.79) \| 0.31 (0.22, 0.43) \| <0.0001 \| 0.61 (0.54, 0.70) \| \| Model 2 \| 1 (reference) \| 0.89 (0.69, 1.15) \| 0.79 (0.61, 1.03) \| 0.75 (0.57, 0.99) \| 0.50 (0.35, 0.72) \| 0.0002 \| 0.74 (0.64, 0.84) \| \| RDW,fL \| ≤ 40.7 \| 40.8–42.4 \| 42.5–44.0 \| 44.1–46.7 \| ≥ 46.8 \| \| \| \| Deaths/N \| 67/373 \| 84/373 \| 88/381 \| 119/385 \| 115/391 \| \| \| \| Crude model \| 1 (reference) \| 1.33 (0.96, 1.83) \| 1.37 (1.00, 1.88) \| 1.96 (1.45, 2.64) \| 1.89 (1.40, 2.56) \| <0.0001 \| 1.12 (1.03, 1.21) \| \| Model 1 \| 1 (reference) \| 1.30 (0.95, 1.80) \| 1.31 (0.96, 1.81) \| 1.84 (1.36, 2.50) \| 1.75 (1.29, 2.38) \| <0.0001 \| 1.09 (1.00, 1.19) \| \| Model 2 \| 1 (reference) \| 1.16 (0.84, 1.61) \| 1.17 (0.84, 1.62) \| 1.51 (1.11, 2.06) \| 1.45 (1.06, 1.98) \| 0.0046 \| 1.05 (0.95, 1.15) \| ## Risk of all-cause in-hospital mortality according to lymphopenia accompanied by elevation of RDW Apart from identifying the association of lymphopenia and elevation of RDW individually with the risk of AAD all-cause in-hospital mortality, we further explored the outcome of mortality when lymphopenia was accompanied by elevated RDW. Lymphocyte percentage and RDW were again converted from continuous variables into categorical variables according to observed tertiles of the distribution to investigate the interaction among different categories of lymphopenia and elevation of RDW on influencing the risk of AAD mortality. When lymphocyte percentage and RDW categories were combined, there was an obvious association between decreasing lymphocyte percentage and elevated RDW (Fig 4and Table 3). Patients with low lymphocyte percentage and high RDW were at a higher risk of all-cause in-hospital mortality versus those with high lymphocyte percentage and low RDW (HR 1.83, $95\%$ CI 1.19–2.82). Collectively, our data demonstrated that lymphopenia combined with the elevation of RDW could serve as a novel candidate for predicting in-hospital AAD mortality. Fig 4: Risk of in-hospital mortality according to lymphocyte percentage and RDW level.Hazard ratios were adjusted for age, sex, smoking history, hypertension history, diabetes history, aortic valve replacement history, anatomical classification, etiology, aorta diameter, onset time and hospital centers. RDW, red blood cell distribution width. TABLE_PLACEHOLDER:Table 3 ## Secondary analyses Several secondary analyses were conducted to examine whether various stratified Cox proportional hazards analyses showed consistent results. As shown in S2 Table, the association of lymphocyte percentage with in-hospital mortality was robust across the strata of age, sex, smoking history, hypertension history, diabetes history, anatomical classification, aorta diameter, onset time, hospital centers, acute kidney injury, procedure of operation, transfusions, stroke or coma, limb ischemia, total cholesterol level, and low-density lipoprotein level. Importantly, patients in the strata undergoing endovascular management showed a stronger association between lymphopenia and in-hospital mortality of AAD patients. When we again performed stratified analyses for RDW according to age, sex, smoking history, hypertension history, diabetes history, aortic valve replacement history, anatomical classification, etiology, aorta diameter, onset time, hospital centers, acute kidney injury, procedure of operation, transfusions, stroke or coma, limb ischemia, total cholesterol level, and low-density lipoprotein level, only diabetes history tended to show an increased risk of all-cause in-hospital mortality, whereas other results were consistent in different variables stratified (S3 Table). Additionally, propensity score–matched population was used to fit Cox proportional hazard regression models for further testing the stability and reliability of the results. The baseline characteristics of the populations generated by PSM appear in S4 Table, the association of lymphopenia and elevated RDW with the risk of in-hospital mortality in AAD was similar with the results in Table 2 (S5 Table). ## Discussion We sought to explore the association between lymphopenia and elevated RDW of AAD patients and the risk of all-cause in-hospital mortality. Additionally, we aimed to establish the extent to which the associated risk of these two variables is additive. We found that lymphopenia and elevation of RDW were associated with the risk of all-cause in-hospital mortality in AAD patients, and those with these two indicators were at a higher risk of mortality. Taken together, our data suggest that lymphopenia and RDW elevation might reflect immune dysregulation and can be viewed as a multidimensional entity to predict the risk of all-cause in-hospital mortality in AAD patients. In previous studies, immune and inflammatory mechanisms have been widely considered to mediate cardiovascular diseases and affect their prognosis [26, 27], while the main mechanism of sporadic AAD is inflammation [3, 5–8]. These data prompted us to generate a novel hypothesis that immune dysregulation in AAD patients might have an influence on their outcomes. Immune dysregulation is an overarching term used to characterize an array of autoimmune and inflammatory conditions [28], which manifest as abnormalities of immune molecules and cells. Compared with cytokines and innate immune cells, such as neutrophils and monocyte-macrophages, lymphocytes serving as adaptive immune cells have been more easily neglected by clinicians in their routine practice. In our study, we demonstrated the significant association between lymphopenia and the risk of all-cause in-hospital mortality in AAD patients. The hypothesis that immune dysregulation of AAD patients is related to their prognosis was corroborated and presented by lymphopenia, which has been underappreciated in routine clinical work. Because mortality in the AAD population is mainly driven by noninfectious causes, our study supports the notion that immune status is indeed associated with the outcome of cardiovascular disease. Lymphopenia reflect adverse inflammatory, and the increase of inflammatory mediators such as tumor necrosis factor and interleukin 1β may reduce levels of circulating T cells [29]. Moreover, AAD patients experience stress, resulting in excess levels of cortisol and catecholamine, and abnormal hormone levels also cause lymphopenia [30, 31]. As to why the immune dysregulation might cause lymphopenia, different mechanisms have been proposed, showing that lymphopenia might be caused by the redistribution of T cells and increased susceptibility of T cells to apoptosis [32–34], while more studies are needed to substantiate these viewpoints. Besides the abnormality of immune cells, immune dysregulation also leads to impaired erythropoiesis via adverse inflammation [35–37]. RDW characterizes the heterogeneity of circulating red blood cells (RBCs) and has been used to differentiate the causes of anemia [38], while it is also utilized to predict the outcomes of cardiovascular and noncardiovascular diseases [37–40]. Furthermore, previous studies have shown that RDW is associated with the levels of interleukin 6 in heart failure and tumor necrosis factor in coronary artery lesions of Kawasaki disease [41]. Given the above-presented description, RDW manifests as another indicator of immune dysregulation and might predict the prognosis of various disorders. Previous multiple studies have identified strong associations between RDW and clinical prognosis in various populations with cardiovascular and noncardiovascular diseases [42–44]. As for the mechanisms, inflammation might lead to altered iron homeostasis and erythropoietin resistance, ultimately causing the elevation of RDW [45, 46]. In this study, the elevation of RDW was identified to be significantly associated with the risk of all-cause in-hospital mortality in AAD patients, which further confirmed the hypothesis of the association between immune dysregulation and the prognosis of AAD patients. To our knowledge, a small number of related studies have been previously performed, demonstrating similar conclusions to ours. Wei Luo et al. [ 34] have reported that lymphopenia correlated with poor outcomes in type A aortic dissection patients undergoing surgery. The researchers enrolled a total of 335 patients from two hospitals in one southern Chinese city and found that pre-operative lymphopenia, particularly CD4+ T lymphopenia, correlated with poor prognosis via apoptosis. Additionally, Cheng Jiang et al. [ 47] have stated that increased RDW was associated with poor outcomes in type B aortic dissection patients undergoing endovascular aortic repair. The investigators included a total of 678 patients from three hospitals in Guangdong province in south China and found that an RDW of > $13.5\%$ on admission was independently associated with increased long-term mortality. Compared to the present study, both of the above-mentioned studies circumscribed the characteristic of the included research patients. Only patients with a specific anatomical classification and undergoing a specific treatments were enrolled in the studies. However, study populations chosen with special features might have limited external generalizability of their results. In the current study, a larger research population was enrolled from a broader territory in China. Patients with all types of aortic dissections based on the anatomical classification and treatments were included to investigate the association between the above-described two indicators and prognosis, and we identified the suitability of results for patients with different features by several secondary analyses. Furthermore, we creatively found a significant interaction between lymphopenia and elevation of RDW for jointly predicting the risk of all-cause in-hospital mortality in AAD patients, and there should be an underlying logical link among them. Theoretically, immune dysregulation might result in abnormal changes of immune cells and immune molecules. In addition to the abnormal changes of immune cells, lymphopenia might reflect adverse inflammatory, metabolic, or neuroendocrine stressors [29]. However, abnormal changed immune molecules, also regarded as inflammatory cytokines, together with activated neurohumoral and adrenergic systems, are both involved in the pathophysiology of RDW elevation in cardiovascular and cerebrovascular diseases [48]. Taken together, the combination of lymphopenia and elevation of RDW might represent a more severe immune phenotype to help in identifying the extremely high-risk AAD patients. However, further research is needed to unveil the potential causes and underlying mechanisms. Similar to the association between lymphopenia and elevated RDW of AAD patients and their prognosis, hemogram parameters are also important in the evaluation of the other cardiovascular diseases. The prognostic value of hematological indices in stable coronary artery disease (CAD) has demonstrated that the mean platelet volume (MPV) and MPV-to-platelet ratio (MPR) might be associated with the degree of collateral development (CCD) in chronic stable CAD. However, the negative association was also found between RDW and inadequate CCD [49]. Further studies have elucidated those similar indicators including MPV-to-lymphocyte ratio and platelet distribution width (PDW) could serve as a marker for CCD in patients with stable angina and non-ST-elevation myocardial infarction (NSTEMI) [50, 51]. Together, our work and the above-mentioned studies illustrated the extensive clinical value of hemogram parameters in assessing cardiovascular diseases, warranting clinical promotion and popularization. Aortic dissection is literally defined as the results of the separation (dissection) of aortic wall layers and is caused by a tear in the intimal layer of the aorta. Once the structural properties of the aorta are compromised, existing dissections are aggravated by mechanical stress caused by blood flow until death [6]. Most of life-threatening clinical features that have been reported in the international registry of AAD [2], such as hypotension and shock, pericardial effusion and tamponade, and periaortic hematoma and brain injury, are caused by aortic rupture, for which the potential reason is compromised aorta. Immune dysregulation, which has been previously considered as a disordered immune response resulting in widespread inflammation and multiorgan damage [11], is also involved in the incidence and development of compromised aorta. As to the underlying mechanisms, several immune-related mechanisms have been suggested in previous studies. Firstly, an immune infiltrate has been found within the middle and outer tunics of dissected aortic specimens [52]. Secondly, single-cell transcriptome analysis has revealed dynamic cell populations in control and aneurysmal human aortic tissue in previous study. The tissues of thoracic aortic aneurysm which can lead to aortic dissection, rupture, and other life-threatening complications had fewer nonimmune cells and more immune cells, especially T lymphocytes, than control tissues [53]. Thirdly, the recall and activation of macrophages inside the middle tunic have also been observed and considered as key events in the early phases of AAD, and macrophages could release metalloproteinases (MMPs) and pro-inflammatory cytokines which lead to matrix degradation. Finally, the imbalance between the production of MMPs and MMP tissue inhibitors is pivotal in the extracellular matrix degradation underlying aortic wall remodeling and tearing in dissections [52]. However, more researches is still needed to explore the relationship between immune dysregulation and ADD-related mortality in the future. Several limitations of our study deserve special consideration. First, this was a retrospective observational study. Although the multicenter cohort study represents a large sample in which we were able to minimize bias, we could not exclude residual confounding, especially by unavailable variables. Second, as with all observational studies, the extent to which these associations might be causal in nature could not be assessed. Third, due to the absent data of follow-up data, we could not evaluate the long-term prognosis of AAD patients and analyze the secondary outcomes, such as relapse of aortic dissection and developing coronary artery and cerebrovascular diseases, and other various complications. Fourth, we assessed death by the discharge record of EMRs, and the specific cause of death could not be clarified with some patients; hence, all-cause in-hospital mortality was the only outcome for analysis in this study. Fifth, as only the Chinese Han population was enrolled in the study, ethnic differences might limit the external generalizability of the results. Despite these limitations, we believe that this study provides beneficial support for clinicians in identifying AAD patients at increased risk of death. ## Conclusions In this multicenter retrospective cohort study, we investigated the association among lymphopenia and elevation of RDW in AAD patients and the risk of all-cause in-hospital mortality, while Lymphopenia and RDW elevation might suggest immune dysregulation and be used to identify and quantify immunologic abnormalities in the AAD population. The presence of lymphopenia and elevation of RDW was associated with significantly increased mortality, and the risk was further heightened when these two indicators were combined. Assessment of immune status derived from routine tests, which have been severely neglected by clinicians in daily clinical practice, might provide novel perspectives for identifying the risk of mortality. More studies are required to confirm whether the earlier immune intervention might reduce the mortality of AAD by alleviating lymphopenia and elevated RDW. ## References 1. Milewicz DM, Ramirez F. Therapies for Thoracic Aortic Aneurysms and Acute Aortic Dissections. Arterioscler Thromb Vasc Biol (2019) 39 126-36. DOI: 10.1161/ATVBAHA.118.310956 2. Evangelista A, Isselbacher EM, Bossone E, Gleason TG, Eusanio MD, Sechtem U. Insights From the International Registry of Acute Aortic Dissection: A 20-Year Experience of Collaborative Clinical Research. Circulation (2018) 137 1846-60. DOI: 10.1161/CIRCULATIONAHA.117.031264 3. Bossone E, Eagle KA. Epidemiology and management of aortic disease: aortic aneurysms and acute aortic syndromes.. Nat Rev Cardiol (2021) 18 331-48. DOI: 10.1038/s41569-020-00472-6 4. Nienaber CA, Clough RE. Management of acute aortic dissection. Lancet (2015) 385 800-11. DOI: 10.1016/S0140-6736(14)61005-9 5. Luo W, Wang Y, Zhang L, Ren P, Zhang C, Li Y. Critical Role of Cytosolic DNA and Its Sensing Adaptor STING in Aortic Degeneration, Dissection, and Rupture.. Circulation (2020) 141 42-66. DOI: 10.1161/CIRCULATIONAHA.119.041460 6. Nienaber CA, Clough RE, Sakalihasan N, Suzuki T, Gibbs R, Mussa F. Aortic dissection. Nat Rev Dis Primers (2016) 2 16053. DOI: 10.1038/nrdp.2016.53 7. Xu H, Du S, Fang B, Li C, Jia X, Zheng S. VSMC-specific EP4 deletion exacerbates angiotensin II-induced aortic dissection by increasing vascular inflammation and blood pressure. Proc Natl Acad Sci U S A (2019) 116 8457-62. DOI: 10.1073/pnas.1902119116 8. Shen YH, LeMaire SA, Webb NR, Cassis LA, Daugherty A, Lu HS. Aortic Aneurysms and Dissections Series. Arterioscler Thromb Vasc Biol (2020) 40 e37-e46. DOI: 10.1161/ATVBAHA.120.313991 9. Sun L, Wang X, Saredy J, Yuan Z, Yang X, Wang H. Innate-adaptive immunity interplay and redox regulation in immune response. Redox Biol (2020) 37 101759. DOI: 10.1016/j.redox.2020.101759 10. Chen L, Deng H, Cui H, Fang J, Zuo Z, Deng J. Inflammatory responses and inflammation-associated diseases in organs.. Oncotarget (2018) 9 7204-18. DOI: 10.18632/oncotarget.23208 11. Tay MZ, Poh CM, Renia L, MacAry PA, Ng LFP. The trinity of COVID-19: immunity, inflammation and intervention. Nat Rev Immunol (2020) 20 363-74. DOI: 10.1038/s41577-020-0311-8 12. Meng X, Yang J, Dong M, Zhang K, Tu E, Gao Q. Regulatory T cells in cardiovascular diseases.. Nat Rev Cardiol (2016) 13 167-79. DOI: 10.1038/nrcardio.2015.169 13. Bozkurt B, Kamat I, Hotez PJ. Myocarditis With COVID-19 mRNA Vaccines.. Circulation (2021) 144 471-84. DOI: 10.1161/CIRCULATIONAHA.121.056135 14. Gupta A, Madhavan MV, Sehgal K, Nair N, Mahajan S, Sehrawat TS. Extrapulmonary manifestations of COVID-19. Nat Med (2020) 26 1017-32. DOI: 10.1038/s41591-020-0968-3 15. Ferrucci L, Fabbri E. Inflammageing: chronic inflammation in ageing, cardiovascular disease, and frailty.. Nat Rev Cardiol. (2018) 15 505-22. DOI: 10.1038/s41569-018-0064-2 16. Al-Kindi SG, Attizzani GF, Decicco AE, Alkhalil A, Nmai C, Longenecker CT. Lymphocyte counts are dynamic and associated with survival after transcatheter aortic valve replacement. Structural Heart (2018) 2 557-64 17. Bartoli-Leonard F, Zimmer J, Aikawa E. Innate and adaptive immunity: the understudied driving force of heart valve disease. Cardiovasc Res (2021) 117 2506-24. DOI: 10.1093/cvr/cvab273 18. Warny M, Helby J, Nordestgaard BG, Birgens H, Bojesen SE. Incidental lymphopenia and mortality: a prospective cohort study.. CMAJ. (2020) 192 E25-E33. DOI: 10.1503/cmaj.191024 19. Giamarellos-Bourboulis EJ, Netea MG, Rovina N, Akinosoglou K, Antoniadou A, Antonakos N. Complex Immune Dysregulation in COVID-19 Patients with Severe Respiratory Failure. Cell Host Microbe (2020) 27 992-1000 e3. DOI: 10.1016/j.chom.2020.04.009 20. Jamal M, Bangash HI, Habiba M, Lei Y, Xie T, Sun J. Immune dysregulation and system pathology in COVID-19. Virulence (2021) 12 918-36. DOI: 10.1080/21505594.2021.1898790 21. Schroth J, Weber V, Jones TF, Del Arroyo AG, Henson SM, Ackland GL. Preoperative lymphopaenia, mortality, and morbidity after elective surgery: systematic review and meta-analysis. Br J Anaesth (2021) 127 32-40. DOI: 10.1016/j.bja.2021.02.023 22. Danese E, Lippi G, Montagnana M. Red blood cell distribution width and cardiovascular diseases. J Thorac Dis (2015) 7 E402-11. DOI: 10.3978/j.issn.2072-1439.2015.10.04 23. Ferreira JP, Lamiral Z, Bakris G, Mehta C, White WB, Zannad F. Red cell distribution width in patients with diabetes and myocardial infarction: An analysis from the EXAMINE trial. Diabetes Obes Metab (2021) 23 1580-7. DOI: 10.1111/dom.14371 24. Liu Y, Han M, Zhao J, Kang L, Ma Y, Huang B. Systematic Review and Meta-analysis of Current Literature on Isolated Abdominal Aortic Dissection. Eur J Vasc Endovasc Surg (2020) 59 545-56. DOI: 10.1016/j.ejvs.2019.05.013 25. Sen I D, ’Oria M, Weiss S, Bower TC, Oderich GS, Kalra M. Incidence and natural history of isolated abdominal aortic dissection: A population-based assessment from 1995 to 2015. J Vasc Surg (2021) 73 1198-204. DOI: 10.1016/j.jvs.2020.07.090 26. Libby P, Mallat Z, Weyand C. Immune and inflammatory mechanisms mediate cardiovascular diseases from head to toe. Cardiovasc Res (2021) 117 2503-5. DOI: 10.1093/cvr/cvab332 27. Lawler PR, Bhatt DL, Godoy LC, Luscher TF, Bonow RO, Verma S. Targeting cardiovascular inflammation: next steps in clinical translation. Eur Heart J (2021) 42 113-31. DOI: 10.1093/eurheartj/ehaa099 28. Mauracher AA, Gujer E, Bachmann LM, Gusewell S, Pachlopnik Schmid J. Patterns of Immune Dysregulation in Primary Immunodeficiencies: A Systematic Review. J Allergy Clin Immunol Pract (2021) 9 792-802 e10. DOI: 10.1016/j.jaip.2020.10.057 29. Zidar DA, Al-Kindi SG, Liu Y, Krieger NI, Perzynski AT, Osnard M. Association of Lymphopenia With Risk of Mortality Among Adults in the US General Population. JAMA Netw Open (2019) 2 e1916526. DOI: 10.1001/jamanetworkopen.2019.16526 30. Zierath D, Tanzi P, Shibata D, Becker KJ. Cortisol is More Important than Metanephrines in Driving Changes in Leukocyte Counts after Stroke. J Stroke Cerebrovasc Dis (2018) 27 555-62. DOI: 10.1016/j.jstrokecerebrovasdis.2017.09.048 31. Liesz A, Ruger H, Purrucker J, Zorn M, Dalpke A, Mohlenbruch M. Stress mediators and immune dysfunction in patients with acute cerebrovascular diseases.. PLoS One (2013) 8 e74839. DOI: 10.1371/journal.pone.0074839 32. Adamo S, Chevrier S, Cervia C, Zurbuchen Y, Raeber ME, Yang L. Profound dysregulation of T cell homeostasis and function in patients with severe COVID-19. Allergy (2021) 76 2866-81. DOI: 10.1111/all.14866 33. Cao C, Yu M, Chai Y. Pathological alteration and therapeutic implications of sepsis-induced immune cell apoptosis. Cell Death Dis (2019) 10 782. DOI: 10.1038/s41419-019-2015-1 34. Luo W, Sun JJ, Tang H, Fu D, Hu ZL, Zhou HY. Association of Apoptosis-Mediated CD4(+) T Lymphopenia With Poor Outcome After Type A Aortic Dissection Surgery.. Front Cardiovasc Med (2021) 8 747467. DOI: 10.3389/fcvm.2021.747467 35. Sio A, Chehal MK, Tsai K, Fan X, Roberts ME, Nelson BH. Dysregulated hematopoiesis caused by mammary cancer is associated with epigenetic changes and hox gene expression in hematopoietic cells. Cancer Res (2013) 73 5892-904. DOI: 10.1158/0008-5472.CAN-13-0842 36. Weiss G, Ganz T, Goodnough LT. Anemia of inflammation. Blood (2019) 133 40-50. DOI: 10.1182/blood-2018-06-856500 37. Xanthopoulos A, Giamouzis G, Dimos A, Skoularigki E, Starling RC, Skoularigis J. Red Blood Cell Distribution Width in Heart Failure: Pathophysiology, Prognostic Role, Controversies and Dilemmas.. J Clin Med (2022) 11. DOI: 10.3390/jcm11071951 38. Salvagno GL, Sanchis-Gomar F, Picanza A, Lippi G. Red blood cell distribution width: A simple parameter with multiple clinical applications. Crit Rev Clin Lab Sci (2015) 52 86-105. DOI: 10.3109/10408363.2014.992064 39. Feng GH, Li HP, Li QL, Fu Y, Huang RB. Red blood cell distribution width and ischaemic stroke. Stroke Vasc Neurol (2017) 2 172-5. DOI: 10.1136/svn-2017-000071 40. Talarico M, Manicardi M, Vitolo M, Malavasi VL, Valenti AC, Sgreccia D. Red Cell Distribution Width and Patient Outcome in Cardiovascular Disease: A ’’Real-World’’ Analysis.. J Cardiovasc Dev Dis (2021) 8. DOI: 10.3390/jcdd8100120 41. Li J, Li DE, Hu M, Huang H, Xu S, Li H. Red blood cell distribution width and tumor necrosis factor-alpha for the early prediction of coronary artery lesion in Kawasaki disease: a retrospective study. Eur J Pediatr (2022) 181 903-9. DOI: 10.1007/s00431-021-04252-3 42. Lorente L, Martin MM, Abreu-Gonzalez P, Sole-Violan J, Ferreres J, Labarta L. Red blood cell distribution width during the first week is associated with severity and mortality in septic patients. PLoS One (2014) 9 e105436. DOI: 10.1371/journal.pone.0105436 43. Duchnowski P, Szymanski P, Orlowska-Baranowska E, Kusmierczyk M, Hryniewiecki T. Raised red cell distribution width as a prognostic marker in aortic valve replacement surgery. Kardiol Pol (2016) 74 547-52. DOI: 10.5603/KP.a2015.0213 44. Al-Kindi SG, Zidar DA, McComsey GA, Longenecker CT. Association of Anisocytosis with Markers of Immune Activation and Exhaustion in Treated HIV.. Pathog Immun. (2017) 2 138-50. DOI: 10.20411/pai.v2i1.199 45. Ganz T, Nemeth E. Hepcidin and iron homeostasis. Biochim Biophys Acta (2012) 1823 1434-43. DOI: 10.1016/j.bbamcr.2012.01.014 46. Madu AJ, Ughasoro MD. Anaemia of Chronic Disease: An In-Depth Review. Med Princ Pract (2017) 26 1-9. DOI: 10.1159/000452104 47. Jiang C, Liu A, Huang L, Liu Q, Liu Y, Geng Q. Red Blood Cell Distribution Width: A Prognostic Marker in Patients With Type B Aortic Dissection Undergoing Endovascular Aortic Repair.. Front Cardiovasc Med (2022) 9 788476. DOI: 10.3389/fcvm.2022.788476 48. Li N, Zhou H, Tang Q. Red Blood Cell Distribution Width: A Novel Predictive Indicator for Cardiovascular and Cerebrovascular Diseases. Dis Markers (2017) 2017 7089493. DOI: 10.1155/2017/7089493 49. Sincer I, Gunes Y, Mansiroglu AK, Cosgun M, Aktas G. Association of mean platelet volume and red blood cell distribution width with coronary collateral development in stable coronary artery disease. Postepy Kardiol Interwencyjnej (2018) 14 263-9. DOI: 10.5114/aic.2018.78329 50. Ornek E, Kurtul A. Relationship of mean platelet volume to lymphocyte ratio and coronary collateral circulation in patients with stable angina pectoris. Coron Artery Dis (2017) 28 492-7. DOI: 10.1097/MCA.0000000000000530 51. Sincer I, Mansiroglu AK, Aktas G, Gunes Y, Kocak MZ. Association between Hemogram Parameters and Coronary Collateral Development in Subjects with Non-ST-Elevation Myocardial Infarction. Rev Assoc Med Bras (1992) (2020) 66 160-5. DOI: 10.1590/1806-9282.66.2.160 52. Cifani N, Proietta M, Tritapepe L, Di Gioia C, Ferri L, Taurino M. Stanford-A acute aortic dissection, inflammation, and metalloproteinases: a review.. Ann Med. (2015) 47 441-6. DOI: 10.3109/07853890.2015.1073346 53. Li Y, Ren P, Dawson A, Vasquez HG, Ageedi W, Zhang C. Single-Cell Transcriptome Analysis Reveals Dynamic Cell Populations and Differential Gene Expression Patterns in Control and Aneurysmal Human Aortic Tissue. Circulation (2020) 142 1374-88. DOI: 10.1161/CIRCULATIONAHA.120.046528
--- title: Differential proteomic analysis and pathogenic effects of outer membrane vesicles derived from Acinetobacter baumannii under normoxia and hypoxia authors: - Sachio Suzuki - Phawinee Subsomwong - Kouji Narita - Noriaki Kawai - Takahito Ishiai - Wei Teng - Rojana Sukchawalit - Akio Nakane - Sadatomo Tasaka - Krisana Asano journal: PLOS ONE year: 2023 pmcid: PMC10016710 doi: 10.1371/journal.pone.0283109 license: CC BY 4.0 --- # Differential proteomic analysis and pathogenic effects of outer membrane vesicles derived from Acinetobacter baumannii under normoxia and hypoxia ## Abstract Acinetobacter baumannii is a major causative agent of nosocomial infections and its outer membrane vesicles (AbOMVs) have been shown to be involved in pathogenicity by transporting virulence factors and transferring information for communication between pathogens and host cells. Despite the fact that the infected sites of A. baumannii such as lungs and skin soft tissues are hypoxic, most studies on AbOMV virulence have used AbOMVs prepared under aerobic conditions. The present study aims to elucidate the protein profile and pathogenic impact of AbOMVs released under hypoxic condition. AbOMVs were isolated from A. baumannii under normoxic and hypoxic conditions, and their protein profiles were compared. The different effects of both normoxic and hypoxic AbOMVs in cytokine response from mouse macrophages, cytotoxicity to the human lung epithelial cells, and bacterial invasion were then investigated. Our results showed that A. baumannii under hypoxia released larger amounts of OMVs with different protein profiles. Although the cytotoxic effect of AbOMVs from normoxia and hypoxia were comparable, AbOMVs from normoxia induced higher TNF-α production and invasion of Staphylococcus aureus and Pseudomonas aeruginosa than those from hypoxia. On the other hand, AbOMVs significantly enhanced A. baumannii invasion into lung epithelial cells in a dose-dependent manner. These results clearly demonstrate that AbOMVs released from normoxic and hypoxic have different impacts in pathogenesis. This finding provides new insight into the complex interactions between A. baumannii, coinfecting pathogens and host cells via OMVs, in particular the different pathogenic effects of AbOMVs under normoxic and hypoxic conditions. ## Introduction Acinetobacter baumannii is an aerobic Gram-negative coccobacillus that is one of the causative agents of nosocomial infections [1]. It causes ventilator-associated pneumonia, blood stream infections, urinary tract infections, meningitis, and wound infections [2]. In recent years, the emergence of multidrug-resistant A. baumannii strains which are more difficult to treat than antibiotic-susceptible species has become one critical priority pathogen, and the World Health Organization is focusing on this issue [3–5]. Outer membrane vesicles (OMVs) released from Gram-negative bacteria have attracted attention as a virulence factor. The OMVs are generated by budding off from the outer membrane and are spherical in shape, 20–200 nm in diameter, surrounded by a lipid bilayer. These vesicles are composed of lipopolysaccharides (LPS), membrane proteins, lipids, cilia, cytosolic proteins, DNA, and RNA [6, 7]. The ligands on the membrane surface can direct OMVs to bind multiple targets. Therefore, it has been suggested that the functions of OMVs include the transport of virulence factors such as outer membrane proteins, phospholipases and lipopolysaccharide, information transfer and communication between pathogens and host cells [8, 9]. In addition, OMVs are considered to be stable vehicles because the enclosed structure of OMVs is able to protect the functional genes, DNA, RNA, proteins and other substances from degrading enzymes such as DNases, RNases and proteases [10]. The pathogenicity of A. baumannii during infectious process is associated with adhesion and invasion into the host cells, biofilm formation, induction of host cell death, and stimulation of host immune response [11, 12]. A major virulence factor is outer membrane protein A (OmpA), which is associated with adhesion, invasion, and host cell apoptosis [13]. Since OMVs can deliver the virulence factors to host cells, there have been several reports on the presence of virulence factors including OmpA in AbOMVs and their pathogenicity such as host cell death and immune responses [14]. However, most studies on AbOMV virulence have used AbOMVs prepared from aerobic conditions [15], despite the fact that the infected sites such as lungs and skin soft tissues are hypoxic [16]. Importantly, there are no reports on the amount or pathogenicity of AbOMVs prepared from hypoxic conditions, even though other bacteria such as Pseudomonas aeruginosa has been shown to produce large amount of OMVs when cultured under stress conditions such as hypoxia [17, 18]. The skin and lungs are the most common sites of A. baumannii infection in which other bacteria such as Staphylococcus aureus and P. aeruginosa often coexist. The polymicrobial infections by these pathogens are associated with severe prognosis of the diseases. One reason for this may be that the antibiotics are less effective. Carbapenem-resistant A. baumannii has been reported to protect other Gram-negative bacteria from exposure to β-lactam antibiotics [19]. It has also been reported that carbapenem-resistant A. baumannii enhanced carbapenem resistance in carbapenem-susceptible S. aureus [20]. Thus, these bacteria are thought to interact with each other and increase virulence as well as tolerance. However, the direct effect of AbOMVs on virulence of these coexisting pathogens has not been reported. In this study, we investigated whether there is difference in the production and protein profiles of AbOMVs prepared under hypoxic condition (hAbOMVs) and under normoxic condition (nAbOMVs). In addition to the production and proteomic analysis, the pathogenic effects of hAbOMVs and nAbOMVs on host immune response, cytotoxicity, and A. baumannii invasion were compared. Moreover, the effects of these AbOMVs on the invasion of coexisting pathogens, S. aureus and P. aeruginosa, were also examined. ## Bacterial strains and growth conditions A. baumannii ATCC19606, S. aureus ATCC1718, and P. aeruginosa ATCC15692 were used in this study. They were cultured at 37°C in tryptic soy broth (TSB; BD Bioscience, Sparks, MD). For purification of AbOMVs, A. baumannii was precultured for 16 h in TSB and used to inoculate into 4.8 L of Brain Heart Infusion medium (BD Bioscience) with $0.5\%$ inoculum. The cultures were then placed under normoxic and hypoxic conditions for nAbOMVs and hAbOMVs, respectively. For normoxic condition, the cultures (400 mL medium in 1 L baffled flasks) were shaken at 125 rpm for 24 h, whereas for hypoxic condition, the cultures (400 mL medium in 500 mL normal conical flasks) were placed under static condition for 48 h. Thereafter, the supernatant was collected by centrifugation twice at 3,800 ×g, 4°C for 60 min, filtrated through 0.45 μm filter (Nalgene Rapid-Flow Filters, Thermo Fisher Scientific, Waltham, MA) to remove remaining bacterial cells and kept -80°C until AbOMVs purification. It should be noted that the oxygen concentration in uninoculated medium was measured using a JPB-70A dissolved oxygen analyzer (Shen Zhen Yage Technology, Guang Dong, China) [21]. The oxygen concentration for normoxia and hypoxia was 6.5 mg/L and 1.0 mg/L, respectively. For bacterial infection assay, A. baumannii, S. aureus, and P. aeruginosa were precultured for 16 h and inoculated into 20 mL of TSB. The cultures were incubated at 37°C under aerobic condition by shaking at 125 rpm for 5 h. The bacterial cells were collected, washed twice with phosphate-buffered saline (PBS), and adjusted to an appropriate concentration by calculating via conversion of optical density at 600 nm. ## Cells and culture conditions Mouse macrophage RAW264.7 cells were cultured at 37°C under $5\%$ CO2 in Dulbecco’s Modified Eagle Medium (DMEM; Nissui Pharmaceutical Co., Tokyo, Japan), supplemented with $10\%$ Fetal Calf Serum (FCS; JRH Biosciences, Lenexa, KS), $0.03\%$ L-glutamine (FUJIFILM Wako Pure Chemical Industries, Osaka, Japan), and antibiotic-antifungal combination agent (Gibco™ Antibiotic-Antimycotic; Thermo Fisher Scientific, Waltham, MA). Human alveolar basal epithelial A549 cells were cultures at 37°C under $5\%$ CO2 in Ham’s F-12K medium (FUJIFILM Wako Pure Chemical Corporation), supplemented with $10\%$ FCS, and antibiotic-antifungal combination agent. ## Purification of AbOMVs AbOMVs in the culture supernatant of A. baumannii were harvested by ultracentrifugation at 128,400 ×g, 4°C for 90 min using a Himac CP80NX Preparative Ultracentrifuge (HITACHI, Tokyo, Japan). After removal of the supernatant, crude pellet containing AbOMVs was then washed and suspended in 2 mL of PBS. AbOMVs were further purified using an OptiPrep™ density gradient according to the standard protocol [22] with some modifications. Briefly, a discontinuous iodixanol gradient [$45\%$, $35\%$, $30\%$, $25\%$ and $15\%$ (w/v) iodixanol solutions (OptiPrep™; Sigma Aldrich, St. Louis, MO) in 0.25 M sucrose/10 mM Tris, pH 7.5] was prepared. A 320 μL volume of crude-AbOMVs suspension was overlaid on the discontinuous iodixanol gradient and ultracentrifuged for 16 h at 100,300 ×g, 4°C. Thereafter, six fractions (720 μL each) were collected from the top of the gradient. Then, each fraction was diluted in PBS, and again harvested by ultracentrifugation at 99,800 ×g, 4°C for 3 h. The pellet was washed once and resuspended in an appropriate volume of PBS. Protein concentration of each fraction was measured by Bradford protein assay using Bio-Rad Protein Dye Reagent Concentrate (Bio-Rad Laboratories, Inc., Hercules, CA). The AbOMVs containing fraction was analyzed under transmission electron microscope (JEM-1230, JEOL, Tokyo, Japan) with negative staining (TI blue, Nisshin EM Co. Ltd, Tokyo, Japan). The particle concentration and size distribution of AbOMVs were analyzed using qNano instrument (Izon Science, Oxford, United Kingdom) according to the manufacture’s instruction. Briefly, the purified AbOMVs were diluted in Measurement Electrolytes (qNano reagent kit) containing $0.03\%$ Tween 20 (FUJIFILM Wako Pure Chemical Corporation) and analyzed using NP200 nanopore membranes in comparison with 200-nm calibration particles by running at 0.62 V. Izon Control Suite software version 3.3.2.2001 was used for data analysis. ## Proteomic analysis of AbOMVs Quantitative proteomic analyses of nAbOMVs and hAbOMVs were performed in triplicate by liquid chromatography-tandem mass spectrometry as described previously [23]. Briefly, the acetone-precipitated proteins of AbOMVs were denatured with $50\%$ trifluoroethanol and reduced with 4 mM dithiothreitol. Free cysteine residues were alkylated prior to trypsinization. The peptides were desalted and separated using liquid chromatography. The mass spectrometer was set in an information-dependent acquisition mode. Acquired spectra were searched against the A. baumannii ATCC19606 proteins (NCBI Genbank: CP045110.1). Positive identification was considered when identified proteins and peptides reached a $1\%$ local false discovery rate. The proteomic data obtained in this study are available using accession number PXD037014 and JPST001865 for Proteome Xchang and jPOST Repository, respectively. Differential proteomics between nAbOMVs and hAbOMVs was then analyzed. Compute pI/Mw tool-ExPASy (https://web.expasy.org/compute_pi/), PsortB v.3.0 (https://www.psort.org/psortb/) were used for predicting molecular weight and subcellular localization of each identified protein, respectively. ## Cytokine assay RAW264.7 cells (1 × 106 cells/mL/well) were seeded in 24-well culture plates and stimulated with 12.5 μg/mL AbOMVs for 48 h. The culture supernatant was then collected by centrifugation twice at 5,000 ×g, 4°C for 10 min to remove the residual cells. The titers of TNF-α, IL-6, and IL-10 were measured using Mouse ELISA kits (Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions. ## Cytotoxic effect of AbOMVs A549 cells (1 × 104 cells/100 μL/well) were seeded in 96-well plates and incubated with 0, 1, 5, 10, 25 and 50 μg/mL AbOMVs at 37°C under $5\%$ CO2. At 12 h of incubation, A549 cells were washed once with Hanks’ Balanced Salt Solution (HANKS; Nissui Pharmaceutical Co.) to remove AbOMVs. Cells were then maintained in 100 μL medium prior to cell viability assay. WST-1 Cell Proliferation Reagent (Sigma-Aldrich) was used for determining cell viability was determined using according to the manufacturer’s instruction. Briefly, WST-1 reagent (10 μL) was added into each well directly and incubated at 37°C under $5\%$ CO2. After color development, the optical density at 450 nm was measured using a microplate reader (MULTISKAN Sky, Thermo Fisher Scientific). The absorbance values were subtracted with background controls (medium without A549 cells containing 0, 1, 5, 10, 25 and 50 μg/mL AbOMVs, respectively). The viability of A549 cells without AbOMVs is referred to $100\%$. ## Bacterial infection assay A549 cells (5 × 104 cells/mL/well) were seeded in 24-well plates and incubated at 37°C, under $5\%$ CO2. At 48 h after incubation, A549 cells were washed and infected with A. baumannii, S. aureus, or P. aeruginosa at MOI = 50 in the presence of AbOMVs. The infection time was 5 h for A. baumannii, and 2 h for S. aureus and P. aeruginosa. Afterwards, the bacterial cells were removed, and the cells were washed twice with HANKS solution and once with PBS. To eliminate all extracellular A. baumannii and S. aureus, the A549 cells were treated with $0.2\%$ Lysostaphin (FUJIFILM Wako Pure Chemical Corporation). On the other hand, the A549 cells were treated with 120 μg/mL of gentamicin (Wako Pure Chemical Corporation) to eliminate the extracellular P. aeruginosa. At 1 h of antibiotic treatment, the cells were then washed with HANKS solution and once with PBS again. To enumerate the invading A. baumannii, the A549 cells were lysed with $0.2\%$ (3-[(3-Cholamidopropyl)-dimethylammonio]-1-propanesulfonate (CHAPS; DOJINDO, Kumamoto, Japan) for 1 h. To enumerate the invading S. aureus and P. aeruginosa, the A549 cells were lysed with $1\%$ CHAPS for 15 min. Then, the bacterial cell suspensions were diluted and plated on tryptic soy agars. The colonies were counted at 24 h after incubation. ## Statistical analysis Statistical differences were analyzed using the method mentioned in each figure legend. A P-value less than 0.05 is considered statistically significant. ## Purification of nAbOMVs and hAbOMVs To prepare nAbOMVs and hAbOMVs, A. baumannii was cultured under aerobic and static condition, respectively. The growth of A. baumannii under aerobic condition reached an OD600nm of about 2.0 after 24 h of incubation, and that under static condition reached an OD600nm of about 1.0 after 48 h. The culture supernatants from both conditions were then collected and subsequently purified by step-gradient ultracentrifugation. From six separated fractions (F1 to F6), the purified AbOMVs from each culture condition were detected in F2 as confirmed by negative-staining transmission electron microscopy (Fig 1A and 1B). The particles with an enclosed structure of approximately 200 nm in diameter were observed. Nanoparticle tracking analysis revealed that the average particle sizes of nAbOMVs and hAbOMVs were very similar, with 187 nm and 189 nm, respectively (Table 1). From 1 μg of protein, the number of OMV particles from normoxia and hypoxia was slightly different with 2.71 × 108 and 2.19 × 108 particles, respectively. Obviously, from an equivalent volume of culture supernatant (4.8 L), the protein yield and total particle number of hAbOMVs were approximately 3 times higher than those of nAbOMVs (Table 1 and S1 Dataset). Fig 1: Negative-staining transmission electron micrographs of (A) nAbOMVs and (B) hAbOMVs.A. baumannii ATCC19606 were cultured under normoxic and hypoxic conditions for 24 h and 48 h, respectively. After removal of bacterial cells, nAbOMVs and hAbOMVs in the culture supernatant from normoxic and hypoxic conditions were collected and purified by step-gradient ultracentrifugation. AbOMVs (white arrows) from both conditions were detected in fraction F2. TABLE_PLACEHOLDER:Table 1 ## Differential proteomic analysis between nAbOMVs and hAbOMVs To investigate the differences in protein profiles between nAbOMVs and hAbOMVs, quantitative proteome analysis was performed. A total of 584 proteins were identified from the triplicated preparations of nAbOMVs and hAbOMVs. The differential comparisons between nAbOMVs and hAbOMVs indicated that 30 proteins were significantly higher in nAbOMVs ($P \leq 0.05$) (Table 2). Among these, six proteins are hypothetical proteins with unknown function. Alkyl hydroperoxide reductase C was detected only in nAbOMVs. The proteins enriched more than 10-fold in nAbOMVs were polysaccharide biosynthesis tyrosine autokinase, dihydrolipoamide acetyltransferase component of pyruvate dehydrogenase complex, outer-membrane lipoprotein carrier protein, and polysaccharide biosynthesis export family protein. The proportion of cytosolic, cytoplasmic membrane, periplasmic. outer membrane, extracellular and unknown location proteins was 7:6:3:2:1:11. The proteins that have been shown to associate with A. baumannii virulence were outer membrane receptor FepA, polysaccharide biosynthesis tyrosine autokinase and taurine ABC transporter substrate-binding protein [24]. Table 2 \| Gene name \| Protein \| Predicted MW (kDa) \| Predicted location* \| Fold- change \| P-value \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| ahpC \| Alkyl hydroperoxide reductase C \| 20.8 \| C \| Infinity† \| 0.020 \| \| ptk \| Polysaccharide biosynthesis tyrosine autokinase \| 81.6 \| Cym \| 27.24 \| <0.001 \| \| aceF \| Dihydrolipoamide acetyltransferase component of pyruvate dehydrogenase complex \| 68.4 \| C \| 23.19 \| 0.020 \| \| lolA \| Outer-membrane lipoprotein carrier protein \| 25.3 \| P \| 10.46 \| <0.001 \| \| wza \| Polysaccharide biosynthesis export family protein \| 40.6 \| O \| 10.18 \| 0.019 \| \| icmF \| Type VI secretion protein \| 143.9 \| Cym \| 9.28 \| 0.049 \| \| HMPREF0010_02682 \| Hypothetical protein \| 24.4 \| Unk \| 9.02 \| 0.004 \| \| DOL94_08760 \| Protein CsuA \| 19.7 \| Unk \| 8.49 \| 0.017 \| \| HMPREF0010_02860 \| SPFH/Band 7/PHB domain protein \| 31.0 \| Unk \| 7.26 \| <0.001 \| \| HMPREF0010_02513 \| Penicillin-binding protein 1A \| 94.8 \| E \| 5.18 \| 0.033 \| \| ggt \| Glutathione hydrolase proenzyme \| 69.8 \| P \| 4.77 \| 0.047 \| \| fadA \| Acetyl-CoA C-acyltransferase FadA \| 41.1 \| C \| 4.20 \| 0.009 \| \| acrA \| Acriflavine resistance protein A/MULTISPECIES \| 43.8 \| Unk \| 4.12 \| 0.001 \| \| cydA \| Cytochrome bd oxidase subunit I \| 58.7 \| Cym \| 4.11 \| 0.048 \| \| HMPREF0010_02797 \| Sel1 repeat protein \| 29.6 \| Unk \| 3.96 \| 0.019 \| \| ATCC19606_33910 \| LprI domain-containing protein \| 14.2 \| Unk \| 3.95 \| 0.049 \| \| queF \| NADPH-dependent 7-cyano-7-deazaguanine reductase QueF \| 31.0 \| C \| 3.91 \| 0.030 \| \| mqo \| Malate dehydrogenase \| 60.4 \| C \| 3.35 \| 0.005 \| \| HMPREF0010_00307 \| Hypothetical protein \| 19.1 \| Unk \| 3.35 \| 0.004 \| \| tauA \| Taurine ABC transporter substrate-binding protein \| 38.0 \| P \| 2.92 \| 0.006 \| \| HMPREF0010_02833 \| Hypothetical protein \| 18.8 \| Unk \| 2.88 \| 0.041 \| \| tolA \| Gramicidin S synthase \| 50.6 \| Unk \| 2.75 \| 0.030 \| \| atpB \| ATP synthase subunit A \| 32.4 \| Cym \| 2.70 \| 0.025 \| \| ATCC19606_13250 \| RDD family protein \| 28.5 \| Cym \| 2.62 \| 0.027 \| \| ATCC19606_04350 \| Hypothetical protein \| 38.6 \| Unk \| 2.57 \| 0.004 \| \| hlyD \| RND transporter \| 40.7 \| Cym \| 2.09 \| 0.050 \| \| HMPREF0010_02162 \| Hypothetical protein \| 17.7 \| Unk \| 1.95 \| 0.028 \| \| A7M90_10095 \| Hypothetical protein \| 46.6 \| C \| 1.78 \| 0.040 \| \| dnaK \| Chaperone protein DnaK \| 69.4 \| C \| 1.77 \| 0.006 \| \| HMPREF0010_01517 \| Outer membrane receptor FepA (TonB-dependent siderophore receptor) \| 82.8 \| O \| 1.60 \| 0.045 \| The proteins enriched in hAbOMVs with P value < 0.05 are shown in Table 3. Of these 25 proteins, two are hypothetical proteins with unknown function and several proteins are enzymes involving in metabolism. GntR family transcriptional regulator was detected only in hAbOMVs. The proteins enriched more than 10-fold in hAbOMVs were haloacid dehalogenase (HAD) hydrolase, family IB and methionine aminopeptidase. The proportion of cytosolic, cytoplasmic membrane, periplasmic. outer membrane, extracellular and unknown location proteins was 14:5:0:0:0:6. Among these proteins, outer-membrane lipoprotein LolB, and high-affinity zinc transporter ATPase are lipoproteins that have been shown to associate with A. baumannii virulence [24]. Table 3 \| Gene name \| Protein \| Predicted MW (kDa) \| Predicted location* \| Fold- change \| P-value \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| HMPREF0010_01056 \| GntR family transcriptional regulator \| 24.8 \| C \| Infinity† \| 0.020 \| \| HMPREF0010_02207 \| HAD hydrolase, family IB \| 25.0 \| C \| 21.81 \| <0.001 \| \| map \| Methionine aminopeptidase \| 30.4 \| C \| 19.84 \| 0.035 \| \| gtr6 \| Glycosyl transferase \| 45.1 \| C \| 9.81 \| 0.048 \| \| ftsI \| Peptidoglycan D, D-transpeptidase FtsI \| 67.7 \| Cym \| 8.67 \| 0.041 \| \| HMPREF0010_01031 \| IclR helix-turn-helix domain protein \| 32.0 \| C \| 8.49 \| 0.049 \| \| lolB \| Outer-membrane lipoprotein LolB \| 21.1 \| Unk \| 8.16 \| 0.036 \| \| bioA \| Adenosylmethionine-8-amino-7-oxononanoate aminotransferase \| 47.8 \| C \| 6.70 \| 0.007 \| \| accD \| Acetyl-coenzyme A carboxylase carboxyl transferase subunit beta \| 33.0 \| C \| 6.42 \| 0.012 \| \| rpsS \| 30S ribosomal protein S19 \| 10.2 \| C \| 5.90 \| 0.004 \| \| HMPREF0010_03595 \| DNA helicase \| 55.3 \| C \| 5.50 \| 0.034 \| \| pyrG \| CTP synthase \| 61.0 \| C \| 4.41 \| 0.022 \| \| HMPREF0010_00687 \| YbaB/EbfC family nucleoid-associated protein \| 12.0 \| Unk \| 4.31 \| 0.011 \| \| pyrD \| Dihydroorotate dehydrogenase \| 36.0 \| Cym \| 3.76 \| 0.023 \| \| rpsT \| 30S ribosomal protein S20 \| 9.7 \| C \| 3.74 \| 0.044 \| \| B9X95_03935 \| DUF799 domain-containing protein \| 24.1 \| Unk \| 3.72 \| 0.007 \| \| HMPREF0010_01246 \| Ribonucleotide-diphosphate reductase subunit beta \| 48.9 \| C \| 3.50 \| 0.039 \| \| hslR \| Heat shock protein 15 \| 17.1 \| C \| 3.38 \| 0.004 \| \| rpmB \| 50S ribosomal protein L28 \| 9.1 \| C \| 3.30 \| 0.012 \| \| znuC \| High-affinity zinc transporter ATPase \| 29.5 \| Cym \| 3.10 \| 0.022 \| \| HMPREF0010_00917 \| Hypothetical protein \| 35.0 \| Unk \| 3.07 \| 0.001 \| \| ATCC19606_09150 \| DUF1615 domain-containing protein \| 45.8 \| Unk \| 2.51 \| 0.020 \| \| ATCC19606_25890 \| Hypothetical protein \| 13.1 \| Unk \| 2.38 \| 0.016 \| \| HMPREF0010_02256 \| Phosphate ABC transporter, phosphate-binding protein \| 37.0 \| Cym \| 1.91 \| 0.031 \| \| metQ_1 \| MetQ/NlpA family ABC transporter substrate-binding protein \| 31.3 \| Cym \| 1.42 \| 0.003 \| \| omp38 \| Outer membrane protein OmpA‡ \| 38.4 \| O \| 2.20 \| 0.252 \| Since OmpA is a key virulence factor of A. baumannii, the amount of OmpA in nAbOMVs and hAbOMVs was also focused. As shown in Table 3, the amount of OmpA was 2.2-fold higher in hAbOMVs than that in nAbOMVs, but the statistical analysis shows no significant difference ($$P \leq 0.252$$). These proteomic data demonstrated that the types and amounts of several proteins involved in virulence of A. baumannii differed between nAbOMVs and hAbOMVs. Therefore, we further examined the difference in the effects of nAbOMVs and hAbOMVs on host immune response, cytotoxicity, and bacterial infections. ## Effects of nAbOMVs and hAbOMVs on host immune responses To examine the different effects of nAbOMVs and hAbOMVs on host immune responses, mouse macrophage RAW264.7 cells were prepared and stimulated with 12.5 μg/mL of each AbOMVs. At 48 h of stimulation, cytokines in the culture supernatant were measured by ELISAs. As shown in Fig 2, the purified AbOMVs from both normoxia and hypoxia significantly stimulated the production of TNF-α, IL-6, and IL-10 from RAW264.7 cells compared to non-AbOMV control. These results suggest that AbOMVs from both conditions strongly stimulate host immune responses. In addition, TNF-α production by nAbOMV stimulation was significantly higher than that by hAbOMVs (Fig 2A and S2 Dataset). However, this different effect was not observed in IL-6 and IL-10 production (Fig 2B and 2C, and S2 Dataset). Fig 2: Cytokine production from mouse macrophages after stimulation with nAbOMVs and hAbOMVs.RAW264.7 cells (1 × 106 cells/mL/well) were incubated with 12.5 μg/mL AbOMVs for 48 h at 37°C. The production of (A) TNF-α, (B) IL-6, and (C) IL-10 in the culture supernatant was determined by ELISAs (n = 6 from 2 independent experiments). P-value was calculated using Mann-Whitney U-test. Asterisk () indicates significant difference between nAbOMVs and hAbOMVs (P < 0.01). ns: not significant. ## Cytotoxic effects of nAbOMVs and hAbOMVs to human lung epithelial cells The cytotoxic effects of nAbOMVs and hAbOMVs to A549 lung epithelial cells were examined. At 12 h after incubation of A549 cells with AbOMVs, WST-1 reagent was added, and the absorbance at 450 nm was measured. The cytotoxic effects of nAbOMVs and hAbOMVs to A549 cells are shown in Fig 3 and S3 Dataset. In the presence of AbOMVs, the cell viability was decreased compared to non-AbOMV control. The decrease in cell viability was dose-dependent up to a concentration of 10 μg/mL. At AbOMVs concentrations above 10 μg/mL, the cell viability remained constant at approximately $63\%$. nAbOMVs and hAbOMVs reduced cell viability with a similar trend with no significant difference. Fig 3: Cytotoxic effect of nAbOMVs and hAbOMVs to lung epithelial cells.A549 cells (1 × 104 cells/100 μL/well) were seeded in 96-well plate and incubated with AbOMVs at 37°C under 5%CO2. At 12 h of incubation, 10 μL of WST-1 reagent was added into each well and the absorbance was measured at 450 nm. Cell viability in the reactions without AbOMVs (control) was calculated as 100% (n = 7 from 3 independent experiments). P-value was calculated using Kruskal Wallis H-test with post hoc Mann-Whitney U-test. Asterisks indicates significant difference to non-AbOMV control (: P < 0.05, *: P < 0.01). P-value of difference between nAbOMVs and hAbOMVs was calculated using Mann-Whitney U-test. ns: not significant. ## Effects of nAbOMVs and hAbOMVs on bacterial infections into A549 cells The effect of nAbOMVs and hAbOMVs on A. baumannii infection into the A549 cells was examined. A549 cells were infected with A. baumannii (MOI = 50) in the presence of AbOMVs for 5 h, then the invading A. baumannii was enumerated. The results in Fig 4A and S4 Dataset are expressed as relative numbers to the invading A. baumannii without AbOMVs (set at 1.0). At various concentrations of nAbOMVs, the numbers of A. baumannii invading into A549 cells were constant and comparable to that of non-AbOMV control. In contrast, hAbOMVs promoted the number of A. baumannii invading into A549 cells in a dose-dependent manner. The invading numbers of A. baumannii in the presence of 12.5–50 μg/mL nAbOMVs and hAbOMVs were significantly different. Fig 4: Effect of nAbOMVs and hAbOMVs on bacterial infections into lung epithelial cells.A549 cells (5 × 104 cells/mL/well) were seeded in a 24-well plate. (A) A549 cells were infected with A. baumannii at MOI = 50 in the presence of AbOMVs for 5 h (n = 6 from 2 independent experiments). Bacterial number from the condition without AbOMVs is referred to 1.0 (dotted line). (B) A549 cells were infected with S. aureus or P. aeruginosa at MOI = 50 in the presence of 12.5 μg/mL AbOMVs for 2 h (n = 6 from 2 independent experiments). Invasion efficacy was expressed as the relative number to the condition without AbOMVs, which is referred to 1.0 (dotted line). P-value of difference to non-AbOMV control was calculated using Kruskal Wallis H-test with post hoc Mann-Whitney U-test. : P < 0.05, *: P < 0.01. P-value of difference between nAbOMVs and hAbOMVs was calculated using Mann-Whitney U-test. †: P < 0.05, ‡: P < 0.01. The effects of AbOMVs on infections of the common coexisting pathogens, S. aureus and P. aeruginosa, were also investigated. A549 cells were infected with S. aureus or P. aeruginosa (MOI = 50) in the presence of 12.5 mg/mL of nAbOMVs or hAbOMVs for 2 h, and the invading bacterial cells were then enumerated. The results are expressed as relative numbers to the invading S. aureus or P. aeruginosa without AbOMVs (set at 1.0). As shown in Fig 4B and S4 Dataset, nAbOMVs promoted the invasion of both S. aureus and P. aeruginosa into A549 cells compared to non-AbOMV control (referred to 1.0). On the other hand, hAbOMVs enhanced the invading number of S. aureus but not P. aeruginosa. Overall infection results showed that nAbOMVs significantly promoted invasion of S. aureus and P. aeruginosa more than hAbOMVs. This was inconsistent with their effects on A. baumannii infection (Fig 4A). ## Discussion A. baumannii is a major causative agent of nosocomial infections [2]. At infected sites such as skin and lungs, this microorganism often co-infects with S. aureus and P. aeruginosa, which results in severe prognosis [25, 26]. It has been known that the extracellular membrane vesicles released from bacteria play an important role in bacterial pathogenicity. Based on structure and composition, these vesicles can deliver the virulence factors directly to the host cells and transmit signal information to other pathogens for promoting virulence [14, 27]. The composition and virulence of AbOMVs prepared under aerobic condition, which is suitable for A. baumannii growth, have been reported [28]. However, it has been reported that infected tissues are hypoxic because of increased oxygen consumption due to activation of immune-related cells and decreased oxygen supply due to damage to blood vessels. Thus, while normal lung tissue is in contact with the atmosphere and have a high oxygen concentration, the infected area is thought to be hypoxia [16]. Under this hypoxic stress, bacteria need to alter their gene expression profiles to survive, especially the proteins related to nutrient transport and metabolism [29]. It has been demonstrated that the amount of OMVs released from P. aeruginosa is altered under hypoxia [17]. Therefore, we hypothesized that the amount and protein profile of AbOMVs prepared under hypoxia would differ from those under normoxia. We also expected that the difference in the protein profiles in the AbOMVs would alter their role in virulence. Therefore, in the present study, AbOMVs under normoxic and hypoxic conditions were purified and their different effects on host immune response, cytotoxicity and bacterial infections were investigated. As shown in Table 1, there was no significant difference in the particle size and particle concentration (per 1 μg protein) between the purified nAbOMVs and hAbOMVs. These particle sizes were comparable to that of A. baumannii OMVs isolated by Sun Li, et al. [ 15]. However, the protein yield as well as the total particle number of hAbOMVs were approximately 3-fold higher than those of nAbOMVs. These obtaining yields were constant from triplicated preparations. Together with the differences in growth of A. baumannii under normoxic and hypoxic conditions, our results indicate that the hypoxia, a stress condition for A. baumannii, remarkably promotes AbOMVs production over normoxic condition. Differential proteomic analysis was performed to observe differences in the protein profiles of nAbOMVs and hAbOMVs. As shown in Tables 2 and 3, 30 proteins were significantly enriched in nAbOMVs and 25 proteins were significantly enriched in hAbOMVs. The enriched proteins in nAbOMVs that have been shown to play a role in A. baumannii virulence are the outer membrane receptor FepA (TonB-dependent siderophore receptor), polysaccharide biosynthesis tyrosine autokinase and taurine ABC transporter substrate-binding protein. FepA has been shown to be involved in iron-scavenging system and promote the growth of A. baumannii [30, 31]. The tyrosine autokinase is markly elevated to export the exopolysaccharide during biofilm formation [32], and the taurine ABC transporter substrate-binding protein is involved in the transport of taurine as a sulfur source and in competition for taurine utilization between the nosocomial pathogens and host [33]. Proteins enriched in hAbOMVs that have been shown to be involved in the virulence of A. baumannii include outer-membrane lipoprotein LolB and high-affinity zinc transporter ATPase. The outer-membrane lipoprotein LolB constitutes the Lol lipoprotein transport system and transports lipoproteins to the outer membrane in especially LPS-deficient A. baumannii [34]. The high-affinity zinc transporter ATPase ZnuC constitutes the ZnuABC transporter to acquire zinc within the hostile environment of the host [35]. In addition to these known proteins, the hypothetical proteins with unknown function might also contribute to the virulence of AbOMVs. Moreover, OMVs from Gram-negative bacteria are known to carry LPS, peptidoglycan as well as lipoproteins to induce inflammatory response via pattern recognition receptor-mediated interactions [36, 37]. Therefore, we further compared the cytokine responses from mouse macrophages after nAbOMVs and hAbOMVs stimulations. As shown in Fig 2, the production of TNF-α, IL-6 and IL-10 from RAW264.7 cells was significantly increased after stimulations with nAbOMVs and hAbOMVs. Although there was no significant difference in the production of IL-6 and IL-10 between nAbOMVs and hAbOMVs stimulation, the TNF-α production caused by nAbOMVs stimulation was significantly higher than that by hAbOMVs. A previous study has suggested that OmpA in OMVs is a key factor for cytokine stimulation because the OMVs from ΔOmpA mutant of A. baumannii induce lower levels of TNF-α and IL-6 compared to OMVs of the wild-type strain [14]. Although our proteomic analysis revealed that OmpA seems to be enriched in hAbOMVs for (2.20-fold change), there was no statistically significant difference (P value = 0.252). In contrast, the proportion of polysaccharide biosynthesis-related proteins and the periplasmic proteins was enriched in nAbOMVs. Thus, it is possible that polysaccharides and peptidoglycans in periplasm may be enriched in nAbOMVs, and these molecules may contribute to the high titer of TNF-α due to nAbOMVs stimulation. It has been shown that AbOMVs exhibit the cytotoxic effect to the host cells [38], and a key toxic factor is OmpA because Skerniškytė et al. demonstrated that the cytotoxicity of AbOMVs from A. baumannii lacking OmpA decreased [14]. Since our proteomic analysis revealed that the amount of OmpA in nAbOMVs and hAbOMVs was not significantly different (P value = 0.252), it was not surprising that the cytotoxic effect of nAbOMVs and hAbOMVs to human lung epithelial cells was comparable (Fig 3). Interestingly, the cytotoxic effect of both AbOMVs was not increased at the concentration above 10 μg/mL. Further experiments are required to explain the details of this mechanism. The effect of nAbOMVs and hAbOMVs on bacterial infection was also examined to determine whether nAbOMVs and hAbOMVs promote A. baumannii invasion. It has been shown that OmpA is required for A. baumannii to adhere to and invade the epithelial cells [12]. The number of ΔompA A. baumannii infections into epithelial cells significantly decreased compared to the wild type, and pretreatment of recombinant OmpA to the host cells inhibits the interaction of A. baumannii with the epithelial cells [12]. Since OmpA was comparably detected in both nAbOMVs and hAbOMVs, we expected that both AbOMVs may inhibit A. baumannii internalization. Surprisingly, our results showed that nAbOMVs did not inhibit A. baumannii invasion even the concentration was up to 50 μg/mL. In contrast, hAbOMVs significantly enhanced A. baumannii invasiveness in a dose-dependent manner (Fig 4A). These results suggest that some proteins enriched in hAbOMVs (in Table 3) may act as effectors to regulate cytoskeleton dynamics and induce A. baumannii uptake into lung epithelial cells. In particular, under stressful hypoxic condition, A. baumannii may highly release these effectors together with OMVs to promote its infectivity. A. baumannii often causes infections of the skin and lungs where P. aeruginosa and S. aureus usually coexist. We hypothesized that there may be AbOMV-mediated interactions between these bacteria. Therefore, the effect of AbOMVs on S. aureus and P. aeruginosa infection was also examined. In contrast to the effects of AbOMVs on A. baumannii infection, nAbOMVs promoted S. aureus and P. aeruginosa invasion into A549 epithelial cells. Although hAbOMVs also promoted S. aureus invasion, these vesicles did not affect P. aeruginosa invasiveness. The invasion mechanism of S. aureus into the epithelial cells is suggested to be mediated by fibronectin that forms bridging between host cell and S. aureus surface [39]. In addition, the invasion mechanism of P. aeruginosa is mediated by the formation of bacterial aggregates on the surface of epithelial cells [40]. Since these mechanisms are different from that of A. baumannii which invades into the epithelial cells via a zipper-like mechanism [12, 41], the different effectors in nAbOMVs and hAbOMVs may reflect their different effects on these bacterial infections. During infection, A. baumannii probably competes other bacterial species [42]. In particular, under stress of hypoxic condition, A. baumannii may have highly competitive with P. aeruginosa because both of them preferentially use aerobic respiration [43, 44]. Therefore, a high hAbOMVs production with the effective molecules from hypoxic condition would be a beneficial strategy for A. baumannii to promote its own survival and outcompete P. aeruginosa. It is important to noted that OMVs do not contain only proteins or lipoproteins. The other bioactive molecules such as LPS, peptidoglycan, DNA and RNA in OMVs are also possible to contribute to these effects. Although further investigation of the molecular mechanisms of AbOMVs involved in bacterial infection into the host cells is required, we expected that the proteins reported in this study would provide great benefits for progression in the research field of OMVs and their relationship to infection. Taken all together, our results indicated AbOMVs from normoxia and hypoxia exhibited different effects to promote the infections of A. baumannii and other bacteria. ## Conclusion This study demonstrated that A. baumannii under hypoxic condition produces higher amount of OMVs than that under normoxic condition. The protein profile and the pathogenic effect of AbOMVs under normoxic and hypoxic are also different. Although the cytotoxic effect of nAbOMVs and hAbOMVs to human lung epithelial cells was comparable, the OMVs released from A. baumannii under normoxia promoted higher TNF-α production and enhanced S. aureus and P. aeruginosa internalization into lung epithelial cells. On the other hand, OMVs release from A. baumannii under hypoxia promoted invasion of A. baumannii into lung epithelial cells. This finding provides new insight into the complex interactions between infecting pathogens and host mediated by OMVs, in particular the different effect under normoxic and hypoxic conditions. ## References 1. Roca I, Espinal P, Vila-Farrés X, Vila J. The. Front Microbiol (2012) 3 148. DOI: 10.3389/fmicb.2012.00148 2. Peleg AY, Seifert H, Paterson DL. Clin Microbiol Rev (2008) 21 538-82. DOI: 10.1128/CMR.00058-07 3. Dijkshoorn L, Nemec A, Seifert H. An increasing threat in hospitals: multidrug-resistant. Nat Rev Microbiol (2007) 5 939-51. DOI: 10.1038/nrmicro1789 4. Tsioutis C, Kritsotakis IE, Karageorgos AS, Stratakou S, Psarologakis C, Kokkini S. Clinical epidemiology, treatment and prognostic factors of extensively drug-resistant. Int J Antimicrob Agents (2016) 48 492-497. DOI: 10.1016/j.ijantimicag.2016.07.007 5. Shlase DM, Bradford PA. Antibiotics―from there to where? How the antibiotic miracle is threatened by resistance and a broken market and what we can do about it. Pathog Immun (2018) 3 19-43. DOI: 10.20411/pai.v3i1.231 6. Kuehn MJ, Kesty NC. Bacterial outer membrane vesicles and the host-pathogen interaction. Genes Dev (2005) 19 2645-55. DOI: 10.1101/gad.1299905 7. Kadurugamuwa JL, Beveridge TJ. Virulence factors are released from. J Bacteriol (1995) 177 3998-4008. DOI: 10.1128/jb.177.14.3998-4008.1995 8. Jun SH, Lee JH, Kim BR, Kim SI, Park TI, Lee JC. PLoS One (2013) 8 e71751. DOI: 10.1371/journal.pone.0071751 9. Ellis TN, Kuehn MJ. Virulence and immunomodulatory roles of bacterial outer membrane vesicles. Microbiol Mol Biol Rev (2010) 74 81-94. DOI: 10.1128/MMBR.00031-09 10. Caruana JC, Walper SA. Bacterial membrane vesicles as mediators of microbe—microbe and microbe—host community interactions. Front Microbiol (2020) 11 432. DOI: 10.3389/fmicb.2020.00432 11. Choi CH, Lee EY, Lee YC, Park TI, Kim HJ, Hyun SH. Outer membrane protein 38 of. Cell Microbiol (2005) 7 1127-38. DOI: 10.1111/j.1462-5822.2005.00538.x 12. Choi CH, Lee JS, Lee YC, Park TI, Lee JC. BMC Microbiol (2008) 8 216. DOI: 10.1186/1471-2180-8-216 13. Gaddy JA, Tomaras AP, Actis LA. The. Infect Immun (2009) 77 3150-60. DOI: 10.1128/IAI.00096-09 14. Skerniškytė J, Karazijaitė E, Lučiūnaitė A, Sužiedėlienė E. OmpA protein-deficient. Pathogens (2021) 10 407. DOI: 10.3390/pathogens10040407 15. Li S, Chen DQ, Ji L, Sun S, Jin Z, Jin ZL. Development of different methods for preparing. Front Immunol (2020) 11 1069. DOI: 10.3389/fimmu.2020.01069 16. Cummins EP, Keogh CE, Crean D, Taylor CT. The role of HIF in immunity and inflammation. Mol Aspects Med (2016) 47-48. DOI: 10.1016/j.mam.2015.12.004 17. Toyofuku M, Zhou S, Sawada I, Takaya N, Uchiyama H, Nomura N. Membrane vesicle formation is associated with pyocin production under denitrifying conditions in. Environ Microbiol (2014) 16 2927-38. DOI: 10.1111/1462-2920.12260 18. Macdonald IA, Kuehn MJ. Stress-induced outer membrane vesicle production by. J Bacteriol (2013) 195 2971-81. DOI: 10.1128/JB.02267-12 19. Liao YT, Kuo SC, Lee YT, Chen CP, Lin SW, Shen LJ. Sheltering effect and indirect pathogenesis of carbapenem-resistant. Antimicrob Agents Chemother (2014) 58 3983-90. DOI: 10.1128/AAC.02636-13 20. Smith NM, Ang A, Tan F, Macias K, James S, Sidhu J. Interaction of. Antimicrob Agents Chemother (2021) 65 e02414-20. DOI: 10.1128/AAC.02414-20 21. Dekic S, Hrenovic J, van Wilpe E, Venter C, Goic-Barisic I. Survival of emerging pathogen. Water Sci Technol (2019) 80 1581-1590. DOI: 10.2166/wst.2019.408 22. Théry C, Amigorena S, Raposo G, Clayton A. Isolation and characterization of exosomes from cell culture supernatants and biological fluids. Curr Protoc Cell Biol (2006). DOI: 10.1002/0471143030.cb0322s30 23. Asano K, Hirose S, Narita K, Subsomwong P, Kawai N, Sukchawalit R. Extracellular vesicles from methicillin resistant. Emerg Microbes Infect (2021) 10 2000-2009. DOI: 10.1080/22221751.2021.1991239 24. Kwon SO, Gho YS, Lee JC, Kim SI. Proteome analysis of outer membrane vesicles from a clinical. FEMS Microbiol Lett (2009) 297 150-6. DOI: 10.1111/j.1574-6968.2009.01669.x 25. Castellanos N, Nakanouchi J, Yüzen DI, Fung S, Fernandez JS, Barberis C. A study on. Curr Microbiol (2019) 76 842-847. DOI: 10.1007/s00284-019-01696-7 26. Bhargava N, Sharma P, Capalash N. N-acyl homoserine lactone mediated interspecies interactions between. Biofouling (2012) 28 813-22. DOI: 10.1080/08927014.2012.714372 27. Chatterjee S, Mondal A, Mitra S, Basu S. J Antimicrob Chemother (2017) 72 2201-2207. DOI: 10.1093/jac/dkx131 28. Jin JS, Kwon SO, Moon DC, Gurung M, Lee JH, Kim SI. PLoS One (2011) 6 e17027. DOI: 10.1371/journal.pone.0017027 29. Gil-Marqués ML, Pachón J, Smani Y. iTRAQ-based quantitative proteomic analysis of. Microbiol Spectr (2022) 10 e0232821. DOI: 10.1128/spectrum.02328-21 30. Moynié L, Luscher A, Rolo D, Pletzer D, Tortajada A, Weingart H. Structure and function of the PiuA and PirA siderophore-drug receptors from. Antimicrob Agents Chemother (2017) 61 e02531-16. DOI: 10.1128/AAC.02531-16 31. Morris FC, Dexter C, Kostoulias X, Uddin MI, Peleg AY. The mechanisms of disease caused by. Front Microbiol (2019) 10 1601. DOI: 10.3389/fmicb.2019.01601 32. Choudhary M, Kaushik S, Kapil A, Shrivastava R, Vashistt J. Decoding. Arch Microbiol (2022) 204 200. DOI: 10.1007/s00203-022-02807-y 33. Vallenet D, Nordmann P, Barbe V, Poirel L, Mangenot S, Bataille E. Comparative analysis of Acinetobacters: three genomes for three lifestyles. PLoS One (2008) 3 e1805. DOI: 10.1371/journal.pone.0001805 34. Henry R, Vithanage N, Harrison P, Seemann T, Coutts S, Moffatt JH. Colistin-resistant, lipopolysaccharide-deficient. Antimicrob Agents Chemother (2012) 56 59-69. DOI: 10.1128/AAC.05191-11 35. Hesse LE, Lonergan ZR, Beavers WN, Skaar EP. The. Infect Immun (2019) 87 e00746-19. DOI: 10.1128/IAI.00746-19 36. Glauser MP. The inflammatory cytokines. New developments in the pathophysiology and treatment of septic shock. Drugs (1996) 52 9-17. DOI: 10.2165/00003495-199600522-00004 37. Liang MD, Bagchi A, Warren HS, Tehan MM, Trigilio JA, Beasley-Topliffe LK. Bacterial peptidoglycan-associated lipoprotein: a naturally occurring toll-like receptor 2 agonist that is shed into serum and has synergy with lipopolysaccharide. J Infect Dis (2005) 191 939-48. DOI: 10.1086/427815 38. Li ZT, Zhang RL, Bi XG, Xu L, Fan M, Xie D. Outer membrane vesicles isolated from two clinical. Microb Pathog (2015) 81 46-52. DOI: 10.1016/j.micpath.2015.03.009 39. Sinha B, Fraunholz M. Int J Med Microbiol (2010) 300 170-5. DOI: 10.1016/j.ijmm.2009.08.019 40. Lepanto P, Bryant DM, Rossello J, Datta A, Mostov KE, Kierbel A. Cell Microbiol (2011) 13 1212-22. DOI: 10.1111/j.1462-5822.2011.01611.x 41. Lee JC, Koerten H, van den Broek P, Beekhuizen H, Wolterbeek R, van den Barselaar M. Adherence of. Res Microbiol (2006) 157 360-6. DOI: 10.1016/j.resmic.2005.09.011 42. Carruthers MD, Nicholson PA, Tracy EN, Munson RS. PLoS One (2013) 8 e59388. DOI: 10.1371/journal.pone.0059388 43. Alvarez-Ortega C, Harwood CS. Responses of. Mol Microbiol (2007) 65 153-65. DOI: 10.1111/j.1365-2958.2007.05772.x 44. Howard A, O’Donoghue M, Feeney A, Sleator RD. Virulence (2012) 3 243-50. DOI: 10.4161//viru.19700
--- title: 'Compensatory eating after exercise in everyday life: Insights from daily diary studies' authors: - Natalie M. Reily - Rebecca T. Pinkus - Lenny R. Vartanian - Kate Faasse journal: PLOS ONE year: 2023 pmcid: PMC10016725 doi: 10.1371/journal.pone.0282501 license: CC BY 4.0 --- # Compensatory eating after exercise in everyday life: Insights from daily diary studies ## Abstract There is considerable variability in how successful people are in losing weight via exercise programs. Experimental research suggests that greater food intake after exercise may be one factor underlying this variability, but no studies have assessed patterns of post-exercise eating behaviour over time in naturalistic settings. Thus, we aimed to assess how exercise and contextual factors (e.g., hunger, presence of others) influence the healthiness and amount of food eaten after exercise in two daily diary studies. In Study 1, participants ($$n = 48$$) reported their food intake and exercise daily for 28 days. For each meal, they provided a brief description of the food(s) eaten which were then categorised as healthy, unhealthy, or mixed (neither healthy nor unhealthy) by two independent coders. Study 2 used the same method, but participants ($$n = 55$$) also reported the portion size of each meal. Hierarchical linear modelling showed that in Study 1, contrary to expectations, post-exercise meals were less likely to be unhealthy (relative to mixed) than were random meals from non-exercise days (OR = 0.63, $$p \leq .011$$), and that participants ate proportionally fewer unhealthy meals on exercise days compared to non-exercise days (b = -4.27, $$p \leq .004$$). Study 2 replicated these findings, and also found that participants consumed larger meals after exercise in comparison to random meals from non-exercise days ($b = 0.25$, $p \leq .001$). Participants were not consistently engaging in compensatory eating by eating less healthily after exercise compared to on non-exercise days, but they did eat larger portions post-exercise. This work highlights the need for naturalistic methods of assessing compensatory eating, and has the potential to facilitate development of strategies to improve health behaviour regulation. ## Introduction Globally, $42\%$ of adults are actively trying to lose weight, and an additional $23\%$ are trying to maintain their current weight [1]. Among the most common strategies used for weight control are monitoring dietary intake and engaging in physical activity and exercise [2]. Despite these strategies being frequently used, the average amount of weight people lose is modest, and the weight loss is difficult to sustain over time [3–5]. A meta-analysis estimated weight loss with diet interventions or diet and exercise interventions to be only 1.6 kg on average [6]. One possible reason for why weight control attempts are often unsuccessful is that people who start exercising might also compensate for that exercise by engaging in other behaviours that make losing weight more difficult. Specifically, people may alter their eating habits in response to exercise by increasing their food intake or eating less healthily after having exercised [7–9]. This compensatory eating behaviour might underlie some of the variability observed in the amount of weight loss in response to exercise [3, 5], and undermine fitness, health and weight goals. Many studies have investigated compensatory eating after exercise in laboratory settings. A robust meta-analysis of laboratory studies showed that exercise led to a slight but nonsignificant increase in subsequent energy intake compared to intake after a control session not involving exercise [10]. In contrast to these studies, other studies have demonstrated effects of compensatory eating after exercise by influencing participants’ perceptions of exercise [11–14]. For example, one study had participants exercise on a stationary bicycle until they had burned 120 kcal, but participants were falsely informed that they had either burned 50 kcal or 265 kcal [13]. Participants who were told that they had burned 265 kcal consumed more food in a subsequent taste test compared to those who were told that they had burned 50 kcal. Another study had participants complete one of three 5 min tasks: exercising (by doing step-ups), imagining exercising (imagining walking up stairs), or imagining attending a classical music concert (no exercise; [12]). Intake for participants in the actual exercise condition and the no-exercise control group was not significantly different, but those that imagined exercising consumed fewer calories than the other two groups. Other studies using within-subjects designs have also predominately found no group-level differences in consumption after exercise compared to consumption after rest [15, 16]. For example, in one study, participant completed two laboratory sessions, one-week apart [16]. The two sessions involved cycling on a stationary bicycle for 50 min (exercise condition) or reading quietly for 50 min (no-exercise condition) and then eating lunch in the laboratory. There was no difference in overall intake between conditions, but there was considerable variability in participants’ eating behaviour after they had exercised compared to when they had not exercise. Specifically, about $50\%$ of participants were classified as “compensators” because they consumed more food after exercise compared to no exercise, whereas the rest of the participants were classified as “non-compensators” because they either did not alter their food intake or they ate less after exercise compared to after no exercise. Other studies have also found variability in eating behaviour after exercise compared to rest, ranging from up to approximately 250 kcal more after exercise than rest for some studies, and up to approximately 200 kcal less after exercise than rest for other studies [17, 18]. An additional complexity is that people’s patterns of intake after exercise may not be consistent across exercise sessions [19, 20]. For example, Unick et al. [ 2015] measured participants’ food intake after exercise and after rest in a within-subjects design, but participants completed three pairs of testing sessions, each separated by one week [20]. Importantly, they identified that the dichotomous classification of people as compensators and non-compensators was not stable over time, with only $21\%$ of their sample consistently classified across all three pairs of testing sessions. Given that most studies of compensatory eating have examined eating behaviour after one or a few exercise sessions at most, the frequency of compensatory eating over time remains unknown. One promising cluster of methodologies that can be used to assess compensatory behaviour over an extended time periods is “daily life methods”, which assess the occurrence of particular behaviours, events, or experiences in naturalistic settings [21]. Sampling can vary in frequency from once per day (daily diary studies; [22]) to several times a day (ecological momentary assessment, experience sampling; [23, 24]), and in duration from a few days to weeks or months, depending on the estimated frequency of the behaviour of interest [25]. Daily life methods are a more ecologically-valid method because they allow participants to report on experiences in their typical environment, rather than in a laboratory scenario. In most compensatory eating laboratory experiments, participants are served a limited range of foods, which is in stark contrast to the modern food environment in which food is often varied, saliently advertised, and ubiquitously available. Similarly, in everyday life, people can choose from a variety of exercise activities rather than being prescribed a specific mode of exercise. Laboratory studies of compensatory eating therefore might be problematic because having choice around parameters of the exercise leads to lower subsequent food intake in comparison to no choice [26]. To date, only one study has explored compensatory eating in free-living conditions [27]. In that study, participants took part in an 8-week walking intervention, with food intake measured via 24-hour dietary recall on one exercise day and one non-exercise day. Food intake was greater in the three hours after exercise, indicative of compensatory eating, although there was no difference in total energy intake on the exercise day compared to the non-exercise day. These results highlight the need for further work to examine compensatory eating in naturalistic settings. Daily life methods can also assess the frequency of particular events because of their ability to track people over extended time periods [25]. This kind of repeated measurement approach can provide insight into individual variability in patterns of post-exercise eating behaviour, which remains a blind spot in the current literature due to a focus on consumption after a single exercise session. Further, daily life methods lend themselves to a comprehensive exploration of the parameters around post-exercise eating behaviour. For example, contextual factors such as social cues [28] and other external cues such as eating location, food availability, and portion size [29] have been shown to influence eating behaviour. It is possible that people may be more likely to engage in post-exercise compensatory eating under certain contextual conditions, and an extended daily diary approach would allow for assessment of the contextual effects. Two recent daily life methods studies provide some information about the relationship between exercise and eating behaviour [30, 31], even though neither was specifically designed to examine compensatory eating after exercise. Dohle and Hofmann [2019] asked participants to complete five short surveys per day for one week about their recent health behaviours, with the aim of examining the relationships among healthy and unhealthy behaviours [30]. One notable finding was that, when participants indicated that unhealthy eating was linked to a previous healthy behaviour, that previous healthy behaviour was most often exercise or healthy eating. However, a limitation of this study was that the contingencies between behaviours that were captured were based on participants’ subjective self-reports of the relationships between their current behaviour and a previous behaviour. Therefore, any additional patterns of behaviour that participants themselves were not aware of might not have been detected. Another study by Grenard et al. [ 2013] found that having exercised for at least 60 min on a particular day was associated with increased likelihood of consuming sweetened beverages but was unrelated to sweet and salty snack consumption [31]. Note, however, that those researchers did not specifically examine food consumption immediately after exercise. Overall, these two studies suggest that daily life methods could be a useful approach to exploring post-exercise compensatory eating. The primary aim of the current research was to use a daily diary approach to test whether people engage in compensatory eating after exercise in everyday life. In Study 1, we compared the healthiness of meals on exercise days to non-exercise days, and the healthiness of post-exercise meals to random meals drawn from non-exercise days. Study 2 aimed to replicate the findings of Study 1, as well as to examine the size of meals eaten on exercise days compared to non-exercise days, and the size of post-exercise meals compared to random meals. A secondary aim of both studies was to explore whether any contextual factors (e.g., food availability, feeling hungry) or characteristics of exercise (e.g., type, duration and intensity) influenced the healthiness (Study 1 and 2) and portion size (Study 2) of post-exercise meals. ## Study 1 Participants reported their food intake and exercise behaviour at the end of each day for 28 days. Given that it was important to capture multiple instances of both exercise and compensatory eating, a reasonably long duration of 28 days was used. Once-per-day sampling was chosen over sampling multiple times per day to minimise participant burden and ensure good compliance over this longer duration [22]. Previous laboratory research has produced mixed findings for compensatory eating, and there is evidence of variability in post-exercise eating behaviour (e.g., [16, 17]) suggesting that people may only compensate some of the time. Thus, although we hypothesised that, across days, participants would eat more unhealthily on exercise days compared to non-exercise days on average, we also expected some within- and between-person variability. Regarding the secondary aim, given the lack of prior research in the area, we did not have any specific hypotheses about the influence of contextual factors. ## Participants Participants were Australian community members ($$n = 61$$) who were recruited via online advertisements. Sample sizes of at least 30 participants are recommended for daily diary studies with survey days nested within participants [32, 33]. To be eligible to participate, participants had to be over 18 years old and have a mobile phone with internet access. They also had to have exercised between 8–20 times in the last 28 days (i.e., approximately 2–5 times per week). Moderate exercisers were selected given that infrequent exercisers (0–1 days per week) were unlikely to provide sufficient exercise sessions across the survey period, and frequent exercisers (6–7 days per week) were unlikely to provide sufficient non-exercise days across the survey period. Participants also had to report that they had eaten less healthily after at least $30\%$ of their reported exercise sessions ($36.7\%$ of potential participants were deemed ineligible for failing to meet this threshold). This requirement was put in place to ensure that sufficient opportunities for compensatory eating were captured across the 28-day study period. Participants were required to complete at least 20 surveys to receive full reimbursement, and those who completed fewer than 18 surveys ($$n = 13$$) were excluded, leaving a final sample of 48 participants (34 women, 14 men). The mean age of the sample was 28.71 years (SD = 10.05; range = 18–59). The mean body mass index (BMI; kg/m2) was 24.34 (SD = 3.95; range = 17.58–39.18). Regarding ethnicity, $44\%$ identified as White, $44\%$ identified as Asian, $2\%$ identified as Aboriginal/Pacific Islander, and $10\%$ identified as “other”. Most participants ($75\%$) wanted to lose weight, and $60\%$ were currently dieting or watching what they ate. Participants ($$n = 67$$) were either Australian community members recruited via online advertisements ($$n = 48$$) or undergraduate students at an Australian university ($$n = 19$$). Eligibility requirements were the same as Study 1 and, as with Study 1, participants who completed fewer than 18 of the 28 end-of-day surveys ($$n = 12$$) were excluded from analyses, leaving a final sample of 55 participants (40 women, 15 men). The mean age of the sample was 23.49 years (SD = 8.38, range = 18–62) and the mean BMI was 23.00 (SD = 1.52; 16.97–36.31). Regarding ethnicity, $67\%$ identified as Asian, $22\%$ identified as White, and $11\%$ identified as “other”. The majority of the sample wanted to lose weight ($69\%$), and $45\%$ were dieting or watching what they ate. From the prescreening eligibility measures, participants reported exercising on average 3.00 times per week (SD = 0.84 times) in the previous four weeks and eating less healthily after $49.64\%$ of their exercise sessions (SD = $14.27\%$). ## Daily surveys The daily surveys were modeled on dietary assessment tools which require participants to systematically report their meals in sequential order [34, 35]. Many of these tools are as accurate as traditional prompted 24-hour recall via phone or interview [36], but they are also time consuming to complete. In order to encourage participant compliance and avoid undue participant burden, we used a briefer measure that allowed us to capture multiple instances of compensatory eating with less precision (rather than a few instances with high precision). At the end of each day, participants reported their food intake for the day by entering their meals one-by-one in the order that they ate them. For each eating occasion, they specified the type of meal (main meal or snack), a brief description of the food(s) eaten, and time of the eating occasions (to the nearest 30 min). Descriptions of food(s) eaten were coded for healthiness by the research team (details below). Next, participants were asked whether or not they had exercised on that day. Participants were informed that exercise in the context of the study referred to structured and purposeful physical activity (rather than incidental physical activity). If they had exercised, they were then asked to specify the exercise type (aerobic/cardio, strength, balance/flexibility, sport, or combination of these options), intensity (vigorous, moderate, or low), start time (to the nearest 30 min) and duration (in 30-min increments). Participants were then given the option to add additional exercise sessions if needed. Next, participants were asked follow-up questions about one target eating occasion from that day. If they had exercised, these questions were about the first eating occasion after the first exercise session they had reported. If participants had not eaten after exercising, or had not exercised that day, they were asked about a randomly selected eating occasion. For the target eating occasion, participants were asked how healthy the eating occasion was (more healthy than normal, the same as normal, less healthy than normal). They then answered 11 yes/no “context” questions about the circumstances surrounding the meal. Specifically, participants indicated whether or not they were: feeling hungry, feeling stressed, in a bad mood, feeling tired, eating alone, in a rush, eating at home, planning to eat the food, or experiencing cravings for that food; and whether or not they had other food options, and whether or not the food was readily available. Finally, participants were shown a list of all their eating occasions from that day. They were asked to rate each meal they had consumed in terms of whether it was less healthy than normal, the same as normal, or more healthy than normal. ## Procedure After providing written informed consent, participants completed demographic information in an initial questionnaire, including age, gender, height and weight (used to calculate BMI), and ethnicity. They were also asked whether or not they were dieting, and whether they wanted to lose weight, stay the same weight, or gain weight. Participants also read instructions about the end-of-day surveys and provided their usual bedtimes. After completing this questionnaire, participants were contacted by the researcher to confirm their end-of-day survey start date and mobile phone number and to provide an opportunity for participants to ask any questions. Each night (1 hr before their nominated bedtime), participants were sent SMS messages containing the end-of-day survey link, which they completed on their mobile phones ($95\%$ of surveys were completed on time, i.e., the same evening). At the end of the 28-day period, participants were sent debriefing and recompense information. Participants were compensated a maximum of AUD $70 if they completed the initial questionnaire and at least 5 surveys in all 4 weeks of the study. The study protocol was approved by UNSW Sydney’s Human Research Ethics Advisory Panel (HC3107). The study protocol was approved by the university’s Human Research Ethics Advisory Panel and the procedure was identical to Study 1 with the following exceptions: ## Coding of meal healthiness The research team coded the descriptions that participants provided about the foods they had eaten in terms of how healthy each eating occasion was (note that the terms “eating occasion” and “meal” are used interchangeably in this paper). Meal descriptions were coded into four categories (healthy, unhealthy, mixed, and “exclude”) using the most recent Australian Guide to Healthy Eating [37]. Healthy meals were those that consisted of foods from the five recommended food groups: grains, lean meats, reduced-fat diary, fruit, and vegetables (e.g., “wholegrain toast with poached eggs and avocado”). Unhealthy meals consisted of foods from the discretionary foods section of the guide, which included alcohol, high sugar or high fat products, and fast food (e.g., “chocolate choc chip muffin”). Mixed meals were those that were neither clearly healthy nor clearly unhealthy (e.g., “sandwich”), and also included meals that consisted of both healthy and unhealthy food items (e.g., “stir fry with vegetables and brown rice, ice cream”). Finally, meals were excluded from analyses if they consisted only of items with zero or negligible calories (e.g., “water” or “herbal tea”) or if the food could not be identified due to typographic error (e.g., “lentens”; $$n = 14$$; post-exercise meals = 1; random meals = 13). Two coders independently coded a random subset of $20\%$ of the total surveys (1,030 meals). Cohen’s kappa for this initial subset was.52, indicating weak inter-rater agreement [38]. The coders discussed discrepancies between meal classifications and resolved them with the input of a third researcher. Some systematic differences in coding were identified (e.g., zero calorie beverages, unfamiliar foods) that appeared to account for a substantial proportion of the disagreement. After resolving discrepancies and refining the coding scheme, the two coders coded a second subset of $10\%$ of the total surveys (511 meals). Cohen’s kappa for the second subset was.81, indicating strong inter-rater agreement. Discrepancies were again resolved with a third researcher. Given the substantial agreement in the second subset, Coder 1 then coded the remaining surveys using the refined coding scheme. The meal coding categorisation was the same as described for Study 1. Two coders independently coded all 1,397 surveys (5,585 meals). Cohen’s kappa was.74, indicating moderate agreement [38]. Discrepancies were resolved with a third coder who was blind to the coding completed by the first two coders. A total of 13 meals were excluded because they could not be coded (post-exercise meals = 2; random meals = 11). As in Study 1, meal healthiness as coded from participants’ meal descriptions was significantly correlated with self-reports of meal healthiness, $r = .69$, $p \leq .001.$ ## Statistical analysis Preliminary descriptive analyses pertaining to the frequency of eating and exercise across the 28 days were computed using SPSS 25. Due to the multilevel structure of the data (daily surveys nested within participants), the primary analyses were carried out using the multilevel modeling software package HLM7 [39]. In these analyses, variables from the end-of-day surveys (exercise vs. non-exercise days, post-exercise vs. random meals, contextual factors) are Level-1 variables, whereas variables pertaining to participant characteristics (individual differences) are Level-2 variables. Models with categorical outcomes were analysed using hierarchical generalised linear modelling (HGLM), and models with continuous outcomes were analysed using hierarchical linear modelling (HLM). The primary analyses examined the healthiness of eating occasions based on codes (unhealthy, mixed, or healthy) derived from participants’ descriptions of the foods they ate. Meal healthiness as coded from participants’ meal descriptions was significantly correlated with their self-reports of meal healthiness, $r = .49$, $p \leq .001.$ The coded measure of healthiness was used for the analysis as a relatively more objective assessment of the foods consumed given that participants’ subjective ratings of relative meal healthiness are more likely to be inaccurate or influenced by demand characteristics or reporting biases. Compensatory eating was examined in two different ways. First, compensatory eating was explored at the meal level by comparing the healthiness of the subset of post-exercise meals ($$n = 437$$ meals) to a subset of randomly-selected meals drawn from non-exercise days ($$n = 593$$ meals). Second, compensatory eating was assessed at the day level in the full dataset by testing whether exercise (yes/no) was a predictor of the percentage of meals on a particular day that were unhealthy. As an additional analysis, we also tested whether exercise predicted the number of main meals consumed per day and the number of snacks consumed per day. Regarding the secondary aim, the post-exercise meal subset was examined to determine whether the healthiness of post-exercise meals was predicted by any contextual factors or characteristics of the exercise. The same analyses were also carried out in the non-exercise day random meal subset to determine whether the same contextual factors predicted healthiness of meals on non-exercise days as for post-exercise meals. The main analyses examined whether exercise was a predictor of the healthiness of the meals consumed and, separately, whether exercise was a predictor of the portion size of the meals consumed. Healthiness was assessed at both the meal level and day level, as in Study 1. Regarding amount of food consumed, compensatory eating was again explored at the meal level such that the subset of post-exercise meals ($$n = 442$$ meals) was compared to randomly-selected meals from non-exercise days ($$n = 812$$ meals) to determine whether exercise predicted the portion size of the meal (self-reported by participants, 1 = small, 5 = big). Compensatory eating was also examined at the day level by testing whether exercise (yes/no) was a predictor of the proportion of meals within a day that were rated as “somewhat big” or “big”. As in Study 1, we tested whether exercise (yes/no) predicted the number of main meals and snacks consumed (i.e., the frequency of eating). The next set of analyses focused on the post-exercise meal subset to determine whether the healthiness (and, separately, portion size) was predicted by any contextual factors or characteristics of the exercise, in line with Study 1. ## Descriptive findings The mean number of surveys completed per participant was 25.25 surveys (SD = 2.44). There were a total of 1,214 recorded end-of-day surveys capturing 5,115 total eating occasions. On average, participants ate 2.62 main meals (i.e., breakfast, lunch, or dinner; SD = 0.37) and 1.55 snacks (SD = 1.13) per day. Exercise sessions. There were 608 total days on which participants reported exercising, and participants exercised an average of 12.67 days during the 28-day survey period (SD = 4.80 days). Participants predominately reported engaging in only one exercise session per day ($$n = 514$$), although there were some days on which participants completed two sessions ($$n = 85$$) or more than two sessions ($$n = 8$$). The analyses focused on the first exercise occasion, of which 346 sessions were described as aerobic/cardio, 87 were strength, 53 were balance/flexibility, 27 were sport, and 94 were a combination. The modal duration of these sessions was 30–60 min. Regarding intensity of exercise, $22.73\%$ of sessions were vigorous, $47.12\%$ were moderate, and $30.15\%$ were low intensity. Meals. There were 438 eating occasions that followed an exercise session (there were no post-exercise meals on 170 days). Of the post-exercise meals that could be coded for healthiness ($$n = 437$$), $25.86\%$ were unhealthy, $40.73\%$ were mixed, and $33.41\%$ were healthy. One random meal was captured from each of the non-exercise days ($$n = 606$$) to be compared to the post-exercise meal subset. Of those meals that could be coded for healthiness ($$n = 593$$), $37.61\%$ were unhealthy, $36.76\%$ were mixed, and $25.63\%$ were healthy. Individual variability. There was considerable individual variability in patterns of eating behaviour. When comparing the proportion of post-exercise meals that were coded as unhealthy to the proportion of random meals on non-exercise days that were coded as unhealthy for each participant (averaging across days), 11 participants showed a pattern of compensatory eating behaviour (greater proportion of unhealthy meals after exercise compared to random meals on non-exercise days), 24 showed the opposite pattern (smaller proportion of unhealthy meals after exercise compared to random meals), and 11 showed eating behaviour that was indifferent to exercise (similar proportions—no more than $10\%$ difference—in unhealthy post-exercise meals and unhealthy non-exercise day random meals; see Fig 1). Fig 1: Individual variability in patterns of eating (Study 1).The y axis shows a difference score for each participant: the proportion of post-exercise meals that were unhealthy minus the proportion of random meals that were unhealthy. A positive score indicates that, on average, the participant consumed a greater proportion of unhealthy meals post exercise than at random meals (i.e., compensatory eating). Dotted lines indicate the boundaries (± 0.10) used to mark a pattern of eating that was indifferent to exercise (similar proportion of unhealthy meals after exercise and at random meals). Participant scores are ordered in descending order from most compensatory eating to least compensatory eating. Difference scores could only be calculated for 46 of the 48 participants because two did not have any post-exercise meals. There were a total of 1,397 surveys recorded. Participants completed 25.40 surveys (SD = 2.73) on average across the 28-day survey period, and $93\%$ of surveys were completed on time (i.e., the same evening). The total number of eating occasions recorded was 5,585, with participants consuming 2.59 main meals (SD = 1.29) and 1.40 snacks (SD = 1.29) per day on average. Exercise sessions. There were 573 total days on which participants reported exercising, and participants exercised an average of 10.42 days (SD = 5.50 days) during the 28-day survey period. Participants predominately reported engaging in only one exercise session per day ($$n = 469$$), although there were some days on which participants completed two sessions ($$n = 94$$) or more than two sessions ($$n = 10$$). The analyses focused on the first exercise occasion of the day, of which $50.96\%$ of sessions were described as aerobic/cardio, $19.02\%$ were strength, $13.96\%$ were balance or flexibility, $6.81\%$ were sport, and $9.25\%$ were a combination. The modal duration of these exercise sessions was 30–60 min. Regarding exercise intensity, $19.37\%$ of sessions were vigorous, $42.41\%$ were moderate, and $38.22\%$ were low intensity. Meals. There were 444 eating occasions that fell after an exercise session (there were no post-exercise meals on 129 days). Of the post-exercise meals that could be coded for healthiness ($$n = 442$$), $26.24\%$ were unhealthy, $45.02\%$ were mixed, and $28.73\%$ were healthy. Regarding self-reported portion size, $9.00\%$ of post-exercise meals were small, $12.84\%$ were somewhat small, $48.42\%$ were moderate, $20.95\%$ were somewhat big, and $8.78\%$ were big. One random meal was captured from each of the non-exercise days ($$n = 823$$) to compare to the post-exercise meal subset. Of those meals that could be coded for healthiness ($$n = 812$$), $34.98\%$ were unhealthy, $34.73\%$ were mixed, and $30.29\%$ were healthy. Regarding self-reported portion size, $12.41\%$ of random meals were small, $18.25\%$ were somewhat small, $46.23\%$ were moderate, $15.69\%$ were somewhat big, and $7.42\%$ were big (see S1 Fig in S1 File). Individual variability. There was considerable individual variability in the pattern of eating behaviours in terms of both the healthiness of meals consumed and the amount of food consumed. For each participant, the proportion of post-exercise meals coded as unhealthy was compared to the proportion of non-exercise day random meals coded as unhealthy. Eleven participants showed a pattern of compensatory eating behaviour (eating proportionally more unhealthy post-exercise meals than random meals) whereas 24 showed the opposite pattern (eating proportionally fewer unhealthy post-exercise meals than random meals), and 14 showed an indifferent eating pattern (similar proportions—no more than $10\%$ difference—in unhealthy post-exercise meals and unhealthy non-exercise day random meals; see Fig 2). Fig 2: Individual variability in patterns of eating—Healthiness (Study 2).The y axis shows a difference score for each participant: the proportion of post-exercise meals that that were unhealthy minus the proportion of random meals that were unhealthy. A positive score indicates that, on average, the participant consumed a greater proportion of unhealthy meals post exercise than at random meals (i.e., compensatory eating). Dotted lines indicate the boundaries (± 0.10) used to mark an eating pattern that was indifferent to exercise (similar proportion of unhealthy meals after exercise and at random meals). Participant scores are ordered in descending order from most compensatory eating to least compensatory eating. Difference scores could only be calculated for 49 of the 55 participants because six did not have any post-exercise meals. The proportion of post-exercise meals classified as big or somewhat big was also compared to the proportion of non-exercise day random meals classified as big or somewhat big for each participant. Sixteen participants showed a pattern of compensatory eating (eating proportionally more meals classified as big after exercise than at random meals), whereas 15 showed the opposite pattern (eating proportionally fewer meals classified as big after exercise than at random meals), and 18 showed an indifferent eating pattern (i.e., within a $10\%$ difference in the proportion of big meals post exercise than at random meals; see Fig 3). Fig 3: Individual variability in patterns of eating—Amount (Study 2).The y axis shows a difference score for each participant: the proportion of post-exercise meals that were big or somewhat big minus the proportion of random meals that were big or somewhat big. A positive score indicates that, on average, the participant consumed a greater proportion of big meals post exercise than at random meals (i.e., compensatory eating). Dotted lines indicate the boundaries (± 0.10) used to mark an eating pattern that was indifferent to exercise (similar proportion of big meals after exercise and at random meals). Participant scores are ordered in descending order from most compensatory eating to least compensatory eating. There was no significant correlation between difference scores for the healthiness of meals and difference scores for the portion size of meals ($r = .22$, $$p \leq .135$$). This lack of correlation suggests that the extent to which an individual showed compensatory eating in terms of eating unhealthily after exercise (compared to random meals) was unrelated to the extent to which they consumed larger portions post exercise (compared to at random meals). ## Exercise and healthiness of meals The primary outcome variable of interest was the healthiness of eating occasions (unhealthy, healthy, and mixed). Mixed was used as the reference category, such that results describe the relative likelihood that eating occasions were unhealthy compared to mixed and the relative likelihood that eating occasions were healthy compared to mixed. Relative likelihoods are described in the following results as odds ratios. To simplify the presentation, comparisons between unhealthy and mixed will be referred to as “eating unhealthily” and comparisons between healthy and mixed will be referred to as “eating healthily.” First, to determine whether there was any evidence of compensatory eating at the meal level, the healthiness of meals was compared for post-exercise eating occasions and random eating occasions drawn from non-exercise days. Exercise (1 = post-exercise eating occasion, 0 = random eating occasion on non-exercise day) was a significant predictor of eating unhealthily such that, contrary to expectations, participants were relatively less likely to eat unhealthily at post-exercise eating occasions (predicted odds = 0.60) compared to random eating occasions on non-exercise days (predicted odds = 0.96). Exercise was not a significant predictor of eating healthily (see Table 1). Table 1 \| Unnamed: 0 \| Unnamed: 1 \| Odds ratio \| Lower limit \| Upper limit \| b \| SE \| t \| p \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Unhealthy vs. mixed meals \| Unhealthy vs. mixed meals \| \| \| \| \| \| \| \| \| \| Intercepta \| 0.96 \| 0.74 \| 1.25 \| -0.04 \| 0.13 \| -0.31 \| 0.757 \| \| \| Post-exercise eating occasion \| 0.63 b \| 0.44 \| 0.9 \| -0.46 \| 0.18 \| -2.54 \| 0.011 \| \| Healthy vs. mixed meals \| Healthy vs. mixed meals \| \| \| \| \| \| \| \| \| \| Intercepta \| 0.63 \| 0.46 \| 0.85 \| -0.47 \| 0.15 \| -3.11 \| 0.003 \| \| \| Post-exercise eating occasion \| 1.10 \| 0.78 \| 1.55 \| 0.09 \| 0.18 \| 0.53 \| 0.597 \| Second, compensatory eating was examined at the day level by assessing whether exercise was a predictor of the percentage of eating occasions within a day that were unhealthy. Exercise (1 = yes, 0 = no) was a significant predictor of the percentage of unhealthy eating occasions, such that participants ate proportionally fewer unhealthy meals on exercise days (predicted value = $30.68\%$) than on non-exercise days (predicted value = $34.95\%$; see Table 2). Table 2 \| Unnamed: 0 \| b \| SE \| t \| p \| \| --- \| --- \| --- \| --- \| --- \| \| Intercepta \| 34.95 \| 2.56 \| 13.66 \| < .001 \| \| Exercise \| -4.27 b \| 1.43 \| -2.99 \| .004 \| At the meal level, exercise was a significant predictor of eating unhealthily such that participants were relatively less likely to eat unhealthily at post-exercise eating occasions (predicted odds = 0.54) compared to random eating occasions on non-exercise days (predicted odds = 0.99). Exercise was also a significant predictor of eating healthily such that participants were relatively less likely to eat healthily at post-exercise eating occasions (predicted odds = 0.60) than at random eating occasions on non-exercise days (predicted odds = 0.85; see Table 4). Table 4 \| Unnamed: 0 \| Unnamed: 1 \| Odds ratio \| Lower limit \| Upper limit \| b \| SE \| t \| p \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Unhealthy vs. mixed meals \| Unhealthy vs. mixed meals \| \| \| \| \| \| \| \| \| \| Intercepta \| 0.99 \| 0.77 \| 1.26 \| -0.01 \| 0.12 \| -0.12 \| .904 \| \| \| Post-exercise eating occasion \| 0.55 \| 0.39 \| 0.77 \| -0.6 \| 0.17 \| -3.51 \| < .001 \| \| Healthy vs. mixed meals \| Healthy vs. mixed meals \| \| \| \| \| \| \| \| \| \| Intercepta \| 0.85 \| 0.68 \| 1.07 \| -0.16 \| 0.11 \| -1.4 \| .168 \| \| \| Post-exercise eating occasion \| 0.71 \| 0.54 \| 0.92 \| -0.35 \| 0.13 \| -2.59 \| .010 \| At the day level, exercise was a significant predictor of the percentage of unhealthy eating occasions, such that participants ate proportionally fewer unhealthy meals on exercise days (predicted value = $30.52\%$) than on non-exercise days (predicted value = $33.90\%$; see Table 5). Table 5 \| Unnamed: 0 \| b \| SE \| t \| p \| \| --- \| --- \| --- \| --- \| --- \| \| Intercepta \| 33.9 \| 2.05 \| 16.54 \| < .001 \| \| Exercise \| -3.39 \| 1.69 \| -2.0 \| .046 \| ## Exercise and frequency of meals Exercise was a significant predictor of the number of main meals consumed such that participants ate relatively more main meals on exercise days (predicted event rate = 2.72 main meals) compared to non-exercise days (predicted event rate = 2.55 main meals). However, exercise was not a significant predictor of the number of snacks (see Table 3). Table 3 \| Unnamed: 0 \| Unnamed: 1 \| Event rate ratio \| Lower limit \| Upper limit \| b \| SE \| t \| p \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Main meals \| Main meals \| \| \| \| \| \| \| \| \| \| Intercepta \| 2.55 \| 2.46 \| 2.64 \| 0.94 \| 0.02 \| 51.94 \| < .001 \| \| \| Exercise \| 1.07 b \| 1.01 \| 1.12 \| 0.06 \| 0.03 \| 3.09 \| .002 \| \| Snacks \| Snacks \| \| \| \| \| \| \| \| \| \| Intercepta \| 1.54 \| 1.47 \| 1.61 \| 0.43 \| 0.12 \| 3.65 \| < .001 \| \| \| Exercise \| 1.06 \| 1.0 \| 1.13 \| 0.06 \| 0.07 \| 0.84 \| .404 \| ## Predictors of post-exercise meal healthiness To explore post-exercise eating behaviour further, contextual factors and characteristics of exercise were examined as predictors of meal healthiness in the post-exercise meal subset ($$n = 437$$). Contextual predictors. The 11 contextual factors referring to parameters of the meal were tested as predictors of post-exercise eating occasion healthiness for the subset of post-exercise eating occasions. Predictors were first examined individually, and then any significant predictors were entered simultaneously into an overall model. Given the exploratory nature of these analyses, no adjustment was made for multiple comparisons, and therefore these results should be interpreted with caution. There were five significant individual predictors: Participants were relatively more likely to eat unhealthily when they were not hungry (predicted odds = 1.04) compared to when they were hungry (predicted odds = 0.47). Participants were also relatively more likely to eat unhealthily when they were in a bad mood (predicted odds = 2.34) than when they were not in a bad mood (predicted odds = 0.52). Unplanned meals were relatively more likely to be unhealthy (predicted odds = 1.10) than were planned meals (predicted odds = 0.29), and participants were also more likely to choose unhealthy options at post-exercise meals when they had to go out of their way to obtain the food (predicted odds = 0.98) compared to when it was readily available (predicted odds = 0.49). Finally, participants were more likely to eat unhealthily after exercise when they were not at home (predicted odds = 0.87) compared to when they were at home (predicted odds = 0.39). There were only two significant predictors of eating healthily (relative to mixed) after exercise. Participants were more likely to eat healthily when there were no other food options (predicted odds = 0.74) compared to when other food options were available (predicted odds = 0.38). Post-exercise meals were also more likely to be healthy when participants felt they were not in a rush (predicted odds = 0.75) compared to when they were in a rush (predicted odds = 0.32; see S1 Table in S1 File). All seven of the significant predictors (i.e., both those that predicted eating unhealthily and those that predicted eating healthily) were then entered together into an overall model. Meal planning was the only predictor that remained significant in the overall model for eating unhealthily (b = -1.13, SE = 0.22, $p \leq .001$), and having no other food options was the only predictor that remained significant for eating healthily (b = -0.65, SE = 0.27, $$p \leq .017$$), indicating that these factors predicted a significant proportion of the variance in healthiness over and above the other predictors. Contextual factors and characteristics of exercise were examined as predictors of meal healthiness in the post-exercise meal subset. Contextual predictors. For the post-exercise eating occasion subset, there were two significant individual predictors of eating unhealthily. Participants were more likely to eat unhealthily when the meal was not planned (predicted odds = 0.77) compared to when the meal was planned (predicted odds = 0.39). They were also more likely to eat unhealthily when they had cravings for the food (predicted odds = 0.97) compared to when they did not have cravings (predicted odds = 0.38). None of the predictors were significant for eating healthily (see S4 Table in S1 File). Both individual predictors remained significant when entered together into an overall model, indicating that they each predicted a significant proportion of the variance in healthiness over and above that explained by the other predictor. For meal planning, b = -0.72, SE = 0.31, $$p \leq .020$$, and for cravings, $b = 0.97$, SE = 0.28, $p \leq .001.$ ## Comparison to non-exercise day random meal sample The same contextual-factors analyses were then repeated in the random eating occasion sample from non-exercise days ($$n = 593$$) to examine whether the predictors that were identified for post-exercise meal healthiness were specific to meals following exercise, or whether the same predictors also predicted healthiness of meals that did not follow exercise. Regarding eating unhealthily, none of the contextual factors that predicted unhealthy eating for post-exercise meals (feeling hungry, being in a bad mood, planning to eat the food, food availability, and eating at home) predicted healthiness of non-exercise random meals (ps >.050). Similarly, none of the contextual factors that predicted meal healthiness in the post-exercise meal sample predicted meal healthiness in the non-exercise random meals (ps >.050). ( See S2 Table in S1 File for full results.) Characteristics of exercise. Characteristics of exercise (exercise intensity, exercise duration, exercise type) were tested as predictors of healthiness of post-exercise meals. However, none of the characteristics of exercise were significant predictors of post-exercise meal healthiness for eating unhealthily or for eating healthily (ps >.050; see S3 Table in S1 File). The contextual factors analyses were repeated for the random eating occasion sample from non-exercise days ($$n = 812$$). As in the post-exercise sample, unplanned meals were more likely to be unhealthy (predicted odds = 1.44) than were planned meals (predicted odds = 0.55), and participants were more likely to eat unhealthily when they had cravings for the food (predicted odds = 1.60) rather than no cravings (predicted odds = 0.73; see S5 Table in S1 File). Characteristics of exercise. Characteristics of exercise (exercise intensity, exercise duration, exercise type) were tested as predictors of meal healthiness of post-exercise meals. Neither exercise duration nor intensity were significant predictors of post-exercise meal healthiness for eating unhealthily. However, for exercise type, one of the dummy-coded predictors was significant for eating unhealthily. Combination exercise was relatively more likely to be followed by an unhealthy meal (predicted odds = 0.89) than was balance/flexibility exercise (predicted odds = 0.15). However, none of the other types of exercise (cardio, strength, or sport) significantly predicted post-exercise meal healthiness (in comparison to combination exercise). Regarding eating healthily, none of the characteristics of exercise were significant predictors of eating healthily compared to mixed (see S6 Table in S1 File). As with the post-exercise meal sample, hunger, eating alone, and food availability were all significant predictors of portion size in the non-exercise random meal sample. Participants ate relatively larger portions when hungry (predicted value = 2.98) compared to not hungry (predicted value = 2.63), when eating with others (predicted value = 2.99) compared to eating alone (predicted value = 2.71), and when they had to go out of their way to obtain the food (predicted value = 3.21) compared to when it was readily available (predicted value = 2.73; see S8 Table in S1 File). Characteristics of exercise. Characteristics of exercise (intensity, type, duration) were tested as predictors of post-exercise portion size. None of the predictors were significant (ps >.050; see S9 Table in S1 File). ## Discussion Over the 28-day study period, there was no evidence that participants compensated for exercise by eating less healthily when averaging across participants. Rather, participants were relatively less likely to eat unhealthily after exercise compared to random meals drawn from non-exercise days. Participants also ate proportionally fewer unhealthy meals on exercise days compared to non-exercise days. Of note, however, there was also individual variability in patterns of eating behaviour over time. Although participants in this study did not, on average, compensate by eating less healthily after exercise, they did eat relatively more main meals on exercise days than on non-exercise days (snack intake was not influenced by exercise). This finding suggests that people might be compensating for their exercise by changing the amount of food that they eat, rather than by making unhealthy food choices after exercise. If the meals that people eat on exercise days are the same size (or larger) than what they usually eat, and they are also eating more meals overall, then this would result in a net increase in total food consumed on exercise days. It is also possible, however, that people eat smaller meals throughout the day on exercise days, but do not consume more food overall. This issue is explored in Study 2. A secondary aim of the current study was to explore whether any contextual factors or characteristics of exercise influenced the healthiness of post-exercise meals. For eating unhealthily, only meal planning remained a significant predictor in the overall model, such that such that unplanned meals were relatively more likely to be unhealthy. For eating healthily, having no other food options available remained significant in the overall model. There was no evidence that the characteristics of exercise (intensity, duration, type) predicted the healthiness of post-exercise meals. Further research is needed to substantiate these exploratory findings. As in Study 1, participants were less likely to eat an unhealthy meal after exercise than they were to eat an unhealthy meal at random meals on non-exercise days, and participants also ate proportionally fewer unhealthy meals on exercise days than on non-exercise days. In addition, participants were also less likely to eat a healthy meal compared to a mixed meal, which might suggest that participants ate less “extremely” after exercise or reflect between-participant variability in eating patterns. Study 2 also examined the amount of food eaten. We found that participants consumed larger meals post-exercise than at random meals, in contrast with evidence from a meta-analysis which found that energy intake did not significantly increase after exercise [10]. However, this meta-analysis included only laboratory studies, unlike our naturalistic method of assessment. We also found that participants consumed a smaller proportion of big meals on exercise days than on non-exercise days. That is, there was compensatory eating with regard to consumption of larger portions after exercise at the meal level, but not at the day level, which might suggest that people may be making up for larger post-exercise meals by eating smaller meals the rest of the day. Note, however, that the day-level effect was borderline significant, so these results must be interpreted with caution. Adjusting for large post-exercise meals by eating less at other meals also seems unlikely because studies examining portion size have consistently found that, when people overeat due to the experimentally-manipulated presence of large portion sizes, they do not tend to modify their portion sizes to be smaller at subsequent meals [4, 40, 42]. Regarding contextual predictors, post-exercise meals and random meals were both more likely to be unhealthy when the meal was not planned, and when cravings were present. Similarly, regarding predictors of meal size, both post-exercise and random meals were likely to be larger when participants reported that they were hungry, eating with others, and when they had to go out of their way to obtain the food. ## Study 2 Study 1 focused on the healthiness of the meals consumed, but it is also important to consider the size of the meals consumed. For example, people might compensate by eating a larger amount of food after exercise, rather than by eating less healthily. In line with this idea, Study 1 also found that participants consumed more main meals (but not more snacks) on exercise days compared to non-exercise days. Consistent consumption of larger portions post exercise could hinder weight-loss attempts by offsetting some of the caloric deficit created through exercise [10] and potentially even lead to weight gain [40]. A pattern of consuming larger portions of food after exercise might be particularly detrimental for those who tend to eat unhealthily after exercise, given that poor diet has numerous associated health risks (e.g., [41]). Previous laboratory studies have shown that participants consumed larger amounts of unhealthy food after exercise when the exercise was perceived as more effortful [13, 14]. However, the foods provided to participants were predominately unhealthy, and it is possible that this unhealthy eating effect might be an artefact of predominately unhealthy food being available. That is, consumption of larger amounts of unhealthy food after exercise in these studies might simply reflect consumption of a greater amount of whatever food was available, and not necessarily a motivated increase in unhealthy eating. Therefore, assessing the portion size of meals consumed in everyday life will provide further insight into whether people are eating larger portions after exercise, and broaden current understanding of patterns of post-exercise eating behaviour in general. The primary aim of Study 2 was to replicate and extend Study 1 by adding a measure of portion size to test whether participants were compensating by eating a larger amount of food after exercise. As in the previous study, participants were asked to report their food intake and exercise behaviour at the end of each day for 28 days. In line with the findings of the previous study, it was hypothesised that participants would, on average, eat less unhealthily at post-exercise meals compared to random meals, and on exercise days compared to non-exercise days. We did not have an expectation about whether exercise would predict the portion size of meals at the post-exercise eating occasion and on exercise days in general. Consistent with Study 1, a secondary aim was to examine whether any contextual factors (e.g., food availability, feeling hungry) or characteristics of exercise (e.g., type, duration, and intensity) influenced the healthiness and size of post-exercise meals. ## Exercise and amount of food consumed Portion size of meals. At the meal level, exercise was a significant predictor of portion size, such that participants reported consuming larger meals post exercise (predicted value = 3.10) than at random meals on non-exercise days (predicted value = 2.86; see Table 6). Table 6 \| Unnamed: 0 \| b \| SE \| t \| p \| \| --- \| --- \| --- \| --- \| --- \| \| Intercepta \| 2.86 \| 0.06 \| 44.74 \| < .001 \| \| Post-exercise eating occasion \| 0.25 \| 0.07 \| 3.46 \| < .001 \| At the day level, exercise was a borderline significant predictor of portion size ($$p \leq .050$$). However, the pattern contrasted with the meal-level findings: A relatively smaller proportion of meals were classified as big or somewhat big on exercise days (predicted value = $23.47\%$) than on non-exercise days (predicted value = $25.97\%$; see Table 7). Table 7 \| Unnamed: 0 \| b \| SE \| t \| p \| \| --- \| --- \| --- \| --- \| --- \| \| Intercepta \| 25.97 \| 2.46 \| 10.53 \| < .001 \| \| Exercise \| -2.49 \| 1.27 \| -1.96 \| .050 \| Frequency of meals. Exercise was a significant predictor of the number of main meals consumed such that participants ate relatively more main meals on exercise days (predicted event rate = 2.66 main meals) than on non-exercise days (predicted event rate = 2.55 main meals). However, exercise was not a significant predictor of the number of snacks consumed (see Table 8). Table 8 \| Unnamed: 0 \| Unnamed: 1 \| Event rate ratio \| Lower limit \| Upper limit \| b \| SE \| t \| p \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Main meals \| Main meals \| \| \| \| \| \| \| \| \| \| Intercepta \| 2.55 \| 2.44 \| 2.68 \| 0.94 \| 0.02 \| 39.61 \| < .001 \| \| \| Exercise \| 1.04 \| 1.01 \| 1.07 \| 0.04 \| 0.02 \| 2.46 \| .014 \| \| Snacks \| Snacks \| \| \| \| \| \| \| \| \| \| Intercepta \| 1.38 \| 1.15 \| 1.64 \| 0.32 \| 0.09 \| 3.63 \| < .001 \| \| \| Exercise \| 1.07 \| 0.99 \| 1.16 \| 0.07 \| 0.04 \| 1.81 \| .071 \| ## Predictors of post-exercise meal portion size Predictors of post-exercise meal portion size were examined using the self-reported measure of portion size (1 = small, 5 = big), which was treated as a continuous outcome variable. Contextual predictors. The 11 contextual factors referring to parameters of the meal were tested as predictors of post-exercise portion size. Hunger was a significant predictor such that participants ate relatively larger portions after exercise when they were hungry (predicted value = 3.18) compared to when they were not hungry (predicted value = 2.66). They also ate larger portions when eating with other people (predicted value = 3.27) compared to when eating alone (predicted value = 2.83) and when they had to go out of their way to obtain the food (predicted value = 3.39) compared to when it was readily available (predicted value = 2.98; see S7 Table in S1 File). When all three of these individually significant predictors were added into an overall model, hunger ($b = 0.48$, SE = 0.11, $p \leq .001$), eating alone (b = -0.37, SE = 0.09, $p \leq .001$), and food availability (b = -0.31, SE = 0.12, $$p \leq .007$$) all remained significant. ## General discussion Previous research on compensatory eating has not examined consistency in post-exercise eating patterns over time, despite preliminary evidence to suggest variability in compensatory eating both between individuals (e.g., [15]), and within the same individual across multiple exercise occasions (e.g., [20]). The present studies fill this gap by investigating compensatory eating behaviour after exercise over 28 days in a naturalistic setting using daily diary methods. Across both studies, there was no indication that participants consistently compensated for exercise by eating less healthily afterwards. Instead, participants ate more healthily on average at post-exercise meals compared to random meals, and ate proportionally more healthy meals on exercise days compared to non-exercise days. These findings are at odds with lab studies that demonstrate evidence of compensatory eating after influencing perceptions of exercise [11, 13, 14]. However, the findings are consistent with other research showing no difference in intake after exercise compared to after rest [15], or no difference in intake after exercise compared to no-exercise [12, 43]. Notably, there was also substantial variability across participants in the healthiness of post-exercise meals. In both studies, approximately $25\%$ of participants tended to compensate by eating more unhealthily on exercise days than they did on non-exercise days, while others showed the opposite pattern, and a third group’s eating behaviour appeared not to be affected by exercise. These results build on previous research which has demonstrated that there is variability in people’s eating behaviour after a single exercise occasion [16, 17], and suggests that there may be meaningful differences in people’s patterns of exercise-associated eating behaviour when examining behaviour over longer time periods. The results of Study 2 also showed that participants consumed relatively bigger portions at post-exercise meals than they did at random meals on non-exercise days over a 28-day period. Although inconsistent with evidence from a meta-analysis of laboratory studies [10], this finding is consistent with the only previous study to investigate post-exercise eating behaviour in a naturalistic setting, which found that people consumed more food in the three hours after exercise than in an equivalent period of time on non-exercise days [27]. When considered in combination with the findings pertaining to the healthiness of post-exercise meals, these results suggest that, on average, people might be consuming larger portions after exercise, but that the type of food consumed post exercise might be more healthy than meals on non-exercise days. Thus, even if people are eating larger portions after exercise, this may not be of great concern because they may be consuming relatively nutritious foods. However, there does appear to be variability in the healthiness of meals consumed post exercise, and consuming larger portions could be more problematic for some people if those portions frequently consist of unhealthy foods. Consumption of unhealthy food also has negative health implications, regardless of weight [41]. Both studies also found that participants ate more main meals on exercise days, but that exercise was not a significant predictor of number of snacks consumed. This finding may reflect variation in how people define meals and snacks, which may depend on the timing or patterning of eating occasions, the nutritional composition of eating occasions, and the context of eating occasions [44, 45]. For example, if healthy foods are relatively more likely to be classified as a main meal than a snack, then the consumption of more main meals on exercise days might reflect a change in classification of meals versus snacks, rather than a true increase in total number of eating occasions. Similarly, if larger portions of the same kind of food are more likely to be classified as main meals than snacks, the finding that more main meals were consumed on exercise days might reflect the consumption of larger portions after exercise. A secondary aim of both studies was to examine whether any contextual factors or characteristics of exercise itself influenced the healthiness (Study 1 and 2) and portion size (Study 2) of post-exercise meals, but there was no consistent pattern to the results across studies. In Study 1, lack of meal planning was a unique predictor of eating unhealthily that was specific to post-exercise meals. Although lack of meal planning and having cravings were significant predictors in Study 2, they were not unique to post-exercise meals (i.e., they predicted the healthiness of random meals as well). For the portion size of post-exercise meals, larger meals were more likely when participants were feeling hungry, eating with others, and when food for that meal was not readily available. Again, these predictors were not specific to post-exercise meals because they also significantly predicted portion size for random meals on non-exercise days. In both studies, none of the characteristics of exercise (intensity, duration, type) predicted the healthiness or the portion size of post-exercise meals, with the exception that one of the dummy-coded predictors for exercise type was a significant predictor of coded meal healthiness in Study 2. ## Limitations and future directions The current work built on previous daily life method studies assessing contingencies between health behaviours [30, 31] by exploring post-exercise eating behaviour over an extended time period. In both our studies, the samples consisted predominately of women, many of whom reported wanting to lose weight and were currently dieting or watching what they ate. We also limited our study to people who were moderately frequent exercisers and self-reported eating unhealthily after exercise at least some of the time so that we could have enough instances of the behaviours of interest to be able to address our research questions. However, by only including participants who exercised at least moderately frequently, we may not have captured the patterns of post-exercise eating of people who exercise only occasionally or who exercise almost every day. By only including participants who self-reported eating unhealthily at least some of the time, our results may actually overestimate the occurrence of post-exercise unhealthy eating. Further research is needed to test the generalisability of these findings in more diverse samples. Daily diaries have the advantage of providing an ecologically valid way of exploring behaviour. However, one limitation associated with these methods is that simply asking participants to record their food intake and exercise behaviour could lead to changes in their behaviour compared to when their behaviour is not monitored [46]. There is currently no viable non-invasive alternative to asking participants to self-monitor food intake, but future studies could potentially use wearable activity trackers to monitor exercise (for a review, see [47]). Another limitation related to the diary method used is that, in order to increase compliance and maintain engagement over the 28-day study period, we opted for a brief report of food intake and exercise, rather than obtaining detailed assessments (with its associated participant burden). Despite strong concordance between the coded healthiness of the meals and participants’ self-reports of those meals, future research could assess dietary intake, portion size, and physical activity using more detailed and precise assessments (such as computer-based food recognition technology, once it becomes more refined and feasible; [48, 49]). More precise measures of food intake and physical activity could provide further detail about the nature of mixed meals and how meal composition may be influenced by exercise or exercise type. These more detailed assessment approaches could also provide information about the interplay between the size and the healthiness of meals consumed post-exercise, including whether exercise influences overall energy intake in addition to the size or healthiness of the meal alone. A limitation associated with our analytical approach was the choice to use random meals on non-exercise days as our comparison, given that participants’ exercise sessions (and subsequent meals after exercising) were likely not occurring at random times. It is possible that the timing of exercise in the day might have influenced the size or healthiness of post-exercise meals. If, for example, for some participants exercise systematically occurred prior to a main meal (which may be larger than a snack) or prior to dessert (which may be less healthy than other meals), then random meals drawn from any time throughout the day are not an optimal comparison. However, given that there was some variation in the timing of exercise both within and between participants, our analytic approach is unlikely to have had a systematic impact on the overall results, but rather to have introduced noise into the data. In order to address this limitation, future research could examine compensatory eating in daily life using artificial intelligence or adaptive modelling approaches to identify “yoked” meals on non-exercise days that most closely matched the timing of post-exercise meals for each participant, rather than comparing post-exercise meals to random meals. These studies also raise important questions for future research around the time course of compensatory eating. It is possible that compensatory eating might occur at other times, rather than at the meal immediately following exercise. That is, people might compensate by eating unhealthily or eating more food at a later meal that day, or even at a meal the next day. Although eating behaviour on exercise days was compared to eating behaviour on non-exercise days, this measure might not have been precise enough to capture any temporal links between exercise and compensatory eating that occurred at a later time. It is also possible that people may engage in compensatory eating when they are expecting to exercise later in the day (a sort of pre-compensation). Indeed, one study has shown that some people increase their food intake when they are anticipating future exercise [50]. Future research should therefore explore the time course of compensatory eating with greater precision. ## Conclusion The current studies used daily diary methods to explore compensatory eating after exercise in everyday life over an extended time period. Using this approach, we found that participants were less likely to eat unhealthily after exercise compared to at random meals on non-exercise days, and a smaller proportion of meals were unhealthy on exercise days than on non-exercise days. Study 2 also found that participants consumed larger portions of food at post-exercise meals compared to random meals on non-exercise days. Considered together, these findings suggest that, on average, people might eat larger portions of healthier food after exercise in their everyday life. There was, however, considerable individual variability in patterns of eating behaviour post exercise. Future research should corroborate these findings in more diverse samples and with more precise dietary assessment methods in order to understand the composition of post-exercise meals, and the factors that influence between- and within-individual differences in exercise-related eating habits. Broadening current understanding of compensatory eating after exercise has the potential to facilitate development of strategies to improve health behaviour regulation. ## References 1. Santos I, Sniehotta FF, Marques MM, Carraça EV, Teixeira PJ. Prevalence of personal weight control attempts in adults: a systematic review and meta-analysis.. Obes Rev. (2017) 18 32-50. DOI: 10.1111/obr.12466 2. Timperio A, Cameron-Smith D, Burns C, Crawford D. The public’s response to the obesity epidemic in Australia: weight concerns and weight control practices of men and women.. Public Health Nutr (2000) 3 417-24. DOI: 10.1017/s1368980000000483 3. Franz MJ, VanWormer JJ, Crain AL, Boucher JL, Histon T, Caplan W. Weight-loss outcomes: A systematic review and meta-analysis of weight-loss clinical trials with a minimum 1-year follow-up. Journal of the American Dietetic Association (2007) 107 1755-67. DOI: 10.1016/j.jada.2007.07.017 4. Jeffery RW, Drewnowski A, Epstein LH, Stunkard AJ, Wilson GT, Wing RR. Long-term maintenance of weight loss: current status.. Health Psychol (2000) 19 5-16. DOI: 10.1037/0278-6133.19.suppl1.5 5. Shaw KA, Gennat HC, O’Rourke P, Del Mar C. Exercise for overweight or obesity.. Cochrane Database of Systematic Reviews. (2006) 6. Wu T, Gao X, Chen M, Van Dam RM. Long-term effectiveness of diet-plus-exercise interventions vs. diet-only interventions for weight loss: a meta-analysis. Obesity Reviews (2009) 10 313-23. DOI: 10.1111/j.1467-789X.2008.00547.x 7. King NA, Caudwell P, Hopkins M, Byrne NM, Colley R, Hills AP. Metabolic and behavioral compensatory responses to exercise interventions: Barriers to weight loss.. Obesity (2007) 15 1373-83. DOI: 10.1038/oby.2007.164 8. King NA, Horner K, Hills AP, Byrne NM, Wood RE, Bryant E. Exercise, appetite and weight management: understanding the compensatory responses in eating behaviour and how they contribute to variability in exercise-induced weight loss. British Journal of Sports Medicine (2012) 46 315. DOI: 10.1136/bjsm.2010.082495 9. Melanson EL, Keadle SK, Donnelly JE, Braun B, King NA. Resistance to exercise-induced weight loss: compensatory behavioral adaptations.. Medicine & Science in Sports & Exercise (2013) 45. DOI: 10.1249/MSS.0b013e31828ba942 10. Schubert MM, Desbrow B, Sabapathy S, Leveritt M. Acute exercise and subsequent energy intake. A meta-analysis. Appetite (2013) 63 92-104. DOI: 10.1016/j.appet.2012.12.010 11. Fenzl N, Bartsch K, Koenigstorfer J. Labeling exercise fat-burning increases post-exercise food consumption in self-imposed exercisers. Appetite (2014) 81 1-7. DOI: 10.1016/j.appet.2014.05.030 12. Inauen J, Radtke T, Rennie L, Scholz U, Orbell S. Transfer or compensation?. Swiss Journal of Psychology (2018) 77 59-67 13. McCaig DC, Hawkins LA, Rogers PJ. Licence to eat: Information on energy expended during exercise affects subsequent energy intake. Appetite (2016) 107 323-9. DOI: 10.1016/j.appet.2016.08.107 14. Werle COC, Wansink B, Payne CR. Is it fun or exercise? The framing of physical activity biases subsequent snacking. Marketing Letters (2015) 26 691-702 15. Donnelly JE, Herrmann SD, Lambourne K, Szabo AN, Honas JJ, Washburn RA. Does increased exercise or physical activity alter ad-libitum daily energy intake or macronutrient composition in healthy adults. ? A systematic review (2014) 9 e83498 16. Finlayson G, Bryant E, Blundell JE, King NA. Acute compensatory eating following exercise is associated with implicit hedonic wanting for food. Physiology & Behavior (2009) 97 62-7. DOI: 10.1016/j.physbeh.2009.02.002 17. Hopkins M, Blundell JE, King NA. Individual variability in compensatory eating following acute exercise in overweight and obese women. British Journal of Sports Medicine (2014) 48 1472. DOI: 10.1136/bjsports-2012-091721 18. Unick JL, Otto AD, Goodpaster BH, Helsel DL, Pellegrini CA, Jakicic JM. Acute effect of walking on energy intake in overweight/obese women. Appetite (2010) 55 413-9. DOI: 10.1016/j.appet.2010.07.012 19. Brown GL, Lean ME, Hankey CR. Reproducibility of 24-h post-exercise changes in energy intake in overweight and obese women using current methodology. British Journal of Nutrition (2012) 108 191-4. DOI: 10.1017/S0007114511005575 20. Unick JL O, ’Leary KC, Dorfman L, Thomas JG, Strohacker K, Wing RR. Consistency in compensatory eating responses following acute exercise in inactive, overweight and obese women. British Journal of Nutrition (2015) 113 1170-7. DOI: 10.1017/S000711451500046X 21. Reis HT, Mehl MR, Conner TS. (2012) 3-21 22. Gunthert KC, Wenze SJ, Mehl MR, Conner TS. Handbook of research methods for studying daily life: The Guilford Press (2012) 144-59 23. Shiffman S, Stone AA, Hufford MR. Ecological momentary assessment.. Annual Review of Clinical Psychology (2008) 4 1-32 24. Stone AA, Shiffman S. Ecological momentary assessment in behavioral medicine.. Annals of Behavioral Medicine (1994) 16 199-202 25. Stone AA, Shiffman S. Capturing momentary, self-report data: A proposal for reporting guidelines. Annals of Behavioral Medicine (2002) 24 236-43. DOI: 10.1207/S15324796ABM2403_09 26. Beer NJ, Dimmock JA, Jackson BEN, Guelfi KJ. Providing choice in exercise influences food intake at the subsequent meal. Medicine & Science in Sports & Exercise (2017) 49. DOI: 10.1249/MSS.0000000000001330 27. Lyon KM. Predictors of acute dietary compensation among sedentary women after free-living moderate intensity exercise (2014) 28. Herman CP, Roth DA, Polivy J. Effects of the presence of others on food intake: a normative interpretation.. Psychol Bull. (2003) 129 873-86. DOI: 10.1037/0033-2909.129.6.873 29. Wansink B.. Environmental factors that increase the food intake and consumption volume of unknowing consumers Annual Review of Nutrition (2004) 24 455-79 30. Dohle S, Hofmann W. Consistency and balancing in everyday health behaviour: An ecological momentary assessment approach.. Applied Psychology: Health and Well-Being (2019) 11 148-69. DOI: 10.1111/aphw.12148 31. Grenard JL, Stacy AW, Shiffman S, Baraldi AN, MacKinnon DP, Lockhart G. Sweetened drink and snacking cues in adolescents. A study using ecological momentary assessment. Appetite (2013) 67 61-73. PMID: 23583312 32. Ohly S, Sonnentag S, Niessen C, Zapf D. Diary studies in organizational research. Journal of Personnel Psychology (2010) 9 79-93 33. Scherbaum CA, Ferreter JM. Estimating statistical power and required sample sizes for organizational research using multilevel modeling. Organizational Research Methods (2008) 12 347-67 34. Eldridge AL, Piernas C, Illner A-K, Gibney MJ, Gurinović MA, De Vries JHM. Evaluation of new technology-based tools for dietary intake assessment—An ILSI Europe dietary intake and exposure task force evaluation.. Nutrients (2019) 11 35. McClung HL, Ptomey LT, Shook RP, Aggarwal A, Gorczyca AM, Sazonov ES. Dietary intake and physical activity assessment: Current tools, techniques, and technologies for use in adult populations. American Journal of Preventive Medicine (2018) 55 e93-e104. DOI: 10.1016/j.amepre.2018.06.011 36. Shim JS, Oh K, Kim HC. Dietary assessment methods in epidemiologic studies.. Epidemiol Health (2014) 36 e2014009. DOI: 10.4178/epih/e2014009 37. 37National Health and Medical Research Council. Australian Dietary Guidelines. National Health and Medical Research Council. 2013.. Australian Dietary Guidelines. National Health and Medical Research Council (2013) 38. McHugh ML. Interrater reliability: the kappa statistic. Biochem Med (Zagreb) (2012) 22 276-82. DOI: 10.1016/j.jocd.2012.03.005 39. Raudenbush SW, Bryk AS, Cheong YF, Congdon RT, du Toit M. Hierarchical linear and nonlinear modeling user manual: User guide for Scientific Software International’s (S.S.I.) program: Scientific Software International;. (2016) 40. French SA, Mitchell NR, Wolfson J, Harnack LJ, Jeffery RW, Gerlach AF. Portion size effects on weight gain in a free living setting.. Obesity (2014) 22 1400-5. DOI: 10.1002/oby.20720 41. Micha R, Peñalvo JL, Cudhea F, Imamura F, Rehm CD, Mozaffarian D. Association between dietary factors and mortality from heart disease, stroke, and Type 2 Diabetes in the United States. JAMA (2017) 317 912-24. DOI: 10.1001/jama.2017.0947 42. Rolls BJ, Roe LS, Meengs JS. The effect of large portion sizes on energy intake is sustained for 11 days.. Obesity (2007) 15 1535-43. PMID: 17557991 43. Coelho J, Roefs A, Havermans R, Salvy S-J, Jansen A. Effects of exercising before versus after eating on dieting and exercise evaluations: A preliminary investigation.. Canadian Journal of Behavioural Science / Revue canadienne des sciences du comportement (2011) 43 63-7 44. Gatenby SJ. Eating frequency: methodological and dietary aspects. British Journal of Nutrition (1997) 77 S7-S20. DOI: 10.1079/bjn19970100 45. Leech RM, Worsley A, Timperio A, McNaughton SA. Understanding meal patterns: definitions, methodology and impact on nutrient intake and diet quality.. Nutrition Research Reviews (2015) 28 1-21. DOI: 10.1017/S0954422414000262 46. Nelson RO, Hayes SC. Theoretical explanations for reactivity in self-monitoring.. Behavior Modification (1981) 5 3-14 47. Shin G, Jarrahi MH, Fei Y, Karami A, Gafinowitz N, Byun A. Wearable activity trackers, accuracy, adoption, acceptance and health impact: A systematic literature review. Journal of Biomedical Informatics (2019) 93 103153. DOI: 10.1016/j.jbi.2019.103153 48. Akpa EAH, Suwa H, Arakawa Y, Yasumoto K. Smartphone-based food weight and calorie estimation method for effective food journaling.. SICE Journal of Control, Measurement, and System Integration. (2017) 10 360-9 49. Liu C, Cao Y, Luo Y, Chen G, Vokkarane V, Ma Y. DeepFood: Deep learning-based food image recognition for computer-aided dietary assessment. Inclusive Smart Cities and Digital Health (2016) 50. Sim AY, Lee LL, Cheon BK. When exercise does not pay: Counterproductive effects of impending exercise on energy intake among restrained eaters. Appetite (2018) 123 120-7. DOI: 10.1016/j.appet.2017.12.017
--- title: 'Maternal characteristics associated with referral to obstetrician-led care in low-risk pregnant women in the Netherlands: A retrospective cohort study' authors: - Susan Niessink-Beckers - Corine J. Verhoeven - Marleen J. Nahuis - Lisanne A. Horvat-Gitsels - Janneke T. Gitsels-van der Wal journal: PLOS ONE year: 2023 pmcid: PMC10016726 doi: 10.1371/journal.pone.0282883 license: CC BY 4.0 --- # Maternal characteristics associated with referral to obstetrician-led care in low-risk pregnant women in the Netherlands: A retrospective cohort study ## Abstract ### Background In the Netherlands, maternity care is divided into midwife-led care (for low-risk women) and obstetrician-led care (for high-risk women). Referrals from midwife-led to obstetrician-led care have increased over the past decade. The majority of women are referred during their pregnancy or labour. Referrals are based on a continuous risk assessment of the health and characteristics of mother and child, yet referral for non-medical factors and characteristics remain unclear. This study investigated which maternal characteristics are associated with women’s referral from midwife-led to obstetrician-led care. ### Materials and methods A retrospective cohort study in one midwife-led care practice in the Netherlands included 1096 low-risk women during January 2015–17. The primary outcomes were referral from midwife-led to obstetrician-led care in [1] the antepartum period and [2] the intrapartum period. In total, 11 maternal characteristics were identified. Logistic regression models of referral in each period were fitted and stratified by parity. ### Results In the antepartum period, referral among nulliparous women was associated with an older maternal age (aOR, 1.07; $95\%$CI, 1.05–1.09), being underweight (0.45; 0.31–0.64), overweight (2.29; 1.91–2.74), or obese (2.65; 2.06–3.42), a preconception period >1 year (1.34; 1.07–1.66), medium education level (0.76; 0.58–1.00), deprivation (1.87; 1.54–2.26), and sexual abuse (1.44; 1.14–1.82). Among multiparous women, a referral was associated with being underweight (0.40; 0.26–0.60), obese (1.61; 1.30–1.98), a preconception period >1 year (1.71; 1.27–2.28), employment (1.38; 1.19–1.61), deprivation (1.23; 1.03–1.46), highest education level (0.63; 0.51–0.80), psychological problems (1.24; 1.06–1.44), and one or multiple consultations with an obstetrician (0.68; 0.58–0.80 and 0.64; 0.54–0.76, respectively). In the intrapartum period, referral among nulliparous women was associated with an older maternal age (1.02; 1.00–1.05), being underweight (1.67; 1.15–2.42), a preconception period >1 year (0.42; 0.31–0.57), medium or high level of education (2.09; 1.49–2.91 or 1.56; 1.10–2.22, respectively), sexual abuse (0.46; 0.33–0.63), and multiple consultations with an obstetrician (1.49; 1.15–1.94). Among multiparous women, referral was associated with an older maternal age (1.02; 1.00–1.04), being overweight (0.65; 0.51–0.83), a preconception period >1 year (0.33; 0.17–0.65), non-Dutch ethnicity (1.98; 1.61–2.45), smoking (0.75; 0.57–0.97), sexual abuse (1.49; 1.09–2.02), and one or multiple consultations with an obstetrician (1.34; 1.06–1.70 and 2.09; 1.63–2.69, respectively). ### Conclusions This exploratory study showed that several non-medical maternal characteristics of low-risk pregnant women are associated with referral from midwife-led to obstetrician-led care, and how these differ by parity and partum period. ## Introduction Multiple countries worldwide provide midwife-led care, e.g., Australia, Canada, New Zealand, the United Kingdom and the Netherlands [1]. Midwife-led care is a model where “the midwife is the lead professional in the planning, organisation and delivery of care given to a woman from initial booking to the postnatal period” [2]. Other models of care are obstetrician-led care, family doctor-led care or shared care. The Dutch maternity care system is divided into two echelons: primary and secondary care. In primary care, known as midwife-led care, midwives provide care for low-risk pregnant women during the antepartum, intrapartum and postpartum periods. Over $87\%$ of all pregnant women in the Netherlands start their prenatal care in primary care [3]. Women will remain in primary care if they are healthy and no complications occur. In cases where pathology occurs in the antepartum, intrapartum, or postpartum period, women are referred to secondary care, also known as obstetrician-led care. In secondary care, the care will be provided by obstetricians or hospital-based midwives [4, 5]. In the Netherlands, the number of referrals in the intrapartum period has increased from $27\%$ to $41\%$ over the past 12 years [3, 6]. The List of Obstetric Indications (LOI) provides guidelines for determining whether a woman should receive midwife-led or obstetrician-led care, mainly based on medical and obstetric history [4, 7]. The main antepartum indications for a referral are gestational diabetes, pregnancy-induced hypertension, and previous caesarean section [3, 8]. The intrapartum indications include a request for medical pain relief such as epidural analgesia, the presence of meconium-stained amniotic fluid, and failure to progress in the first or second stage of labour [9–12]. In particular, the number of referrals for a request for medical pain relief has increased; in 2004, only $4\%$ of women received epidural analgesia during labour compared to $21\%$ in 2017 [3]. Dutch primary care midwives are not qualified to care for women who receive an epidural in the Netherlands. Therefore, they are referred to obstetrician-led care. Numerous studies have determined the influence of risk factors such as body mass index (BMI) or maternal age on perinatal outcomes [13–17]. However, few studies have investigated the association between perinatal outcomes and a wide range of maternal characteristics that are readily available in clinical records, such as a history or presence of psychological problems or sexual abuse [18–25]. Moreover, the associations between specifically non-medical maternal factors and referral towards obstetric care are still unknown. Awareness of all maternal characteristics, including non-medical factors affecting the chance of a referral, will help healthcare professionals provide individualised preventive care [26, 27]. Before we can intervene on those factors, we need to know which non-medical factors increase the likelihood of referral to obstetrician-led care. Therefore, our study investigated which non-medical maternal characteristics are associated with women’s referral from midwife-led to obstetrician-led care in the antepartum and intrapartum periods, as these may influence perinatal outcomes. ## Study design and participants This retrospective cohort study took place in one large midwifery practice in an urban region near Amsterdam, the Netherlands. The study period was from January 2015 to January 2017 [8]. The study sample included women with a singleton pregnancy who received midwife-led care after the first trimester. Those who had a miscarriage were excluded and those who were referred to obstetrician-led care in the first trimester or received only postnatal care. Informed consent was obtained verbally and noted in their medical records in their presence [8]. The Medical Ethics Committee of the Amsterdam University Medical Centres (location VUmc) (FWA00017598) approved the study (ref. 2018.019). ## Measures The two primary outcomes were defined as a referral from midwife-led to obstetrician-led care in the antepartum (yes/no) and, given no previous referral, in the intrapartum period (yes/no). Maternal non-medical characteristics—the independent variables of interest—were based on literature on the effect of these characteristics on morbidity and mortality of mother and child [13, 15–18, 21, 23–25, 28]. These characteristics were obtained from the women’s medical records where they were noted by the midwife at the beginning of the antepartum period. Dichotomous variables included: preconception period (≤1 year/>1 year)—the period the woman was trying to conceive—, employment (yes/no), ethnicity (Dutch/non-Dutch)—based on the mothers country of birth—, lived in a deprived area (no/yes)—based on a zip code classified as a deprived area [29]—, smoking (no/yes), psychological problems (past and previous) (no/yes), and a history of sexual abuse (no/yes). Education level was categorised into low (primary education, pre-vocational secondary education or secondary vocational education), medium (senior general secondary education or pre-university education) and high (higher professional education or university education). BMI and number of consultations with an obstetrician were categorised due to non-normal distributions. Underweight was defined as having a BMI of <18.5, healthy weight was defined as a BMI of 18.5–24.9, overweight was defined as a BMI of 25.0–29.9, and obesity was defined as a BMI of ≥30.0 [30]. The number of consultations with an obstetrician—a standalone consult in obstetric care without women discontinuing midwife-led care—was defined as none, one, and ≥2. Maternal age was a continuous variable based on the date of birth. Compared to multiparous women, nulliparous women are more likely to be referred to obstetrician-led care both during the antepartum and intrapartum periods [3, 9, 31]. Therefore, parity was identified as an effect modifier. The reasons for referral were grouped by the antepartum and intrapartum periods and were based on regional protocols and the LOI.https://www.ncbi.nlm.nih.gov/pubmed/30153431 ## Statistical analyses The study population’s baseline characteristics were summarised by means and standard deviations for normally distributed continuous variables, and frequencies and percentages for categorical variables, including dichotomous ones. Missing data patterns were explored by fitting logistic regression models to understand potential selection bias and dealt with by multiple imputations [32, 33]. Logistic regression models of referral in the antepartum and intrapartum periods were stratified by parity. The leanest models were obtained using backward elimination with a significance level set at α = 0.05. The logistic model assumptions of linearity, independence of errors, and multicollinearity were checked by looking at interactions between predictor and its log transformation, Durbin-Watson tests, and variance inflation factors, respectively. Model performance was assessed by specificity, sensitivity, and total accuracy [34–38]. The sensitivity analysis included repeating the model fittings on the subset with complete information (i.e., complete case analysis). All analyses were performed in SPSS version 24. ## Descriptive analysis The study sample included 1096 participants with 520 ($47\%$) nulliparous and 576 ($53\%$) multiparous women. A total of 448 ($41\%$) participants were referred to obstetrician-led care during the antepartum period, $39\%$ of nulliparous and $42\%$ of multiparous women. Among the 617 women who started labour in midwife-led care, 31 ($5\%$) were moved to another practice or clinic in the country by the end of the antepartum period and 287 ($47\%$) were referred to obstetrician-led care in the intrapartum period, $62\%$ of nulliparous and $33\%$ of multiparous women. The baseline characteristics by partum period and parity are presented in Table 1. Table 1 \| a: Characteristics of the study population concerning a referral in the antepartum period \| a: Characteristics of the study population concerning a referral in the antepartum period.1 \| a: Characteristics of the study population concerning a referral in the antepartum period.2 \| a: Characteristics of the study population concerning a referral in the antepartum period.3 \| a: Characteristics of the study population concerning a referral in the antepartum period.4 \| a: Characteristics of the study population concerning a referral in the antepartum period.5 \| a: Characteristics of the study population concerning a referral in the antepartum period.6 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| \| \| Parity \| Parity \| Parity \| Parity \| \| \| \| Total \| Nulliparous \| Nulliparous \| Multiparous \| Multiparous \| \| \| \| Total \| Not referred \| Referred \| Not referred \| Referred \| \| \| \| n = 1096 (100.0%) \| n = 316 (28.8%) \| n = 204 \| n = 332 \| n = 244 \| \| \| \| n = 1096 (100.0%) \| n = 316 (28.8%) \| (18.6%) \| (30.3%) \| (22.3%) \| \| \| \| mean (SD) \| mean (SD) \| mean (SD) \| mean (SD) \| mean (SD) \| \| Maternal age \| \| 29.2 (5.0) \| 26.7 (4.6) \| 28.3 (4.8) \| 30.9 (4.4) \| 31.1 (4.7) \| \| \| \| n (%) \| n (% of no nulliparous) \| n (% of yes nulliparous) \| n (% of no multiparous) \| n (% of yes multiparous) \| \| BMI* \| BMI* \| BMI* \| BMI* \| BMI* \| BMI* \| BMI* \| \| \| <18.5 \| 61 (5.6) \| 30 (9.6) \| 7 (3.4) \| 19 (5.8) \| 5 (2.1) \| \| \| 18.5–24.9 \| 611 (56.1) \| 205 (65.3) \| 100 (49.0) \| 182 (55.2) \| 124 (51.2) \| \| \| 25–29.9 \| 282 (25.9) \| 57 (18.2) \| 65 (31.9) \| 90 (27.3) \| 70 (28.9) \| \| \| ≥30 \| 136 (12.5) \| 22 (7.0) \| 32 (15.7) \| 39 (11.8) \| 43 (17.8) \| \| Preconception period* \| Preconception period* \| Preconception period* \| Preconception period* \| Preconception period* \| Preconception period* \| Preconception period* \| \| \| ≤ 1 year \| 920 (90.2) \| 264 (88.9) \| 159 (82.0) \| 294 (95.5) \| 203 (91.9) \| \| \| > 1 year \| 100 (9.8) \| 33 (11.1) \| 35 (18.0) \| 14 (4.5) \| 18 (8.1) \| \| Education level* \| Education level* \| Education level* \| Education level* \| Education level* \| Education level* \| Education level* \| \| \| low \| 132 (12.5) \| 31 (10.1) \| 20 (10.1) \| 44 (13.8) \| 37 (15.9) \| \| \| medium \| 501 (47.4) \| 145 (47.1) \| 81 (40.9) \| 153 (47.8) \| 122 (52.6) \| \| \| high \| 425 (40.2) \| 132 (42.9) \| 97 (49.0) \| 123 (38.4) \| 73 (31.5) \| \| Employment* \| Employment* \| Employment* \| Employment* \| Employment* \| Employment* \| Employment* \| \| \| yes \| 734 (67.8) \| 228 (73.1) \| 153 (75.7) \| 200 (60.6) \| 153 (64.0) \| \| \| no \| 349 (32.2) \| 84 (26.9) \| 49 (24.3) \| 130 (39.4) \| 86 (36.0) \| \| Ethnicity* \| Ethnicity* \| Ethnicity* \| Ethnicity* \| Ethnicity* \| Ethnicity* \| Ethnicity* \| \| \| Dutch \| 508 (46.4) \| 162 (51.3) \| 102 (50.7) \| 141 (42.5) \| 101 (42.6) \| \| \| non-Dutch \| 586 (53.6) \| 154 (48.7) \| 99 (49.3) \| 191 (57.5) \| 136 (57.4) \| \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| \| \| yes \| 228 (20.8) \| 49 (15.6) \| 51 (25.0) \| 69 (20.8) \| 59 (24.3) \| \| \| no \| 866 (79.2) \| 266 (84.4) \| 153 (75.0) \| 263 (79.2) \| 184 (75.7) \| \| Smoking* \| Smoking* \| Smoking* \| Smoking* \| Smoking* \| Smoking* \| Smoking* \| \| \| yes \| 253 (23.2) \| 80 (25.4) \| 54 (26.6) \| 64 (19.3) \| 55 (22.6) \| \| \| no \| 839 (76.8) \| 235 (74.6) \| 149 (73.4) \| 267 (80.7) \| 188 (77.4) \| \| Psychological problems* \| Psychological problems* \| Psychological problems* \| Psychological problems* \| Psychological problems* \| Psychological problems* \| Psychological problems* \| \| \| yes \| 309 (28.3) \| 83 (26.5) \| 61 (29.9) \| 90 (27.2) \| 75 (30.9) \| \| \| no \| 782 (71.7) \| 230 (73.5) \| 143 (70.1) \| 241 (72.8) \| 168 (69.1) \| \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| \| \| yes \| 135 (12.4) \| 38 (12.1) \| 29 (14.2) \| 38 (11.5) \| 30 (12.3) \| \| \| no \| 957 (87.6) \| 276 (87.9) \| 175 (85.8) \| 293 (88.5) \| 213 (87.7) \| \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| \| \| none \| 503 (46.1) \| 142 (45.4) \| 101 (49.8) \| 136 (41.0) \| 124 (50.8) \| \| \| 1 \| 330 (30.2) \| 100 (31.9) \| 56 (27.6) \| 108 (32.5) \| 66 (27.0) \| \| \| > 1 \| 259 (23.7) \| 71 (22.7) \| 46 (22.7) \| 88 (26.5) \| 54 (22.1) \| \| b: Characteristics of the study population concerning a referral in the intrapartum period \| b: Characteristics of the study population concerning a referral in the intrapartum period \| b: Characteristics of the study population concerning a referral in the intrapartum period \| b: Characteristics of the study population concerning a referral in the intrapartum period \| b: Characteristics of the study population concerning a referral in the intrapartum period \| b: Characteristics of the study population concerning a referral in the intrapartum period \| b: Characteristics of the study population concerning a referral in the intrapartum period \| \| \| \| \| Parity \| Parity \| Parity \| Parity \| \| \| \| Total \| Nulliparous \| Nulliparous \| Multiparous \| Multiparous \| \| \| \| Total \| Not referred \| Referred \| Not referred \| Referred \| \| \| \| n = 617 (100.0%) \| n = 112 (18.2%) \| n = 179 (29.0%) \| n = 218 (35.3%) \| n = 108 (17.5%) \| \| \| \| mean (SD) \| mean (SD) \| mean (SD) \| mean (SD) \| mean (SD) \| \| Maternal age \| \| 28.9 (4.8) \| 26.6 (4.2) \| 26.8 (4.6) \| 30.8 (4.2) \| 30.8 (4.5) \| \| \| \| n (%) \| n (% of no nulliparous) \| n (% of yes nulliparous) \| n (% of no multiparous) \| n (% of yes multiparous) \| \| BMI* \| BMI* \| BMI* \| BMI* \| BMI* \| BMI* \| BMI* \| \| \| <18.5 \| 45 (7.3) \| 8 (7.2) \| 18 (10.1) \| 13 (6.0) \| 6 (5.7) \| \| \| 18.5–24.9 \| 372 (60.6) \| 75 (67.6) \| 117 (65.4) \| 118 (54.1) \| 62 (58.5) \| \| \| 25–29.9 \| 141 (23.0) \| 21 (18.9) \| 31 (17.3) \| 64 (29.4) \| 25 (23.6) \| \| \| ≥30 \| 56 (9.1) \| 7 (6.3) \| 13 (7.3) \| 23 (10.6) \| 13 (12.3) \| \| Preconception period* \| Preconception period* \| Preconception period* \| Preconception period* \| Preconception period* \| Preconception period* \| Preconception period* \| \| \| ≤ 1 year \| 533 (92.5) \| 89 (83.2) \| 153 (92.2) \| 193 (95.1) \| 98 (98.0) \| \| \| > 1 year \| 43 (7.5) \| 18 (16.8) \| 13 (7.8) \| 10 (4.9) \| 2 (2.0) \| \| Education level* \| Education level* \| Education level* \| Education level* \| Education level* \| Education level* \| Education level* \| \| \| low \| 72 (12.0) \| 17 (15.2) \| 13 (7.5) \| 28 (13.0) \| 14 (14.1) \| \| \| medium \| 284 (47.4) \| 47 (42.0) \| 87 (50.3) \| 96 (44.7) \| 54 (54.5) \| \| \| high \| 243 (40.6) \| 48 (42.9) \| 73 (42.2) \| 91 (42.3) \| 31 (31.3) \| \| Employment* \| Employment* \| Employment* \| Employment* \| Employment* \| Employment* \| Employment* \| \| \| yes \| 410 (67.0) \| 84 (75.0) \| 130 (73.9) \| 135 (62.2) \| 61 (57.0) \| \| \| no \| 202 (33.0) \| 28 (25.0) \| 46 (26.1) \| 82 (37.8) \| 46 (43.0) \| \| Ethnicity* \| Ethnicity* \| Ethnicity* \| Ethnicity* \| Ethnicity* \| Ethnicity* \| Ethnicity* \| \| \| Dutch \| 288 (46.7) \| 60 (55.6) \| 90 (50.3) \| 124 (57.7) \| 34 (31.5) \| \| \| non-Dutch \| 329 (53.3) \| 48 (44.4) \| 89 (49.7) \| 91 (42.3) \| 74 (68.5) \| \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| Lived in deprived area* \| \| \| yes \| 116 (18.8) \| 17 (15.2) \| 31 (17.3) \| 41 (18.8) \| 27 (25.0) \| \| \| no \| 501 (81.2) \| 95 (84.8) \| 148 (82.7) \| 177 (81.2) \| 81 (75.0) \| \| Smoking* \| Smoking* \| Smoking* \| Smoking* \| Smoking* \| Smoking* \| Smoking* \| \| \| yes \| 136 (22.1) \| 27 (24.1) \| 46 (25.7) \| 43 (19.8) \| 20 (18.5) \| \| \| no \| 480 (77.9) \| 85 (75.9) \| 133 (74.3) \| 174 (80.2) \| 88 (81.5) \| \| Psychological problems* \| Psychological problems* \| Psychological problems* \| Psychological problems* \| Psychological problems* \| Psychological problems* \| Psychological problems* \| \| \| yes \| 165 (26.9) \| 31 (27.9) \| 46 (25.8) \| 56 (25.7) \| 32 (29.9) \| \| \| no \| 449 (73.1) \| 80 (72.1) \| 132 (74.2) \| 162 (74.3) \| 75 (70.1) \| \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| History of sexual abuse* \| \| \| yes \| 68 (11.1) \| 17 (15.3) \| 14 (7.8) \| 22 (10.1) \| 15 (14.0) \| \| \| no \| 547 (88.9) \| 94 (84.7) \| 165 (92.2) \| 196 (89.9) \| 92 (86.0) \| \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| Consultation obstetric care* \| \| \| none \| 256 (41.7) \| 50 (45.0) \| 73 (41.2) \| 99 (45.4) \| 34 (31.5) \| \| \| 1 \| 200 (32.6) \| 36 (32.4) \| 58 (32.8) \| 71 (32.6) \| 35 (32.4) \| \| \| > 1 \| 158 (25.7) \| 25 (22.5) \| 46 (26.0) \| 48 (22.0) \| 39 (36.1) \| The most common reasons for referral in the antepartum period were gestational diabetes ($13\%$) and pregnancy-induced hypertension ($6\%$) (Fig 1A). In the intrapartum period, the most common reason for referral were a request for pain relief ($13\%$) and a failure to progress in the first stage of labour ($12\%$) (Fig 1B). Fig 1: Reasons for referral antepartum & intrapartum.(A) Reasons for referral in the antepartum period. (B) Reasons for referral in the intrapartum period. There was limited missing data (<$1\%$), except for education ($3\%$) and preconception period ($7\%$), with overall $11\%$ missing observations. The proportion of missing data did not differ by referral and parity in the antepartum period; aOR of 0.86 ($95\%$CI, 0.46–1.57) for nulliparous women and 1.34 (0.81–2.21) for multiparous women. However, it was higher among multiparous women who were referred in the intrapartum period (2.52; 1.25–5.13). This means that in the complete-data sample of the intrapartum period, referred multiparous women were underrepresented (S1 Table). ## Antepartum The full and final (leanest) models of referral to obstetrician-led care in the antepartum period by parity are presented in Table 2A and their model performances in S2 Table; only the final models are described. Among nulliparous women, older women were more likely to be referred (aOR, 1.07; $95\%$CI, 1.05–1.09). Compared to women of healthy weight, those who were underweight were less likely to be referred (0.45; 0.31–0.64), whilst those who were overweight or obese were more likely to be referred (2.29; 1.91–2.74 or 2.65; 2.06–3.42, respectively). Women were also more likely to be referred when the preconception period was longer than a year (1.34; 1.07–1.66), they lived in a deprived area (1.87; 1.54–2.26), and had a history of sexual abuse (1.44; 1.14–1.82). Table 2 \| a: Association with referral antepartum (n = 1096) \| a: Association with referral antepartum (n = 1096).1 \| a: Association with referral antepartum (n = 1096).2 \| a: Association with referral antepartum (n = 1096).3 \| a: Association with referral antepartum (n = 1096).4 \| \| --- \| --- \| --- \| --- \| --- \| \| Variable \| Nulliparous \| Nulliparous \| Multiparous \| Multiparous \| \| Variable \| Full model \| Final model \| Full model \| Final model \| \| \| OR (95%CI) \| OR (95%CI) \| OR (95%CI) \| OR (95%CI) \| \| Maternal age (years) \| 1.07 (1.05–1.09) \| 1.07 (1.05–1.09) * \| 1.00 (0.99–1.02) \| \| \| BMI (kg/m2) \| \| \| \| \| \| <18.5 \| 0.45 (0.31–0.64) \| 0.45 (0.31–0.64) * \| 0.38 (0.25–0.58) \| 0.40 (0.26–0.60) * \| \| 18.5–24.9 \| 1.00 \| 1.00 \| 1.00 \| 1.00 \| \| 25.0–29.9 \| 2.27 (1.89–2.72) \| 2.29 (1.91–2.74) * \| 1.08 (0.92–1.27) \| 1.08 (0.92–1.27) \| \| ≥ 30.0 \| 2.64 (2.05–3.41) \| 2.65 (2.06–3.42) * \| 1.60 (1.30–1.97) \| 1.61 (1.30–1.98) * \| \| Preconception period (≤ 1 year/> 1 year) \| 1.35 (1.08–1.69) \| 1.34 (1.07–1.66) * \| 1.70 (1.27–2.28) \| 1.71 (1.27–2.28) * \| \| Education level \| \| \| \| \| \| Low \| 1.00 \| 1.00 \| 1.00 \| 1.00 \| \| Medium \| 0.74 (0.57–0.98) \| 0.76 (0.58–1.00) * \| 0.88 (0.71–1.08) \| 0.86 (0.70–1.06) \| \| High \| 0.93 (0.70–1.24) \| 0.94 (0.71–1.24) \| 0.65 (0.52–0.82) \| 0.63 (0.51–0.80) * \| \| Employment (no/yes) \| 1.09 (0.90–1.33) \| \| 1.43 (1.21–1.68) \| 1.38 (1.19–1.61) * \| \| Ethnicity (Dutch/non-Dutch) \| 0.98 (0.83–1.17) \| \| 1.08 (0.92–1.27) \| \| \| Lived in a deprived area (no/yes) \| 1.89 (1.54–2.32) \| 1.87 (1.54–2.26) * \| 1.20 (1.01–1.44) \| 1.23 (1.03–1.46) * \| \| Smoking (no/yes) \| 1.10 (0.92–1.33) \| \| 1.12 (0.94–1.34) \| \| \| Psychological problems (no/yes) \| 0.93 (0.77–1.12) \| \| 1.20 (1.01–1.42) \| 1.24 (1.06–1.44) * \| \| History of sexual abuse (no/yes) \| 1.48 (1.15–1.90) \| 1.44 (1.14–1.82) * \| 1.14 (0.90–1.44) \| \| \| Consultation obstetric care \| \| \| \| \| \| \| 1.00 \| \| 1.00 \| 1.00 \| \| 1 \| 0.88 (0.73–1.05) \| \| 0.68 (0.58–0.81) \| 0.68 (0.58–0.80) * \| \| > 1 \| 0.93 (0.77–1.14) \| \| 0.64 (0.53–0.76) \| 0.64 (0.54–0.76) * \| \| b: Association with referral intrapartum (n = 617) \| b: Association with referral intrapartum (n = 617) \| b: Association with referral intrapartum (n = 617) \| b: Association with referral intrapartum (n = 617) \| b: Association with referral intrapartum (n = 617) \| \| Variable \| Nulliparous \| Nulliparous \| Multiparous \| Multiparous \| \| Variable \| Full model \| Final model \| Full model \| Final model \| \| \| OR (95%CI) \| OR (95%CI) \| OR (95%CI) \| OR (95%CI) \| \| Maternal age (years) \| 1.03 (1.01–1.04) \| 1.02 (1.00–1.05) * \| 1.02 (1.00–1.05) \| 1.02 (1.00–1.04) * \| \| BMI (kg/m2) \| \| \| \| \| \| <18.5 \| 1.71 (1.18–2.48) \| 1.67 (1.15–2.42) * \| 0.86 (0.56–1.33) \| 0.84 (0.55–1.30) \| \| 18.5–24.9 \| 1.00 \| 1.00 \| 1.00 \| 1.00 \| \| 25.0–29.9 \| 1.00 (0.76–1.31) \| 0.98 (0.75–1.28) \| 0.65 (0.51–0.83) \| 0.65 (0.51–0.83) * \| \| ≥ 30.0 \| 1.15 (0.76–1.74) \| 1.14 (0.75–1.73) \| 0.84 (0.60–1.16) \| 0.83 (0.60–1.15) \| \| Preconception period (≤ 1 year/> 1 year) \| 0.42 (0.30–0.57) \| 0.42 (0.31–0.57) * \| 0.35 (0.18–0.68) \| 0.33 (0.17–0.65) * \| \| Education level \| \| \| \| \| \| Low \| 1.00 \| 1.00 \| 1.00 \| \| \| Medium \| 2.20 (1.56–3.11) \| 2.09 (1.49–2.91) * \| 1.26 (0.92–1.74) \| \| \| High \| 1.74 (1.20–2.52) \| 1.56 (1.10–2.22) * \| 0.79 (0.56–1.11) \| \| \| Employment (no/yes) \| 0.83 (0.64–1.08) \| \| 1.13 (0.90–1.43) \| \| \| Ethnicity (Dutch/non-Dutch) \| 1.05 (0.84–1.32) \| \| 1.98 (1.58–2.49) \| 1.98 (1.61–2.45) * \| \| Lived in a deprived area (no/yes) \| 1.05 (0.79–1.39) \| \| 1.21 (0.93–1.56) \| \| \| Smoking (no/yes) \| 1.16 (0.90–1.48) \| \| 0.73 (0.56–0.95) \| 0.75 (0.57–0.97) * \| \| Psychological problems (no/yes) \| 0.96 (0.75–1.23) \| \| 1.07 (0.85–1.36) \| \| \| History of sexual abuse (no/yes) \| 0.46 (0.32–0.65) \| 0.46 (0.33–0.63) * \| 1.50 (1.08–2.09) \| 1.49 (1.09–2.02) * \| \| Consultation obstetric care \| \| \| \| \| \| \| 1.00 \| 1.00 \| 1.00 \| 1.00 \| \| 1 \| 1.09 (0.86–1.38) \| 1.10 (0.87–1.39) \| 1.32 (1.04–1.69) \| 1.34 (1.06–1.70) * \| \| > 1 \| 1.44 (1.10–1.88) \| 1.49 (1.15–1.94) * \| 2.08 (1.62–2.66) \| 2.09 (1.63–2.69) * \| Among multiparous women, compared to those with a healthy weight, underweight women were less likely to receive a referral (0.40; 0.26–0.60) than obese women (1.61; 1.30–1.98). Women were also more likely to be referred if their preconception period was longer than one year (1.71; 1.27–2.28), they had a history of psychological problems (1.24; 1.06–1.44), they worked during pregnancy (1.38; 1.19–1.61), and lived in a deprived area (1.23; 1.03–1.46). In contrast, women were less likely to be referred when they had a high as opposed to a low education level (0.63; 0.51–0.80), and had one or more consultations in obstetric care (0.68; 0.58–0.80 or 0.64; 0.54–0.76, respectively). ## Intrapartum The full and final (leanest) models of referral to obstetrician-led care in the intrapartum period by parity are presented in Table 2B and their model performances in S2 Table; only the final models are described. Among nulliparous women, older women had increased odds of being referred (1.02; 1.00–1.05). Compared to women of healthy weight, those who were underweight were more likely to be referred (1.67; 1.15–2.42) as well as women with multiple consultations during the pregnancy (1.49; 1.15–1.94). Women who had a higher education level had increased odds of a referral compared to the lowest level: medium education level (2.09; 1.49–2.91) and high education level (1.56; 1.10–2.22). Women were less likely to be referred when their preconception period was more than one year (0.42; 0.31–0.57) or they had a history of sexual abuse (0.46; 0.33–0.63). Among multiparous women, older women had increased odds of being referred (1.02; 1.00–1.04). Women were more likely to be referred when they were from a non-Dutch ethnic group (1.98; 1.61–2.45), had a history of sexual abuse (1.49; 1.09–2.02), or had one or more consultations in obstetrician-led care (1.34; 1.06–1.70 or 2.09; 1.63–2.69, respectively). The odds of being referred were lower in those who were overweight compared to those being of a healthy weight (0.65; 0.51–0.83), had a preconception period of more than one year (0.33; 0.17–0.65), or were smokers (0.75; 0.57–0.97). ## Discussion Our study investigated the possible associations between multiple maternal characteristics and referral from midwife-led to obstetrician-led care during the antepartum and intrapartum periods. In addition to the more commonly researched maternal characteristics such as BMI and age, we showed that non-medical characteristics such as employment, education level, history of sexual abuse and consultations in obstetrician-led care are associated with referral, and differ by parity and partum period. Overall, maternal characteristics associated with referral during the antepartum period were age, BMI, preconception period, education level, employment, living in a deprived area, psychological problems, history of sexual abuse, and consultations in obstetric care. In the intrapartum period, maternal characteristics associated with referral were age, BMI, preconception period, education level, ethnicity, smoking, history of sexual abuse, and consultations in obstetrician-led care. In agreement with other studies, we found that age, BMI, ethnicity, smoking and living in a deprived area are associated with referral [18, 23, 25, 39–41]. A meta-analysis found that women with a preconception period of more than one year had an increased risk for preterm birth, low birth weight and small-for-gestational-age [21], which are all reasons for referral and thus agree with our findings in the antepartum period. However, in the intrapartum period, the effect was the opposite, which might be explained by the fact that over $50\%$ of these women were already referred to obstetrician-led care in the antepartum period. The number of consultations in obstetrician-led care (without transfer of care) had a mixed effect on referral to obstetrician-led care; this increased the likelihood of referral for both nulliparous and multiparous women in the intrapartum period, whereas in the antepartum period it had no significant effect for nulliparous women and a decreased effect for multiparous women. This has not been studied before, but we hypothesise that it comes down to the different reasons for consultations in obstetrician-led care. For instance, multiparous women with a history of postpartum haemorrhage or manual removal of the placenta need a consultation during their current pregnancy. However, they were not at risk for medical complications in the antepartum period. Multiparous but not nulliparous women who worked during pregnancy had higher odds of being referred to obstetrician-led care in the antepartum period. Although we cannot explain the different effects by parity, a meta-analysis showed that physically demanding work is significantly associated with hypertension or preeclampsia and preterm birth, which are reasons for referral [22]. Nulliparous women with a medium or high education level compared to women with a low education level had a higher chance of being referred to obstetric care in the intrapartum period. This is contrary to other studies that demonstrate a lower education level is associated with more medical complications during the intrapartum period [19, 23]. The relatively small sample size could be a reason for this contradicting result, as only 27 women had a low education level. A history of sexual abuse had a mixed effect on referral to obstetrician-led care; during the antepartum period, it was associated with higher odds for nulliparous women and no association for multiparous women, whilst during the intrapartum period it was associated with lower odds for nulliparous women and higher odds for multiparous women. This is not consistent with the literature, which shows that a history of sexual abuse is associated with more psychological problems and adverse perinatal outcomes [20, 42]. Overall, for the antepartum and intrapartum periods, the associations between maternal characteristics and a referral differed. Most of the characteristics in the antepartum and intrapartum periods were comparable, except for a BMI≥ 30; in the antepartum period, $13\%$ of the women had a BMI ≥ 30 compared to $9\%$ of the women in the intrapartum period. Since a BMI ≥ 30 is associated with more complications during pregnancy, these women might already have been referred to obstetrician-led care, which could explain the difference in the associations [17]. ## Strength and limitations The strength of our study is the availability of a large number of maternal characteristics, including non-medical ones such as psychosocial and lifestyle factors. Data were collected directly from the midwife practice’s medical records rather than from routinely registered data, resulting in more reliable data. Our low-risk study population included only one midwifery practice. Nationally, the rate of referrals varies by practice and location. A nationwide retrospective cohort study reported that intrapartum referrals of nulliparous or multiparous women range between 55–$68\%$ and 20–$32\%$, respectively (in our study $62\%$ and $33\%$, respectively). The care providers’ assessment of risk and uncertainty, as well as regional guidelines, may impact the referral rate [43]. Therefore, we cannot generalise our results to all low-risk women in the Netherlands. Our study had a higher proportion of non-Dutch women and smokers compared to the Dutch maternity population, and this could lead to more referrals to obstetrician-led care [44–47]. During our study period, all women with gestational diabetes were referred to obstetrician-led care. The current policy in this region is that women who only need a diet to stabilise their glucose values are not referred to obstetrician-led care. This policy is no longer in use in this region. Finally, our study had limited missing data, which were appropriately addressed by multiple imputations [32, 33]. ## Practical implications and future research Our study showed that multiple maternal characteristics are associated with a referral to obstetrician-led care in this particular midwifery practice. In particular, there are a number of non-medical characteristics associated with a referral that can affect the course of pregnancy and birth, but these are currently not considered in maternity care [26, 27]. Non-medical issues could be addressed at an earlier stage, preferably during preconception or at the beginning of pregnancy. For instance, addressing the lack of social support is important during pregnancy, as this is associated with adverse perinatal outcomes [48, 49]. Therefore, we would advocate more facilities for midwives, as they often provide close care to women and know their social environment. This would allow midwives to treat women with non-medical factors more intensely. Care models such as midwife-led care, which emphasise continuity of care, are in themselves important for the well-being of the mother and child [1]; and other midwife-led care models such as CenteringPregnancy (group sessions), facilitate social support and enable women to improve their self-confidence [50]. The women in this study had access to other caregivers. Continuity of care might be beneficial for non-medical maternal characteristics, particularly since case-load midwifery care is associated with a lower referral rate [51]. Support from other caregivers is also accessible in the midwife-led care model, whereby the midwife can support women with the help of a psychologist, dietitian, or welfare worker. As preventive care focused on non-medical issues may benefit medical care, we advocate further research into the relationship between non-medical maternal characteristics and referral, as well as how midwives can improve midwife-led care for women with psychosocial or lifestyle issues (non-medical characteristics). ## Conclusion To our knowledge, this is the first study that illustrates a large number of non-medical maternal characteristics of low-risk pregnant women that are associated with a referral from midwife-led to obstetrician-led care in the antepartum and intrapartum periods, both for nulliparous and multiparous women. In particular, certain characteristics such as living in a deprived area, unemployment and a history of sexual abuse might benefit from other care models or interventions. These include case-load care or resilience-enhancing interventions such as CenteringPregnancy, as well as the supportive care of a welfare worker or psychologist. We advocate further research to increase awareness of the influence non-medical characteristics have on referral, as well as research about interventions that could improve modifiable maternal characteristics in the preconception and/or antepartum period. ## Ethics approval and consent to participate Informed consent was obtained verbally and noted in women’s medical records under women’s supervision [8]. All methods were performed in accordance with the relevant guidelines and regulations (Declaration of Helsinki). ## References 1. Sandall J, Soltani H, Gates S, Shennan A, Devane D. Midwife-led continuity models versus other models of care for childbearing women.. The Cochrane database of systematic reviews. (2016.0) 4 Cd004667. DOI: 10.1002/14651858.CD004667.pub5 2. 2Royal College of Obstetricians and Gynaecologists. The National Sentinel Caesarean Section Audit Report. London: RCOG Clinical Effectiveness Support Unit, 2001. [ISBN 1–900364–66–2.].. The National Sentinel Caesarean Section Audit Report (2001.0) 3. 3Perined. Perinatal Care in the Netherlands 2002–2021. In Dutch, available from: https://www.perined.nl/onderwerpen/publicaties-perined/jaarboek-zorg Accessed februari 2018. 4. Amelink-Verburg MP, Buitendijk SE. Pregnancy and Labour in the Dutch Maternity Care System: What Is Normal? The Role Division Between Midwives and Obstetricians.. Journal of Midwifery & Women’s Health. (2010.0) 55 216-25. DOI: 10.1016/j.jmwh.2010.01.001 5. Offerhaus PM. Patterns in primary midwife-led care in the Netherlands. Trends and variation in intrapartum referrals.. Radboud University Nijmegen (2015.0) 9-10 6. Nederland SPR. Stichting Perinatale Registratie Nederland. Perinatal Care in the Netherlands 2004.. Drukkery Tessink 2007 91-5 7. 7De Koninklijke Nederlandse Organisatie van Verloskundige—Verloskundig vademecum. 2003 VDA-groep, Apeldoorn. Available from: https://www.knov.nl/serve/file/knov.nl/knov_downloads/769/file/Verloskundig%20Vademecum%202003.pdf. Accessed 3 Jan 2018. 8. Gitsels-van der Wal JT, Gitsels LA, Hooker A, van Weert B, Martin L, Feijen-de Jong EI. Determinants and underlying causes of frequent attendance in midwife-led care: an exploratory cross-sectional study.. BMC pregnancy and childbirth. (2019.0) 19 203. DOI: 10.1186/s12884-019-2316-5 9. Amelink-Verburg MP, Rijnders ME, Buitendijk SE. A trend analysis in referrals during pregnancy and labour in Dutch midwifery care 1988–2004. BJOG: an international journal of obstetrics and gynaecology (2009.0) 116 923-32. DOI: 10.1111/j.1471-0528.2009.02202.x 10. Offerhaus PM, de Jonge A, van der Pal-de Bruin KM, Hukkelhoven CW, Scheepers PL, Lagro-Janssen AL. Change in primary midwife-led care in the Netherlands in 2000–2008: a descriptive study of caesarean sections and other interventions among 789,795 low risk births.. Midwifery (2014.0) 30 560-6. DOI: 10.1016/j.midw.2013.06.013 11. Perdok H, Jans S, Verhoeven C, van Dillen J, Mol BW, de Jonge A. Intrapartum referral from primary to secondary care in the Netherlands: a retrospective cohort study on management of labor and outcomes.. Birth (Berkeley, Calif). (2015.0) 42 156-64. DOI: 10.1111/birt.12160 12. Schuit E, Hukkelhoven CW, van der Goes BY, Overbeeke I, Moons KG, Mol BW. Risk indicators for referral during labor from community midwife to gynecologist: a prospective cohort study.. The journal of maternal-fetal & neonatal medicine: the official journal of the European Association of Perinatal Medicine, the Federation of Asia and Oceania Perinatal Societies, the International Society of Perinatal Obstet. (2016.0) 29 3304-11. DOI: 10.3109/14767058.2015.1124080 13. Carolan M.. Maternal age >/ = 45 years and maternal and perinatal outcomes: a review of the evidence.. Midwifery (2013.0) 29 479-89. DOI: 10.1016/j.midw.2012.04.001 14. Khalil A, Syngelaki A, Maiz N, Zinevich Y, Nicolaides KH. Maternal age and adverse pregnancy outcome: a cohort study.. Ultrasound in obstetrics & gynecology: the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. (2013.0) 42 634-43. DOI: 10.1002/uog.12494 15. Lean SC, Derricott H, Jones RL, Heazell AEP. Advanced maternal age and adverse pregnancy outcomes: A systematic review and meta-analysis.. PloS one. (2017.0) 12 e0186287. DOI: 10.1371/journal.pone.0186287 16. Sheen JJ, Wright JD, Goffman D, Kern-Goldberger AR, Booker W, Siddiq Z. Maternal age and risk for adverse outcomes. American journal of obstetrics and gynecology (2018.0) 219 390. DOI: 10.1016/j.ajog.2018.08.034 17. Stubert J, Reister F, Hartmann S, Janni W. The Risks Associated With Obesity in Pregnancy.. Deutsches Arzteblatt international. (2018.0) 115 276-83. DOI: 10.3238/arztebl.2018.0276 18. Agyemang C, Vrijkotte TG, Droomers M, van der Wal MF, Bonsel GJ, Stronks K. The effect of neighbourhood income and deprivation on pregnancy outcomes in Amsterdam, The Netherlands. Journal of epidemiology and community health (2009.0) 63 755-60. DOI: 10.1136/jech.2008.080408 19. Cantarutti A, Franchi M, Monzio Compagnoni M, Merlino L, Corrao G. Mother’s education and the risk of several neonatal outcomes: an evidence from an Italian population-based study.. BMC pregnancy and childbirth. (2017.0) 17 221. DOI: 10.1186/s12884-017-1418-1 20. Dahlen HG, Munoz AM, Schmied V, Thornton C. The relationship between intimate partner violence reported at the first antenatal booking visit and obstetric and perinatal outcomes in an ethnically diverse group of Australian pregnant women: a population-based study over 10 years. BMJ Open (2018.0) 8 e019566. DOI: 10.1136/bmjopen-2017-019566 21. Maclagan L, Messerlian C, Basso O. Infertility and the risk of adverse pregnancy outcomes: a systematic review and meta-analysis. Human Reproduction (2012.0) 28 125-37. DOI: 10.1093/humrep/des347 22. Mozurkewich EL, Luke B, Avni M, Wolf FM. Working conditions and adverse pregnancy outcome: a meta-analysis.. Obstetrics & Gynecology. (2000.0) 95 623-35. DOI: 10.1016/s0029-7844(99)00598-0 23. Savitz DA, Kaufman JS, Dole N, Siega-Riz AM, Thorp JM, Kaczor DT. Poverty, education, race, and pregnancy outcome.. Ethnicity & disease. (2004.0) 14 322-9. PMID: 15328932 24. Staneva A, Bogossian F, Pritchard M, Wittkowski A. The effects of maternal depression, anxiety, and perceived stress during pregnancy on preterm birth: A systematic review. Women and birth: journal of the Australian College of Midwives (2015.0) 28 179-93. DOI: 10.1016/j.wombi.2015.02.003 25. Zwart JJ, Jonkers MD, Richters A, Ory F, Bloemenkamp KW, Duvekot JJ. Ethnic disparity in severe acute maternal morbidity: a nationwide cohort study in the Netherlands.. European journal of public health (2011.0) 21 229-34. DOI: 10.1093/eurpub/ckq046 26. Vos AA, Leeman A, Waelput AJM, Bonsel GJ, Steegers EAP, Denktaş S. Assessment and care for non-medical risk factors in current antenatal health care.. Midwifery (2015.0) 31 979-85. DOI: 10.1016/j.midw.2015.06.008 27. Lagendijk J, Vos AA, Bertens LCM, Denktas S, Bonsel GJ, Steyerberg EW. Antenatal non-medical risk assessment and care pathways to improve pregnancy outcomes: a cluster randomised controlled trial. European journal of epidemiology (2018.0) 33 579-89. DOI: 10.1007/s10654-018-0387-7 28. Nesari M, Olson JK, Vandermeer B, Slater L, Olson DM. Does a maternal history of abuse before pregnancy affect pregnancy outcomes? A systematic review with meta-analysis.. BMC pregnancy and childbirth. (2018.0) 18 404. DOI: 10.1186/s12884-018-2030-8 29. 29Vektis. Postcodetabel achterstandswijken. In Dutch, available from: https://tog.vektis.nl/TogDownloads.aspx. Accessed May 2022. 30. 30World Health Organisation—BMI. Available from: https://wwweurowhoint/en/health-topics/disease-prevention/nutrition/a-healthy-lifestyle/body-mass-index-bmi. Accessed 10 Aug 2021. 31. Offerhaus PM, Hukkelhoven CW, de Jonge A, van der Pal-de Bruin KM, Scheepers PL, Lagro-Janssen AL. Persisting rise in referrals during labor in primary midwife-led care in the Netherlands.. Birth (Berkeley, Calif). (2013.0) 40 192-201. DOI: 10.1111/birt.12055 32. Donders AR, van der Heijden GJ, Stijnen T, Moons KG. Review: a gentle introduction to imputation of missing values. Journal of clinical epidemiology (2006.0) 59 1087-91. DOI: 10.1016/j.jclinepi.2006.01.014 33. Sterne JAC, White IR, Carlin JB, Spratt M, Royston P, Kenward MG. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ (2009.0) 338 b2393. DOI: 10.1136/bmj.b2393 34. Halligan S, Altman DG, Mallett S. Disadvantages of using the area under the receiver operating characteristic curve to assess imaging tests: a discussion and proposal for an alternative approach.. European radiology. (2015.0) 25 932-9. DOI: 10.1007/s00330-014-3487-0 35. Twisk JWR. Inleiding in de toegepaste biostatistiek (2016.0) 256-73 36. Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N. Assessing the performance of prediction models: a framework for traditional and novel measures.. Epidemiology (2010.0) 21 128-38. DOI: 10.1097/EDE.0b013e3181c30fb2 37. Fischer JE, Bachmann LM, Jaeschke R. A readers’ guide to the interpretation of diagnostic test properties: clinical example of sepsis.. Intensive care medicine. (2003.0) 29 1043-51. DOI: 10.1007/s00134-003-1761-8 38. Swets JA. Measuring the accuracy of diagnostic systems. Science (New York, NY). (1988.0) 240 1285-93. DOI: 10.1126/science.3287615 39. Rode L, Nilas L, Wøjdemann K, Tabor A. Obesity-Related Complications in Danish Single Cephalic Term Pregnancies.. Obstetrics & Gynecology. (2005.0) 105 537-42. DOI: 10.1097/01.AOG.0000152304.39492.1c 40. Schimmel MS, Bromiker R, Hammerman C, Chertman L, Ioscovich A, Granovsky-Grisaru S. The effects of maternal age and parity on maternal and neonatal outcome. Archives of gynecology and obstetrics (2015.0) 291 793-8. DOI: 10.1007/s00404-014-3469-0 41. Wang Y, Tanbo T, Abyholm T, Henriksen T. The impact of advanced maternal age and parity on obstetric and perinatal outcomes in singleton gestations. Archives of gynecology and obstetrics (2011.0) 284 31-7. DOI: 10.1007/s00404-010-1587-x 42. Brunton R, Dryer R. Child Sexual Abuse and Pregnancy: A Systematic Review of the Literature.. Child Abuse & Neglect. (2021.0) 111 104802. DOI: 10.1016/j.chiabu.2020.104802 43. Seijmonsbergen-Schermers AE, Zondag DC, Nieuwenhuijze M, van den Akker T, Verhoeven CJ, Geerts CC. Regional variations in childbirth interventions and their correlations with adverse outcomes, birthplace and care provider: A nationwide explorative study.. PloS one. (2020.0) 15 e0229488. DOI: 10.1371/journal.pone.0229488 44. 44Centraal Bureau voor de Statistiek. 2019. Available from: https://www.cbs.nl/en-gb. Accessed 20 Apr 2019. 45. 45C.I. Lanting, J.P. van Wouwe, P. van Dommelen, K.M. van der Pal-de Bruin. Roken tijdens de zwangerschap. TNO: 2016. In Dutch. Available from: https://www.tno.nl/media/6211/factsheet_roken_tijdens_de_zwangerschap.pdf. Accessed 20 Apr 2019. 46. 46M. Tuithof, R. Siauw, S. van Dorsselaer, K. Monshouwer—Factsheet Monitor Zwangerschap en Middelengebruik. Trimbos instituut: 2017. In Dutch. Available from: file:///C:/Users/Susan/Downloads/factsheet-monitor-zwangerschap—en-middelengebruik_3.pdf. Accessed 20 May 2020. 47. de Graaf H., Wijssen C.. Seksuele gezondheid in Nederland (2017.0) 48. Oakley A, Rajan L, Grant A. Social support and pregnancy outcome. British journal of obstetrics and gynaecology (1990.0) 97 155-62. DOI: 10.1111/j.1471-0528.1990.tb01741.x 49. Orr ST. Social support and pregnancy outcome: a review of the literature.. Clinical obstetrics and gynecology. (2004.0) 47 842-55. DOI: 10.1097/01.grf.0000141451.68933.9f 50. Kweekel L, Gerrits T, Rijnders M, Brown P. The Role of Trust in CenteringPregnancy: Building Interpersonal Trust Relationships in Group-Based Prenatal Care in The Netherlands.. Birth (Berkeley, Calif). (2017.0) 44 41-7. DOI: 10.1111/birt.12260 51. Offerhaus P, Jans S, Hukkelhoven C, de Vries R, Nieuwenhuijze M. Women’s characteristics and care outcomes of caseload midwifery care in the Netherlands: a retrospective cohort study.. BMC Pregnancy and Childbirth (2020.0) 20 517. DOI: 10.1186/s12884-020-03204-3
--- title: Prevalence of Overweight and Obesity in Jamaica From 2000 to 2016 journal: Cureus year: 2023 pmcid: PMC10016753 doi: 10.7759/cureus.34907 license: CC BY 3.0 --- # Prevalence of Overweight and Obesity in Jamaica From 2000 to 2016 ## Abstract The prevalence of overweight and obesity in Jamaica has been steadily increasing over the past decade and is now a significant health issue. This paper focuses on the trends in the prevalence of overweight and obesity in Jamaica from 2000 to 2016. Overweight and obesity prevalence in adults increased from $43.8\%$ in 2000 to $55.5\%$ in 2016, from $34.2\%$ in 2000 to $47.4\%$ in 2016in adult males, and from $53.0\%$ in 2000 to $63.6\%$ in 2016 in adult females. In children/adolescents aged 10 to 19 years, the prevalence of obesity has doubled between 2000 and 2016. The data shows that the prevalence of overweight and obesity in children/adolescents increased from $5\%$ in 2000 to $11.4\%$ in 2016, from $4.4\%$ in 2000 to $11.0\%$ in 2016 in boys, and from $5.5\%$ in 2000 to $11.9\%$ in 2016 in girls. ## Introduction The term overweight is defined as a body mass index (BMI) of over 25 kg/m2, while obesity is defined as a BMI of over 30 kg/m2 [1]. Overweight and obesity have become significant global public health concerns with severe health, psychological, and economic burdens [2]. The prevalence of overweight and obesity has been steadily increasing over the past four decades in both developed and developing countries [3]. According to the World Health Organization (WHO), in 2016, more than 1.9 billion adults were overweight, and 650 million were obese. Being overweight and obese is associated with numerous health complications. The common ones are inflammation, diabetes mellitus, and cardiovascular disease. However, multiple studies have shown that there is an associated cancer burden in people with excess body weight; the ones with the most evidence are breast cancer, endometrial cancer, esophageal adenocarcinoma, and kidney cancers [4]. There is also a link between asthma and obesity in childhood [5]. Studies have also shown that severely obese people are at high risk for depression due to poor body image [6]. Obesity can also have a severe financial burden on a country’s economy [7]. Based on a study conducted in 2015 the direct cost of diabetes mellitus in Jamaica was between US 567 million and US 765 million dollars [8]. In 2017, Jamaica estimated that cardiovascular diseases and diabetes combined will cost US 77 billion dollars over the next 15 years, this involves treatment costs and the loss of productivity from persons who are affected within those two categories alone [9]. These are all health complications that are associated with being overweight and obese. The prevalence of overweight and obesity has increased substantially and is now a significant health concern for Jamaica. In 2016 $24.7\%$, approximately one in four adults in Jamaica were obese. That same year the country ranked 55 out of 191 countries worldwide based on the percentage of the adult population that was obese and ranked fourth among Caribbean Community and Common Market (CARICOM) member states [10]. This report aims to describe the trends in the prevalence of overweight and obesity in Jamaica from 2000 to 2016. Data from the Pan American Health Organization (PAHO) shows a steady increase in the prevalence of overweight and obesity in Jamaica. Analyzing and highlighting the relevance of this data is vital to help reduce the prevalence of overweight and obesity throughout the next decade. Therefore, this may aid in establishing a solid foundation for better analysis of the root problem in the region so that prevention and treatment strategies may be better implemented. ## Materials and methods Study design This was a secondary analysis study of data obtained from the PAHO database [11]. The data available for the prevalence of overweight and obesity from 2000 to 2016 among the Jamaican population was analyzed and summarized for both the adult and children/adolescent cohorts. Data collection The data used was collected from the PAHO core indicators database. The PAHO core indicators database provides the latest data on health indicators for 49 countries and territories in the Region of the Americas. The primary sources used to create these health indicators are demographic censuses, national health information systems, population surveys, and data from health facilities [12]. Data analysis Values were expressed as a percentage of the population. The data obtained was organized and tabulated within Microsoft Excel (Microsoft Corp., Redmond, WA, USA). Further analysis to obtain the p-value and correlation of data was done using the SPSS (IBM Corp., Armonk, NY, USA) statistics tool. ## Results An overview of the data obtained from the PAHO on the prevalence of overweight and obesity from 2000 to 2016 in the Jamaican population is presented in Table 1. Figure 1 highlights the trend of overweight and obesity among adults, while Figure 2 highlights the trend of overweight and obesity among children/adolescents from 2000 to 2016. The data showed that the prevalence of overweight and obesity among the Jamaican population had been steadily increasing from 2000 to 2016, with the prevalence of overweight and obesity among the children/adolescents population doubling. From 2000 to 2016, there was a $13.2\%$ and $10.2\%$ increase in the prevalence of overweight and obesity in adult males and adult females, respectively. The average prevalence of overweight and obesity over the 16 years for adult males was $40.71\%$ (standard deviation (SD)=4.16), while for adult females it was $49.69\%$ (SD=3.70). In the children/adolescents population, there was a $6.6\%$ increase in the male cohort and a $6.4\%$ increase in the female cohort. The average prevalence of overweight and obesity over the 16 years for boys was $7.43\%$ (SD=2.07), while for girls it was $8.45\%$ (SD=2.02). Overall, the increase in the prevalence of overweight and obesity in the adult population was $10.2\%$ with an average of $58.12\%$ (SD=3.23) and $6.4\%$ in the children/adolescents population with an average of $7.99\%$ (SD=2.04). ## Discussion Being overweight and obese are serious and important issues for Jamaica. Obesity-related comorbidities such as diabetes mellitus and cardiovascular diseases have become one of the leading causes of death in the Jamaican population, with cardiovascular diseases alone accounting for $27\%$ of deaths under 70 years [13]. This indicates that appropriate prevention and treatment strategies are both crucial issues for the country. Adolescents/children with obesity are more likely to develop other serious health issues. Childhood obesity is associated with a higher chance of more aggressive asthma attacks, premature death, and disability in adulthood. Additionally, obese children may experience psychological issues such as depression [6,14]. Given the long-term impact of obesity on children and the rapid rate of increase in its prevalence, acknowledging and tackling this issue is essential for the sustained health of Jamaican youth. In 2000, the prevalence of overweight and obesity was $34.2\%$ in adult males, and $53.0\%$ in adult females. In 2016 these percentages rose to $47.4\%$ in adult males and $63.2\%$ in adult females, which shows that the prevalence of overweight and obesity is much higher in adult females compared to adult males. Another crucial point to consider is that while the prevalence of obesity and overweight is higher in adult females, the absolute increase in adult males was $13.2\%$ ($p \leq 0.01$). In comparison, it was $10.2\%$ in adult females ($p \leq 0.01$), suggesting that the prevalence of overweight and obesity is increasing faster in adult males compared to adult females. In 2016, the prevalence for both genders was $55.5\%$, an $11.7\%$ increase from the prevalence for both genders in 2000, which was $43.8\%$. For some context, the global prevalence of overweight and obesity in 2016 was $39.0\%$, which would put the prevalence of overweight and obesity among adults in Jamaica at $16.5\%$ higher than the global average. In 2000, the prevalence of overweight and obesity in adolescents aged 10 to 19 years was $4.4\%$ in males and $5.5\%$ in females. These figures climbed to $11.0\%$ and $11.9\%$ in 2016. Females consistently maintained a higher prevalence of overweight and obesity throughout the 16 years. However, the absolute increase in the prevalence amongst males in this cohort is a significant observation to note as well. There was a $6.6\%$ increase in males ($p \leq 0.01$); in comparison, females had a lower absolute increase of $6.4\%$ ($p \leq 0.01$), indicating that even though the prevalence of obesity and overweight was higher in females, the rate of increase was slightly higher in males. In 2016 the prevalence for both genders was $11.4\%$, a $6.4\%$ increase from the prevalence for both genders in 2000 which was $5.0\%$. The global prevalence of overweight and obesity among adolescents in 2016 was $18\%$ [1]. Jamaica’s prevalence of overweight and obesity was only $6.6\%$ lower than the global average. In 2011 the ministry of health, per the ministry of education in Jamaica, conducted a health-promoting school survey [15]. The objective of this study was to gather data to aid in establishing policies and programs for school health. The study looked at various indicators that affect students' health in 60 schools across the 14 parishes of the country. This was an excellent initiative to promote proper health awareness among children/adolescents. However, this strategy should also be implemented in other sectors of the country. Health education should be provided to all people at all levels on the critical impacts of overweight and obesity on children and adults, thus increasing the public's awareness. This may be achieved through multilevel community and school-based interventions [16]. Additionally, the prevalence of obesity may also be managed by implementing fitness programs in the curriculum of educational facilities and workplaces. A major limitation of this study is that the data analyzed and presented could only show the prevalence of overweight and obesity in the Jamaican population from a general perspective [17]. The data available do not allow for analysis of the prevalence of overweight and obesity in various subgroups and geographical locations across the country. ## Conclusions The data reflects a significant increase in the prevalence of overweight and obesity in adults and children/adolescents in the Jamaican population between 2000 and 2016. For both age groups, females consistently had a higher prevalence than males. However, the data for males in both cohorts reflected a more significant rate of increase in the prevalence of overweight and obesity between 2000 and 2016. ## References 1. Obesity. (2023) 2. Williams EP, Mesidor M, Winters K, Dubbert PM, Wyatt SB. Overweight and obesity: prevalence, consequences, and causes of a growing public health problem. Curr Obes Rep (2023) 4 363-370. PMID: 26627494 3. Ng M, Fleming T, Robinson M. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet (2014) 384 766-781. PMID: 24880830 4. Sung H, Siegel RL, Torre LA. Global patterns in excess body weight and the associated cancer burden. CA Cancer J Clin (2019) 69 88-112. PMID: 30548482 5. Matricardi PM, Grüber C, Wahn U, Lau S. The asthma-obesity link in childhood: open questions, complex evidence, a few answers only. Clin Exp Allergy (2007) 37 476-484. PMID: 17430342 6. Dixon JB, Dixon ME, O'Brien PE. Depression in association with severe obesity: changes with weight loss. Arch Intern Med (2003) 163 2058-2065. PMID: 14504119 7. Tremmel M, Gerdtham UG, Nilsson PM, Saha S. Economic burden of obesity: a systematic literature review. Int J Environ Res Public Health (2017) 14 8. Barcelo A, Arredondo A, Gordillo-Tobar A, Segovia J, Qiang A. The cost of diabetes in Latin America and the Caribbean in 2015: evidence for decision and policy makers. J Glob Health (2017) 7 20410 9. Cardiovascular diseases and diabetes to cost $77 billion over next 15 years, Jamaica Information Service. (2023) 10. Most obese countries in the world. (2023) 11. Core indicators dashboard, PAHO/EIH Open Data. (2023) 12. Pan American Health Organization. Health Indicators. Conceptual and operational considerations. Health Indicators. Conceptual and operational considerations (2018) 13. Noncommunicable diseases data portal, Country profile. (2023) 14. Sagar R, Gupta T. Psychological aspects of obesity in children and adolescents. Indian J Pediatr (2018) 85 554-559. PMID: 29150753 15. Chin D, McFarlane N L. Health promoting school survey 2011. (2013) 16. Vo L, Albrecht SS, Kershaw KN. Multilevel interventions to prevent and reduce obesity. Curr Opin Endocr Metab Res (2019) 4 62-69. PMID: 31538131 17. Wickham RJ. Secondary analysis research. J Adv Pract Oncol (2019) 10 395-400. PMID: 33343987
--- title: 'Discovery of Novel Human Constitutive Androstane Receptor Agonists with the Imidazo[1,2-a]pyridine Structure' authors: - Ivana Mejdrová - Jan Dušek - Kryštof Škach - Alžbeta Stefela - Josef Skoda - Karel Chalupský - Klára Dohnalová - Ivona Pavkova - Thales Kronenberger - Azam Rashidian - Lucie Smutná - Vojtěch Duchoslav - Tomas Smutny - Petr Pávek - Radim Nencka journal: Journal of Medicinal Chemistry year: 2023 pmcid: PMC10017030 doi: 10.1021/acs.jmedchem.2c01140 license: CC BY 4.0 --- # Discovery of Novel Human Constitutive Androstane Receptor Agonists with the Imidazo[1,2-a]pyridine Structure ## Abstract The nuclear constitutive androstane receptor (CAR, NR1I3) plays significant roles in many hepatic functions, such as fatty acid oxidation, biotransformation, liver regeneration, as well as clearance of steroid hormones, cholesterol, and bilirubin. CAR has been proposed as a hypothetical target receptor for metabolic or liver disease therapy. Currently known prototype high-affinity human CAR agonists such as CITCO (6-(4-chlorophenyl)imidazo[2,1-b][1,3]thiazole-5-carbaldehyde-O-(3,4-dichlorobenzyl)oxime) have limited selectivity, activating the pregnane X receptor (PXR) receptor, a related receptor of the NR1I subfamily. We have discovered several derivatives of 3-(1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine that directly activate human CAR in nanomolar concentrations. While compound 39 regulates CAR target genes in humanized CAR mice as well as human hepatocytes, it does not activate other nuclear receptors and is nontoxic in cellular and genotoxic assays as well as in rodent toxicity studies. Our findings concerning potent human CAR agonists with in vivo activity reinforce the role of CAR as a possible therapeutic target. ## Introduction The constitutive androstane receptor (CAR, NR1I3) is a ligand-activated transcription factor belonging to the nuclear receptor subfamily NR1I. Human CAR is dominantly expressed in hepatocytes. While the endogenous ligands of human CAR are obscure, a number of naturally occurring steroids such as androstanol, androstenol, and 5β-pregnane-3,20-dione have been proposed as endogenous inverse agonists in supraphysiological concentrations.1,2 Recent animal studies with a mouse agonist suggest that CAR plays an important role in the metabolism of glucose, lipids, and fatty acids as well as in the endobiotic metabolism of bile acids, cholesterol, bilirubin, and thyroid hormones.3 It has been proposed in several independent animal studies that CAR activation may ameliorate glucose homeostasis and insulin sensibility in the treatment of type 2 diabetes.4,5 In addition, since CAR activation affects the expression of lipogenic genes in mice, this might also be a promising therapeutic intervention in the treatment of human obesity, steatosis, or hypercholesterolemia,4,6−9 although contradictory and species-specific reports also exist.10−12 CAR activators have been also proposed as a potential therapy for steatohepatitis or liver regeneration.13,14 So far, only two human CAR crystal structures with a human agonist bound have been reported.15 The CAR ligand-binding domain (LBD) cavity has a mostly hydrophobic and flexible character with a pocket size of 675 Å.3,8,16 The hydrophobic cavity suggests that human CAR ligands are mostly highly lipophilic compounds. Human CAR displays unique properties in comparison with other nuclear receptors as well as its rodent orthologues. CAR variant 1 (wtCAR, CAR1, and wild-type CAR) exhibits strong constitutive activity that can be further activated by agonists or repressed by inverse agonists. In addition, both direct LBD-dependent and LBD-independent activation are known for CAR. Human CAR is present in at least three transcript variants (wtCAR, CAR2, and CAR3) in the liver, which differ in their ligand-dependent activation and basal constitutive activities. The wild-type variant CAR (348 AA, NM_005122.4, and transcript variant 3) features high constitutive activity in the regulation of basal expression of target genes and high sensitivity for inverse agonists. This variant represents about $40\%$ of CAR transcripts in the liver parenchyma. The variant CAR3, also called CAR-SV2 (353 AA, XM_005245697.4, transcript variant X4), which has an insertion of the five amino acids APYLT into the LBD, represents $50\%$ of transcripts. CAR3 has low constitutive activity but is highly inducible by ligands and much more active in the upregulation of CAR target genes in the liver. The transcript variant CAR2 (352 AA, NM_001077480.2) is a minor variant with moderate induction activity. The exact physiological functions of the variants are obscure, but several selective activators of individual variants have been described in the literature.8,17−19 There are no highly potent, specific, and drug-like (with suitable physicochemical and ADME properties) agonists of the human CAR receptor without off-target effects that can be therapeutically used or can serve as a tool in therapeutic intervention with human CAR ligands. The unique properties of human CAR, mainly its hydrophobic pocket and high constitutive activity, make the discovery of specific ligands difficult.20 Therefore, determining suitable drug candidate molecules targeting human CAR and high-affinity endogenous ligands remains problematic.21 The only compound known to date is 6-(4-chlorophenyl)imidazo[2,1-b]thiazole-5-carbaldehyde O-(3,4-dichlorobenzyl)oxime (CITCO, 1), which is a potent human—but not a mouse—CAR agonist.22 However, this highly lipophilic compound also significantly activates the related pregnane X receptor (NR1I2, PXR) of the same subfamily through π–π interactions with the W299 residue.22−24 This may exert an unfavorable effect on glycemia and liver steatosis.25 On the contrary, the prototype mouse CAR ligand 1,4-bis[(3,5-dichloropyridine-2-yl)oxy]benzene (TCPOBOP) does not activate human CAR.26 Different strategies have been used in high-content CAR ligand screenings recently performed, including nuclear translocation assays with an adenoviral-enhanced yellow fluorescent protein-tagged hCAR (Ad/EYFP-hCAR) vector in hepatocytes,27,28 mammalian one-hybrid assays using a fusion protein of CAR or its LBD,21,23,29−31 and assays employing stable luciferase reporter cell lines expressing wtCAR and treated with an inverse agonist,32 as well as with a CAR3-selective screening method combined with other CAR assays.33 In addition, studies employing pharmacophore computational modeling and the virtual screening of chemical databases have been performed.30,34 In the past, several CAR activators with various structural features have been discovered in the screened libraries (Figure 1A) or after modification of the lead compound CITCO [1], Figure 1B.27−29,33−37 Figure 1: CAR activators discovered by screening chemical libraries (A) or modifications of CITCO as the lead compound (B). TCPOBOP is a mouse CAR ligand. These human CAR ligands, however, still have limited potency to activate human CAR in nanomolar concentrations in comparison with the prototype high-affinity CAR ligand CITCO. Limited studies are currently being undertaken which explore structure–activity relationship variations by systematic synthesis on the human CAR ligand after the initial hit compound discovery or modification of the human CAR agonist CITCO as a template. Recently, Liang et al. specifically modified the 4-chlorophenyl, imidazothiazole, and 3,4-dichlorphenyl groups of CITCO.36 Especially, their discovered compound (E)-6-(4-chlorophenyl)imidazo[2,1-b]oxazole-5-carbaldehyde O-(3,4-dichlorobenzyl)oxime and compound DL5050, ((E)-6-(naphthalen-2-yl)imidazo[2,1-b]oxazole-5-carbaldehyde O-(3,4-dichlorobenzyl)oxime) with the imidazoxazole core exert increased potency and selectivity for human CAR activation over human PXR in a human CAR1-expressing reporter cell line and primary human hepatocytes (PHH).36 In an additional study, Liang et al. synthesized a library of CITCO analogues with the 6-(4-chlorophenyl)imidazo[2,1-b]oxazole core as well as with modified 4-chlorophenyl or 3,4-dichlorophenyl rings with a variety of substituted arene moieties. In all these novel compounds, the oxime linker of CITCO, which might cause chemical instability, was replaced by groups such as amine, amide, imine, and ether.37 In their study, compound DL5016 (N-((6-(naphthalen-2-yl)imidazo[2,1-b]oxazol-5-yl)methyl)-2,3-dihydro-1H-inden-2-amine), which has an EC50 value of 0.66 μM in cellular reporter assays, appeared as an efficient and selective human CAR agonist with lower PXR activation than CITCO. In addition, the ligand was shown to induce receptor translocation into the nucleus, to upregulate the expression of the human CAR target gene, and to enhance the efficacy of cyclophosphamide-based cytotoxicity to non-Hodgkin lymphoma cells (Figure 1).37 Very recently, DL7076 (CN06) has been discovered as a dual activator of the CAR and nuclear factor erythroid 2-related factor 2 (Nrf2).38 Recently, we have described 2-(3-methoxyphenyl)quinazoline derivatives modified at position 4 with 4-methoxy, 4-methylthio, or 4(1H)-thione) moieties as potent but nonspecific human CAR ligands also activating PXR and vitamin D receptors.39 Similarly, a human CAR agonist FL81 (5-(3,4-dimethoxybenzyl)-3-phenyl-4,5-dihydroisoxazole) discovered by another group also activates the PXR receptor to some extent.23 Interestingly, in recent years, several inverse agonists of human CAR have been discovered with IC50 in submicromolar/nanomolar concentrations such as PK11195 (1-(2-chlorophenyl)-N-methyl-N-(1-methylpropyl)-3-isoquinolinecarboxamide),40 S07662 (1-[(2-methylbenzofuran-3-yl)methyl]-3-(thiophen-2-ylmethyl) urea),21 and CINPA1 ([5-[(diethylamino)acetyl]-10,11-dihydro-5H-dibenz[b,f]azepin-3-yl]carbamic acid ethyl ester).41 In the present work, we aimed to discover selective human CAR agonists that do not activate PXR or other nuclear receptors but still possess suitable ADME properties for further experiments in human hepatocyte cellular models or application in humanized CAR mouse models. In a library of kinase inhibitors, we found two lead structures (compounds 2 and 3) which appeared to be analogues to the known human CAR ligand CITCO. ## Results and Discussion We initially synthesized two analogues 2 and 3 of a known yet unspecific CAR agonist, CITCO, by modifying the middle flexible oxime linker to the triazole ring (Figure 2). The oxime moiety is unstable under acidic conditions, which may complicate its use in vivo. The triazole ring offered stability, less flexibility, and good accessibility via an undemanding click reaction. Because of the synthetic feasibility and possibility to expand the variety of derivatives via a single reaction, the CuAAC reaction has been chosen. Different substitution patterns of blue, green, and red areas allowed us to explore the SAR of the compounds regarding the binding site, the bioavailability of the prepared compounds, and selectivity/specificity toward the key receptors. Our decisions were also based on preliminary docking data (e.g., Figure S-3). Figure 2: Known human CAR ligand compound 1 (CITCO) and two lead structures (compounds 2 and 3) with areas of modification in substituted phenyl ring (blue), central heterocyclic linker (red), and substituted benzyl ring (green). We found that analogues 2 and 3 significantly activate both CAR and PXR. Their activities and affinities toward human CAR were similar to those of CITCO in both the recombinant CAR LBD-dependent TR-FRET assay and cellular luciferase reporter assays. Their potency toward PXR was, however, more significant in comparison with the compound CITCO (Figure 3). Compound 2 displayed less cytotoxicity in COS-1 cells than compound 3 (Table S-1). Figure 3: Lead compounds 2 and 3 significantly activate CAR and PXR in the TR-FRET LanthaScreen CAR coactivator assay (CAR TR-FRET), in the CAR LBD assembly assay (CAR AA), or in the PXR-responsive luciferase reporter assay. EC50 (in μM) values were obtained based on sigmoidal dose–response fitting. Activities of CITCO and rifampicin, a PXR agonist, at 10 μM are set to be 100%. Compound 2 with the original imidazo[2,1-b]thiazole moiety of CITCO and compound 3 with imidazo[1,2-a]pyridine moiety were further modified in three key areas: blue (substituted phenyl ring), red (central heterocyclic linker), and green (substituted benzyl ring) (Figure 2). ## Design and Synthesis of the First Generation of Novel CAR Ligands The first modification of the phenyl and later benzyl ring led to two series A and B based on the compounds 2 and 3, respectively. The synthesis of the key compounds 2 and 3 as well as their modified analogues with a preserved triazole central heterocyclic linker is illustrated in Scheme 1. Scheme 1: Preparation of the Lead Compounds and AnaloguesReagents and conditions: (a) for example, 2-bromo-1-(4-chlorophenyl)ethan-1-one, NaHCO3, EtOH, 70 °C, o.n.; (b) NIS, DCM, 25 °C; (c) TMS-acetylene, CuI, TEA, Pd(PPh3)2Cl2, DMF, 0–25 °C; (d) 4-(azidomethyl)-1,2-dichlorobenzene, CuSO4·5H2O, KF, Na-ascorbate, THF/H2O (1:1), 0–25 °C, 1 h. The synthesis of both series started from 2-aminothiazole 4 (for series A) or 2-aminopyridine 5 (for series B). Cyclization with appropriate phenylacetyl chloride with various substitution patterns (Table 1) in EtOH at 70 °C led to 6-substituted imidazo[2,1-b]thiazoles 6a–i or 2-substituted imidazo[1,2-a]pyridines 7a–i. Iodination with NIS in DCM led to the iodinated intermediates 8a–i and 9a–i, respectively, in high to quantitative yields. A subsequent Sonogashira reaction with TMS-acetylene under Pd(PPh3)2Cl2 catalysis resulted in compounds 10a–i and 11a–i. Lastly, a triazole ring was formed via a CuAAC click reaction, yielding final compounds 2, 12g, and 14a–f and 3, 13b–i, and 15a–m (Table 1). Table 1 \| comp. \| yielda (%) \| comp \| yielda (%).1 \| comp.1 \| yielda (%).2 \| comp.2 \| yielda (%).3 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| 2 \| 90.0 \| 13f \| 74.0 \| 14d \| 87 \| 15f \| 82 \| \| 12g \| 92.0 \| 13g \| 90.0 \| 14e \| 80 \| 15g \| 80 \| \| 3 \| 57.0 \| 13h \| 89.0 \| 14f \| 87 \| 15h \| 80 \| \| 13b \| 84.0 \| 13i \| 92.0 \| 15a \| 92 \| 15i \| 92 \| \| 13c \| 70.0 \| 14a \| 88.0 \| 15b \| 81 \| 15j \| 93 \| \| 13d \| 74.0 \| 14b \| 86.0 \| 15c \| 75 \| 15k \| 88 \| \| 13e \| 86.0 \| 14c \| 86.0 \| 15d \| 78 \| 15l \| 78 \| \| \| \| \| \| 15e \| 83 \| 15m \| 92 \| ## Biology The compounds with a substituted phenyl ring (12g–13i) appeared as potent agonists of the CAR with nanomolar EC50 in a CAR TR-FRET assay. However, these compounds also significantly activated PXR, and most of them decreased the viability of COS-1 or HepG2 cells (Tables 2 and S-1). Table 2 \| Comp. \| CAR TR-FRET EC50 (μM) \| CAR AA EC50 (μM) \| CAR3 % CITCO activityb \| PXR % RIF activityb \| \| --- \| --- \| --- \| --- \| --- \| \| 2 \| 0.003 ± 0.0001 \| 1.16 ± 0.5 \| 78 ± 4 \| 136 ± 8 \| \| 3 \| 0.005 ± 0.001 \| 1.35 ± 0.5 \| 167 ± 12 \| 53 ± 4 \| \| 12g \| 0.016 \| Nd \| 411 ± 13 \| 57 ± 5 \| \| 13b \| 0.0003 \| ndtox. \| 184 ± 21 \| 47 ± 5 \| \| 13c \| 0.0003 \| ndtox. \| ndtox. \| 68 ± 7 \| \| 13d \| 0.001 \| ndtox. \| ndtox. \| 55 ± 4 \| \| 13e \| 0.001 \| 1.34 ± 0.2 \| ndtox. \| 42 ± 3 \| \| 13f \| 0.062 \| Nd \| 166 ± 10 \| 128 ± 8 \| \| 13g \| 0.002 \| ndtox. \| 323 ± 27 \| 43 ± 4 \| \| 13h \| nd# \| ndtox. \| ndtox. \| ndtox. \| \| 13i \| <0.001 \| ndtox \| nd tox. \| 89 ± 5 \| \| 14a \| 0.007 \| Nd \| 143 \| 86 ± 10 \| \| 14b \| 0.432 \| 0.15 ± 0.02 \| 171 \| 111 ± 7 \| \| 14c \| 0.656 \| Nd \| 170 \| 109 ± 10 \| \| 14d \| 0.007 \| Nd \| 91 ± 9 \| 215 ± 21 \| \| 14e \| 0.01 \| Nd \| 109 ± 9 \| 170 ± 13 \| \| 14f \| >5 \| Nd \| 15 ± 2 \| 87 ± 8 \| \| 15a \| 0.002 \| 0.05 ± 0.01 \| 78 ± 4 \| 120 ± 10 \| \| 15b \| 0.019 \| Nd \| 123 ± 7 \| 89 ± 9 \| \| 15c \| 0.040 \| 0.12 ± 0.01 \| 135 ± 11 \| 117 ± 10 \| \| 15d \| 0.001 \| 0.46 ± 0.02 \| 176 ± 12 \| 195 ± 12 \| \| 15e \| 1.38 \| Nd \| 134 ± 7 \| 212 ± 21 \| \| 15f \| 0.06 \| 0.12 ± 0.01 \| 33 ± 4 \| 59 ± 8 \| \| 15g \| 0.011 \| 2.76 ± 0.2 \| 62 ± 5 \| 25 ± 2 \| \| 15h \| 0.08 \| 0.04 ± 0.007 \| 44 ± 7 \| 68 ± 7 \| \| 15i \| 0.009 \| 3.05 ± 0.8 \| 18 ± 2 \| 5 ± 0.2 \| \| 15j \| no activity \| no activity \| 3 ± 0.3 \| 16 ± 2 \| \| 15k \| >5 \| no activity \| 33 ± 4 \| 21 ± 3 \| \| 15l \| >5 \| 0.50 ± 0.01 \| 73 ± 6 \| 29 ± 4 \| \| 15m \| no activity \| no activity \| 12 ± 0.4 \| 20 ± 4 \| \| CITCO \| 0.012 ± 0.004 \| 0.69 ± 0.04 \| \| \| \| 10 μM \| \| \| 396 ± 25 \| 27 ± 5 \| \| 1 μM \| \| \| 100% \| 8 ± 1 \| \| Rifampicin 10 μM \| \| \| \| 100% \| Compounds 14a–14f were also found as potent agonists of both CAR and PXR. Similarly, compounds 15a–h displayed significant activation of CAR and PXR with some moderate effects on cellular viability. Among these compounds, 15d was found as a highly efficient CAR agonist, while at the same time, it significantly activated PXR with Emax higher than that of rifampicin (Figure 4 and Table 2). Showing an opposite result, compounds 15i–m have marginal activities on the CAR and weak activity toward PXR (Table 2). Figure 4: Activities of compounds 15i and 15d to stimulate the CAR in the TR-FRET LanthaScreen CAR Coactivator assay (CAR TR-FRET), in the CAR LBD assembly assay (CAR AA), or in the PXR-responsive luciferase reporter assay. EC50 values (μM) were obtained based on sigmoidal dose–response curve fitting. Activities of CITCO and rifampicin at 10 μM are set to be 100%. Compound 15i appeared to be a selective CAR ligand. However, its activity in the TR-FRET CAR coactivator assay was negligible, and its activity in the CAR LBD assembly assay was weaker compared to CITCO (Figure 4). Thus, compound 15i demonstrates the phenomenon that some high-potency compounds in the TR-FRET assay with nanomolar EC50 but less efficacy in cellular assays are, in fact, partial agonists of the CAR. These compounds do not reach the maximal activity (Emax) of full agonists such as CITCO or compound 15d (Figure 4). We should also consider the possibility that the tested compounds are likely distributed into cell membranes in cellular assays, which results in lower potency (higher EC50 in CAR AA and CAR3 assays) in comparison with the in vitro TR-FRET assay. In contrast, compound 15j with a sulfonyl pyrrolidine moiety does not possess any activity to the CAR. We suppose that it is too bulky to fit into the CAR LBD domain (Table 2). We can conclude that the substitution of the phenyl ring with a lipophilic moiety increased activities for both the CAR and PXR. Similarly, lipophilic substitution or no substitution on the benzyl ring increased the nonselective activation of CAR and PXR. Compound 15d was found as an efficient dual CAR/PXR agonist (Figure 4). Interestingly, compounds 15f and 15h displayed high potency for wtCAR in TR-FRET and CAR AA assays (with EC50 in the nanomolar range), but they were less potent in the CAR3 assay, suggesting some selectivity for the wtCAR variant over the CAR3 variant (Table 2). When we tested the compounds with the modified middle heterocyclic linker, we found that the central moiety also contributes to both CAR and PXR activation, although there were no dramatic variations in the effects of different heterocycles. Out of the series of compounds, compound 19B showed higher relative selectivity to the CAR as it significantly activates CAR and CAR3, but it tends to activate PXR from 5 μM (Figure 5). Similarly, compound 21 has high activity toward the CAR in the CAR LBD assembly assay, but it has a significant activity to PXR at a 1 μM concentration. Interestingly, in the CAR TR-FRET assay, the compound seems to be a partial agonist with the Emax lower than that of CITCO or compound 19B (Table 4, Figure 5). Figure 5: Relative activities of compounds 19B and 21 to activate CAR in TR-FRET CAR coactivator assays (CAR TR-FRET), in the CAR LBD assembly assay (CAR AA), or the PXR-responsive luciferase reporter assay. EC50 values (μM) were obtained based on sigmoidal dose–response curve fitting. Activities of CITCO and rifampicin at 10 μM are set to be $100\%$. TABLE_PLACEHOLDER:Table 4 No significant cytotoxicity was observed for these compounds in COS-1 and HepG2 cells (Table S-2). When we tested derivatives of compound 3 (Series B) with acyl moieties in the meta position of the benzyl ring with the preserved triazole ring as a linker (37–44), we found that the moieties significantly contribute to CAR activation but not to PXR activation. Carboxylic acid itself in the meta position [38] resulted in a complete loss of CAR activation. However, amides, as well as esters, significantly activated CAR in all assays. Compound 37 appeared as the most efficient to activate the CAR in the TR-FRET LanthaScreen CAR coactivation assay and highly efficient to activate a CAR LBD assembly assay with an EC50 lower than that of CITCO (EC50 = 0.4 and 152 nM vs 12 and 690 nM, respectively). Importantly, a methylester (compound 37), amides (39, 40, and 41) as well as N-methoxy-N-methylamide (compound 42) all have minimal (37 and 41) or no activity (39, 40, and 42) to activate PXR at 10 μM (Figure 6 and Table 7). Other compounds from the set also display low activation of PXR. These compounds were also noncytotoxic in viability assays (Table S-3). Figure 6: Activation of CAR and PXR in TR-FRET LanthaScreen CAR coactivation assay (CAR TR-FRET), in the CAR LBD assembly assay (CAR AA), or the PXR-responsive luciferase assay. EC50 values (μM) were obtained based on sigmoidal dose–response curve fitting. Activities of CITCO and rifampicin at 10 μM are set to be $100\%$. TABLE_PLACEHOLDER:Table 7 When we looked in detail at the CAR agonists 39, 40, 41, and 42 without significant PXR activation, their potencies in the TR-FRET LanthaScreen CAR coactivation assay were by an order of magnitude lower (and EC50 higher) than that of CITCO and compound 39 seems to be a partial agonist of the CAR in the assay. In the case of CAR LBD assembly and CAR3 variant assays, these compounds activated the CAR LBD with lower but still comparable affinities in comparison with CITCO. This phenomenon may be explained by the different activation of the CAR LBD by these compounds via another coactivator than with PGC1α, which is involved in the TR-FRET CAR coactivation assay. Indeed, SRC-1 (NCOA1) along with other coactivators are important for the coactivation of CAR8 and we can suppose an array of different coactivators in CAR variants activation in cellular assays. We may also suppose the intracellular accumulation of these compounds, for example, via an uptake mechanism, which may increase their potencies in cellular CAR LBD assembly and CAR3 variant assays, but not in the TR-FRET LanthaScreen CAR coactivation assay. Interestingly, compounds 39, 40, and 42 moderately deactivated the PXR-responsive construct in concentrations higher than 10 μM. Compounds with substitution of the meta position of the benzyl ring with other substituents (43–51) and with the preserved triazole ring as a linker retained efficient CAR activation with high potency (EC50 below 0.1 μM) (Table 7), although these compounds also activate PXR to some degree. Some of these display substantial effects on cellular viability (Table S-3), which may affect cellular assays. With a methoxyethyl moiety, compound 48 was found to activate the CAR LBD assembly assay with the lowest EC50 = 24.9 nM; however, the compound is not selective for the CAR and significantly activates PXR (EC50 = 4.34 ± 1 μM). Compound 48 was also highly potent in the activation of the CAR3 variant in the CAR3 variant assay (Table 7). Interestingly, some compounds such as 45 and 47 had high potency for wtCAR in the TR-FRET and CAR AA assays, but they were less potent in the CAR3 assay, suggesting some selectivity for the wtCAR variant. Compounds 60–61 with a replaced methylene part of the linker with O and NH (Table 6) lost the activity to the CAR and retained a weak activity to PXR. ## Design and Synthesis of the Second Generation of Novel CAR Ligands For the next series of compounds, we focused on the middle heterocyclic linker. The other rings (phenyl and benzyl) maintained the substitution pattern of compounds 2 and 3. The triazole ring was replaced by several heterocycles with one to three heteroatoms, such as thiadiazol or oxazole. The complete list is shown in Table 3. Table 3 \| comp. \| yielda (%) \| comp..1 \| yielda (%).1 \| \| --- \| --- \| --- \| --- \| \| 16A \| 52 \| 20.0 \| 66.0 \| \| 16B \| 50 \| 21.0 \| 21.0 \| \| 17 \| 6 \| 22.0 \| 32.0 \| \| 18 \| 21 \| 23.0 \| 73.0 \| \| 19A \| 61 \| 24.0 \| 95.0 \| \| 19B \| 62 \| \| \| Compounds 16A, 16B, 17, and 18 originated from the same precursors 25 or 26, which were synthesized by a condensation reaction of 4 or 5 with ethyl 3-(4-chlorophenyl)-3-oxopropanoate in the presence of CBr4. The ethyl ester moiety of precursors 25 and 26 was hydrolyzed using LiOH·H2O, and the obtained acid derivatives 27 and 28 were treated with EDC and HOBt at 25 °C, followed by the addition of substituted N′-hydroxyacetimidamide at 80 °C to yield compounds 16A and 16B (Scheme 2). Scheme 2: Preparation of Novel Middle-Ring Heterocyclic AnaloguesReagents and conditions: (a) ethyl 3-(4-chlorophenyl)-3-oxopropanoate, CBr4, CH3CN, 80 °C, o.n., 82%; (b) N2H4·H2O (3 equiv), EtOH, reflux, o.n., 87%; (c) ethyl 2-(4-chlorophenyl)imidazo[1,2-a]pyridine-3-carboxylate, HATU, DIPEA, DMF, 25 °C, o.n., 90%; (d) tosyl chloride (1.5 equiv), TEA (3 equiv), DCM, 0 °C for 17, 6%; Lawesson’s reagent (3 equiv), toluene, 100 °C, o.n. for 18, 21%; (e) LiOH·H2O, THF/H2O 4:1, 25 °C, 3 h, quant.; (f) (E)-2-(3,4-dichlorophenyl)-N′-hydroxyacetimidamide, EDC, HOBt, DMF, 25–80 °C, o.n., 52% resp 50%. In order to synthesize compounds 17 and 18, the ester derivative 26 was reacted with an excess of hydrazine hydrate in EtOH, providing compound 29 and further acylated with 2-(3,4-dichlorophenyl)acetic acid by means of the peptide coupling reagent HATU in DMF. Ring-closing reaction of 30 with tosyl chloride at 25 °C or Lawesson’s reagent at 100 °C overnight led to final compounds 17 and 18 respectively, although at very low yields (Scheme 2). Thiazole analogues 19A and 19B were prepared from intermediates 6a and 7a, which were reacted with an excess of chloroacetyl chloride in dry dioxane at 70 °C for 30 min and then heated up to 100 °C overnight, followed by cyclization with ethanethioamide (1.5 equiv) in EtOH at reflux (Scheme 3). Scheme 3: Preparation of Novel Thiazole Analogues of the Lead CompoundsReagents and conditions: (a) chloroacetyl chloride (3 equiv), dioxane, 70 °C, 30 min, 100 °C, o.n., 91% resp 76%; (b) 2-(3,4-dichlorophenyl)ethanethioamide, EtOH, reflux, o.n., 61% resp 62%. Sonogashira reaction of 9a with TMS acetylene under Pd(PPh3)2Cl2 catalysis provided intermediate 11a, following deprotection of the TMS group with K2CO3 in MeOH yielded compound 33. Pretreatment of phenylacetaldehyde with hydroxylamine hydrochloride and following reaction with intermediate 33 in the presence of chloramine T and CuI at 25 °C provided compound 20. The click reaction of compound 33 with TMSN3 and CuI in a DMF/MeOH mixture under an inert atmosphere led to the unsubstituted triazole derivative 34 (Scheme 4). Scheme 4: Preparation of Oxazole Derivative 20 and Metabolite M3 (34)Reagents and conditions: (a) K2CO3, MeOH, 25 °C, 2 h, 90%; (b) phenylacetaldehyde, NH2OH·HCl, NaOH, H2O/t-BuOH, CuI, chloramine T, 25 °C, o.n., 66%; (c) TMSN3, CuI, DMF/MeOH 10:1, 70 °C, 84%. Pyrrole and pyrazole derivatives (21 and 22) were obtained in two-step synthesis starting from iodinated precursor 9a, which was coupled in a Suzuki reaction under Pd(PPh3)4 catalysis with (1-(tert-butoxycarbonyl)-1H-pyrrol-3-yl)boronic acid or 4-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)-1H-pyrazole with Pd(dppf)Cl2 as a catalyst, respectively, followed by a substitution reaction with benzyl chloride under basic reaction conditions (Scheme 5). Compounds 23 and 24 were synthesized from intermediate 11a via a click reaction with 4-azido-1,2-dichlorobenzene or 4-(2-azidoethyl)-1,2-dichlorobenzene according to Scheme 1. Scheme 5: Synthesis of Pyrrole and Pyrazole Derivatives 21 and 22Reagents and conditions: (a) (1-(tert-butoxycarbonyl)-1H-pyrrol-3-yl)boronic acid, Pd(PPh3)4, Na2CO3, dioxane/H2O, 90 °C, o.n., 35%; (b) 4-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)-1H-pyrazole, Pd(dppf)Cl2, Na2CO3, dioxane/H2O, 90 °C, o.n., 40%; (c) 1,2-dichloro-4-(chloromethyl)benzene, NaH, DMF, 25 °C, o.n., 21%; (d) 1,2-dichloro-4-(chloromethyl)benzene, CH3CN, K2CO3, 25 °C, o.n., 32%. ## Design and Synthesis of the Third Generation of Novel CAR Ligands Since the most promising biological results were found with compound 3 derivatives, in the next step, we addressed the modification of the benzyl ring with the main emphasis on the meta position while maintaining the triazole ring. The chlorine atom was replaced with a series of acyl molecules (Table 5). This resulted in improved estimated water solubility and bioavailability of the compounds. The key compound 37 was prepared by the same means as compound 3 (Scheme 1) with a minor modification in the azide coupling partner. Straightforward hydrolysis led to acid analogue 38, which was converted to amide analogues 39–42 via acylchloride 52 (Scheme 6). Scheme 6: Synthesis of Lead Compound 3 Analogues with Benzyl Ring ModificationReagents and conditions: (a) LiOH·H2O, THF/H2O, 25 °C, 2 h, 92%; (b) SOCl2, toluene; (c) NHR1R2, DIPEA, 88–95%; (d,e) MeMgBr or EtMgBr, THF, 0–25 °C, 72% resp. 69%; (f) LAH (1 equiv), THF, 5 °C, 2 h, 75%; (g) K2CO3, MeOH, 25 °C, 2 h; (h) CH3I, K2CO3, DMF, 65%; (i) 45, NH2OH·HCl, DCM, 2 h, 90%; (j) HCl/diethylether, THF, 20 min, 0 °C, quant. In order to increase bioavailability, compound 39 was converted to its HCl salt 39 HCl (Scheme 6). The subsequent reaction of N-methoxy-N-methylbenzamide derivative 42 with a Grignard reagent or LAH at low temperature yielded ketone [43,44] or aldehyde analogues 45, which upon the reaction with hydroxylamine provided derivative 46. Finally, the reduction of the ester derivative with subsequent methylation provided compounds 47 and 48 (Scheme 6). N-acyl derivative 51 was synthesized in two steps from nitro derivative 49 after reduction and the succeeding acylation reaction (Scheme 7). Scheme 7: Synthetic Pathway of Nitro, Amino, and Acetylamino DerivatesReagents and conditions: (a) 4-(azidomethyl)-1-chloro-2-nitrobenzene, CuSO4·5H2O, KF, Na-ascorbate, THF/H2O, 25 °C, 1 h, 83% (b) AcOH, Fe, MeOH, reflux, 79%; (c) Ac2O, pyridine, dioxane, 25 °C, o.n., 90%. Next, we decided to broaden the number of examples of heterocyclic linkers with six-membered heterocycle pyridine and to replace the original triazole with an aryl ring. Syntheses of these compounds started from iodinated precursor 9a, which was coupled with aryl/pyridyl boronic acid (Scheme 8), followed by the Negishi coupling reaction with benzylzinc bromide catalyzed by Pd, providing compounds 55 and 56, respectively. Unfortunately, these compounds were barely soluble, so we did not test them. Scheme 8: Preparation of Six-Membered Heterocyclic AnaloguesReagents and conditions: (a) (4-bromophenyl)boronic acid or (6-chloropyridin-3-yl)boronic acid, dioxane/H2O mixture (4:1), Na2CO3, Pd(dppf)Cl2·DCM, 95 °C, o.n., 59% resp 63%; (b) Pd2dba3, XantPhos, benzylzinc bromide solution 0.5 M in THF, THF, 60 °C, o.n., 72% resp 75%. Moreover, the methylene part of the linker was exchanged for O and NH. Similarly, to the previously mentioned compounds, intermediate 9a was coupled with appropriate boronic acid (Scheme 9), providing intermediate compounds 57 and 58 with a free amino group and methoxy group, respectively. Demethylation of compound 58 afforded compound 59 with a free hydroxy group. Both compounds 57 and 59 were coupled with 5-bromo-2-chlorobenzamide by way of Buchwald or Ullmann coupling conditions, yielding compounds 60 and 61 (Scheme 9, Table 6). Scheme 9: Synthesis of Pyridine Derivatives with a N/O LinkerReagents and conditions: (a) 2-aminopyridine-5-boronic acid pinacol ester, dioxane, Na2CO3 in 2 mL of H2O, Pd(dppf)Cl2, 90 °C, o.n., $53\%$; (b) 5-bromo-2-chlorobenzamide, dioxane, NatBuO, XantPhos, Pd2dba3, 100 °C, o.n., $51\%$; (c) (6-methoxypyridin-3-yl)boronic acid, dioxane, Na2CO3, H2O, Pd(dppf)Cl2, 90 °C, o.n., $53\%$; (d) 4 M HCl/dioxane, 95 °C, o.n., $73\%$; (e) 5-bromo-2-chlorobenzamide, BPPO, CuI, K3PO4, DMF, 110 °C, o.n., $6\%$. TABLE_PLACEHOLDER:Table 6 ## Characterization of Induction Properties of Selected Candidates in Human Hepatocyte Models and Their Interactions with Human CAR Variants Next, we decided to analyze our novel selective CAR agonists 37, 39, 40, 41, and 42 to determine whether they could upregulate CYP2B6 gene mRNA, the typical CAR target gene, in PHH from one donor. We found that all compounds could significantly upregulate CYP2B6 mRNA. Compounds 39 and 42 tend to be the most potent with a 1 μM concentration in the experiments. These data suggest that the compounds are metabolically stable in metabolically competent hepatocyte cells and that they enter hepatocytes to activate the CAR (Figure 7A). Figure 7: Induction of CAR target genes in primary human hepatocyte models and interactions of selected candidates with CAR transcription variants and mouse CAR. (A) PHHs were treated with compounds 37, 39, 40, 41, and 42 together with CITCO for 24 h. The expression of CYP2B6 mRNA, a prototype CAR target gene, was analyzed using RT-qPCR in technical triplicates. Data are presented as mRNA fold induction to control (vehicle-treated) samples. (B) Translocation experiments with EGFP-hCAR + Ala chimera in COS-1 cells treated with tested compounds (10 μM) for 24 h before confocal microscopy. Data are presented as % of cells with specific cytoplasm or mixed/nuclear localization of pEGFP-hCAR + Ala chimeric protein. (C) Interactions of compounds 37, 39, 40, 41, and 42 with human wtCAR in the CAR LBD assembly assay (wtCAR AA) or with wtCAR inhibited with PK11195 (0.1 μM), with CAR2 or CAR3 variants, or with mouse Car (mCar). (D) PHHs from five donors were treated with compound 39 and CITCO (1 and 10 μM, respectively) for 48 h. CAR target genes CYP2B6, CYP3A4, and CYP2C9 mRNA expression have been studied using RT-qPCR. Data are presented as fold induction to control (vehicle-treated) samples. Western blotting experiments with primary human (BioIVT) treated with comp. 39, rifampicin (rif), CITCO, and PXR antagonist SPA70 (10 μM) for 48 h. Monoclonal anti-CYP2B6 antibody (PA5-35032) was used to detect CYP2B6 protein. (E,F) HepaRG cells and HepaRG KO CAR cells without functional CAR activity were treated with phenobarbital (500 μM), CITCO, rifampicin (10 μM), or compound 39 at a 1 μM concentration for 48 h. CYP2B6 and CYP3A4 mRNA expression have been analyzed using RT-qPCR. (G) LS174T cells expressing PXR, but without functional CAR, were treated with compound 39, rifampicin (rif), CITCO, SPA70, or compound 39 at a 10 μM concentration for 48 h. CYP2B6 and CYP3A4 mRNA expression has been analyzed using RT-qPCR. (H) Luciferase gene reporter assay with the CYP3A4 gene promoter construct (p3A4-luc) in HepG2 cells transfected with either PXR or CAR3 expression constructs. Cells were treated for 24 h before analysis. (I) Dose–response activation of CAR2 and CAR3 variants with compound 39 in luciferase reporter gene assays. p < 0.05 and *p < 0.01-significant CYP2B6 or CYP3A4 mRNA upregulation, p3A4-luc activation or EGFP-hCAR + Ala fusion protein nuclear translocation to control samples; f-statistically significant effect of SPA70 on rifampicin-mediated CYP3A4 mRNA expression or activation of the p3A4-luc luciferase construct. In translocation experiments with the EGFP-hCAR + Ala chimera, we examined the tested compounds to determine whether they stimulate cytoplasm-to-nuclear translocation of the activated human CAR with extra alanine in the LBD (CAR + A).42 We noted that mainly CITCO and compounds 37, 39, and 40 significantly decrease the number of cells with specific cytoplasm EGFP-hCAR + Ala localization, and they increase the portion of cells with nuclear localization of the CAR chimera (Figures 7B and S-1). In these experiments, compound 39 appeared as the most promising candidate. In agreement with cellular assays or induction experiments in PHHs, tested compounds have similar activities in comparison with CITCO (comp. 1) or compound 37, which are high-affinity CAR agonists in TR-FRET CAR assays. These data suggest that the cellular environment and signaling have a significant determination on CAR activation. In the next experiments, we sought to determine whether the discovered selective CAR agonists 37, 39, 40, 41, and 42 interact with wild-type CAR (wtCAR), human CAR variants 2 (CAR2) and 3 (CAR3), as well as with mouse CAR orthologue in luciferase reporter assays. Efficacy to activate wtCAR was assessed using the CAR LBD assembly assay (CAR AA) or with a wtCAR expression vector that was inhibited with PK111195 (0.1 μM), a known CAR inhibitor. We found that compound 37 is highly efficient in the stimulation of the variant CAR2 and other variants of CAR in comparison with CITCO ($100\%$ activity). Compound 39 significantly activated wtCAR in the CAR LBD assembly assay and the CAR3 variant in the gene reporter assay. Its activity in the assay with wtCAR and its inhibitor PK11195, however, was low, suggesting a weak efficacy to compete with the PK111195 inhibitor in the CAR LBD. Other candidate compounds have lower potency in comparison to CITCO ($100\%$ activity) in the activation of CAR variants. Compound 42 appeared as a combined agonist of wtCAR and its variants in all assays. Only compound 37 was found to stimulate the mouse CAR when compared to the mouse ligand TCPOBOP (Figure 7C). In the follow-up studies, we examined the stability of compound 37 in human and mouse microsomes and in plasma. We found that compound 37 is unstable in both mouse and human microsomes with t$\frac{1}{2}$ = 4.78 ± 1.31 min and t$\frac{1}{2}$ = 6.38 ± 0.67 min, respectively. Importantly, we found that compound 37 is also unstable in mouse plasma as well with t$\frac{1}{2}$ = 22.76 ± 0.03 min (Figure S-2). ## Selection of the Candidate for Animal Studies and Detailed Characterization of Compound 39 In the next experiments, we studied the most efficient compound 39 in five PHHs from five different donors to determine whether it could upregulate CYP2B6, CYP3A4, and CYP2C9 mRNA. These genes are significantly, but not exclusively, regulated via the CAR in human hepatocytes. Despite high variability in response in different hepatocyte preparations, we found that compound 39 has similar activity to induce these genes in comparison with CITCO (Figure 7D). Western blotting experiments in PHHs (BioIVT) treated with comp. 39, rifampicin, CITCO, and PXR antagonist SPA70 (10 μM) for 48 h revealed that compound 39 up-regulates CYP2B6 protein and that the upregulation is not abolished by the PXR antagonist SPA70 (Figure 7D, inserted panel). To confirm that compound 39 induces CYP2B6 mRNA via the activated CAR, we performed experiments with HepaRG and its KO CAR counterpart cell line without CAR expression. We observed the upregulation of CYP2B6 and CYP3A4 mRNA only in the HepaRG cells but not in the HepaRG KO CAR cells after treatment with both CITCO and compound 39 (Figure 7E,F). In the next experiments, CYP2B6 and CYP3A4 mRNA expression were analyzed in LS174T cells using RT-qPCR. LS174T cells express endogenous PXR but lack functional CAR.43 We did not observe any significant induction of these genes by compound 39 in these cells (Figure 7G). Then, we performed luciferase gene reporter assays with the CYP3A4 gene promoter construct (p3A4-luc) in HepG2 cells transfected with either PXR or CAR3 expression constructs. Compound 39 activated the luciferase construct only in the presence of CAR3, and the PXR antagonist SPA70 had no significant effect on the activation (Figure 7H). Finally, we examined the dose–response activation of CAR2 and CAR3 variants with compound 39 (Figure 7I). Unfortunately, the profiles of the dose–response curves did not reach the plateau phase and did not allow us to calculate EC50 and Emax values in the range of concentrations up to 30 μM (Figure 7I). Based on the data, we can conclude that compound 37 is the most active ligand for all CAR variants. Compound 39 displayed the most significant activity in the induction experiments in PHH and HepaRG cells irrespective of their lower affinities to wtCAR or CAR3 variants in CAR TR-FRET and cellular assays as well as marginal activity toward the CAR2 variant. In addition, we found that compound 39 does not induce CYP2B6 or CYP3A4 mRNA via PXR activation. Next, we considered the physicochemical properties of selected compounds such as molecular weight (Mw), Log S (the solubility of a substance, measured in mol/L), and Log P (the partition coefficient is a ratio of concentrations of nonionized compound between water and octanol). Compound 39 is the smallest and less lipophilic candidate compound with better-predicted water solubility among the selected candidate compounds (Table 8). Table 8 \| compound \| Mw \| log P \| Log S \| \| --- \| --- \| --- \| --- \| \| 37 \| 478.33 \| 5.76 \| –7.52 \| \| 39 \| 463.32 \| 4.85 \| –6.96 \| \| 40 \| 477.35 \| 5.08 \| –7.22 \| \| 41 \| 491.38 \| 5.32 \| –7.34 \| \| 42 \| 493.35 \| 5.28 \| –7.39 \| Therefore, we decided to use compound 39 in further experiments with other nuclear receptor assays and in humanized CAR mice. ## Novel CAR Ligands Interact with His203 and Occupy a Hydrophobic Pocket in Human wtCAR-LBD For the modeling analyses, we docked the compounds 37, 39, 40, and 48 in human wtCAR-LBD using CITCO as a reference (Figure 8A,B). Furthermore, to explore the wtCAR-LBD conformational dynamics and the interactions of these novel compounds, we conducted 25 μs of all-atom MD simulations (5 μs for each system plus CITCO). We studied the differences in the protein–ligand interactions among the systems, in comparison to CITCO. We observed the relevant role of the hydrogen bond interaction of H203 with the phenylimidazole ring in the novel compounds ranging from ∼65 to $90\%$. This interaction was observed in particular for compounds 39 and 40 for ∼65 and $90\%$, respectively (Figure 8C,D; Figures S-4–S-7). In addition to H203, T225 and D228 also have a relevant role in compound 39 stabilization. These interactions were formed between the amide of compound 39 and the T225 and D228 backbone oxygen, (located on H6) for ∼35 and $28\%$ of the simulation time, respectively (Figure 8C,D; Figure S-7). It is noteworthy that in the wtCAR/CITCO simulations, these additional polar interactions are not observed. Figure 8: (A) Overview of the CAR LBD structure (wtCAR as the reference structure) and the small-molecule ligand (39) used in this study. The regions of interest are highlighted as follows: H2′-H3 loop (residues 140–153), dark gray; H3 loop (residues 157–178), light blue; H5 (residues 196–209), pale green; β sheets (residues 217–223), pink; H10 (residues 308–333), light brown; H11 (residues 336–339), light orange; H12 (residues 341–348), dark brown. The rectangular area denotes the location of the ligand-binding pocket (LBP) and the residues forming the LBP. The main residues participating in ligand binding are depicted in the stick model with a transparent molecular surface. Residues are colored according to their respective regions (see left structure). (B) 2D structure of CITCO and compound 39. (C) Representative snapshots of LBP with CITCO and compound 39 are shown. The green dashed line represents the hydrogen bond. (D) Frequencies of protein–ligand hydrogen bonds and protein–ligand hydrophobic interactions in percent are shown on the right of the panel. (E) Close view of H3 and H12 zooming in K195 (on H4) and S348 (on H12). The green dashed line represents the hydrogen bond between K195 and S348. Distance between H3 and H12 (center of mass) is represented in the left box plot. Distance between K195 and S348 (oxygen atoms) is represented in the right box plot. The black line in each box represents the median value. (F) Hydrogen bond between the Y326 oxygen atom and N165 polar group is shown as the green dashed line. Color codes are the same as in panel A. Distance between N165 and Y326 (oxygen atoms) is represented in the right box plot. In addition to the hydrogen bond interactions, all novel compounds show high hydrophobic interaction frequency with F161 (∼$100\%$), the H203 imidazole ring (∼70 to $100\%$), and Y224 (90–$100\%$, except for the compound 40, which is around $20\%$), and lower interaction with C202, F234, Y326, and L242 (no interaction with compound 40) (Figure 8C,D; Figures S-7 and S-8). These interactions were similarly observed with CITCO. Also, some interactions are compound-specific such as I164 with compound 39, L206 with compounds 37 and 40, F217 with compounds 37, 40, and 48, and L239 with compounds 40 and 48. Overall, we observed that all the novel compounds adopted U-shaped conformations similar to CITCO within the wtCAR-LBD (Figure S-6). This conformation is mainly supported by hydrophobic interactions, with an exclusive interaction for compound 39 with I164, and extra T225 and D228 hydrogen bonds for compound 39, which stabilizes the compound within human wtCAR-LBD. ## H12 Positioned in Close Vicinity of H3 MDs revealed no direct interaction between CITCO and residues from H12. In this regard, we then proceeded to investigate the changes in geometry and dynamicity of this region relative to the LBD with novel compounds and CITCO. For this purpose, we calculated the distance between H12 and H3 (center of mass of each helix). The result showed that all novel compounds can stabilize the conformation of H12 in the close vicinity of H3 similar to CITCO (Figure 8E, Figure S-9A). This geometry is known to initiate receptor activation.15 It has been reported that H12 stays away from the pocket due to the barrier formed by hydrophobic residues in the LBD,44 where H11 directs the H12 in this active position.15 Previous studies also indicate that the free carboxylate of the H12 C-terminus interacts with the K195 side chain (on H4), leading to further H12 stabilization.15 To assess this phenomenon over the simulation time, we next calculated the distance between the carboxylate group of the H12 C-terminus and the polar group of K195 (Figure 8E; Figure S-9B). The median value for this distance in both wtCAR/CITCO and wtCAR/compounds 37, 39, and 40 stands around 3.1 Å, with a further distribution with compound 48. This geometry enables the hydrogen bond formation between the H3 and H12 regions, providing extra stability to the systems. Taken together, this supports our result in terms of the high binding affinity and potency of our novel compounds. ## Further Geometry Stabilization through N165–Y326 Interaction Along with the closeness of H12 and H3, and the interaction between the H4 and H12 C-terminus, the stabilization of the systems comes through the hydrogen bond interaction between N165 (H3) and Y326 (H10). Both CITCO and 39 show relatively similar rigidity in this region (Figure 8F). The same trends are also observed with other novel compounds (Figure S-8C) with further distribution in the presence of compounds 40 and 48. Although this interaction has been previously observed in the crystal structure with CITCO,15 MD data indicates that it is also relevant for our novel compounds. Taken together, our docking data followed by microsecond timescale all-atom MD simulations revealed that CITCO and compound 39 interact with wtCAR-LBD mainly by hydrophobic contacts and that stronger polar contacts were formed between compound 39 and wtCAR-LBD compared to CITCO due to hydrogen bond interactions between comp. 39 amide moiety and T225 and D228 backbone oxygen. Interestingly, previous findings report that no specific hydrogen bonds are required for CITCO stability inside the CAR.18,45 Analyses of the MD trajectories showed that the interaction between compound 39 and I164 besides the higher interaction frequency with Y326 (hydrophobic interaction) compared to that of the CITCO (Figure 8D) could highlight the critical role of H3 and H10 in protein stabilization. Of note, H10 lies on the heterodimerization interface where RXRα binds to the CAR. Our MD data also revealed that the H12 region is ordered and stable upon ligand binding. This event has been earlier reported as a driving force for CAR constitutive activity15 and, therefore, supports the agonistic effect of compound 39. ## Selectivity of Compound 39 to Other Nuclear Receptors Next, we sought to determine whether compound 39 is selective to the human CAR and whether it activates other nuclear receptors, for which a set of luciferase reporter assays was employed. We confirmed the selectivity of compound 39 for CAR as with no other nuclear receptor or the transcription factor aryl hydrocarbon receptor (AhR) was significantly activated by the compound at 10 μM concentration (Figure 9). Figure 9: Luciferase reporter assays for human nuclear receptors LXRα, LXRβ, TH, FXR, GR, PPARα, PPARδ/β, PPARγ, VDR, AR, Erα, and ERβ and for the AhR transcription factor were used to confirm the selectivity of compound 39. Specific ligands (GW3965, thyroxin, obeticholic acid, dexamethasone, fenofibrate, GW501516, rosiglitazone, 3-methylcholantrene, calcitriol, testosterone, and estradiol) have been used in various luciferase reporter assays. Compound 1 (CITCO) and compound 39 have been tested at 10 μM in HepG2 cells treated for 24 h. ## Microsomal Stability Experiments and Pilot Animal Pharmacokinetic Study In the following experiments, we evaluated both the plasma and microsomal stability of compound 39 HCl in human plasma, human liver microsomes, as well as liver fraction S9 in time intervals of up to 120 min (Figure 10A,B; Table S-5). We found that compound 39 is highly stable in human plasma (t$\frac{1}{2}$ ≥ 240 min). However, we observed a significant decline of compound 39 concentration in human microsomes as well as fraction S9 (t$\frac{1}{2}$ = 38.04 min and t$\frac{1}{2}$ = 42.4 min, respectively) (Figure 10B; Table S-5). Figure 10: Plasma and microsomal stability experiments and single-dose pharmacokinetics in C57BL/6N mice. The stability of compound 39 in human plasma (A) and human microsomes with S9 fraction (B) were analyzed after 2 h of treatment. (C) Pharmacokinetics (PK) after single-dose application of compound 39 as hydrochloric salt either via i.v. or peroral application (10 mg/kg, n = 4) were analyzed in mice over 480 min. (D) Metabolites M1 (comp. 41), M2 (comp. 40), and M3 (comp. 34, 2-(4-chlorophenyl)-3-(1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine) of compound 39 were observed after i.v. application. Samples have been analyzed using HPLC-MS/MS. (E) Inhibition of CYP3A4, CYP2B6, and CYP1A2 enzymes in microsomes. Compound 39 was tested in the concentration range from 0.1 nM up to 30 μM. Relative activity data were fitted, and dose–response curves were used to obtain IC50. We also evaluated the plasma protein binding of compound 39 in both human and mouse plasma, determining that $98\%$ of compound 39 is bound to human plasma proteins (Table S-6). We observed very similar properties of compound 39 in mouse plasma and mouse hepatic microsomes (Tables S-5–S-7). In a pilot single-dose pharmacokinetic study, we found fast absorption of compound 39 HCl hydrochloric salt after p.o. application in gavage, although the compound was rapidly eliminated from the plasma (Figure 10C; Table S-9). Significantly, we detected traces of metabolites for compound 39 after i.v. application in plasma. Metabolites M1 and M2 represent compounds 41 and 40. Both compounds are N-methylated derivatives of compound 39 with significant CAR activity. Minor metabolite M3 (compound 34) is 2-(4-chlorophenyl)-3-(1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine, indicating that hepatic metabolic enzymes may attack the methylene bridge between the heterocycle and phenyl rings (Figure 10D). The metabolite is inactive with respect to the CAR activation, and the metabolite was not observed in human liver microsomes with the S9 fraction (data not shown). These data suggest that compound 39 is the main active compound, and it is likely eliminated intact as the parent compound. Nevertheless, further detailed pharmacokinetic studies should focus on the distribution, biliary elimination, and phase II metabolic clearance of the compound. In addition, we conducted another examination to determine whether compound 39 inhibits the activities of major human cytochrome P450 enzymes. We found that compound 39 has a minor effect on major cytochrome P450 enzymes. Compound 39 inhibits enzymatic activities of CYP3A4 (with IC50 = 16.08 μM) and CYP1A2 (IC50 = 21.07 μM) in higher micromolar concentrations, but the compound has no activity on the CYP2B6 enzyme up to 30 μM concentration (Figure 10E). ## Effects of Compound 39 in CAR Humanized Mice Next, we treated humanized PXR/CAR/CYP3A mice with compound 39 to study the regulation of CAR target genes after a single i.p. application. We found that compound 39 significantly upregulates Cyp2b10 mRNA and protein, and human CYP3A4 mRNA in the humanized model, but significantly decreases the expression of genes Scd1 and G6pc after a single dose of 1 mg/kg. The latter genes are critically involved in triglyceride synthesis and gluconeogenesis in the liver. CITCO appeared more potent to induce Cyp2b10 mRNA but less potent to upregulate CYP3A4 mRNA expression, confirming the high efficiency of compound 39 to regulate the key CAR targets genes in murine hepatocytes. We also observed the trend of a decrease of Srebp1 and Fasn mRNA expression after compound 39 application (1 mg/kg) (Figure 11A), which agrees with data observed with the mouse CAR ligand TCPOBOP. This suggests that the human CAR ligand 39 recapitulates the significant effect of the murine ligand TCPOBOP on the regulation of lipid metabolism.4,5 Figure 11: In vivo effects of compound 39 on liver CAR target genes involved in the intermediary metabolism of glucose, lipids and bile acids, hepatocyte proliferation, and apoptosis in humanized PXR/CAR/CYP3A mice (n = 4) after single i.p. application of the dose 1 or 10 mg/kg. Mice were sacrificed 36 h after application; livers were subjected to RT-qPCR analysis (A,B), western blotting analysis with anti-Cyp2b10 antibody, or were weighted (C). Blood samples were analyzed for biochemical parameters (D). p < 0.05-significant effect vs control (vehicle-treated) mice.* We did not observe upregulation of the genes involved in rodent liver proliferation after CAR activation and liver weight gain in the experiments (Figure 11B,C). Nevertheless, long-term studies are needed to examine liver hypertrophy and hyperplasia after repeated treatment with compound 39. In analyzing blood biochemistry data after the single-dose application of compound 39 (dose 10 mg/kg), we observed a statistically significant decrease in plasma low-density lipoprotein (LDL) levels. This is consistent with results found with the mouse CAR ligand TCPOBOP in wild-type mice, indicating a positive effect of CAR activation on LDL plasma levels.9 We also observed a decrease in bile acid and total bilirubin (bilirubin-T) plasma levels after the application of compound 39, although these effects were not statistically significant. Neither glucose, plasma triglycerides (TG), HDL lipoproteins, nor liver injury biomarkers (AST, ALT, and LDH) was significantly affected by compound 39 after the single-dose application (Figure 11D). These results of the pilot single-dose pharmacokinetic study suggest that compound 39 is a novel effective human CAR agonist in animal experiments, a finding which warrants further repeated-dose long-term proof-of-concept studies. ## Toxicity Studies of Compound 39 We observed no cytotoxicity in HepG2, COS-1 (Table S-3), HepaRG, HepaRG KO CAR, or in the PHHs after 48 h treatment (data not presented). Furthermore, in the Repeated Dose 7 day Oral Toxicity Study in Rodents (EMA/CPMP/ICH/$\frac{286}{1995}$, 2009 guidelines), no significant signs of toxicity were observed after the 7 days of oral administration of compound 39 HCl into rats. In particular, no significant changes in body weight, changes in behavior, gross pathology, hematology, and biochemistry parameters were observed after the 7 days of oral administration of the compound 39 HCl in all groups (groups with 1, 10, and 30 mg/kg b.w.) when compared to the control group. We also tested the cardiotoxicity of compound 39 in a modified hERG fluorescence polarization assay. We did not observe any binding of compound 39 to hERG up to 20 μM (Supporting Information, Chapter 10). Finally, we did not observe any frame-shift or base-pair substitution mutagenicity of compound 39 in a modified Ames fluctuation assay performed on Salmonella typhimuriumTA100 and TA98 strains at a concentration of 1 and 10 μM (Table S-10). ## Conclusions Attempts to delineate the therapeutic implications of the CAR in humans have been hindered by the significant overlap in the pharmacology of human CAR and PXR receptors and the lack of a highly selective and potent human CAR agonist with suitable ADME properties. In this work, we used a rational design of novel selective human CAR agonists. We applied a bioisosteric approach to the central part of the hit molecules 2 and 3 to prepare new ligands for this human nuclear receptor. We were thus able to design a series of novel compounds that differed significantly in both nominal activities as CAR agonists and selectivity toward the PXR receptor as well as enhanced stability in comparison with the model compound CITCO. Based on our results, we performed a careful multiparametric selection of suitable candidates for further pharmacodynamic and pharmacokinetic studies. We found that the imidazo[1,2-a]pyridine core with the 1,2,3-triazole linker can be used for the further design of specific human CAR ligands. Replacement of the flexible oxime linker of CITCO with the triazole ring offered stability, less flexibility, and good accessibility via an undemanding click reaction. Modification of the 3,4-dichlorphenyl moiety of the hit compound 3 with amides (analogues 39–42) resulted in CAR ligands without agonistic activities to PXR (Scheme 6, Table 7, and Figure 6). Although extremely potent CAR agonists emerged in the resulting library of compounds, we also had to consider their metabolic stability and activity toward PXR. As a result, we decided to use compound 39 for further experiments, which, although not among our most potent CAR agonists, exhibited a desirable CAR/PXR profile and reasonable metabolic stability, allowing subsequent in vivo experiments. Using this chemical tool, which we have shown to have no observable toxicity or genotoxic potential, we were able to prove that compound 39 significantly activates the human CAR, both in vitro in human hepatocyte models and CAR humanized mice. Significantly, we noted that compound 39 regulates typical CAR target genes involved in xenobiotic (Cyp2b10), lipid (Scd1), or glucose (G6pc) metabolism, and it decreases plasma LDL lipoproteins even after a single dose in humanized PXR/CAR/CYP3A$\frac{4}{3}$A7 mice. In summary, our work identifies a selective CAR receptor agonist for which we have demonstrated both in vitro and in vivo activities in relevant models for the human CAR. Compound 39 thus warrants further preclinical studies in humanized CAR models or human hepatocyte models to better understand the unique function of the human CAR without confounding off-target effects on PXR receptor activation. ## Synthesis of Novel Ligands General chemical procedures: NMR spectra were measured on a Bruker AVANCE II-600 and/or Bruker AVANCE II-500 instruments (600.1 or 500.0 MHz for 1H and 150.9 or 125.7 MHz for 13C) in hexadeuterodimethyl sulfoxide and referenced to the solvent signal (δ 2.50 and 39.70, respectively). Mass spectra were measured on a LTQ Orbitrap XL (Thermo Fischer Scientific) using electrospray ionization (ESI) and a GCT Premier (Waters) using EI. The elemental analyses were obtained on a Perkin Elmer CHN Analyzer 2400, Series II Sys (PerkinElmer), and X-ray fluorescence spectrometer SPECTRO iQ II (SPECTRO Analytical Instruments, Germany). Column chromatography and thin-layer chromatography (TLC) were performed using Silica gel 60 (Fluka) and Silufol Silica gel 60 F254 foils (Merck), respectively. The purity of newly synthesized compounds was >$95\%$, confirmed by UPLC-MS. Solvents were evaporated at 2 kPa and a bath temperature of 30–60 °C. The compounds were dried at 13 Pa and 50 °C. ## General Procedure I: Cyclization of Heterocycle 2-Aminothiazole [3] or 2-aminopyridine [4] was dissolved in EtOH, and substituted or unsubstituted bromoacetophenone derivative (1 equiv) was added, followed by the addition of NaHCO3 (1 equiv). The reaction mixture was heated at 70 °C overnight. After the completion of the reaction (monitored by TLC or UPLC), the solvent was evaporated to a minimal volume, and the residue was diluted with EtOAc and washed with water. The water phase was extracted twice more with EtOAc, and the combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography (eluent petrol ether/EtOAc or EtOAc/MeOH). ## General Procedure II: Iodination 2-Substituted imidazo[1,2-a]pyridines or imidazo[2,1-b]thiazoles were dissolved in CH3CN (5 mL/mmol) and NIS (1.05 equiv) was added in one portion. The suspension was stirred at 25 °C, and the conversion was monitored by TLC. After the completion of the reaction (1–4 h), the reaction mixture was diluted with EtOAc and washed with a saturated Na2S2O3 solution. The inorganic phase was extracted twice more with EtOAc; combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography, with the mobile phase petrol ether/EtOAc (10:$50\%$). ## General Procedure III: Sonogashira Coupling 3-Iodoimidazo[1,2-a]pyridines or 5-iodoimidazo[2,1-b]thiazoles were placed in a dried round-bottom flask, diluted with dry DMF, degassed at 0 °C, and flushed with argon. CuI (10 mol %) and Pd(PPh3)2Cl2 (5 mol %) were added, the mixture was properly degassed, dry TEA (3 equiv) was added, and the mixture was degassed again. Finally, TMS-acetylene (5 equiv) was added in one portion. The reaction mixture was stirred at 25 °C under an argon atmosphere. After the completion of the reaction (monitored by TLC), the mixture was diluted if necessary with CHCl3 and filtered over Celite. The filtrate was washed with water; the water phase was extracted twice more with CHCl3; the combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography (mobile phase petrol ether/EtOAc). ## General Procedure IV: Click Reaction Trimethylsilyl(ethynyl)imidazo[2,1-b]thiazole or trimethylsilyl(ethynyl)imidazo[1,2-a]pyridine derivatives were dissolved in THF/H2O mixture (1:1) and an appropriate azido intermediate (1 equiv) was added. The reaction mixture was degassed at 0 °C, refilled with argon, and CuSO4·5H2O (10 mol %), KF (1 equiv), were Na-ascorbate (1 equiv) were added in one portion. The reaction mixture was stirred at 25 °C and monitored by TLC. After the completion of the reaction, the mixture was diluted with EtOAc and washed with water. The water phase was extracted twice more with EtOAc; combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography (eluent petrol ether/EtOAc or EtOAc/MeOH). ## General Procedure V: Ester Hydrolysis The methyl or ethyl ester derivative was dissolved in THF/H2O 2:1, and LiOH·H2O (4 equiv) was added in one portion. The reaction mixture was stirred at 25 °C and monitored by TLC. After the completion of the reaction, the mixture was extracted with EtOAc, and the water phase was acidified to pH 2 and extracted again with EtOAc. The organic phase was dried over sodium sulfate and purified by reverse-phase flash column chromatography. ## General Procedure VI: Amide Preparation 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-benzoic acid [38] was placed in a round-bottom flask, and toluene was added (5 mL) followed by the addition of thionyl chloride (0.5 mL, in excess). The reaction mixture was stirred at 90 °C overnight. The reaction mixture was evaporated to dryness, coevaporated with toluene, and used directly for the next step without any purification. The intermediate 2-chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)benzoyl chloride [53] was dissolved in dry DCM and cooled in an ice bath. An appropriate amine (1.2 equiv) was added, followed by the addition of DIPEA (1.5 or 2 equiv in case of amine salts). The reaction mixture was stirred at 25 °C and monitored by TLC or LCMS. After completion of the reaction, the mixture was diluted with DCM, washed with water, and purified by reverse-phase flash CC or flash column chromatography. ## General Procedure VII: Ketone Preparation Using the Grignard Reagent 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-N-methoxy-N-methylbenzamide was dissolved in dry THF (6 mL), cooled in an ice bath, degassed, and refilled with argon. The Grignard reagent (3 M in DEE, 2 equiv) was added in one portion, and the mixture was allowed to warm to 25 °C and stirred overnight. For the LAH reduction in case of an aldehyde, the mixture was stirred at 5 °C for 2 h. The mixture was quenched with a saturated NH4Cl solution and extracted with EtOAc. Combined organic phases were dried over sodium sulfate, evaporated, and purified if necessary by flash column chromatography (petrol ether/EtOAc 70:$100\%$). ## 6-(4-Chlorophenyl)-5-(1-(3,4-dichlorobenzyl)-1H-1,2,3-triazol-4-yl)imidazo[2,1-b]thiazole (2) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 268 mg ($90\%$). 1H NMR (500 MHz, DMSO-d6): δ 8.49 (s, 1H), 8.04 (d, $J = 4.5$ Hz, 1H), 7.67–7.70 (m, 4H), 7.42–7.45 (m, 2H), 7.39 (d, $J = 4.5$ Hz, 1H), 7.34 (ddm, $J = 2.1$ Hz, $J = 8.3$ Hz, 1H), 5.69 (br s, 2H). 13C NMR (101 MHz, DMSO): δ 149.63, 142.82, 137.13, 136.99, 133.30, 132.39, 131.49, 131.24, 131.19, 130.42, 129.30, 128.73, 128.66, 123.62, 120.02, 114.52, 113.92, 51.85. HRMS: calcd for [M + H], 459.99518; found, 459.99522. ## 2-(4-Chlorophenyl)-3-(1-(3,4-dichlorobenzyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (3) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: Yield: 130 mg ($57\%$). 1H NMR (500 MHz, DMSO-d6): δ 8.53 (s, 1H), 8.48 (dt, $J = 1.1$ Hz, $J = 6.9$ Hz, 1H), 7.67–7.71 (m, 5H), 7.41–7.44 (m, 2H), 7.39 (ddd, $J = 1.2$ Hz, $J = 6.8$ Hz, $J = 9.1$ Hz, 1H), 7.35 (dd, $J = 2.1$ Hz, $J = 8.4$ Hz, 1H), 7.00 (td, $J = 1.2$ Hz, $J = 6.8$ Hz, 1H), 5.75 (s, 2H). 13C NMR (101 MHz, CDCl3): δ 144.95, 142.52, 137.05, 136.40, 133.04, 132.86, 131.53, 131.28, 131.21, 130.34, 129.67, 128.71, 128.59, 126.30, 125.95, 125.45, 117.13, 113.38, 111.57, 51.94. HRMS: calcd for [M + H], 454.03876; found, 454.03886. ## 6-(4-Chlorophenyl)imidazo[2,1-b]thiazole (6a) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$60\%$). Yield: 981 mg ($82\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.26 (s, 1H), 7.94 (d, $J = 4.4$ Hz, 1H), 7.90–7.80 (m, 2H), 7.48–7.39 (m, 2H), 7.28 (d, $J = 4.4$ Hz, 1H). 13C NMR (101 MHz, DMSO): δ 149.88, 145.54, 133.64, 131.78, 129.09, 126.85, 120.52, 113.85, 110.30. EI MS: calcd for [M + H], 234.0018; found, 234.0020. ## 6-(3,4-Dichlorophenyl)imidazo[2,1-b]thiazole (6g) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 878 mg ($65\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.37 (s, 1H), 8.06 (d, $J = 2.0$ Hz, 1H), 7.97 (d, $J = 4.5$ Hz, 1H), 7.82 (dd, $J = 8.4$, 2.1 Hz, 1H), 7.64 (d, $J = 8.4$ Hz, 1H), 7.30 (d, $J = 4.5$ Hz, 1H). 13C NMR (101 MHz, DMSO): δ 149.83, 143.93, 135.18, 131.65, 131.04, 129.21, 126.38, 124.93, 120.26, 113.98, 110.90. HRMS: calcd for [M + Na], 268.97015; found, 268.97029. ## 2-(4-Chlorophenyl)imidazo[1,2-a]pyridine (7a) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$60\%$). Yield: 1.49 g ($88\%$). 1H NMR (401 MHz, chloroform-d): δ 8.12 (dt, $J = 1.2$ Hz, $J = 6.8$ Hz, 1H), 7.93–7.87 (m, 2H), 7.84 (d, $J = 0.7$ Hz, 1H), 7.64 (dq, $J = 1.0$ Hz, $J = 9.2$ Hz, 1H), 7.44–7.40 (m, 2H), 7.20 (ddd, $J = 1.3$ Hz, $J = 6.8$ Hz, $J = 9.2$ Hz, 1H), 6.80 (td, $J = 1.2$ Hz, $J = 6.8$ Hz, 1H). 13C NMR (101 MHz, CDCl3): δ 145.86, 144.82, 133.79, 132.45, 129.02, 127.38, 125.73, 125.03, 117.69, 112.72, 108.31. EI MS: calcd for [M + H], 228.0454; found, 228.0456. ## 2-(p-Tolyl)imidazo[1,2-a]pyridine (7b) The title compound was prepared according to the described procedure. The identity and purity were confirmed by NMR and HRMS. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 853 mg ($76\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.50 (dt, $J = 6.8$, 1.2 Hz, 1H), 8.33 (d, $J = 0.7$ Hz, 1H), 7.88–7.82 (m, 2H), 7.56 (dq, $J = 9.1$, 1.0 Hz, 1H), 7.27–7.19 (m, 3H), 6.87 (td, $J = 6.7$, 1.2 Hz, 1H), 2.33 (s, 3H). 13C NMR (101 MHz, DMSO): δ 145.21, 144.96, 137.43, 131.62, 129.75, 127.24, 125.97, 125.23, 116.99, 112.61, 109.10, 21.33. HRMS: calcd for [M + H], 209.10732; found, 209.10745. ## 2-(4-Ethylphenyl)imidazo[1,2-a]pyridine (7c) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 1.02 g ($74\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.51 (dt, $J = 6.8$, 1.2 Hz, 1H), 8.34 (d, $J = 0.7$ Hz, 1H), 7.90–7.85 (m, 2H), 7.56 (dq, $J = 9.1$, 1.0 Hz, 1H), 7.29–7.26 (m, 2H), 7.23 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 6.88 (td, $J = 6.7$, 1.2 Hz, 1H), 2.63 (q, $J = 7.6$ Hz, 2H), 1.21 (t, $J = 7.6$ Hz, 3H). 13C NMR (101 MHz, DMSO): δ 145.21, 144.97, 143.79, 131.88, 128.56, 127.25, 126.04, 125.24, 117.00, 112.63, 109.14, 28.44, 16.01. HRMS: calcd for [M + H], 223.12298; found, 223.12301. ## 2-(4-Fluorophenyl)imidazo[1,2-a]pyridine (7d) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 988 mg ($88\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.52 (dt, $J = 1.2$, 7.0 Hz, 1H), 8.38 (s, 1H), 8.04–7.96 (m, 2H), 7.57 (d, $J = 9.1$ Hz, 1H), 7.31–7.21 (m, 3H), 6.89 (td, $J = 1.3$, 6.8 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 145.30, 143.91, 130.93, 127.37, 125.51, 117.06, 112.80, 109.41. 13C NMR (101 MHz, DMSO-d6): δ 162.29 (d, $J = 244.2$ Hz), 130.93, 127.97 (d, $J = 8.2$ Hz), 116.05 (d, $J = 21.5$ Hz). HRMS: calcd for [M + H], 213.08225; found, 213.08226. ## 2-(4-(Trifluoromethyl)phenyl)imidazo[1,2-a]pyridine (7e) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 1.07 g ($77\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.57–8.53 (m, 1H), 8.21–8.15 (m, 1H), 7.82–7.77 (m, 1H), 7.61 (dq, $J = 9.1$, 1.0 Hz, 0H), 7.29 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 6.93 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 145.20, 142.84, 138.12, 128.06, 127.75, 127.30, 126.21, 125.89, 125.86, 125.82, 125.78, 125.71, 117.04, 112.84, 110.73. HRMS: calcd for [M + H], 263.07906; found, 263.07907. ## 2-(2,4-Dichlorophenyl)imidazo[1,2-a]pyridine (7f) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 634 mg ($76\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.64 (d, $J = 0.7$ Hz, 1H), 8.61 (dt, $J = 6.8$, 1.2 Hz, 1H), 8.29 (d, $J = 8.5$ Hz, 1H), 7.71 (d, $J = 2.2$ Hz, 1H), 7.60 (dq, $J = 9.1$, 1.0 Hz, 1H), 7.54 (dd, $J = 8.6$, 2.2 Hz, 1H), 7.30 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 6.94 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 144.26, 139.80, 132.92, 132.22, 131.75, 131.67, 130.15, 128.08, 127.69, 126.21, 117.15, 113.42, 113.01. HRMS: calcd for [M + H], 263.01373; found, 263.01390. ## 2-(3,4-Dichlorophenyl)imidazo[1,2-a]pyridine (7g) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 533 mg ($87\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.54–8.49 (m, 2H), 8.18 (d, $J = 2.0$ Hz, 1H), 7.93 (dd, $J = 8.4$, 2.0 Hz, 1H), 7.68 (d, $J = 8.4$ Hz, 1H), 7.58 (dq, $J = 9.1$, 1.0 Hz, 1H), 7.27 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 6.91 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 145.10, 141.99, 134.88, 131.74, 131.13, 130.02, 127.25, 127.23, 125.73, 125.70, 116.94, 112.84, 110.47. HRMS: calcd for [M + H], 263.01373; found, 263.01393. ## 4-(Imidazo[1,2-a]pyridin-2-yl)benzonitrile (7h) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 952 mg ($82\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.56 (s, 1H), 8.54 (dt, $J = 6.8$, 1.2 Hz, 1H), 8.18–8.12 (m, 2H), 7.91–7.85 (m, 2H), 7.60 (dq, $J = 9.1$, 1.0 Hz, 1H), 7.28 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 6.92 (td, $J = 6.7$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 145.40, 142.63, 138.68, 133.03, 127.44, 126.39, 126.10, 119.33, 117.14, 113.13, 111.41, 110.06. HRMS: calcd for [M + H], 220.08692; found, 220.08686. ## 2-(4-Methoxyphenyl)imidazo[1,2-a]pyridine (7i) The title compound was prepared according to General Procedure I. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 200 mg ($84\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.49 (dt, $J = 6.8$, 1.2 Hz, 1H), 7.92–7.87 (m, 2H), 7.54 (dq, $J = 9.1$, 1.0 Hz, 1H), 7.21 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 7.03–6.98 (m, 2H), 6.86 (td, $J = 6.7$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 159.48, 145.19, 144.90, 127.33, 127.16, 127.00, 125.09, 116.86, 114.59, 112.51, 108.48, 55.60. HRMS: calcd for [M + H], 240.07675; found, 240.07674. ## 6-(4-Chlorophenyl)-5-iodoimidazo[2,1-b]thiazole (8a) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 421 mg ($89\%$). 1H NMR (401 MHz, DMSO-d6): δ 7.98 (m, 2H), 7.87 (d, $J = 4.5$ Hz, 1H), 7.52 (m, 2H), 7.42 (d, $J = 4.5$ Hz, 1H). 13C NMR (101 MHz, DMSO): δ 149.88, 145.54, 133.64, 131.78, 129.09, 126.85, 120.52, 113.85, 110.30. HRMS: calcd for [M + H], 360.90577; found, 360.90586. ## 6-(3,4-Dichlorophenyl)-5-iodoimidazo[2,1-b]thiazole (8g) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 514 mg ($96\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.17 (d, $J = 2.1$ Hz, 1H), 7.99 (dd, $J = 8.5$, 2.1 Hz, 1H), 7.90 (d, $J = 4.5$ Hz, 1H), 7.75 (d, $J = 8.5$ Hz, 1H), 7.46 (d, $J = 4.5$ Hz, 1H). 13C NMR (101 MHz, DMSO): δ 150.98, 145.37, 134.98, 131.61, 131.21, 130.34, 128.53, 126.97, 120.67, 114.97, 60.96. HRMS: calcd for [M + H], 394.86679; found, 394.86699. ## 2-(4-Chlorophenyl)-3-iodoimidazo[1,2-a]pyridine (9a) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 352 mg ($98\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.42 (dt, $J = 6.9$, 1.2 Hz, 1H), 8.11–8.06 (m, 2H), 7.63 (dt, $J = 9.0$, 1.2 Hz, 1H), 7.60–7.55 (m, 2H), 7.38 (ddd, $J = 9.0$, 6.8, 1.3 Hz, 1H), 7.09 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, CDCl3): δ 147.52, 145.52, 132.99, 132.85, 129.73, 128.67, 127.35, 126.47, 117.16, 113.93, 63.82. HRMS: calcd for [M + H], 354.94935; found, 354.94944. ## 3-Iodo-2-(p-tolyl)imidazo[1,2-a]pyridine (9b) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 501 mg ($94\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.41 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.98–7.92 (m, 2H), 7.61 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.36 (ddd, $J = 9.0$, 6.7, 1.3 Hz, 1H), 7.31 (d, $J = 8.0$ Hz, 2H), 7.06 (td, $J = 6.8$, 1.2 Hz, 1H), 2.36 (s, 3H). 13C NMR (101 MHz, DMSO): δ 147.70, 147.12, 137.95, 131.41, 129.41, 128.32, 127.44, 126.35, 117.29, 113.92, 63.23, 21.36. HRMS: calcd for [M + H], 335.00397; found, 335.00397. ## 2-(4-Ethylphenyl)-3-iodoimidazo[1,2-a]pyridine (9c) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 1.23 g ($86\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.41 (dt, $J = 6.9$, 1.1 Hz, 1H), 8.00–7.95 (m, 2H), 7.62 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.39–7.36 (m, 3H), 7.35–7.31 (m, 2H), 7.07 (td, $J = 6.8$, 1.2 Hz, 1H), 2.66 (q, $J = 7.6$ Hz, 2H), 1.22 (t, $J = 7.6$ Hz, 3H). 13C NMR (101 MHz, DMSO): δ 147.41, 146.83, 143.95, 131.35, 128.10, 127.93, 127.16, 126.08, 117.00, 113.65, 62.95, 28.15, 15.67. HRMS: calcd for [M + H], 349.01962; found, 349.01982. ## 2-(4-Fluorophenyl)-3-iodoimidazo[1,2-a]pyridine (9d) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 1.42 g ($97\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.40 (dt, $J = 6.9$, 1.1 Hz, 1H), 8.12–8.06 (m, 2H), 7.62 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.40–7.30 (m, 3H), 7.07 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 162.15 (d, $J = 245.5$ Hz), 147.44, 145.88, 130.42 (d, $J = 3.0$ Hz), 130.15 (d, $J = 8.3$ Hz), 127.25, 126.28, 117.05, 115.48 (d, $J = 21.5$ Hz), 113.78, 63.34. HRMS: calcd for [M + H], 338.97890; found, 338.97902. ## 3-Iodo-2-(4-(trifluoromethyl)phenyl)imidazo[1,2-a]pyridine (9e) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 1.25 g ($92\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.46 (d, $J = 6.9$ Hz, 1H), 8.31 (d, $J = 8.0$ Hz, 2H), 7.88 (d, $J = 8.1$ Hz, 2H), 7.67 (d, $J = 9.0$ Hz, 1H), 7.42 (dd, $J = 8.9$, 6.9 Hz, 1H), 7.12 (t, $J = 6.8$ Hz, 1H). 13C NMR (101 MHz, DMSO): δ 147.90, 145.35, 138.31, 128.87, 128.65 127.74, 126.96, 125.81 ($J = 3.78$ Hz), 124.77, 117.61, 114.38, 65.13. HRMS: calcd for [M + H], 388.97570; found, 388.97582. ## 2-(2,4-Dichlorophenyl)-3-iodoimidazo[1,2-a]pyridine (9f) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 238 mg ($89\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.40 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.79 (d, $J = 2.0$ Hz, 1H), 7.64 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.59–7.51 (m, 2H), 7.40 (ddd, $J = 9.0$, 6.8, 1.3 Hz, 1H), 7.12 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 147.39, 146.72, 134.57, 134.42, 134.34, 132.87, 129.62, 127.72, 127.46, 126.52, 117.62, 114.32, 67.98. HRMS: calcd for [M + H], 388.91037; found, 388.91071. ## 2-(3,4-Dichlorophenyl)-3-iodoimidazo[1,2-a]pyridine (9g) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 638 mg ($98\%$). 1H NMR (401 MHz, chloroform-d): δ 8.24 (dt, $J = 6.9$, 1.1 Hz, 1H), 8.22 (d, $J = 2.1$ Hz, 1H), 7.96 (dd, $J = 8.4$, 2.1 Hz, 1H), 7.66 (dt, $J = 9.0$, 1.2 Hz, 1H), 7.55 (d, $J = 8.4$ Hz, 1H), 7.33 (ddd, $J = 9.1$, 6.8, 1.3 Hz, 1H), 6.99 (td, $J = 6.9$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 179.90, 147.82, 144.25, 134.93, 131.66, 131.24, 131.09, 129.70, 128.14, 127.74, 127.08, 117.55, 114.43, 64.93. HRMS: calcd for [M + H], 388.91037; found, 388.91052. ## 4-(3-Iodoimidazo[1,2-a]pyridin-2-yl)benzonitrile (9h) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 245 mg ($90\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.46 (dt, $J = 7.0$, 1.1 Hz, 1H), 8.31–8.26 (m, 2H), 8.00–7.95 (m, 2H), 7.66 (dt, $J = 9.1$, 1.1 Hz, 1H), 7.41 (ddd, $J = 9.1$, 6.8, 1.2 Hz, 1H), 7.11 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 147.64, 144.64, 138.54, 132.60, 128.49, 127.49, 126.85, 119.02, 117.37, 114.20, 110.54, 65.39. HRMS: calcd for [M + H], 345.98357; found, 345.98370. ## 3-Iodo-2-(4-methoxyphenyl)imidazo[1,2-a]pyridine (9i) The title compound was prepared according to General Procedure II. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 238 mg ($89\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.39 (dt, $J = 6.9$, 1.1 Hz, 1H), 8.05–7.96 (m, 2H), 7.60 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.35 (ddd, $J = 9.0$, 6.7, 1.2 Hz, 1H), 7.09–7.05 (m, 3H), 3.82 (s, 3H). 13C NMR (101 MHz, DMSO): δ 159.68, 147.67, 146.99, 129.70, 127.38, 126.59, 126.27, 117.16, 114.27, 113.83, 62.62, 55.66. HRMS: calcd for [M + H], 350.99888; found, 350.99893. ## 6-(4-Chlorophenyl)-5-((trimethylsilyl)ethynyl)imidazo[2,1-b]thiazole (10a) The title compound was prepared according to General Procedure III (Scheme 1). Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 458 mg ($61\%$). 1H NMR (500 MHz, DMSO-d6): δ 8.10 (m, 2H), 7.92 (d, $J = 4.4$ Hz, 1H), 7.51–7.54 (m, 2H), 7.44 (d, $J = 4.4$ Hz, 1H), 0.31. 13C NMR (101 MHz, DMSO): δ 149.86, 145.03, 133.72, 132.31, 128.81, 127.46, 119.32, 115.60, 106.93, 105.19, 93.51, −0.17. EI MS: calcd for [M + H], 330.0414; found, 330.0416. ## 2-(4-Chlorophenyl)-3-((trimethylsilyl)ethynyl)imidazo[1,2-a]pyridine (11a) The title compound was prepared according to General Procedure III. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 625 mg ($56\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.41 (dt, $J = 6.8$, 1.2 Hz, 1H), 8.28–8.21 (m, 2H), 7.70 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.59–7.53 (m, 2H), 7.46 (ddd, $J = 9.0$, 6.8, 1.3 Hz, 1H), 7.14 (td, $J = 6.8$, 1.2 Hz, 1H), 0.34 (s, 9H). 13C NMR (101 MHz, DMSO): δ 146.17, 144.70, 133.43, 132.04, 128.88, 128.28, 127.82, 125.82, 117.30, 114.28, 108.95, 104.06, 93.27, −0.10. HRMS: calcd for [M + H], 325.09223; found, 325.09232. ## 2-(p-Tolyl)-3-((trimethylsilyl)ethynyl)imidazo[1,2-a]pyridine (11b) The title compound was prepared according to General Procedure III. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 117 mg ($64\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.39 (d, $J = 6.8$ Hz, 1H), 8.16 (d, $J = 7.8$ Hz, 2H), 7.69 (d, $J = 9.0$ Hz, 1H), 7.43 (dd, $J = 8.9$, 6.9 Hz, 1H), 7.30 (d, $J = 7.9$ Hz, 2H), 7.12 (t, $J = 6.8$ Hz, 1H), 2.36 (s, 3H), 0.34 (s, 9H). 13C NMR (101 MHz, DMSO): δ 147.99, 144.94, 138.80, 130.71, 129.64, 127.79, 126.95, 125.98, 117.43, 114.32, 108.68, 103.79, 94.13, 21.38, 0.24. HRMS: calcd for [M + H], 305.14685; found, 305.14690. ## 2-(4-Ethylphenyl)-3-((trimethylsilyl)ethynyl)imidazo[1,2-a]pyridine (11c) The title compound was prepared according to General Procedure III. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 126 mg ($63\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.39 (dt, $J = 6.8$, 1.2 Hz, 1H), 8.22–8.16 (m, 2H), 7.69 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.42 (ddd, $J = 9.0$, 6.8, 1.3 Hz, 1H), 7.35–7.29 (m, 2H), 7.11 (td, $J = 6.8$, 1.2 Hz, 1H), 2.65 (q, $J = 7.6$ Hz, 2H), 1.20 (t, $J = 7.6$ Hz, 3H), 0.34 (s, 9H). 13C NMR (101 MHz, DMSO): δ 147.66, 144.72, 144.65, 130.69, 128.12, 127.44, 126.73, 125.64, 117.14, 113.98, 108.33, 103.51, 93.84, 28.18, 15.57, −0.06. HRMS: calcd for [M + H], 319.16250; found, 319.16255. ## 2-(4-Fluorophenyl)-3-((trimethylsilyl)ethynyl)imidazo[1,2-a]pyridine (11d) The title compound was prepared according to General Procedure III. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 250 mg ($69\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.41 (dd, $J = 6.9$, 1.4 Hz, 1H), 8.32–8.25 (m, 2H), 7.70 (dd, $J = 9.1$, 1.4 Hz, 1H), 7.45 (ddt, $J = 9.0$, 6.8, 1.3 Hz, 1H), 7.38–7.31 (m, 2H), 7.13 (tt, $J = 6.9$, 1.3 Hz, 1H), 0.36–0.31 (m, 9H). 13C NMR (101 MHz, DMSO): δ 162.46 (d, $J = 246.2$ Hz), 146.53, 144.70, 129.74 (d, $J = 3.0$ Hz), 128.79 (d, $J = 8.4$ Hz), 127.74, 125.81, 117.24, 117.24, 115.81 (d, $J = 21.6$ Hz), 114.21, 114.21, 108.58, 103.68, 93.46, −0.05. HRMS: calcd for [M + H], 309.12178; found, 309.12195. ## 2-(4-(Trifluoromethyl)phenyl)-3-((trimethylsilyl)ethynyl)imidazo[1,2-a]pyridine (11e) The title compound was prepared according to General Procedure III. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 123 mg ($54\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.45 (tt, $J = 6.8$, 1.0 Hz, 3H), 7.91–7.82 (m, 2H), 7.74 (dt, $J = 9.1$, 1.1 Hz, 1H), 7.49 (ddd, $J = 9.1$, 6.8, 1.3 Hz, 1H), 7.17 (td, $J = 6.8$, 1.2 Hz, 1H), 0.37 (s, 9H). 13C NMR (101 MHz, DMSO): δ 145.80, 145.12, 137.36, 137.35, 129.25, 128.93, 128.38, 127.42, 126.23, 126.08 (q, $J = 3.9$ Hz), 117.80, 114.82, 109.61, 105.20, 93.24, 0.18. HRMS: calcd for [M + H], 359.11859; found, 359.11862. ## 2-(2,4-Dichlorophenyl)-3-((trimethylsilyl)ethynyl)imidazo[1,2-a]pyridine (11f) The title compound was prepared according to General Procedure III. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 98 mg ($64\%$). 1H NMR (401 MHz, chloroform-d): δ 8.30 (dt, $J = 6.8$, 1.2 Hz, 1H), 7.64 (m, 2H), 7.54 (d, $J = 2.1$ Hz, 1H), 7.38–7.27 (m, 2H), 6.97 (td, $J = 6.8$, 1.2 Hz, 1H), 0.27 (s, 9H). 13C NMR (101 MHz, CDCl3): δ 146.79, 144.80, 135.05, 134.32, 133.22, 131.39, 130.12, 126.95, 126.52, 125.55, 117.98, 113.42, 108.14, 92.17, 0.04. HRMS: calcd for [M + H], 321.14177; found, 321.14179. ## 2-(3,4-Dichlorophenyl)-3-((trimethylsilyl)ethynyl)imidazo[1,2-a]pyridine (11g) The title compound was prepared according to General Procedure III and used as crude in the next step without purification. ## 4-(3-((Trimethylsilyl)ethynyl)imidazo[1,2-a]pyridin-2-yl)benzonitrile (11h) The title compound was prepared according to General Procedure III. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 170 mg ($65\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.44 (dt, $J = 6.8$, 1.2 Hz, 1H), 8.42–8.38 (m, 2H), 7.99–7.94 (m, 2H), 7.74 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.49 (ddd, $J = 9.1$, 6.8, 1.3 Hz, 1H), 7.17 (td, $J = 6.8$, 1.1 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 145.16, 144.88, 137.51, 132.85, 128.24, 127.06, 125.97, 118.91, 117.56, 114.62, 110.97, 109.67, 105.24, 92.83, −0.14. HRMS: calcd for [M + H], 316.12645; found, 316.12656. ## 2-(4-Methoxyphenyl)-3-((trimethylsilyl)ethynyl)imidazo[1,2-a]pyridine (11i) The title compound was prepared according to General Procedure III. Mobile phase petrol ether/EtOAc (10:$50\%$). Yield: 98 mg ($64\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.38 (dt, $J = 6.8$, 1.2 Hz, 1H), 8.26–8.18 (m, 2H), 7.67 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.42 (ddd, $J = 9.0$, 6.8, 1.3 Hz, 1H), 7.11 (td, $J = 6.8$, 1.2 Hz, 1H), 7.08–7.03 (m, 2H), 3.82 (s, 3H), 0.34 (s, 9H). 13C NMR (101 MHz, DMSO): δ 159.83, 147.63, 144.60, 128.13, 127.19, 125.75, 125.49, 116.93, 114.07, 113.72, 108.09, 102.92, 94.03, 55.32, −0.04, −0.06, −0.08. HRMS: calcd for [M + H], 321.14177; found, 321.14179. ## 5-(1-(3,4-Dichlorobenzyl)-1H-1,2,3-triazol-4-yl)-6-(3,4-dichlorophenyl)imidazo[2,1-b]thiazole (12g) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (30:$60\%$). Yield: 51 mg ($92\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.57 (s, 1H), 8.01 (d, $J = 4.5$ Hz, 1H), 7.87 (d, $J = 2.0$ Hz, 1H), 7.71 (d, $J = 2.1$ Hz, 1H), 7.69–7.64 (m, 2H), 7.61 (d, $J = 8.4$ Hz, 1H), 7.40 (d, $J = 4.5$ Hz, 1H), 7.36 (dd, $J = 8.3$, 2.1 Hz, 1H), 5.70 (s, 2H). 13C NMR (101 MHz, DMSO): δ 149.77, 141.34, 136.86, 136.81, 135.00, 131.52, 131.39, 131.24, 131.21, 130.87, 130.44, 130.14, 128.90, 128.65, 127.44, 123.91, 119.85, 114.89, 114.40, 51.92. HRMS: calcd for [M + H], 493.95620; found, 493.95625. ## 3-(1-(3,4-Dichlorobenzyl)-1H-1,2,3-triazol-4-yl)-2-(p-tolyl)imidazo[1,2-a]pyridine (13b) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 140 mg ($84\%$).1H NMR (401 MHz, DMSO-d6): δ 8.47 (dd, $J = 7.1$, 1.5 Hz, 1H), 8.45 (d, $J = 1.5$ Hz, 0H), 7.72–7.65 (m, 1H), 7.56–7.51 (m, 1H), 7.39–7.33 (m, 1H), 7.16 (d, $J = 7.8$ Hz, 1H), 6.97 (td, $J = 6.9$, 1.7 Hz, 0H), 5.74 (s, 1H), 2.32 (s, 2H). 13C NMR (101 MHz, DMSO): δ 144.82, 143.87, 137.48, 137.11, 136.79, 131.51, 131.27, 131.23, 131.16, 130.25, 129.17, 128.54, 125.88, 125.74, 125.30, 116.96, 113.06, 110.91, 51.86, 21.01. HRMS: calcd for [M + H], 434.09338; found, 434.09355. ## 3-(1-(3,4-Dichlorobenzyl)-1H-1,2,3-triazol-4-yl)-2-(4-ethylphenyl)imidazo[1,2-a]pyridine (13c) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 350 mg ($70\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.48–8.43 (m, 2H), 7.72–7.65 (m, 3H), 7.59–7.54 (m, 2H), 7.38–7.33 (m, 2H), 7.20–7.15 (m, 2H), 6.97 (td, $J = 6.9$, 1.2 Hz, 1H), 5.74 (s, 2H), 2.61 (q, $J = 7.6$ Hz, 2H), 1.19 (t, $J = 7.6$ Hz, 3H). 13C NMR (101 MHz, DMSO): δ 144.85, 143.88, 143.75, 137.12, 136.79, 131.52, 131.24, 131.19, 130.26, 128.56, 127.95, 127.90, 125.90, 125.86, 125.28, 116.98, 113.09, 110.92, 51.91, 28.10, 15.55. HRMS: calcd for [M + H], 448.10903; found, 448.10914. ## 3-(1-(3,4-Dichlorobenzyl)-1H-1,2,3-triazol-4-yl)-2-(4-fluorophenyl)imidazo[1,2-a]pyridine (13d) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (30:$80\%$). Yield: 211 mg ($74\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.50 (s, 1H), 8.48 (dt, $J = 7.0$, 1.2 Hz, 1H), 7.75–7.65 (m, 6H), 7.41–7.32 (m, 2H), 7.24–7.15 (m, 2H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H), 5.74 (s, 2H). 13C NMR (101 MHz, DMSO): δ 162.40 (d, $J = 245.1$ Hz), 145.14, 143.14, 137.30, 136.80, 131.80, 131.53, 131.47, 130.91 (d, $J = 3.1$ Hz), 130.62, 130.31 (d, $J = 8.3$ Hz), 128.81, 126.41, 126.10, 125.68, 117.33, 115.82 (d, $J = 21.5$ Hz), 113.53, 111.49, 52.21. HRMS: calcd for [M + H], 438.06831; found, 438.06851. ## 3-(1-(3,4-Dichlorobenzyl)-1H-1,2,3-triazol-4-yl)-2-(4-(trifluoromethyl)phenyl)imidazo[1,2-a]pyridine (13e) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (30:$80\%$). Yield: 143 mg ($86\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.52 (dt, $J = 7.0$, 1.2 Hz, 1H), 8.38 (s, 1H), 8.04–7.96 (m, 2H), 7.57 (d, $J = 9.1$ Hz, 1H), 7.31–7.21 (m, 3H), 6.89 (td, $J = 6.8$, 1.3 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 162.01 (d, $J = 244.2$ Hz), 145.01, 143.63, 130.65, 127.68 (d, $J = 8.2$ Hz), 127.09, 125.23, 116.78, 115.77 (d, $J = 21.5$ Hz), 112.51, 109.12. HRMS: calcd for [M + H], 488.06511; found, 488.06511. ## 3-(1-(3,4-Dichlorobenzyl)-1H-1,2,3-triazol-4-yl)-2-(2,4-dichlorophenyl)imidazo[1,2-a]pyridine (13f) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (40:$80\%$). Yield: 120 mg ($74\%$). 1H NMR (401 MHz, DMSO-d6): δ 9.11 (dt, $J = 7.0$, 1.1 Hz, 1H), 8.02 (s, 1H), 7.71 (dt, $J = 9.1$, 1.2 Hz, 1H), 7.68 (d, $J = 2.1$ Hz, 1H), 7.64 (d, $J = 8.3$ Hz, 1H), 7.58 (d, $J = 8.3$ Hz, 1H), 7.54 (d, $J = 2.1$ Hz, 1H), 7.52 (dd, $J = 8.3$, 2.1 Hz, 1H), 7.42 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 7.25 (dd, $J = 8.3$, 2.1 Hz, 1H), 7.11 (td, $J = 6.8$, 1.2 Hz, 1H), 5.67 (s, 2H). 13C NMR (101 MHz, DMSO): δ 144.69, 140.47, 137.27, 137.05, 134.19, 133.95, 133.64, 132.58, 131.46, 131.09, 130.00, 129.45, 128.30, 127.75, 126.19, 126.02, 123.26, 117.31, 113.72, 113.61, 51.61. HRMS: calcd for [M + H], 417.06837; found, 417.06839. ## 3-(1-(3,4-Dichlorobenzyl)-1H-1,2,3-triazol-4-yl)-2-(3,4-dichlorophenyl)imidazo[1,2-a]pyridine (13g) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (30:$60\%$). Yield: 56 mg ($90\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.62 (s, 1H), 8.46 (dt, $J = 7.0$, 1.2 Hz, 1H), 7.85 (d, $J = 1.9$ Hz, 1H), 7.74–7.69 (m, 3H), 7.69–7.66 (m, 1H), 7.65 (d, $J = 2.0$ Hz, 1H), 7.62 (d, $J = 8.4$ Hz, 1H), 7.41 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 7.37 (dd, $J = 8.3$, 2.1 Hz, 1H), 7.02 (td, $J = 6.8$, 1.2 Hz, 1H), 5.76 (s, 2H). 13C NMR (101 MHz, DMSO): δ 144.96, 141.00, 136.95, 136.02, 134.78, 131.55, 131.40, 131.26, 130.91, 130.65, 130.39, 129.30, 128.58, 127.85, 126.58, 126.09, 125.46, 117.20, 113.57, 112.06, 51.97. HRMS: calcd for [M + H], 487.99978; found, 487.99980. ## 4-(3-(1-(3,4-Dichlorobenzyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridin-2-yl)benzonitrile (13h) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (30:$80\%$). Yield: 189 mg ($89\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.59 (s, 1H), 8.44 (dt, $J = 7.0$, 1.1 Hz, 1H), 7.89–7.81 (m, 4H), 7.75–7.69 (m, 3H), 7.42 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 7.36 (dd, $J = 8.3$, 2.1 Hz, 1H), 7.03 (td, $J = 6.8$, 1.2 Hz, 1H), 5.75 (s, 2H). 13C NMR (101 MHz, DMSO): δ 178.80, 132.59, 131.51, 131.28, 131.21, 130.39, 128.58, 128.51, 126.70, 126.27, 125.49, 117.31, 113.66, 52.00. HRMS: calcd for [M + H], 445.07298; found, 445.07299. ## 3-(1-(3,4-Dichlorobenzyl)-1H-1,2,3-triazol-4-yl)-2-(4-methoxyphenyl)imidazo[1,2-a]pyridine (13i) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (30:$80\%$). Yield: 385 mg ($92\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.46 (d, $J = 6.5$ Hz, 1H), 7.72–7.64 (m, 1H), 7.63–7.57 (m, 1H), 7.38–7.32 (m, 1H), 6.96 (td, $J = 6.8$, 1.2 Hz, 0H), 6.95–6.90 (m, 1H), 5.74 (s, 1H), 3.78 (s, 2H). 13C NMR (101 MHz, DMSO): δ 159.28, 144.78, 143.76, 137.11, 136.88, 131.49, 131.23, 131.14, 130.23, 129.21, 128.52, 126.50, 125.77, 125.70, 125.23, 116.84, 114.00, 112.96, 110.44, 55.28, 51.88. HRMS: calcd for [M + H], 450.08829; found, 450.08826. ## 5-(1-(4-Chlorobenzyl)-1H-1,2,3-triazol-4-yl)-6-(4-chlorophenyl)imidazo[2,1-b]thiazole (14a) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$60\%$). Yield: 136 mg ($88\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.45 (s, 1H), 8.03 (d, $J = 4.5$ Hz, 1H), 7.75–7.64 (m, 2H), 7.49–7.41 (m, 4H), 7.40–7.36 (m, 3H), 5.68 (s, 2H). 13C NMR (101 MHz, DMSO): δ 149.57, 142.75, 137.05, 135.01, 133.28, 133.07, 132.35, 130.04, 129.25, 128.95, 128.69, 123.45, 119.96, 114.46, 113.94, 52.40. HRMS: calcd for [M + H], 426.03415; found, 426.03423. ## 6-(4-Chlorophenyl)-5-(1-(3,4-dimethoxybenzyl)-1H-1,2,3-triazol-4-yl)imidazo[2,1-b]thiazole (14b) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 120 mg ($86\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.40 (s, 1H), 8.01 (d, $J = 4.5$ Hz, 1H), 7.71–7.64 (m, 2H), 7.45–7.40 (m, 2H), 7.38 (d, $J = 4.5$ Hz, 1H), 7.05 (d, $J = 2.0$ Hz, 1H), 6.95 (d, $J = 8.2$ Hz, 1H), 6.90 (dd, $J = 2.0$, 8.2 Hz, 1H), 5.56 (s, 2H), 3.74 (s, 6H). 13C NMR (101 MHz, DMSO): δ 149.80, 149.24, 142.95, 137.18, 133.60, 132.60, 129.53, 128.95, 128.45, 123.44, 121.13, 120.20, 114.75, 114.35, 112.48, 112.30, 56.00, 55.97, 53.45. HRMS: calcd for [M + H], 452.09425; found, 452.09438. ## 6-(4-Chlorophenyl)-5-(1-(4-methoxybenzyl)-1H-1,2,3-triazol-4-yl)imidazo[2,1-b]thiazole (14c) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 121 mg ($86\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.40 (s, 1H), 8.02 (d, $J = 4.5$ Hz, 1H), 7.76–7.61 (m, 3H), 7.45–7.41 (m, 3H), 7.38 (d, $J = 4.5$ Hz, 1H), 7.33 (d, $J = 8.7$ Hz, 2H), 7.01–6.88 (m, 3H), 5.59 (s, 2H), 3.75 (s, 3H). 13C NMR (101 MHz, DMSO): δ 159.63, 149.81, 142.96, 137.20, 133.60, 132.60, 130.06, 129.52, 128.97, 128.22, 123.42, 120.23, 114.60, 114.34, 55.62, 53.06. HRMS: calcd for [M + H], 422.08369; found, 422.08372. ## 5-(1-Benzyl-1H-1,2,3-triazol-4-yl)-6-(4-chlorophenyl)imidazo[2,1-b]thiazole (14d) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 154 mg ($87\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.45 (s, 1H), 8.03 (d, $J = 4.5$ Hz, 1H), 7.75–7.65 (m, 2H), 7.45–7.41 (m, 1H), 7.40–7.32 (m, 8H), 5.68 (s, 2H). 13C NMR (101 MHz, DMSO): δ 149.84, 143.02, 137.27, 136.33, 133.59, 132.62, 129.54, 129.25, 128.97, 128.64, 128.34, 123.75, 120.24, 114.75, 114.29, 53.49. HRMS: calcd for [M + Na], 392.07312; found, 392.07318. ## 6-(4-Chlorophenyl)-5-(1-(pyridin-2-ylmethyl)-1H-1,2,3-triazol-4-yl)imidazo[2,1-b]thiazole (14e) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 143 mg ($80\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.57 (ddd, $J = 4.8$, 1.9, 1.0 Hz, 1H), 8.46 (s, 1H), 8.05 (d, $J = 4.5$ Hz, 1H), 7.85 (td, $J = 7.7$, 1.8 Hz, 1H), 7.77–7.69 (m, 2H), 7.47–7.42 (m, 2H), 7.41–7.35 (m, 2H), 7.33 (dd, $J = 7.8$, 1.1 Hz, 1H), 5.81 (s, 2H). 13C NMR (101 MHz, DMSO): δ 154.95, 149.61, 149.53, 142.69, 137.55, 136.85, 133.31, 132.31, 129.26, 128.67, 124.16, 123.44, 122.24, 119.94, 114.47, 114.02, 54.64. HRMS: calcd for [M + H], 393.06837; found, 393.06845. ## 4-((4-(6-(4-Chlorophenyl)imidazo[2,1-b]thiazol-5-yl)-1H-1,2,3-triazol-1-yl)methyl)-benzonitrile (14f) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 87 mg ($87\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.49 (s, 1H), 8.05 (d, $J = 4.5$ Hz, 1H), 7.92–7.79 (m, 2H), 7.73–7.66 (m, 2H), 7.52–7.47 (m, 2H), 7.47–7.42 (m, 2H), 7.39 (d, $J = 4.5$ Hz, 1H), 5.80 (s, 2H). 13C NMR (101 MHz, DMSO): δ 149.61, 142.82, 141.48, 137.13, 133.27, 132.93, 132.37, 129.28, 128.83, 128.72, 123.76, 120.00, 118.72, 114.49, 113.87, 111.11, 52.56. HRMS: calcd for [M + H], 487.99978; found, 487.99993. ## 3-(1-(4-Chlorobenzyl)-1H-1,2,3-triazol-4-yl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridine (15a) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 145 mg ($92\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.49 (s, 1H), 8.45 (dt, $J = 7.0$, 1.2 Hz, 1H), 7.71–7.66 (m, 3H), 7.51–7.47 (m, 2H), 7.45–7.36 (m, 5H), 7.00 (td, $J = 6.8$, 1.2 Hz, 1H), 5.73 (s, 2H). 13C NMR (101 MHz, DMSO): δ 149.57, 142.75, 137.05, 135.01, 133.28, 133.07, 132.35, 130.04, 129.25, 128.95, 128.69, 123.45, 119.96, 114.46, 113.94, 52.40. HRMS: calcd for [M + H], 420.07773; found, 420.07765. ## 2-(4-Chlorophenyl)-3-(1-(3,4-dimethoxybenzyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (15b) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (30:$60\%$). Yield: 111 mg ($81\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.47–8.42 (m, 2H), 7.71–7.66 (m, 3H), 7.43–7.36 (m, 3H), 7.05 (d, $J = 1.9$ Hz, 1H), 7.03–6.95 (m, 2H), 6.91 (dd, $J = 2.0$, 8.2 Hz, 1H), 5.62 (s, 2H), 3.75 (d, $J = 3.8$ Hz, 6H). 13C NMR (101 MHz, DMSO): δ 149.27, 149.24, 145.12, 142.55, 136.41, 133.26, 133.11, 129.92, 128.94, 128.51, 126.60, 125.78, 125.63, 121.03, 117.34, 113.67, 112.40, 112.33, 112.05, 56.00, 55.97, 53.53. HRMS: calcd for [M + Na], 446.13783; found, 446.13786. ## 2-(4-Chlorophenyl)-3-(1-(4-methoxybenzyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (15c) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 96 mg ($75\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.50–8.35 (m, 2H), 7.71–7.66 (m, 3H), 7.44–7.31 (m, 5H), 7.02–6.93 (m, 1H), 5.63 (s, 2H), 3.76 (s, 3H). 13C NMR (101 MHz, DMSO): δ 159.64, 145.18, 142.68, 136.46, 133.33, 133.07, 129.96, 129.90, 128.95, 128.27, 126.51, 125.76, 125.62, 117.40, 114.64, 113.63, 112.01, 55.63, 53.16. HRMS: calcd for [M + H], 416.12726; found, 416.12730. ## 3-(1-Benzyl-1H-1,2,3-triazol-4-yl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridine (15d) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 139 mg ($78\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.50 (s, 1H), 8.45 (dt, $J = 1.1$, 6.9 Hz, 1H), 7.69 (m, 3H), 7.44–7.37 (m, 8H), 7.00 (td, $J = 1.2$, 6.8 Hz, 2H), 5.73 (s, 2H). 13C NMR (101 MHz, DMSO): δ 145.20, 142.73, 136.53, 136.40, 133.32, 133.09, 129.92, 129.29, 128.95, 128.65, 128.24, 126.54, 126.12, 125.63, 117.40, 113.65, 111.95, 53.57. HRMS: calcd for [M + H], 386.11670; found, 386.11681. ## 2-(4-Chlorophenyl)-3-(1-(pyridin-2-ylmethyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (15e) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (20:$50\%$). Yield: 147 mg ($83\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.58 (dt, $J = 4.7$, 1.4 Hz, 1H), 8.52 (s, 1H), 8.47 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.85 (td, $J = 7.7$, 1.8 Hz, 1H), 7.77–7.71 (m, 2H), 7.68 (dt, $J = 9.1$, 1.2 Hz, 1H), 7.44–7.39 (m, 2H), 7.39–7.33 (m, 3H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H), 5.87 (s, 2H). 13C NMR (101 MHz, DMSO): δ 155.00, 149.65, 144.91, 142.41, 137.55, 136.14, 133.05, 132.80, 129.64, 128.63, 126.55, 126.16, 125.31, 123.45, 122.21, 117.10, 113.29, 111.70, 54.75. HRMS: calcd for [M + H], 387.11195; found, 387.11198. ## 4-((4-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)benzonitrile (15f) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (40:$80\%$). Yield: 148 mg ($82\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.53 (s, 1H), 8.48 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.94–7.88 (m, 2H), 7.72–7.67 (m, 3H), 7.53–7.48 (m, 2H), 7.47–7.42 (m, 2H), 7.39 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 7.01 (td, $J = 6.8$, 1.2 Hz, 1H), 5.85 (s, 2H). 13C NMR (101 MHz, DMSO): δ 145.23, 142.82, 141.84, 136.69, 133.31, 133.27, 133.14, 129.94, 129.06, 129.00, 126.57, 126.42, 125.70, 119.02, 117.41, 113.65, 111.82, 111.44, 52.96. HRMS: calcd for [M + H], 411.11195; found, 411.11215. ## 2-(4-Chlorophenyl)-3-(1-(4-(methylthio)benzyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (15g) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (40:$100\%$). Yield: 72 mg ($80\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.46 (s, 1H), 8.44 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.72–7.65 (m, 3H), 7.44–7.35 (m, 3H), 7.33–7.27 (m, 3H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H), 5.67 (s, 2H), 2.47 (s, 3H). 13C NMR (101 MHz, DMSO): δ 144.91, 142.43, 138.60, 136.24, 133.02, 132.81, 132.52, 129.62, 128.76, 128.67, 126.28, 126.24, 125.67, 125.34, 117.10, 113.35, 111.67, 52.88, 14.76. HRMS: calcd for [M + H], 431.09715; found, 431.09723. ## 2-(4-Chlorophenyl)-3-(1-(1-phenylethyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (15h) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (40:$100\%$). Yield: 72 mg ($80\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.58 (s, 1H), 8.41 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.69 (dd, $J = 8.4$, 1.7 Hz, 3H), 7.44–7.32 (m, 9H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H), 6.08 (q, $J = 7.1$ Hz, 1H), 1.96 (d, $J = 7.1$ Hz, 3H). 13C NMR (101 MHz, DMSO): δ 144.88, 142.41, 141.14, 136.02, 133.00, 132.75, 129.52, 128.95, 128.58, 128.24, 126.43, 126.20, 125.28, 124.60, 117.08, 113.30, 111.70, 59.89, 21.25. HRMS: calcd for [M + H], 400.13235; found, 400.13214. ## 2-(4-Chlorophenyl)-3-(1-(4-(methylsulfonyl)benzyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (15i) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (40:$100\%$). Yield: 613 mg ($92\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.52 (dt, $J = 6.8$, 1.1 Hz, 1H), 8.07 (s, 1H), 7.98 (m, 2H), 7.72 (dt, $J = 9.1$, 1.1 Hz, 1H), 7.69 (m, 2H), 7.60 (m, 2H), 7.45 (m, 2H), 7.42 (ddd, $J = 9.1$, 6.8, 1.1 Hz, 1H), 7.04 (td, $J = 6.8$, 1.1 Hz, 1H), 5.97 (s, 2H), 3.23 (s, 3H). 13C NMR (101 MHz, DMSO-d6): δ 145.13, 143.06, 141.54, 140.72, 137.61, 135.29, 133.03, 132.90, 129.83, 128.91, 128.72, 127.68, 126.50, 125.39, 117.20, 113.61, 111.13, 57.56, 43.65. HRMS: calcd for [M - H], 462.07970; found, 462.07935. ## 2-(4-Chlorophenyl)-3-(1-(4-(pyrrolidin-1-ylsulfonyl)benzyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (15j) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (40:$100\%$). Yield: 144 mg ($93\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.49 (d, $J = 6.6$ Hz, 1H), 8.03 (s, 1H), 7.85 (m, 2H), 7.67–7.73 (m, 3H), 7.57 (m, 2H), 7.43 (m, 2H), 7.40 (m, 1H), 7.02 (t, $J = 6.6$ Hz, 1H), 5.94 (s, 2H), 3.17 (m, 4H), 1.66 (m, 4H). 13C NMR (101 MHz, DMSO-d6): δ 144.97, 142.99, 140.4, 137.50, 136.50, 134.97, 132–132.89 (m), 129.62, 128.55, 127.63, 126.13, 125.07, 116.98, 113.26, 110.97, 57.43, 47.71, 24.65. HRMS: calcd for [M – H], 517.12190; found, 517.12134. ## 2-(4-Chlorophenyl)-3-(1-(4-nitrobenzyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (15k) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (40:$100\%$). Yield: 253 mg ($88\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.56 (s, 1H), 8.49 (dt, $J = 6.9$, 1.2 Hz, 1H), 8.31–8.21 (m, 2H), 7.73–7.67 (m, 3H), 7.62–7.54 (m, 2H), 7.47–7.41 (m, 2H), 7.39 (ddd, $J = 9.0$, 6.7, 1.3 Hz, 1H), 7.00 (td, $J = 6.8$, 1.3 Hz, 1H), 5.92 (s, 2H). 13C NMR (101 MHz, DMSO): δ 147.73, 145.24, 143.78, 142.85, 136.75, 133.30, 133.14, 129.95, 129.39, 129.00, 126.55, 126.44, 125.70, 124.43, 117.40, 113.64, 111.80, 52.70. HRMS: calcd for [M - H], 430.09; found, 429.555. ## 4-((4-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)aniline (15l) 2-(4-Chlorophenyl)-3-(1-(4-nitrobenzyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (15k) was dissolved in MeOH and AcOH (7 equiv), and Fe (3.5 equiv) was added. The reaction mixture was stirred at reflux until the completion of the reaction. After cooling to 25 °C, the mixture was extracted with EtOAc, washed with NaHCO3 solution, and the organic phase was dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography, mobile phase cyclohexane/EtOAc (40–$100\%$). Yield: 123 mg ($78\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.41 (dt, $J = 7.0$, 1.3 Hz, 1H), 8.39 (s, 1H), 7.72–7.65 (m, 3H), 7.43–7.32 (m, 3H), 7.10–7.04 (m, 2H), 6.98 (td, $J = 6.8$, 1.3 Hz, 1H), 6.59–6.53 (m, 2H), 5.48 (s, 2H), 5.20 (s, 2H). 13C NMR (101 MHz, DMSO): δ 149.31, 145.14, 142.57, 136.35, 133.35, 133.06, 129.86, 129.63, 128.90, 126.44, 125.58, 125.47, 122.83, 117.37, 114.23, 113.57, 112.10, 53.72. HRMS: calcd for [M + H], 401.12760; found, 401.12735. ## N-(4-((4-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)phenyl)-acetamide (15m) 4-((4-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)aniline was dissolved in dioxane, and Ac2O (1.5 equiv) was added followed by pyridine (1.5 equiv). The reaction mixture was stirred at 25 °C overnight. After the completion of the reaction, the mixture was evaporated, diluted with EtOAc, and washed with water. The organic phase was dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography. Mobile phase cyclohexane/EtOAc (40–$100\%$). Yield: 63 mg ($92\%$). 1H NMR (401 MHz, DMSO-d6): δ 10.02 (s, 1H), 8.45 (s, 1H), 8.45–8.42 (m, 1H), 7.72–7.66 (m, 3H), 7.63–7.58 (m, 2H), 7.39 (s, 2H), 7.39–7.34 (m, 1H), 7.33–7.28 (m, 2H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H), 5.65 (s, 2H), 2.04 (s, 3H). 13C NMR (101 MHz, DMSO): δ 168.54, 144.88, 142.37, 139.48, 136.20, 133.04, 132.78, 130.37, 129.59, 128.65, 128.62, 126.18, 125.58, 125.34, 119.31, 117.08, 113.30, 111.69, 66.52, 53.00, 24.17. HRMS: calcd for [M + H], 443.13816; found, 443.13773. ## 5-(6-(4-Chlorophenyl)imidazo[2,1-b]thiazol-5-yl)-3-(3,4-dichlorobenzyl)-1,2,4-oxadiazole (16A) 6-(4-Chlorophenyl)imidazo[2,1-b]thiazole-5-carboxylic acid was dissolved in dry DMF, degassed, and refilled with argon. EDC (1 equiv) and HOBt (1 equiv) were added in one portion, and the mixture was stirred at 25 °C for 30 min. Then, a solution of (E)-2-(3,4-dichlorophenyl)-N′-hydroxyacetimidamide in dry DMF was added, and the reaction mixture was stirred at 80 °C overnight. After cooling to 25 °C, the mixture was diluted with EtOAc, washed with NaHCO3 solution, and water, and the organic phase was dried over sodium sulfate. The product was isolated by flash column chromatography, mobile phase petrol ether/EtOAc (50:$70\%$). Yield: 250 mg ($52\%$). 1H NMR (500 MHz, DMSO-d6): δ 8.33 (d, $J = 4.4$ Hz, 1H), 7.94–7.91 (m, 2H), 7.71 (d, $J = 2.1$ Hz, 1H), 7.63 (d, $J = 4.0$ Hz, 1H), 7.62 (s, 1H), 7.57–7.53 (m, 2H), 7.42–7.39 (m, 1H), 4.23 (s, 2H). 13C NMR (126 MHz, DMSO): δ 168.78, 167.94, 153.92, 150.51, 137.29, 134.31, 132.08, 131.65, 131.45, 131.25, 131.09, 130.18, 130.10, 128.70, 121.58, 117.20, 109.75, 30.64. HRMS: calcd for [M + H], 460.97919; found, 460.97921. ## 5-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-3-(3,4-dichlorobenzyl)-1,2,4-oxadiazole (16B) 2-(4-Chlorophenyl)imidazo[1,2-a]pyridine-3-carboxylic acid was dissolved in dry DMF, degassed, and refilled with argon. EDC (1 equiv) and HOBt (1 equiv) were added in one portion, and the mixture was stirred at 25 °C for 30 min. Then, a solution of (E)-2-(3,4-dichlorophenyl)-N′-hydroxyacetimidamide in dry DMF was added, and the reaction mixture was stirred at 80 °C overnight. After cooling to 25 °C, the mixture was diluted with EtOAc and washed with NaHCO3 solution and water, and the organic phase was dried over sodium sulfate. The product was isolated by flash column chromatography, mobile phase petrol ether/EtOAc (50:$70\%$). Yield: 252 mg ($50\%$). 1H NMR (401 MHz, DMSO-d6): δ 9.41 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.90 (dt, $J = 9.0$, 1.2 Hz, 1H), 7.88–7.84 (m, 2H), 7.71 (d, $J = 2.0$ Hz, 1H), 7.67 (ddd, $J = 9.0$, 6.9, 1.3 Hz, 1H), 7.63 (d, $J = 8.3$ Hz, 1H), 7.60–7.54 (m, 2H), 7.42 (dd, $J = 8.3$, 2.1 Hz, 1H), 7.36 (td, $J = 7.0$, 1.3 Hz, 1H), 4.27 (s, 2H). 13C NMR (101 MHz, DMSO): δ 168.27, 167.90, 150.18, 147.19, 137.02, 134.29, 132.16, 131.54, 131.33, 131.19, 130.82, 129.90, 129.77, 129.38, 128.39, 128.01, 117.58, 115.47, 30.46. HRMS: calcd for [M + H], 455.02277; found, 455.02288. ## 2-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-5-(3,4-dichlorobenzyl)-1,3,4-oxadiazole (17) 2-(4-Chlorophenyl)-N′-(2-(3,4-dichloro-phenyl)acetyl)imidazo[1,2-a]pyridine-3-carbohydrazide was dissolved in dry DCM, and tosyl chloride (1.5 equiv) was added, followed by TEA (3 equiv) at 0 °C. The reaction mixture was stirred at 25 °C overnight. The mixture was diluted with water and extracted with EtOAc, and the organic phase was washed with a saturated NaHCO3 solution. The organic phase was dried over sodium sulfate. The product was isolated by flash column and RP-flash column chromatography; mobile phase hexane/EtOAc (30:$60\%$), and (H2O/CH3CN 10:$80\%$). Yield: 16 mg ($6\%$). 1H NMR (500 MHz, DMSO-d6): δ 9.30 (dt, $J = 7.0$, 1.2 Hz, 1H), 7.82 (dt, $J = 9.0$, 1.2 Hz, 1H), 7.79–7.75 (m, 2H), 7.61–7.57 (m, 3H), 7.43–7.39 (m, 2H), 7.31–7.25 (m, 2H), 4.35 (s, 2H). 13C NMR (126 MHz, DMSO): δ 163.81, 158.08, 147.83, 146.98, 135.75, 134.31, 132.41, 131.77, 131.62, 131.32, 131.12, 130.70, 129.92, 128.65, 128.52, 127.96, 117.70, 115.12, 106.70, 30.17. HRMS: calcd for [M + H], 455.02277; found, 455.02282. ## 2-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-5-(3,4-dichlorobenzyl)-1,3,4-thiadiazole (18) A round-bottom flask was charged with 2-(4-chlorophenyl)-N′-(2-(3,4-dichlorophenyl)acetyl)imidazo[1,2-a]pyridine-3-carbohydrazide, degassed, and refilled with argon. Dry toluene was added, and the mixture was degassed once more and refilled with argon. Lawesson’s reagent (3 equiv) was added, and the mixture was stirred at 100 °C overnight. After cooling to 25 °C, the mixture was diluted with water and extracted with EtOAc. An organic phase was dried over sodium sulfate and evaporated. A residue was purified by flash column chromatography. Mobile phase H2O/MeOH (30:$100\%$). Yield: 44 mg ($21\%$). 1H NMR (500 MHz, DMSO-d6): δ 9.35 (dt, $J = 7.0$, 1.2 Hz, 1H), 7.85 (dt, $J = 9.0$, 1.2 Hz, 1H), 7.78–7.73 (m, 2H), 7.65–7.57 (m, 3H), 7.42–7.38 (m, 2H), 7.31 (ddd, $J = 8.2$, 4.5, 1.7 Hz, 2H), 4.35 (s, 2H). 13C NMR (126 MHz, DMSO): δ 163.75, 158.07, 147.63, 146.86, 135.84, 134.23, 132.26, 131.72, 131.62, 131.43, 131.16, 130.56, 130.10, 128.84, 128.55, 128.08, 117.70, 115.31, 106.65, 29.99. HRMS: calcd for [M + Na], 477.00472; found, 477.00467. ## 6-(4-Chlorophenyl)-5-(2-(3,4-dichlorobenzyl)thiazol-4-yl)imidazo[2,1-b]thiazole (19A) 2-Chloro-1-(6-(4-chlorophenyl)imidazo[2,1-b]thiazol-5-yl)ethan-1-one [34] was dissolved in EtOH and 2-(3,4-dichlorophenyl)ethanethioamide (1.5 equiv) was added. A reaction mixture was stirred at reflux overnight. After cooling to 25 °C, the mixture was purified by RP-flash column chromatography. Mobile phase H2O/CH3CN (20:$80\%$). Yield: 213 mg ($61\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.01 (d, $J = 4.5$ Hz, 1H), 7.71 (d, $J = 2.0$ Hz, 1H), 7.68–7.61 (m, 4H), 7.42–7.36 (m, 4H), 4.46 (s, 2H). 13C NMR (101 MHz, DMSO): δ 169.35, 149.37, 143.99, 143.09, 139.54, 133.79, 132.56, 131.60, 131.53, 131.24, 130.15, 130.04, 129.84, 128.83, 120.43, 118.35, 117.72, 114.49, 37.55. HRMS: calcd for [M + H], 475.96110; found, 475.96130. ## 4-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-2-(3,4-dichlorobenzyl)thiazole (19B) 2-Chloro-1-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)ethan-1-one [35] was dissolved in EtOH, and 2-(3,4-dichlorophenyl)ethanethioamide (1.5 equiv) was added. A reaction mixture was stirred at reflux overnight. After cooling to 25 °C, the mixture was purified by RP-flash column chromatography. Mobile phase: H2O/CH3CN (20:$80\%$). Yield: 112 mg ($62\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.42 (d, $J = 6.9$ Hz, 1H), 7.83 (s, 1H), 7.74 (d, $J = 2.1$ Hz, 1H), 7.67 (q, $J = 8.1$ Hz, 4H), 7.45–7.35 (m, 4H), 6.99 (t, $J = 6.9$ Hz, 1H), 4.51 (s, 2H). 13C NMR (101 MHz, DMSO): δ 169.57, 144.48, 143.11, 142.15, 139.27, 133.16, 132.66, 131.29, 131.25, 130.96, 129.86, 129.73, 129.68, 128.56, 126.10, 125.41, 121.39, 117.09, 115.60, 113.16, 37.34. HRMS: calcd for [M + H], 470.00468; found, 470.00488. ## 3-Benzyl-5-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)isoxazole (20) Phenylacetaldehyde (0.045 mL, 1 equiv) was dissolved in an H2O/t-BuOH mixture (4 mL) and NH2OH·HCl (30 mg, 1 equiv), followed by NaOH (20 mg, 1 equiv) addition. The reaction mixture was stirred at rt for 3 h; then, 2-(4-chlorophenyl)-3-ethynylimidazo[1,2-a]pyridine 33 (100 mg, 0.39 mmol), chloramine T (120 mg, 1 equiv), and CuI (8 mg, 10 mol %) were added; and the reaction mixture was stirred overnight. The reaction mixture was diluted with water and extracted with EtOAc. Combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by RP-flash column chromatography. Mobile phase H2O/CH3CN (10:$100\%$). Yield: 99 mg ($66\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.48 (dt, $J = 6.9$, 1.1 Hz, 1H), 7.76 (dt, $J = 9.1$, 1.2 Hz, 1H), 7.68–7.63 (m, 2H), 7.51–7.46 (m, 3H), 7.35 (m, 4H), 7.26 (ddt, $J = 6.9$, 4.6, 3.1 Hz, 1H), 7.12 (td, $J = 6.9$, 1.2 Hz, 1H), 6.82 (s, 1H), 4.11 (s, 2H). 13C NMR (101 MHz, DMSO): δ 163.90, 160.02, 146.13, 145.18, 137.73, 133.78, 132.59, 130.33, 129.23, 129.09, 129.04, 127.82, 127.16, 126.15, 117.66, 114.67, 105.32, 31.94. Anal. ( C23H16ClN3O·0.75H2O): C, H, N. HRMS: calcd for [M + H], 386.10547; found, 386.10564. EA: C, 69.17; H, 4.42; N, 10.52. Found: C, 68.94; H, 4.11; N, 10.41. ## 2-(4-Chlorophenyl)-3-(1-(3,4-dichlorobenzyl)-1H-pyrrol-2-yl)imidazo[1,2-a]pyridine (21) 2-(4-Chlorophenyl)-3-(1H-pyrrol-2-yl)imidazo[1,2-a]pyridine 35 (76 mg, 0.26 mmol) was dissolved in dry DMF (2 mL), degassed, and refilled with argon. NaH (10 mg, 1.3 equiv, $60\%$ in mineral oil) was added, and the mixture was stirred at 25 °C for 30 min. 1,2-Dichloro-4-(chloromethyl)benzene (0.05 mL, 1.3 equiv) was added, and the reaction mixture was stirred at 25 °C overnight. Then, the mixture was diluted with water and extracted with EtOAc. Combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography, mobile phase petrol ether/EtOAc (15:$50\%$). Yield: 25 mg ($21\%$). 1H NMR (401 MHz, DMSO-d6): δ 7.57 (m, 3H), 7.47 (d, $J = 6.8$ Hz, 1H), 7.39 (dd, $J = 2.7$, 1.7 Hz, 1H), 7.38–7.33 (m, 2H), 7.27 (ddd, $J = 9.0$, 6.7, 1.2 Hz, 1H), 7.17 (d, $J = 8.2$ Hz, 1H), 6.74 (td, $J = 6.8$, 1.1 Hz, 1H), 6.64 (d, $J = 2.0$ Hz, 1H), 6.58 (dd, $J = 8.3$, 2.0 Hz, 1H), 6.51 (dd, $J = 3.6$, 1.7 Hz, 1H), 6.43–6.38 (m, 1H), 4.80 (d, $J = 15.4$ Hz, H), 4.53 (d, $J = 15.4$ Hz, 1H). 13C NMR (101 MHz, DMSO): δ 144.60, 142.39, 138.69, 132.84, 132.59, 130.86, 130.34, 129.80, 128.70, 128.62, 127.99, 127.04, 126.00, 125.48, 123.94, 118.69, 116.79, 113.98, 112.70, 112.19, 109.16, 49.78. HRMS: calcd for [M + H], 452.04826; found, 452.04842. ## 2-(4-Chlorophenyl)-3-(1-(3,4-dichlorobenzyl)-1H-pyrazol-4-yl)imidazo[1,2-a]pyridine (22) 2-(4-Chlorophenyl)-3-(1H-pyrazol-4-yl)imidazo[1,2-a]pyridine 36 (30 mg, 0.1 mmol) was dissolved in CH3CN (2 mL), and the solution was degassed and refilled with argon. K2CO3 (15 mg, 1.1 equiv) was added, followed by the addition of 1,2-dichloro-4-(chloromethyl)benzene (0.02 mL, 1 equiv). The reaction mixture was refluxed overnight under an argon atmosphere. The product was isolated from preparative TLC; mobile phase: petrol ether/EtOAc 3:2. Yield: 15 mg ($32\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.26 (s, 0H), 8.09 (dd, $J = 6.9$, 1.5 Hz, 0H), 7.80 (s, 0H), 7.73–7.68 (m, 1H), 7.68–7.60 (m, 1H), 7.58 (d, $J = 2.1$ Hz, 1H), 7.41 (dd, $J = 8.5$, 1.7 Hz, 1H), 7.36–7.24 (m, 1H), 6.97–6.90 (m, 1H), 5.49 (s, 1H). 13C NMR (101 MHz, DMSO): δ 144.74, 141.23, 140.90, 139.00, 133.78, 132.57, 132.37, 131.36, 130.87, 130.02, 129.79, 129.36, 128.82, 128.38, 125.81, 124.78, 117.28, 113.42, 113.20, 109.08, 54.16. HRMS: calcd for [M + H], 453.04351; found, 453.04356. ## 2-(4-Chlorophenyl)-3-(1-(3,4-dichlorophenyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (23) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (30:$100\%$). Yield: 225 mg ($73\%$). 1H NMR (401 MHz, DMSO-d6): δ 9.31 (s, 1H), 8.49 (dt, $J = 6.9$, 1.1 Hz, 2H), 8.37 (d, $J = 2.5$ Hz, 1H), 8.07 (dd, $J = 8.8$, 2.5 Hz, 1H), 7.95 (d, $J = 8.8$ Hz, 1H), 7.85–7.78 (m, 3H), 7.73 (dt, $J = 9.1$, 1.0 Hz, 2H), 7.48–7.44 (m, 3H), 7.44–7.39 (m, 1H), 7.02 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 145.10, 142.84, 136.98, 136.19, 132.88, 132.84, 132.55, 132.06, 131.49, 129.65, 128.73, 126.53, 125.45, 124.34, 122.30, 120.57, 117.14, 113.39, 110.89. HRMS: calcd for [M + H], 440.02311; found, 440.02321. ## 2-(4-Chlorophenyl)-3-(1-(3,4-dichlorophenethyl)-1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (24) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (30:$100\%$). Yield: 138 mg ($95\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.33 (s, 0H), 8.25 (d, $J = 6.9$ Hz, 1H), 7.69 (d, $J = 9.0$ Hz, 1H), 7.64 (d, $J = 8.5$ Hz, 1H), 7.58–7.49 (m, 1H), 7.39 (d, $J = 8.5$ Hz, 2H), 7.21–7.13 (m, 1H), 6.99 (t, $J = 6.6$ Hz, 1H), 4.78 (t, $J = 6.7$ Hz, 1H), 3.26 (t, $J = 6.7$ Hz, 1H). 13C NMR (101 MHz, DMSO): δ 144.84, 142.21, 139.06, 135.60, 132.98, 132.77, 131.08, 131.05, 130.64, 129.49, 129.46, 129.40, 128.61, 126.20, 125.77, 125.00, 117.13, 113.27, 111.64, 50.51, 34.70. HRMS: calcd for [M + H], 468.05441; found, 468.05450. ## Ethyl 6-(4-Chlorophenyl)imidazo[2,1-b]thiazole-5-carboxylate (25) 2-Aminothiazole (3 equiv) was dissolved in CH3CN, and ethyl 3-(4-chlorophenyl)-3-oxopropanoate (1 equiv) was added as a solution in CH3CN, followed by the addition of CBr4 (2 equiv). The reaction mixture was stirred at 80 °C overnight. After the completion of the reaction, the mixture was evaporated to a minimal volume, diluted with water, and extracted with EtOAc. Combined organic phases were dried over sodium sulfate and purified by flash column chromatography. Mobile phase petrol ether/EtOAc (25:$50\%$). Yield: 365 mg ($82\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.22 (d, $J = 4.5$ Hz, 1H), 7.89–7.85 (m, 2H), 7.54 (d, $J = 4.5$ Hz, 2H), 7.52–7.48 (m, 2H), 4.30 (q, $J = 7.1$ Hz, 2H), 1.27 (t, $J = 7.1$ Hz, 3H). 13C NMR (101 MHz, DMSO): δ 159.18, 152.65, 151.43, 133.50, 132.49, 131.49, 127.90, 121.94, 116.12, 114.53, 60.83, 14.12. HRMS: calcd for [M + H], 307.03025; found, 307.03034. ## Ethyl 2-(4-Chlorophenyl)imidazo[1,2-a]pyridine-3-carboxylate (26) 2-Aminopyridine (3 equiv) was dissolved in CH3CN, and ethyl 3-(4-chlorophenyl)-3-oxopropanoate (1 equiv) was added as a solution in CH3CN followed by the addition of CBr4 (2 equiv). A reaction mixture was stirred at 80 °C overnight. After the completion of the reaction, the mixture was evaporated to a minimal volume, diluted with water, and extracted with EtOAc. Combined organic phases were dried over sodium sulfate and purified by flash column chromatography. Spectral characteristics match those described in the literature.46 ## 6-(4-Chlorophenyl)imidazo[2,1-b]thiazole-5-carboxylic Acid (27) The title compound was prepared according to General Procedure V with a minor modification. The product was filtered as a precipitate after acidification. Yield: 286 mg ($90\%$). Spectral characteristics matched those described in the literature.47 ## 2-(4-Chlorophenyl)imidazo[1,2-a]pyridine-3-carboxylic Acid (28) The title compound was prepared according to General Procedure V with a minor modification. The product was filtered as a precipitate after acidification. Yield: 548 mg (quant.). 1H NMR (401 MHz, DMSO-d6): δ 9.40–9.32 (m, 1H), 7.83–7.75 (m, 3H), 7.59–7.52 (m, 1H), 7.50 (dd, $J = 9.0$, 2.5 Hz, 2H), 7.21 (td, $J = 6.9$, 1.3 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 162.11, 151.23, 146.75, 133.77, 133.70, 132.35, 129.28, 128.92, 128.80, 128.04, 127.77, 117.53, 115.02, 112.51. HRMS: calcd for [M + H], 273.04253; found, 273.04262. ## 2-(4-Chlorophenyl)imidazo[1,2-a]pyridine-3-carbohydrazide (29) Ester derivative was dissolved in absolute EtOH and N2H4·H2O (10 equiv) was added. The reaction mixture was refluxed overnight. The reaction mixture was cooled to 25 °C, and the formed precipitate was filtered, washed with EtOH, and dried. Yield: 412 mg ($87\%$). 1H NMR (401 MHz, DMSO-d6): δ 9.70 (s, 1H), 8.57 (d, $J = 7.0$ Hz, 1H), 7.92–7.77 (m, 2H), 7.66 (d, $J = 9.1$ Hz, 1H), 7.56–7.43 (m, 2H), 7.40 (t, $J = 8.0$ Hz, 1H), 7.04 (t, $J = 6.9$ Hz, 1H), 4.67 (s, 2H). 13C NMR (101 MHz, DMSO): δ 160.38, 144.68, 143.00, 132.96, 132.64, 129.97, 128.57, 126.85, 126.37, 117.04, 115.62, 113.53. HRMS: calcd for [M + H], 287.06942; found, 287.06950. ## 2-(4-Chlorophenyl)-N′-(2-(3,4-dichlorophenyl)acetyl)imidazo[1,2-a]pyridine-3-carbohydrazide (30) 2-(4-Chlorophenyl)imidazo[1,2-a]pyridine-3-carbohydrazide was dissolved with dry DMF and ethyl 2-(4-chlorophenyl)imidazo[1,2-a]pyridine-3-carboxylate (1 equiv) was added. The reaction mixture was degassed and refilled with argon. HATU (1.2 equiv) was added, followed by the addition of DIPEA (1.2 equiv). The reaction mixture was stirred at 25 °C and monitored by UPLC. After the completion of the reaction, the solution was evaporated to a minimal volume and purified by reverse-phase flash column chromatography. Mobile phase petrol ether/EtOAc (40:$80\%$). Yield: 298 mg ($90\%$). 1H NMR (401 MHz, DMSO-d6): δ 10.49 (s, 1H), 10.43 (s, 1H), 8.80 (dt, $J = 7.0$, 1.2 Hz, 1H), 8.06–7.98 (m, 2H), 7.95 (s, 1H), 7.71 (dt, $J = 9.0$, 1.2 Hz, 1H), 7.65–7.60 (m, 2H), 7.49 (dd, $J = 9.0$, 2.3 Hz, 2H), 7.44 (ddd, $J = 9.0$, 6.8, 1.3 Hz, 1H), 7.34 (dd, $J = 8.3$, 2.1 Hz, 1H), 7.10 (td, $J = 6.9$, 1.2 Hz, 1H), 3.63 (s, 2H). 13C NMR (101 MHz, DMSO): δ 169.21, 160.31, 145.02, 144.01, 136.78, 133.29, 132.28, 131.35, 130.97, 130.62, 130.18, 129.86, 129.58, 128.59, 127.33, 126.68, 117.16, 114.63, 113.89, 35.98. HRMS: calcd for [M + H], 473.03334; found, 473.03327. ## 2-Chloro-1-(6-(4-chlorophenyl)imidazo[2,1-b]thiazol-5-yl)ethan-1-one (31) 6-(4-Chlorophenyl)imidazo[2,1-b]thiazole (1 mmol) was dissolved in dry dioxane (4 mL), and chloroacetyl chloride (3 equiv) was added in one portion. The reaction mixture was stirred at 70 °C under an argon atmosphere for 30 min and then at 100 °C overnight. After cooling to 25 °C, a precipitate was formed. The suspension was diluted with a saturated NaHCO3 solution and extracted with EtOAc. The organic phase was dried over sodium sulfate and purified by flash column chromatography, mobile phase petrol ether/EtOAc (30:$60\%$). Yield: 273 mg ($91\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.46 (d, $J = 4.4$ Hz, 1H), 7.73–7.68 (m, 2H), 7.64 (d, $J = 4.4$ Hz, 1H), 7.62–7.57 (m, 2H), 4.41 (s, 2H). 13C NMR (101 MHz, DMSO): δ 180.76, 154.65, 153.23, 134.82, 133.38, 132.00, 129.02, 122.62, 122.19, 117.57, 47.18. HRMS: calcd for [M + H], 310.98072; found, 310.98089. ## 2-Chloro-1-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)ethan-1-one (32) 2-(4-Chlorophenyl)imidazo[1,2-a]pyridine (1 mmol) was dissolved in dry dioxane (4 mL) and chloroacetyl chloride (3 equiv) was added in one portion. The reaction mixture was stirred at 70 °C under an argon atmosphere for 30 min and then at 100 °C overnight. After cooling to 25 °C, a precipitate was formed. The suspension was diluted with a saturated NaHCO3 solution and extracted with EtOAc. The organic phase was dried over sodium sulfate and purified by flash column chromatography; mobile phase petrol ether/EtOAc (30:$60\%$). Yield: 492 mg ($76\%$). 1H NMR (401 MHz, DMSO-d6): δ 9.63 (d, $J = 6.9$ Hz, 1H), 7.94 (d, $J = 8.9$ Hz, 1H), 7.85 (t, $J = 7.9$ Hz, 1H), 7.73 (d, $J = 8.2$ Hz, 2H), 7.65 (d, $J = 8.2$ Hz, 2H), 7.45 (t, $J = 6.9$ Hz, 1H), 4.36 (s, 2H). 13C NMR (101 MHz, DMSO): δ 182.37, 151.37, 145.94, 135.44, 132.49, 132.19, 132.15, 129.23, 129.22, 119.42, 117.34, 116.76, 47.93. HRMS: calcd for [M + H], 305.02429; found, 305.02434. ## 2-(4-Chlorophenyl)-3-ethynylimidazo[1,2-a]pyridine (33) 2-(4-Chlorophenyl)-3-((trimethylsilyl)ethynyl)imidazo[1,2-a]pyridine (11, 513 mg, 1.58 mmol) was dissolved in MeOH and K2CO3 (435 mg, 2 equiv) was added in one portion. The reaction was stirred at 25 °C and monitored by TLC. After the completion of the reaction, the mixture was diluted with DCM and washed with water. Combined organic phases were dried over sodium sulfate and evaporated, and the residue was purified by flash column chromatography. Mobile phase petrol ether/EtOAc (30:$50\%$). Yield: 362 mg ($90\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.46 (dt, $J = 6.8$, 1.2 Hz, 1H), 8.28–8.20 (m, 2H), 7.71 (dt, $J = 9.0$, 1.1 Hz, 1H), 7.60–7.52 (m, 2H), 7.45 (ddd, $J = 9.0$, 6.8, 1.3 Hz, 1H), 7.12 (td, $J = 6.8$, 1.2 Hz, 1H), 5.44 (s, 1H). 13C NMR (101 MHz, DMSO): δ 145.95, 144.71, 133.40, 131.99, 128.97, 128.32, 128.28, 127.70, 125.77, 117.27, 114.16, 103.47, 94.39, 72.47. HRMS: calcd for [M + Na], 253.05270; found, 253.05273. ## 2-(4-Chlorophenyl)-3-(1H-1,2,3-triazol-4-yl)imidazo[1,2-a]pyridine (34) 2-(4-Chlorophenyl)-3-ethynylimidazo[1,2-a]pyridine 33 (100 mg, 0.396 mmol) was dissolved in DMF/MeOH (10:1) mixture, degassed, and purged with argon. To this solution, CuI (10 mol %) and TMSN3 (1 equiv) were added and the mixture was stirred at 70 °C overnight. After the completion of the reaction, the mixture was diluted with EtOAc and washed with water, and the combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography, mobile phase cyclohexane/EtOAc (10:$70\%$). Yield: 98 mg ($84\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.42 (dt, $J = 7.0$, 1.2 Hz, 1H), 8.13 (s, 1H), 7.76–7.66 (m, 3H), 7.47–7.41 (m, 2H), 7.37 (ddd, $J = 9.0$, 6.7, 1.2 Hz, 1H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 145.21, 142.71, 135.51, 133.42, 133.08, 130.22, 129.90, 129.90, 128.95, 126.44, 125.60, 117.35, 113.58, 112.24. HRMS: calcd for [M + H], 296.06975; found, 296.06964. ## 2-(4-Chlorophenyl)-3-(1H-pyrrol-2-yl)imidazo[1,2-a]pyridine (35) 2-(4-Chlorophenyl)-3-iodoimidazo[1,2-a]pyridine (430 mg, 1.21 mmol) was dissolved in dioxane (12 mL) and (1-(tert-butoxycarbonyl)-1H-pyrrol-3-yl)boronic acid (260 mg, 1 equiv) was added, followed by a solution of Na2CO3 (392 mg, 3 equiv) in 3 mL of H2O. A reaction mixture was degassed and refilled with argon. Pd(PPh3)4 (72 mg, $5\%$ mol) was added, and the mixture was degassed again and refilled with argon. The reaction mixture was stirred at 90 °C overnight. After cooling to 25 °C, the mixture was diluted with water and extracted with EtOAc. Combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography; mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 103 mg ($35\%$). 1H NMR (401 MHz, DMSO-d6): δ 11.28 (s, 1H), 7.99 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.69–7.62 (m, 3H), 7.42–7.37 (m, 2H), 7.32 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 7.07 (td, $J = 2.7$, 1.5 Hz, 1H), 6.93 (td, $J = 6.8$, 1.2 Hz, 1H), 6.40–6.37 (m, 1H), 6.34 (dt, $J = 3.3$, 2.5 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 144.27, 141.30, 133.34, 132.24, 128.58, 128.51, 125.77, 124.66, 120.79, 117.38, 116.91, 114.57, 112.91, 111.12, 109.48. HRMS: calcd for [M + H], 294.07925; found, 294.07926. ## 2-(4-Chlorophenyl)-3-(1H-pyrazol-4-yl)imidazo[1,2-a]pyridine (36) 2-(4-Chlorophenyl)-3-iodoimidazo[1,2-a]pyridine (334 mg, 0.94 mmol) was dissolved in dioxane (8 mL) and 4-(4,4,5,5-tetramethyl-1,3,2-dioxaborolan-2-yl)-1H-pyrazole (183 mg, 1 equiv) was added, followed by a solution of Na2CO3 (3 equiv) in 2 mL of H2O. A reaction mixture was degassed and refilled with argon. Pd(dppf)Cl2 (40 mg, $5\%$ mol) was added, and the mixture was degassed again and refilled with argon. The reaction mixture was stirred at 90 °C overnight. After cooling to 25 °C, the mixture was diluted with water and extracted with EtOAc. Combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography; mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 112 mg ($40\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.54 (d, $J = 6.7$ Hz, 1H), 8.45 (s, 1H), 8.00 (d, $J = 8.3$ Hz, 2H), 7.59 (d, $J = 9.1$ Hz, 1H), 7.51 (d, $J = 8.3$ Hz, 2H), 7.29–7.22 (m, 1H), 6.92 (t, $J = 6.7$ Hz, 1H). HRMS: calcd for [M + H], 295.07450; found, 295.07467. ## Methyl 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)-methyl)benzoate (37) The title compound was prepared according to General Procedure IV. Mobile phase petrol ether/EtOAc (60:$90\%$). Yield: 456 mg ($84\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.54 (s, 1H), 8.46 (dt, $J = 7.0$, 1.1 Hz, 1H), 7.84–7.81 (m, 1H), 7.72–7.67 (m, 3H), 7.65 (d, $J = 8.3$ Hz, 1H), 7.56 (dd, $J = 8.3$, 2.3 Hz, 1H), 7.44–7.40 (m, 2H), 7.40–7.36 (m, 1H), 7.00 (td, $J = 6.8$, 1.2 Hz, 1H), 5.79 (s, 2H), 3.87 (s, 3H). 13C NMR (101 MHz, DMSO): δ 165.39, 144.95, 142.46, 136.39, 135.56, 133.01, 132.84, 132.82, 131.82, 131.52, 130.65, 130.46, 129.62, 128.68, 126.29, 125.94, 125.39, 117.12, 113.37, 111.55, 52.89, 52.07. HRMS: calcd for [M + H], 478.08328; found, 478.08321. ## 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-benzoic Acid (38) The title compound was prepared according to General Procedure V. Mobile phase: H2O/MeOH (10:$80\%$). Yield: 528 mg ($92\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.56 (s, 1H), 8.46 (d, $J = 6.9$ Hz, 1H), 7.78 (d, $J = 2.3$ Hz, 1H), 7.73–7.67 (m, 3H), 7.62 (d, $J = 8.3$ Hz, 1H), 7.52 (dd, $J = 8.3$, 2.3 Hz, 1H), 7.41 (dd, $J = 18.9$, 8.6 Hz, 3H), 7.01 (t, $J = 6.8$ Hz, 1H), 5.79 (s, 2H). 13C NMR (101 MHz, DMSO): δ 166.58, 144.94, 142.42, 136.36, 135.43, 133.00, 132.85, 132.14, 131.83, 131.66, 131.37, 130.37, 129.57, 128.70, 126.28, 125.98, 125.36, 117.11, 113.35, 111.54, 52.11. HRMS: calcd for [M + H], 430.10653; found, 430.10615. ## 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)benzamide (39) The title compound was prepared according to General Procedure VI. Mobile phase: H2O/MeOH (30:$100\%$). Yield: 251 mg ($88\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.56 (s, 1H), 8.51–8.43 (m, 1H), 8.00–7.84 (m, 1H), 7.70 (td, $J = 6.8$, 2.1 Hz, 4H), 7.53 (t, $J = 7.3$ Hz, 1H), 7.49–7.34 (m, 4H), 7.00 (qd, $J = 6.8$, 6.1, 1.1 Hz, 1H), 5.76 (s, 2H). 13C NMR (101 MHz, DMSO): δ 167.85, 140.71, 137.56, 135.11, 134.72, 134.37, 133.41, 133.00, 130.31, 130.17, 129.63, 129.36, 128.39, 127.21, 127.06, 126.55, 117.19, 113.25, 113.19, 52.31. HRMS: calcd for [M + H], 463.08354; found, 463.08359. ## 3-(1-(3-Carbamoyl-4-chlorobenzyl)-1H-1,2,3-triazol-4-yl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridin-1-ium Chloride (39HCl) 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)benzamide was dissolved in THF and cooled in an ice bath, and HCl 1 M in ether was added while stirring vigorously. The formed salt was stirred at 25 °C for 20 min, diluted with dry diethylether, filtered, and washed with more diethylether. EA: C, $54.96\%$; H, $3.51\%$; N, $16.54\%$; Cl, $21.48\%$. ## 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-N-methylbenzamide (40) The title compound was prepared according to General Procedure VI. Mobile phase: H2O/MeOH (30:$100\%$). Yield: 87 mg ($89\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.57 (s, 1H), 8.47 (dt, $J = 7.0$, 1.1 Hz, 1H), 8.41 (q, $J = 4.4$ Hz, 1H), 7.72–7.67 (m, 3H), 7.55 (d, $J = 8.2$ Hz, 1H), 7.47–7.41 (m, 4H), 7.39 (ddd, $J = 9.1$, 6.7, 1.2 Hz, 1H), 7.00 (td, $J = 6.8$, 1.2 Hz, 1H), 5.75 (s, 2H), 2.77 (d, $J = 4.6$ Hz, 3H). 13C NMR (101 MHz, DMSO): δ 166.58, 144.96, 142.43, 137.62, 136.35, 135.20, 133.01, 132.89, 130.40, 130.24, 129.86, 129.60, 128.75, 128.44, 126.31, 125.94, 125.40, 117.13, 113.38, 111.57, 52.22, 26.17. HRMS: calcd for [M + H], 477.09919; found, 477.09927. ## 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-N,N-dimethylbenzamide (41) The title compound was prepared according to General Procedure VI. Mobile phase: H2O/MeOH (30:$100\%$). Yield: 228 mg ($95\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.54 (s, 1H), 8.46 (dt, $J = 6.9$, 1.2 Hz, 1H), 8.31 (s, 1H), 7.71–7.66 (m, 3H), 7.58 (d, $J = 8.3$ Hz, 1H), 7.43 (dd, $J = 8.6$, 2.2 Hz, 3H), 7.38 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 7.35 (d, $J = 2.2$ Hz, 1H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H), 5.76 (s, 2H), 3.02 (s, 3H), 2.75 (s, 3H). 13C NMR (101 MHz, DMSO): δ 166.79, 144.94, 142.44, 136.80, 136.38, 135.78, 133.01, 132.85, 130.03, 130.02, 129.57, 129.06, 128.70, 127.44, 126.26, 125.92, 125.37, 117.11, 113.34, 111.55, 52.23, 37.65, 34.22. HRMS: calcd for [M + H], 491.11484; found, 491.11493. ## 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-N-methoxy-N-methylbenzamide (42) The title compound was prepared according to General Procedure VI with a minor modification. 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-benzoic acid (528 mg, 1.13 mmol) was dissolved in dry DCM (8 mL) and cooled in an ice bath. Oxalyl chloride (0.2 mL, 2 equiv) was added followed by a catalytic amount of DMF. The reaction mixture was stirred at 25 °C overnight. The solvent was evaporated, and crude acyl chloride was used in the next step without further purification. Crude acyl chloride was dissolved in dry DCM (10 mL), and N,O-dimethylhydroxylamine hydrochloride (112 mg, 1 equiv) was added followed by TEA (0.32 mL, 2 equiv). The reaction mixture was stirred at 25 °C for 3 h. Then the mixture was diluted with DCM, washed with water, and purified by flash column chromatography, mobile phase petrol ether/EtOAc (60:$100\%$). Yield: 521 ($91\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.55 (s, 1H), 8.50–8.39 (m, 1H), 7.69 (dd, $J = 8.7$, 2.0 Hz, 3H), 7.58 (d, $J = 8.1$ Hz, 2H), 7.48–7.40 (m, 4H), 7.38 (ddd, $J = 9.0$, 6.7, 1.3 Hz, 1H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H), 5.78 (s, 2H), 3.39 (s, 3H), 3.29 (s, 3H). 13C NMR (101 MHz, DMSO): δ 166.91, 144.93, 142.44, 136.35, 135.93, 135.22, 133.01, 132.85, 130.22, 129.83, 129.57, 129.40, 128.69, 127.15, 126.24, 125.90, 125.34, 117.11, 113.32, 111.55, 61.20, 52.22, 31.99. HRMS: calcd for [M + H], 507.10976; found, 507.10983. ## 1-(2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-phenyl)ethan-1-one (43) The title compound was prepared according to General Procedure VII. Mobile phase: H2O/MeOH (30:$100\%$). Yield: 68 mg ($72\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.54 (s, 1H), 8.47 (dt, $J = 6.9$, 1.1 Hz, 1H), 7.73–7.67 (m, 4H), 7.61 (d, $J = 8.3$ Hz, 1H), 7.50 (dd, $J = 8.3$, 2.2 Hz, 1H), 7.45–7.41 (m, 2H), 7.38 (ddd, $J = 9.0$, 6.8, 1.3 Hz, 1H), 7.00 (td, $J = 6.8$, 1.1 Hz, 1H), 5.78 (s, 2H), 2.59 (s, 3H). 13C NMR (101 MHz, DMSO): δ 199.81, 144.92, 142.46, 139.01, 136.36, 135.53, 133.01, 132.82, 131.95, 131.18, 129.76, 129.62, 129.14, 128.68, 126.25, 125.89, 125.38, 117.11, 113.32, 111.56, 52.20, 30.64. HRMS: calcd for [M + H], 462.08829; found, 462.08840. ## 1-(2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)phenyl)propan-1-one (44) The title compound was prepared according to General Procedure VII. Mobile phase: H2O/MeOH (30:$100\%$). Yield: 51 mg ($69\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.54 (s, 1H), 8.49–8.42 (m, 1H), 7.72–7.67 (m, 3H), 7.65 (d, $J = 2.1$ Hz, 1H), 7.60 (d, $J = 8.3$ Hz, 1H), 7.50–7.46 (m, 1H), 7.44–7.36 (m, 3H), 7.00 (td, $J = 6.8$, 1.1 Hz, 1H), 5.77 (s, 2H), 2.93 (q, $J = 7.2$ Hz, 2H), 1.12–1.02 (m, 3H). 13C NMR (101 MHz, DMSO): δ 203.21, 144.92, 142.44, 139.56, 136.35, 135.51, 133.01, 132.82, 131.52, 130.94, 129.60, 129.30, 128.67, 128.46, 126.26, 125.91, 125.38, 117.11, 113.32, 111.56, 52.23, 35.75, 8.03. HRMS: calcd for [M + H], 476.10394; found, 476.10407. ## 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)benzaldehyde (45) The title compound was prepared according to General Procedure VII with LAH. Crystalized from ACN. Yield: 158 mg ($75\%$). 1H NMR (401 MHz, DMSO-d6): δ 10.34 (s, 1H), 8.56 (s, 1H), 8.47 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.83 (t, $J = 1.4$ Hz, 1H), 7.72–7.67 (m, 5H), 7.45–7.41 (m, 2H), 7.38 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 6.99 (td, $J = 6.8$, 1.3 Hz, 1H), 5.83 (s, 2H). 13C NMR (101 MHz, DMSO): δ 189.72, 144.93, 142.46, 136.39, 136.33, 136.13, 135.17, 133.00, 132.83, 132.36, 131.43, 129.59, 128.79, 128.70, 126.24, 125.95, 125.35, 117.11, 113.33, 111.52, 52.06. HRMS: calcd for [M + H], 448.07264; found, 448.07224. ## (E)-2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)benzaldehyde Oxime (46) 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)benzaldehyde [45] was dissolved in DCM, and NH2OH·HCl (1.2 equiv) was added, followed by the addition of TEA (1.2 equiv). The reaction mixture was stirred at 25 °C and washed with water, and the organic phase was dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography. Mobile phase petrol ether/EtOAc (50:$100\%$). Yield: 75 mg ($90\%$). 1H NMR (401 MHz, DMSO-d6): δ 11.75 (s, 1H), 8.51 (s, 1H), 8.46 (dt, $J = 7.0$, 1.2 Hz, 1H), 8.36 (s, 1H), 7.78 (d, $J = 2.2$ Hz, 1H), 7.71–7.66 (m, 4H), 7.56 (d, $J = 8.3$ Hz, 1H), 7.43–7.35 (m, 4H), 7.00 (td, $J = 6.8$, 1.3 Hz, 1H), 5.77 (s, 2H). 13C NMR (101 MHz, DMSO): δ 145.22, 144.72, 142.75, 136.69, 136.00, 133.29, 133.13, 132.37, 131.02, 130.78, 130.63, 129.89, 128.97, 126.54, 126.35, 126.16, 125.63, 117.41, 113.67, 111.85, 52.64. HRMS: calcd for [M + H], 463.08354; found, 463.08301. ## (2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-phenyl)methanol (47) Methyl 2-chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)-methyl)benzoate (160 mg, 0.334 mmol) was dissolved in dry THF (6 mL), cooled in an ice bath, and degassed. A flask was refilled with argon, and LAH (0.35 mL, 1 equiv) was added. A reaction mixture was allowed to warm to 25 °C and stirred overnight. After the completion of the reaction, the mixture was carefully quenched with ice and extracted with EtOAc. Combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography (cyclohexane/EtOAc 40:$70\%$). Yield: 85 mg ($57\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.52 (s, 1H), 8.46 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.73–7.67 (m, 4H), 7.51 (d, $J = 2.3$ Hz, 1H), 7.47–7.41 (m, 3H), 7.38 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 7.30–7.26 (m, 1H), 7.00 (td, $J = 6.8$, 1.2 Hz, 1H), 5.76 (s, 2H), 5.51 (t, $J = 5.4$ Hz, 1H), 4.57 (dt, $J = 5.4$, 0.9 Hz, 2H). 13C NMR (101 MHz, DMSO): δ 144.93, 142.41, 140.22, 136.29, 135.16, 133.00, 132.82, 130.72, 129.59, 129.31, 128.71, 127.79, 127.22, 126.24, 125.87, 125.33, 117.11, 113.34, 111.60, 60.22, 52.73. HRMS: calcd for [M + H], 450.08829; found, 450.08835. ## 3-(1-(4-Chloro-3-(methoxymethyl)benzyl)-1H-1,2,3-triazol-4-yl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridine (48) (2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)phenyl)methanol (85 mg, 0.188 mL) was dissolved in dry THF (4 mL) and NaH (8 mg, 1.1 equiv) was added. After 10 min, CH3I (0.013 mL, 1 equiv) was added and the mixture was stirred at 25 °C overnight. The mixture was purified by flash column chromatography (cyclohexane/EtOAc 30:$80\%$). Yield: 35 mg ($40\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.51 (s, 1H), 8.46 (dt, $J = 7.0$, 1.3 Hz, 1H), 7.71–7.65 (m, 3H), 7.51–7.45 (m, 2H), 7.40 (dd, $J = 8.6$, 2.0 Hz, 2H), 7.35 (d, $J = 1.3$ Hz, 0H), 7.32 (dd, $J = 8.2$, 2.3 Hz, 1H), 6.99 (td, $J = 6.8$, 1.3 Hz, 1H), 5.75 (s, 2H), 4.48 (s, 2H), 3.36 (d, $J = 1.8$ Hz, 3H, water overlapping). 13C NMR (101 MHz, DMSO): δ 144.92, 142.42, 136.41, 136.33, 135.26, 133.01, 132.81, 131.72, 129.69, 129.59, 128.71, 128.64, 128.46, 126.20, 125.79, 125.30, 117.10, 113.31, 111.60, 70.77, 58.30, 52.58. HRMS: calcd for [M + H], 464.10394; found, 464.10396. ## 3-(1-(4-Chloro-3-nitrobenzyl)-1H-1,2,3-triazol-4-yl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridine (49) The title compound was prepared according to General Procedure IV (Scheme 1). Mobile phase petrol ether/EtOAc (40:$100\%$). Yield: 835 mg ($83\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.58 (s, 1H), 8.50 (dt, $J = 6.9$, 1.2 Hz, 1H), 8.15 (d, $J = 2.1$ Hz, 1H), 7.82 (d, $J = 8.4$ Hz, 1H), 7.73–7.65 (m, 4H), 7.43–7.38 (m, 2H), 7.36 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 6.98 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 147.65, 144.93, 142.50, 136.94, 136.53, 133.62, 132.99, 132.85, 132.31, 129.63, 128.62, 126.15, 125.94, 125.41, 125.39, 125.10, 117.06, 113.23, 111.50, 51.70. HRMS: calcd for [M – H], 465.06281; found, 465.06250. ## 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-aniline (50) 3-(1-(4-Chloro-3-nitrobenzyl)-1H-1,2,3-triazol-4-yl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridine [49] was dissolved in MeOH and AcOH (7 equiv), and Fe (3.5 equiv) was added. The reaction mixture was stirred at reflux until the completion of the reaction. After cooling to 25 °C, the mixture was extracted with EtOAc, washed with a NaHCO3 solution, and the organic phase was dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography. Mobile phase cyclohexane/EtOAc (40:$100\%$). Yield: 490 mg ($79\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.46 (dt, $J = 7.0$, 1.2 Hz, 1H), 8.44 (s, 1H), 7.73–7.66 (m, 3H), 7.44–7.39 (m, 2H), 7.37 (ddd, $J = 9.1$, 6.7, 1.2 Hz, 1H), 7.21 (d, $J = 8.2$ Hz, 1H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H), 6.71 (d, $J = 2.1$ Hz, 1H), 6.52–6.47 (m, 1H), 5.59 (s, 2H), 5.50 (s, 2H). 13C NMR (101 MHz, DMSO): δ 145.11, 144.88, 142.37, 136.29, 135.75, 132.96, 132.81, 132.80, 129.65, 129.60, 129.49, 128.65, 128.61, 126.19, 126.14, 125.74, 125.33, 117.07, 116.94, 116.02, 114.31, 113.30, 111.65, 52.95. HRMS: calcd for [M + H], 435.08863; found, 435.08815. ## N-(2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-phenyl)acetamide (51) 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)-aniline [50] was dissolved in dioxane and Ac2O (1.5 equiv) was added followed by pyridine (1.5 equiv). The reaction mixture was stirred at 25 °C overnight. After the completion of the reaction, the mixture was evaporated, diluted with EtOAc, and washed with water. The organic phase was dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography. Mobile phase cyclohexane/EtOAc (30:$100\%$). Yield: 75 mg ($90\%$). 1H NMR (401 MHz, DMSO-d6): δ 9.58 (s, 1H), 8.54 (s, 1H), 8.46 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.73–7.67 (m, 4H), 7.54 (d, $J = 8.3$ Hz, 1H), 7.46–7.41 (m, 2H), 7.38 (ddd, $J = 9.0$, 6.7, 1.3 Hz, 1H), 7.17 (dd, $J = 8.3$, 2.2 Hz, 1H), 6.99 (td, $J = 6.8$, 1.2 Hz, 1H), 5.74 (s, 2H), 2.11 (s, 3H). 13C NMR (101 MHz, DMSO): δ 169.27, 145.22, 142.68, 136.54, 135.83, 135.80, 133.25, 133.11, 130.25, 129.85, 129.00, 127.90, 126.58, 126.28, 125.78, 125.66, 125.41, 117.35, 114.34, 113.64, 111.84, 52.79, 23.86. HRMS: calcd for [M + H], 477.09919; found, 477.09866. ## 2-Chloro-5-((4-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)-1H-1,2,3-triazol-1-yl)methyl)benzoyl Chloride (52) The title compound was prepared according to General Procedure V and used in the next step without further purification. ## 3-(4-Bromophenyl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridine (53) 2-(4-Chlorophenyl)-3-iodoimidazo[1,2-a]pyridine (300 mg, 0.846 mmol) was combined with (4-bromophenyl)boronic acid (1 equiv) and diluted with dioxane/H2O mixture (4:1; 8 mL). Na2CO3 (3 equiv) was added, and the mixture was degassed and refilled with argon. Pd(dppf)Cl2·DCM (5 mol %) was added, and the mixture was degassed again and refilled with argon. The mixture was stirred at 95 °C overnight (Scheme 8). After the completion of the reaction, the mixture was diluted with water and extracted with EtOAc. The organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography. Mobile phase: cyclohexane/EtOAc (20:$60\%$). Yield: 193 mg ($59\%$). 13C NMR (101 MHz, DMSO): δ 144.73, 140.86, 133.34, 133.20, 133.17, 132.72, 129.61, 129.22, 128.92, 128.75, 126.10, 124.32, 123.04, 117.42, 113.42. ## 2-(4-Chlorophenyl)-3-(6-chloropyridin-3-yl)imidazo[1,2-a]pyridine (54) 2-(4-Chlorophenyl)-3-iodoimidazo[1,2-a]pyridine (300 mg, 0.846 mmol) was combined with (6-chloropyridin-3-yl)boronic acid (1.3 equiv) and diluted with a dioxane/H2O mixture (4:1; 8 mL). Na2CO3 (3 equiv) was added, and the mixture was degassed and refilled with argon. Pd(dppf)Cl2·DCM (5 mol %) was added and the mixture was degassed again and refilled with argon. The mixture was stirred at 95 °C overnight (Scheme 8). After the completion of the reaction, the mixture was diluted with water and extracted with EtOAc. The organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography. Mobile phase: cyclohexane/EtOAc (15:$60\%$). Yield: 183 mg ($63\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.53 (dd, $J = 2.5$, 0.8 Hz, 1H), 8.17 (dt, $J = 6.9$, 1.2 Hz, 1H), 8.06 (dd, $J = 8.2$, 2.5 Hz, 1H), 7.74 (dd, $J = 8.2$, 0.8 Hz, 1H), 7.68 (dt, $J = 9.1$, 1.2 Hz, 1H), 7.58–7.52 (m, 2H), 7.42–7.33 (m, 3H), 6.93 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 151.85, 151.07, 145.19, 142.40, 142.01, 133.04, 132.95, 129.75, 129.07, 126.53, 125.76, 125.37, 124.70, 117.40, 116.95, 113.59. ## 3-(4-Benzylphenyl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridine (55) 3-(4-Bromophenyl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridine [53] was dissolved in anhydrous THF (6 mL) under an argon atmosphere. Pd2dba3 (5 mol %) was added followed by the addition of XantPhos (15 mol %). To the mixture, a benzylzinc bromide solution 0.5 M in THF (2 equiv) was added and the mixture was stirred at 60 °C overnight (Scheme 8). The mixture was evaporated to a minimal volume, adsorbed onto silica, and purified by flash column chromatography followed by reverse-phase flash column chromatography. Mobile phase: cyclohexane/EtOAc (30:$100\%$) and H2O/ACN (30:$100\%$). Yield: 143 mg ($72\%$). 1H NMR (401 MHz, DMSO-d6): δ 7.95 (d, $J = 6.8$ Hz, 1H), 7.63 (d, $J = 9.0$ Hz, 1H), 7.61–7.55 (m, 2H), 7.42 (d, $J = 8.0$ Hz, 2H), 7.37 (d, $J = 8.2$ Hz, 2H), 7.35–7.25 (m, 6H), 7.24–7.17 (m, 1H), 6.83 (td, $J = 6.8$, 1.2 Hz, 1H), 4.04 (s, 2H). 13C NMR (101 MHz, DMSO): δ 144.17, 142.47, 140.85, 140.15, 133.33, 132.22, 130.77, 130.13, 129.17, 129.03, 128.71, 128.50, 126.78, 126.30, 125.51, 123.96, 120.97, 117.06, 112.93, 66.52. ## 3-(6-Benzylpyridin-3-yl)-2-(4-chlorophenyl)imidazo[1,2-a]pyridine (56) 2-(4-Chlorophenyl)-3-(6-chloropyridin-3-yl)imidazo[1,2-a]pyridine [54] was dissolved in anhydrous THF (6 mL) under an argon atmosphere. Pd2dba3 (5 mol %) was added, followed by the addition of XantPhos (15 mol %). To the mixture, a benzylzinc bromide solution 0.5 M in THF (2 equiv) was added and the mixture was stirred at 60 °C overnight (Scheme 8). The mixture was evaporated to a minimal volume, adsorbed onto silica, and purified by flash column chromatography followed by reverse-phase flash column chromatography. Mobile phase: cyclohexane/EtOAc (30:$100\%$) and H2O/ACN (30:$100\%$). Yield: 158 mg ($75\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.57 (dd, $J = 2.3$, 0.9 Hz, 1H), 8.06 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.89 (dd, $J = 8.0$, 2.3 Hz, 1H), 7.67 (dt, $J = 9.1$, 1.1 Hz, 1H), 7.57–7.52 (m, 2H), 7.48 (dd, $J = 8.1$, 0.9 Hz, 1H), 7.39–7.30 (m, 7H), 7.23 (ddt, $J = 8.6$, 6.2, 1.8 Hz, 1H), 6.88 (td, $J = 6.8$, 1.2 Hz, 1H), 4.21 (s, 2H). 13C NMR (101 MHz, DMSO): δ 161.60, 151.00, 144.97, 141.61, 139.81, 139.36, 133.35, 132.78, 129.65, 129.58, 128.97, 126.78, 126.27, 124.52, 124.11, 123.37, 118.26, 117.41, 113.49, 66.82. ## 5-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)pyridin-2-amine (57) 2-(4-Chlorophenyl)-3-iodoimidazo[1,2-a]pyridine (0.2 g, 0.56 mmol) was combined with 2-aminopyridine-5-boronic acid pinacol ester (1.1 equiv) and dissolved in dioxane (8 mL) followed by a solution of Na2CO3 in 2 mL of H2O. A reaction mixture was degassed and refilled with argon. Pd(dppf)Cl2 (5 mol %) was added, and the mixture was degassed again and refilled with argon. The reaction mixture was stirred at 90 °C overnight (Scheme 9). After cooling to 25 °C, the mixture was diluted with water and extracted with EtOAc. Combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography; mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 100 mg ($53\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.00 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.96 (d, $J = 2.3$ Hz, 1H), 7.71–7.65 (m, 2H), 7.62 (dt, $J = 9.1$, 1.2 Hz, 1H), 7.47 (dd, $J = 8.5$, 2.4 Hz, 1H), 7.44–7.37 (m, 2H), 7.30 (ddd, $J = 9.0$, 6.7, 1.3 Hz, 1H), 6.88 (td, $J = 6.8$, 1.2 Hz, 1H), 6.64 (d, $J = 8.5$ Hz, 1H), 6.38 (s, 2H). 13C NMR (101 MHz, DMSO): δ 160.23, 149.97, 144.19, 140.26, 139.39, 133.53, 132.09, 128.97, 128.58, 125.43, 124.19, 119.37, 117.01, 112.82, 112.32, 108.79. HRMS: calcd for [M + H], 321.0907; found, 321.0909. ## 2-(4-Chlorophenyl)-3-(6-methoxypyridin-3-yl)imidazo[1,2-a]pyridine (58) 2-(4-Chlorophenyl)-3-iodoimidazo[1,2-a]pyridine (0.2 g, 0.56 mmol) was combined with (6-methoxypyridin-3-yl)boronic acid (1.1 equiv) and dissolved in dioxane (8 mL), followed by a solution of Na2CO3 in 2 mL of H2O. The reaction mixture was degassed and refilled with argon. Pd(dppf)Cl2 (5 mol %) was added, and the mixture was degassed again and refilled with argon. The reaction mixture was stirred at 90 °C overnight. After cooling to 25 °C, the mixture was diluted with water and extracted with EtOAc. Combined organic phases were dried over sodium sulfate and evaporated. The residue was purified by flash column chromatography; mobile phase petrol ether/EtOAc (20:$70\%$). Yield: 100 mg ($53\%$). 1H NMR (401 MHz, DMSO-d6): δ 8.29 (dd, $J = 2.5$, 0.8 Hz, 1H), 8.03 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.85 (dd, $J = 8.5$, 2.4 Hz, 1H), 7.66 (dt, $J = 9.0$, 1.2 Hz, 1H), 7.63–7.57 (m, 2H), 7.42–7.36 (m, 2H), 7.33 (ddd, $J = 9.0$, 6.7, 1.3 Hz, 1H), 7.05 (dd, $J = 8.6$, 0.8 Hz, 1H), 6.90 (td, $J = 6.8$, 1.3 Hz, 1H), 3.95 (s, 3H), 13C NMR (101 MHz, DMSO): δ 164.28, 149.39, 144.77, 142.09, 141.25, 133.43, 132.65, 129.46, 128.97, 126.10, 124.57, 118.88, 118.25, 117.33, 113.36, 112.12, 53.94. HRMS: calcd for [M + H], 336.0904; found, 336.0902. ## 5-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)pyridin-2-ol (59) 2-(4-Chlorophenyl)-3-(6-methoxypyridin-3-yl)imidazo[1,2-a]pyridine 58 was treated with 4M HCl/dioxane at 95 °C overnight. After completion of the reaction, the solvent was evaporated and the residue neutralized with a saturated NaHCO3 solution and extracted with EtOAc. The organic phase was dried over sodium sulfate, evaporated, and purified by RP-flash column chromatography. Mobile phase H2O/MeOH 15:$90\%$. Yield: 72 mg ($73\%$). 1H NMR (401 MHz, DMSO-d6): δ 12.04 (s, 1H), 8.12 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.74–7.69 (m, 2H), 7.67–7.59 (m, 2H), 7.48–7.41 (m, 3H), 7.32 (ddd, $J = 9.0$, 6.7, 1.2 Hz, 1H), 6.92 (td, $J = 6.8$, 1.2 Hz, 1H), 6.52 (dd, $J = 9.4$, 0.7 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 162.36, 144.57, 143.64, 140.96, 138.82, 133.48, 132.60, 129.35, 129.01, 126.01, 124.89, 121.73, 117.63, 117.22, 113.19, 106.40. HRMS: calcd for [M + H], 322.07417; found, 322.07397. ## 2-Chloro-5-((5-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)pyridin-2-yl)amino)benzamide (60) 5-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)pyridin-2-amine was combined with 5-bromo-2-chlorobenzamide (1 equiv) and diluted with dioxane, and the mixture was degassed and refilled with argon. To the solution, NatBuO (1.5 equiv), XantPhos (10 mol %), and Pd2dba3 (5 mol %) were added and the mixture was stirred at 100 °C overnight. The mixture was filtered through Celite, and the solvent was evaporated to a minimal volume and purified by flash chromatography followed by RP-flash column chromatography. Mobile phase EtOAc, then H2O/MeOH (20:$100\%$). Yield: 45 mg ($51\%$). 1H NMR (401 MHz, DMSO-d6): δ 9.67 (s, 1H), 8.30 (dd, $J = 2.3$, 0.8 Hz, 1H), 8.10 (dt, $J = 6.9$, 1.2 Hz, 1H), 7.90–7.82 (m, 3H), 7.71 (dd, $J = 8.6$, 2.4 Hz, 1H), 7.68–7.63 (m, 3H), 7.58–7.55 (m, 1H), 7.43–7.39 (m, 2H), 7.38 (d, $J = 8.8$ Hz, 1H), 7.33 (ddd, $J = 9.1$, 6.7, 1.3 Hz, 1H), 7.04 (dd, $J = 8.7$, 0.8 Hz, 1H), 6.91 (td, $J = 6.8$, 1.2 Hz, 1H). 13C NMR (101 MHz, DMSO): δ 168.53, 155.68, 149.29, 144.40, 140.74, 140.26, 139.88, 137.53, 133.32, 132.27, 129.89, 129.16, 128.68, 125.71, 124.33, 120.63, 119.95, 118.64, 117.93, 117.06, 115.77, 113.00, 112.08. HRMS: calcd for [M + H], 474.08829; found, 474.08786. ## 2-Chloro-5-((5-(2-(4-chlorophenyl)imidazo[1,2-a]pyridin-3-yl)pyridin-2-yl)oxy)benzamide (61) 5-(2-(4-Chlorophenyl)imidazo[1,2-a]pyridin-3-yl)pyridin-2-ol (1.5 equiv) was combined with 5-bromo-2-chlorobenzamide (50 mg, 1 equiv) followed by BPPO (2 mol %), CuI (2 mol %), and K3PO4 (2 equiv). The mixture was degassed and refilled with argon, and dry DMF (1 mL) was added. The mixture was degassed and refilled with argon and heated up to 110 °C overnight. The mixture was filtrated over Celite, dissolved in MeOH, and purified by RP-flash column chromatography. Mobile phase: H2O/CH3CN (20:$100\%$). The product was crystallized from THF. Yield: 4 mg ($6\%$). 1H NMR (500 MHz, DMSO-d6): δ 8.36 (d, $J = 6.9$ Hz, 1H), 8.05 (d, $J = 2.5$ Hz, 1H), 7.96 (s, 1H), 7.85–7.80 (m, 2H), 7.71 (s, 1H), 7.67–7.62 (m, 4H), 7.50 (dd, $J = 9.4$, 2.6 Hz, 1H), 7.48–7.43 (m, 2H), 7.33 (ddd, $J = 8.9$, 6.8, 1.2 Hz, 1H), 6.94 (td, $J = 6.8$, 1.2 Hz, 1H), 6.69 (d, $J = 9.4$ Hz, 1H). 13C NMR (126 MHz, DMSO): δ 167.32, 160.73, 144.41, 143.38, 141.21, 140.86, 139.01, 137.62, 133.12, 132.42, 130.28, 129.69, 129.48, 129.23, 128.81, 127.32, 125.87, 125.10, 121.99, 116.95, 116.82, 112.87, 107.21. HRMS: calcd for [M + H], 474.06503; found, 474.06512. ## Chemicals for Biological Experiments Compound 1 (CITCO, (6-(4-chlorophenyl)imidazo [2,1-b][1,3]thiazole-5-carbaldehyde-O-(3,4-dichloro-benzyl)oxime)), rifampicin (rif), TCPOBOP, and PK11195 were obtained from Sigma-Aldrich (St. Louis, Missouri, United States, now Merck), which is now known as Merck (Darmstadt, Germany). Phenobarbital (Luminal 200 mg/mL injection) was manufactured by Desitin Pharma spol. s.r.o. ( Prague, Czech Republic). Ligands for nuclear receptors (GW3965, thyroxin, obeticholic acid, dexamethasone, fenofibrate, GW501516, rosiglitazone, 3-methylcholanthrene, calcitriol, testosterone, and estradiol) were purchased from Sigma-Aldrich (now Merck). The prototype ligands were used at 100 nM (dexamethasone, calcitriol), 1 μM (thyroxin), or 10 μM concentrations. The compounds were dissolved in DMSO, and the final concentration of DMSO in the entire reaction mixture or cultivation media was $0.1\%$. ## Cell Culture Human hepatocellular carcinoma HepG2 and monkey fibroblast-like COS-1 cell lines were cultured as we have described before.48 All experiments were performed between passages 5–13 after thawing. CAR Knockout HepaRG and parent HepaRG cells were obtained from Sigma-Aldrich, now Merck (Darmstadt, Germany). The cell lines were cultivated and differentiated in the same manner on the 12-well plates. For each experiment, the HepaRG cells were seeded at a density of 26,600 cells/cm2 and kept in William’s medium supplemented with 5 μg/mL insulin, 50 μM hydrocortisone, $10\%$ HyClone fetal serum (GE Healthcare Life Sciences, Pittsburgh, USA). 14 days after seeding, the HepaRG cells were differentiated into hepatocyte-like cells using $1.5\%$ DMSO in culture media for another 14 days.49 LS174T, an epithelial Caucasian colon adenocarcinoma cell line, was obtained from (Merck Life Science spol. s r.o., Prague, 87060401-1VL). The line has functional PXR but very low CAR nuclear receptor expression.43 The cell line has been cultivated for induction experiments as we described before.50 ## Primary Human Hepatocytes The PHHs (human hepatocytes in monolayer-long-term cultures) were obtained from Biopredic (Rennes, France) (batch HEP220965, 45-year-old female, Caucasian; HEP220966, 53-year-old female, Caucasian; HEP220969, 78-year-old female, Caucasian; HEP220971, 46-year-old male, Caucasian, HEP220976, 73-year-old male, African, HEP220980, 84-year-old male, Caucasian). The cells were cultivated according to the manufacturer’s protocol. The PHHs were treated with selected test compounds for 24 or 48 h in the use medium at the concentrations of 1, 5, or 10 μM. Cryopreserved human hepatocytes (HJK) were purchased from BioIVT (Westbury, New York, USA). PHHs were cultured in William’s E medium (ThermoFisher Scientific, Waltham, Massachusetts, USA) supplemented with insulin (10 μg/mL)–transferrin (5.5 μg/mL)–sodium selenite (6.7 ng/mL) (Thermo Fisher Scientific), l-glutamine (2 mM)–penicillin (100 U/mL)–streptomycin (100 μg/mL) (Sigma-Aldrich), 100 nM dexamethasone, and $10\%$ fetal bovine serum (HyClone). Western blotting experiments have been performed with total cellular or tissue lysates (20 μg) with polyclonal anti-CYP2B6 (PA5-35032, dilution 1:1500), antibeta actin recombinant rabbit monoclonal antibody (MA5-32540, clone JF53-10), and CYP3A4 polyclonal antibody (PA1-343, 1:2000) (all from Thermo Fisher). For protein analysis, PHHs have been treated for 48 h. ## RT-qPCR RT-qPCR was used to examine CAR target gene expression in PHH, HepaRG cells, or in mouse liver samples. Total RNA, reverse transcription, and qPCR were performed, and mRNA expression data was analyzed as we have described before.51 All RT-qPCR experiments were performed in triplicate samples, and data are presented as fold induction to vehicle-treated cells with the same reagents we have described before.48,52 PCR TaqMan probes for murine genes have been listed in our previous reports.52 TaqMan probes for CYP3A4, CYP2C9, and CYP2B6 human genes as well as for reference genes B2M and GADPH genes were obtained from Thermo Fisher: CYP3A4 (Hs00604506_m1), CYP2C9 (Hs02383631_s1), CYP2B6 (Hs04183483_g1), GAPDH (Hs02758991_g1), and B2M (Hs00984230_m1). ## Luciferase Assays A human CAR LBD assembly assay (CAR AA) was performed according to protocols published by Carazo and Pavek with two hybrid expression constructs encoding helices 3–12 (pCAR-C/VP16) and helix 1 (pCAR-N/GAL4) parts of human CAR LBD.53 Cells were treated for 24 h with tested compounds (range of concentration from 1 nM up to 30 μM). Data are presented as relative activity (%) to compound 1 (CITCO) at 10 μM. The half-maximal effective concentration (EC50) to activate CAR in the assay was calculated from at least six points of dose–response curves using the GraphPad Prism software. Luciferase gene reporter assays to determine interactions with CAR and its variants or with PXR were performed as we have described before in HepG2 or COS-1 cells.39,48 Using these assays, the relative activation of the CAR3 variant in comparison with the activity of compound 1 (CITCO) at 1 μM concentration or the relative value of PXR activation (% of rifampicin-mediated PXR activation at 10 μM concentration) were determined (Tables 2, 4, and 7). The CYP2B6-luc reporter plasmid (originally entitled B-1.6k/PB/XREM) was kindly donated by Dr. Hongbing Wang (University of Maryland School of Pharmacy, Baltimore, MD, USA) and was used in assays with all human CAR variants as well as with the mouse Car expression vector. Expression vectors (based on pcDNA3.1+/C-(K)-DYK vector) for CAR variant 3 (CAR3, 353 AA, CloneID OHu34914, XM_005245697.4, transcript variant X4, mRNA), CAR variant 2 (CAR2, 352 AA, Clone ID OHu10438, NM_001077480.2), and CAR wild type (wtCAR, 348 AA, Clone ID OHu09315, NM_005122.4, transcript variant 3) were purchased from Genscript (Piscataway, NJ, USA). The mouse Car expression vector pCMV6-mCar (NM_009803) was obtained from OriGene Technologies, Rockville, MD, USA). Empty expression vectors were used in control experiments. In addition, other expression constructs for a ligand-activated CAR transcription variant 3 (CAR3, pTracer-CMV2-CAR3) was a kind gift from Dr. C. J. Omiecinski (Pennsylvania State University, State College, PA, USA), which was used to validate our results with the commercial construct. The CYP3A4 promoter luciferase reporter construct (p3A4-luc) and PXR expression vector for transient transfection luciferase assays have been described before.50 The p3A4-luc plasmid bears a distal XREM (−7836/-7208) and a basal promoter sequence (prPXRE, −362/+53) from the CYP3A4 gene promoter region. Transient transfection experiments with various nuclear receptor-responsive luciferase assays have been performed as we described before with the same protocol and plasmids.54,55 ## CYP Enzymatic Activity Assays Human recombinant CYP3A4, CYP2B6, and CYP1A2 enzymes expressed from cDNA using baculovirus-infected insect cells with human CYP450 reductase and cytochrome b5 in a microsomal fraction (CYP450-Glo CYP3A4 Assay, CYP450-Glo CYP2B6 Assay, and CYP450-Glo CYP1A2 Assay, Promega, Hercules, CA) were used to evaluate the interaction of compound 39 with these enzymes in vitro according to protocols we published before.56 ## TR-FRET CAR Coactivator Binding Assay The LanthaScreen TR-FRET CAR Coactivator Binding Assay Kit, goat (Thermo Fisher Scientific, Catr. No PV4836) with GST-tagged human CAR LBD and a fluorescein-labeled PGC1α coactivator peptide was used with slight modifications of the manufacturer’s protocol as we have reported before.53 The half-maximal effective concentration (EC50) to activate CAR LBD in the assay was calculated from at least six points (range of 10 pM to 10 μM) from at least two experiments ($$n = 2$$–3) using the GraphPad Prism software. ## Translocation Assay Nuclear translocation of pEGFP-hCAR + Ala chimera in COS-1 SV40-transformed African green monkey kidney cells was performed as we have described before with the construct generated in the same report.48 The method is a modification of the method originally described by Chen et al.42 ## Animal Experiments Humanized PXR-CAR-CYP3A$\frac{4}{3}$A7 mice (model 11585) were obtained from Taconic (Rensselaer, NY) and kept in a temperature-controlled and light-controlled facility with a 12 h light–dark cycling. All animals had free access to a commercially available laboratory chow diet (Velaz, Prague, Czech Republic). Male 9–14 week-old animals ($$n = 4$$ per group) were randomized into four groups (control; compound 39 1 mg/kg; compound 39 10 mg/kg; compound 1 10 mg/kg), and these compounds were administered as a single application intraperitoneally in a $5\%$ glycerol formulation in saline. Animals were sacrificed 36 h after the administration, and livers were removed, weighted, and snap-frozen in liquid nitrogen for further total RNA isolation. All animal studies were performed in accordance with the European Directive $\frac{86}{609}$/EEC, and they were approved by the Czech Central Commission for Animal Welfare. ## Human and Mouse Plasma Protein Binding, Metabolic Stability in Human or Mouse Liver Microsomes, and Human Liver S9 Fraction Protocols for the plasma protein binding assay and metabolic stability testing in human and mouse liver microsomes or the S9 fraction are described in the Supporting Information, Chapters 6 and 7. ## Pharmacokinetic Study after Single-Dose Application PK studies were performed in C57BL/6N male mice after 10 mg/kg application of compound 39 via either i.v. or gavage application ($$n = 4$$ per group) using HPLC-MS/MS analysis. Blood samples were taken in the following intervals: 10, 120, 240, 480, 720, and 1440 min. Detailed protocols and PK parameters calculation are described in the Supporting Information, Chapter 8. ## Genotoxicity Testing and 7 Day Oral Toxicity Study in Rats The protocols for the assays are described in the Supporting Information, Chapters 9 and 11. ## hERG Fluorescence Polarization Assay The hERG fluorescence polarization assay was performed as described in the Supporting Information, Chapter 10. ## Molecular Modeling Receptor and Ligand Preparation: The crystal structure of the hCAR model was retrieved from the RCSB Protein Data Bank (www.rcsb.org) (PDB code: 1XVP).15 All ligands for docking were drawn using Maestro (2020.2) and prepared using LigPrep to generate the three-dimensional conformation, adjust the protonation state to physiological pH (7.4), and calculate the partial atomic charges, with the force-field OPLS3e. We employed a standard docking to accommodate the compounds 37, 39, 40, and 48 within the CAR’s LBD (PDB ID: 1XVP; resolution: 2.0 Å, cocrystallized with compound 1,15 amino acid numbering follows the crystal structure), using Glide.57 Ligands were docked within a grid around 12 Å from the centroid of the cocrystallized ligand generating 10 poses per ligand. To validate the docking obtained for test ligands, and also to evaluate the capability of the docking algorithm to locate the ligands within the LBD, we redocked the cocrystal ligand (compound 1, a full agonist) inside the CAR LBD. Next, the seven systems (four test compounds plus compound 1) were prepared and minimized by adding hydrogens, adjusting the protonation states of amino acids, and fixing missing side-chain atoms and protein loops using Maestro PrepWizard 2020.2. The molecular dynamics simulation protocol and respective analyses can be found in the Supporting Information, Chapter 2. For each ligand, simulations of five 1 μs independent replicas were carried out, resulting in 25 μs worth of simulations for all five systems. ## Statistical Analysis Data are presented as the means and SD from at least three independent experiments ($$n = 3$$). A one-way analysis of variance (ANOVA) with Dunnett’s post hoc test was applied. GraphPad Prism ver. 9.3.1. Software (GraphPad Software, Inc., San Diego, CA, United States) was used to perform statistical analysis. EC50 indicates the xenobiotic concentration required to achieve half-maximum activation, and relative Emax represents the overall maximal calculated activation produced by the tested compound (i.e., maximal efficacy). The activities of compound 1 and rifampicin at 10 μM were set to be $100\%$ in the dose–response calculations. IC50 represents the half-maximal inhibitory concentration in the viability MTT assay or in cytochrome P450 inhibition assays. A p-value of <0.05 was considered to be statistically significant. ## References 1. Forman B. M., Tzameli I., Choi H. S., Chen J., Simha D., Seol W., Evans R. M., Moore D. D.. Androstane metabolites bind to and deactivate the nuclear receptor CAR-beta. Nature (1998) 395 612-615. DOI: 10.1038/26996 2. Moore L. B., Parks D. J., Jones S. A., Bledsoe R. K., Consler T. G., Stimmel J. B., Goodwin B., Liddle C., Blanchard S. G., Willson T. M., Collins J. L., Kliewer S. A.. Orphan nuclear receptors constitutive androstane receptor and pregnane X receptor share xenobiotic and steroid ligands. J. Biol. Chem. (2000) 275 15122-15127. DOI: 10.1074/jbc.m001215200 3. Mackowiak B., Hodge J., Stern S., Wang H.. The Roles of Xenobiotic Receptors: Beyond Chemical Disposition. Drug Metab. Dispos. (2018) 46 1361. DOI: 10.1124/dmd.118.081042 4. Dong B., Saha P. K., Huang W., Chen W., Abu-Elheiga L. A., Wakil S. J., Stevens R. D., Ilkayeva O., Newgard C. B., Chan L., Moore D. D.. Activation of nuclear receptor CAR ameliorates diabetes and fatty liver disease. Proc. Natl. Acad. Sci. U. S. A. (2009) 106 18831-18836. DOI: 10.1073/pnas.0909731106 5. Gao J., He J., Zhai Y., Wada T., Xie W.. The constitutive androstane receptor is an anti-obesity nuclear receptor that improves insulin sensitivity. J. Biol. Chem. (2009) 284 25984-25992. DOI: 10.1074/jbc.m109.016808 6. Gao J., Xie W.. Targeting xenobiotic receptors PXR and CAR for metabolic diseases. Trends Pharmacol. Sci. (2012) 33 552-558. DOI: 10.1016/j.tips.2012.07.003 7. Jiang M., Xie W.. Role of the constitutive androstane receptor in obesity and type 2 diabetes: a case study of the endobiotic function of a xenobiotic receptor. Drug Metab. Rev. (2013) 45 156-163. DOI: 10.3109/03602532.2012.743561 8. Molnár F., Kublbeck J., Jyrkkarinne J., Prantner V., Honkakoski P.. An update on the constitutive androstane receptor (CAR). Drug Metabol. Drug Interact. (2013) 28 79-93. DOI: 10.1515/dmdi-2013-0009 9. Režen T., Tamasi V., Lovgren-Sandblom A., Bjorkhem I., Meyer U. A., Rozman D.. Effect of CAR activation on selected metabolic pathways in normal and hyperlipidemic mouse livers. BMC Genom. (2009) 10 384. DOI: 10.1186/1471-2164-10-384 10. Marmugi A., Lukowicz C., Lasserre F., Montagner A., Polizzi A., Ducheix S., Goron A., Gamet-Payrastre L., Gerbal-Chaloin S., Pascussi J. M., Moldes M., Pineau T., Guillou H., Mselli-Lakhal L.. Activation of the Constitutive Androstane Receptor induces hepatic lipogenesis and regulates Pnpla3 gene expression in a LXR-independent way. Toxicol. Appl. Pharmacol. (2016) 303 90-100. DOI: 10.1016/j.taap.2016.05.006 11. Breuker C., Moreau A., Lakhal L., Tamasi V., Parmentier Y., Meyer U., Maurel P., Lumbroso S., Vilarem M. J., Pascussi J. M.. Hepatic expression of thyroid hormone-responsive spot 14 protein is regulated by constitutive androstane receptor (NR1I3). Endocrinology (2010) 151 1653-1661. DOI: 10.1210/en.2009-1435 12. Maglich J. M., Lobe D. C., Moore J. T.. The nuclear receptor CAR (NR1I3) regulates serum triglyceride levels under conditions of metabolic stress. J. Lipid Res. (2009) 50 439-445. DOI: 10.1194/jlr.m800226-jlr200 13. Baskin-Bey E. S., Anan A., Isomoto H., Bronk S. F., Gores G. J.. Constitutive androstane receptor agonist, TCPOBOP, attenuates steatohepatitis in the methionine choline-deficient diet-fed mouse. World J. Gastroenterol. (2007) 13 5635-5641. DOI: 10.3748/wjg.v13.i42.5635 14. Tschuor C., Kachaylo E., Limani P., Raptis D. A., Linecker M., Tian Y., Herrmann U., Grabliauskaite K., Weber A., Columbano A., Graf R., Humar B., Clavien P. A.. Constitutive androstane receptor (Car)-driven regeneration protects liver from failure following tissue loss. J. Hepatol. (2016) 65 66-74. DOI: 10.1016/j.jhep.2016.02.040 15. Xu R. X., Lambert M. H., Wisely B. B., Warren E. N., Weinert E. E., Waitt G. M., Williams J. D., Collins J. L., Moore L. B., Willson T. M., Moore J. T.. A structural basis for constitutive activity in the human CAR/RXRalpha heterodimer. Mol. Cell (2004) 16 919-928. DOI: 10.1016/j.molcel.2004.11.042 16. Ingraham H. A., Redinbo M. R.. Orphan nuclear receptors adopted by crystallography. Curr. Opin. Struct. Biol. (2005) 15 708-715. DOI: 10.1016/j.sbi.2005.10.009 17. Mackowiak B., Wang H.. Mechanisms of xenobiotic receptor activation: Direct vs. indirect. Biochim. Biophys. Acta (2016) 1859 1130-1140. DOI: 10.1016/j.bbagrm.2016.02.006 18. Chai S. C., Cherian M. T., Wang Y. M., Chen T.. Small-molecule modulators of PXR and CAR. Biochim. Biophys. Acta (2016) 1859 1141-1154. DOI: 10.1016/j.bbagrm.2016.02.013 19. Ross J., Plummer S. M., Rode A., Scheer N., Bower C. C., Vogel O., Henderson C. J., Wolf C. R., Elcombe C. R.. Human constitutive androstane receptor (CAR) and pregnane X receptor (PXR) support the hypertrophic but not the hyperplastic response to the murine nongenotoxic hepatocarcinogens phenobarbital and chlordane in vivo. Toxicol. Sci. (2010) 116 452-466. DOI: 10.1093/toxsci/kfq118 20. Chai S. C., Lin W., Li Y., Chen T.. Drug discovery technologies to identify and characterize modulators of the pregnane X receptor and the constitutive androstane receptor. Drug Discov. Today (2019) 24 906-915. DOI: 10.1016/j.drudis.2019.01.021 21. Küblbeck J., Jyrkkärinne J., Molnár F., Kuningas T., Patel J., Windshügel B., Nevalainen T., Laitinen T., Sippl W., Poso A., Honkakoski P.. New in vitro tools to study human constitutive androstane receptor (CAR) biology: discovery and comparison of human CAR inverse agonists. Mol. Pharm. (2011) 8 2424-2433. DOI: 10.1021/mp2003658 22. Maglich J. M., Parks D. J., Moore L. B., Collins J. L., Goodwin B., Billin A. N., Stoltz C. A., Kliewer S. A., Lambert M. H., Willson T. M., Moore J. T.. Identification of a novel human constitutive androstane receptor (CAR) agonist and its use in the identification of CAR target genes. J. Biol. Chem. (2003) 278 17277-17283. DOI: 10.1074/jbc.m300138200 23. Küblbeck J., Laitinen T., Jyrkkärinne J., Rousu T., Tolonen A., Abel T., Kortelainen T., Uusitalo J., Korjamo T., Honkakoski P., Molnár F.. Use of comprehensive screening methods to detect selective human CAR activators. Biochem. Pharmacol. (2011) 82 1994-2007. DOI: 10.1016/j.bcp.2011.08.027 24. Lin W., Bwayi M., Wu J., Li Y., Chai S. C., Huber A. D., Chen T.. CITCO Directly Binds to and Activates Human Pregnane X Receptor. Mol. Pharmacol. (2020) 97 180-190. DOI: 10.1124/mol.119.118513 25. Hakkola J., Bernasconi C., Coecke S., Richert L., Andersson T. B., Pelkonen O.. Cytochrome P450 Induction and Xeno-Sensing Receptors Pregnane X Receptor, Constitutive Androstane Receptor, Aryl Hydrocarbon Receptor and Peroxisome Proliferator-Activated Receptor alpha at the Crossroads of Toxicokinetics and Toxicodynamics. Basic Clin. Pharmacol. Toxicol. (2018) 123 42. DOI: 10.1111/bcpt.13004 26. Tzameli I., Pissios P., Schuetz E. G., Moore D. D.. The xenobiotic compound 1,4-bis[2-(3,5-dichloropyridyloxy)]benzene is an agonist ligand for the nuclear receptor CAR. Mol. Cell. Biol. (2000) 20 2951-2958. DOI: 10.1128/mcb.20.9.2951-2958.2000 27. Li H., Chen T., Cottrell J., Wang H.. Nuclear translocation of adenoviral-enhanced yellow fluorescent protein-tagged-human constitutive androstane receptor (hCAR): a novel tool for screening hCAR activators in human primary hepatocytes. Drug Metab. Dispos. (2009) 37 1098-1106. DOI: 10.1124/dmd.108.026005 28. Mackowiak B., Li L., Lynch C., Ziman A., Heyward S., Xia M., Wang H.. High-content analysis of constitutive androstane receptor (CAR) translocation identifies mosapride citrate as a CAR agonist that represses gluconeogenesis. Biochem. Pharmacol. (2019) 168 224-236. DOI: 10.1016/j.bcp.2019.07.013 29. Imai J., Yamazoe Y., Yoshinari K.. Novel cell-based reporter assay system using epitope-tagged protein for the identification of agonistic ligands of constitutive androstane receptor (CAR). Drug Metab. Pharmacokinet. (2013) 28 290-298. DOI: 10.2133/dmpk.dmpk-12-rg-112 30. Jyrkkärinne J., Windshügel B., Rönkkö T., Tervo A. J., Küblbeck J., Lahtela-Kakkonen M., Sippl W., Poso A., Honkakoski P.. Insights into ligand-elicited activation of human constitutive androstane receptor based on novel agonists and three-dimensional quantitative structure-activity relationship. J. Med. Chem. (2008) 51 7181-7192. DOI: 10.1021/jm800731b 31. Dring A. M., Anderson L. E., Qamar S., Stoner M. A.. Rational quantitative structure-activity relationship (RQSAR) screen for PXR and CAR isoform-specific nuclear receptor ligands. Chem. Biol. Interact. (2010) 188 512-525. DOI: 10.1016/j.cbi.2010.09.018 32. Lynch C., Zhao J., Wang H., Xia M.. Quantitative High-Throughput Luciferase Screening in Identifying CAR Modulators. Methods Mol. Biol. (2016) 1473 33-42. DOI: 10.1007/978-1-4939-6346-1_4 33. Keminer O., Windshügel B., Essmann F., Lee S. M. L., Schiergens T. S., Schwab M., Burk O.. Identification of novel agonists by high-throughput screening and molecular modelling of human constitutive androstane receptor isoform 3. Arch. Toxicol. (2019) 93 2247-2264. DOI: 10.1007/s00204-019-02495-6 34. Lynch C., Pan Y., Li L., Ferguson S. S., Xia M., Swaan P. W., Wang H.. Identification of novel activators of constitutive androstane receptor from FDA-approved drugs by integrated computational and biological approaches. Pharm. Res. (2013) 30 489-501. DOI: 10.1007/s11095-012-0895-1 35. Burk O., Piedade R., Ghebreghiorghis L., Fait J. T., Nussler A. K., Gil J. P., Windshügel B., Schwab M.. Differential effects of clinically used derivatives and metabolites of artemisinin in the activation of constitutive androstane receptor isoforms. Br. J. Pharmacol. (2012) 167 666-681. DOI: 10.1111/j.1476-5381.2012.02033.x 36. Liang D., Li L., Lynch C., Diethelm-Varela B., Xia M., Xue F., Wang H.. DL5050, a Selective Agonist for the Human Constitutive Androstane Receptor. ACS Med. Chem. Lett. (2019) 10 1039-1044. DOI: 10.1021/acsmedchemlett.9b00079 37. Liang D., Li L., Lynch C., Mackowiak B., Hedrich W. D., Ai Y., Yin Y., Heyward S., Xia M., Wang H., Xue F.. Human constitutive androstane receptor agonist DL5016: A novel sensitizer for cyclophosphamide-based chemotherapies. Eur. J. Med. Chem. (2019) 179 84-99. DOI: 10.1016/j.ejmech.2019.06.031 38. Stern S., Liang D., Li L., Kurian R., Lynch C., Sakamuru S., Heyward S., Zhang J., Kareem K. A., Chun Y. W., Huang R., Xia M., Hong C. C., Xue F., Wang H.. Targeting CAR and Nrf2 improves cyclophosphamide bioactivation while reducing doxorubicin-induced cardiotoxicity in triple-negative breast cancer treatment. JCI Insight (2022) 7. DOI: 10.1172/jci.insight.153868 39. Smutny T., Nova A., Drechslerová M., Carazo A., Hyrsova L., Hrušková Z. R., Kuneš J., Pour M., Špulák M., Pavek P.. 2-(3-Methoxyphenyl)quinazoline Derivatives: A New Class of Direct Constitutive Androstane Receptor (CAR) Agonists. J. Med. Chem. (2016) 59 4601-4610. DOI: 10.1021/acs.jmedchem.5b01891 40. Li L., Chen T., Stanton J. D., Sueyoshi T., Negishi M., Wang H.. The peripheral benzodiazepine receptor ligand 1-(2-chlorophenyl-methylpropyl)-3-isoquinoline-carboxamide is a novel antagonist of human constitutive androstane receptor. Mol. Pharmacol. (2008) 74 443-453. DOI: 10.1124/mol.108.046656 41. Cherian M. T., Lin W., Wu J., Chen T.. CINPA1 is an inhibitor of constitutive androstane receptor that does not activate pregnane X receptor. Mol. Pharmacol. (2015) 87 878-889. DOI: 10.1124/mol.115.097782 42. Chen T., Tompkins L. M., Li L., Li H., Kim G., Zheng Y., Wang H.. A single amino acid controls the functional switch of human constitutive androstane receptor (CAR) 1 to the xenobiotic-sensitive splicing variant CAR3. J. Pharmacol. Exp. Ther. (2010) 332 106-115. DOI: 10.1124/jpet.109.159210 43. Burk O., Arnold K. A., Nussler A. K., Schaeffeler E., Efimova E., Avery B. A., Avery M. A., Fromm M. F., Eichelbaum M.. Antimalarial artemisinin drugs induce cytochrome P450 and MDR1 expression by activation of xenosensors pregnane X receptor and constitutive androstane receptor. Mol. Pharmacol. (2005) 67 1954-1965. DOI: 10.1124/mol.104.009019 44. Shan L., Vincent J., Brunzelle J. S., Dussault I., Lin M., Ianculescu I., Sherman M. A., Forman B. M., Fernandez E. J.. Structure of the murine constitutive androstane receptor complexed to androstenol: a molecular basis for inverse agonism. Mol. Cell (2004) 16 907-917. DOI: 10.1016/s1097-2765(04)00728-2 45. Suino K., Peng L., Reynolds R., Li Y., Cha J. Y., Repa J. J., Kliewer S. A., Xu H. E.. The nuclear xenobiotic receptor CAR: structural determinants of constitutive activation and heterodimerization. Mol. Cell (2004) 16 893-905. DOI: 10.1016/s1097-2765(04)00727-0 46. Trapani G., Franco M., Ricciardi L., Latrofa A., Genchi G., Sanna E., Tuveri F., Cagetti E., Biggio G., Liso G.. Synthesis and binding affinity of 2-phenylimidazo[1,2-alpha]pyridine derivatives for both central and peripheral benzodiazepine receptors. A new series of high-affinity and selective ligands for the peripheral type. J. Med. Chem. (1997) 40 3109-3118. DOI: 10.1021/jm970112+ 47. Palagiano F., Arenare L., Luraschi E., Caprariis P., Abignente E., D’Amico M., Filippelli W., Rossi F.. Research on heterocyclic compounds. XXXIV. Synthesis and SAR study of some imidazo[2,1-b]thiazole carboxylic and acetic acids with antiinflammatory and analgesic activities. Eur. J. Med. Chem. (1995) 30 901-909. DOI: 10.1016/0223-5234(96)88309-7 48. Skoda J., Dusek J., Drastik M., Stefela A., Dohnalova K., Chalupsky K., Smutny T., Micuda S., Gerbal-Chaloin S., Pavek P.. Diazepam Promotes Translocation of Human Constitutive Androstane Receptor (CAR) via Direct Interaction with the Ligand-Binding Domain. Cells (2020) 9 2532. DOI: 10.3390/cells9122532 49. Hyrsova L., Smutny T., Carazo A., Moravcik S., Mandikova J., Trejtnar F., Gerbal-Chaloin S., Pavek P.. The pregnane X receptor down-regulates organic cation transporter 1 (SLC22A1) in human hepatocytes by competing for (″squelching″) SRC-1 coactivator. Br. J. Pharmacol. (2016) 173 1703-1715. DOI: 10.1111/bph.13472 50. Pavek P., Pospechova K., Svecova L., Syrova Z., Stejskalova L., Blazkova J., Dvorak Z., Blahos J.. Intestinal cell-specific vitamin D receptor (VDR)-mediated transcriptional regulation of CYP3A4 gene. Biochem. Pharmacol. (2010) 79 277-287. DOI: 10.1016/j.bcp.2009.08.017 51. Skoda J., Dusek J., Drastik M., Stefela A., Dohnalova K., Chalupsky K., Smutny T., Micuda S., Gerbal-Chaloin S., Pavek P.. Diazepam Promotes Translocation of Human Constitutive Androstane Receptor (CAR) via Direct Interaction with the Ligand-Binding Domain. Cells (2020) 9 2532. DOI: 10.3390/cells9122532 52. Dusek J., Skoda J., Holas O., Horvatova A., Smutny T., Linhartova L., Hirsova P., Kucera O., Micuda S., Braeuning A., Pavek P.. Stilbene compound trans-3,4,5,4 -tetramethoxystilbene, a potential anticancer drug, regulates constitutive androstane receptor (Car) target genes, but does not possess proliferative activity in mouse liver. Toxicol. Lett. (2019) 313 1-10. DOI: 10.1016/j.toxlet.2019.05.024 53. Carazo A., Pávek P.. The Use of the LanthaScreen TR-FRET CAR Coactivator Assay in the Characterization of Constitutive Androstane Receptor (CAR) Inverse Agonists. Sensors (2015) 15 9265-9276. DOI: 10.3390/s150409265 54. Stefela A., Vrzal R., Pavek P.. (E)-7-ethylidene-lithocholic acid (7-ELCA) is a potent dual farnesoid X receptor (FXR) antagonist and GPBAR1 agonist inhibiting FXR-induced gene expression in hepatocytes and stimulating glucagon-like peptide-1 secretion from enteroendocrine cells. Front. Pharmacol. (2021) 12 1980. DOI: 10.3389/fphar.2021.713149 55. Dvořák Z., Vrzal R., Pavek P., Ulrichova J.. An evidence for regulatory cross-talk between aryl hydrocarbon receptor and glucocorticoid receptor in HepG2 cells. Physiol. Res. (2008) 57 427-435. DOI: 10.33549/a10.33549/physiolres.931090 56. Dusek J., Carazo A., Trejtnar F., Hyrsova L., Holas O., Smutny T., Micuda S., Pavek P.. Steviol, an aglycone of steviol glycoside,sweeteners, interacts with the pregnane X (PXR) and aryl hydrocarbon (AHR) receptors in detoxification regulation. Food Chem. Toxicol. (2017) 109 130-142. DOI: 10.1016/j.fct.2017.09.007 57. Friesner R. A., Murphy R. B., Repasky M. P., Frye L. L., Greenwood J. R., Halgren T. A., Sanschagrin P. C., Mainz D. T.. Extra precision glide: docking and scoring incorporating a model of hydrophobic enclosure for protein-ligand complexes. J. Med. Chem. (2006) 49 6177-6196. DOI: 10.1021/jm051256o
--- title: 'Role of self-management program based on 5A nursing model in quality of life among patients undergoing hemodialysis: a Randomized Clinical Trial' authors: - Sahar Keivan - Abdolali Shariati - Mojtaba Miladinia - Mohammad Hosein Haghighizadeh journal: BMC Nephrology year: 2023 pmcid: PMC10017059 doi: 10.1186/s12882-023-03108-2 license: CC BY 4.0 --- # Role of self-management program based on 5A nursing model in quality of life among patients undergoing hemodialysis: a Randomized Clinical Trial ## Abstract ### Introduction Various nursing models are usually employed to achieve self-management and improve the quality of life in chronic conditions. Given its person-based characteristics, the 5 A nursing model can improve the quality of life of hemodialysis patients. ### Purpose This study aimed to determine the role of a self-management program based on the 5 A nursing model in the quality of life of patients undergoing hemodialysis. ### Materials and methods This clinical trial was conducted on hemodialysis patients in Iran. Random sampling was adopted to assign 60 patients to intervention and control groups. After the participants completed a demographic questionnaire and the Kidney Disease Quality of Life–Short Form (KDQOL–SF), routine measures were taken in the control group. However, the 5 A nursing model was implemented in the intervention group for three months. The self-care program was executed in face-to-face sessions or via phone calls and SMSs. After three months, the quality of life was measured again in both groups. ### Findings There were significant differences after the intervention between the intervention and control groups in specific dimensions of quality of life, such as cognitive functions, symptoms, sleep, dialysis, social support, and renal complications ($P \leq 0.05$). The two groups also had significant differences in the general scores of quality of life ($P \leq 0.05$). ### Conclusion The 5 A self-management intervention as a person-based model could improve self-care in hemodialysis patients. Nurses can implement this model to mitigate care costs, enhance interventions, and improve patients’ quality of life. ### Trial registration Iranian Registry of Clinical Trials (IRCT20211103052955N1; $\frac{19}{11}$/2021). ## Introduction Chronic kidney disease is a public health threat worldwide. The global prevalence of chronic kidney failure is 262 cases per one million. Hemodialysis patients experience various side effects such as insomnia, skin irritation, headaches, blood pressure disorders, vascular complications, muscle cramps, itches, nausea, and vomiting, which can affect different dimensions of their quality of life [1, 2]. Although many developments have been made in hemodialysis, this complicated process requires a caregiver team and many instructions to improve the quality of life among patients with hemodialysis [3]. Since the quality of life is affected in patients with hemodialysis, appropriate care methods should be adopted to mitigate the effects through lifestyle moderation. Hence, self-management interventions can be instrumental tools to support the necessary lifestyle changes in hemodialysis [4]. The self-management program is a rehabilitation method in which patients play a critical role. All healthcare activities focus on patients to achieve self-decision-making, maximize independence, and improve personal health based on abilities and lifestyles by enhancing the quality of life [5]. Self-management refers to personal abilities to control symptoms, physical outcomes, treatments, and social-psychological effects of chronic cases such as dialysis patients who need to control their lifestyle changes [5, 6]. Studies have shown that self-management interventions such as training patients and using care models can effectively improve disease symptoms and patients’ quality of life, especially regarding compliance with medical prescriptions, functional instructions, and patient satisfaction [7]. Different nursing models are usually employed to achieve self-management and improve the quality of life in chronic cases [8]. As a behavioral modification model, the 5 A nursing model is an evidence-based approach designed to change behavior and achieve self-management. Presented by Glasgow et al. [ 2003], this model offers nursing intervention through assessment, advice, agreement, assistance, and arrangement stages. In fact, this model provides a valuable framework for executing self-management interventions [9]. This model has improved care outcomes in some chronic cases; however, no improvements were reported in some settings [10–12]. The chronic nature of renal failure and dependence on hemodialysis for survival can impose high costs on patients and decline the quality of their lives. Thus, patients worry about their abilities to do daily tasks and live normally. Due to these patients’ various complications, comprehensive nursing interventions are essential, emphasizing rehabilitation programs. Therefore, necessary arrangements should be made to allow patients to take responsibility for improving self-efficacy and enhancing self-management behaviors. There are weaknesses in conventional education strategies. At the same time, the patient’s active participation in the therapeutic process and hemodialysis is essential. On the other hand, although the 5 A model has been assessed for the quality of life of patients with acute coronary syndrome, hypertension, and diabetes patients, the researcher did not find any study investigating the effect of the self-management program based on the 5 A model on the quality of life of hemodialysis patients. Hence, this study aimed to determine the role of a self-management program based on the 5 A nursing model in the quality of life of hemodialysis patients. ## Design and setting This parallel-group randomized clinical trial was conducted in the Dialysis Center of Imam Ali Hospital, affiliated with Ahvaz Jundishapur University of Medical Sciences in Khuzestan Province, Iran, in 2021–2022 (Iranian Registry of Clinical Trials; IRCT20211103052955N1; $\frac{19}{11}$/2021). ## Participants In this trial, the convenience sampling method was adopted to select 60 patients undergoing hemodialysis. The inclusion criteria were literacy, no history of psychological disorders, age of 18 to 65 years old, and no history of cognitive disorders such as Alzheimer’s disease. The participants were randomly assigned to intervention and control groups with 30 members each. The permuted block randomization technique was employed to allocate the participants. The blocks were randomly determined as 2, 4, 6, and 8, whereas a statistician prepared the randomization list. The sampling attrition rate was considered $10\%$. Based on a similar study [13] and the following formula, 30 participants were allocated to each group with a $95\%$ confidence interval (CI) and a test power of $80\%$. Moreover, participants were free to leave the study in case of unwillingness or the emergence of serious problems (Fig. 1). The principal investigator and statistician were blinded. Fig. 1CONSORT flowchart \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\times }_{1}=$\frac{11}{5}$$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\times }_{2}=$\frac{9}{67}$$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }_{1}=$\frac{2}{12}$$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }_{2}=$\frac{2}{6}$$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = \frac{{\left({\sigma _1^2 + \sigma _2^2} \right){{\left[{{z_{1 - \alpha /2}} + {z_{1 - \beta }}} \right]}^2}}}{{{{\left({{M_1} - {M_2}} \right)}^2}}}$$\end{document} ## Intervention The intervention group received a self-management program based on the 5 A model, which was implemented in five stages through face-to-face meetings, phone calls, and SMSs in three months. Figure 2; Table 1 present the 5 A model steps and their implementation methods in detail [13]. Moreover, the control group received the routine hospital program, including conventional care and training measures. Fig. 2The 5 A model algorithm Table 1The 5 A Model Steps in Hemodialysis PatientsStep 1AssessIn this step, patients were analyzed in face-to-face interviews regarding risk factors, history of diseases, renal complications, compliance with pharmaceutical prescription, sleeping status, nutrition, type of activity, and case information. Step 2 AdviseIn this step, the previous analysis results were considered to inform patients of the diagnosed health risks and emphasize the benefits of behavioral modification. Step 3 AgreeAn agreement was reached between the patients and the researcher. Given the diagnosed problems, appropriate behavioral goals were agreed upon with patients, and a practical program was designed for each goal. The criterial need was set between 0 and 1 for each behavioral goal so that patients could determine their trust in the program implementation. These criteria were registered in the behavioral modification form, and the patients were asked to record their performance status in each behavioral goal weekly for 12 weeks. Step 4 AssistThe patients were instructed on how to control the consumption of liquids, how to implement care measures through vascular access, the importance of sports and necessary levels of physical activities, how to take care of their skin, and compliance with their diets and prescriptions in face-to-face sessions. The patients were asked to perform the sports exercises daily and then record their performance results. Proportionate to the needs of patients, a personal training session was arranged respecting the patient’s willingness (in-person or via phone calls) to reiterate and emphasize the instructions. Step 5 ArrangeIn this step, the functions of patients were followed-up for three months. In fact, they were followed up via phone calls or SMSs three times a week from the fifth week to the 12th week every week to ensure that the patients complied with the intervention within the first four weeks. These efforts were made to remind patients of the practical program and solve any potential problems. Furthermore, each patient’s progress was followed up every four weeks in one face-to-face session to modify goals or useful programs by reaching new agreements or encouraging patients to keep up the intervention. ## Data Collection A three-section tool was employed for data collection. The first section gathered data on clinical and demographic variables such as age, gender, educational attainment, and hemodialysis treatment duration. The second section comprised the Kidney Disease Quality of Life–Short Form (KDQOL–SF). In the third section, a checklist of the 5 A model steps was employed to only record the 5 A model steps within a unified framework. The KDQOL–SF is a self-administered questionnaire consisting of general and specific dimensions regarding the quality of life, created by Hayes et al. in 1994. The general dimension of quality of life included two other dimensions (i.e., physical health and psychological health) and eight areas. The physical dimension included four areas: general health, physical performance, physical role, and physical pain, whereas the psychological dimension comprised emotional role, social performance, psychological health, and happiness. Moreover, the specific dimension had 11 areas, including symptoms and problems, effects of renal diseases, burden of renal diseases, cognitive performance, quality of social relationships, social support, sleeping status, employment status, sexual problems, and satisfaction with care and personnel. This multidimensional tool is reliable and valid, including all dimensions of the SF–36 Questionnaire plus the kidney disease variables. It also has high internal consistency and correlation. Each dimension is scored from 0 to 100. Scores above 50 on each dimension and area indicate good quality of life. The reliability and validity of the Persian version of KDQOL–SF were confirmed by Yekaninejad et al. in Iran (α = 0.77–0.9) [14]. This questionnaire was completed before the intervention and three months after the intervention. ## Data Analysis Data analysis was performed in SPSS 18 at a significance level of 0.05 using the independent t-test, paired t-test, and ANOVA. However, nonparametric tests were conducted in case of a non-normal distribution or a ordinal qualitative variable. ## Results The age variable is expressed separately as mean and standard deviation for the control and intervention groups. The average age was 52.83 ± 8.71 years in the control group and 52.66 ± 9.23 years in the intervention group, without a statistically significant difference ($$P \leq 0.943$$). The findings indicated that most patients were male ($56\%$), married ($78\%$), and urban residents ($98\%$). There were no significant differences between the two groups regarding demographic variables such as age, marital status, place of residence, educational attainment, and history of renal transplants ($P \leq 0.05$). However, the two groups differed significantly in terms of some demographic features, such as employment status ($$P \leq 0.009$$), economic status ($$P \leq 0.011$$), and the number of dialysis sessions ($$P \leq 0.001$$) (Table 2). The distortive effects of these variables were analyzed through ANCOVA ($P \leq 0.05$); hence, it could be concluded that the intervention caused intergroup differences and that the abovementioned factors did not affect the results. Table 2Distribution of frequency and percentage of demographic characteristics of patients in intervention and control groupsvariableClassificationintervention groupcontrol groupp-valuenumber (percentage)number (percentage) Gander man(53.3)1660.0))180.602woman(46.7)14(40.0)12 Marriage status single(3.3)1(13.3)40.369married(83.3)25(73.3)22divorced(13.3)4(13.3)4 Address city(96.7)29(100.0)300.313village(3.3)1(0.0)0 Education Literacy for reading and writing(53.3)16(36.7)110.189Under diploma(30.0)9(20.0)6Diploma(13.3)4(13.3)4university(3.3)1[0]0 Job status free(70.0)21(36.7)110.009freelance job(26.7)8(26.7)8Government(0.0)07(13.3)retired(3.3)1(23.3)7 Economic status Income equals expenses(28.6)25(50.0)150.011More income than expenses1(3.4)(6.7)2Income less than expenses3(10.3)(43.3)13 Number of dialysis sessions 2 times a week23(76.7)(40.0)240.0013 times a week7(23.3)(60.0)36 Kidney transplant history no(14.3)4(10.3)30.650yes(85.7)24(89.7)26 According to the results, there were significant differences between the two groups before the intervention concerning the scores of some specific dimensions of quality of life, such as renal complications, dialysis, patient satisfaction, and quality of social interactions. The distortive effects of these variables were analyzed through ANCOVA. Given the P-value ($$P \leq 0.001$$), it could be concluded that the intervention caused intergroup differences, and the abovementioned factors did not affect the results. The independent t-test indicated no significant differences between the two groups after the intervention regarding the specific dimensions of quality of life, such as sexual performance, employment status, quality of social interactions, and patient satisfaction. However, the two groups significantly differed in other dimensions, such as cognitive performance, symptoms, sleeping status, dialysis, social support, and renal complications (Table 3). Table 3Intergroup comparison of specific aspects of quality of life of patients before and after intervention in intervention and control groupsvariableIntervention groupControlgroup*p-valuep-valueBefore interventionAfter interventionBefore interventionAfter interventionmean ± standard deviationmean ± standard deviationmean ± standard deviationmean ± standard deviation Symptom.problem 58.26 ± 14.3972.36 ± 10.6759.37 ± 18.7960.06 ± 19.490.0040.798 Effect of kidney disease 9.58 ± 17.3442.50 ± 13.3728.12 ± 26.4027.29 ± 25.610.0060.002 Work status 5.00 ± 20.125.55 ± 21.1813.79 ± 32.4414.28 ± 32.930.2490.214 Cognitive function 51.33 ± 22.3672.66 ± 14.8958.44 ± 25.9457.11 ± 23.130.0030.260 Quality of social interaction 46.44 ± 20.5270.44 ± 14.6161.77 ± 25.4861.66 ± 25.580.1540.013 Sexual function 17.91 ± 18.4729.16 ± 23.2931.94 ± 33.1330.60 ± 32.140.8440.050 Sleep 46.83 ± 15.3868.58 ± 12.3850.25 ± 12.6649.41 ± 12.530.0010.352 Social support 79.88 ± 25.7394.44 ± 11.0171.10 ± 25.8766.66 ± 25.890.0010.197 Dialysis staff encouragement 90.83 ± 17.3498.33 ± 7.1466.37 ± 29.7163.75 ± 25.710.0010.001 Patient satisfaction 36.66 ± 16.0262.21 ± 13.0859.44 ± 22.6056.66 ± 23.810.0010.001 p-value- Before intervention p-value After intervention Concerning the general dimensions of quality of life before the intervention, there were significant differences between the two groups regarding some dimensions, such as energy level, emotions and feelings, role limitation due to emotional and physical problems, and general health. The distortive effects of these variables were analyzed through ANCOVA, which indicated that the intervention caused intergroup differences and that the above mentioned factors did not affect the results. According to the independent t-test, there were no significant differences between the two groups after the intervention regarding the general dimensions of quality of life, such as role limitation due to emotional and physical problems, pain, and general health. However, the two groups differed significantly in other general scales of quality of life, such as physical performance, emotions and feelings, social performance, general health status, and energy level. The intragroup comparison of the general dimensions of quality of life indicated that all dimensions in both groups changed significantly after the intervention (Table 4). Table 4Between-group and intra-group comparison of the general dimensions of patients’ quality of life before and after the intervention in the intervention and control groupsvariableIntervention groupControl groupp-valuep-valueBefore interventionAfter interventionBefore interventionAfter interventionmean ± standard deviationmean ± standard deviationmean ± standard deviationmean ± standard deviation Physical function 55.50 ± 16.8872.33 ± 19.4655.66 ± 28.0054.33 ± 28.090.9780.978 p-value * 0.0010.001 Role limitation due to physical problems 90.83 ± 26.6561.66 ± 24.3365.83 ± 39.1064.65 ± 37.510.0050.005 p-value * 0.0010.001 Pain 34.33 ± 22.2851.66 ± 20.9841.41 ± 26.3841.00 ± 25.240.2660.266 p-value * 0.0010.001 General health 14.00 ± 17.2434.16 ± 20.4338.83 ± 22.4637.00 ± 21.990.0010.001 p-Value * 0.0010.001 Emotional well-being 39.73 ± 17.7968.53 ± 8.2553.73 ± 18.4555.33 ± 19.510.0040.004 p-value * 0.0010.001 Role limitation due to emotional problems 91.11 ± 23.0444.44 ± 23.7058.88 ± 39.8159.52 ± 38.860.0010.001 p-value * 0.0090.009 Social function 47.50 ± 17.4965.41 ± 12.1448.33 ± 22.1949.16 ± 22.720.8720.872 p-value * 0.0010.001 Energy-Fatigue 27.50 ± 21.6064.00 ± 9.8639.83 ± 21.7141.00 ± 23.130.0310.031 p-value * 0.0010.001 General health 52.00 ± 9.2472.00 ± 7.1454.33 ± 22.0753.00 ± 21.670.5950.595 p-value *** 0.0010.001 p-value- Before intervention p-value After intervention p-value Intragroup comparison ## Discussion This study aimed to determine the role of a self-management program based on the 5 A nursing intervention model in the quality of life of hemodialysis patients. According to the findings, there were significant differences between the two groups after the intervention regarding the specific dimensions of life quality, such as cognitive performance, symptoms, sleeping status, dialysis, social support, and renal complications. Farbod et al. [ 2019] analyzed the effects of a 5 A self-management model on the quality of life of patients with hypertension. They reported that the mean scores of different areas of quality of life (physical, psychological, social, and environmental) and the total score of quality of life increased significantly in the intervention group as opposed to the control group after the intervention [15]. Khoshkhu et al. [ 2021] analyzed the effects of a 5 A-based program on self-care and quality of life among patients with hypertension. Their results indicated significant differences between the intervention and control groups after the intervention [16]. In a study entitled Effects of 5 A Care Model on Quality of Life and Fatigue in Patients Undergoing Chemotherapy, Zhang et al. [ 2021] indicated that the proposed model increased patients’ quality of life significantly and reduced their fatigue [17]. In another study entitled Effects of 5 A-Based Self-Management Program on Fatigue and Dyspnea of Patients with Heart Failure, Hajmohammadi et al. [ 2021] reported that the proposed model decreased fatigue and shortness of breath significantly among patients [14]. Hence, in line with the recent research findings, other studies have also found this model effective in certain factors, such as the quality of life and self-care among patients with chronic diseases. In this study, many dimensions of quality of life were improved significantly in the intervention group as opposed to the control group. The total score of the general quality of life significantly increased among the patients in the intervention group; however, it significantly decreased in the control group. These findings indicated a significant difference between the two groups regarding the quality of life scores in all dimensions. In other words, most of the scores were improved in the intervention group; however, the scores declined in the control group. In fact, the intervention improved the general quality of life scores significantly in most dimensions. Ginanjar et al. [ 2017] used cell phones and the 5 A model to enhance the quality of life among patients with hypertension. According to their results, the scores of patients increased on all SF–36 dimensions in the intervention group as opposed to the control group. They also reported significant differences in physical and emotional dimensions and pain [18]. In line with other studies, this survey indicated the favorable effects of the 5 A model on the quality of life of many patients with chronic conditions due to its person-based characteristics. The declined quality of life scores in the control group could be attributed to the specific sampling conditions. Many patients had constraints in attending the group sessions and doctor visits due to the public instructions during the COVID-19 pandemic, which could affect their quality of life. Contrary to the survey by Javanoosh et al. [ 2016] examining the effect of the self-management program based on the 5 A model on the quality of life of the elderly with acute coronary syndrome, the results showed an increase in the average scores of all aspects of the quality of life in the intervention group, but, this increase was not statistically significant. This lack of significance can be due to the old age of the participants, as many factors are influential in their quality of life, and many functions and activities are reduced and declined per se due to the nature of acute coronary syndrome [19]. In addition to imposing high costs and excessive burdens on society due to side effects and dependence on health centers, hemodialysis can cause unfavorable changes to the quality of life. The first step in the 5 A model is based on a patient’s personal assessment, and each patient’s conditions and problems are considered in all steps. Hence, this model should be applied to hemodialysis patients more often. Moreover, non-attendance follow-ups via phone can help identify at-risk cases and prevent re-hospitalization or the emergence of risk factors in patients before the onset of a condition. The researcher did not find any study investigating the effect of the self-management program based on the 5 A model on the quality of life of hemodialysis patients, and this study is one of the newest in this field. Therefore, we recommend using this model to improve the quality of life of dialysis patients. Also, we need more studies with larger samples. This study faced certain limitations. For instance, there was only one dialysis center in the designated city; thus, the research team had to perform sampling in only one ward. Despite the iterated research emphasis, this limitation may have led to the distribution and sharing of information between the two groups, accounting for the insignificance of some dimensions. Moreover, many factors the researcher could not control may have affected the quality of life. ## Conclusion The findings indicated that the 5 A nursing model improved the quality of life of hemodialysis patients. Many care services provided by healthcare centers, especially for patients with chronic conditions, are offered within a routine framework without considering each patient’s needs. For instance, some hemodialysis patients may have polyuria, whereas others may develop oliguria, requiring special care and instructions. These details are sometimes ignored. Therefore, self-management models such as the proposed one can improve these patients’ self-care and quality of life. However, the follow-up period was short in this study; therefore, it is recommended to prolong the follow-up period in subsequent studies. ## References 1. Shamsizadeh M, Ranjbaran F, Sharifian P. The Effect of an Educational Program based on the teach back Method on the quality of life in Hemodialysis Patients: a clinical Trial Study*. J Crit Care Nurs* (2021.0) 14 1-11 2. Javanvash Z, Mojdekanloo M, Rastaghi S, Rad M. The effect model-based self-management program 5A on quality of life of elderly patients with acute coronary syndrome Bojnourd Year 1395. J Sabzevar Univ Med Sci (2018.0) 25 75-82 3. 3.Jafari F, Hashemi N, Reisi M. The effect of diet training on variations in blood pressure, weight, and some biochemical factors in hemodialysis patients: a clinical trial. Journal of Clinical Nursing and Midwifery. 2015;3. 4. Taheri P, Varmaghani M, Nazari N, Sharifi F, Fakhrzadeh H, Arzaghi SM. Health status of elderly people n east Azerbaijan; a cross sectional study. Iran J diabet metabolism (2017.0) 16 249-60 5. Bagheri M, Bagheri M, Niknami S. The effect of educational intervention on knowledge and selfcare of elderly people with type 2 diabet. J Gerontol (2018.0) 3 1-10 6. Dehghan B, Shoghi M, SeidFatemi N. Effect of self-management training with Group discussion method on self-esteem of adolescents with Hemodialysis. J Pediatr Nurs (2019.0) 5 74-82 7. Shabibi P, Zavareh MSA, Sayehmiri K, Qorbani M, Safari O, Rastegarimehr B. Effect of educational intervention based on the Health Belief Model on promoting self-care behaviors of type-2 diabetes patients. Electron physician (2017.0) 9 5960. DOI: 10.19082/5960 8. Bikbov B, Purcell CA, Levey AS, Smith M, Abdoli A, Abebe M. Global, regional, and national burden of chronic kidney disease, 1990–2017: a systematic analysis for the global burden of Disease Study 2017. The lancet (2020.0) 395 709-33. DOI: 10.1016/S0140-6736(20)30045-3 9. 9.Glasgow RE, Davis CL, Funnell MM, Beck A. Implementing practical interventions to support chronic illness self-management. The joint commission journal on quality and safety. 2003 Nov 1;29(11):563 – 74. 10. Naji Esfehani F, Seirafi MR, Mujembari AK. The effectiveness of behavioral activation intervention on increasing Self-Care Behaviors and Life Expectancy in the Elderly. Aging Psychol (2020.0) 6 93-105 11. Barati F, Sadeghmoghadam L, Sajjadi M, Sh N, Bahri N. Validation of the Persian version of self-care tools for hypertension among older adults. Med Glas (Zenica) (2019.0) 16 338-43 12. 12.Lashkari F, Brazparandjani S, Latifi SM, Chahkhoei M, Khalili A, Paymard A et al. The effect of collaborative care model on the fatigue in patients undergoing maintenance hemodialysis: A randomized clinical trial. 2016. 13. Bastani M, Ghasemi G, sadeghi M. The effect of selected Core Stability exercises on restless legs syndrome and quality of life in the Elderly Undergoing Hemodialysis. Qom Univ Med Sci J (2018.0) 12 48-58. DOI: 10.29252/qums.12.8.48 14. 14.Hajmohamadi M, Sabzvari S, Jahani Y, Imani-Goghary Z. Investigating the Effectiveness of Self-management Program based on 5A Model on Fatigue and Dyspnea in Patients with Heart Failure. 2021. 15. 15.Azar FEF, Solhi M, Azadi NA, Ziapour A, Lebni JY, Sharma M et al. The Effect of Educational Intervention Based on Model 5A Self-Management Theory on Life Quality in Hypertensive Patients. 2019. 16. Khoshkhoo M, Sajjadi M, Mansoorian MR, Ajamzibad H. Effects of 5A model-based intervention on self-care and quality of life in Elderly People with Hypertension. Iran J Ageing (2021.0) 16 348-61 17. Zhang X, Lai M, Wu D, Luo P, Fu S. The effect of 5A nursing intervention on living quality and self-care efficacy of patients undergoing chemotherapy after hepatocellular carcinoma surgery. Am J Translational Res (2021.0) 13 6638 18. 18.Saputri GZ, Akrom ED. Improving Outpatient’s Quality of Life Through Patient Adherence of Antihypertensive Therapy Using “Mobile Phone (SMS) and Brief Counseling–5A” in Polyclinic of Internal Medicine at PKU Muhammadiyah Bantul Hospital, Yogyakarta.Indonesian Journal of Clinical Pharmacy Volume. 2017;6(2). 19. Moradi M, Nasiri M, Jahanshahi M, Hajiahmadi M. The effects of a self-management program based on the 5A model on self-efficacy among older men with hypertension. Nurs Midwife Stud (2019.0) 8 21-7. DOI: 10.4103/nms.nms_97_17
--- title: 'Prevalence of hypertension and determinants of treatment-seeking behaviour among adolescents and young adults in India: an analysis of NFHS-4' authors: - Yuvaraj Krishnamoorthy - Sathish Rajaa - Sudheera Sulgante - Palanivel Chinnakali - Nidhi Jaswal - Sonu Goel journal: Journal of Public Health (Oxford, England) year: 2022 pmcid: PMC10017093 doi: 10.1093/pubmed/fdac006 license: CC BY 4.0 --- # Prevalence of hypertension and determinants of treatment-seeking behaviour among adolescents and young adults in India: an analysis of NFHS-4 ## Abstract ### Background Previous evidences have reported that almost three-fourth of young hypertensives are not seeking care for their condition leading to severe complications. This study was conducted to assess the determinants of treatment-seeking behaviour among the young hypertensives in India. ### Methods The National Family Health Survey-4 data were analysed. Sampling weights and clustering was accounted using svyset command. Screening, awareness, prevalence and control status were reported with $95\%$ confidence interval (CI). Poisson regression was done to identify the determinants of treatment-seeking behaviour. ### Results In total, $13.8\%$ of younger adults had hypertension, $51.1\%$ were aware of their status and $19.5\%$ sought treatment. Participants in 15–19 years (adjusted Prevalence Ratio (aPR) = 0.70) and 20–29 years (aPR = 0.63), male gender (aPR = 0.84), Muslim religion (aPR = 1.14), urban region (aPR = 0.87), secondary (aPR = 0.88) and higher education (aPR = 0.86), residing in Northern (aPR = 0.79), Central (aPR = 0.76), Southern region (aPR = 0.65), preferring home treatment, medical shop or any other care (aPR = 0.63) were significant determinants of treatment-seeking behaviour. ### Conclusion More than 1 in 10 younger adults in India have hypertension and only half of them were aware of their status and one-fifth sought treatment. Adolescents, males, Hindus, urban population, higher education and residing in Northern, Central and Southern region had poor treatment-seeking behaviour. ## Introduction Non-communicable diseases (NCDs), also known as lifestyle diseases, have become one of the leading causes of deaths worldwide.1 Almost $60\%$ of all the deaths and $50\%$ of all the morbidity burden is contributed by the major NCDs globally.2 Amongst them, hypertension (HTN) has emerged as an important public health concern. Owing to the lack of recognizable signs and symptoms in hypertension, most of the times people remain unaware of this condition. Undiagnosed and untreated HTN for longer time can lead to severe form of disease leading to serious complications and mortality. WHO has rated HTN as one of the major causes of premature mortality worldwide.3 Henceforth, consistency in screening for early diagnosis and adequate management with life style modification is needed for effective reduction in the prevalence of high blood pressure (BP). Several strategies have been developed throughout the world to lower the hypertension burden. Government of India has also introduced several initiatives such as ‘National programme for prevention and control of cancer, diabetes, cardiovascular diseases and stroke (NPCDCS)’, population level screening of hypertension and ‘India Hypertension Control Initiative (IHCI)’ to decrease the deaths and disability.4,5 Most of these strategies are focussed on middle aged and older age group people. However, there is an increasing trend of hypertension among the people belonging to younger age group over the past decade.6 Young hypertension can be defined as the hypertension occurring in people <40 years of age.7 Almost one-third of young hypertension is caused by some underlying secondary causes.8 Though secondary hypertension can be cured when compared with the essential hypertension, which usually lasts lifelong, secondary hypertension causes severe end organ damage more than the essential hypertension.8 Hence, a holistic approach to evaluate the younger population for suspected hypertension is important. Another important dimension that requires focus is the awareness about one’s own hypertension status and treatment-seeking behaviour among the younger population. Since hypertension is considered an iceberg disease, all the adults aged ≥ 18 years should be screened for raised BP. However, previous evidences from India showed that only half of the people with hypertension knew their hypertension status.9 Even after diagnosis, almost three-fourth are not seeking care from health facility leading to severe complications and end organ damage.9 Even though, several studies were conducted around India to determine the burden of young hypertension,10–12 only few studies tried to explore the treatment-seeking behaviour of the young hypertension13 and there was no large-scale survey assessing the care-seeking behaviour among young hypertensives. National Family Health Survey-4 (NFHS-4) data provide sufficient opportunity to study about the prevalence of hypertension among adolescents and younger adults in India.14 In addition to the point estimates, understanding the treatment-seeking behaviour and its determinants might help in devising new strategies to increase the treatment coverage rate. Hence, this study was conducted as a secondary data analysis of NFHS-4 data to determine the prevalence of hypertension the determinants of treatment-seeking behaviour among adolescents and young adults in India. ## Study design A secondary data analysis was conducted using nationally representative data of NFHS-4 gathered from the Demographic Health Survey (DHS) programme. NFHS survey has been conducted to capture data on various health indicators the Indian population. ## Study setting India, the second most populated country in the world, is having a population of roughly 130 crores and is divided into 30 states and 6 union territories (UT). Each state is sub-divided into districts and further districts into census enumeration blocks (CEB)/wards in urban area and villages/taluk in rural area. In India, opportunistic screening for hypertension should be done for all the individuals aged ≥18 years for the diagnosis of young hypertension.15 India was also the first country worldwide to adopt the NCD global monitoring framework and action plan, which included nine targets including the target on reducing the burden of hypertension.16 ## Participants Participants < 40 years (15–39 years) were taken as study population for the current study analysis. Variables extracted for these study population consists of the following: independent variables extracted were socio-demographic characteristics such as age, gender, education, wealth index, marital status, type of residence, religion, caste/tribe, geographical region, health insurance coverage and general healthcare seeking behaviour. Dependent variable was the treatment-seeking behaviour among the young hypertensives. ## Data sources Self-reported case of hypertension was ascertained in the survey based on ‘yes’ or ‘no’ response to the question whether the respondents were told that they have high BP on ≥2 occasions by the physician. Newly diagnosed hypertension was defined by the Joint National Committee (JNC’s) eight hypertension guidelines.17 All the participants underwent three BP measurements during the survey. Participants with systolic blood pressure (SBP) reading ≥ 140 mmHg and/or diastolic blood pressure (DBP) reading ≥ 90 mmHg while taking the average of second and third measurements were identified as newly diagnosed cases. For the assessment of treatment-seeking behaviour, question on whether sought treatment and currently on medication was asked among the respondents who were told that they had high BP on ≥2 occasions by a physician. ## Study size In NFHS-4 survey, two-stage sampling method was used for the selection of villages (in rural areas) and CEBs (in urban areas). The household selection process, data sources, its validation and data collection procedure have been described comprehensively as a separate report.14 In total, 699 686 females and 112 122 males have completed the questionnaire. Since the current study focuses on young hypertension, the responses of 631 876 participants (15–39 years) who were interviewed regarding the hypertension status were included into the analysis. ## Statistical methods We obtained the dataset from the DHS website in.dta format and it was imported into STATA 14.2 (StataCorp, College Station, TX, USA) for analysing the data. Variables necessary for the analysis such as age, gender, education, wealth index, marital status, type of residence, religion, caste/tribe, geographical region, health insurance coverage and general healthcare seeking behaviour were kept in the dataset and all other variables were dropped for ease of analysis. These co-variates were selected after reviewing the previous literature on similar studies, in addition to obtaining the opinion of public health experts in this regard. Then, sampling weights were adjusted while performing the analysis, to account the differential probabilities of participant selection. Sample design and clustering was also accounted using svyset command. Screening, prevalence, awareness and control status of young hypertension (belonging to age group < 40 years) were reported with $95\%$ confidence interval (CI). Univariable and multivariable Poisson regression was performed to identify the determinants of treatment-seeking behaviour among young hypertensives. Unadjusted and adjusted prevalence ratio (PR) with $95\%$CI was reported. Variables with P value < 0.20 in the univariable model were considered into the multivariable regression model. Variables with P value < 0.05 in the multivariable model were considered statistically significant determinants of treatment-seeking behaviour. Poisson regression was performed instead of the commonly employed logistic regression technique as the logistic regression only provides the odds ratio, whereas PR can be estimated by the Poisson regression (better measure for reporting the effect estimate in cross-sectional surveys). In addition, it is difficult to infer the odds ratio for cross-sectional studies as there is confusion between odds or risk, leading to erroneous quantitative interpretation. ## Results In total, 631 876 participants aged 15–39 years were monitored for hypertension in the NFHS-4 survey. As shown in Table 1, majority of the participants ($43.3\%$) belonged to the age group between 20 and 29 years. Females consisted of $87.5\%$ of the study population. Most of the participants ($67\%$) were currently married; more than three-fourth belonged to Hindu religion; more than two-third had secondary to higher educational qualification; majority ($66.6\%$) were living in rural area. Almost one-fourth of the study participants belonged to Central region. Some form of health insurance scheme covered only one-fourth of the participants; more than half of them generally seek care in private healthcare facilities. Table 1 \| Socio-demographic characteristics \| Frequency, N (unweighted proportion %) \| Weighted proportion (95% CI) \| \| --- \| --- \| --- \| \| Age category (in years) \| Age category (in years) \| Age category (in years) \| \| 15–19 \| 144 211 (22.8) \| 22.4 (22.3–22.6) \| \| 20–29 \| 271 957 (43.0) \| 43.3 (43.1–43.5) \| \| 30–39 \| 215 708 (34.1) \| 34.3 (34.1–34.5) \| \| Gender \| Gender \| Gender \| \| Male \| 80 046 (12.7) \| 12.5 (12.3–12.8) \| \| Female \| 551 830 (87.3) \| 87.5 (87.2–87.7) \| \| Marital status (N = 631 846) \| Marital status (N = 631 846) \| Marital status (N = 631 846) \| \| Never married \| 207 205 (32.8) \| 30.9 (30.7–31.1) \| \| Currently married \| 411 407 (65.1) \| 67.0 (66.7–67.2) \| \| Widowed/separated/divorced \| 13 234 (2.1) \| 2.1 (2.0–2.2) \| \| Religion \| Religion \| Religion \| \| Hindu \| 465 383 (73.6) \| 80.0 (79.4–80.5) \| \| Muslim \| 90 359 (14.3) \| 14.5 (14.0–15.0) \| \| Christian \| 45 808 (7.3) \| 2.2 (2.1–2.4) \| \| Othersa \| 30 326 (4.8) \| 3.2 (3.1–3.5) \| \| Education status (N = 630 701) \| Education status (N = 630 701) \| Education status (N = 630 701) \| \| No formal education \| 122 522 (19.5) \| 19.0 (18.7–19.3) \| \| Primary \| 79 083 (12.5) \| 12.5 (12.3–12.6) \| \| Secondary \| 344 555 (54.6) \| 53.7 (53.4–54.0) \| \| Higher \| 84 541 (13.4) \| 14.8 (14.5–15.1) \| \| Caste/tribe (N = 605 513) \| Caste/tribe (N = 605 513) \| Caste/tribe (N = 605 513) \| \| Scheduled castec \| 115 990 (19.2) \| 21.8 (21.4–22.2) \| \| Scheduled tribec \| 115 602 (19.1) \| 9.7 (9.4–10.0) \| \| Other backward class \| 245 039 (40.5) \| 44.9 (44.4–45.5) \| \| None of the above \| 125 788 (20.8) \| 22.9 (22.4–23.4) \| \| Don’t know \| 3094 (0.5) \| 0.7 (0.6–0.8) \| \| Wealth index \| Wealth index \| Wealth index \| \| Poorest (I quintile) \| 120 440 (19.1) \| 18.0 (17.6–18.4) \| \| Poorer (II quintile) \| 136 768 (21.6) \| 20.1 (19.8–20.4) \| \| Middle (III quintile) \| 134 933 (21.4) \| 21.0 (20.7–21.2) \| \| Richer (IV quintile) \| 125 323 (19.8) \| 21.0 (20.7–21.3) \| \| Richest (V quintile) \| 114 412 (18.1) \| 19.9 (19.4–20.4) \| \| Residence \| Residence \| Residence \| \| Urban \| 182 812 (28.9) \| 33.4 (32.5–34.3) \| \| Rural \| 449 064 (71.1) \| 66.6 (65.7–67.5) \| \| Geographical region \| Geographical region \| Geographical region \| \| North \| 129 312 (20.5) \| 13.8 (13.3–14.3) \| \| Central \| 172 211 (27.2) \| 24.8 (24.1–25.5) \| \| East \| 112 689 (17.8) \| 22.1 (21.4–22.9) \| \| Northeast \| 87 228 (13.8) \| 3.5 (3.3–3.6) \| \| West \| 51 687 (8.2) \| 14.2 (13.5–15.0) \| \| South \| 78 749 (12.5) \| 21.6 (20.8–22.4) \| \| Covered by health insurance/health scheme \| Covered by health insurance/health scheme \| Covered by health insurance/health scheme \| \| Yes \| 159 644 (25.3) \| 27.5 (27.0–28.0) \| \| No \| 468 681 (74.2) \| 72.0 (71.5–72.4) \| \| Don’t know \| 3551 (0.6) \| 0.5 (0.4–0.6) \| \| General healthcare seeking behaviour \| General healthcare seeking behaviour \| General healthcare seeking behaviour \| \| Public \| 323 839 (51.3) \| 43.9 (43.4–44.4) \| \| Private \| 287 727 (45.5) \| 52.3 (51.9–52.8) \| \| Othersb \| 20 310 (3.2) \| 3.7 (3.5–3.9) \| Hypertension care cascade among adolescents and younger adults are provided in Fig. 1. We found that $57.8\%$ ($95\%$ CI: 57.4–$58.1\%$) were screened for hypertension previously. The level of screening was better with higher age groups as only $27.9\%$ adolescents were screened, whereas $64\%$ those belonging to 20–29 years and $69.4\%$ of those belonging to 30–39 years were screened. In total, $13.8\%$ ($95\%$ CI: 13.6–$14.1\%$) of the adolescents and younger adults had hypertension (self-reported/newly diagnosed). The burden of hypertension was also higher with each increasing age group with only $5.5\%$ of adolescents had hypertension, whereas nearly $21\%$ of those belonging to 30–39 years had hypertension. Amongst these young hypertensives, $51.1\%$ ($95\%$ CI: 50.2–$52.1\%$) of the participants were aware of their hypertension status. The level of awareness was better among those participants belonging to 20–29 years age group ($57\%$) when compared with those belonging to 15–19 years ($46.9\%$) or 30–39 years ($47.5\%$). Amongst these self-reported participants, only $19.5\%$ ($95\%$ CI: 18.7–$20.3\%$) sought treatment and currently on medications (Table 2). Fig. 1: Hypertension care cascade pathway among adolescents and younger adults (15–39 years) in NFHS 4, India. TABLE_PLACEHOLDER:Table 2 Table 3 shows the determinants of treatment-seeking behaviour among young hypertensives in India. All the variables from the univariable model were included in the multivariable model as they had P value < 0.20. In the adjusted analysis, age, gender, religion, residence, education, caste/tribe, geographical region and general healthcare seeking behaviour were found to be significant determinants of treatment-seeking behaviour. Table 3 \| Characteristics \| Total \| Sought treatment (weighted %) \| Unadjusted prevalence ratio (95% CI) \| P value \| Adjusted prevalence ratio (95% CI) \| P value.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Age category (in years) \| Age category (in years) \| Age category (in years) \| Age category (in years) \| Age category (in years) \| Age category (in years) \| Age category (in years) \| \| 15–19 \| 3586 \| 17.2 \| 0.71 (0.64–0.79) \| <0.001 \| 0.70 (0.61–0.80) \| <0.001 \| \| 20–29 \| 18 285 \| 14.8 \| 0.61 (0.57–0.65) \| <0.001 \| 0.63 (0.59–0.67) \| <0.001 \| \| 30–39 \| 21 509 \| 24.2 \| Ref \| — \| Ref \| — \| \| Gender \| Gender \| Gender \| Gender \| Gender \| Gender \| Gender \| \| Male \| 3909 \| 16.4 \| 0.83 (0.74–0.93) \| 0.001 \| 0.84 (0.75–0.95) \| 0.005 \| \| Female \| 39 471 \| 19.8 \| Ref \| — \| Ref \| — \| \| Marital status \| Marital status \| Marital status \| Marital status \| Marital status \| Marital status \| Marital status \| \| Never married \| 6610 \| 16.1 \| Ref \| — \| Ref \| — \| \| Currently married \| 35 586 \| 20.0 \| 1.24 (1.13–1.36) \| <0.001 \| 0.97 (0.86–1.08) \| 0.58 \| \| Widowed/separated/divorced \| 1182 \| 22.6 \| 1.40 (1.18–1.67) \| <0.001 \| 0.98 (0.81–1.19) \| 0.88 \| \| Religion \| Religion \| Religion \| Religion \| Religion \| Religion \| Religion \| \| Hindu \| 30 245 \| 18.6 \| Ref \| — \| Ref \| — \| \| Muslim \| 7026 \| 23.3 \| 1.25 (1.15–1.36) \| <0.001 \| 1.14 (1.04–1.25) \| 0.007 \| \| Christian \| 3513 \| 20.7 \| 1.11 (0.93–1.33) \| 0.25 \| 1.15 (0.95–1.39) \| 0.15 \| \| Othersa \| 2596 \| 23.2 \| 1.25 (1.08–1.43) \| 0.002 \| 1.04 (0.91–1.21) \| 0.54 \| \| Residence \| Residence \| Residence \| Residence \| Residence \| Residence \| Residence \| \| Urban \| 14 883 \| 18.1 \| 0.88 (0.81–0.96) \| 0.006 \| 0.87 (0.79–0.96) \| 0.005 \| \| Rural \| 28 497 \| 20.4 \| Ref \| — \| Ref \| — \| \| Education status (N = 43 279) \| Education status (N = 43 279) \| Education status (N = 43 279) \| Education status (N = 43 279) \| Education status (N = 43 279) \| Education status (N = 43 279) \| Education status (N = 43 279) \| \| No formal education \| 7922 \| 23.7 \| Ref \| — \| Ref \| — \| \| Primary \| 5440 \| 22.2 \| 0.94 (0.85–1.03) \| 0.17 \| 0.99 (0.90–1.09) \| 0.91 \| \| Secondary \| 23 173 \| 18.4 \| 0.78 (0.72–0.83) \| <0.001 \| 0.88 (0.81–0.96) \| 0.003 \| \| Higher \| 6744 \| 16.6 \| 0.70 (0.63–0.79) \| <0.001 \| 0.86 (0.76–0.97) \| 0.02 \| \| Caste/tribe (N = 41 095) \| Caste/tribe (N = 41 095) \| Caste/tribe (N = 41 095) \| Caste/tribe (N = 41 095) \| Caste/tribe (N = 41 095) \| Caste/tribe (N = 41 095) \| Caste/tribe (N = 41 095) \| \| Scheduled caste \| 7997 \| 18.3 \| 0.80 (0.72–0.89) \| <0.001 \| 0.92 (0.83–1.02) \| 0.11 \| \| Scheduled tribe \| 6676 \| 22.2 \| 0.97 (0.87–1.09) \| 0.64 \| 0.95 (0.84–1.07) \| 0.41 \| \| Other backward class \| 17 173 \| 17.3 \| 0.75 (0.70–0.82) \| <0.001 \| 0.89 (0.82–0.96) \| 0.005 \| \| None of the above \| 9249 \| 22.9 \| Ref \| — \| Ref \| — \| \| Wealth index \| Wealth index \| Wealth index \| Wealth index \| Wealth index \| Wealth index \| Wealth index \| \| Poorest (I quintile) \| 5011 \| 23.1 \| Ref \| — \| Ref \| — \| \| Poorer (II quintile) \| 7760 \| 21.1 \| 0.91 (0.83–1.00) \| 0.06 \| 0.99 (0.90–1.10) \| 0.92 \| \| Middle (III quintile) \| 9688 \| 17.8 \| 0.77 (0.69–0.85) \| <0.001 \| 0.91 (0.82–1.02) \| 0.10 \| \| Richer (IV quintile) \| 10 720 \| 18.6 \| 0.81 (0.72–0.89) \| <0.001 \| 0.99 (0.89–1.12) \| 0.96 \| \| Richest (V quintile) \| 10 201 \| 19.3 \| 0.83 (0.75–0.93) \| 0.001 \| 1.05 (0.91–1.20) \| 0.52 \| \| Geographical region \| Geographical region \| Geographical region \| Geographical region \| Geographical region \| Geographical region \| Geographical region \| \| North \| 11 009 \| 20.5 \| 0.73 (0.66–0.82) \| <0.001 \| 0.79 (0.69–0.91) \| 0.001 \| \| Central \| 7691 \| 19.0 \| 0.68 (0.62–0.76) \| <0.001 \| 0.76 (0.67–0.86) \| <0.001 \| \| East \| 6995 \| 24.2 \| 0.87 (0.78–0.97) \| 0.01 \| 0.94 (0.83–1.06) \| 0.34 \| \| Northeast \| 6669 \| 27.9 \| Ref \| — \| Ref \| — \| \| West \| 2094 \| 25.5 \| 0.91 (0.79–1.06) \| 0.25 \| 1.02 (0.86–1.20) \| 0.81 \| \| South \| 8922 \| 14.7 \| 0.53 (0.47–0.59) \| <0.001 \| 0.65 (0.57–0.75) \| <0.001 \| \| Covered by health insurance/health scheme (n = 43 139) \| Covered by health insurance/health scheme (n = 43 139) \| Covered by health insurance/health scheme (n = 43 139) \| Covered by health insurance/health scheme (n = 43 139) \| Covered by health insurance/health scheme (n = 43 139) \| Covered by health insurance/health scheme (n = 43 139) \| Covered by health insurance/health scheme (n = 43 139) \| \| Yes \| 12 548 \| 16.7 \| 0.79 (0.74–0.86) \| <0.001 \| \| 0.05 \| \| No \| 30 591 \| 21.0 \| Ref \| — \| Ref \| — \| \| General healthcare seeking behaviour \| General healthcare seeking behaviour \| General healthcare seeking behaviour \| General healthcare seeking behaviour \| General healthcare seeking behaviour \| General healthcare seeking behaviour \| General healthcare seeking behaviour \| \| Public \| 23 188 \| 19.0 \| 0.94 (0.87–1.00) \| 0.07 \| 0.98 (0.91–1.05) \| 0.54 \| \| Private \| 19 296 \| 20.2 \| Ref \| — \| Ref \| — \| \| Othersb \| 896 \| 14.3 \| 0.71 (0.57–0.88) \| 0.002 \| 0.63 (0.51–0.79) \| <0.001 \| Participants belonging to 15–19 years age group (aPR = 0.70; $95\%$CI: 0.61–0.80) and 20–29 years age group (aPR = 0.63; $95\%$CI: 0.59–0.67) had lesser treatment-seeking behaviour when compared to the participants belonging to higher age group (30–39 years). Males had lesser treatment-seeking behaviour for hypertension (aPR = 0.84; $95\%$CI: 0.75–0.95) when compared to females. Muslims had significantly better treatment-seeking behaviour for hypertension when compared to Hindus (aPR = 1.14; $95\%$CI: 1.04–1.25). A smaller number of participants in urban areas had sought treatment for hypertension (aPR = 0.87; $95\%$CI: 0.79–0.96) compared with people in rural residence. Participants having secondary (aPR = 0.88; $95\%$CI: 0.81–0.96) and higher educational qualification (aPR = 0.86; $95\%$CI: 0.76–0.97) had significantly lower treatment-seeking behaviour when compared to participants with no formal education. Participants belonging to Northern (aPR = 0.79; $95\%$CI: 0.69–0.91), Central (aPR = 0.76; $95\%$CI: 0.67–0.86), Southern region (aPR = 0.65; $95\%$CI: 0.57–0.75) had significantly lower treatment-seeking behaviour when compared to participants in the Northeast region. Participants who prefer home treatment, medical shop or other form of care for seeking general healthcare (aPR = 0.63; $95\%$CI: 0.51–0.79) had significantly lower treatment-seeking behaviour compared to those who seek general healthcare from private health facilities. ## Main findings of the study In the current study, the burden of young hypertension in India was $13.8\%$. We have found that $42.2\%$ of the young adults were never screened for hypertension before. Among the young hypertensives, only $51.1\%$ knew about their status and only $19.5\%$ of them sought treatment. We also found that the adolescents, males, Hindus, urban population, higher education and residing in Northern, Central and Southern region had poor treatment-seeking behaviour. ## What is already known on this topic The overall prevalence of young hypertension in India was found to be $13.8\%$ ($95\%$ CI: 13.6–$14.1\%$). Similar finding was found in the large prospective SITE study (Screening India’s twin epidemic) where $12.7\%$ of the individuals below 40 years had hypertension.6 Large-scale South Asian study conducted by Prasad et al. showed that $11.9\%$ of young adults had hypertension.18 Other small-scale studies around India showed that the prevalence of young hypertension ranging from 10 to $17\%$.19–21 There is an increasing burden of young hypertension over the past decade in India because of adopting unhealthy lifestyle changes like unhealthy dietary habits, physical inactivity and tobacco and alcohol use at an earlier age. A retrospective analysis among young hypertensives showed that ~$30\%$ of hypertension among younger adults was due to secondary causes.22 ## What this study adds In the current study, we have found that $42.2\%$ of the young adults were never screened for hypertension before. Among the young hypertensives diagnosed in the study, only $51.1\%$ knew about their hypertension status and only $19.5\%$ of them sought treatment and currently on medications. Existing guidelines on hypertension in India has provided comprehensive guide on screening, diagnosis, assessment and management of hypertension.15 In addition, introduction of newer programmes for population level screening of hypertension and awareness campaigns are being conducted.15,16 Almost half of the younger adults were never screened for hypertension or unaware of their hypertensive status and ~$80\%$ of young hypertensives did not seek treatment. This shows that our country is slow or lagging behind in the secondary prevention of hypertension, which involves early diagnosis by screening and adequate treatment of the condition. Hence, the combination of active mode of population level and high-risk group screening and the passive mode of opportunistic screening of all the younger adults for hypertension will help to improve the scenario. Dropout from seeking care should be avoided by following up the patient through frontline workers like ASHA, ANM or AWW during their regular home visits. Once diagnosed, primary health facility nearest to the patient’s residence need to be informed in carrying out the above-mentioned activities. We also identified certain determinants of poor treatment-seeking behaviour for hypertension, which will help devise targeted interventions for early initiation of treatment and increase the coverage among younger hypertensives. We found that participant belonging to 15–19 years and 20–29 years had poor treatment-seeking behaviour compared with 30–39-year-old participants with hypertension. Possible reasons for such finding could be the desire to handle their own problems, lower perceived susceptibility and perceived need for seeking help and financial concerns.23 However, further qualitative exploration is required to understand and address their concerns in seeking treatment for hypertension. We also found that males had poor treatment-seeking behaviour for hypertension compared to females. Males tend to have poor health-seeking behaviour in general irrespective of the condition.24,25 It has also been reported that the men only seek care only during emergencies or in the later stages of chronic illnesses.26,27 Hence, males remain an important high-risk group that need to be targeted for improving their care-seeking behaviour especially for chronic conditions like hypertension. We also found that the urban population, participants with higher educational qualifications and people belonging to the Southern region had poor treatment-seeking behaviour for hypertension. Some of these findings are surprising given the higher awareness about the disease condition and its complications amongst people with higher educational qualifications. In urban area, irrespective of the higher accessibility to healthcare facilities, there is predominance of the private sector making it difficult for the urban poor to access and afford the care. Hence, concerted efforts should be made to strengthen the public health system, especially the primary healthcare facilities and door-step services for the needy.28,29 Though, the Southern region contains some of the best-performing states in terms of almost all the health indicators in the country, the younger population was found to have poor treatment-seeking behaviour for hypertension in our study. Hence, it is important to explore the reasons behind such finding through qualitative survey. A major strength of the study is analysing data from the nationally representative survey with a higher response rate to determine the estimate of hypertension cascade of care amongst young hypertensives. This increases the generalisability of the results as the sample was larger and representative of the younger population in India. The current study contributes to the limited evidence available regarding the treatment-seeking behaviour among the younger hypertensives. Current study also has certain implications for public health practice. The current study provides valuable insights into the treatment-seeking behaviour among young hypertensives in India. It emphasizes the importance of identifying the adolescents and younger adults at high-risk of dropping out from the treatment. As most health promotion and screening strategies are targeted towards the middle-aged and elderly population, there is a need to develop newer strategies or adopt successful strategies from other countries to reduce the dropout and improve the treatment initiation among younger adults. Life course approach is one such strategy, which was proven to be effective in preventing the development of NCDs and emphasizes about the importance of early diagnosis and initiation of treatment. Interventions implemented at schools, colleges and workplaces are important to reduce the dropout rate, targeting adolescents and younger adults. However, further interventional study needs to be done to help devise newer interventions for improving their treatment-seeking behaviour. ## Limitations of the study The study's limitations were the cross-sectional design, which makes it difficult to infer the causal relationship. The data on diagnosis of young hypertension and their treatment-seeking behaviour were collected during the household survey without cross-checking the details from the health facility. Hence, there can be an underestimation of the known case of hypertension or their treatment-seeking behaviour. We also could not explore the role of certain important factors like perceived susceptibility, perceived severity across the selected age groups in poor treatment-seeking behaviour, perceived need for seeking help, etc., because of data limitations.23 ## Conclusion The current study found that more than 1 in 10 younger adults in India have hypertension. Only half of these hypertensive participants were aware of their hypertension status and only one-fifth of these sought treatments and were currently on medication. Adolescents, males, Hindus, urban population, participants with higher educational qualification and belonging to Northern, Central and Southern region had poor treatment-seeking behaviour for hypertension. Hence, a holistic approach to screen all adolescents and younger adults should be made for early diagnosis and early initiation of treatment to prevent the development of complications in the future. ## Conflict of interest None declared. ## Ethics and consent The study was ethically approved from the Institute’s Ethical Committee, Postgraduate Institute of Medical Education and Research (PGIMER), Chandigarh (PGI/IEC/$\frac{2019}{002357}$). ## Funding The Resource Centre funded this study for Cardiovascular Health under the project ‘Strengthening Management of Hypertension Services’ through Capacity Building, media and communication and stakeholder engagement program. The funding agency obtained authorization to use the data for conducting the study. ## References 1. 1. World Health Organization . Deaths from Cardiovascular Diseases and Diabetes [Internet]. WHO [cited 2021 July 1] Available from: http://www.who.int/gho/ncd/mortality_morbidity/cvd/en/. Deaths from Cardiovascular Diseases and Diabetes (2021) 2. Ghaffar A, Reddy KS, Singhi M. Burden of non-communicable diseases in South Asia. Br Med J. (2004) 328 807-10. PMID: 15070638 3. Mackay J, Mensah G. Atlas of Heart Disease and Stroke (2004) 4. Krishnan A, Gupta V, Ritvik NB. How to effectively monitor and evaluate NCD programmes in India. Indian. J Community Med. (2011) 36 S57-62 5. Kaur P, Kunwar A, Sharma M. India hypertension control initiative-hypertension treatment and blood pressure control in a cohort in 24 sentinel site clinics. J Clin Hypertens (Greenwich). (2021) 23 720-9. PMID: 33369074 6. Joshi SR, Saboo B, Vadivale M. SITE investigators prevalence of diagnosed and undiagnosed diabetes and hypertension in India: results from the screening India's twin epidemic (SITE) study. Diabetes Technol Ther (2012) 14 8-15. PMID: 22050271 7. Sukor N. Clinical approach to young hypertension. Brunei Int Med J. (2013) 9 82 8. Kamath SA. Young hypertensive: how and how much to investigate?. Med Update. (2008) 18 570-7 9. Anchala R, Kannuri NK, Pant H. Hypertension in India: a systematic review and meta-analysis of prevalence, awareness, and control of hypertension. J Hypertens. (2014) 32 1170. PMID: 24621804 10. Midha T, Krishna V, Shukla R. Correlation between hypertension and hyperglycemia among young adults in India. World J Clin Cases. (2015) 3 171. PMID: 25685764 11. Reddy VS, Jacob GP, Ballala K. A study on the prevalence of hypertension among young adults in a coastal district of Karnataka, South India. Int J Healthcare Biomed Res. (2015) 3 32-9 12. Zafar KS, Ram VS, Kumar M. The prevalence of hypertension among young adults in a rural population of North India. Int J Res Med Sci. (2017) 5 4869 13. Yuvaraj BY, Nagendra Gowda MR, Umakantha AG. Prevalence, awareness, treatment, and control of hypertension in rural areas of Davanagere. Indian J Community Med. (2010) 35 138. PMID: 20606939 14. National Family Health Survey – 4. (2017) 15. 15. Ministry of Health and Family Welfare . Screening, Diagnosis, Assessment, and Management of Primary Hypertension in Adults in India: Quick Reference Guide. [Internet]. WHO May 2016 (18 November 2021, date last accessed) Available from: https://nhm.gov.in/images/pdf/guidelines/nrhm-guidelines/stg/Hypertension_full.pdf. Screening, Diagnosis, Assessment, and Management of Primary Hypertension in Adults in India: Quick Reference Guide (Internet) 16. 16. Ministry of Health and Family Welfare . National Multisectoral Action Plan for Prevention and Control of Common Noncommunicable Diseases (2017–22) [Internet]. WHO [cited 2021 Nov 18] Available from: https://main.mohfw.gov.in/sites/default/files/National%20Multisectoral%20Action%20Plan%20%28NMAP%29%20for%20Prevention%20and%20Control%20of%20Common%20NCDs%20%282017-22%29_1.pdf. National Multisectoral Action Plan for Prevention and Control of Common Noncommunicable Diseases (2017–22) (2021) 17. James PA, Oparil S, Carter BL. 2014 evidence-based guideline for the Management of High Blood Pressure in Adults: report from the panel members appointed to the eighth Joint National Committee (JNC 8) 2014 guideline for Management of High Blood Pressure 2014 guideline for Management of High Blood Pressure. JAMA. (2014) 311 507-20. PMID: 24352797 18. Prasad M, Flowers E, Mathur A. Effectiveness of a community screening program for metabolic syndrome and cardiovascular risk factor identification in young South Asians adults. Diabetes Metab Syndr Clin Res Rev. (2015) 9 38-41 19. Rao CR, Kamath VG, Shetty A. High blood pressure prevalence and significant correlates: a quantitative analysis from Coastal Karnataka. India. ISRN Prev Med. (2013) 2013 574973. PMID: 24967139 20. Shukla AN, Madan T, Thakkar BM. Prevalence and predictors of undiagnosed hypertension in an apparently healthy western indian population. Adv Epidemiol. (2015) 2015 649184 21. Aggarwal A, Aggarwal S, Sarkar PG, Sharma V. Predisposing factors to premature coronary artery disease in young (age ≤ 45 years) smokers: a single center retrospective case control study from India. J Cardiovasc Thorac Res. (2014) 6 15-9. PMID: 24753826 22. Camelli S, Bobrie G, Postel-Vinay N. Prevalence of secondary hypertension in young hypertensive adults. J Hypertens. (2015) 33Suppl 1:e47 23. Gulliver A., Griffiths K.M., Christensen H.. Perceived barriers and facilitators to mental health help-seeking in young people: a systematic review. BMC Psychiatry. (2010) 10 113. PMID: 21192795 24. Galdas PM, Cheater F, Marshall P. Men and health help-seeking behaviour: literature review. J Adv Nurs. (2005) 49 616-23. PMID: 15737222 25. Addis ME, Mahalik JR. Men, masculinity, and the contexts of help seeking. American Psychologist. (2003) 58 5-14. PMID: 12674814 26. Engaging men and boys in changing gender-based inequity in health: evidence from programme interventions. Geneva (2007) 27. Loenen T, Berg MJ, Faber MJ, Westert GP. Propensity to seek healthcare in different healthcare systems: analysis of patient data in 34 countries. BMC Health Serv Res. (2015) 15 28. Palepu S, Yadav K, Ahamed F. Acute morbidity profile and treatment seeking behaviour among people residing in an urban resettlement colony in Delhi, India. Nepal J Epidemiol. (2018) 8 716-24. PMID: 30867975 29. Bhojani U, Thriveni B, Devadasan R. Out-of-pocket healthcare payments on chronic conditions impoverish urban poor in Bangalore. India. BMC Public Health. (2012) 12 990. PMID: 23158475
--- title: Intermittent F-actin Perturbations by Magnetic Fields Inhibit Breast Cancer Metastasis authors: - Xinmiao Ji - Xiaofei Tian - Shuang Feng - Lei Zhang - Junjun Wang - Ruowen Guo - Yiming Zhu - Xin Yu - Yongsen Zhang - Haifeng Du - Vitalii Zablotskii - Xin Zhang journal: Research year: 2023 pmcid: PMC10017101 doi: 10.34133/research.0080 license: CC BY 4.0 --- # Intermittent F-actin Perturbations by Magnetic Fields Inhibit Breast Cancer Metastasis ## Abstract F-actin (filamentous actin) has been shown to be sensitive to mechanical stimuli and play critical roles in cell attachment, migration, and cancer metastasis, but there are very limited ways to perturb F-actin dynamics with low cell toxicity. Magnetic field is a noninvasive and reversible physical tool that can easily penetrate cells and human bodies. Here, we show that $\frac{0.1}{0.4}$-T 4.2-Hz moderate-intensity low-frequency rotating magnetic field-induced electric field could directly decrease F-actin formation in vitro and in vivo, which results in decreased breast cancer cell migration, invasion, and attachment. Moreover, low-frequency rotating magnetic fields generated significantly different effects on F-actin in breast cancer vs. noncancerous cells, including F-actin number and their recovery after magnetic field retrieval. Using an intermittent treatment modality, low-frequency rotating magnetic fields could significantly reduce mouse breast cancer metastasis, prolong mouse survival by 31.5 to $46.0\%$ ($P \leq 0.0001$), and improve their overall physical condition. Therefore, our work demonstrates that low-frequency rotating magnetic fields not only can be used as a research tool to perturb F-actin but also can inhibit breast cancer metastasis through F-actin modulation while having minimum effects on normal cells, which reveals their potential to be developed as temporal-controlled, noninvasive, and high-penetration physical treatments for metastatic cancer. ## Introduction Cytoskeleton is a fibrous network of protein filaments, including microtubules, actin, and intermediate filaments, which are not only responsible for cell morphology maintenance and organelle position but also key components for multiple cellular processes, including membrane trafficking, signal transduction, and cell migration and division. As one of the major components of cytoskeleton, actin is the most abundant cellular protein in most types of eukaryote cells. G-actin (globular actin) monomers in cells can polymerize into F-actin (filamentous actin), which is mainly found in the cellular cortex, stress fibers, and pseudopodia. It has been known that in almost all steps of cancer metastatic spread, the actin reorganization and reassembly are involved [1]. Since metastasis is the leading cause of cancer patient lethality, there are numerous efforts on developing treatment method to interfere with actin and its regulators in cancer research. For example, inhibition of the actin-bundling protein fascin effectively inhibits cancer metastasis in mice [2,3], indicating the effectiveness of targeting of actin regulation proteins for anticancer therapy. In the last few decades, physical methods have been rapidly developing strategies for alternative cancer treatment because they have multiple advantages, including low toxicity and high temporal control. For example, tumor-treating field utilizes low-intensity medium-frequency alternating electric fields to interfere with the mitotic spindles in dividing cancer cells and has been approved by Food and Drug Administration to be used on glioblastoma [4–6]. Compared to electric field, magnetic field is less invasive and can provide better tissue penetration. It has been shown that even a weak magnetic field could affect new tissue formation in vivo [7] and a combination of static magnetic and electric field or a static magnetic field alone could both alleviate type 2 diabetes in mice [8,9]. Moreover, magnetic field has also been shown to be able to affect both purified actin and cellular actin [10–21]. For example, the 4-Hz oscillating magnetic fields could interrupt F-actin network in mesenchymal stem cells to inhibit their differentiation [21]. This is consistent with the well-known fact that F-actin is sensitive to mechanical stimuli by driving mechanical forces into biochemical signaling [22–25]. Here we chose 0.1-T and 0.4-T 250 r/m (~4.2 Hz) moderate intensity low-frequency rotating magnetic fields (LF-RMFs) to investigate their effects on breast cancer metastasis, the most common invasive cancer and the second leading cancer death in women. Combining multiple in vitro and in vivo experiments, comparing 2 breast cancer and 2 noncancer cell lines, as well as theoretical calculations, we found that LF-RMFs can directly but differentially affect F-actin in cancer vs. noncancer cells, which consequently decreases breast cancer cell attachment, migration, and invasion, as well as the reduced breast cancer metastasis in mice. ## LF-RMFs reduce F-actin in breast cancer cells To examine the effect of LF-RMFs on breast cancer cells, we custom-designed 3 sets of cell incubators to be used on 3 instruments: a sham control, a 0.1-T (Max) LF-RMF, and a 0.4 T (Max) LF-RMF (Fig. 1A to C and Fig. S1). Breast cancer MDA-MB231 and MCF-7 cells were plated to allow attachment before they were exposed to sham, 0.1-T LF-RMF, or 0.4 T LF-RMF. Immunofluorescence was performed after 6 h of exposure, which shows that the actin stress fibers were obviously disrupted by LF-RMFs, especially at 0.4 T (Fig. 1D and E and Fig. S2). Moreover, we quantified the F-actin filament number at the cell periphery vs. at the surrounding areas (Fig. 1F) and found that the F-actin in the cell center seems to be more obviously affected compared to the surrounding area (Fig. 1G). We also examined different time points at 5 and 20 min, and it is obvious that the F-actin in MDA-MB231 cells can be disrupted by LF-RMFs at as early as 5 min (Fig. 1H). In contrast, the microtubules in MDA-MB231 cells were not obviously affected by 0.4-T LF-RMF (Fig. S3). Fig. 1.: LF-RMFs reduce F-actin in breast cancer cells. (A) Illustration of experimental setup and design of the customized cell incubator. (B) Magnetic field intensity on the top of the “0.4-T LF-RMF” equipment. Red circle indicates where the cell culture dish was placed. (C) Magnetic field intensity changes at where the cells are. (D) Immunofluorescence images of MDA-MB231 cells after 6 h of sham or 0.1-T or 0.4-T 4.2-Hz RMF exposure. (E) Quantification of F-actin filament length. The length of 500 F-actin filaments was measured. Data are represented as means ± SD. (F) Areas were defined as center of the cells vs. surrounding areas. (G) Quantification of F-actin numbers in the center of the cells vs. surrounding areas. Data are represented as means ± SEM. (H) Immunofluorescence images of MDA-MB231 cells after sham or 0.1-T or 0.4-T 4.2-Hz RMF exposure for 5 or 20 min.**P < 0.0001. Scale bars: 20 μm. ## Time- and cell type-dependent effects of LF-RMFs on cellular F-actin Since the actin distribution seemed different when the cells were treated with LF-RMFs for 6 h (Fig. 1D) vs. 5 or 20 min (Fig. 1H), we used a series of different time points to examine the effects of LF-RMFs on cellular actin. Using MDA-MB231 cells, it is obvious that the stress fibers were significantly reduced as early as 5 min (Fig. 2A), but there is an obvious switch from stress fibers to lamellipodia/ruffles from 10 to 60 min. At 90 min, the cells shrank and rounded up, similar to the effects of other actin drugs. We also tested 2 human noncancer cell lines, the normal breast MCF10A cells (Fig. 2B) and retinal pigment epithelial RPE1 cells (Fig. S4). The results show that the stress fibers in MCF10A and RPE1 cells are much more robust than MDA-MB231 because they are more resistant to LF-RMF perturbations. Moreover, they did not have the stress fibers to lamellipodia/ruffles switch or the roundup phenotypes at a later stage (Fig. 2B and Fig. S4) as did MDA-MB231 cells. Fig. 2.: Time- and cell type-dependent effects of LF-RMFs on cellular F-actin. (A) MDA-MB231 breast cancer cells and (B) MCF10A breast noncancer cells were treated with sham, 0.1-T RMF, or 0.4-T RMF for different time points. Cells were fixed and stained with phalloidin and DAPI. Scale bar: 20 μm. Next, we further side-by-side compared 3 different cell types—MDA-MB231 cells, MCF10A cells, and RPE1 cells—for different treatment time and/or recover time after magnetic field retrieval (Fig. 3). It is apparent that the F-actin changes were very obvious in the breast cancer MDA-MB231 cells, but not noncancer MCF10A cells or RPE1 cells (Fig. 3A to C and Fig. S5), confirming that the MDA-MB231 cells are much more sensitive to LF-RMFs compared with the 2 noncancerous cells. We also examined the recovery after 1.5 h of 0.4-T LF-RMF treatment. Our results show that the obvious cellular actin changes in MDA-MB231 caused by LF-RMF did not recover even after 24 h after LF-RMF retrieval, while the actin changes in MCF10A and RPE1 cells can soon recover (Fig. 3D to F and Fig. S6). Fig. 3.: LF-RMFs differentially and reversibly affect F-actin in cancer vs. noncancerous cells. (A) MDA-MB231, (B) MCF10A, and (C) RPE1 cells were treated with 0.1- or 0.4-T LF-RMFs for 5 min or 0.4 T LF-RMF for 60 min. (D) MDA-MB231, (E) MCF10A, and (F) RPE1 cells were treated with 0.4-T LF-RMF for 1.5 h, with or without recovery for 12 or 24 h. Cells were stained with phalloidin and DAPI. Scale bar: 20 μm. To further examine the differential sensitivity of actin fibers in cancer vs. noncancer cells to perturbations, we used different concentrations of cytochalasin D (cytoD), as well as different recovery times after drug retrieval. It seems that cytoD induced a more significant effect on MDA-MB231 and MCF-7 breast cancer cells compared with the noncancer RPE1 cells (Fig. 4A and Fig. S7). Moreover, the RPE1 cells can fully recover after drug retrieval but not the MDA-MB231 and MCF-7 breast cancer cells (Fig. 4B and Fig. S8). These confirm that the F-actin in breast cancer cells is much more sensitive to drug or magnetic field perturbations compared with the noncancer cells. Fig. 4.: Cytochalasin D (cytoD) differentially and reversibly affects F-actin in cancer vs. noncancerous cells. (A) MDA-MB231 and RPE1 cells were treated with 0, 50, or 100 nM cytoD for 2 h before they were fixed and stained with phalloidin (red) and DAPI (blue). (B) MDA-MB231 and RPE1 cells were treated with 100 nM cytoD for 2 h, with or without additional washout to allow recovery for 12 or 24 h, before they were fixed and stained with phalloidin and DAPI. Scale bar: 20 μm. ## LF-RMFs significantly inhibit MDA-MB231 breast cancer cell migration, invasion, attachment, and spreading in vitro Since F-actin is essential for cell migration and invasion, we did 3 in vitro experiments to test whether LF-RMFs can affect MDA-MB231 breast cancer cell migration and invasion. We used the wound healing migration assay (Fig. S9), Transwell migration assay (Fig. 5A) and Transwell invasion assay with Matrigel (Fig. 5B), a mixture of laminin and collagen, to provide a 3-dimensional cell culture condition. It is obvious that both 0.1-T and 0.4-T LF-RMFs have significant inhibition effects on MDA-MB231 cell migration and invasion (Fig. 5A and B and Fig. S9). Fig. 5.: LF-RMFs significantly inhibit MDA-MB231 breast cancer cell migration, invasion, attachment and spreading in vitro. (A) Experimental illustration, representative images, and quantification of relative numbers of migrated cells in Transwell migration assays in vitro. Data are represented as means ± SEM. (B) Transwell invasion assays with Matrigel. Data are represented as means ± SEM. (C) Immunofluorescent images of MDA-MB231 cells stained with phalloidin (green), anti-tubulin antibody (red), and DAPI (blue) after 6 h of RMF treatment. Scale bar: 20 μm. (D) Quantification of cell area assessed through phalloidin staining in MDA-MB231 cells after 6 h of sham or RMF treatment. Both MDA-MB231 and MCF-7 cells were quantified. Data are represented as means ± SD. (E) Quantification of relative attached cell number after 18 h of sham or RMF treatment. Data are represented as means ± SD. (F) Quantification of cell attachment experiments using fibronectin (Fn)-coated plates show that LF-RMFs reduced both MDA-MB231 and MCF-7 breast cancer cell attachment after 18 h of treatment. Data are represented as means ± SD. Experiments have been repeated 3 times. P < 0.05, P < 0.01, P < 0.001, and **P <0.0001. Next, we examined the effects of LF-RMF on breast cancer cell attachment and spreading, in which F-actin also plays a central role. We exposed MDA-MB231 and MCF-7 breast cancer cells to LF-RMFs right after they were plated on coverslips in the tissue culture plates. After 6 h, the cells were fixed and stained. We observed that both MDA-MB231 and MCF-7 (Figs. 5C and S2) had decreased attachment and spreading after LF-RMF treatment, especially the 0.4 T. Quantification of the cell area shows that both 0.1-T and 0.4 T LF-RMFs reduced the cell area (Fig. 5D). In addition, we tested the relative attached cell numbers after 18 h of LF-RMF treatment and found that the attached cell numbers were also reduced (Fig. 5E). Moreover, to mimic the in vivo cell attachment condition, we tested cell attachment on surfaces coated with one of the extracellular matrix proteins, fibronectin (Fn). We found that the cell attachment on Fn-coated surface was also reduced in both types of breast cancer cell lines by LF-RMFs (Fig. 5F). Therefore, our results indicate that MDA-MB231 and MCF-7 cell attachment can be reduced by 0.1-T and 0.4-T LF-RMFs, especially the 0.4-T LF-RMF. ## LF-RMFs inhibit breast cancer metastasis, prolong survival time, and improve physical conditions of mice bearing MDA-MB231 breast cancer cells Since our cellular assays show that multiple processes involved in cancer metastasis are inhibited by LF-RMFs, including cell migration, invasion, attachment, and spreading, we next investigated whether LF-RMFs can affect breast cancer metastasis in vivo. The LF-RMF acts on both cancer cells and normal cells. Although the normal cells are much less sensitive to this perturbation, they are also affected to some extent. However, the normal cells have a much better recovery ability after magnetic field retrieval compared with the cancer cells (Fig. 3D to F and Fig. S6). We hypothesize that using intermittent treatment, we can achieve a maximum perturbation on the cancer cells and a minimal effect on the normal cells. We used female BALB/c (nu/nu) nude mice and injected 5 × 106 MDA-MB231 cells into their tail veins to develop breast cancer cell metastasis mice model. The mice were then treated with sham, 0.1-T LF-RMF, or 0.4-T LF-RMF for 6 h/d, for 136 d (Fig. 6A and B) ($$n = 12$$ for each group). Their body weight (Fig. S10) and water and food consumption (Fig. 6C) were measured every day, and their blood samples were collected for blood routine test on the 45th day after intravenous injection (Fig. S11). No statistical abnormality in the blood was revealed, but we observed obvious improvement of their physical conditions after LF-RMF treatment. To get a more accurate evaluation, we performed multiple behavior tests, including the balance beam test, grip test, and open-field test, which show that LF-RMFs could improve the motor coordination, muscular strength, and exploratory activity of MDA-MB231–bearing mice (Fig. S12). These results and the increased food consumption of LF-RMF-treated mice indicate that LF-RMFs could improve the overall health conditions of these breast cancer cell metastatic mice. Fig. 6.: LF-RMFs significantly inhibit MDA-MB231 breast cancer metastasis and prolonged the mice survival time. (A) Magnetic field induction as a function of time. (B) Experimental procedure. MDA-MB231 cells were injected intravenously through the tail vein. (C) Average daily food and water intake of mice in different groups. Data are represented as means ± SD. (D) LF-RMFs significantly prolonged survival time of mice bearing MDA-MB231 breast cancer cells. Data are represented as means ± SEM. (E) LF-RMFs significantly reduced number of tumor metastasis nodules on the lung tissues. The lung tissues collected before day 140 were analyzed for metastasis module numbers. For sham control, n = 11; for 0.1 T, n = 6; for 0.4 T, n = 5. Data are represented as means ± SEM. (F) Representative lung section HE staining shows that LF-RMF-treated mice have relative normal lung tissues. (G) Representative lung section Ki-67 staining shows that LF-RMF-treated mice have much reduced proliferating cancer modules in their lung tissues. P < 0.001, *P < 0.0001; ns, not significant. After 136 d of LF-RMF treatment, the mice were fed normally, without more LF-RMF treatment, until all mice in the sham control group died before day 165 (Fig. 6D). However, there were still 4 and 6 mice alive in the 0.1-T and 0.4-T RMF-treated groups at that time, respectively. We continued to feed them and monitored their growth until day 185 before the remaining mice were sacrificed. Statistical results show that the 0.1-T LF-RMF has prolonged the mice survival time by $31.5\%$ (average survival time of 0.1-T LF-RMF-treated mice is 129.33 d vs. 98.33 d for the sham control; $$n = 12$$; $P \leq 0.0001$ by Student t test) and the 0.4-T LF-RMF has prolonged the mice survival time by $46.0\%$ (average survival time of 0.4-T LF-RMF-treated mice is 143.58 d vs. 98.33 d for the sham control; $$n = 12$$; $P \leq 0.0001$ by Student t test) (Fig. 6D). Since MDA-MB231 breast cancer cells were injected into these mice to induce metastasis, we examined the mice tissues to see whether LF-RMFs affect their metastatic condition. It is obvious that the number of metastatic nodules on the lungs of these mice was significantly reduced by LF-RMFs (Fig. 6E). We further examined the lung tissues with HE (hematoxylin and eosin) (Fig. 6F), and Ki-67 staining (Fig. 6G), a commonly used marker for cell proliferation in cancer. Our HE staining results show that both 0.1-T and 0.4-T LF-RMFs could reduce the lung metastasis, while the 0.4-T LF-RMF has a more significant effect than the 0.1-T one (Fig. 6F). The lung tissues in the 0.4-T-treated mice had remained their normal appearance. Moreover, 0.4 T has a more significant effect than 0.1 T in reducing the Ki-67 staining in lung tissues (Fig. 6G). We have also examined 3 additional cancer markers, the proliferating cell nuclear antigen, epidermal growth factor receptor, and Vimentin, which all confirmed that the LF-RMFs could inhibit the breast cancer metastasis in lung tissues, and the 0.4-T LF-RMF had a more significant effect than the 0.1-T LF-RMF (Fig. S13). ## RHO activity and cell cycle progression were disrupted by LF-RMFs in breast cancer cells It has been reported by Wosik et al. [ 20] that the magnetic force produced by medium intensity gradient static magnetic field could cause the rearrangement of actin cytoskeleton of macrophages and change cell morphology, which was similar to Rho inhibition. Since Rho guanosine triphosphatases (GTPases) play important roles in cell adhesion, migration, and invasion [26–28], we next examined whether LF-RMF could affect Rho guanosine triphosphatase activity in breast cancer cells. We first performed Rho activation assays after treating MCF-7 cells for 1 or 3 h but did not observe obvious changes (Fig. 7A). However, when we prolonged the treatment time to 4.5 h, the Rho activity was reduced, which is even more obvious at 6 h (Fig. 7B). We also treated MDA-MB231 with LF-RMF for 4.5 h and observed reduced Rho activity as well (Fig. 7C). In addition, we examined some cell attachment- and invasion-related markers and found that the focal adhesion structural protein Talin and Tensin 2 levels were both decreased by LF-RMF treatment (Fig. 7D). Moreover, not surprisingly, since LF-RMFs affect actin and related signaling pathways, they also increase the binucleated cells (Fig. 7E) and arrest cell cycle progression in MDA-MB231 cells (Fig. 7F). Fig. 7.: RHO activity and cell cycle progression were disrupted by LF-RMFs in breast cancer cells. (A to C) RHO activation assay shows that RMF could reduce RHO activation in MCF7 and MDA-MB231 cells after 4.5 h. Neg, negative control. Pos, positive control. (D) Representative Western blots and the quantification show that Talin and Tensin 2 levels are reduced by LF-RMF treatment in MCF-7 cells. Data are represented as means ± SEM. n = 3. (E) Quantification of the percentage of binucleated and multinucleated MDA-MB231 cells. Data are represented as means ± SEM. (F) The cell cycle distribution after being treated with sham, 0.1-T LF-RMF, or 0.4-T LF-RMF for 30 h. Data are represented as means ± SD. P < 0.05 and *P < 0.01. ## LF-RMFs directly affect F-actin polymerization and depolymerization Since LF-RMFs can affect cellular F-actin in as short as 5 min (Figs. 1 to 3), while the RHO activity needs a much longer time (Fig. 7A to C), we hypothesize that actin may be the direct target of LF-RMFs. Next, we performed in vitro actin polymerization assays using pyrene-actin to measure its dynamics and used transmission electron microscopy (TEM) to observe their assembled structures (Fig. 8A to C). Because of the technical limitations, we cannot monitor this polymerization process while the reactions were exposed to LF-RMF, so we initiated the actin polymerization reactions in the presence of sham, 0.1-T LF-RMF, or 0.4-T LF-RMF for 5 min and then used a fluorescent spectrophotometer to monitor the formation process of F-actin fibers, which was reflected by the fluorescence level of pyrene-actin. Our results show that even the short treatment of actin polymerization reaction by 0.4-T LF-RMF for only 5 min was enough to reduce the level of F-actin polymerization (Fig. 8B), which is consistent with our previously observed cellular F-actin changes in cells (Figs. 1 to 3). Fig. 8.: LF-RMFs directly affect actin polymerization in vitro. (A) Experimental procedures illustration. (B) Pyrene-actin assay shows that the 0.4-T LF-RMF treatment for 5 min can directly inhibit actin polymerization in vitro. (C) Representative TEM images of F-actin assembled in sham, 0.1-T LF-RMF, or 0.4-T LF-RMF. Scale bar: 200 nm. (D) DLS shows the particle size distribution of actin after 10 min of assembly in sham, 0.1-T LF-RMF, or 0.4-T LF-RMF. Data are represented as means ± SEM. **P < 0.0001. (E) Actin polymerization with or without electric field. The red arrows represent the electric dipole moments of G-actin. Blue arrows are the electric field lines. Shadow area is an F-actin domain with parallel orientation of the dipole moments. In this area, the G-actin binding energy is reduced by the dipole-dipole interaction energy. (F) Sketch of a fraction of F-actin containing a domain with parallel orientation of dipole moments of G-actin monomers (4 central green circles). Calculated vector field, E, of 8 electric dipoles (↑↓↑↑↑↑↓↑) is shown by the blue arrows. In the domain vicinity, the energy of the stray electric field weakens the G-actin binding energy in the chain. On the horizontal scale, unity corresponds to the length of 1 G-actin monomer, 2.7 nm. To confirm the effects of RMFs on actin polymerization, we used TEM and dynamic light scattering (DLS) to examine the actin polymerization in the absence or presence of RMFs for 10 min. Both experiments show that F-actin formation is inhibited by RMF treatment (Fig. 8C and D). In the TEM experiment, the actin filaments are abundant and robust in the sham group, which were greatly reduced by RMF treatment (Fig. 8C). In the DLS experiment, the total particle size in the sham group has the highest average size (573.77 nm), which was significantly reduced by RMF treatment (Fig. 8D). We hypothesize that RMF-induced electric field may cause these effects. When a cell dish is placed in the RMF, an electromotive force V rises, which reads asV=−d(BS)dt=−πRrωB[1]where B is the magnetic induction of the RMF, $R = 20$ cm is the rotation radius, $r = 9$ mm is the cell dish radius, and ω = 2πν is the angular velocity of rotation. The derivative of the magnetic flux with respect to time refers to how the magnetic field changes with the time. The cells we used are attached to the bottom of the tissue culture plate so that they can be considered as flat 2-dimensional disks. By definition V=∮E→dl→=2πrE, from which, considering Eq. 1, the induced electric field amplitude, E reads asE=πνBR[2] Estimation of Eq. 2 for ν = 4.2 Hz, $R = 20$ cm, and $B = 0.4$ T gives $E = 1.1$ V/m. It is known that the building block G actin polymerizes into semiflexible F-actin filaments (Fig. 8E). When actin polymerizes in the presence of electric field, there is a torque on the actin dipole moment, M→=p→E→, which tends to align the G-actins with the field and can lead to the appearance of domains with the parallel orientation of the dipole moments of G-actin (Fig. 8E). To analyze the role of a domain with the parallel-oriented dipole moments in F-actin stability, we calculated the distribution (vector field) of the electric field of the dipole chain consisting of 8 G-actin monomers oriented as shown on Fig. 8F. Here, the domain of 4 G-actin monomers, with parallel orientation of the dipole moments, generates a local stray electric field, E (the blue quasi-parallel field lines) with a positive energy that is proportional to E2, thereby weakening the negative binding G-G actin energy. The calculations were performed with the Wolfram Mathematica software. To evaluate the contribution of the domain formation in F-actin, we can consider 2 parallel dipoles with moment $$p \leq 75$$ Debye = 75⋅3.336 × 10−30 C $m = 2.5$ × 10−28 C m, separated by the distance $l = 2.7$ nm (monomer length). Their interaction energy isWdip=p24πε0l3[3]where ε0 is the permittivity of free space. Estimations of Eq. 3 give Wdip = 2.85 ⋅10−20 J, which is larger than the thermal fluctuation energy, kBT (where kB is the Boltzmann constant and T is the temperature): Wdip/ kBT = 6.9. Thus, these dipoles (G-actin monomers) repulse each other with the energy 17.16 kJ/mol (=Wdip NA, where NA is the Avogadro’s number). The GG actin repulsion force can be estimated as FGG ≈ Wdip/$l = 10$ pN. Importantly, a force of the same order of value (15 to 27 pN) can significantly change both the bond lifetimes of G-actin–G-actin (GG) and G-actin–F-actin (GF) interactions [29]. For the effect of the GG dipole actin interaction on the binding energy, since the binding energy of G-actin monomers is W = −27.81 kJ/mol [30], the effective binding energy is reduced by G-actin dipole-dipole repulsion and reads asWef=W+p2NA4πε0l3≈−10.21kJ/mol[4] This implies that the ratio of the effective binding energy to the thermal fluctuation energy is Wef/kBT = 4.1. Therefore, the bond survival probability is decreased by the GG dipole interaction. Thus, the estimated contribution of the dipole-dipole interaction to the binding energy of actin (Eq. 4) allows us to conclude that in the presence of an electric field induced by the RMF, F-actin becomes less stable. In other words, in the presence of the RMF, an F-actin assembly operates just above the thermal fluctuation energy limit, Wef/kBT = 4.1, and a small mechanical energy input or a force impulse will disturb the stability of F-actin filaments by breaking it into parts. ## Discussion The actin dipole moments are randomly oriented when there is no electric field, while an electric field tends to align them parallelly. Here, we show that 0.1- and 0.4-T, 4.2-Hz LF-RMF-induced electric field could directly decrease F-actin stability. The estimated change of the actin binding energy provoked by the LF-RMF is sufficient to destabilize F-actin polymerization process. This destabilization leads to F-actin shortening and polymerization dynamics changes, which consequently reduces the ability of cells to adhere to a surface and spread. In vivo, these effects manifest themselves in reducing breast cancer metastasis and increasing the survival rate of tumor-bearing mice (Fig. 9). Fig. 9.: A model illustrates that LF-RMFs decrease F-actin to inhibit breast cancer cell attachment, migration, and invasion, which reduces breast cancer metastasis in vivo. It is interesting that we found that the cellular actin filaments in breast cancer cells are much more sensitive and respond to LF-RMFs very differently from noncancerous cells. First of all, the MDA-MB231 cancer cells are much more sensitive to LF-RMFs, whose stress fibers are obviously reduced as early as 5 min after LF-RMF treatment. Secondly, the MDA-MB231 cancer cells have time-dependent stress fiber to lamellipodia switch, while the 2 noncancerous cells do not. This is consistent with the metastatic characteristic of the MDA-MB231 cancer cells. Thirdly, the actin abnormalities and cell shape changes in the MDA-MB231 cancer cells are not recoverable at 24 h after treatment, while the subtle changes in the 2 noncancerous cells can fully recover. Therefore, when used intermittently, LF-RMF could interrupt F-actin in breast cancer cells while having minimum effects on noncancerous cells. Compared to current actin drugs, many of which have obvious cell toxicity, the high-temporal control, low-toxicity, and high-penetration characteristics of magnetic fields make them a superior research tool to investigate actin dynamics in cells and animals and inhibit breast cancer metastasis and maybe other types of cancer as well. Because of the intrinsic diamagnetic anisotropy of peptide bond [31], the diamagnetic anisotropy of filamentous cytoskeletons can be additive so that they are generally susceptible to external magnetic fields, both in vitro [32,33] and in cells [34–39]. Multiple studies have shown that magnetic fields can target the microtubules and mitotic spindle [37,38,40–45]. F-actin is known to be sensitive to mechanical stimuli and plays essential roles in multiple physiological and pathological processes, including cancer metastasis. High-magnetic field-induced F-actin changes have been shown to affect cell behavior [46]. On the basis of theoretical and experimental evidences [21,47], the 4-Hz oscillating magnetic fields could interrupt F-actin network in mesenchymal stem cells to inhibit their differentiation [21], which is consistent with a previous study showing that the relaxation time of F-actin network in cells is about 0.1 to 1 s [47]. Therefore, we think F-actin network may have resonance with external magnetic field of 1- to 10-Hz frequency. In fact, it has been shown in Dictyostelium cells that there is an intrinsic oscillatory process in the cellular actin system, which has a clear resonance at a stimulation period of 20 s in that specific cell type [48]. Although we only tested the frequency of 4.2 Hz (max rotation speed) in this study because of the limitation of our equipment, we hypothesize that the F-actin network of the studied cells may have resonance with external LF-RMF of 1- to 10-Hz frequencies. In fact, some LF-RMF studies using a faster rotation speed of 7.5 Hz, also at 0.4 T, have shown to have anticancer potentials in mice bearing lung cancer, melanoma, and hepatocellular carcinoma [49–51]. We hypothesize that LF-RMF-induced F-actin disruption is likely to be one of the fundamental mechanisms for these cancer metastasis inhibitions. Although we did not observe obvious microtubule cytoskeleton in this study, the effects of LF-RMFs of different intensities and frequencies should also be systematically addressed in the future. It is possible that microtubules may be responsive to magnetic field parameters that are different from the actin cytoskeleton. It should be mentioned that although magnetic fields have superior tissue penetration, it also means that they do not have a precisely control-focused area. However, since the majority of our bodies are composed of very weak diamagnetic materials, including water, lipids, and proteins, which are not sensitive to weak to moderate magnetic fields. Therefore, weak to moderate magnetic fields do not have obvious perturbations on them. More importantly, we revealed that the F-actin in noncancerous breast cells is much less sensitive than that in breast cancer cells, which indicate that the normal cells in our human bodies are less likely to be agitated by these magnetic fields. In summary, our study shows that $\frac{0.1}{0.4}$-T, 4.2-Hz RMFs could interrupt actin filament to inhibit breast cancer cell attachment and migration in vitro and reduce breast cancer metastasis in vivo. The survival rate and physical conditions of breast cancer tumor-bearing mice were also significantly improved by RMF treatment. This intermittent LF-RMF treatment (a few hours per day) has the potential to be developed into a promising cancer treatment method, which could effectively inhibit cancer metastasis while having a minimum effect on normal cells. The advantage of its high penetration through tissues, the easy removal of magnetic field at any time, as well as its local applications to specific part of the bodies make it a very unique and powerful physical tool that can afford high temporal and spatial control over a large variety of cancer types. Therefore, people are encouraged to investigate more cancer cell types, magnetic field frequencies, as well as increased magnetic field intensities to further explore the future clinical potentials of these low-frequency RMFs, which could likely lead to significantly improved therapeutic outcomes. ## Cell culture The cell lines MDA-MB231 (RRID: CVCL_0062) and MCF-7 (RRID: CVCL_0031) were from American Type Culture Collection. Cell line identity was confirmed and cultured in Dulbecco’s modified Eagle medium (DMEM) supplemented with $10\%$ fetal bovine serum (CLARK Bioscience, FB25015). ## Cell synchronization assay MDA-MB231 cells were plated on coverslips or in 35-mm cell culture dishes at 1 × 105 cells/ml for 24 h. Single thymidine was used for synchronization. Thymidine (2.5 mM; Sigma-Aldrich, T1895) was added to cells for 17 h before washed with phosphate-buffered saline (PBS) to remove thymidine. After adding fresh DMEM, the cells were treated with RMF or sham immediately for another 30 h. The treated cells were collected for cell cycle detection or immunofluorescence analysis. ## Cell cycle analysis MDA-MB231 cells blocked by single thymidine were released in fresh DMEM and immediately exposed to RMF or sham for 30 h. Cells were collected and trypsinized with $0.25\%$ trypsin/EDTA and washed with ice-cold PBS 3 times before they were fixed with −20 oC $70\%$ ethanol overnight. After fixation, the cells were washed with PBS to remove ethanol, and stained with propidium iodide solution (BD Pharmingen, San Diego, CA, USA). The data were acquired by flow cytometry (CytoFLEX, Beckman Coulter, Brea, CA, USA) and analyzed by MODFIT LT (Verity Software House, LA, CA, USA, RRID: SCR_016106). ## Immunofluorescence Cells plated on coverslips were washed with PBS and fixed in $4\%$ formaldehyde at room temperature for 20 min. After block by AbDil-Tx (TBS [$0.1\%$ Triton X-100, $2\%$ bovin serum albumin, and $0.05\%$ sodium azide]) for 30 min at room temperature, cells were subjected to immunofluorescence using β-tubulin antibody (TransGen Biotech, HC101-02) and Alexa Fluor 488-conjugated anti-mouse IgG (Invitrogen, A-21202). F-actin was stained with Alexa Fluor™ 594 phalloidin (Invitrogen, A12381) for 0.5 h at room temperature. Cells were then stained with 300 nM 4′,6-diamidino-2-phenylindole (DAPI) at room temperature for 2 min before mounted with Pro-Long Gold Antifade (Invitrogen, P36930). Images were taken using an Olympus (SpinSR10, Olympus, Tokyo, Japan) fluorescence microscope. Percentage of binuclear and multinuclear cells was quantified on the basis of the immunofluorescence images. ## Cell culture wound closure assay Cells were plated at 3× 105 cells/ml in a 3.5-cm cell culture dish and grew to $100\%$ confluence in a 37 °C, $5\%$ CO2 incubator for 24 h. Then, a vertical wound scratch throughout the middle of cell monolayer was made using a 200-μl pipette tip. The media and cell debris were carefully aspirated before replacing with culture media without fetal calf serum (FCS). An initial picture was taken immediately by an inverted microscope before the cell culture dishes were exposed to sham, 0.1-T RMF, or 0.4-T RMF for 24 h. Pictures were taken every 6 h to record the process of wound closure. ## Cell migration and invasion assay Migration of MDA-MB231cells were assessed by using 24-well chemotaxis chamber (Costar, Corning, cat# 3422) with a polycarbonate membrane of 8-μm pore size. Briefly, the cells were starved for 26 h in DMEM without FCS. Then, 100-μl starved MDA-MB231 cells in serum-free medium at a concentration of 3 × 105 cells/ml were seeded on the upper compartments. Six hundred microliters of the fresh medium with $10\%$ FCS was added into the bottom of the lower chamber. The prepared 24-well plates were respectively exposed to sham, 0.1-T RMF, or 0.4-T RMF. After 14 h of incubation, the nonmigrated cells on the upper surface of the membrane filter were removed, and the migrated cells attached to the lower surface of the filter were fixed with $4\%$ formaldehyde solution and stained by $0.2\%$ crystal purple solution. Migrated cells were quantitatively assessed by counting the number of cells under a light microscope. Relative migrated cell number was calculated by normalization to the sham control group. The cell invasion assay was performed like the migration assay with slight modification. Briefly, the membrane filters of the 24-well chemotaxis chamber were coated with Matrigel (Becton Dickinson), and the invasion time was extended to 16 h. ## Rotating magnetic field setup A pair of permanent magnets (rare-earth permanent magnet N42M) were equipped on a rotor with opposite poles facing up, which rotate at 4.2 Hz (the maximum speed of the instrument) clockwise. Both the magnets and the rotor were contained in a 44 cm × 41 cm × 52 cm (L × W × H) square box. In the sham control group, a pair of iron cubes were used instead. The distances between the magnets and the box surface were adjusted, and the maximum magnetic field intensities at locations of the cell culture plates were ~0.1 and ~0.4 T, respectively. Magnetic flux density distributions were measured by LakeShore 410 Gaussmeter (Lake Shore Cryotronics, Westerville, OH). For each group, 2 mouse cages (each containing 6 mice) were placed on the surface of the box. The magnetic flux densities on the bottom of the mice cage are within 0.01- to 0.4-T range. Mouse cages of all 3 groups—the sham group (iron cube) and the 0.1-T and 0.4-T LF-RMF groups—were put on the instruments for 6 h (9AM-3PM)/d, 7 d/week, at 4.2 Hz for 4.5 months. ## Custom-designed cell incubator for RMF We designed and constructed a set of biological sample incubation system (Fig. S1A), with accurate temperature, gas, and humidity control. A nonmagnetic stainless steel cylinder with a 300-mm outer diameter and a 200-mm inner diameter was the main part of the device, and the cylinder wall was hollow for flowing circulating water. A tube with a 10-mm inner diameter on the top cover of cylinder was used for flowing $5\%$ CO2, and two 10-mm-inner-diameter tubes on both sides of cylinder were used for flowing circulating water. An electronic thermometer was used to monitor the temperature of sample chamber, which could be controlled by thermal conduction from the circulating temperature-controlled water, which flowed the cavity of cylinder wall. The temperature of the sample chamber can be controlled at 37 °C by adjusting the water temperature. ## Measuring F-actin in the central vs. peripheral areas We used ImageJ software (ImageJ, RRID:SCR_003070) to measure the F-actin distribution in cells. Firstly, we drew 3 to 5 parallel lines perpendicular to the long axis in the cell, with both ends reach the cell edge. The length of the line segment was measured (L). Then, we drew 2 points at $\frac{1}{4}$L inward from both ends of the line segment and connected the points to form a circle. The area inside of the circle is considered as the central area of the cell, and the area outside of the circle is considered as the peripheral area of the cell. ## Pyrene-actin assay The actin mixture (rabbit skeletal muscle actin mixed with $10\%$ pyrene-actin, Cytoskeleton, Cat. # AKL99 and #AP05) was diluted to 3.0 μM with G-buffer (5 mM Tris-HCl [pH 8.0], 0.2 mM CaCl2, 0.2 mM adenosine triphosphate [ATP], and 0.5 mM dithiothreitol) and left on ice for 1 h to depolymerize actin. The reactions were then centrifuged at 14,000 rpm at 4 °C for 30 min to remove residual nucleation centers. For polymerization assays, the supernatants were pipetted to a 96-well black opaque plate and measured in a fluorescent spectrophotometer to establish a baseline. Polymerization was induced by adding 10 × polymerization buffer (500 mM KCl, 20 mM MgCl2, and 10 mM ATP) to each well and mixing, followed by sham or LF-RMF treatment at room temperature for 5 min. The plate was returned to the spectrophotometer and was scanned every 30 s for 1 h. For depolymerization assays, supernatants were prepared and polymerized at room temperature for 1 h before they were centrifuged to pellet the polymerized actin (F-actin). F-actin was suspended in F-buffer (G-buffer + $\frac{1}{10}$th the volume of 10× polymerization buffer) and added to a 96-well black plate. The baseline fluorescent was measured as described above. Depolymerization was induced by adding 2 μM Latrunculin A (Sigma-Aldrich, Cat. # L5163, RRID: SCR_004098), followed by a rotating magnetic field treatment at room temperature for 5 min. The plate was returned to the spectrophotometer and was scanned every 30 s for 1 h. The kinetics of polymerization and depolymerization were monitored using a SpectraMax i3x plate reader (Molecular Devices, San Jose, CA) with excitation and emission wavelengths of 360 and 420 nm, respectively. ## Animal assays The animal protocols were approved by the ethical and humane committee of Hefei Institutes of Physical Science, Chinese Academy of Sciences. Specific-pathogen-free-grade 4-week-old female BALB/c (nu/nu) nude mice (RRID: IMSRAPB: 4790) were from Nanjing University-Nanjing Institute of Biomedicine. Mice were maintained in a sterile environment with light, humidity, and temperature control. After feeding for a week, experimental metastasis assays were performed by injecting 5 × 106 MDA-MB231 cells/200 μl into mice tail veins. After intravenous injection of MDA-MB231 cells, we randomly divided 36 mice into 3 groups: sham, 0.1-T group, and 0.4-T group. For each group, $$n = 12$.$ The mice were treated with sham or magnetic field on the second day after intravenous injection, from 9:00 AM to 3:00 PM every day, for 136 d. Weight and daily water and food consumption were measured every day. Mouse blood samples were collected from the eye orbit on the 45th day after intravenous injection for blood routine test. After 136 d, the mice were fed normally until their natural death (for the sham control group) or until 185 d. The lung and liver tissues were collected and processed for HE and Ki-67 staining. ## Rho protein activation assays Amounts of RHOA and RHOA-guanosine triphosphate (GTP) were analyzed with the Rho Activation Assay Biochem Kit (Cytoskeleton, BK036). Cells were seeded at appropriate cell density (5 × 104 cells/ml for MDA-MB231 cells and 1 × 105 cells/ml for MCF-7 cells) in 100-mm dishes and grew for 3 d. LF-RMF treatment was performed when the cells were approximately $30\%$ confluent. After LF-RMF treatment, all media was aspirated and the dishes were placed on ice immediately. Cells were rinsed with 10 ml of ice-cold PBS buffer to remove serum before they were lysed by 250 μl of ice-cold lysis buffer containing 1× Protease Inhibitor Cocktail (Roche) on ice for 5 min. Then, the cell lysates were harvested by a cell scraper. Cell lysates were transferred into the prelabeled sample tubes on ice and clarified by centrifugation at 10,000 × g, 4 °C for 1 min. Supernatants (20 μl) were used for protein concentration measurement by a protein concentration determination kit (P0009, Beyotime Biotechnology), and a 50 μg sample was used for Western blot quantitation of total RhoA in each sample. Equivalent amount of protein lysate (800 μg) was added to a 50 μg of rhotekin-RBD beads immediately for the pull-down assay. The reactions were incubated at 4 °C on a rotator or rocker for 1 h. Subsequently, the rhotekin-RBD beads were pelleted by centrifugation at 5,000 g at 4 °C for 1 min, and $90\%$ of the supernatant was removed without disturbed the bead pellet. The beads were washed once with 500 μl of Wash Buffer and pelleted by centrifugation at 5,000 g at 4 °C for 3 min. Then, the supernatant was carefully removed without disturbing the bead pellet. Twenty microliters of 2 × Laemmli sample buffer was added to each tube to thoroughly resuspend the beads. The bead samples were boiled for 2 min and analyzed by SDS-PAGE and Western blot analysis. Eight hundred micrograms of supernatant was loaded with GTPγS as a positive control for the pull-down assay. For the negative control, supernatant was loaded with GDP in place of the GTPγS. ## Actin polymerization examined by TEM and DLS The actin proteins (rabbit skeletal muscle actin, Cytoskeleton, Cat. # AKL99) were diluted to 3.0 μM with G-buffer (5 mM Tris-HCl [pH 8.0], 0.2 mM CaCl2, 0.2 mM ATP, and 0.5 mM dithiothreitol) before they were left on ice for 1 h and centrifuged at 14,000 rpm, 4 °C for 30 min to remove residual actin oligomers. Actin polymerization was induced by adding 10 × polymerization buffer (500 mM KCl, 20 mM MgCl2, and 10 mM ATP), followed by sham or LF-RMF treatment at room temperature for 10 min before they were processed for TEM or DLS measurement. For TEM experiment, the sample were added to a 200-mesh grid (20 s for ion sputtering) and incubated for 90 s. After the grid was dried, $1\%$ uranyl acetate was added to the grid to allow incubation for 90 s. Then, the grid was stained by the dye solution 3 times before it was air-dried, and then the images were collected by a TEM (Talos F200X, FEI) operated at 200 kV. For the DLS experiment, the particle size distributions were examined by a Zetasizer Nano-ZSE Particle Analyzer (Malvern Instrument, Ltd., UK). All processes were carried out at room temperature. ## Statistical analysis Statistical significance was determined by 2-tailed paired or unpaired Student t test. Statistical data are presented as means ± SEM (standard error of the mean) or SD (standard deviation). Sample size (n) and P value are specified in the text or figure legends. To reduced conscious or subconscious experimenter bias, we performed most data analysis in a blinded way by independent researchers. Their results are pooled together for statistical analysis. ## Data Availability All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors. ## References 1. Nurnberg A, Kitzing T, Grosse R. Nucleating actin for invasion. Nat Rev Cancer (2011) 11 177-187. PMID: 21326322 2. Shan DD, Chen L, Njardarson JT, Gaul C, Ma X, Danishefsky SJ, Huang XY. Synthetic analogues of migrastatin that inhibit mammary tumor metastasis in mice. Proc Natl Acad Sci USA (2005) 102 3772-3776. PMID: 15728385 3. Chen L, Yang SY, Jakoncic J, Zhang JJL, Huang X-Y. Migrastatin analogues target fascin to block tumour metastasis. Nature (2010) 476 1062-1066 4. Kirson ED, Gurvich Z, Schneiderman R, Dekel E, Itzhaki A, Wasserman Y, Schatzberger R, Palti Y. Disruption of cancer cell replication by alternating electric fields. Cancer Res (2004) 64 3288-3295. PMID: 15126372 5. Pless M, Weinberg U. Tumor treating fields: Concept, evidence and future. Expert Opin Investig Drugs (2011) 20 1099-1106 6. Davies AM, Weinberg U, Palti Y. Tumor treating fields: A new frontier in cancer therapy. Ann N Y Acad Sci (2013) 1291 86-95. PMID: 23659608 7. Van Huizen AV, Morton JM, Kinsey LJ, Von Kannon DG, Saad MA, Birkholz TR, Czajka JM, Cyrus J, Barnes FS, Beane WS. Weak magnetic fields alter stem cell-mediated growth. Sci Adv (2019) 5. PMID: 30729158 8. Carter CS, Huang SC, Searby CC, Cassaidy B, Miller MJ, Grzesik WJ, Piorczynski TB, Pak TK, Walsh SA, Acevedo M. Exposure to static magnetic and electric fields treats type 2 diabetes. Cell Metab (2020) 32 561-574.e7. PMID: 33027675 9. Yu B, Liu J, Cheng J, Zhang L, Song C, Tian X, Fan Y, Lv Y, Zhang X. A static magnetic field improves iron metabolism and prevents high-fat-diet/streptozocin-induced diabetes. Innovation (Camb) (2021) 2. PMID: 34557734 10. Torbet J, Dickens MJ. Orientation of skeletal muscle actin in strong magnetic fields. FEBS Lett (1984) 173 403-406. PMID: 6745445 11. Mo WC, Zhang Z-J, Wang D-L, Liu Y, Bartlett PF, He R-Q. Shielding of the geomagnetic field alters actin assembly and inhibits cell motility in human neuroblastoma cells. Sci Rep (2016) 6. PMID: 27029216 12. Gartzke J, Lange K. Cellular target of weak magnetic fields: Ionic conduction along actin filaments of microvilli. Am J Physiol Cell Physiol (2002) 283 C1333-C1346. PMID: 12372794 13. Coletti D, Teodori L, Albertini MC, Rocchi M, Pristerà A, Fini M, Molinaro M, Adamo S. Static magnetic fields enhance skeletal muscle differentiation in vitro by improving myoblast alignment. Cytometry A (2007) 71 846-856. PMID: 17694560 14. Dini L, Dwikat M, Panzarini E, Vergallo C, Tenuzzo B. Morphofunctional study of 12-. Bioelectromagnetics (2009) 30 352-364. PMID: 19189300 15. Gioia L, Saponaro I, Bernabò N, Tettamanti E, Mattioli M, Barboni B. Chronic exposure to a 2 mT static magnetic field affects the morphology, the metabolism and the function of in vitro cultured swine granulosa cells. Electromagn Biol Med (2013) 32 536-550. PMID: 23631680 16. Valiron O, Peris L, Rikken G, Schweitzer A, Saoudi Y, Remy C, Job D. Cellular disorders induced by high magnetic fields. J Magn Reson Imaging (2005) 22 334-340. PMID: 16106367 17. Qian AR, Hu LF, Gao X, Zhang W, di SM, Tian ZC, Yang PF, Yin DC, Weng YY, Shang P. Large gradient high magnetic field affects the association of MACF1 with actin and microtubule cytoskeleton. Bioelectromagnetics (2009) 30 545-555. PMID: 19475564 18. Zhang J, Meng X, Ding C, Xie L, Yang P, Shang P. Regulation of osteoclast differentiation by static magnetic fields. Electromagn Biol Med (2017) 36 8-19. PMID: 27355421 19. Wang Z, Hao F, Ding C, Yang Z, Shang P. Effects of static magnetic field on cell biomechanical property and membrane ultrastructure. Bioelectromagnetics (2014) 35 251-261. PMID: 24619812 20. Wosik J, Chen W, Qin K, Ghobrial RM, Kubiak JZ, Kloc M. Magnetic field changes macrophage phenotype. Biophys J (2018) 114 2001-2013. PMID: 29694876 21. Zablotskii V, Lunov O, Novotná B, Churpita O, Trošan P, Holáň V, Syková E, Dejneka A, Kubinová Š. Down-regulation of adipogenesis of mesenchymal stem cells by oscillating high-gradient magnetic fields and mechanical vibration. Appl Phys Lett (2014) 105 22. Harris AR, Jreij P, Fletcher DA. Mechanotransduction by the actin cytoskeleton: converting mechanical stimuli into biochemical signals. Annu Rev Biophys (2018) 47 617-631 23. Roy NH, Burkhardt JK. The actin cytoskeleton: A mechanical intermediate for signal integration at the immunological synapse. Front Cell Dev Biol (2018) 6. PMID: 30283780 24. Gouget CLM, Hwang YY, Barakat AI. Model of cellular mechanotransduction via actin stress fibers. Biomech Model Mechanobiol (2016) 15 331-344. PMID: 26081725 25. Wang N. Review of cellular mechanotransduction. J Phys D Appl Phys (2017) 50. PMID: 29097823 26. Makrodouli E, Oikonomou E, Koc M, Andera L, Sasazuki T, Shirasawa S, Pintzas A. BRAF and RAS oncogenes regulate Rho GTPase pathways to mediate migration and invasion properties in human colon cancer cells: A comparative study. Mol Cancer (2011) 10 118. PMID: 21943101 27. Friedl P, Wolf K, Zegers MM. Rho-directed forces in collective migration. Nat Cell Biol (2014) 16 208-210. PMID: 24576897 28. Nobes CD, Hall A. Rho, rac, and cdc42 GTPases regulate the assembly of multimolecular focal complexes associated with actin stress fibers, lamellipodia, and filopodia. Cell (1995) 81 53-62. PMID: 7536630 29. Lee H, Eskin SG, Ono S, Zhu C, McIntire LV. Force-history dependence and cyclic mechanical reinforcement of actin filaments at the single molecular level. J Cell Sci (2019) 132. PMID: 30659118 30. Unlu A. Computational prediction of actin–actin interaction. Mol Biol Rep (2014) 41 355-364. PMID: 24242338 31. Pauling L. Diamagnetic anisotropy of the peptide group. Proc Natl Acad Sci USA (1979) 76 2293-2294. PMID: 287071 32. Vassilev PM, Dronzine RT, Vassileva MP, Georgiev GA. Parallel arrays of microtubules formed in electric and magnetic fields. Biosci Rep (1982) 2 1025-1029. PMID: 7165793 33. Bras W, Diakun GP, Díaz JF, Maret G, Kramer H, Bordas J, Medrano FJ. The susceptibility of pure tubulin to high magnetic fields: A magnetic birefringence and x-ray fiber diffraction study. Biophys J (1998) 74 1509-1521. PMID: 9512047 34. Denegre JM, Valles JM, Lin K, Jordan WB, Mowry KL. Cleavage planes in frog eggs are altered by strong magnetic fields. Proc Natl Acad Sci USA (1998) 95 14729-14732. PMID: 9843957 35. Valles JM. Model of magnetic field-induced mitotic apparatus reorientation in frog eggs. Biophys J (2002) 82 1260-1265. PMID: 11867443 36. Valles JM, Wasserman SRRM, Schweidenback C, Edwardson J, Denegre JM, Mowry KL. Processes that occur before second cleavage determine third cleavage orientation in. Exp Cell Res (2002) 274 112-118. PMID: 11855862 37. Zhang L, Hou Y, Li Z, Ji X, Wang Z, Wang H, Tian X, Yu F, Yang Z, Pi L. 27 T ultra-high static magnetic field changes orientation and morphology of mitotic spindles in human cells. eLife (2017) 6. PMID: 28244368 38. Luo Y, Ji X, Liu J, Li Z, Wang W, Chen W, Wang J, Liu Q, Zhang X. Moderate intensity static magnetic fields affect mitotic spindles and increase the antitumor efficacy of 5-FU and Taxol. Bioelectrochemistry (2016) 109 31-40. PMID: 26775206 39. Mo WC, Liu Y, Cooper HM, He RQ. Altered development of. Bioelectromagnetics (2012) 33 238-246. PMID: 21853450 40. Mershin A, Kolomenski AA, Schuessler HA, Nanopoulos DV. Tubulin dipole moment, dielectric constant and quantum behavior: Computer simulations, experimental results and suggestions. Biosystems (2004) 77 73-85. PMID: 15527947 41. Bras W, Torbet J, Diakun GP, Rikken GL, Diaz JF. The diamagnetic susceptibility of the tubulin dimer. J Biophys (2014) 2014. PMID: 24701206 42. Glade N, Tabony J. Brief exposure to high magnetic fields determines microtubule self-organisation by reaction-diffusion processes. Biophys Chem (2005) 115 29-35. PMID: 15848281 43. Liu Y, Guo Y, Valles JM, Tang JX. Microtubule bundling and nested buckling drive stripe formation in polymerizing tubulin solutions. Proc Natl Acad Sci USA (2006) 103 10654-10659. PMID: 16818889 44. Guo Y, Liu Y, Oldenbourg R, Tang JX, Valles JM. Effects of osmotic force and torque on microtubule bundling and pattern formation. Phys Rev E Stat Nonlinear Soft Matter Phys (2008) 78 45. Wang DL, Wang XS, Xiao R, Liu Y, He RQ. Tubulin assembly is disordered in a hypogeomagnetic field. Biochem Biophys Res Commun (2008) 376 363-368. PMID: 18782559 46. Eguchi Y, Ueno S. Stress fiber contributes to rat Schwann cell orientation under magnetic field. IEEE T Magn (2005) 41 4146-4148 47. Brangwynne CP, Koenderink GH, MacKintosh FC, Weitz DA. Nonequilibrium microtubule fluctuations in a model cytoskeleton. Phys Rev Lett (2008) 100. PMID: 18517833 48. Westendorf C, Negrete J, Bae AJ, Sandmann R, Bodenschatz E, Beta C. Actin cytoskeleton of chemotactic amoebae operates close to the onset of oscillations. Proc Natl Acad Sci USA (2013) 110 3853-3858. PMID: 23431176 49. Ren J, Ding L, Xu Q, Shi G, Li X, Li X, Ji J, Zhang D, Wang Y, Wang T. LF-MF inhibits iron metabolism and suppresses lung cancer through activation of P53-miR-34a-E2F1/E2F3 pathway. Sci Rep (2017) 7. PMID: 28389657 50. Nie YZ, Du L, Mou Y, Xu Z, Weng L, Du Y, Zhu Y, Hou Y, Wang T. Effect of low frequency magnetic fields on melanoma: tumor inhibition and immune modulation. BMC Cancer (2013) 13 582. PMID: 24314291 51. Nie YZ, Chen Y, Mou Y, Weng L, Xu Z, Du Y, Wang W, Hou Y, Wang T. Low frequency magnetic fields enhance antitumor immune response against mouse H22 hepatocellular carcinoma. PLOS ONE (2013) 8. PMID: 24278103
--- title: 'Smoking, alcohol consumption, and 24 gastrointestinal diseases: Mendelian randomization analysis' authors: - Shuai Yuan - Jie Chen - Xixian Ruan - Yuhao Sun - Ke Zhang - Xiaoyan Wang - Xue Li - Dipender Gill - Stephen Burgess - Edward Giovannucci - Susanna C Larsson journal: eLife year: 2023 pmcid: PMC10017103 doi: 10.7554/eLife.84051 license: CC BY 4.0 --- # Smoking, alcohol consumption, and 24 gastrointestinal diseases: Mendelian randomization analysis ## Abstract ### Background: Whether the positive associations of smoking and alcohol consumption with gastrointestinal diseases are causal is uncertain. We conducted this Mendelian randomization (MR) to comprehensively examine associations of smoking and alcohol consumption with common gastrointestinal diseases. ### Methods: Genetic variants associated with smoking initiation and alcohol consumption at the genome-wide significance level were selected as instrumental variables. Genetic associations with 24 gastrointestinal diseases were obtained from the UK Biobank, FinnGen study, and other large consortia. Univariable and multivariable MR analyses were conducted to estimate the overall and independent MR associations after mutual adjustment for genetic liability to smoking and alcohol consumption. ### Results: Genetic predisposition to smoking initiation was associated with increased risk of 20 of 24 gastrointestinal diseases, including 7 upper gastrointestinal diseases (gastroesophageal reflux, esophageal cancer, gastric ulcer, duodenal ulcer, acute gastritis, chronic gastritis, and gastric cancer), 4 lower gastrointestinal diseases (irritable bowel syndrome, diverticular disease, Crohn’s disease, and ulcerative colitis), 8 hepatobiliary and pancreatic diseases (non-alcoholic fatty liver disease, alcoholic liver disease, cirrhosis, liver cancer, cholecystitis, cholelithiasis, and acute and chronic pancreatitis), and acute appendicitis. Fifteen out of 20 associations persisted after adjusting for genetically predicted alcohol consumption. Genetically predicted higher alcohol consumption was associated with increased risk of duodenal ulcer, alcoholic liver disease, cirrhosis, and chronic pancreatitis; however, the association for duodenal ulcer did not remain statistically significant after adjustment for genetic predisposition to smoking initiation. ### Conclusions: This study provides MR evidence supporting causal associations of smoking with a broad range of gastrointestinal diseases, whereas alcohol consumption was associated with only a few gastrointestinal diseases. ### Funding: The Natural Science Fund for Distinguished Young Scholars of Zhejiang Province; National Natural Science Foundation of China; Key Project of Research and Development Plan of Hunan Province; the Swedish Heart Lung Foundation; the Swedish Research Council; the Swedish Cancer Society. ## eLife digest People who smoke cigarettes or drink large amounts of alcohol are more likely to develop disorders with their digestive system. But it is difficult to prove that heavy drinking or smoking is the primary cause of these gastrointestinal diseases. For example, it is possible that having a digestive disorder makes people more likely to take up these habits to reduce pain or discomfort caused by the illness (an effect known as reverse causation). The association may also be the result of confounding factors, such as age or diet, which contribute to digestive problems as well as the health outcomes of smoking and drinking. Additionally, many people who smoke also drink alcohol and vice versa, making it challenging to determine if one or both behaviors contribute to the disease. One solution is to employ Mendelian randomization which uses genetics to determine if two variables are linked. Using this statistical approach, Yuan, Chen, Ruan et al. investigated if people who display genetic variants that predispose someone to becoming a smoker or drinker are at greater risk of developing certain digestive disorders. This reduces the possibility of confounding and reverse causation, as any association between genetic variants will have been present since birth, and will have not been impacted by external factors. Yuan, Chen, Ruan et al. used data from two studies that had collected the genetic and health information of thousands of people living in the United Kingdom or Finland. The analyses revealed that genetic variants associated with cigarette smoking increase the risk of 20 of the 24 gastrointestinal diseases investigated. This risk persisted for most of the disorders, even after adjusting for genes linked with alcohol consumption. Further analysis showed that genetic variants linked to heavy drinking increase the risk of duodenal ulcer, alcoholic liver disease, cirrhosis, and chronic pancreatitis. However, accounting for smoking-linked genes eliminated the relationship with duodenal ulcer. These findings suggest that smoking has detrimental effects on gastrointestinal health. Reducing the number of people who start smoking or encouraging smokers to quit may help prevent digestive diseases. Even though there were fewer associations between heavy alcohol consumption and gastrointestinal illness, further studies are needed to investigate this relationship in more depth. ## Introduction Tobacco smoking and alcohol consumption are leading causes of the global burden of disease and are major contributors to premature mortality (GBD 2016 Alcohol Collaborators, 2018; GBD 2016 Alcohol Collaborators, 2020). Gastrointestinal diseases account for considerable health care use and expenditures, and a holistic approach to lifestyle interventions may result in more health gains and less economic burdens (Peery et al., 2022). Population-based studies have identified tobacco smoking as a risk factor for several gastrointestinal diseases, including gastroesophageal reflux disease (Eusebi et al., 2018), esophageal cancer (Fund WCR and Research AIfC, 2007), Crohn’s disease (Piovani et al., 2019), liver cancer (McGee et al., 2019), and pancreatitis (Yadav and Whitcomb, 2010). Evidence on the association between tobacco smoking and risk of other gastrointestinal diseases is limited and inconsistent. Like smoking, heavy alcohol consumption has been associated with increased risk of several gastrointestinal outcomes, including gastritis (Bujanda, 2000), gastric cancer (Laszkowska et al., 2021), colorectal cancer (McNabb et al., 2020), cirrhosis (Simpson et al., 2019), liver cancer (McGee et al., 2019), and pancreatitis (Yadav and Whitcomb, 2010). However, whether these associations are all causal remains unestablished, since most of the evidence was obtained from observational studies where the results may be biased by reverse causality and confounding. Of note, even though reverse causality may not be an issue in the studies for any of studied gastroenterological outcomes, it might exist for certain gastroenterological diseases causing pain, which smoker patients may try to increase smoking dose to mitigate via an intake of higher levels of nicotine. In addition, as smoking and alcohol consumption are phenotypically and genetically correlated (Roberts et al., 2020; Liu et al., 2019), the independent impacts of smoking and alcohol consumption on gastrointestinal diseases are unclear. Establishing the causal association of tobacco smoking and alcohol consumption with gastrointestinal diseases is crucial, as this provides further evidence for subsequent recommending public policies and clinical interventions. Mendelian randomization (MR) is an epidemiological approach that utilizes genetic variants as an instrument to strengthen the causal inference in an exposure-outcome association (Davey Smith and Hemani, 2014). MR is by nature not prone to confounding since genetic variants are randomly assorted at conception and thus unrelated to environmental and self-adopted factors that usually act as confounders. Additionally, this method can minimize reverse causality since fixed alleles are unaffected by the onset and progression of disease. Previous MR studies have examined the associations of smoking and alcohol consumption with several gastrointestinal diseases (Yuan and Larsson, 2022a; Larsson et al., 2020; Yuan and Larsson, 2022b; Yuan et al., 2022c; Chen et al., 2022; Yuan et al., 2021). Nevertheless, whether smoking and alcohol consumption exert influence on a wide range of gastrointestinal outcomes has not been investigated in a comprehensive way. A thorough investigation on the gastrointestinal consequences of smoking and alcohol drinking is of great importance to develop non-pharmacological interventions on gastrointestinal diseases. Here, we conducted an MR investigation of the associations of smoking and alcohol consumption with the risk of common gastrointestinal diseases to fill in above knowledge gaps. ## Materials and methods Figure 1 shows the study design overview. The study was based on publicly available genome-wide association studies (GWAS), and the detailed information on used studies was presented in Supplementary file 1A. The genetic associations were estimated using data from the UK Biobank study (Sudlow et al., 2015), the FinnGen study (Kurki et al., 2022; https://www.finngen.fi/en), and several large consortia. The summary effect estimates were combined using meta-analysis for each gastrointestinal disease from different data resources. Included studies had been approved by corresponding institutional review boards and ethical committees, and consent forms had been signed by all participants. Figure 1.: Overview of the present study design.GERA, Genetic Epidemiology Research on Aging; IIBDGC, the International Inflammatory Bowel Disease Genetics Consortium; MR, Mendelian randomization; MR-PRESSO, Mendelian randomization pleiotropy residual sum and outlier; SNP, single nucleotide polymorphism. ## Instrumental variable selection A total of 378 and 99 single nucleotide polymorphisms (SNPs) associated with smoking initiation (a binary phenotype indicating whether an individual had ever being a regular smoker, 1,232,091 individuals of European descent) and alcohol consumption (log-transformed drinks per week, 941,280 individuals of European descent) at the genome-wide significance threshold ($p \leq 5$ × 10–8) were identified by the GWAS and Sequencing Consortium of Alcohol and Nicotine use (GSCAN) study (Liu et al., 2019). These SNPs explained approximately 2.3 and $0.3\%$ of the variation in smoking initiation and alcohol consumption, respectively (Liu et al., 2019). SNPs in linkage disequilibrium (defined as r2 >0.01 or clump distance <10,000 kb) and with the weaker associations with the exposure were removed, leaving 314 independent SNPs as instrumental variables for smoking initiation and 84 for alcohol consumption. Smoking initiation and alcohol consumption shared two index genetic variants, which were rs1713676 and rs11692435. Considering partial sample overlap (around $30\%$) between the GSCAN study with full data and the UK Biobank study (Liu et al., 2019), we performed sensitivity analyses for smoking initiation and alcohol consumption using summary statistics data from the analysis excluding the UK Biobank and 23andMe. For a supplementary analysis of smoking behavior, we used 126 SNPs associated with a lifetime smoking index that considered smoking duration, heaviness, and cessation (Wootton et al., 2020). The set of genetic instruments captured around $0.36\%$ of the variance in lifetime smoking (Wootton et al., 2020). We also conducted a sensitivity analysis using rs1229984 in ADH1B gene that encodes alcohol dehydrogenase 1B enzyme as the genetic instrument for alcohol consumption to minimize pleiotropy. Detailed information on used SNPs is presented in Supplementary file 1B. ## Gastrointestinal disease data sources Genetic associations with 24 gastrointestinal diseases were obtained from the UK Biobank study (Sudlow et al., 2015), the FinnGen study (Kurki et al., 2022), and two large consortia, including the International Inflammatory Bowel Disease Genetics Consortium (IIBDGC) (Liu et al., 2015) and Genetic Epidemiology Research on Aging (GERA) (Guindo-Martínez et al., 2021). Included outcomes were classified into four major categories according to the disease onset site: [1] upper gastrointestinal diseases (gastroesophageal reflux disease, esophageal cancer, gastric ulcer, acute gastritis, chronic gastritis, and gastric cancer); [2] lower gastrointestinal diseases (irritable bowel disease, celiac disease, diverticular disease, Crohn’s disease, ulcerative colitis, and colorectal cancer); [3] hepatobiliary and pancreatic diseases (non-alcoholic fatty liver disease, alcoholic liver disease, cirrhosis, liver cancer, cholangitis, cholecystitis, cholelithiasis, acute pancreatitis, chronic pancreatitis, and pancreatic cancer); and [4] other (acute appendicitis). The UK Biobank study is a large multicenter cohort study of 500,000 participants recruited in the United Kingdom between 2006 and 2010 (Sudlow et al., 2015). We used the summary statistics of European ancestry from GWAS conducted by Lee lab, where the gastrointestinal outcomes were defined by codes of the International Classification of Diseases 9th Revision (ICD-9) and ICD-10 (Zhou et al., 2020). Genetic associations were estimated by logistic regression with adjustment for sex, birth year, and the first four genetic principal components. For the FinnGen study, we used summary-level data on the genetic associations with gastrointestinal diseases from the last publicly available R7 data release (Kurki et al., 2022). The FinnGen study is a nationwide genetic study where genetic and electronic health record data were collected. The gastrointestinal diseases were ascertained by the codes of the ICD-8, ICD-9, and ICD-10. Genome-wide association analyses were adjusted for sex, age, genetic components, and genotyping batch. Summary-level genetic data on Crohn’s disease (5956 cases and 14,927 controls) and ulcerative colitis (6968 cases and 20,464 controls) were additionally obtained from the IIBDGC (Liu et al., 2015), and data on irritable bowel syndrome (3117 cases and 53,520 controls) were obtained from the GERA (Guindo-Martínez et al., 2021). Detailed diagnostic codes are listed in Supplementary file 1C. ## Statistical analysis Data were harmonized to omit ambiguous SNPs with non-concordant alleles and palindromic SNPs with ambiguous minor allele frequency (>0.42 and <0.58) were removed from the analysis. The primary MR analyses were performed by the multiplicative random-effects inverse-variance weighted (IVW) method, which provides the most precise estimates though assuming that all SNPs are valid instruments. The analysis of rs1229984 for alcohol consumption was conducted by the Wald method. Estimates for each association from different sources were combined using fixed-effects meta-analysis, and the heterogeneity of the associations from different data sources was evaluated by the I2 statistic. Heterogeneity among SNPs’ estimates in each association was assessed by Cochran’s Q value. Multivariable MR analyses were conducted to mutually adjust for smoking initiation and alcohol consumption. To detect potential unbalanced pleiotropy (horizontal pleiotropy) and examine the consistency of the associations, three sensitivity analyses including the weighted median (Yavorska and Burgess, 2017), MR-Egger (Burgess and Thompson, 2017), and MR pleiotropy residual sum and outlier (MR-PRESSO) (Verbanck et al., 2018) analyses were performed. The weighted median method can provide consistent estimates when more than $50\%$ of the weight comes from valid instrument variants (Yavorska and Burgess, 2017). The MR-Egger intercept test can detect unmeasured pleiotropy, and MR-Egger regression can generate estimates after accounting for horizontal pleiotropy albeit with less precision (Burgess and Thompson, 2017). The MR-PRESSO method can identify SNP outliers and provide results identical to that from IVW after removal of outliers (Verbanck et al., 2018). The F-statistic was estimated to quantify instrument strength, and an F-statistic >10 suggested a sufficiently strong instrument. Power analysis was performed using an online tool (Brion et al., 2013). The Benjamini-Hochberg correction that controls the false discovery rate was applied to correct for multiple testing. The association with a nominal p-value <0.05 but Benjamini-Hochberg adjusted p-value >0.05 was regarded suggestive, and the association with a Benjamini-Hochberg adjusted p-value <0.05 was deemed significant. All analyses were two-sided and performed using the TwoSampleMR (Hemani et al., 2018), MendelianRandomization (Yavorska and Burgess, 2017), and MRPRESSO R packages (Verbanck et al., 2018) in R software 4.1.2. ## Results The F-statistic for each genetic variant was above 10, suggesting a good strength of used genetic instruments (Supplementary file 1B). Most associations were well powered (Supplementary file 1D). For smoking initiation, there was $80\%$ power to detect the smallest odds ratio (OR) ranging from 1.08 to 1.40 for included outcomes. Although power was lower for alcohol consumption, it was adequate to detect a moderate effect size for most common gastrointestinal diseases. ## Smoking and gastrointestinal diseases Genetic predisposition to smoking initiation was associated with 20 of the 24 studied gastrointestinal diseases, and all these associations remained after multiple comparison correction (Table 1 and Supplementary file 1E). In detail, genetic liability to smoking initiation was positively associated with seven upper gastrointestinal diseases: gastroesophageal reflux (OR, 1.28; $95\%$ confidence interval [CI], 1.20–1.37; $$p \leq 4.09$$ × 10−14), esophageal cancer (OR, 1.67; $95\%$ CI, 1.24–2.25; $$p \leq 6.84$$ × 10−4), gastric ulcer (OR, 1.54; $95\%$ CI, 1.37–1.72; $$p \leq 3.83$$ × 10−14), duodenal ulcer (OR, 1.53; $95\%$ CI, 1.34–1.75; $$p \leq 8.47$$ × 10−10), acute gastritis (OR, 1.29; $95\%$ CI, 1.09–1.53; $$p \leq 0.003$$), chronic gastritis (OR, 1.33; $95\%$ CI, 1.18–1.49; $$p \leq 1.55$$ × 10–6), and gastric cancer (OR, 1.42; $95\%$ CI, 1.13–1.79; $$p \leq 0.003$$); genetic liability to smoking initiation was positively associated with four lower gastrointestinal diseases: irritable bowel syndrome (OR, 1.22; $95\%$ CI, 1.12–1.32; $$p \leq 3.50$$ × 10−6), diverticular disease (OR, 1.25; $95\%$ CI, 1.18–1.33; $$p \leq 5.23$$ × 10−14), Crohn’s disease (OR, 1.25; $95\%$ CI, 1.11–1.40; $$p \leq 3.03$$ × 10−4), and ulcerative colitis (OR, 1.15; $95\%$ CI, 1.04–1.26; $$p \leq 0.004$$); genetic liability to smoking initiation was positively associated with eight hepatobiliary and pancreatic diseases: non-alcoholic fatty liver disease (OR, 1.49; $95\%$ CI, 1.26–1.76; $$p \leq 3.82$$ × 10−6), alcoholic liver disease (OR, 1.99; $95\%$ CI, 1.65–2.41; $$p \leq 1.49$$ × 10−12), cirrhosis (OR, 1.68; $95\%$ CI, 1.40–2.02; $$p \leq 3.39$$ × 10−8), liver cancer (OR, 1.57; $95\%$ CI, 1.13–2.17; $$p \leq 0.007$$), cholecystitis (OR, 1.47; $95\%$ CI, 1.29–1.68; $$p \leq 4.71$$ × 10−9), cholelithiasis (OR, 1.20; $95\%$ CI, 1.13–1.27; $$p \leq 5.75$$ × 10−9), acute pancreatitis (OR, 1.39; $95\%$ CI, 1.23–1.56; $$p \leq 6.71$$ × 10−8), and chronic pancreatitis (OR, 1.38; $95\%$ CI, 1.17–1.64; $$p \leq 1.79$$ × 10−4); genetic liability to smoking initiation was positively associated with acute appendicitis (OR, 1.15; $95\%$ CI, 1.08–1.23; $$p \leq 1.27$$ × 10−5). Results were consistent in sensitivity analyses. An indication of horizontal pleiotropy was observed in the analysis of esophageal cancer in the FinnGen study (p for MR-Egger intercept <0.05, Supplementary file 1F). Although MR-PRESSO detected one to three outliers, the associations persisted and remained significant after removal of these out-lying SNPs (Supplementary file 1F). When using the genetic variants for smoking initiation based on data without the UK Biobank and 23andMe studies, the associations attenuated slightly albeit remained significant after multiple comparisons (Supplementary file 1L and Supplementary file 1G). All associations were replicated in the supplementary analysis of the lifetime smoking index (Supplementary file 1G). After correcting for multiple testing, genetically predicted lifetime smoking index was significantly associated with 17 of 24 gastrointestinal diseases, where the patterns of associations were generally similar to the analysis for smoking initiation (Supplementary file 1M and Supplementary file 1G). In distinction to the analysis of smoking initiation, genetically predicted lifetime smoking index was not significantly associated with acute gastritis, gastric cancer, Crohn’s disease, and ulcerative colitis, whereas genetically predicted lifetime smoking index was significantly associated with pancreatic cancer (OR, 2.09; $95\%$ CI, 1.30–3.36). Table 1. \| Disease \| Disease.1 \| Total cases \| Total controls \| UVMR \| UVMR.1 \| UVMR.2 \| MVMR adjusted for alcohol consumption \| MVMR adjusted for alcohol consumption.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Disease \| Disease \| Total cases \| Total controls \| OR (95% CI) \| p Value \| I2 (95% CI) \| OR (95% CI) \| p Value \| \| Upper gastrointestinal diseases \| Gastroesophageal reflux \| 34135 \| 634629 \| 1.28 (1.20, 1.37) \| 4.09 × 10-14* \| 46.24 \| 1.65 (1.35, 2.02) \| 1.38 × 10-6* \| \| Upper gastrointestinal diseases \| Esophageal cancer \| 1130 \| 702116 \| 1.67 (1.24, 2.25) \| 6.84 × 10-4* \| 22.68 \| 4.78 (2.10, 10.90) \| 1.97 × 10-4* \| \| Upper gastrointestinal diseases \| Gastric ulcer \| 8651 \| 666879 \| 1.54 (1.37, 1.72) \| 3.83 × 10-14* \| 44.96 \| 1.95 (1.40, 2.71) \| 7.31 × 10-5* \| \| Upper gastrointestinal diseases \| Duodenal ulcer \| 5713 \| 666879 \| 1.53 (1.34, 1.75) \| 8.47 × 10-10* \| 0.00 \| 1.64 (1.07, 2.52) \| 0.024 \| \| Upper gastrointestinal diseases \| Acute gastritis \| 3048 \| 643478 \| 1.29 (1.09, 1.53) \| 0.003* \| 0.00 \| 1.54 (0.91, 2.62) \| 0.106 \| \| Upper gastrointestinal diseases \| Chronic gastritis \| 7975 \| 643478 \| 1.33 (1.18, 1.49) \| 1.55 × 10-6* \| 77.04 \| 1.33 (0.96, 1.86) \| 0.091 \| \| Upper gastrointestinal diseases \| Gastric cancer \| 1608 \| 701472 \| 1.42 (1.13, 1.79) \| 0.003* \| 0.00 \| 2.29 (1.14, 4.59) \| 0.020 \| \| Lower gastrointestinal diseases \| Irritable bowel disease \| 15718 \| 641489 \| 1.22 (1.12, 1.32) \| 3.50 × 10-6* \| 11.84 \| 1.43 (1.10, 1.85) \| 0.008* \| \| Lower gastrointestinal diseases \| Celiac disease \| 4808 \| 631700 \| 0.82 (0.66, 1.02) \| 0.071 \| 0.00 \| 0.87 (0.53, 1.43) \| 0.590 \| \| Lower gastrointestinal diseases \| Diverticular disease \| 50065 \| 587969 \| 1.25 (1.18, 1.33) \| 5.23 × 10-14* \| 67.29 \| 1.56 (1.30, 1.87) \| 1.41 × 10-6* \| \| Lower gastrointestinal diseases \| Crohn’s disease \| 10846 \| 645718 \| 1.25 (1.11, 1.40) \| 3.03 × 10-4* \| 0.00 \| 1.48 (1.01, 2.16) \| 0.042 \| \| Lower gastrointestinal diseases \| Ulcerative colitis \| 16770 \| 651255 \| 1.15 (1.04, 1.26) \| 0.004* \| 0.00 \| 0.94 (0.71, 1.25) \| 0.677 \| \| Lower gastrointestinal diseases \| Colorectal cancer \| 9519 \| 686953 \| 1.03 (0.92, 1.14) \| 0.632 \| 29.94 \| 1.03 (0.76, 1.39) \| 0.841 \| \| Hepatobiliary and pancreatic diseases \| Non-alcoholic fatty liver disease \| 3242 \| 707631 \| 1.49 (1.26, 1.76) \| 3.82 × 10-6* \| 0.00 \| 2.11 (1.15, 3.88) \| 0.016* \| \| Hepatobiliary and pancreatic diseases \| Alcoholic liver disease \| 2955 \| 680369 \| 1.99 (1.65, 2.41) \| 1.49 × 10-12* \| 92.68 \| 2.26 (1.26, 4.03) \| 0.006 \| \| Hepatobiliary and pancreatic diseases \| Cirrhosis \| 5904 \| 706200 \| 1.68 (1.40, 2.02) \| 3.39 × 10-8* \| 0.00 \| 1.92 (1.06, 3.47) \| 0.032 \| \| Hepatobiliary and pancreatic diseases \| Liver cancer \| 714 \| 702008 \| 1.57 (1.13, 2.17) \| 0.007* \| 0.00 \| 1.96 (0.73, 5.25) \| 0.183 \| \| Hepatobiliary and pancreatic diseases \| Cholangitis \| 1708 \| 664749 \| 1.02 (0.80, 1.29) \| 0.892 \| 0.00 \| 1.31 (0.61, 2.84) \| 0.489 \| \| Hepatobiliary and pancreatic diseases \| Cholecystitis \| 5893 \| 664749 \| 1.47 (1.29, 1.68) \| 4.71 × 10-9* \| 84.72 \| 2.38 (1.57, 3.60) \| 4.14 × 10-5* \| \| Hepatobiliary and pancreatic diseases \| Cholelithiasis \| 42510 \| 664749 \| 1.20 (1.13, 1.27) \| 5.75 × 10-9* \| 0.00 \| 1.33 (1.02, 1.73) \| 0.035 \| \| Hepatobiliary and pancreatic diseases \| Acute pancreatitis \| 6634 \| 679713 \| 1.39 (1.23, 1.56) \| 6.71 × 10–8* \| 79.71 \| 1.55 (1.04, 2.31) \| 0.031 \| \| Hepatobiliary and pancreatic diseases \| Chronic pancreatitis \| 3173 \| 679713 \| 1.38 (1.17, 1.64) \| 1.79 × 10–4* \| 0.00 \| 1.27 (0.74, 2.16) \| 0.384 \| \| Hepatobiliary and pancreatic diseases \| Pancreatic cancer \| 1643 \| 701472 \| 1.00 (0.79, 1.26) \| 0.999 \| 67.21 \| 2.08 (1.06, 4.10) \| 0.034 \| \| Other \| Acute appendicitis \| 25361 \| 690149 \| 1.15 (1.08, 1.23) \| 1.27 × 10–5* \| 0.00 \| 1.15 (0.92, 1.44) \| 0.221 \| In multivariable MR analysis adjusted for genetically predicted alcohol consumption, the associations between genetically predicted smoking initiation and gastrointestinal diseases were consistent with that from univariable MR analysis (Table 1 and Supplementary file 1H). However, the associations became stronger with wider CIs, in particular the associations for gastrointestinal reflux, esophageal cancer, gastric ulcer, irritable bowel syndrome, diverticular disease, non-alcoholic fatty liver disease, alcoholic liver disease, and cholecystitis (Table 1). In addition, the association for pancreatic cancer became suggestive significant from null. ## Alcohol consumption and gastrointestinal diseases Genetically predicted alcohol consumption was nominally positively associated with esophageal cancer (OR, 2.86; $95\%$ CI, 1.18–6.91; $$p \leq 0.020$$), duodenal ulcer (OR, 1.92; $95\%$ CI, 1.23–3.00; $$p \leq 0.004$$), alcoholic liver disease (OR, 14.35; $95\%$ CI, 7.69–26.81; $$p \leq 6.32$$ × 10−17), cirrhosis (OR, 2.96; $95\%$ CI, 1.50–5.85; $$p \leq 0.002$$), and chronic pancreatitis (OR, 2.96; $95\%$ CI, 1.80–4.89; $$p \leq 2.13$$ × 10−5), and nominally inversely associated with irritable bowel disease (OR, 0.73; $95\%$ CI 0.57–0.93; $$p \leq 0.012$$) (Table 2). After Benjamini-Hochberg correction, the associations for duodenal ulcer, alcoholic liver disease, cirrhosis, and chronic pancreatitis remained (Supplementary file 1E). Results were consistent in sensitivity analyses, and no horizontal pleiotropy was detected (Supplementary file 1I). One outlier was detected in the analysis of duodenal ulcer in the FinnGen study, and the association slightly changed after removal of this outlier (Supplementary file 1I). Results were consistent in the sensitivity analysis, where the genetic associations with alcohol consumption were obtained from the genome-wide association analysis excluding the UK Biobank and 23andMe studies (Supplementary file 1N and Supplementary file 1G). The associations were directionally consistent albeit with wider CIs in the analysis, where alcohol consumption was instrumented by rs1229984 (Supplementary file 1J). The associations for alcoholic liver disease, cirrhosis, and chronic pancreatitis persisted after adjustment for genetic liability to smoking initiation and multiple testing correction (Table 2 and Supplementary file 1H). Table 2. \| Disease \| Disease.1 \| Total cases \| Total controls \| UVMR \| UVMR.1 \| UVMR.2 \| MVMR adjusted for smoking initiation \| MVMR adjusted for smoking initiation.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Disease \| Disease \| Total cases \| Total controls \| OR (95% CI) \| p Value \| I2 (95% CI) \| OR (95% CI) \| p Value \| \| Upper gastrointestinal diseases \| Gastroesophageal reflux \| 34135 \| 634629 \| 0.99 (0.81, 1.21) \| 0.893 \| 46.24 \| 0.88 (0.72, 1.08) \| 0.219 \| \| Upper gastrointestinal diseases \| Esophageal cancer \| 1130 \| 702116 \| 2.86 (1.18, 6.91) \| 0.020 \| 22.68 \| 1.28 (0.59, 2.82) \| 0.533 \| \| Upper gastrointestinal diseases \| Gastric ulcer \| 8651 \| 666879 \| 1.30 (0.95, 1.77) \| 0.098 \| 44.96 \| 1.06 (0.77, 1.47) \| 0.721 \| \| Upper gastrointestinal diseases \| Duodenal ulcer \| 5713 \| 666879 \| 1.92 (1.23, 3.00) \| 0.004* \| 0.00 \| 1.54 (1.01, 2.34) \| 0.045 \| \| Upper gastrointestinal diseases \| Acute gastritis \| 3048 \| 643478 \| 0.99 (0.58, 1.69) \| 0.960 \| 0.00 \| 0.88 (0.52, 1.48) \| 0.621 \| \| Upper gastrointestinal diseases \| Chronic gastritis \| 7975 \| 643478 \| 1.33 (0.90, 1.95) \| 0.147 \| 77.04 \| 1.33 (0.93, 1.89) \| 0.115 \| \| Upper gastrointestinal diseases \| Gastric cancer \| 1608 \| 701472 \| 1.57 (0.75, 3.30) \| 0.233 \| 0.00 \| 1.59 (0.79, 3.21) \| 0.194 \| \| Lower gastrointestinal diseases \| Irritable bowel disease \| 15718 \| 641489 \| 0.73 (0.57, 0.93) \| 0.012 \| 11.84 \| 0.74 (0.57, 0.97) \| 0.027 \| \| Lower gastrointestinal diseases \| Celiac disease \| 4808 \| 631700 \| 0.69 (0.44, 1.07) \| 0.097 \| 0.00 \| 1.04 (0.64, 1.68) \| 0.887 \| \| Lower gastrointestinal diseases \| Diverticular disease \| 50065 \| 587969 \| 0.95 (0.79, 1.13) \| 0.553 \| 67.29 \| 0.94 (0.79, 1.13) \| 0.527 \| \| Lower gastrointestinal diseases \| Crohn’s disease \| 10846 \| 645718 \| 0.91 (0.62, 1.32) \| 0.613 \| 0.00 \| 0.74 (0.53, 1.05) \| 0.088 \| \| Lower gastrointestinal diseases \| Ulcerative colitis \| 16770 \| 651255 \| 1.11 (0.82, 1.50) \| 0.509 \| 0.00 \| 0.88 (0.67, 1.15) \| 0.358 \| \| Lower gastrointestinal diseases \| Colorectal cancer \| 9519 \| 686953 \| 1.09 (0.76, 1.55) \| 0.649 \| 29.94 \| 1.28 (0.95, 1.72) \| 0.098 \| \| Hepatobiliary and pancreatic diseases \| Non-alcoholic fatty liver disease \| 3242 \| 707631 \| 1.20 (0.63, 2.28) \| 0.574 \| 0.00 \| 0.99 (0.54, 1.79) \| 0.962 \| \| Hepatobiliary and pancreatic diseases \| Alcoholic liver disease \| 2955 \| 680369 \| 14.35 (7.69, 26.81) \| 6.32 × 10-17* \| 92.68 \| 9.60 (5.28, 17.46) \| 1.25 × 10-13* \| \| Hepatobiliary and pancreatic diseases \| Cirrhosis \| 5904 \| 706200 \| 2.96 (1.50, 5.85) \| 0.002* \| 0.00 \| 2.41 (1.29, 4.52) \| 0.006* \| \| Hepatobiliary and pancreatic diseases \| Liver cancer \| 714 \| 702008 \| 1.16 (0.43, 3.11) \| 0.775 \| 0.00 \| 0.76 (0.29, 2.02) \| 0.585 \| \| Hepatobiliary and pancreatic diseases \| Cholangitis \| 1708 \| 664749 \| 0.96 (0.44, 2.08) \| 0.912 \| 0.00 \| 0.72 (0.33, 1.55) \| 0.397 \| \| Hepatobiliary and pancreatic diseases \| Cholecystitis \| 5893 \| 664749 \| 1.36 (0.91, 2.03) \| 0.132 \| 84.72 \| 0.96 (0.64, 1.45) \| 0.862 \| \| Hepatobiliary and pancreatic diseases \| Cholelithiasis \| 42510 \| 664749 \| 1.02 (0.75, 1.39) \| 0.878 \| 0.00 \| 1.03 (0.79, 1.35) \| 0.801 \| \| Hepatobiliary and pancreatic diseases \| Acute pancreatitis \| 6634 \| 679713 \| 1.36 (0.91, 2.03) \| 0.128 \| 79.71 \| 1.17 (0.78, 1.75) \| 0.456 \| \| Hepatobiliary and pancreatic diseases \| Chronic pancreatitis \| 3173 \| 679713 \| 2.96 (1.80, 4.89) \| 2.13 × 10-5* \| 0.00 \| 3.24 (1.86, 5.64) \| 3.18 × 10-5 \| \| Hepatobiliary and pancreatic diseases \| Pancreatic cancer \| 1643 \| 701472 \| 0.63 (0.32, 1.26) \| 0.193 \| 67.21 \| 0.79 (0.40, 1.56) \| 0.496 \| \| Other \| Acute appendicitis \| 25361 \| 690149 \| 0.80 (0.63, 1.01) \| 0.063 \| 0.00 \| 0.77 (0.61, 0.97) \| 0.024 \| ## Discussion We conducted a comprehensive MR investigation to examine the causal role of smoking and alcohol consumption in 24 gastrointestinal diseases, and the result summary of this comprehensive analysis is shown in Figure 2 and Supplementary file 1K. We found robust associations between genetic predisposition to smoking and increased risk of 15 gastrointestinal outcomes independent of alcohol consumption, showing an extensive impact on gastrointestinal health. In contrast, genetically predicted alcohol consumption was robustly and predominantly associated with increased risk of liver and pancreatic diseases, including alcoholic liver disease, cirrhosis, and chronic pancreatitis after adjustment for smoking. Figure 2.:** Summary of associations of genetically predicted smoking initiation, lifetime smoking, and alcohol consumption with 24 gastrointestinal diseases.UVMR, univariable Mendelian randomization; MVMR, multivariable Mendelian randomization. The numbers in the box are the odds ratios for associations of exposure for gastrointestinal diseases. The association with a p-value <0.05 but Benjamini-Hochberg adjusted p-value >0.05 was regarded suggestive, and the association with a Benjamini-Hochberg adjusted p-value <0.05 was deemed significant. Corroborating and extending the previous observational studies, our MR investigation strengthened the evidence that smoking has a detrimental effect on gastrointestinal health and increases the risk of a broad range of gastrointestinal diseases, including gastroesophageal reflux disease (Eusebi et al., 2018), esophageal cancer (Castro et al., 2018), gastric and duodenal ulcer (Kato et al., 1992), gastritis (Nordenstedt et al., 2013), gastric cancer (Zhang et al., 2020), irritable bowel syndrome (Talley et al., 2021), diverticular disease (Aune et al., 2017), Crohn’s disease (Piovani et al., 2019), cirrhosis (Liu et al., 2009), liver cancer (McGee et al., 2019), cholelithiasis (Aune et al., 2016), acute and chronic pancreatitis (Aune et al., 2019), and acute appendicitis (Montgomery et al., 1999). In line with previous MR studies, the current MR study also found that smoking was associated with increased risk of gastroesophageal reflux disease (Yuan and Larsson, 2022a), esophageal cancer (Larsson et al., 2020), gastric cancer (Larsson et al., 2020), diverticular disease (Yuan and Larsson, 2022b) non-alcoholic fatty liver disease (Yuan et al., 2022c), cholelithiasis (Chen et al., 2022), and acute and chronic pancreatitis (Yuan et al., 2021). As for ulcerative colitis, traditional observational studies revealed a decreased risk among current smokers (Piovani et al., 2019; Park et al., 2019); however, a recent MR analysis including 12,366 ulcerative colitis cases did not verify this inverse association in the analysis where smoking initiation was instrumented by 363 SNPs (Georgiou et al., 2021). Based on data from three independent populations, our study provided genetic evidence that smoking was a causal risk factor for ulcerative colitis in the analysis including 16,770 cases. Observational studies found that smoking was associated with an increased risk of colorectal cancer in a dose-dependent manner (Botteri et al., 2020), whereas the positive association was not observed in an MR analysis (Larsson et al., 2020). The current study was in line with the above MR study and found no strong association between smoking initiation and colorectal cancer risk. Nevertheless, a previous MR analysis with a 52,775 colorectal cancer cases found that genetic prediction to lifetime smoking index was positively associated with risks of colorectal cancer (Dimou et al., 2021), which might imply that our null finding might be caused by insufficient power due to a relatively small sample size. Smoking has been identified as a well-established risk factor for pancreatic cancer (Mizrahi et al., 2020). Interestingly, despite a null finding on the association of genetic liability to smoking initiation and pancreatic cancer in univariable MR analysis, the association became stronger and suggestively significant after adjusting for genetically predicted alcohol consumption. This might be explained by an inverse association between moderate alcohol consumption and pancreatic cancer. In addition, an adverse effect of smoking on pancreatic cancer was observed when using a smoking index as genetic instrument for lifetime smoking exposure. Our findings also provide novel evidence on the associations of smoking with the higher risk of cholecystitis and alcoholic liver disease independently of alcohol consumption, which need to be verified. The pathogenic role of alcohol in alcoholic liver disease is well established and was confirmed also in our MR analysis. Our MR evidence along with previous observational studies also supported alcohol consumption as a risk factor for esophageal cancer (Yu et al., 2018), cirrhosis (Roerecke et al., 2019), and chronic pancreatitis (Samokhvalov et al., 2015). Noteworthy, the association between alcohol consumption and esophageal cancer became positively nonsignificant in multivariable MR, which possibly explained by the synergistic effect of alcohol and smoking. However, the association between alcohol consumption and duodenal ulcer has been scarcely studied. A meta-analysis including a small number of studies with relatively small sample sizes indicated that alcohol consumption was not associated with duodenal ulcer (Ryan-Harshman and Aldoori, 2004). This null finding is likely due to insufficient power. Alcohol drinking has been associated with increased risk of gastric, colorectal, and liver cancer as well as acute pancreatitis (Bagnardi et al., 2015). These associations were not supported by our MR study. A possible explanation for this inconsistent findings is that heavy alcohol drinking is commonly associated with an unhealthy lifestyle and meager nutrition (Klatsky, 2001), which might exert confounding effects that could not be ruled out in previous observational studies. Another possible reason is that the U-shaped association could not be detected in MR analysis. For example, light drinking may be associated with decreased risk of these diseases (McNabb et al., 2020). In addition, it is also possible that the null associations observed in present study might be a consequence of inadequate power given SNPs used to mimic alcohol consumption explained a small phenotypic variance. In agreement with previous studies, our MR investigation demonstrated no associations of alcohol consumption with the development of gastroesophageal reflux, Crohn’s disease, or ulcerative colitis (Eusebi et al., 2018; Piovani et al., 2019; Georgiou et al., 2021). Many mechanisms have been proposed to support the observed positive associations between smoking and gastrointestinal diseases. Tobacco smoking has been shown to augment the production of numerous pro-inflammatory cytokines and decrease the levels of anti-inflammatory cytokines (Arnson et al., 2010), which might mediate a variety of inflammation-associated gastrointestinal diseases. In addition, smoking may also generate impacts on the immune system, including inhibition of the function of circulatory dendritic cells (Givi et al., 2015) and alteration signaling of Toll-like receptors (Noakes et al., 2006), which might contribute to the autoimmune disease and occurrence of neoplasm. The underlying mechanisms behind the associations of alcohol consumption with gastrointestinal diseases have not been fully understood. In addition to direct mucosal damage, the metabolites of ethanol are accountable for a part of the inflammation of alcohol drinking on the liver (Mandrekar and Szabo, 2009) and the gastrointestinal tract (Bishehsari et al., 2017). This study investigated the impacts of smoking and alcohol consumption on a wide range of gastrointestinal disease. Based on our findings, promoting public awareness of the adverse impacts of tobacco smoking and alcohol consumption on gastrointestinal diseases is of particular importance and should be used as prevention strategies to lower gastrointestinal disease burden because these two factors are modifiable behavioral factors as possible targets of the pharmacal (Leone et al., 2020) and behavioral interventions. In addition, our results may help facilitate the guidelines of gastrointestinal disease prevention and the management of certain patients who have a subsequent high risk of gastrointestinal disease, like those with obesity and diabetes (Camilleri et al., 2017; Krishnan et al., 2013). The major strength of the present study is MR design, which minimized bias from confounding and reverse causality and thus improved the causal inference in the associations of smoking and alcohol consumption with gastrointestinal diseases. We also used several independent outcome sources and combined the estimates, which increased statistical power as well as strengthened our findings by the observed consistency of results. Another strength is that we confined our analysis within the individuals of European ancestry, which minimized the population stratification bias. This study also has several limitations. A major limitation of MR design is horizontal pleiotropy, which means that the used SNPs exert effects on the outcomes not via the exposure but via alternative pathways. However, in this study, the bias caused by pleiotropic effects should be minimal since we observed no indications of horizontal pleiotropy in MR-Egger analysis, consistent results from a series of sensitivity analyses, and robust associations from multivariable MR analysis with mutual adjustment. Another limitation is the relatively small phenotypic variance of alcohol consumption (approximately $0.2\%$), which resulted in inadequate power to detect weak associations for certain uncommon gastrointestinal diseases. There are several limitations of using summary-level data. First, we could not evaluate the nonlinear associations between alcohol consumption and gastrointestinal diseases without individual-level data. We could not differentiate the associations of smoking and alcohol consumption on the pathological subtypes of certain gastroenterological diseases, like esophageal cancer, based on summary-level data. For example, heavy alcohol consumption was associated with a high risk of squamous esophageal cancer (Abnet et al., 2018), but the associations were inconsistent for adenocarcinoma esophageal cancer (Coleman et al., 2018), which needs further investigation. Stratification analysis on sex was unlikely to be performed. In addition, we could not interpret and rescale the associations in a comparable scale to traditional observational studies because the unit of the exposure phenotypes was fixed in the corresponding genome-wide association analyses. An additional limitation is that our analysis was confined to the European populations, and thus whether the observed associations can be generalized to other populations remains unknown. For alcohol consumption, it has been reported that there were substantial behavioral and genetic differences across ethnic groups. For example, East Asian individuals drink much less alcohol compared to other races, which appears to be related to ALDH2 gene (Jorgenson et al., 2017). A further potential limitation is that the UK Biobank study was included in both the exposure and outcome datasets, which might cause MR estimates toward the observational associations. However, the used instrumental variants were proven to be strongly associated with the exposure (F-statistic >10) (Burgess et al., 2016), and the associations were replicated in the FinnGen study. Moreover, the associations remained stable in the sensitivity analyses using the genetic associations with exposures from the data excluding the UK Biobank and 23andMe studies. All of these indicated that the bias due to sample overlap was limited. In conclusion, this MR study suggested that smoking is a risk factor for a broad range of gastrointestinal diseases independent of alcohol consumption. Alcohol consumption on the other hand seemed to be an independent risk factor for only a few gastrointestinal diseases, including alcoholic liver disease, cirrhosis, and chronic pancreatitis, but we cannot rule out weak associations with other diseases. These findings provide genetic evidence on supporting reducing tobacco smoking and possibly excessive alcohol consumption in particular to prevent gastrointestinal diseases. ## Funding Information This paper was supported by the following grants: ## Data availability Data analyzed in the current study are publicly available GWAS summary-level data. The specific information and link could be found in Table S1, Supplementary file 1. The code and curated data for the current analysis are available at https://github.com/XixianRuan/smoking_gi (copy archived at swh:1:rev:1f31c18364102366be9ed05770e4a0f23de078f6). The following previously published datasets were used: KurkiMI 2022FinnGen: Unique genetic insights from combining isolated population and national health register dataThe FinnGen studyfinngen MengzhenL 2019Data Related to Association studies of up to 1.2 million individuals yield new insights into the genetic etiology of tobacco and alcohol useData Repository for University of Minnesota (DRUM)$\frac{10.13020}{3}$b1n-ff32 ## References 1. Abnet CC, Arnold M, Wei WQ. Epidemiology of esophageal squamous cell carcinoma. Gastroenterology (2018) 154 360-373. DOI: 10.1053/j.gastro.2017.08.023 2. Arnson Y, Shoenfeld Y, Amital H. Effects of tobacco smoke on immunity, inflammation and autoimmunity. Journal of Autoimmunity (2010) 34 J258-J265. DOI: 10.1016/j.jaut.2009.12.003 3. Aune D, Vatten LJ, Boffetta P. Tobacco smoking and the risk of gallbladder disease. European Journal of Epidemiology (2016) 31 643-653. DOI: 10.1007/s10654-016-0124-z 4. Aune D, Sen A, Leitzmann MF, Tonstad S, Norat T, Vatten LJ. Tobacco smoking and the risk of diverticular disease-a systematic review and meta-analysis of prospective studies. Colorectal Disease (2017) 19 621-633. DOI: 10.1111/codi.13748 5. Aune D, Mahamat-Saleh Y, Norat T, Riboli E. Tobacco smoking and the risk of pancreatitis: a systematic review and meta-analysis of prospective studies. Pancreatology (2019) 19 1009-1022. DOI: 10.1016/j.pan.2019.09.004 6. Bagnardi V, Rota M, Botteri E, Tramacere I, Islami F, Fedirko V, Scotti L, Jenab M, Turati F, Pasquali E, Pelucchi C, Galeone C, Bellocco R, Negri E, Corrao G, Boffetta P, La Vecchia C. Alcohol consumption and site-specific cancer risk: a comprehensive dose-response meta-analysis. British Journal of Cancer (2015) 112 580-593. DOI: 10.1038/bjc.2014.579 7. Bishehsari F, Magno E, Swanson G, Desai V, Voigt RM, Forsyth CB, Keshavarzian A. Alcohol and gut-derived inflammation. Alcohol Research (2017) 38 163-171. PMID: 28988571 8. Botteri E, Borroni E, Sloan EK, Bagnardi V, Bosetti C, Peveri G, Santucci C, Specchia C, van den Brandt P, Gallus S, Lugo A. Smoking and colorectal cancer risk, overall and by molecular subtypes: a meta-analysis. The American Journal of Gastroenterology (2020) 115 1940-1949. DOI: 10.14309/ajg.0000000000000803 9. Brion M-JA, Shakhbazov K, Visscher PM. Calculating statistical power in Mendelian randomization studies. International Journal of Epidemiology (2013) 42 1497-1501. DOI: 10.1093/ije/dyt179 10. Bujanda L. The effects of alcohol consumption upon the gastrointestinal tract. The American Journal of Gastroenterology (2000) 95 3374-3382. DOI: 10.1111/j.1572-0241.2000.03347.x 11. Burgess S, Davies NM, Thompson SG. Bias due to participant overlap in two-sample Mendelian randomization. Genetic Epidemiology (2016) 40 597-608. DOI: 10.1002/gepi.21998 12. Burgess S, Thompson SG. Interpreting findings from Mendelian randomization using the MR-egger method. European Journal of Epidemiology (2017) 32 377-389. DOI: 10.1007/s10654-017-0255-x 13. Camilleri M, Malhi H, Acosta A. Gastrointestinal complications of obesity. Gastroenterology (2017) 152 1656-1670. DOI: 10.1053/j.gastro.2016.12.052 14. Castro C, Peleteiro B, Lunet N. Modifiable factors and esophageal cancer: a systematic review of published meta-analyses. Journal of Gastroenterology (2018) 53 37-51. DOI: 10.1007/s00535-017-1375-5 15. Chen L, Yang H, Li H, He C, Yang L, Lv G. Insights into modifiable risk factors of cholelithiasis: a mendelian randomization study. Hepatology (2022) 75 785-796. DOI: 10.1002/hep.32183 16. Coleman HG, Xie SH, Lagergren J. The epidemiology of esophageal adenocarcinoma. Gastroenterology (2018) 154 390-405. DOI: 10.1053/j.gastro.2017.07.046 17. Davey Smith G, Hemani G. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Human Molecular Genetics (2014) 23 R89-R98. DOI: 10.1093/hmg/ddu328 18. Dimou N, Yarmolinsky J, Bouras E, Tsilidis KK, Martin RM, Lewis SJ, Gram IT, Bakker MF, Brenner H, Figueiredo JC, Fortner RT, Gruber SB, van Guelpen B, Hsu L, Kaaks R, Kweon S-S, Lin Y, Lindor NM, Newcomb PA, Sánchez M-J, Severi G, Tindle HA, Tumino R, Weiderpass E, Gunter MJ, Murphy N. Causal effects of lifetime smoking on breast and colorectal cancer risk: Mendelian randomization study. Cancer Epidemiology, Biomarkers & Prevention (2021) 30 953-964. DOI: 10.1158/1055-9965.EPI-20-1218 19. Eusebi LH, Ratnakumaran R, Yuan Y, Solaymani-Dodaran M, Bazzoli F, Ford AC. Global prevalence of, and risk factors for, gastro-oesophageal reflux symptoms: a meta-analysis. Gut (2018) 67 430-440. DOI: 10.1136/gutjnl-2016-313589 20. Fund WCR Research AIfC 2007Food, nutrition, physical activity, and the prevention of cancer: a global perspectiveAmer Inst for Cancer Research. Food, nutrition, physical activity, and the prevention of cancer: a global perspective (2007) 21. Alcohol use and burden for 195 countries and territories, 1990-2016: a systematic analysis for the global burden of disease study 2016. Lancet (2018) 392 1015-1035. DOI: 10.1016/S0140-6736(18)31310-2 22. Global burden of 87 risk factors in 204 countries and territories, 1990-2019: a systematic analysis for the global burden of disease study 2019. Lancet (2020) 396 1223-1249. DOI: 10.1016/S0140-6736(20)30752-2 23. Georgiou AN, Ntritsos G, Papadimitriou N, Dimou N, Evangelou E. Cigarette smoking, coffee consumption, alcohol intake, and risk of Crohn’s disease and ulcerative colitis: a Mendelian randomization study. Inflammatory Bowel Diseases (2021) 27 162-168. DOI: 10.1093/ibd/izaa152 24. Givi ME, Folkerts G, Wagenaar GTM, Redegeld FA, Mortaz E. Cigarette smoke differentially modulates dendritic cell maturation and function in time. Respiratory Research (2015) 16. DOI: 10.1186/s12931-015-0291-6 25. Guindo-Martínez M, Amela R, Bonàs-Guarch S, Puiggròs M, Salvoro C, Miguel-Escalada I, Carey CE, Cole JB, Rüeger S, Atkinson E, Leong A, Sanchez F, Ramon-Cortes C, Ejarque J, Palmer DS, Kurki M, Aragam K, Florez JC, Badia RM, Mercader JM, Torrents D. The impact of non-additive genetic associations on age-related complex diseases. Nature Communications (2021) 12. DOI: 10.1038/s41467-021-21952-4 26. Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D, Laurin C, Burgess S, Bowden J, Langdon R, Tan VY, Yarmolinsky J, Shihab HA, Timpson NJ, Evans DM, Relton C, Martin RM, Davey Smith G, Gaunt TR, Haycock PC. The MR-Base platform supports systematic causal inference across the human phenome. eLife (2018) 7. DOI: 10.7554/eLife.34408 27. Jorgenson E, Thai KK, Hoffmann TJ, Sakoda LC, Kvale MN, Banda Y, Schaefer C, Risch N, Mertens J, Weisner C, Choquet H. Genetic contributors to variation in alcohol consumption vary by race/ethnicity in a large multi-ethnic genome-wide association study. Molecular Psychiatry (2017) 22 1359-1367. DOI: 10.1038/mp.2017.101 28. Kato I, Nomura AM, Stemmermann GN, Chyou PH. A prospective study of gastric and duodenal ulcer and its relation to smoking, alcohol, and diet. American Journal of Epidemiology (1992) 135 521-530. DOI: 10.1093/oxfordjournals.aje.a116319 29. Klatsky AL. Diet, alcohol, and health: a story of connections, confounders, and cofactors. The American Journal of Clinical Nutrition (2001) 74 279-280. DOI: 10.1093/ajcn/74.3.279 30. Krishnan B, Babu S, Walker J, Walker AB, Pappachan JM. Gastrointestinal complications of diabetes mellitus. World Journal of Diabetes (2013) 4 51-63. DOI: 10.4239/wjd.v4.i3.51 31. Kurki MI, Karjalainen J, Palta P, Sipilä TP, Kristiansson K, Donner K, Reeve MP, Laivuori H, Aavikko M, Kaunisto MA, Loukola A, Lahtela E, Mattsson H, Laiho P, Della Briotta Parolo P, Lehisto A, Kanai M, Mars N, Rämö J, Kiiskinen T, Heyne HO, Veerapen K, Rüeger S, Lemmelä S, Zhou W, Ruotsalainen S, Pärn K, Hiekkalinna T, Koskelainen S, Paajanen T, Llorens V, Gracia-Tabuenca J, Siirtola H, Reis K, Elnahas AG, Aalto-Setälä K, Alasoo K, Arvas M, Auro K, Biswas S, Bizaki-Vallaskangas A, Carpen O, Chen CY, Dada OA, Ding Z, Ehm MG, Eklund K, Färkkilä M, Finucane H, Ganna A, Ghazal A, Graham RR, Green E, Hakanen A, Hautalahti M, Hedman Å, Hiltunen M, Hinttala R, Hovatta I, Hu X, Huertas-Vazquez A, Huilaja L, Hunkapiller J, Jacob H, Jensen JN, Joensuu H, John S, Julkunen V, Jung M, Junttila J, Kaarniranta K, Kähönen M, Kajanne RM, Kallio L, Kälviäinen R, Kaprio J, Kerimov N, Kettunen J, Kilpeläinen E, Kilpi T, Klinger K, Kosma VM, Kuopio T, Kurra V, Laisk T, Laukkanen J, Lawless N, Liu A, Longerich S, Mägi R, Mäkelä J, Mäkitie A, Malarstig A, Mannermaa A, Maranville J, Matakidou A, Meretoja T, Mozaffari SV, Niemi MEK, Niemi M, Niiranen T, O’Donnell CJ, Obeidat M, Okafo G, Ollila HM, Palomäki A, Palotie T, Partanen J, Paul DS, Pelkonen M, Pendergrass RK, Petrovski S, Pitkäranta A, Platt A, Pulford D, Punkka E, Pussinen P, Raghavan N, Rahimov F, Rajpal D, Renaud NA, Riley-Gillis B, Rodosthenous R, Saarentaus E, Salminen A, Salminen E, Salomaa V, Schleutker J, Serpi R, Shen H, Siegel R, Silander K, Siltanen S, Soini S, Soininen H, Sul JH, Tachmazidou I, Tasanen K, Tienari P, Toppila-Salmi S, Tukiainen T, Tuomi T, Turunen JA, Ulirsch JC, Vaura F, Virolainen P, Waring J, Waterworth D, Yang R, Nelis M, Reigo A, Metspalu A, Milani L, Esko T, Fox C, Havulinna AS, Perola M, Ripatti S, Jalanko A, Laitinen T, Mäkelä T, Plenge R, McCarthy M, Runz H, Daly MJ, Palotie A. FinnGen: Unique Genetic Insights from Combining Isolated Population and National Health Register Data. medRxiv (2022). DOI: 10.1101/2022.03.03.22271360 32. Larsson SC, Carter P, Kar S, Vithayathil M, Mason AM, Michaëlsson K, Burgess S, Tsilidis KK. Smoking, alcohol consumption, and cancer: a mendelian randomisation study in UK biobank and international genetic consortia participants. PLOS Medicine (2020) 17. DOI: 10.1371/journal.pmed.1003178 33. Laszkowska M, Rodriguez S, Kim J, Hur C. Heavy alcohol use is associated with gastric cancer: analysis of the National health and nutrition examination survey from 1999 to 2010. The American Journal of Gastroenterology (2021) 116 1083-1086. DOI: 10.14309/ajg.0000000000001166 34. Leone FT, Zhang Y, Evers-Casey S, Evins AE, Eakin MN, Fathi J. Initiating pharmacologic treatment in tobacco-dependent adults. An Official American Thoracic Society Clinical Practice Guideline. Am J Respir Crit Care Med (2020) 202 e5-e31. DOI: 10.1164/rccm.202005-1982ST 35. Liu B, Balkwill A, Roddam A, Brown A, Beral V, Million Women Study C. Separate and joint effects of alcohol and smoking on the risks of cirrhosis and gallbladder disease in middle-aged women. American Journal of Epidemiology (2009) 169 153-160. DOI: 10.1093/aje/kwn280 36. Liu JZ, van Sommeren S, Huang H, Ng SC, Alberts R, Takahashi A, Ripke S, Lee JC, Jostins L, Shah T, Abedian S, Cheon JH, Cho J, Dayani NE, Franke L, Fuyuno Y, Hart A, Juyal RC, Juyal G, Kim WH, Morris AP, Poustchi H, Newman WG, Midha V, Orchard TR, Vahedi H, Sood A, Sung JY, Malekzadeh R, Westra H-J, Yamazaki K, Yang S-K, Barrett JC, Alizadeh BZ, Parkes M, Bk T, Daly MJ, Kubo M, Anderson CA, Weersma RK. Association analyses identify 38 susceptibility loci for inflammatory bowel disease and highlight shared genetic risk across populations. Nature Genetics (2015) 47 979-986. DOI: 10.1038/ng.3359 37. Liu M, Jiang Y, Wedow R, Li Y, Brazel DM, Chen F, Datta G, Davila-Velderrain J, McGuire D, Tian C, Zhan X, Choquet H, Docherty AR, Faul JD, Foerster JR, Fritsche LG, Gabrielsen ME, Gordon SD, Haessler J, Hottenga J-J, Huang H, Jang S-K, Jansen PR, Ling Y, Mägi R, Matoba N, McMahon G, Mulas A, Orrù V, Palviainen T, Pandit A, Reginsson GW, Skogholt AH, Smith JA, Taylor AE, Turman C, Willemsen G, Young H, Young KA, Zajac GJM, Zhao W, Zhou W, Bjornsdottir G, Boardman JD, Boehnke M, Boomsma DI, Chen C, Cucca F, Davies GE, Eaton CB, Ehringer MA, Esko T, Fiorillo E, Gillespie NA, Gudbjartsson DF, Haller T, Harris KM, Heath AC, Hewitt JK, Hickie IB, Hokanson JE, Hopfer CJ, Hunter DJ, Iacono WG, Johnson EO, Kamatani Y, Kardia SLR, Keller MC, Kellis M, Kooperberg C, Kraft P, Krauter KS, Laakso M, Lind PA, Loukola A, Lutz SM, Madden PAF, Martin NG, McGue M, McQueen MB, Medland SE, Metspalu A, Mohlke KL, Nielsen JB, Okada Y, Peters U, Polderman TJC, Posthuma D, Reiner AP, Rice JP, Rimm E, Rose RJ, Runarsdottir V, Stallings MC, Stančáková A, Stefansson H, Thai KK, Tindle HA, Tyrfingsson T, Wall TL, Weir DR, Weisner C, Whitfield JB, Winsvold BS, Yin J, Zuccolo L, Bierut LJ, Hveem K, Lee JJ, Munafò MR, Saccone NL, Willer CJ, Cornelis MC, David SP, Hinds DA, Jorgenson E, Kaprio J, Stitzel JA, Stefansson K, Thorgeirsson TE, Abecasis G, Liu DJ, Vrieze S. Association studies of up to 1.2 million individuals yield new insights into the genetic etiology of tobacco and alcohol use. Nature Genetics (2019) 51 237-244. DOI: 10.1038/s41588-018-0307-5 38. Mandrekar P, Szabo G. Signalling pathways in alcohol-induced liver inflammation. Journal of Hepatology (2009) 50 1258-1266. DOI: 10.1016/j.jhep.2009.03.007 39. McGee EE, Jackson SS, Petrick JL, Van Dyke AL, Adami H-O, Albanes D, Andreotti G, Beane-Freeman LE, Berrington de Gonzalez A, Buring JE, Chan AT, Chen Y, Fraser GE, Freedman ND, Gao Y-T, Gapstur SM, Gaziano JM, Giles GG, Grant EJ, Grodstein F, Hartge P, Jenab M, Kitahara CM, Knutsen SF, Koh W-P, Larsson SC, Lee I-M, Liao LM, Luo J, Milne RL, Monroe KR, Neuhouser ML, O’Brien KM, Peters U, Poynter JN, Purdue MP, Robien K, Sandler DP, Sawada N, Schairer C, Sesso HD, Simon TG, Sinha R, Stolzenberg-Solomon R, Tsugane S, Wang R, Weiderpass E, Weinstein SJ, White E, Wolk A, Yuan J-M, Zeleniuch-Jacquotte A, Zhang X, Zhu B, McGlynn KA, Campbell PT, Koshiol J. Smoking, alcohol, and biliary tract cancer risk: a pooling project of 26 prospective studies. Journal of the National Cancer Institute (2019) 111 1263-1278. DOI: 10.1093/jnci/djz103 40. McNabb S, Harrison TA, Albanes D, Berndt SI, Brenner H, Caan BJ, Campbell PT, Cao Y, Chang-Claude J, Chan A, Chen Z, English DR, Giles GG, Giovannucci EL, Goodman PJ, Hayes RB, Hoffmeister M, Jacobs EJ, Joshi AD, Larsson SC, Le Marchand L, Li L, Lin Y, Männistö S, Milne RL, Nan H, Newton CC, Ogino S, Parfrey PS, Petersen PS, Potter JD, Schoen RE, Slattery ML, Su Y-R, Tangen CM, Tucker TC, Weinstein SJ, White E, Wolk A, Woods MO, Phipps AI, Peters U. Meta-Analysis of 16 studies of the association of alcohol with colorectal cancer. International Journal of Cancer (2020) 146 861-873. DOI: 10.1002/ijc.32377 41. Mizrahi JD, Surana R, Valle JW, Shroff RT. Pancreatic cancer. The Lancet (2020) 395 2008-2020. DOI: 10.1016/S0140-6736(20)30974-0 42. Montgomery SM, Pounder RE, Wakefield AJ. Smoking in adults and passive smoking in children are associated with acute appendicitis. Lancet (1999) 353. DOI: 10.1016/S0140-6736(05)74951-5 43. Noakes PS, Hale J, Thomas R, Lane C, Devadason SG, Prescott SL. Maternal smoking is associated with impaired neonatal toll-like-receptor-mediated immune responses. The European Respiratory Journal (2006) 28 721-729. DOI: 10.1183/09031936.06.00050206 44. Nordenstedt H, Graham DY, Kramer JR, Rugge M, Verstovsek G, Fitzgerald S, Alsarraj A, Shaib Y, Velez ME, Abraham N, Anand B, Cole R, El-Serag HB. Helicobacter pylori-negative gastritis: prevalence and risk factors. The American Journal of Gastroenterology (2013) 108 65-71. DOI: 10.1038/ajg.2012.372 45. Park S, Chun J, Han K-D, Soh H, Kang EA, Lee HJ, Im JP, Kim JS. Dose-Response relationship between cigarette smoking and risk of ulcerative colitis: a nationwide population-based study. Journal of Gastroenterology (2019) 54 881-890. DOI: 10.1007/s00535-019-01589-3 46. Peery AF, Crockett SD, Murphy CC, Jensen ET, Kim HP, Egberg MD, Lund JL, Moon AM, Pate V, Barnes EL, Schlusser CL, Baron TH, Shaheen NJ, Sandler RS. Burden and cost of gastrointestinal, liver, and pancreatic diseases in the United States: update 2021. Gastroenterology (2022) 162 621-644. DOI: 10.1053/j.gastro.2021.10.017 47. Piovani D, Danese S, Peyrin-Biroulet L, Nikolopoulos GK, Lytras T, Bonovas S. Environmental risk factors for inflammatory bowel diseases: an umbrella review of meta-analyses. Gastroenterology (2019) 157 647-659. DOI: 10.1053/j.gastro.2019.04.016 48. Roberts W, Verplaetse T, Peltier MKR, Moore KE, Gueorguieva R, McKee SA. Prospective association of e-cigarette and cigarette use with alcohol use in two waves of the population assessment of tobacco and health. Addiction (2020) 115 1571-1579. DOI: 10.1111/add.14980 49. Roerecke M, Vafaei A, Hasan OSM, Chrystoja BR, Cruz M, Lee R, Neuman MG, Rehm J. Alcohol consumption and risk of liver cirrhosis: a systematic review and meta-analysis. American Journal of Gastroenterology (2019) 114 1574-1586. DOI: 10.14309/ajg.0000000000000340 50. Ryan-Harshman M, Aldoori W. How diet and lifestyle affect duodenal ulcers. Review of the evidence. Canadian Family Physician (2004) 50 727-732. PMID: 15171675 51. Samokhvalov AV, Rehm J, Roerecke M. Alcohol consumption as a risk factor for acute and chronic pancreatitis: a systematic review and a series of meta-analyses. EBioMedicine (2015) 2 1996-2002. DOI: 10.1016/j.ebiom.2015.11.023 52. Simpson RF, Hermon C, Liu B, Green J, Reeves GK, Beral V, Floud S. Alcohol drinking patterns and liver cirrhosis risk: analysis of the prospective UK Million women study. The Lancet. Public Health (2019) 4 e41-e48. DOI: 10.1016/S2468-2667(18)30230-5 53. Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, Downey P, Elliott P, Green J, Landray M, Liu B, Matthews P, Ong G, Pell J, Silman A, Young A, Sprosen T, Peakman T, Collins R. Uk Biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLOS Medicine (2015) 12. DOI: 10.1371/journal.pmed.1001779 54. Talley NJ, Powell N, Walker MM, Jones MP, Ronkainen J, Forsberg A, Kjellström L, Hellström PM, Aro P, Wallner B, Agréus L, Andreasson A. Role of smoking in functional dyspepsia and irritable bowel syndrome: three random population-based studies. Alimentary Pharmacology & Therapeutics (2021) 54 32-42. DOI: 10.1111/apt.16372 55. Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nature Genetics (2018) 50 693-698. DOI: 10.1038/s41588-018-0099-7 56. Wootton RE, Richmond RC, Stuijfzand BG, Lawn RB, Sallis HM, Taylor GMJ, Hemani G, Jones HJ, Zammit S, Davey Smith G, Munafò MR. Evidence for causal effects of lifetime smoking on risk for depression and schizophrenia: a Mendelian randomisation study. Psychological Medicine (2020) 50 2435-2443. DOI: 10.1017/S0033291719002678 57. Yadav D, Whitcomb DC. The role of alcohol and smoking in pancreatitis. Nature Reviews. Gastroenterology & Hepatology (2010) 7 131-145. DOI: 10.1038/nrgastro.2010.6 58. Yavorska OO, Burgess S. MendelianRandomization: an R package for performing Mendelian randomization analyses using summarized data. International Journal of Epidemiology (2017) 46 1734-1739. DOI: 10.1093/ije/dyx034 59. Yu C, Tang H, Guo Y, Bian Z, Yang L, Chen Y, Tang A, Zhou X, Yang X, Chen J, Chen Z, Lv J, Li L. Hot tea consumption and its interactions with alcohol and tobacco use on the risk for esophageal cancer: a population-based cohort study. Annals of Internal Medicine (2018) 168 489-497. DOI: 10.7326/M17-2000 60. Yuan S, Giovannucci EL, Larsson SC. Gallstone disease, diabetes, calcium, triglycerides, smoking and alcohol consumption and pancreatitis risk: Mendelian randomization study. NPJ Genomic Medicine (2021) 6. DOI: 10.1038/s41525-021-00189-6 61. Yuan S, Larsson SC. Adiposity, diabetes, lifestyle factors and risk of gastroesophageal reflux disease: a mendelian randomization study. European Journal of Epidemiology (2022a) 37 747-754. DOI: 10.1007/s10654-022-00842-z 62. Yuan S, Larsson SC. Genetically predicted adiposity, diabetes, and lifestyle factors in relation to diverticular disease. Clinical Gastroenterology and Hepatology (2022b) 20 1077-1084. DOI: 10.1016/j.cgh.2021.06.013 63. Yuan S, Chen J, Li X, Fan R, Arsenault B, Gill D, Giovannucci EL, Zheng JS, Larsson SC. Lifestyle and metabolic factors for nonalcoholic fatty liver disease: mendelian randomization study. European Journal of Epidemiology (2022c) 37 723-733. DOI: 10.1007/s10654-022-00868-3 64. Zhang Y-B, Pan X-F, Chen J, Cao A, Zhang Y-G, Xia L, Wang J, Li H, Liu G, Pan A. Combined lifestyle factors, incident cancer, and cancer mortality: a systematic review and meta-analysis of prospective cohort studies. British Journal of Cancer (2020) 122 1085-1093. DOI: 10.1038/s41416-020-0741-x 65. Zhou W, Zhao Z, Nielsen JB, Fritsche LG, LeFaive J, Gagliano Taliun SA, Bi W, Gabrielsen ME, Daly MJ, Neale BM, Hveem K, Abecasis GR, Willer CJ, Lee S. Scalable generalized linear mixed model for region-based association tests in large biobanks and cohorts. Nature Genetics (2020) 52 634-639. DOI: 10.1038/s41588-020-0621-6
--- title: Network-based multi-omics integration reveals metabolic at-risk profile within treated HIV-infection authors: - Flora Mikaeloff - Marco Gelpi - Rui Benfeitas - Andreas D Knudsen - Beate Vestad - Julie Høgh - Johannes R Hov - Thomas Benfield - Daniel Murray - Christian G Giske - Adil Mardinoglu - Marius Trøseid - Susanne D Nielsen - Ujjwal Neogi journal: eLife year: 2023 pmcid: PMC10017104 doi: 10.7554/eLife.82785 license: CC BY 4.0 --- # Network-based multi-omics integration reveals metabolic at-risk profile within treated HIV-infection ## Abstract Multiomics technologies improve the biological understanding of health status in people living with HIV on antiretroviral therapy (PWH). Still, a systematic and in-depth characterization of metabolic risk profile during successful long-term treatment is lacking. Here, we used multi-omics (plasma lipidomic, metabolomic, and fecal 16 S microbiome) data-driven stratification and characterization to identify the metabolic at-risk profile within PWH. Through network analysis and similarity network fusion (SNF), we identified three groups of PWH (SNF-1–3): healthy (HC)-like (SNF-1), mild at-risk (SNF-3), and severe at-risk (SNF-2). The PWH in the SNF-2 ($45\%$) had a severe at-risk metabolic profile with increased visceral adipose tissue, BMI, higher incidence of metabolic syndrome (MetS), and increased di- and triglycerides despite having higher CD4+ T-cell counts than the other two clusters. However, the HC-like and the severe at-risk group had a similar metabolic profile differing from HIV-negative controls (HNC), with dysregulation of amino acid metabolism. At the microbiome profile, the HC-like group had a lower α-diversity, a lower proportion of men having sex with men (MSM) and was enriched in Bacteroides. In contrast, in at-risk groups, there was an increase in Prevotella, with a high proportion of MSM, which could potentially lead to higher systemic inflammation and increased cardiometabolic risk profile. The multi-omics integrative analysis also revealed a complex microbial interplay of the microbiome-associated metabolites in PWH. Those severely at-risk clusters may benefit from personalized medicine and lifestyle intervention to improve their dysregulated metabolic traits, aiming to achieve healthier aging. ## Introduction Antiretroviral therapy (ART) has improved the immune profile by suppressing viral replication and reducing the morbidity and mortality of people living with HIV (PWH). Yet living with HIV under ART induces a strong metabolic perturbation in the body due to virus persistence, immune activation, chronic low-grade inflammation, and treatment toxicity, mostly with older antiretrovirals (Yoshimura, 2017). The biological shifts due to a mixed effect of drugs and viruses are also highly personalized depending on the patient genetic background, age, sex, immunological, and lifestyle factors (Pelchen-Matthews et al., 2018). Long-term HIV infection, even with viral suppression, is associated with an accentuated onset of non-AIDS-related comorbidities (Deeks, 2011). Consequently, diseases of the aged population appear in relatively young HIV patients, including cardiovascular disease, liver-kidney disease, and neurocognitive and metabolic disorders (Nasi et al., 2017). Systems biological analyses are valuable methodologies for systematically understanding pathology and identifying potential novel treatment strategies (Karahalil, 2016). Microbiome studies have provided enormous knowledge about the microbial association with HIV status, sexual practice, and gender (Zhou et al., 2020; Gelpi et al., 2020; Noguera-Julian et al., 2016), and the possible interplay between HIV-related gut microbiota, immune dysfunction, and comorbidities like metabolic syndrome (MetS), and visceral adipose tissue (VAT) accumulation (Gelpi et al., 2020). Our extensive metabolomics studies from three different cohorts from India (Babu et al., 2019), Cameroon (Mikaeloff et al., 2022), and Denmark (Gelpi et al., 2021) with more than 500 PWH have indicated disrupted amino acid (AA) metabolism in PWH with ART (PWH) following prolonged treatment that plays the central role in the comorbidities such as MetS (Gelpi et al., 2021). The application of integrative omics to understand the disease pathogenesis in PWH under suppressive ART is lacking. To our knowledge, no integrative omics studies have been performed to understand complex biological phenotypes in PWH during prolonged suppressive ART. Multi-omic characterizations may offer insights into understanding the mechanisms underlying biological processes in a specific disease condition. A recent longitudinal study integrating metabolomics, plasma protein biomarkers, and transcriptomics in patients' samples identified potential lipid and amino acid metabolism perturbations in PWH with immune reconstitution inflammatory syndrome (IRIS) (Pei et al., 2021). Our recent network-based integrative plasma lipidomics, metabolic biomarker, and clinical data indicated a coordinated role of clinical parameters like accumulation of visceral adipose tissue (VAT) and exposure to earlier generations of antiretrovirals with glycerolipids and glutamate metabolism in the pathogenesis of PWH with MetS (Olund Villumsen et al., 2021). The present study aimed to identify a molecular data-driven phenotypic patient stratification using network-based integration of plasma metabolomics/lipidomics and fecal microbiota within a cohort of PWH with prolonged suppressive therapy who were at-risk of metabolic complications. We further investigated the underlying factors differing from these profiles and the link to their clinical phenotype to clarify the risk factors for metabolic disease. ## Comprehensive multi-omics characterization of PWH on successful cART In this study, we used untargeted plasma metabolomics (877 metabolites) (Gelpi et al., 2021), lipidomics (977 lipids) (Olund Villumsen et al., 2021), and fecal 16 S rRNA microbiome [241 amplicon sequence variants (ASVs)] data (Gelpi et al., 2020) from 97 PWH from the Copenhagen Comorbidity (COCOMO) cohort (Gelpi et al., 2018) where we have three types of omics data available. Additionally, we included 42 clinical and demographical features comprising lifestyle habits (food, medicine, alcohol, smoking), comorbidities linked to obesity and non-communicable chronic comorbidities (e.g. liver function, kidney function, and diabetes), and HIV-related measurements (viral load, treatment history, CD4 T-cell count, CD8 T-cell counts) (Appendix 1). The PWH were mainly male ($86\%$, $\frac{84}{97}$), of Caucasian ethnic origin ($81\%$, $\frac{79}{97}$), with a median (IQR) age of 54 [48-63] years. The median (IQR) duration of the treatment was 15 [9-18] years. At the time of sample collection, the viral load was below the detection level with successful immune reconstitution [median (IQR) CD4 T-cell count 713 [570-900] cells/µL] (Table 1). Additionally, 20 HIV-negative controls (HC) from the Danish population with similar sex proportions ($90\%$ male, $\frac{18}{20}$) and median age (IQR) of 56 [50-67] years with slightly higher median (IQR) BMI 26 [23-29] compared to the complete cohort [24 [22-27], $$p \leq 0.04$$; Table 1—source data 1]. The HC was used to reference multi-omics and define the HC-like PWH. Table 1. \| Unnamed: 0 \| Complete Cohort \| SNF-1 \| SNF- 2 \| SNF-3 \| P values \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| At-risk Classification \| \| HC-like \| Severe at risk \| Mild \| \| \| N \| 97 \| 19 \| 44 \| 34 \| \| \| Age in years, Median (IQR) \| 54 (48–63) \| 60 (48–68) \| 54 (48–62) \| 54 (51–60) \| 0.75 \| \| Gender, Male, N (%) \| 84 (87) \| 15 (79) \| 40 (91) \| 29 (85) \| 0.36 \| \| Ethnicity Caucasian, N (%) \| 79 (81) \| 15 (79) \| 38 (87) \| 26 (77) \| 0.49 \| \| Mode of transmission, N (%) Homosexual/bisexual Heterosexual Other/unknown \| 63 (65)26 (27)8 (8) \| 9 (47)7 (37)3 (16) \| 36 (81)6 (14)2 (5) \| 18 (53)13 (38)3 (9) \| 0.017 \| \| CD4 Nadir, cells/mL, Median (IQR) \| 235 (123–320) \| 240 (127–330) \| 240 (145–365) \| 223 (42–290) \| 0.49 \| \| CD4 at ART Initiation, cells/mL, Median (IQR) \| 287 (155–410) \| 270 (120–360) \| 318 (192–463) \| 240 (108–320) \| 0.11 \| \| Viral Load at ART initiation, log copies/mL, Median (IQR) \| 5.02 (4.34–5.61) \| 4.87 (4.32–5.5) \| 5.11 (4.74–5.61) \| 4.94 (4.2–5.55) \| 0.35 \| \| CD4 at sampling, cells/mL, Median(IQR) \| 713 (570–900) \| 680 (540–958) \| 762 (689–923) \| 610 (475–819) \| 0.015 \| \| CD8 at sampling, cells/mL, Median (IQR) \| 775 (600–1100) \| 780 (630–879) \| 894 (638–1300) \| 700 (530–870) \| 0.054 \| \| Viral load (<50 copies/mL), N (%) \| 97 (100) \| 19 (100) \| 44 (100) \| 34 (100) \| 1.0 \| \| Duration of treatment in years, median (IQR) \| 15 (9–18) \| 15 (13–18) \| 15 (8–18) \| 14 (7–17) \| 0.73 \| \| Current Treatment, 1st drug, N (%) ABC TDF/TAF Other \| 31 (32)42 (43)24 (25) \| 8 (42)8 (42)3 (16) \| 13 (30)19 (43)12 (27) \| 10 (29)15 (44)9 (27) \| 0.84 \| \| Current Treatment, 3rd drug, N (%) NNRTI PI/r INSTI Other \| 38 (39)18 (19)15 (15)26 (27) \| 8 (42)4 (21)4 (21)3 (16) \| 14 (32)11 (25)6 (14)13 (29) \| 16 (47)3 (9)5 (15)10 (29) \| 0.45 \| \| BMI, Mean (SD) \| 24 (22–27) \| 22 (19–25) \| 26 (23–28) \| 24 (22–27) \| 0.003 \| \| VAT, Median (IQR) \| 89 (36–142) \| 41 (19–106) \| 127 (79–196) \| 69 (26–100) \| 0.0001 \| \| SAT, Median (IQR) \| 111 (70–167) \| 69 (33–115) \| 117 (82–174) \| 119 (83–190) \| 0.02 \| \| MetS, N (%) \| 43 (44) \| 6 (32) \| 31 (70) \| 6 (17) \| 9e-06 \| \| Central obesity, N(%) \| 57 (59) \| 8 (42) \| 32 (73) \| 17 (50) \| 0.033 \| \| Waist circumference (cm) \| 94 (87–101) \| 90 (84–95) \| 100 (91–105) \| 90 (87–97) \| 0.0007 \| \| Hypertension, N (%) \| 49 (51) \| 5 (26) \| 23 (52) \| 21 (62) \| 0.04 \| ## Integrative omics-based similarity network fusion (SNF) identifies three clusters in PWH To stratify the PWH based on their molecular signature, we used Similarity Network Fusion (SNF) that constructs similarity matrices and networks of PWH for each of the omics and fuses them into one network that represents the full spectrum of the underlying data and disease status in PWH (Wang et al., 2014). We identified three clusters of patients, defined as SNF-1 ($$n = 19$$), SNF-2 ($$n = 44$$), and SNF-3 ($$n = 34$$) (Figure 1A). The concordance matrix based on Normalized Mutual Information (NMI) score (0=no mutual information, 1=perfect correlation) showed that lipids had the most influence in the final network (NMI = 0.6), followed by metabolites (NMI = 0.4) and finally, microbiome (NMI = 0.3) (Figure 1B). Clear segregation of the SNF clusters (Figure 1C) was observed on the PCA plot based on the fused network values (Figure 1D) and PCA of single omics for lipidomics (Figure 1—figure supplement 1A) and metabolomics (Figure 1—figure supplement 1B) but not microbiome (Figure 1—figure supplement 1C). Figure 1.: Similarity network fusion-based PWH stratification using lipidomics, metabolomics, and microbiome integration.(A) Scatter plot showing the maximization of Eigen gap and the minimization of rotation cost for optimizing the number of clusters. (B) Concordance matrix between the combined network (SNF) and each omics network based on NMI calculation (0=no mutual information, 1=perfect correlation). (C) SNF-combined similarity network colored by clusters (SNF-1/HC-like=blue, SNF-2/severe at-risk=yellow, SNF-3/mild at-risk=grey) obtained after spectral clustering. Edges' color indicates the strength of the similarity (black = strong, grey = weak). (D) PCA plot of samples based on fused network. Samples are colored by condition. ## Cluster-specific clinical characteristics define a metabolic at-risk group Cluster-specific clinical characteristics of PWH are presented in Table 1. Clusters were not statistically different for age, gender, duration of ART, and type of ART($p \leq 0.05$). On the other hand, SNF-1 had the healthiest profile (herein HC-like group), SNF-3 an intermediate (herein mild at-risk group, and SNF-2 the most severe metabolic perturbations herein severe at-risk group), indicating an at-risk metabolic profile. The severe at-risk group represented patients with high BMI, central obesity, higher VAT, and incidence of MetS (all $p \leq 0.05$) but there was no association with measures of liver damage (alanine aminotransferase, ALT) or reduced kidney function (estimated glomerular filtration rate, eGFR), all $p \leq 0.05$ (Table 1—source data 2). Regardless of disease severity, the severe at-risk group’s patients had a higher CD4+ T-cell count at the time of sample collection and more men who have sex with men (MSM) as transmission mode compared to the other clusters (all $p \leq 0.05$) considered as confounding factors here. The at-risk groups, severe and mild, had a significantly higher subcutaneous adipose tissue (SAT) and incidence of hypertension compared to the HC-like cluster (all $p \leq 0.05$). The HC-like cluster had the lowest BMI, SAT, VAT, and incidence of hypertension (all $p \leq 0.05$). ## Lipids and metabolites highlight clinical differences between patient clusters Next, we performed the differential metabolite and lipid class abundance between the clusters. A similar lipid profile was observed between the HC-like, mild at-risk groups and HC (Figure 2A and B, and Supplementary file 1). Patients from the severe at-risk group showed a significant increase in diglycerides (DAG; Figure 2A) and triglycerides (TAG) (Figure 2B) compared to HC-like, mild at-risk cluster, and HC (all FDR <0.1) as well as other lipids classes which coordinate with their clinical metabolic profile (Figure 2—figure supplement 1). After adjusting for two confounders' modes of HIV transmission and CD4 count at sampling that are different between the clusters, the trends for lipid class remained the same (Figure 2—source data 1). In this analysis, the relation between cluster and ART class was not significant (χ2, FDR = 0.45). Still, we can mention that the three groups had an important proportion of missing data for this variable ($16\%$, $29\%$, and $29\%$, respectively). Figure 2.: Lipidomics and metabolomics, characterization of the PWH clusters.(A) Boxplots of DAG from untargeted lipid classes separated by groups. Significant stars are displayed for each comparison with FDR <0.05, FDR <0.01, FDR <0.001 (limma). (B) Boxplots of TAG from untargeted lipid classes separated by groups. (C) PCA plot of samples after prior standardization based on significant metabolites between at least one pairwise comparison (limma, FDR <0.05). Variance proportions are written on each component axis. Samples are colored by condition. (D) Circular heatmap of the top 159 metabolites (FDR <0.005). Metabolites are represented as slices and labeled around the plot. LogFold Change from significant metabolites between groups is displayed in the first six outer layers. The 7th to 9th layers represents the coefficient of correlation between metabolites and BMI, metabolites and age (Spearman, p <0.1, absolute R>0.15) and the p-value from significant associations between metabolites and gender (χ2, p<0.1). The inner layer represents the pathway of each metabolite. (E) PCA plot based on metabolites differing clusters adjusted for transmission mode and CD4 count. Figure 2—source data 1.Table of differential lipid abundance analysis SNFs by lipids classes by clusters and corrected for transmission mode and CD4 count. To identify the global metabolite impact on the cluster, we performed differential metabolite abundance (DMA) analysis. We kept stringent statistical parameters (FDR <0.005) and identified 159 metabolites with highly different metabolites among the groups (Supplementary file 2). The mild at-risk group and HC had only nine metabolites differing, in line with the high clustering of both groups shown with PCA (Figure 2C). The most perturbations were observed between HC and the HC-like PWH ($\frac{124}{159}$) and HC and severe at-risk group ($\frac{62}{159}$) (Figure 2D). Compared to HC, these clusters showed an up-regulation of the metabolites in the xenobiotics, nucleotides, and amino acid metabolism. In turn, the HC-like and severe at-risk groups showed similar metabolic profiles. Among these 159 metabolites, 50 had a low or moderate association with age and BMI (Spearman correlation, absolute R<0.4, $p \leq 0.1$) and 51 with gender (χ2, $p \leq 0.1$), showing the modest influence of individual characteristics on metabolomics profile. Within the PWH groups, after adjusting for the two confounders, the supervised principal component analysis of the significantly different metabolites ($$n = 217$$) identified distinct clusters of HC-like, mild, and severe at-risk groups (Figure 2E and Supplementary file 3). The DMA identified the similarity of HC-like and severe at-risk groups with only 15 metabolites significantly different (FDR <0.05); most were part of lipid metabolism. Combining the in-depth metabolomics and lipidomic data indicated more personalized risk factors for PWH that the clinical features cannot explain. A complex interplay between the multi-omics layers defines overall health status. ## Sexual preferences influence the clusters' differences driven by the microbiome As the metabolic aberrations were closely linked with the microbiome profile, we investigated the microbiome’s impact on PWH clusters. The α-diversity indices indicated a loss of diversity according to Observed, ACE, se. ACE, Chao1, and Fisher indices in HC-like compared to the severe at-risk group (Mann Whitney, FDR <0.05; Figure 3A, Figure 3—figure supplement 1 and Figure 3—source data 1a). A non-metric multidimensional scaling (NMDS) ordination of the dissimilarity-based index (Bray-Curtis) of diversity at the ASV level was performed to measure the inter-individual differences between groups (β-diversity; Figure 3B). Based on NMDS plot axis coordinate 1, the HC-like group was segregated separately from mild and severe at-risk groups (Mann Whitney, FDR <0.05, Figure 3C). The relative abundance of fecal microbiota was more influenced by the transmission mode than the cluster itself (Figure 3—figure supplement 2A). No other comorbidities on the microbiome profile were observed (Figure 3—figure supplement 2B–D). The severely at-risk group had a significantly higher number of MSM than the other groups (Table 1). While combining severe and mild at-risk groups, there were $69\%$ ($\frac{54}{78}$) MSM in the at-risk clusters and $47\%$ ($\frac{9}{19}$) MSM in the HC-like group. This indicated that sexual preferences and the HIV-1 transmission mode relate to compositional differences in fecal microbiota between clusters. The same effect was observed after correction for transmission mode and CD4 T-count, and alpha diversity did not differ between clusters (Figure 3—source data 1b). Permutational multivariate analysis of variance (PERMANOVA) at the family level showed that the centroids of the HC-like groups were different from the severe at-risk (FDR <0.001) and mild groups (FDR = 0.0054; Figure 3—source data 2), indicating that there is only a location effect as permutation test for homogeneity of multivariate dispersions was not significant between the clusters (FDR >0.05). No statistical difference was observed between the severe and mild at-risk groups in both tests (FDR = 0.38). The HC-like group was enriched in Bacteroides and Lachnospira, while at-risk groups were enriched in Prevotella, Veillonella, and Succinivibrio (Figure 3D–E). These families were also among 54 significantly discriminative features between HC-like and at-risk groups, as shown with linear discriminant analysis effect size (LefSe; Figure 3F). Mann Whitney U test between clusters at the family level also found Prevotellaceae and Bacteroidaceae to be statistically distinct between these clusters (FDR <0.05; Figure 3G). Our data thus support the potential role of the Prevotella and Bacteroides in the cluster separation that the sexual preferences could mediate in PWH than the metabolic risk cluster. Figure 3.: Transmission mode drove cluster differences in microbiome data.(A) Boxplots of alpha diversity indices (Observed, ACE, Chao1, Fisher) separated by HIV cluster. Significant stars are shown for each comparison (Mann-Whitney U test). (B) Non-metric multidimensional scaling (NMDS) plot of Bray-Curtis distances. Samples are colored by clusters. Boxplots based on NMDS1 and NMDS2 are represented. (C) Barplot represents the relative abundance of bacteria at the family level for each patient. Patient information is displayed above the barplot, including cluster, metabolic syndrome (MetS: yes/no), hypertension (yes/no), transmission mode, and gender. (D) Barplot showing the top microbial families by representing their coefficient from PERMANOVA between SNF-1 and SNF-2. (E) Barplot showing the top microbial families between SNF-1 and SNF-3. (F) LEfSe cladogram representing cluster-specific microbial communities to HC-like and to at-risk groups (SNF-2/SNF-3). Top families from PERMANOVA are labeled. (G) Boxplot of relative abundance at family level for Bacteroides (top) and Prevotella (bottom). Significant stars are shown for significant comparisons (Mann-Whitney U test). Figure 3—source data 1.Alpha diversity indices statistics. Figure 3—source data 2.Permutational multivariate analysis of variance at the family level. ## Factor and network analysis indicated the importance of microbiome-associated metabolites To identify the molecular and clinical factors driving SNF cluster separation at the multi-omic level, we employed the Multi-Omic Factor Analysis (MOFA) tool for the multi-omics integration (Argelaguet et al., 2018). After low variance filtering, the MOFA model was built using three views: microbiome with 173 ASVs, metabolome with 676 metabolites, and lipidome with 709 lipids. The model found 15 uncorrelated latent factors (Figure 4—figure supplement 1), that is, combinations of features at the multi-omic level. The total variance was explained at $80\%$ by the lipidome, $22\%$ by the metabolome, and $2\%$ by the microbiome, agreeing with the SNF analysis (Figure 4A). No factor explained most of the variance in the three views (Figure 4B). After, we selected features with the largest weight in each cluster-associated factor (Figure 4C). Features with the most importance based on the top $10\%$ of absolute weight were selected in each view, resulting in 396 features (263 lipids, 111 metabolites, and 22 ASVs). A good cluster separation based on hierarchical clustering of Spearman correlation confirmed the relevance of this subset of features (Figure 4D). We also extracted the top 20 features for each view based on this subset (Figure 4E). Bacteroides and Firmicutes were found in the phylum with the highest weight confirming our results from microbiome analysis and the importance of these microbial communities for cluster separation. Nevertheless, the microbiome had a lower weight than metabolites and lipids in MOFA factors. Among the top 20 metabolite features, three metabolites derived or modified by microbiota (defined as microbiome-associated metabolites; MAM) (3,4−dihydroxybutyrate, 2−oxindole−3−acetate, and indoleacetylglutamine) were found (Figure 4E). To investigate the coordinated role of MAM, we performed the consensus association analysis (Figure 4—figure supplement 2). To balance the different number of features in each of the three omics, we randomly selected 241 metabolites, 241 lipids, and 241 ASVs 1000 times. Significant pairwise correlations (FDR <10–6) found in >$90\%$ of comparisons were used to build a positive co-expression network, and community detection was performed, resulting in a network with 1324 nodes (694 lipids, 536 metabolites, 94 microbial communities), 131863 edges and eight multi-omic communities (N>30). To refine this network, we selected the 396 features based on MOFA differing the most clusters (Figure 4D) in the co-expression network (Figure 4F). The most central communities (Average degree C1=444, Average degree C2=364) were lipid specific (SNF-1, lipids = $\frac{122}{124}$, SNF-2, lipids = $\frac{127}{128}$). In contrast, metabolites enriched communities were sparser with a lower average degree (C3=26, C4=22, C6=10, C7=6) but still connected to lipids with 86 edges between lipids and metabolites. Microbiome-enriched community (c8) did not correlate with metabolites or lipids. However, eight MAMs were found in the network, mostly in c6 ($\frac{5}{21}$), showing that MAMs were highly intercorrelated and could have a potential role in shaping the systemic metabolic and lipid profile. Figure 4.: Factor analysis highlights the essential features for cluster separation and potential microbiome-derived metabolites importance (A) Barplot of total variance explained by MOFA model per view.(B) Variance decomposition plot. The percentage of variance is explained by each factor for each view. (C) External covariate association with factors plot. Association is represented with log10 adjusted p-values from Pearson correlation. (D) Heatmap representing levels of microbial communities, metabolites, and lipids with the higher absolute weight in MOFA factors associated with cluster (F1, F2, F3, F5, F8). Samples are labeled according to the study groups. Data were Z-score transformed. The type of data (lipid, metabolite, microbe) is displayed on the right. (E) Top 20 features with higher absolute weight in MOFA factors associated with cluster (F1, F2, F3, F5, F8) from lipidome, metabolome, and microbiome. Microbiome-derived metabolites and bacterial phylum of interest are colored in blue and red, respectively. (F) MOFA features differing clusters and interactions extracted from the three-layers consensus co-expression network. Microbiome-derived metabolites are labeled. ## MAM is highly associated with clinical features driven by bile acid metabolism and indole derivatives We observed a high correlation among the MAMs (Figure 4F). Therefore, to further investigate their role in PWH, we retrieved 69 metabolites defined as (i) produced by intestinal bacterial mainly part of secondary bile acid metabolism ($$n = 22$$) and (ii) produced by host modified by bacteria ($$n = 47$$, polyamines, propionate, acetate, butyrate, and indole derivatives) as reported (Appendix 2; Postler and Ghosh, 2017). Differential abundance analysis 19 MAMs differed between HC and PWH irrespective of the SNF clusters, and 30 differed between at least one comparison (Figure 5A). The propionate and indole derivates were significantly (FDR <0.05) increased in PWH compared to HC. As observed in the whole metabolomics profile, mild had a more similar profile to HC than HC-like and severe at-risk groups. In contrast, the HC-like and the severe at-risk groups had identical profiles. We performed univariate linear regression to investigate the link between microbiome-derived metabolites and clinical parameters (Figure 5—source data 1). Lithocholate sulfate was associated with obesity-related comorbidities (MetS, SAT, VAT, hypertension, and central obesity) and deoxycholic acid 12-sulfate. Several lifestyle parameters impacted MAM, such as poultry and vegetable intake, smoking, and alcohol. The use of medication as antihypertensives was also associated with three MDMs. Glycolithocholate and glycoursodeoxycholic acid sulfate were linked to HIV-related parameters (CD4 nadir, CD4 at study entry) and patients' demography and lifestyle parameters. The SNF cluster was linked to lithocholate sulfate, 3-ureidopropionate, and imidazole propionate (Figure 5B). Finally, to measure the influence of MAM on plasma metabolomics profile, we only performed association analysis and community detection on metabolomics data (Figure 5C). We obtained a co-expression network with 843 nodes and 15490 edges (FDR <0.02) and observed seven communities (c1-c7) (Figure 4C). The c4 contained all the secondary bile acid metabolites. Though the differential abundance analysis did not show all MAM differences between the SNF clusters and HC, they were highly correlated in PWH, with significant MDMs differing between the groups (Figure 5D). Combining all the data, we showed the essential role of MAMs in the system-level metabolic profile of PWH on successful therapy. Figure 5.: Microbiome-associated metabolites are affected in HIV clusters (A) Heatmap representing abundances of microbiome-derived metabolites differing in at least one comparison.Data were Z-score transformed. Significant logFC (limma, FDR <0.05) of pairwise comparisons between conditions, groups, and under groups of microbiome-derived metabolites are displayed on the right. (B) Cytoscape network showing significant positive and negative associations between clinical parameters and microbiome-derived metabolites (univariate linear regression, FDR <0.05). Clinical parameters are colored based on categories. (C) Co-expression network of metabolomics data in PWH. Metabolites are grouped by communities, and microbiome-derived metabolites are labeled and colored based on the subgroup. (D) The subset of microbiome-derived metabolites from the co-expression network. Non-significant metabolites in all comparisons are displayed with transparency. Significant microbiome-derived metabolites between at least two conditions are labeled. Figure 5—source data 1.Univariate linear regression between clinical parameters and microbiome-derived metabolites differing groups. ## Discussion In this study, we used network and factorization-based integrative analysis of plasma metabolomics, lipidomics, and microbiome profile to characterize clinical phenotypes in the PWH. We identified three different diseases' state-omics phenotypes (HC-like, mild, and severe at-risk) within PWH driven by metabolomics, lipidomics, and microbiome that a single omics or clinical feature could not explain. The integrative omics highlighted the importance of highly intercorrelated microbiome-derived metabolites and their association with the clinical parameters in PWH cluster separation, shaping their systemic health profile. The severe at-risk group (SNF-2) has the at-risk metabolic profile characterized by an increase in TAG and DAG, highest median BMI, MetS incidence, VAT, and SAT, but had a higher CD4 T-cell count at sample collection compared to HC-like and mild at-risk group, which displayed an HC like lipidomic profile. However, the HC-like and severe at-risk group had a similar metabolic profile differing from HC, with dysregulation of AA metabolism. At the microbiome profile, the HC-like group had a lower α-diversity, a lower proportion of MSM, and was enriched in Bacteroides. In contrast, in at-risk groups, there was an increase in Prevotella, with a high proportion of MSM confirming the influence of sexual orientation on the microbiome profile (Noguera-Julian et al., 2016). Our study thus identified a risk group of PWH with successful treatment with a dysregulated metabolic profile potentiate metabolic diseases that could be barriers to healthy aging. Similarity network analysis reduces the high-dimensional nature and different variances of multi-omics data to group patients based on the most similar profile (Wang et al., 2014). One of the main advantages of this method is the possibility to compare the networks' similarities to find out which layer has the most similarity with the final network. The similarity network fusion-based patient stratification has been used primarily in non-communicable diseases like cancer [to identify cancer subtypes (Wang et al., 2014; Chierici et al., 2020) and prognosis (Wang et al., 2021)], respiratory diseases (Narayana et al., 2021), and to study the influence of diet on human health (Burton-Pimentel et al., 2021). Recently we developed SNF-based patient stratification by integrating transcriptomics and metabolomics to define disease severity in COVID-19 that are predictive of the most robust biological features (Ambikan et al., 2022). We also reported the influence of gut microbiota on the systemic metabolic profile associated with disease severity (Albrich et al., 2022). However, no data were presented to stratify the PWH to fingerprint their disease status. The SNF has shown that the most crucial omics layer in cluster separation was lipids (NMI = 0.6), supported by the MOFA analysis. A study reported that ART and HIV reservoirs are responsible for changes in adipose tissue and lipids metabolism in PWH (Lagathu et al., 2019). Dyslipidemia represents the increase in triglycerides, low-density lipoprotein cholesterol (LDL-C), total cholesterol (TC), and decrease of high-density lipoprotein cholesterol (HDL-C) cholesterol in the blood is a well-recognized complication observed in PWH; both naïve (Wang et al., 2016) and after ART initiation leading to cardiovascular diseases and mortality (Bowman and Funderburg, 2019; Fiseha et al., 2021). We found that the severe at-risk individuals ($\frac{44}{97}$) had most lipids classes upregulated, especially TAG, DAG, and CER, compared to the other groups, while HC-like and mild at-risk groups had no difference with HC. The severe at-risk group also has more patients with high BMI, VAT, SAT, and incidence of MetS. DAG and TAG high levels have been linked to cardiovascular events (Bowman and Funderburg, 2019; Stegemann et al., 2014). The TAG levels have been linked to insulin resistance and increased diabetes risk (Bowman and Funderburg, 2019), confirming this cluster group’s qualification as patients with dysregulated lipid profiles and metabolic disease risk. The association of lipid profiles with CD4 counts is still debated. It is positively associated with (Fiseha et al., 2021; Ji et al., 2019), and negatively (Ombeni and Kamuhabwa, 2016) associated with the highly abundant lipid profile. Interestingly, we found the severe at-risk group to have the highest CD4 count and suppressed viremia but have dysregulated lipid profiles that could be reasoned for unhealthy aging and adverse cardio-metabolic health. Therefore, we propose using a holistic view to define the clinical and immunological treatment success of PWH beyond viral suppression and immune reconstitution. The second omics-defining clusters were metabolites (NMI = 0.4). Interestingly, the metabolic profile did not completely overlap with the lipid profile showing the complexity associated with the disease. PWH in the HC-like group differed most from the HC regarding their HC-like clinical parameter, with the lowest BMI, VAT, and SAT. Nevertheless, $32\%$ of PWH in the HC-like group had MetS, half of the severe at-risk group ($70\%$) but double the mild at-risk group ($17\%$), indicating a possible lipid-independent metabolic dysregulation. Still, the mild at-risk group had the profile of the most HC-like, similar to the lipids, despite having a significantly higher number of patients with hypertension than the HC-like group. The HC-like and severe at-risk groups showed an up-regulation of the metabolites in the xenobiotics, nucleotides, and AA metabolism, indicating a potential role of diet. We previously showed that the glutamate metabolism was highly disrupted in PWH with MetS in the same COCOMO cohort (Gelpi et al., 2021), which can be responsible for late immune recovery in short-term ART patients (Rosado-Sánchez et al., 2019). Also, short-chain dicarboxylacylcarnitines (SCDA) and glutamine/valine were higher in PWH with coronary artery disease than in controls (Okeke et al., 2018). In our cohort, we observed glutamate, N-acetyl-glutamate, phenyl-acetyl-glutamate, gamma-glutamylglutamate, and 4-hydroxyglutamate part of the glutamine/glutamate metabolism had higher abundance in severe at-risk groups than the mild at-risk group. N-acetyl-glutamate was increased in the mild at-risk group compared to the HC-like group. The microbiome network had a modest similarity with the final SNF network (NMI = 0.3), and the PCA plot did not observe apparent clustering of patients. Metabolism and immunity of the host are affected by bacteria and disrupted microbiomes linked to illness (Sun et al., 2016). More importantly, there is a high variability of microbiota among individuals based on lifestyle, diet, medication, and physiology (Knight et al., 2018). Increased α-diversity is associated with good health and decreased diversity in several diseases, including HIV (Zhou et al., 2020). A meta-analysis reported that HIV status was not associated with decreased a-diversity in MSM, perhaps due to sexual behaviors, but was decreased in PWH with heterosexual transmission (Tuddenham et al., 2020). Despite having healthy clinical and metabolic profiles, we observed a higher α-diversity in the severe at-risk group compared to the HC-like group, probably driven by a higher prevalence of MSM. In terms of bacterial composition, early studies reported that PWH had a higher abundance of Prevotella and a lower abundance of Bacteroides (Neff et al., 2018), which in subsequent studies were found to be more related to MSM behaviors than HIV status (Zhou et al., 2020; Gelpi et al., 2020; Noguera-Julian et al., 2016; Vujkovic-Cvijin et al., 2020). Our study observed that the severe at-risk group was enriched in Prevotella and depleted in Bacteriodes compared to the HC-like group. Interestingly, the decrease of Bacteroides in obese patients was inversely correlated with serum glutamate (Wu et al., 2021), which was also observed in severe at-risk group patients. On the other hand, some Prevotella species have pro-inflammatory effects, leading to intestinal inflammation, bacterial translocation, and microbiome dysbiosis (Iljazovic et al., 2021). In general, the complete cohort is mainly composed of MSM ($65\%$, $\frac{63}{97}$). As described above, it confirmed that the difference in the microbiome is driven by MSM status in severe at-risk groups, as there was $81\%$ of MSM in that group. The mild at-risk group, even if there is no difference from the severe at-risk group according to PERMANOVA, has the same proportion of MSM as the HC-like group. It has been proposed that early regulation of the MSM-related microbiome could help prevent HIV infection (Zhou et al., 2020). However, the question remains whether the MSM-related microbiome is a potential driving force of metabolic comorbidities or whether MSM is a confounding factor disturbing a potentially clinical signal from a disturbed microbiome. Moreover, an increase in Prevotella could potentially aggravate intestinal and systemic inflammation leading to an increased cardiometabolic risk profile (Iljazovic et al., 2021; Littlefield et al., 2022). Microbial compositions have implications for metabolism and metabolic diseases, notably through the production of MAMs (Agus et al., 2021). Secondary bile acids transformed from primary bile acids by bacteria have a role in lipid digestion. It regulates host metabolism through signaling and can inhibit the production of pro-inflammatory cytokines by immune cells (Postler and Ghosh, 2017). Lipid metabolism, including triglyceride trafficking, is influenced by bile acids through the interaction with the Farnesoid X receptor (FXR) receptor and has been implicated in mice’s metabolic disorders (Schoeler and Caesar, 2019). A bile acid, glycolithocholate was downregulated in PWH compared to controls previously associated with insulin resistance (Diboun et al., 2021). It was negatively associated with food elements such as vegetable intake and choice of fat for cooking, alcohol, and HIV-related parameters such as CD4 levels (nadir and at ART initiation) and HIV duration. High glycodeoxycholate was observed in the at-risk group compared to controls, while glycodeoxycholic acid is negatively associated with insulin resistance (Wu et al., 2021). Glycocholenate sulfate was downregulated in the three clusters compared to controls. All secondary bile acids were shown to be highly intercorrelated in co-expression analysis. Three other bile acids, lithocholate sulfate, glycousodesoxycholic acid sulfate, and deoxycholic acid 12-sulfate, were negatively associated with metabolic perturbations, including MetS, VAT, and central obesity. Acetate, propionates, and butyrate are part of short-chain fatty acids (SCFAs) and are obtained from the fiber bacterial fermentation in the colon that the host’s enzymes cannot digest (Alwin and Karst, 2021). Proprionate derivates were upregulated in HC-like and severe at-risk groups. Acetate and butyrate derivates had a more variable profile. Imidazole propionate (IMP) and 3-ureidopropionate were linked to the SNF clusters. In our study, the IMP was also linked to vegetable intake, reportedly involved in insulin resistance (Agus et al., 2021). The Bacteroides metabolize most of the acetate and propionate from polysaccharides, and Firmicutes produce butyrate (Postler and Ghosh, 2017), which does not explain the relationship within the SNF clusters indicating a more complex interplay between the MAMs and bacterial community in a diseased condition. Tryptophan is converted by bacterial tryptophanase into indole, and indole derivates are involved in the host-microbiota homeostasis (Krautkramer et al., 2021). Indoles derivates were mainly upregulated in the HC-like and severe at-risk groups. Our data thus suggested the role of MDMs in shaping the clinical phenotype and systemic health profile in PWH, which could be a therapeutic target for improving health. Although our study is the first to demonstrate an integrative multi-omics approach to the role of MAMs in systemic alterations in PWH, our study has limitations that merit comments. First, the study is cross-sectional and therefore restricted to predicting dynamic interactions of different omics layers. Second, the microbiome data analysis was done through 16 S methodologies and has a high level of missing data at the genus and species level. Third, although the network-based analysis and the observational data suggest a potential causal association of altered metabolic profile with clinical features, other factors may drive observed effects. Fourth, although this is the largest study to date to perform integrative omics in PWH, the number of samples was relatively low. Finally, microbiome and metabolomics are highly dependent upon an individual’s genetics, environment, and diet. The interaction noted may characterize the epiphenomena of a personalized immune system that can be an avenue for future studies to develop a more personalized model for integrative omics to phenotype the disease states we recently reported (Ambikan et al., 2022). In conclusion, we performed a multi-omics analysis of PWH with different clinical features. We identified the diversity of PWH in HIV-related biological alterations regardless of immunological recovery and virological suppression. A proportion of PWH (severe at-risk group around $45\%$ in the present cohort) showed highly dysregulated lipidomics (increased TAG and DAG) and clinical profile (increased BMI and obesity-related features) with increased Prevotella and decreased Bacteroides, the latter being related to MSM transmission. However, alterations in the metabolomics profile and higher CD4 T-cell count at the time of sample collection indicate a complex systemic interplay between host immunity and metabolic health. It can lead to an aggravated higher inflammation profile leading to a cardiometabolic risk profile among the MSM that might affect healthy aging in this population. Integrative analytical approaches that reflect the overall systemic health profile of PWH may improve patient stratification and individual therapeutic and preventive strategies. Given the complex interplay between the clinical and molecular metabolic profile, the application of the multi-omics data for much larger cohorts of PWH might facilitate a better identification of network perturbations and molecular network connections to detect early disease transition toward metabolic complications at an earlier stage. Developing a more personalized model or targeting the interaction networks rather than individual clinical or omics features may provide novel treatment strategies in countering dysregulated metabolic traits, aiming to achieve healthier aging. ## Patient cohort and multi-omics data The cohort comprises 97 PWH from the Copenhagen Comorbidity (COCOMO) Cohort, a prospective cohort of PWH. We used untargeted metabolomics (Gelpi et al., 2021), a complex lipid profile (Olund Villumsen et al., 2021), and 16 S rRNA microbiome data (Gelpi et al., 2020) reported earlier for the larger cohorts. We also extracted clinical and demographic data from the COCOMO database. The HIV-negative controls (HC) ($$n = 20$$) were used to understand the basal level of omics. Briefly, untargeted metabolomics, which detects the hydrophilic polar compounds, was performed using the Metabolon HD4 Discovery platform (Metabolon Inc, Morrisville, NC 27560, USA) using ultrahigh-performance liquid chromatography/mass spectrometry/mass spectrometry (UHPLC/MS/MS). Untargeted lipidomic was performed through the Complex Lipid Panel technique (Metabolon Inc, Morrisville, NC 27560, USA). The lipid panel covered lipid panels cover Ceramide (CER), Cholesteryl Esters (CE), Diacylglycerols (DAG), Dihydroceramide (DCER), Hexosylceramide (HCER), Lactosylceramide (LCER), Lysophosphatidylcholine (LPC), Lysophosphatidylethanolamine (LPE), Monoacylglycerol (MAG), Phosphatidylcholine (PC), Phosphatidylethanolamine (PE), Phosphatidylinositol (PI), Sphingomyelin (SM), and Triacylglycerols (TAG). ## Omics-driven PWH stratification using Similarity network fusion (SNF) To stratify the PWH into omics-driven clusters, we used the package SNFtool (Wang et al., 2014). Lipids and metabolites with low variance (<0.3) were removed from the data. The microbiome, lipidome, and metabolome were standard normalized before analysis. Pairwise sample distances were calculated with the function dist2 followed by the construction of similarity graphs (number of neighbors, $K = 13$, hyperparameter, alpha = 0.8) for each layer. The similarity network fusion (SNF) was used to all the networks ($K = 13$, number of iterations, $T = 10$) into one. Spectral clustering was applied to the fused network to determine the optimal number of clusters ($C = 3$). The parameters (K, alpha, T, C) were chosen to maximize the Eigengap and minimize rotation cost. The concordance matrix was calculated based on network similarity and measured in normalized mutual information (NMI). ## Lipidomics and metabolomics analysis Untargeted metabolomics and lipidomics were log2 transformed before analysis. Individual lipid data were grouped by lipid classes as in the following.[Classj]=∑$i = 1$n[speciesi][Classj]=Concentration of the lipid class j[speciesj]=Concentration of the molecular species in=number of molecular species of a class j The differential abundance analysis was performed pairwise with the R package limma between groups (HC, SNF-1, SNF-2, SNF-3) for lipidomics and metabolomics in two models, one with only clusters and one with clusters, and corrected for factors that differ between the clusters. Benjamini-Hochberg (BH) adjustment was applied. ## Microbiome analysis Microbiome data analysis was performed using the R package phyloseq (McMurdie and Holmes, 2013). The alpha diversity estimates were calculated using the estimate_richness function and the following measures: Observed, ACE, se. ACE, Chao1, Shannon, Simpson, InvSimpson, and Fisher. NMDS ordinations based on Bray-Curtis distances between all samples were calculated using the ordinate function. The vegan package (Jari Oksanen et al., 2022) was used to perform PERMANOVA. Equal multivariate dispersion was verified using the betadisper function applying Marti Anderson’s PERMDISP2 procedure. Pairwise PERMANOVA test was done between groups using the adonis function, Bray distance, and Bonferroni correction. The cutoff for the adjusted p-value was set up to 0.05. Galaxy module LDA Effect Size (LEfSe) was used to find microbial communities (at genus, family, or higher level) specific to one specific cluster (Segata et al., 2011). The multiclass analysis approach was one against all. First, a non-parametric factorial Kruskal-Wallis (KW) sum-rank test was performed with clusters (cutoff alpha = 0.05), followed by pairwise Wilcoxon rank-sum tests between clusters (cutoff alpha = 0.05), and then effect size calculation for each significant feature was done using discriminant analysis (absolute LDA score >2). Results are represented using a cladogram produced by the module. ## Microbiome-associated metabolites Microbiome-associated metabolites (MAM), groups, and subgroups were retrieved from the previous literature (Postler and Ghosh, 2017) to determine the impact of the microbiome on the metabolism. Univariate linear regression was performed with the function lm between microbiome-derived metabolites and clinical parameters to see the influence of lifestyle on these metabolites. ## Multi-omics factor analysis (MOFA) MOFA was used to determine the weight of each data type and individual features in PWH. Filtered data for SNF was also used for MOFA analysis (Argelaguet et al., 2018). Microbiome data were rarefied by filtering based on variance (>0.2). In addition, the microbiome data were center log-ratio (CLR) transformed to follow a normal distribution. The MOFA model was trained using default parameters, and sample metadata was added to the model. The total variance explained per view was used to see the weight of each omics layer. A correlation plot was used to verify the low correlation between factors. A variance decomposition plot was used to determine the percentage of variance explained by each factor and omics layer. Association analysis of the factors with clinical features was done using the MOFA function correlate_factors_with_covariates and factors associated with the SNF cluster selected. Five and $95\%$ quantile weights for each view were selected for each factor. Pathway analysis was performed on factors using the MOFA function run_enrichment for each view, with the parametric statistical test, FDR-adjusted p-values, and separated positive and negative values. Annotation libraries were made from Metabolon super pathways for metabolomics and lipidomics and Division level for the microbiome. ## Co-expression analysis We used co-expression analysis to measure the interactions between all features in the data. Pairwise Spearman correlations between features were calculated using the R package stat, and the cutoff for FDR of significant correlations was selected to minimize the number of false positives. The positive and negative networks were built using the python igraph (Csárdi and Nepusz, 2005) and compared to random networks of the same size. Leiden community detection was applied to find groups of interconnected features, and the mean degree was calculated to represent the community centrality using the python module leidenag (Blondel et al., 2008). Communities of less than 30 features were excluded. Consensus association analysis was performed to integrate the three layers of omics using 1000 iterations. At each iteration, pairwise correlations between ASVs ($$n = 241$$), 241 metabolites, and 241 lipids selected randomly were run, and significant positive correlations (Spearman, FDR <0.001) were kept as an association. Associations found in $90\%$ of the comparisons over all iterations were kept building the final network as described above. ## General statistics Differences between clusters in clinical parameters were measured using Kruskal–Wallis H test for continuous variables and Chi-Square Test or Fisher’s Exact Test for discrete variables. Deviations were mentioned in all respective analyses. The default p-value cutoff was set to 0.05. Other p-values cutoffs are adapted for a specific analysis depending upon the number of significance and to minimize the false positivity (Team TRDC, 2010). ## Visualization Scatter plots, PCA plots, box plots, NMDS plots, circular heatmap, and bar plots were generated using ggplot2 (Wickham, 2016). Heatmaps were generated using ComplexHeatmap (Gu et al., 2016). Sankey plot was made using the R package ggalluvial (Brunson, 2020). Networks were plotted using Cytoscape v3.6.1 (Shannon et al., 2003). ## Funding Information This paper was supported by the following grants: ## Data availability All of the data generated or analyzed during this study are included in this published article and/or the supplementary materials. Created datasets and code are publicly available. The metabolomics and lipidomics data are available from https://doi.org/10.6084/m9.figshare.14356754.v1 and https://doi.org/10.6084/m9.figshare.14509452.v1. All the codes are available at github: https://github.com/neogilab/HIV_multiomics, (copy archived at swh:1:rev:86aae862497b7dbb3dae4ce2e5a44b0369e0dec0). The following datasets were generated: NeogiU NielsenSD figshare2022Original Scale Metabolomics data: COCOMO10.6084/m9.figshare.14356754.v1 NeogiU NielsenSD figshare2022Original Scale Data: COCOMO_Lipidomics10.6084/m9.figshare.14509452.v1 ## References 1. Agus A, Clément K, Sokol H. Gut microbiota-derived metabolites as central regulators in metabolic disorders. Gut (2021) 70 1174-1182. DOI: 10.1136/gutjnl-2020-323071 2. Albrich WC, Ghosh TS, Ahearn-Ford S, Mikaeloff F, Lunjani N, Forde B, Suh N, Kleger G-R, Pietsch U, Frischknecht M, Garzoni C, Forlenza R, Horgan M, Sadlier C, Negro TR, Pugin J, Wozniak H, Cerny A, Neogi U, O’Toole PW, O’Mahony L. A high-risk gut microbiota configuration associates with fatal hyperinflammatory immune and metabolic responses to SARS-cov-2. Gut Microbes (2022) 14. DOI: 10.1080/19490976.2022.2073131 3. Alwin A, Karst SM. The influence of microbiota-derived metabolites on viral infections. Current Opinion in Virology (2021) 49 151-156. DOI: 10.1016/j.coviro.2021.05.006 4. Ambikan AT, Yang H, Krishnan S, Svensson-Akusjärvi S, Gupta S, Lourda M, Sperk M, Arif M, Zhang C, Nordqvist H, Ponnan SM, Sönnerborg A, Treutiger CJ, O’Mahony L, Mardinoglu A, Benfeitas R, Neogi U. Multiomics personalized network analyses highlight progressive immune disruption of central metabolism associated with COVID-19 severity. SSRN Electronic Journal (2022) 1. DOI: 10.2139/ssrn.3988390 5. Argelaguet R, Velten B, Arnol D, Dietrich S, Zenz T, Marioni JC, Buettner F, Huber W, Stegle O. Multi-Omics factor analysis-a framework for unsupervised integration of multi-omics data sets. Molecular Systems Biology (2018) 14. DOI: 10.15252/msb.20178124 6. Babu H, Sperk M, Ambikan AT, Rachel G, Viswanathan VK, Tripathy SP, Nowak P, Hanna LE, Neogi U. Plasma metabolic signature and abnormalities in HIV-infected individuals on long-term successful antiretroviral therapy. Metabolites (2019) 9. DOI: 10.3390/metabo9100210 7. Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E. Fast unfolding of communities in large networks. Journal of Statistical Mechanics (2008) 2008. DOI: 10.1088/1742-5468/2008/10/P10008 8. Bowman E, Funderburg NT. Lipidome abnormalities and cardiovascular disease risk in HIV infection. Current HIV/AIDS Reports (2019) 16 214-223. DOI: 10.1007/s11904-019-00442-9 9. Brunson J. Ggalluvial: layered grammar for alluvial plots. Journal of Open Source Software (2020) 5. DOI: 10.21105/joss.02017 10. Burton-Pimentel KJ, Pimentel G, Hughes M, Michielsen CC, Fatima A, Vionnet N, Afman LA, Roche HM, Brennan L, Ibberson M, Vergères G. Discriminating dietary responses by combining transcriptomics and metabolomics data in nutrition intervention studies. Molecular Nutrition & Food Research (2021) 65. DOI: 10.1002/mnfr.202000647 11. Chierici M, Bussola N, Marcolini A, Francescatto M, Zandonà A, Trastulla L, Agostinelli C, Jurman G, Furlanello C. Integrative network fusion: a multi-omics approach in molecular profiling. Frontiers in Oncology (2020) 10. DOI: 10.3389/fonc.2020.01065 12. Csárdi G, Nepusz T. The igraph software package for complex network research. InterJournal (2005) 1695 1-9 13. Deeks SG. Hiv infection, inflammation, immunosenescence, and aging. Annual Review of Medicine (2011) 62 141-155. DOI: 10.1146/annurev-med-042909-093756 14. Diboun I, Al-Mansoori L, Al-Jaber H, Albagha O, Elrayess MA. Metabolomics of lean/overweight insulin-resistant females reveals alterations in steroids and fatty acids. The Journal of Clinical Endocrinology and Metabolism (2021) 106 e638-e649. DOI: 10.1210/clinem/dgaa732 15. Fiseha T, Alemu W, Dereje H, Tamir Z, Gebreweld A. Prevalence of dyslipidaemia among HIV-infected patients receiving combination antiretroviral therapy in North shewa, Ethiopia. PLOS ONE (2021) 16. DOI: 10.1371/journal.pone.0250328 16. Gelpi M, Afzal S, Lundgren J, Ronit A, Roen A, Mocroft A, Gerstoft J, Lebech A-M, Lindegaard B, Kofoed KF, Nordestgaard BG, Nielsen SD. Higher risk of abdominal obesity, elevated low-density lipoprotein cholesterol, and hypertriglyceridemia, but not of hypertension, in people living with human immunodeficiency virus (HIV): results from the copenhagen comorbidity in HIV infection study. Clinical Infectious Diseases (2018) 67 579-586. DOI: 10.1093/cid/ciy146 17. Gelpi M, Vestad B, Hansen SH, Holm K, Drivsholm N, Goetz A, Kirkby NS, Lindegaard B, Lebech A-M, Hoel H, Michelsen AE, Ueland T, Gerstoft J, Lundgren J, Hov JR, Nielsen SD, Trøseid M. Impact of human immunodeficiency virus-related gut microbiota alterations on metabolic comorbid conditions. Clinical Infectious Diseases (2020) 71 e359-e367. DOI: 10.1093/cid/ciz1235 18. Gelpi M, Mikaeloff F, Knudsen AD, Benfeitas R, Krishnan S, Svenssson Akusjärvi S, Høgh J, Murray DD, Ullum H, Neogi U, Nielsen SD. The central role of the glutamate metabolism in long-term antiretroviral treated HIV-infected individuals with metabolic syndrome. Aging (2021) 13 22732-22751. DOI: 10.18632/aging.203622 19. Gu Z, Eils R, Schlesner M. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics (2016) 32 2847-2849. DOI: 10.1093/bioinformatics/btw313 20. Iljazovic A, Roy U, Gálvez EJC, Lesker TR, Zhao B, Gronow A, Amend L, Will SE, Hofmann JD, Pils MC, Schmidt-Hohagen K, Neumann-Schaal M, Strowig T. Perturbation of the gut microbiome by Prevotella spp. enhances host susceptibility to mucosal inflammation. Mucosal Immunology (2021) 14 113-124. DOI: 10.1038/s41385-020-0296-4 21. Jari Oksanen FGB, Friendly M, Kindt R, Pierre Legendre DM, Minchin PR, O’Hara RB, Gavin LSP, Stevens MHH. R Package (2022) 22. Ji S, Xu Y, Han D, Peng X, Lu X, Brockmeyer NH, Wu N. Changes in lipid indices in HIV+ cases on HAART. BioMed Research International (2019) 2019. DOI: 10.1155/2019/2870647 23. Karahalil B. Overview of systems biology and omics technologies. Current Medicinal Chemistry (2016) 23 4221-4230. DOI: 10.2174/0929867323666160926150617 24. Knight R, Vrbanac A, Taylor BC, Aksenov A, Callewaert C, Debelius J, Gonzalez A, Kosciolek T, McCall L-I, McDonald D, Melnik AV, Morton JT, Navas J, Quinn RA, Sanders JG, Swafford AD, Thompson LR, Tripathi A, Xu ZZ, Zaneveld JR, Zhu Q, Caporaso JG, Dorrestein PC. Best practices for analysing microbiomes. Nature Reviews. Microbiology (2018) 16 410-422. DOI: 10.1038/s41579-018-0029-9 25. Krautkramer KA, Fan J, Bäckhed F. Gut microbial metabolites as multi-kingdom intermediates. Nature Reviews. Microbiology (2021) 19 77-94. DOI: 10.1038/s41579-020-0438-4 26. Lagathu C, Béréziat V, Gorwood J, Fellahi S, Bastard J-P, Vigouroux C, Boccara F, Capeau J. Metabolic complications affecting adipose tissue, lipid and glucose metabolism associated with HIV antiretroviral treatment. Expert Opinion on Drug Safety (2019) 18 829-840. DOI: 10.1080/14740338.2019.1644317 27. Littlefield KM, Schneider JM, Neff CP, Soesanto V, Siebert JC, Nusbacher NM, Moreno-Huizar N, Cartwright IM, Armstrong AJS, Colgen SP, Lozupone CA, Palmer BE. Elevated inflammatory fecal immune factors in men who have sex with men with HIV associate with microbiome composition and gut barrier function. Frontiers in Immunology (2022) 13. DOI: 10.3389/fimmu.2022.1072720 28. McMurdie PJ, Holmes S. Phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLOS ONE (2013) 8. DOI: 10.1371/journal.pone.0061217 29. Mikaeloff F, Svensson Akusjärvi S, Ikomey GM, Krishnan S, Sperk M, Gupta S, Magdaleno GDV, Escós A, Lyonga E, Okomo MC, Tagne CT, Babu H, Lorson CL, Végvári Á, Banerjea AC, Kele J, Hanna LE, Singh K, de Magalhães JP, Benfeitas R, Neogi U. Trans cohort metabolic reprogramming towards glutaminolysis in long-term successfully treated HIV-infection. Communications Biology (2022) 5. DOI: 10.1038/s42003-021-02985-3 30. Narayana JK, Mac Aogáin M, Ali NABM, Tsaneva-Atanasova K, Chotirmall SH. Similarity network fusion for the integration of multi-omics and microbiomes in respiratory disease. The European Respiratory Journal (2021) 58. DOI: 10.1183/13993003.01016-2021 31. Nasi M, De Biasi S, Gibellini L, Bianchini E, Pecorini S, Bacca V, Guaraldi G, Mussini C, Pinti M, Cossarizza A. Ageing and inflammation in patients with HIV infection. Clinical and Experimental Immunology (2017) 187 44-52. DOI: 10.1111/cei.12814 32. Neff CP, Krueger O, Xiong K, Arif S, Nusbacher N, Schneider JM, Cunningham AW, Armstrong A, Li S, McCarter MD, Campbell TB, Lozupone CA, Palmer BE. Fecal microbiota composition drives immune activation in HIV-infected individuals. EBioMedicine (2018) 30 192-202. DOI: 10.1016/j.ebiom.2018.03.024 33. Noguera-Julian M, Rocafort M, Guillén Y, Rivera J, Casadellà M, Nowak P, Hildebrand F, Zeller G, Parera M, Bellido R, Rodríguez C, Carrillo J, Mothe B, Coll J, Bravo I, Estany C, Herrero C, Saz J, Sirera G, Torrela A, Navarro J, Crespo M, Brander C, Negredo E, Blanco J, Guarner F, Calle ML, Bork P, Sönnerborg A, Clotet B, Paredes R. Gut microbiota linked to sexual preference and HIV infection. EBioMedicine (2016) 5 135-146. DOI: 10.1016/j.ebiom.2016.01.032 34. Okeke NL, Craig DM, Muehlbauer MJ, Ilkayeva O, Clement ME, Naggie S, Shah SH. Metabolites predict cardiovascular disease events in persons living with HIV: a pilot case-control study. Metabolomics (2018) 14. DOI: 10.1007/s11306-018-1318-z 35. Olund Villumsen S, Benfeitas R, Knudsen AD, Gelpi M, Høgh J, Thomsen MT, Murray D, Ullum H, Neogi U, Nielsen SD. Integrative lipidomics and metabolomics for system-level understanding of the metabolic syndrome in long-term treated HIV-infected individuals. Frontiers in Immunology (2021) 12. DOI: 10.3389/fimmu.2021.742736 36. Ombeni W, Kamuhabwa AR. Lipid profile in HIV-infected patients using first-line antiretroviral drugs. Journal of the International Association of Providers of AIDS Care (2016) 15 164-171. DOI: 10.1177/2325957415614642 37. Pei L, Fukutani KF, Tibúrcio R, Rupert A, Dahlstrom EW, Galindo F, Laidlaw E, Lisco A, Manion M, Andrade BB, Sereti I. Plasma metabolomics reveals dysregulated metabolic signatures in HIV-associated immune reconstitution inflammatory syndrome. Frontiers in Immunology (2021) 12. DOI: 10.3389/fimmu.2021.693074 38. Pelchen-Matthews A, Ryom L, Borges ÁH, Edwards S, Duvivier C, Stephan C, Sambatakou H, Maciejewska K, Portu JJ, Weber J, Degen O, Calmy A, Reikvam DH, Jevtovic D, Wiese L, Smidt J, Smiatacz T, Hassoun G, Kuznetsova A, Clotet B, Lundgren J, Mocroft A. Aging and the evolution of comorbidities among HIV-positive individuals in a european cohort. AIDS (2018) 32 2405-2416. DOI: 10.1097/QAD.0000000000001967 39. Postler TS, Ghosh S. Understanding the holobiont: how microbial metabolites affect human health and shape the immune system. Cell Metabolism (2017) 26 110-130. DOI: 10.1016/j.cmet.2017.05.008 40. Rosado-Sánchez I, Rodríguez-Gallego E, Peraire J, Viladés C, Herrero P, Fanjul F, Gutiérrez F, Bernal E, Pelazas R, Leal M, Veloso S, López-Dupla M, Blanco J, Vidal F, Pacheco YM, Rull A. Glutaminolysis and lipoproteins are key factors in late immune recovery in successfully treated HIV-infected patients. Clinical Science (2019) 133 997-1010. DOI: 10.1042/CS20190111 41. Schoeler M, Caesar R. Dietary lipids, gut microbiota and lipid metabolism. Reviews in Endocrine & Metabolic Disorders (2019) 20 461-472. DOI: 10.1007/s11154-019-09512-0 42. Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS, Huttenhower C. Metagenomic biomarker discovery and explanation. Genome Biology (2011) 12. DOI: 10.1186/gb-2011-12-6-r60 43. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N, Schwikowski B, Ideker T. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Research (2003) 13 2498-2504. DOI: 10.1101/gr.1239303 44. Stegemann C, Pechlaner R, Willeit P, Langley SR, Mangino M, Mayr U, Menni C, Moayyeri A, Santer P, Rungger G, Spector TD, Willeit J, Kiechl S, Mayr M. Lipidomics profiling and risk of cardiovascular disease in the prospective population-based bruneck study. Circulation (2014) 129 1821-1831. DOI: 10.1161/CIRCULATIONAHA.113.002500 45. Sun Y, Ma Y, Lin P, Tang Y-W, Yang L, Shen Y, Zhang R, Liu L, Cheng J, Shao J, Qi T, Tang Y, Cai R, Guan L, Luo B, Sun M, Li B, Pei Z, Lu H. Fecal bacterial microbiome diversity in chronic HIV-infected patients in China. Emerging Microbes & Infections (2016) 5. DOI: 10.1038/emi.2016.25 46. Team TRDC 2010R: A Language and Environment for Statistical Computing4.1.2 edVienna, AustriaR Foundation for Statistical Computing. R: A Language and Environment for Statistical Computing (2010) 47. Tuddenham SA, Koay WLA, Zhao N, White JR, Ghanem KG, Sears CL. The impact of human immunodeficiency virus infection on gut microbiota α-diversity: an individual-level meta-analysis. Clinical Infectious Diseases (2020) 70 615-627. DOI: 10.1093/cid/ciz258 48. Vujkovic-Cvijin I, Sortino O, Verheij E, Sklar J, Wit FW, Kootstra NA, Sellers B, Brenchley JM, Ananworanich J, van der Loeff MS, Belkaid Y, Reiss P, Sereti I. Hiv-associated gut dysbiosis is independent of sexual practice and correlates with noncommunicable diseases. Nature Communications (2020) 11. DOI: 10.1038/s41467-020-16222-8 49. Wang B, Mezlini AM, Demir F, Fiume M, Tu Z, Brudno M, Haibe-Kains B, Goldenberg A. Similarity network fusion for aggregating data types on a genomic scale. Nature Methods (2014) 11 333-337. DOI: 10.1038/nmeth.2810 50. Wang Q, Ding H, Xu J, Geng W, Liu J, Guo X, Kang J, Li X, Jiang Y, Shang H. Lipids profile among ART-naïve HIV infected patients and men who have sex with men in China: a case control study. Lipids in Health and Disease (2016) 15. DOI: 10.1186/s12944-016-0297-1 51. Wang T-H, Lee C-Y, Lee T-Y, Huang H-D, Hsu JB-K, Chang T-H. Biomarker identification through multiomics data analysis of prostate cancer prognostication using a deep learning model and similarity network fusion. Cancers (2021) 13. DOI: 10.3390/cancers13112528 52. Wickham H. Ggplot2: Elegant Graphics for Data Analysis (2016). DOI: 10.1007/978-3-319-24277-4 53. Wu J, Wang K, Wang X, Pang Y, Jiang C. The role of the gut microbiome and its metabolites in metabolic diseases. Protein & Cell (2021) 12 360-373. DOI: 10.1007/s13238-020-00814-7 54. Yoshimura K. Current status of HIV/AIDS in the art era. Journal of Infection and Chemotherapy (2017) 23 12-16. DOI: 10.1016/j.jiac.2016.10.002 55. Zhou J, Zhang Y, Cui P, Luo L, Chen H, Liang B, Jiang J, Ning C, Tian L, Zhong X, Ye L, Liang H, Huang J. Gut microbiome changes associated with HIV infection and sexual orientation. Frontiers in Cellular and Infection Microbiology (2020) 10. DOI: 10.3389/fcimb.2020.00434
--- title: Diet-induced loss of adipose hexokinase 2 correlates with hyperglycemia authors: - Mitsugu Shimobayashi - Amandine Thomas - Sunil Shetty - Irina C Frei - Bettina K Wölnerhanssen - Diana Weissenberger - Anke Vandekeere - Mélanie Planque - Nikolaus Dietz - Danilo Ritz - Anne Christin Meyer-Gerspach - Timm Maier - Nissim Hay - Ralph Peterli - Sarah-Maria Fendt - Nicolas Rohner - Michael N Hall journal: eLife year: 2023 pmcid: PMC10017106 doi: 10.7554/eLife.85103 license: CC BY 4.0 --- # Diet-induced loss of adipose hexokinase 2 correlates with hyperglycemia ## Abstract Chronically high blood glucose (hyperglycemia) leads to diabetes and fatty liver disease. Obesity is a major risk factor for hyperglycemia, but the underlying mechanism is unknown. Here, we show that a high-fat diet (HFD) in mice causes early loss of expression of the glycolytic enzyme Hexokinase 2 (HK2) specifically in adipose tissue. Adipose-specific knockout of Hk2 reduced glucose disposal and lipogenesis and enhanced fatty acid release in adipose tissue. In a non-cell-autonomous manner, Hk2 knockout also promoted glucose production in liver. Furthermore, we observed reduced hexokinase activity in adipose tissue of obese and diabetic patients, and identified a loss-of-function mutation in the hk2 gene of naturally hyperglycemic Mexican cavefish. Mechanistically, HFD in mice led to loss of HK2 by inhibiting translation of Hk2 mRNA. Our findings identify adipose HK2 as a critical mediator of local and systemic glucose homeostasis, and suggest that obesity-induced loss of adipose HK2 is an evolutionarily conserved mechanism for the development of selective insulin resistance and thereby hyperglycemia. ## Introduction Vertebrates mediate glucose homeostasis by regulating glucose production and disposal in specific tissues (Roden and Shulman, 2019; Wasserman, 2009). High blood glucose stimulates pancreatic beta cells to secrete the hormone insulin which in turn promotes glucose disposal in skeletal muscle and adipose tissue and inhibits glucose production in liver. Although its contribution to glucose clearance is minor (Jackson et al., 1986; Kowalski and Bruce, 2014), adipose tissue plays a particularly important role in systemic glucose homeostasis (Abel et al., 2001; Shepherd et al., 1993). Adipose-specific knockout of insulin signaling components, such as the insulin receptor, mTORC2, and AKT, results in local and systemic insulin insensitivity (Beg et al., 2017; Cybulski et al., 2009; Frei et al., 2022; Jiang et al., 2003; Kumar et al., 2010; Sakaguchi et al., 2017; Shearin et al., 2016; Tang et al., 2016). However, we and others have reported that diet-induced obesity in mice causes adipose dysfunction, systemic insulin insensitivity, and hyperglycemia despite normal insulin signaling in white adipose tissue (WAT; Figure 1—figure supplement 1A; Shimobayashi et al., 2018; Tan et al., 2015). How does obesity cause systemic insulin insensitivity and hyperglycemia? In other words, how does diet induce hyperglycemia? Here, we show that a high-fat diet induces early loss of the glycolytic enzyme HK2 specifically in adipose tissue. Loss of adipose HK2 leads to reduced glucose disposal by adipose tissue and increased glucose production in liver, ultimately causing glucose intolerance. Loss of adipose HK2 also decreased lipogenesis and increased fatty acid release in WAT. This and related findings in Mexican cavefish and adipose tissue of obese patients suggest that diet-induced downregulation of adipose HK2 contributes to hyperglycemia. ## HK2 is down-regulated in obese mice and humans To determine the molecular basis of diet-induced hyperglycemia in mice, we performed an unbiased proteomic analysis on visceral white adipose tissue (vWAT) isolated from C57BL/6JRj wild-type mice fed a HFD for 4 weeks or normal diet (ND) (Figure 1—figure supplement 1B–D). We detected and quantified 6294 proteins of which 52 and 67 were up- and down-regulated, respectively, in vWAT of HFD-fed mice (Supplementary file 1). The glycolytic enzyme Hexokinase 2 (HK2), expressed in adipose tissue and muscle, was among the proteins significantly down-regulated in vWAT (Figure 1A). We focused on HK2 due to its role in glucose metabolism. HK2 phosphorylates glucose to generate glucose-6-phosphate (G6P), the rate-limiting step in glycolysis, in an insulin-stimulated manner. HK2 is the most abundant (~$80\%$) of the three hexokinase isoforms expressed in vWAT but the only one down-regulated upon HFD (Figure 1A and Figure 1—figure supplement 1E). Quantification by immunoblotting revealed an approximately $60\%$, $90\%$ and $50\%$ reduction in HK2 expression in vWAT, subcutaneous WAT (sWAT), and brown adipose tissue (BAT), respectively, of 4 week HFD mice (Figure 1B–C). Consistent with reduced HK2 expression, hexokinase activity was decreased in WAT of HFD-fed mice (Figure 1D). HK2 expression was also decreased in WAT of ND-fed ob/ob mice, compared to littermate controls (Figure 1—figure supplement 1F), suggesting that loss of HK2 is common to different obesogenic conditions. A longitudinal study of HFD-fed mice revealed that HK2 down-regulation in adipose tissue occurred within one week of HFD and correlated with systemic insulin insensitivity (Figure 1—figure supplement 2A–F). HK2 expression was unchanged in skeletal muscle (Figure 1B). Expression of the glucose transporter GLUT4 was slightly reduced in vWAT at 4 weeks of HFD (Figure 1A and Figure 1—figure supplement 2D). In sWAT, GLUT4 was down-regulated at >1 week of HFD (Figure 1—figure supplement 2E). Confirming earlier observations (Shimobayashi et al., 2018; Tan et al., 2015), insulin signaling was normal in WAT at 4 weeks of HFD (Figure 1—figure supplements 1A and 2D–E). However, in BAT, insulin signaling was significantly downregulated within 1 week of HFD (Figure 1—figure supplement 2F). Shifting from HFD to 2 weeks of ND restored HK2 expression and normal blood glucose (Figure 1E–F and Figure 1—figure supplement 2C). Thus, loss of HK2 expression upon obesogenic conditions is an adipose-specific, transient physiological response that correlates with hyperglycemia. Figure 1.: Loss of HK2 in obese mouse and obese human.(A) The Log2 fold change (FC) of Hexokinase and GLUT4 protein expression in visceral white adipose tissue (vWAT) of normal diet (ND)- and 4 week high-fat diet (HFD)-fed wild-type C57BL/6JRj mice. Multiple t test, q<0.0001. n=5 (ND) and 5 (HFD). (B) Immunoblot analyses of vWAT, subcutaneous WAT (sWAT), brown adipose tissue (BAT), and skeletal muscle from ND- and 4 week HFD-fed mice. CALX serves as a loading control. n=6 (ND) and 6 (HFD). (C) Quantification of panel B. Data is normalized to the loading control. Student’s t test. p<0.01. (D) Hexokinase (HK) activity of vWAT and sWAT from ND- and 4 week HFD-fed mice. Student’s t test, p<0.05, *p<0.01. n=5 (ND) and 5 (HFD). (E–F) Immunoblot analyses of vWAT (E) and sWAT (F) from ND-, 2 week HFD-, and 2 week HFD +2 week ND-fed mice. n=5 (ND), 5 (HFD), and 6 (HFD +ND). Data is normalized to the loading control. One-way ANOVA. p<0.01, p<0.01, *p<0.0001. (G) Hexokinase (HK) activity of vWAT from lean, obese non-diabetic, and obese diabetic patients. Two-way ANOVA, p<0.05. n=27 (lean), 30 (obese), and 14 (obese diabetic). (H) Comparison of vWAT HK activity from low or high HOMA-IR obese non-diabetic patients. Student’s t tests, p<0.05. n=12 (low, HOMA-IR <2.9) and 18 (high, HOMA-IR >2.9). (I) Pearson’s correlation analyses of hexokinase activity and homeostatic assessment for insulin resistance (HOMA-IR) in obese patients. See Figure 1—source data 1. Figure 1—source data 1.Uncropped blots and source data for graphs for Figure 1. Similar to our findings in mice, omental WAT (human vWAT) biopsies from obese non-diabetic and obese diabetic patients displayed a~$30\%$ reduction in hexokinase activity (Figure 1G and Supplementary file 2). Importantly, hexokinase activity was particularly low in obese non-diabetic patients with severe insulin resistance, compared to patients with mild insulin resistance (Figure 1H). Although hexokinase activity negatively correlated with insulin resistance in obese patients, this correlation was not statistically significant (Figure 1I), suggesting that loss of hexokinase activity may not be the only cause of insulin resistance in human. We note that Ducluzeau et al. observed a decrease in HK2 mRNA expression in adipose tissue of diabetic patients (Ducluzeau et al., 2001). Altogether, the above findings suggest that HK2 down-regulation in adipose tissue is a key event, possibly causal, in obesity-induced insulin insensitivity and hyperglycemia in mouse and human. ## A loss-of-function hk2 mutation in hyperglycemic cavefish Mexican cavefish (Astyanax mexicanus), also known as blind fish, are hyperglycemic compared to surface fish from which they are descended (Riddle et al., 2018; Figure 2A–B). Hyperglycemia, although a pathological condition in mouse and human, is a selected trait that presumably allows cave-dwelling fish to survive in nutrient-limited conditions. Among three independently evolved cavefish isolates, Pachón and Tinaja cavefish (names refer to the caves from which the fish were isolated) acquired a loss-of-function mutation in the insulin receptor gene, causing insulin resistance and hyperglycemia (Riddle et al., 2018; Figure 2B). The third isolate, Molino cavefish, is the most hyperglycemic but contains a wild-type insulin receptor gene and displays normal insulin signaling (Riddle et al., 2018). Since the phenotype of Molino fish is similar to that of HFD-fed mice (normal insulin signaling yet hyperglycemic), we hypothesized that this cavefish may be hyperglycemic due to a loss-of-function mutation in the hk2 gene. DNA sequencing of the hk2 gene of surface fish and the three cavefish variants revealed a mutation in the hk2 gene uniquely in Molino. The homozygous, missense mutation in the coding region of the Molino hk2 gene changed highly conserved arginine 42 (R42) to histidine (R42H) (Figure 2C and D). Based on the published structure of HK2 (Nawaz et al., 2018), R42 forms a salt bridge with aspartic acid 272 (D272) to stabilize the conformation of HK2 (Figure 2E), predicting that R42H destabilizes HK2 and is thus a loss-of-function mutation. To test this prediction, we expressed surface fish and Molino HK2 in HEK293T cells which have low intrinsic hexokinase activity. Indeed, Molino HK2 displayed little-to-no hexokinase activity compared to surface fish HK2 (Figure 2F). To test whether R42H causes loss of function in mammalian HK2, we examined hexokinase activity of recombinant mammalian HK2 containing the Molino mutation (HK2-R42H). HK2-R42H displayed ~$50\%$ hexokinase activity compared to HK2-WT, despite similar expression levels (Figure 2G). Thus, the R42H mutation may account for the hyperglycemia in Molino cavefish. In other words, R42H in Molino appears to be a genetically fixed version of what we observe in mice as a physiological down-regulation of HK2 in response to HFD. Although further study is required to link the R42H mutation to the hyperglycemic phenotype in Molino cavefish, the findings in Molino provide orthogonal evidence that the down-regulation of HK2 in mice is physiologically relevant in hyperglycemia. Figure 2.: A loss-of-function HK2 variant in hyperglycemic Mexican cavefish.(A) Surface fish and Mexican cavefish Molino. (B) Comparisons of phenotypes in surface fish, Pachón, Tinaja, and Molino. (C) DNA sequence of the Molino variant. (D) Amino acid sequence alignment of the HK2-R42H mutation within vertebrates. (E) Structural analyses revealed the presence of a salt bridge between Arginine 42 (R42) and Aspartic acid 272 (D272) in the human HK2 (PDB: 2MTZ). (F) HK activity and immunoblot analyses for lysates of HEK293T cells expressing surface or Molino HK2. Student’s t test, *p<0.0001. N=4. (G) HK activity and immunoblots for lysates of HEK293T cells expressing control, HK2-WT, or HK2-R42H. Student’s t test, *p<0.0001. N=4. See Figure 2—source data 1. Figure 2—source data 1.Uncropped blots and source data for graphs for Figure 2. ## Adipose-specific Hk2 knockout causes hyperglycemia The above findings altogether suggest that adipose-specific loss of HK2 may be a cause of hyperglycemia. To test this hypothesis, we first generated a stable HK2 knockdown pre-adipocyte 3T3-L1 cell line (Figure 3A–B). HK2-knockdown pre-adipocytes differentiated normally to produce mature adipocytes (Figure 3—figure supplement 1A–B). The glycolytic rate was lower in HK2-knockdown adipocytes compared to controls, as measured by extracellular acidification rate (ECAR) (Figure 3C) and lactate production (Figure 3D). Furthermore, although basal glucose accumulation did not differ, insulin-stimulated glucose accumulation was $50\%$ lower in HK2-knockdown adipocytes (Figure 3E), despite normal insulin signaling (Figure 3A). The observation that HK2-knockdown has no effect on basal glucose accumulation is due to the fact that HK2 is inactive in the absence of insulin (Gottlob et al., 2001; Miyamoto et al., 2008). Thus, loss of HK2 decreases glucose disposal in insulin-stimulated adipocytes in vitro. Figure 3.: Loss of adipose HK2 causes reduced glucose disposal in adipocytes.(A) Immunoblot analyses of control and HK2 knockdown 3T3-L1 adipocyte lysates. Cells were stimulated with 100 nM insulin for 25 min. N=4. (B) HK activity of control and HK2 knockdown 3T3-L1 adipocytes. Student’s t test, *p<0.001. N=3. (C) Extracellular acidification rate of control and HK2 knockdown 3T3-L1 adipocytes in response to glucose (10 mM) and 2-deoxyglucose (2DG, 50 mM). Two-way ANOVA. p<0.05. N=2. (D) Lactate secreted into media by control and HK2 knockdown 3T3-L1 adipocytes treated with or without 100 nM insulin. One-way ANOVA. p<0.05, p<0.01. N=3. (E) 2DG uptake in control and HK2 knockdown 3T3-L1 adipocytes treated with or without 100 nM insulin. One-way ANOVA. **p<0.0001. N=3. See Figure 3—source data 1. Figure 3—source data 1.Uncropped blots and source data for graphs for Figure 3. To examine further the role of adipose HK2 in glucose homeostasis, and in particular the causality of HK2 loss in insulin insensitivity and hyperglycemia, we generated an adipose-specific Hk2 knockout (AdHk2KO) mouse in which the knockout was induced at 4–5 weeks of age (Figure 4—figure supplement 1A). In AdHk2KO mice, HK2 expression was decreased ~$75\%$ in vWAT,~$70\%$ in sWAT, and ~$65\%$ in BAT but unchanged in skeletal muscle (Figure 4A and Figure 4—figure supplement 1B–D). Glycolytic metabolites were also decreased in WAT (Figure 4B). Furthermore, insulin signaling was normal in adipose tissue (Figure 4A and Figure 4—figure supplement 1B–C) and other tissues (see below) of AdHk2KO mice. ND-fed AdHk2KO mice displayed slightly less fat mass and slightly more lean mass than controls, but no difference in overall body weight (Figure 4—figure supplement 2A–B). No significant difference was observed in the weight of individual organs except for a small decrease in vWAT and a small increase in liver weight (Figure 4—figure supplement 2C–D). vWAT and sWAT from AdHk2KO and control mice were morphologically indistinguishable (Figure 4—figure supplement 2E). AdHk2KO and control mice displayed similar mRNA expression and circulating levels of leptin and adiponectin (Figure 4—figure supplement 2F–I). Expression of brown adipocyte markers (Ucp1 and Dio2) was also unchanged in AdHk2KO mice (Figure 4—figure supplement 2J). The above similarities in adipose tissue from AdHk2KO and control mice are not surprising given that the HK2 knockout was induced at 4–5 weeks of age, after adipose tissue is fully developed. However, contrary to controls, ND-fed AdHk2KO mice were insulin insensitive and severely glucose intolerant as measured by conventional tolerance tests (Figure 4C–D). Moreover, glucose intolerance was observed in adipose-specific heterozygous Hk2 knockout mice (Figure 4—figure supplement 3A–C). These data indicate that partial loss of HK2, as observed in HFD-fed wild-type mice (Figure 1), is sufficient to cause glucose intolerance. We also measured the insulin insensitivity of AdHk2KO mice in hyperinsulinemic-euglycemic clamp conditions (Figure 4E–F and Figure 4—figure supplement 3D–F). Compared to control mice, AdHk2KO mice required significantly less or no infusion of glucose to maintain euglycemia (Figure 4E–F), despite similar hyperinsulinemia (Figure 4—figure supplement 3D). We note that the glycolytic rate before insulin infusion is higher in AdHk2KO mice (Figure 4—figure supplement 3F). The reason for this increased basal glycolytic rate is unknown. The above findings confirm that loss of HK2 specifically in adipose tissue is sufficient to cause insulin insensitivity and thereby hyperglycemia, even in ND-fed mice. Figure 4.: Loss of adipose HK2 causes hyperglycemia in mice.(A) Immunoblot analyses of HK2 expression and insulin signaling in vWAT of control and adipose-specific Hk2 knockout (AdHk2KO) mice. Mice were fasted overnight or fasted overnight and re-fed for 3 hours. Quantification data is normalized to a loading control. n=6. (B) Fold change (AdHk2KO/control) of metabolites in glycolysis and the pentose phosphate pathway in vWAT of control and AdHk2KO mice. Student’s t test. p<0.05. n=9 (control) and 10 (AdHk2KO). (C) Insulin tolerance test (ITT) on control and AdHk2KO mice. The mice were fasted for 6 hours and injected with insulin (0.5 U/kg body weight). Two-way ANOVA. p<0.05, *p<0.001. n=6 (control) and 10 (AdHk2KO). (D) Glucose tolerance test (GTT) on control and AdHk2KO mice. Mice were fasted for 6 hours and injected with glucose (2 g/kg body weight). Two-way ANOVA for glucose curves and Student’s t test for AUC. p<0.05, p<0.01. n=7 (control) and 12 (AdHk2KO). (E–F) Hyperinsulinemic-euglycemic clamp studies on control and AdHk2KO mice. Mice were fasted for 6 hours. Under insulin clamp, euglycemia was maintained (E) by manipulating glucose infusion rate (F). Bar graph shows average glucose infusion rate under euglycemia. Two-way ANOVA for glucose infusion rate curve and Student’s t test for average glucose infusion rate, p<0.01, ***p<0.0001. n=6 (control) and 7 (AdHk2KO). See Figure 4—source data 1. Figure 4—source data 1.Uncropped blots and source data for graphs for Figure 4. We also investigated HFD-fed AdHk2KO mice. No significant difference was observed in fat mass, body weight, organ weight and insulin tolerance in AdHk2KO mice compared to controls, in mice fed a HFD for 12 weeks (Figure 4—figure supplement 4A–D). The similar phenotype of HFD-fed AdHk2KO and control mice was expected due to the HFD-induced, early loss of HK2 expression in the control mice (Figure 1A–B and Figure 1—figure supplement 2D–F). ## Adipose-specific Hk2 knockout causes selective insulin resistance in liver Consistent with reduced glucose accumulation observed in adipocytes in vitro (Figure 3E), we also observed reduced glucose uptake in adipose tissue in vivo, as measured upon injection of the glucose tracer 14C-2-deoxyglucose during our hyperinsulinemic-euglycemic clamp studies (Figure 5A). However, adipose tissue accounts for only <$5\%$ of glucose disposal (Jackson et al., 1986; Kowalski and Bruce, 2014). Thus, the insulin insensitivity and glucose intolerance observed in AdHk2KO mice (Figure 4C–F) cannot be explained solely by impaired glucose uptake in adipose tissue. Glucose uptake in skeletal muscle, which accounts for 70–$80\%$ of insulin-inducible glucose clearance (Kowalski and Bruce, 2014), was unimpaired in AdHk2KO mice (Figure 5A). Plasma insulin levels were also similar in AdHk2KO and control mice (Figure 5—figure supplement 1A–B). Moreover, insulin signaling was not affected in skeletal muscle and liver of fasted and re-fed AdHk2KO mice (Figure 4—figure supplement 1D and Figure 5—figure supplement 1C). These findings indicate that the severity of the glucose intolerance in AdHk2KO mice is not due to defects in glucose uptake in skeletal muscle, insulin secretion or insulin signaling in peripheral tissues. Figure 5.: Loss of adipose HK2 causes reduced glucose disposal in adipose tissue and de-represses glucose production in liver.(A) Glucose uptake measured at the end of hyperinsulinemic-euglycemic clamp. 2DG was injected 30 min prior to organ collections. The 2DG values were normalized by tissue mass used for the assay. Mann-Whitney test, p<0.05, *p<0.01. n=5 (control) and 5 (AdHk2KO) mice. (B) Endogenous glucose production under basal and hyperinsulinemic-euglycemic clamp conditions. Mann-Whitney test, p<0.05, *p<0.01. n=6 (control) and 7 (AdHk2KO) mice. (C) Pyruvate tolerance test (PTT) on control and AdHk2KO mice. Mice were fasted for 15 hours and injected with pyruvate (2 g/kg body weight). Two-way ANOVA, p<0.05, *p<0.01. n=5 (control) and 6 (AdHk2KO). (D) mRNA levels of gluconeogenic genes (left) and lipogenic genes (rignt) in liver of control and AdHk2KO mice. Multiple t test, p<0.05, *p<0.01. n=9 (control) and 12 (AdHk2KO). (E) De novo-synthesized plasma TG. Mice were treated with 3H-H2O and incorporation of 3H in plasma TG was measured. Student’s t test, p<0.05. n=7 (control) and 5 (AdHk2KO). (F) Plasma triglyceride (TG) levels in control and AdHk2KO mice. Student’s t test, p<0.05. n=8 (control) and 10 (AdHk2KO). (G) Plasma cholesterol, low density lipoprotein (LDL), or high density lipoprotein (HDL) levels in control and AdHk2KO mice. Multiple t test, p<0.01. n=8 (control) and 10 (AdHk2KO). See Figure 5—source data 1. Figure 5—source data 1.Source data for graphs for Figure 5. Another potential explanation for the severity of glucose intolerance in AdHk2KO mice is de-repression of hepatic glucose production. Adipose tissue is known to impinge negatively on glucose production in liver (Abel et al., 2001; Kumar et al., 2010; Tang et al., 2016; Vijayakumar et al., 2017), the main glucose producing organ. We indeed observed increased glucose production in AdHk2KO mice under hyperinsulinemic-euglycemic clamp conditions (Figure 5B). To test further whether the increased glucose production in AdHk2KO mice is due to enhanced gluconeogenesis, we performed a pyruvate tolerance test (PTT). AdHk2KO mice displayed significantly higher production of glucose upon pyruvate injection, compared to controls (Figure 5C). Consistent with the observed increase in pyruvate-dependent glucose production, gluconeogenic genes (G6pc and Pepck) were upregulated in AdHk2KO liver (Figure 5D). These findings suggest that enhanced hepatic glucose production accounts for the severity of glucose intolerance in AdHk2KO mice. Thus, loss of adipose HK2 non-cell-autonomously promotes glucose intolerance by enhancing hepatic glucose production. We observed enhanced glucose production in AdHk2KO mice despite normal systemic insulin signaling, including in liver (Figure 5—figure supplement 1C). This is an apparent paradox because insulin signaling normally inhibits glucose production. To investigate this paradox, we examined hepatic lipogenic gene expression, another readout for insulin action. The insulin-stimulated transcription factors sterol regulatory element-binding protein 1 c (SREBP1c) and carbohydrate responsive element binding protein (ChREBP) activate fatty acid synthesis genes and thereby promote de novo lipogenesis in liver (Horton et al., 2002; Iizuka et al., 2004). Expression of Srebp1c, Mlxipl/Chrebp, and fatty acid synthesis genes (Acly, Scd1, Fasn, and Acc) was maintained, even mildly increased, in liver of AdHk2KO mice (Figure 5D). FASN and ACC protein levels were also slightly increased in liver of AdHk2KO mice (Figure 5—figure supplement 1C). These findings indicate that loss of adipose HK2 does not inhibit hepatic lipogenesis although hepatic glucose production is enhanced. The increased hepatic glucose production and maintained lipogenesis, as observed in AdHk2KO mice, resembles a condition in diabetic patients known as selective insulin resistance (Brown and Goldstein, 2008; see below). Thus, loss of adipose HK2 causes selective insulin resistance and thereby contributes to the pathogenesis of type 2 diabetes. We note that hepatic triglyceride (TG) levels were unchanged in AdHk2KO mice despite increased expression of lipogenic genes and enzymes in liver (Figure 5D and Figure 5—figure supplement 1C–E) and increased de novo TG synthesis (Figure 5E). Most likely, de novo-synthesized hepatic TG in AdHk2KO mice is secreted and delivered to adipose tissue for storage, as suggested by the higher levels of circulating TG and cholesterol in AdHk2KO mice (Figure 5F–G). ## Adipose-specific Hk2 knockout impairs lipogenesis and enhances fatty acid release in adipose tissue How does loss of HK2 in adipose tissue increase hepatic glucose production? Adipose lipogenesis non-cell-autonomously inhibits hepatic glucose production (Vijayakumar et al., 2017). Adipose-specific knockout of the transcription factor ChREBP decreases adipose lipogenesis and thereby increases hepatic glucose production (Ortega-Prieto and Postic, 2019; Vijayakumar et al., 2017). Glucose-derived metabolites have been shown to promote ChREBP activity (Dentin et al., 2012; Kabashima et al., 2003; Kawaguchi et al., 2002; Kim et al., 2016; Li et al., 2010). Based on these observations, we hypothesized that HFD-induced loss of ChREBP activity and thus lipogenesis in adipose tissue may cause increased hepatic glucose production. To test this hypothesis, we examined Mlxipl/Chrebp expression in adipose tissue of AdHk2KO mice. ChREBP has two isoforms. Constitutively expressed ChREBPα promotes transcription of ChREBPβ which then activates lipogenic genes (Herman et al., 2012). Expression of Mlxipl-beta/Chrebp-beta, but not Mlxipl-alpha/Chrebp-alpha, was significantly decreased in adipose tissue of AdHk2KO mice (Figure 6A). Consistent with reduced Mlxipl-beta/Chrebp-beta expression, lipogenic genes and enzymes and ultimately lipogenesis were down-regulated in adipose tissue of AdHk2KO mice (Figure 6A–D). The above findings indicate that loss of adipose HK2 causes loss of ChREBP activity and lipogenesis in adipose tissue. Furthermore, our above findings combined with previous literature (Vijayakumar et al., 2017) suggest that loss of adipose HK2 promotes hepatic glucose production via loss of adipose lipogenesis. We note that adipose tissue weight in AdHk2KO mice is essentially unchanged (Figure 4—figure supplement 2C) despite loss of lipogenesis, likely due to a compensating supply of TG from liver (Figure 5D–F). Figure 6.: Decreased lipogenesis and enhanced fatty acid release in adipose tissue in AdHk2KO mice.(A) mRNA levels of fatty acid synthesis genes in vWAT (top), sWAT(middle), and BAT (bottom) from control and AdHk2KO mice. Multiple t test, p<0.05, p<0.01, p<0.001, *p<0.0001. n=9 (control) and 12 (AdHk2KO) for vWAT and sWAT. n=10 (control) and 10 (AdHk2KO) for BAT. (B) Immunoblot analyses of fatty acid synthesis enzymes in vWAT of control and Adk2KO mice. n=6. (C) Quantification of panel B. Data is normalized to loading controls. Student’s t test. p<0.01. **p<0.0001. (D) De novo lipogenesis of vWAT explants of control and AdHk2KO mice. vWAT explants were treated with 3H-H2O in the absence or presence of 100 nM insulin for 1 hour. Two-way ANOVA, p<0.01, ***p<0.0001. n=8 (control) and 10 (AdHk2KO). (E) Non-esterified fatty acid (NEFA) release of vWAT explants of control and AdHk2KO mice. vWAT explants were treated with or without 10 µM isoproterenol in the absence or presence of 100 nM insulin. Multiple t test, p<0.05. n.s., not significant. n=10 (control) and 12 (AdHk2KO). (F) Plasma NEFA levels in control and AdHk2KO mice before (0 min) and after (15 min) glucose injection. Mice were fasted for 6 hours and injected with glucose (2 g/kg body weight). Two-way ANOVA, p<0.001. n.s., not significant. n=10 (control) and 11 (AdHk2KO). (G) Plasma NEFA levels in control and AdHk2KO mice after (15 min) glucose injection was normalized by plasma NEFA levels at 0 min (before glucose injection) in E. Student’s t test, p<0.01. n=10 (control) and 11 (AdHk2KO). See Figure 6—source data 1. Figure 6—source data 1.Uncropped blots and source data for Figure 6.* Non-esterified fatty acid (NEFA) released from adipose tissue, in addition to loss of adipose lipogenesis, promotes glucose production in liver (Perry et al., 2015; Titchenell et al., 2016). Furthermore, ChREBP in adipose tissue both promotes lipogenesis and inhibits NEFA release (Vijayakumar et al., 2017). Thus, we examined the effect of HK2 loss on NEFA release. Ex vivo, insulin inhibited isoproterenol-induced NEFA release in adipose tissue from control mice, but not in explants from AdHk2KO mice (Figure 6E). In vivo, glucose administration inhibited NEFA release only in control mice (Figure 6F–G) despite similar physiological increases insulin levels in fasted AdHk2KO and control mice (Figure 5—figure supplement 1B), in agreement with the above ex vivo experiment. These findings suggest that loss of HK2 simultaneously inhibits lipogenesis and promotes NEFA release in adipose tissue, both of which contribute to de-repression of hepatic glucose production. ## Mechanisms of HFD-induced HK2 down-regulation Our findings suggest that HFD causes loss of adipose HK2 and thereby triggers hyperglycemia. How does HFD down-regulate adipose HK2? To answer this question, we first examined Hk2 mRNA expression in HFD- and ND-fed mice. HFD-fed mice exhibited decreased Hk2 mRNA in sWAT but not in vWAT or BAT (Figure 7A). This suggests that HK2 synthesis is down-regulated at the transcriptional level in sWAT, and at a post-transcriptional level in vWAT and BAT. To investigate the post-transcriptional mechanism, we measured synthesis of HK2 in vWAT of HFD-fed mice. vWAT explants from HFD- or ND-fed mice were treated with L-azidohomoalanine (AHA), a methionine analog, and AHA-containing polypeptides were purified and quantified by mass spectrometry (Supplementary file 3). The amount of AHA-containing HK2 polypeptides was significantly decreased in vWAT explants from HFD-fed mice (Figure 7B). Thus, HFD down-regulates HK2 in vWAT by inhibiting Hk2 mRNA translation. Figure 7.: Mechanism of HFD-induced HK2 down-regulation in adipose tissue.(A) Hk2 mRNA levels in sWAT, vWAT, and BAT from ND- and 4 week HFD-fed mice. Student’s t test, p<0.01. n=10 (sWAT from ND and HFD), 10 (vWAT from ND and HFD), 8 (BAT from ND), 10 (BAT from HFD). (B) Nascent polypeptides. vWATs isolated from ND- or 4 week HFD-fed mice were labled with L-azidohomoalanine (AHA), and AHA-incorporated polypeptides were measured by mass spectrometer. FASN and ACC serve as positive controls and CALX serves as a negative control. Multiple t test, p<0.05, p<0.01. n=4 (ND) and 4 (HFD). (C) Foxk1 and Foxk2 mRNA levels in sWAT from ND- or 4 week HFD-fed mice. Multiple t test, p<0.01, **p<0.001. n=10 (ND) and 10 (HFD). (D) Foxk1 and Foxk2 mRNA levels in vWAT of ND- or 4 week HFD-fed wild-type C57BL6JRj mice. No significant difference in multiple t test. n=10 (ND) and 10 (HFD). (E) Foxk1 and Foxk2 mRNA levels in BAT of ND- or 4 week HFD-fed wild-type C57BL6JRj mice. No significant difference in multiple t test. n=8 (ND) and 10 (HFD). See Figure 7—source data 1. Figure 7—source data 1.Source data for graphs for Figure 7. Are there any other adipose proteins post-transcriptionally regulated like HK2? In our AHA-proteomics dataset (Supplementary file 3), we identified 155 proteins whose AHA incorporation positively correlated with HK2 AHA-incorporation. Pathway enrichment analysis of the 155 proteins yielded ribosomal proteins, the electron transport chain (ETC), oxidative phosphorylation (OXPHOS), fatty acid synthesis, and glycolysis as the top 5 pathways (Figure 7—figure supplement 1A–D). As demonstrated in Figure 6, downregulation of enzymes in fatty acid synthesis (e.g. FASN and ACC) was mainly due to reduced transcription (Figure 7—figure supplement 1C). However, reduced AHA-incorporation in ribosomal, ETC, and OXPHOS proteins was post-transcriptional (Figure 7—figure supplement 1B), suggesting that these proteins might be regulated in a manner similar to HK2. Regulated protein synthesis in adipose tissue may play an important role in the development of hyperglycemia. It was recently reported that the forkhead transcription factors FOXK1 and FOXK2 promote Hk2 transcription in adipose tissue (Sukonina et al., 2019). We investigated whether the transcriptional down-regulation of Hk2 in sWAT is due to loss of FOXK1 and FOXK2. Foxk1 and Foxk2 expression was significantly decreased in sWAT, but not in vWAT or BAT, of HFD-fed mice (Figure 7C–E). Thus, HFD appears to down-regulate HK2 in sWAT by inhibiting FOXK$\frac{1}{2}$ and, thereby, expression of the Hk2 gene. ## Discussion How obesity causes insulin insensitivity and hyperglycemia is a long-standing question. In this study, we show that HFD causes hyperglycemia by inducing loss of HK2 specifically in adipose tissue. Loss of adipose HK2 is sufficient to cause glucose intolerance, in two ways (Figure 8). First, HK2 loss results in reduced glucose disposal by adipose tissue due to the inability of adipocytes to trap and metabolize non-phosphorylated glucose (Figure 5A). Second, loss of HK2 in adipocytes de-represses glucose production in liver (Figure 5B). We propose that loss of adipose HK2 is a mechanism of diet-induced insulin insensitivity and hyperglycemia. Figure 8.: Diet-induced HK2 in adipose tissue promotes glucose intolerance.(A) HK2 promotes lipogenesis and suppresses NEFA release in adipose tissue (left), suppressing hepatic glucose production (right) and thus maintaining glucose homeostasis. (B) Diet-induced loss of adipose HK2 triggers glucose intolerance via reduced glucose disposal in adipocytes (left) and de-repressed hepatic gluconeogenesis despite maintained lipogenesis (right). In liver, insulin normally represses glucose production to decrease blood glucose and up-regulates lipogenesis to increase energy storage. Paradoxically, in type 2 diabetes, insulin fails to inhibit glucose production yet stimulates lipogenesis, hence the liver is selectively insulin resistant (Brown and Goldstein, 2008). Furthermore, previous findings suggest that selective insulin resistance occurs despite intact hepatic insulin signaling (Titchenell et al., 2016). Consistent with these previous findings, we observed that loss of adipose HK2 causes selective insulin resistance without affecting hepatic insulin signaling. Thus, we propose that loss of adipose HK2 is a mechanism for selective insulin resistance in liver and ultimately diabetes. How does adipose HK2 control hepatic glucose production? As described above, one possible explanation is that decreased lipogenesis and increased NEFA release in adipose tissue, due to loss of HK2, promotes glucose production (Figure 6; Perry et al., 2015; Titchenell et al., 2016; Vijayakumar et al., 2017). Another interesting possibility is that loss of adipose HK2 causes increased hepatic glucose production via the central nervous system (CNS). Adipose tissue has a sensory nervous system (Blaszkiewicz et al., 2019; Fishman and Dark, 1987; Frei et al., 2022; Kreier et al., 2006; Makwana et al., 2021; Song et al., 2009; Wang et al., 2022) which may communicate with the liver via the sympathetic nervous system. Sympathetic activity promotes hepatic glucose production (Niijima and Fukuda, 1973; Shimazu and Fukuda, 1965). The above two models are not mutually exclusive since released bioactive fatty acids could act directly on the liver and/or adipose sensory neurons. A third possibility to explain increased hepatic glucose production is that increased glucose uptake by the liver due to reduced glucose disposal in adipose tissue may stimulate hepatic glucose production, consistent with the demonstration by Kim et al. that hepatic ChREBP promotes glucose production (Kim et al., 2016). The phenotype of AdHk2KO mice is similar to that of adipose-specific GLUT4 knockout mice (Abel et al., 2001). Both knockout mice display reduced lipogenesis in adipose tissue and are hyperglycemic due to decreased glucose disposal in adipose tissue and increased glucose production in the liver (Vijayakumar et al., 2017). The phenotypic similarity of GLUT4 and HK2 knockout mice underscores the importance of adipose glucose metabolism in systemic insulin sensitivity and glucose homeostasis. However, our findings on diet-induced HK2 downregulation provide important new insight on the development of diet-induced hyperglycemia. First, unlike GLUT4, HK2 downregulation in adipose tissue correlates with HFD-induced hyperglycemia in mice (Figure 1—figure supplement 2). Second, it has been shown that glucose or a downstream metabolite(s) promotes ChREBP activity and lipogenesis in metabolic organs including adipose tissue (Ortega-Prieto and Postic, 2019). Two previous studies demonstrated that HFD prevents ChREBP activation and lipogenesis in adipose tissue, which leads to hyperglycemia due to enhanced hepatic glucose production in mice (Herman et al., 2012; Vijayakumar et al., 2017). Importantly, one of these studies showed that HFD inhibits ChREBP activity in adipose tissue independently of GLUT4 expression (Herman et al., 2012), concluding that the underlying mechanism of HFD-induced inhibition of ChREBP and hyperglycemia is still unknown. Our findings suggest that the physiological mechanism of HFD-induced inhibition of adipose ChREBP and lipogenesis, and consequently hyperglycemia, may be loss of HK2. Thus, we propose that the primary defect in adipose tissue of HFD-fed mice is not at the level of GLUT4 expression and glucose transport, but rather at the level of HK2 expression and glucose phosphorylation. Can HFD-resistant HK2 expression in adipose tissue prevent diet-induced insulin insensitivity and hyperglycemia? To this point, our attempts to express HK2 in WAT of HFD-fed mice were unsuccessful (unpublished data), most likely because HK2 expression is tightly controlled at the post-transcriptional level (Figure 7). Importantly, expression of glucokinase (also known as HK4, hepatic hexokinase) in adipose tissue has been shown to prevent insulin insensitivity in HFD-fed mice (Muñoz et al., 2010). Elucidating the molecular mechanism of HFD-induced translational repression of HK2 may lead to a novel strategy in the treatment of insulin insensitivity and type 2 diabetes. A previous study demonstrated that global heterozygous Hk2 knockout mice have normal, even better, glucose tolerance compared to controls (Heikkinen et al., 1999). How can one explain this apparent discrepancy between global Hk2 knockout and our adipose-specific Hk2 knockout? We note that adipose-specific GLUT4 knockout mice display glucose intolerance (Abel et al., 2001), but global GLUT4 knockout mice are glucose tolerant due to a compensatory increase in glucose uptake in liver (Katz et al., 1995; Ranalletta et al., 2005). We observed increased systemic glycolysis in AdHk2KO mice (Figure 4—figure supplement 3F). Global ablation of HK2 mice may provoke an even stronger compensatory response to maintain systemic glucose homeostasis. This may also explain the observation that global heterozygous Hk2 knockout mice display better glucose handling than wild-type controls (Heikkinen et al., 1999). We note that Hk2 knockout did not completely phenocopy to the effect of obesity in mice and humans. While HK2 was decreased approximately $75\%$ in vWAT of AdHk2KO mice (Figure 4A), HK2 expression in HFD-fed mice and HK activity in oWAT from obese patients were decreased only ~$60\%$ and~$30\%$, respectively (Figure 1B and G). Is the reduction in HK2 in obese mice and patients sufficient to contribute to systemic insulin insensitivity and glucose intolerance? A previous study showed that adipose-specific Rab10 knockout mice display reduced GLUT4 translocation to the plasma membrane and thus a ~$50\%$ reduction in insulin-stimulated glucose disposal, in isolated adipocytes (Vazirani et al., 2016). Importantly, adipose-specific Rab10 knockout mice failed to suppress hepatic glucose production and were thus insulin insensitive, indicating that a limited disruption in glucose metabolism in adipose tissue can significantly impact systemic insulin sensitivity and glucose homeostasis. Thus, the partial loss of HK2 observed in obese mice and patients may be sufficient to impact systemic insulin sensitivity and thereby glucose homeostasis. We found an HK2-R42H mutation in naturally hyperglycemic Mexican cavefish. To date, no human monogenic disease has been reported associated with mutations in the HK2 gene. However, 12 R42W and 4 R42Q alleles are reported in the genomAD database (Karczewski et al., 2020). Further studies are required to determine whether these alleles are loss-of-function and associated with diabetes. ## Mouse Wild-type C57BL/6JRj mice were purchased from JANVIER LABS (Le Genest-Saint-lsle, France). Mice carrying Hk2 with exons 4–10 flanked by loxP sites (Hk2fl/fl) were previously described (Patra et al., 2013). Adipoq-CreERT2 mice were provided by Prof. Stefan Offermanns (MPI-HLR, Germany)(Sassmann et al., 2010). Hk2fl/fl mice were crossed with Adipoq-CreERT2 mice, and resulting Cre positive Hk2fl/+ mice were crossed with Hk2fl/fl mice to generate adipocyte-specific Hk2 knockout (Adipoq-CreERT2 positive Hk2fl/fl) mice (AdHk2KO). Hk2 knockout was induced by i.p. injection of 1 mg/mouse tamoxifen (Sigma-Aldrich, St. Louis, Missouri) in corn oil (Sigma-Aldrich, St. Louis, Missouri) for 7 days. Littermate Cre negative animals were used as a control. Control mice were also treated with tamoxifen. Mice were housed at 22 °C in a conventional facility with a 12 hours light/dark cycle with unlimited access to water, and normal diet (ND, KLIBA NAFAG, Kaiseraugst, Switzerland) or high-fat diet (HFD: $60\%$ kcal % fat NAFAG 2127, KLIBA, Kaiseraugst, Switzerland). Only male mice between 6 and 17 weeks of age were used for experiments. ## Plasmids pTwist-surface HK2 and pTwist-Molino HK2 plasmids were purchased from Twist Bioscience (South San Francisco, California). The R42H mutation was introduced into the pLenti-CMV-ratHK2 (DeWaal et al., 2018) by PCR using the oligos 5’atttctaggcACttccggaaggagatggagaaag3’ and 5’cttccggaaGTgcctagaaatctccagaagggtc3’. The desired sequence change was confirmed. Figure 4—figure supplement 1. ## Cell culture 3T3-L1 and HEK293T cells were obtained from ATCC. We perform mycoplasma contamination every three month. The cell lines used were mycoplasma negative. HEK293T cells were cultured in M1 medium composed of DMEM high glucose (Sigma-Aldrich, St. Louis, Missouri) supplemented with 4 mM glutamine (Sigma-Aldrich, St. Louis, Missouri), 1 mM sodium pyruvate (Sigma-Aldrich, St. Louis, Missouri), 1 x penicillin and streptomycin (Sigma-Aldrich, St. Louis, Missouri), and $10\%$ FBS (ThermoFisher Scientific, Waltham, Massachusetts). 3T3-L1 cells were cultured and differentiated as previously described (Zebisch et al., 2012). In brief, 3T3-L1 preadipocyte cells were maintained in M1 medium at 37 °C incubator with $5\%$ CO2. For differentiation, cells were maintained in M1 medium for 2 days after reaching confluence. The cells were then transferred to M2 medium composed of M1 medium supplemented with 1.5 µg/mL insulin (Sigma-Aldrich, St. Louis, Missouri), 0.5 mM IBMX (AdipoGen LIFE SCIENCES, Liestal, Switzerland), 1 µM dexamethasone (Sigma-Aldrich, St. Louis, Missouri), and 2 µM rosiglitazone (AdipoGen LIFE SCIENCES, Liestal, Switzerland), defined as day 0 post-differentiation. After 2 days, the cells were transferred to M3 medium (M1 with 1.5 µg/mL insulin). At day 4 post-differentiation, cells were transferred back to M2 medium. From day 6 post-differentiation, cells were maintained in M3 with medium change every 2 days. For Hk2 knockdown, MISSION shRNA (TRCN0000280118) or control pLKO plasmid were purchased from Sigma-Aldrich (St. Louis, Missouri) and co-transfected with psPAX2 (a gift from Didier Trono: Addgene plasmid # 12260) and pCMV-VSV-G (Stewart et al., 2003; a gift from Robert Weinberg: Addgene plasmid # 8454) into HEK293T cells. Supernatants containing lentivirus were collected one day after transfection, and used to infect undifferentiated 3T3-L1 cells. Transduced cells were selected by puromycin (Thermo Fisher Scientific, Waltham, Massachusetts). For all experiments, 8–14 days post-differentiated cells were used. For HK2 overexpression in HEK293T cells, plasmids were transfected with jetPRIME (Polyplus-transfection, Illkirch-Graffenstaden, France) following manufacturer’s instructions. ## Human biopsies Omental white adipose tissue (oWAT) biopsies were obtained from lean subjects with normal fasting glucose level and body mass index (BMI) <27 kg/m2, from obese non-diabetic subjects with BMI >35 kg/m2 HbA1c<$6.0\%$, and from obese diabetic sujects with BMI >35 kg/m2 HbA1c>$6.1\%$ (Supplementary file 2). All subjects gave informed consent before the surgical procedure. Patients were fasted overnight and underwent general anesthesia. All oWAT specimens were obtained between 8:30 and 12:00 am, snap-frozen in liquid nitrogen, and stored at –80 °C for subsequent use. ## Proteomics Tissues were pulverized and homogenized in a tissue lysis buffer containing 100 mM Tris (VWR, Radnor, Pennsylvania)-HCl (Merck, Burlington, Massachusetts) pH7.5, 2 mM EDTA (Sigma-Aldrich, St. Louis, Missouri), 2 mM EGTA (Sigma-Aldrich, St. Louis, Missouri), 150 mM NaCl (Sigma-Aldrich, St. Louis, Missouri), $1\%$ Triton X-100 (Fluka Chemie, Buchs, Switzerland), cOmplete inhibitor cocktail (Sigma-Aldrich, St. Louis, Missouri) and PhosSTOP (Sigma-Aldrich, St. Louis, Missouri). Proteins were precipitated by trichloroacetic acid (Sigma-Aldrich, St. Louis, Missouri) and the resulting protein pellets were washed with cold acetone (Merck, Burlington, Massachusetts). 25 μg of peptides were labeled with isobaric tandem mass tags (TMT 10-plex, Thermo Fisher Scientific, Waltham, Massachusetts) as described previously (Ahrné et al., 2016). Shortly, peptides were resuspended in 20 μl labeling buffer (2 M urea (Sigma-Aldrich, St. Louis, Missouri), 0.2 M HEPES (Sigma-Aldrich, St. Louis, Missouri), pH 8.3) and 5 μL of each TMT reagent were added to the individual peptide samples followed by a 1 hour incubation at 25 °C. To control for ratio distortion during quantification, a peptide calibration mixture consisting of six digested standard proteins mixed in different amounts was added to each sample before TMT labeling. To quench the labelling reaction, 1.5 μL aqueous 1.5 M hydroxylamine solution (Merck, Burlington, Massachusetts) was added and samples were incubated for another 10 min at 25 °C followed by pooling of all samples. The pH of the sample pool was increased to 11.9 by adding 1 M phosphate buffer pH 12 (Sigma-Aldrich, St. Louis, Missouri) and incubated for 20 min at 25 °C to remove TMT labels linked to peptide hydroxyl groups. Subsequently, the reaction was stopped by adding 2 M hydrochloric acid (Merck, Burlington, Massachusetts) and until a pH <2 was reached. Finally, peptide samples were further acidified using $5\%$ TFA (Thermo Fisher Scientific, Waltham, Massachusetts), desalted using Sep-Pak Vac 1 cc (50 mg) C18 cartridges (Waters, Milford, Massachusetts) according to the manufacturer’s instructions and dried under vacuum. TMT-labeled peptides were fractionated by high-pH reversed phase separation using a XBridge Peptide BEH C18 column (3,5 µm, 130 Å, 1 mm x 150 mm, Waters, Milford, Massachusetts) on an 1260 Infinity HPLC system (Agilent Technologies, Santa Clara, California). Peptides were loaded on column in buffer A (20 mM ammonium formate (Sigma-Aldrich, St. Louis, Missouri) in water, pH 10) and eluted using a two-step linear gradient from $2\%$ to $10\%$ in 5 min and then to $50\%$ buffer B (20 mM ammonium formate (Sigma-Aldrich, St. Louis, Missouri) in $90\%$ acetonitrile (Thermo Fisher Scientific, Waltham, Massachusetts), pH 10) over 55 min at a flow rate of 42 µl/min. Elution of peptides was monitored with a UV detector (215 nm, 254 nm) and a total of 36 fractions were collected, pooled into 12 fractions using a post-concatenation strategy as previously described (Wang et al., 2011) and dried under vacuum. Dried peptides were resuspended in 20 μl of $0.1\%$ aqueous formic acid (Sigma-Aldrich, St. Louis, Missouri) and subjected to LC–MS/MS analysis using a Q Exactive HF Mass Spectrometer fitted with an EASY-nLC 1000 (both Thermo Fisher Scientific, Waltham, Massachusetts) and a custom-made column heater set to 60 °C. Peptides were resolved using a RP-HPLC column (75 μm×30 cm) packed in-house with C18 resin (ReproSil-Pur C18–AQ, 1.9 μm resin; Dr. Maisch GmbH, Ammerbuch, Germany) at a flow rate of 0.2 μL/min. The following gradient was used for peptide separation: from $5\%$ B to $15\%$ B over 10 min to $30\%$ B over 60 min to $45\%$ B over 20 min to $95\%$ B over 2 min followed by 18 min at $95\%$ B. Buffer A was $0.1\%$ formic acid (Sigma-Aldrich, St. Louis, Missouri) in water and buffer B was $80\%$ acetonitrile (Thermo Fisher Scientific, Waltham, Massachusetts), $0.1\%$ formic acid (Sigma-Aldrich, St. Louis, Missouri) in water. The mass spectrometer was operated in DDA mode with a total cycle time of approximately 1 s. Each MS1 scan was followed by high-collision-dissociation (HCD) of the 10 most abundant precursor ions with dynamic exclusion set to 30 s. For MS1, 3e6 ions were accumulated in the Orbitrap over a maximum time of 100 ms and scanned at a resolution of 120,000 FWHM (at 200 m/z). MS2 scans were acquired at a target setting of 1e5 ions, accumulation time of 100 ms and a resolution of 30,000 FWHM (at 200 m/z). Singly charged ions and ions with unassigned charge state were excluded from triggering MS2 events. The normalized collision energy was set to $35\%$, the mass isolation window was set to 1.1 m/z and one microscan was acquired for each spectrum. The acquired raw-files were converted to the mascot generic file (mgf) format using the msconvert tool (part of ProteoWizard, version 3.0.4624 [2013-6-3]). Using the MASCOT algorithm (Version 2.4.1, Matrix Science, Boston Massachusetts), the mgf files were searched against a decoy database containing normal and reverse sequences of the predicted SwissProt entries of Mus musculus (https://www.ebi.ac.uk/, release date $\frac{2014}{11}$/24), the six calibration mix proteins (Ahrné et al., 2016) and commonly observed contaminants (in total 50214 sequences for Mus musculus) generated using the SequenceReverser tool from the MaxQuant software (Version 1.0.13.13). The precursor ion tolerance was set to 10 ppm and fragment ion tolerance was set to 0.02 Da. The search criteria were set as follows: full tryptic specificity was required (cleavage after lysine or arginine residues unless followed by proline), 3 missed cleavages were allowed, carbamidomethylation (C) and TMT6plex (K and peptide N-terminus) were set as fixed modification and oxidation (M) as a variable modification. Next, the database search results were imported into the Scaffold Q+software (version 4.3.2, Proteome Software Inc, Portland, Oregon) and the protein false identification rate was set to $1\%$ based on the number of decoy hits. Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony. Proteins sharing significant peptide evidence were grouped into clusters. Acquired reporter ion intensities in the experiments were employed for automated quantification and statically analysis using a modified version of our in-house developed SafeQuant R script (Ahrné et al., 2016). This analysis included adjustment of reporter ion intensities, global data normalization by equalizing the total reporter ion intensity across all channels, summation of reporter ion intensities per protein and channel, calculation of protein abundance ratios and testing for differential abundance using empirical Bayes moderated t-statistics. Finally, the calculated p-values were corrected for multiple testing using the Benjamini−Hochberg method (q-value). Deregulated proteins were selected by log2(fold change)>0.6 or log2(fold change)<–0.6, q-value <0.01. ## Immunoblots Tissues or cells were homogenized in a lysis buffer containing 100 mM Tris (Merck, Burlington, Massachusetts) pH7.5, 2 mM EDTA (Sigma-Aldrich, St. Louis, Missouri), 2 mM EGTA (Sigma-Aldrich, St. Louis, Missouri), 150 mM NaCl (Sigma-Aldrich, St. Louis, Missouri), $1\%$ Triton X-100 (Fluka Chemie, Buchs, Switzerland), cOmplete inhibitor cocktail (Sigma-Aldrich, St. Louis, Missouri) and PhosSTOP (Sigma-Aldrich, St. Louis, Missouri). Protein concentration was determined by the Bradford assay (Bio-rad), and equal amounts of protein were separated by SDS-PAGE, and transferred onto nitrocellulose membranes (GE Healthcare, Chicago, Illinois). Antibodies used in this study were as follows: AKT (Cat#4685 or Cat#2920), AKT-pS473 (Cat#4060), AKT-pT308 (Cat#13038), PRAS40-pT246 (Cat#2997), PRAS40 (Cat#2691), HK2 (Cat#2867), S6-pS$\frac{240}{244}$ (Cat#5364), S6 (Cat#2217), FASN (Cat#3189), ACC (Cat#3662), HA tag (Cat#3724) and ChREBP (Cat#58069) from Cell Signaling Technology (Danvers, Massachusetts), GLUT4 (Cat# NBP2-22214, Bio-Techne, Abingdon, United Kingdom), HSP90, (Cat#sc-13119, Santa Cruz Biotechnology, Dallas, Texas), CALNEXIN (Cat#ADI-SPA-860-F, Enzo Life Sciences, Farmingdale, New York) and ACTIN (Cat#MAB1501, Merck, Burlington, Massachusetts). For quantification, the specific signals were normalized to a loading control. ## Metabolomics Tissues (25–30 mg) were finely ground in a cryogenic grinding before the extraction. Metabolite extraction was performed, in a mixture ice/dry ice, by a cold two-phase methanol–water–chloroform extraction (Elia et al., 2017; van Gorsel et al., 2019). The samples were resuspended in 900 μl of precooled methanol/water ($\frac{5}{3}$) (v/v) and 100 µL of 13C yeast internal standard. Afterwards, 500 μl of precooled chloroform was added to each sample. Samples were vortexed for 10 min at 4 °C and then centrifuged (max. speed, 10 min, 4 °C). The methanol–water phase containing polar metabolites was separated and dried using a vacuum concentrator at 4 °C overnight and stored at −80 °C. For the detection of the pentose phosphate pathway and glycolysis intermediates by LC-MS, a 1290 Infinity II liquid chromatography (Agilent Technologies, Santa Clara, California) with a thermal autosampler set at 4 °C, coupled to a Q-TOF 6546 mass spectrometer (Agilent Technologies, Santa Clara, California) was used. Samples were resuspended in 100 µL of $50\%$ methanol and a volume of 5 µL and 20 μL of sample were injected on Agilent InfinityLab Poroshell 120 HILIC-Z column, 2.1 mm × 150 mm, 2.7 μm, PEEK-lined. The separation of metabolites was achieved at 50 °C with a flow rate of 0.25 ml/min. A gradient was applied for 32 min (solvent A: 10 mM ammonium acetate in water with 2.5 μM InfinityLab Deactivator Additive, pH = 9 – solvent B: 10 mM ammonium acetate in water/acetonitrile 15:85 (v:v) with 2.5 μM InfinityLab Deactivator Additive, pH = 9) to separate the targeted metabolites (0 min: $96\%$B, 2 min: $96\%$B, 5.5 min: $88\%$B, 8.5 min: $88\%$B, 9 min: $86\%$B, 14 min: $86\%$B, 19 min: $82\%$B; 25 min: $65\%$B, 27 min: $65\%$B, 28 min: $96\%$B; 32 min: $96\%$B). The MS operated in negative full scan mode (m/z range: 50–1200) using a shealth gas temperature of 350 °C (12 L/min) and a gas temperature at 225 °C (13 L/min). The nebulizer was set at 35 psi, the fragmentor at 125 V and the capillary at 3500 V. Data was collected using the Masshunter software (Agilent Technologies, Santa Clara, California) and normalized by 13C yeast internal standard and the protein content. For the detection of Acetyl-CoA by LC-MS, a Dionex UltiMate 3000 LC System (Thermo Fisher Scientific, Waltham, Massachusetts) with a thermal autosampler set at 4 °C, coupled to a Q Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific, Waltham, Massachusetts) was used. Samples were resuspended in 100 µL of $50\%$ MeOH and a volume of 10 μl of sample was injected on a C18 column (Acquity UPLC HSS T3 1.8 µm 2.1x100 mm). The separation of metabolites was achieved at 40 °C with a flow rate of 0.25 ml/min. A gradient was applied for 40 min (solvent A: 10 mM Tributyl-Amine, 15 mM acetic acid – solvent B: Methanol) to separate the targeted metabolites (0 min: $0\%$B, 2 min: $0\%$B, 7 min: $37\%$B, 14 min: $41\%$B, 26 min: $100\%$B, 30 min: $100\%$B, 31 min: $0\%$B; 40 min: $0\%$B). The MS operated in negative full scan mode (m/z range: 70–1050 and 750–850 from 5 to 25 min) using a spray voltage of 4.9 kV, capillary temperature of 320 °C, sheath gas at 50.0, auxiliary gas at 10.0. Data was collected using the Xcalibur software (Thermo Fisher Scientific, Waltham, Massachusetts) and analyzed with Matlab for the correction of natural abundance. Data were normalized by 13C yeast internal standard and the protein content. ## RNA isolation and quantitative RT-PCR Total RNA was isolated with TRIzol reagent (Sigma-Aldrich, St. Louis, Missouri) and RNeasy kit (Qiagen, Hilden, Germany). RNA was reverse-transcribed to cDNA using iScript cDNA synthesis kit (Bio-Rad Laboratories, Hercules, California). Semiquantitative real-time PCR analysis was performed using fast SYBR green (Applied Biosystems, Waltham, Massachusetts). Relative expression levels were determined by normalizing each CT values to Tbp using the ∆∆CT method. The sequence for the primers (Microsynth, Balgach, Switzerland) used in this study was as follows: ## Hexokinase assay For measuring hexokinase activity of surface fish HK2, Molino HK2, wild-type rat HK2 or rat HK2-R42H, HEK293T cells were transfected with plasmids containing these HK2 sequences, and cells were harvested at 24 hours after transfection. Hexokinase activity were measured with cell lysates or tissue lysates with a hexokinase assay kit (Abcam, Cambridge, United Kingdom) following manufacturer’s instructions. The final glucose concentration in the hexokinase assay was 4 mM. The hexokinase activities were normalized to the protein content in lysates. ## Seahorse analyses Measurements were performed with an XF96 Extracellular Flux Analyzer (Seahorse Bioscience of Agilent Technologies, Santa Clara, California) following manufacturer’s instructions. ## Lactate measurement Differentiated adipocytes were starved serum overnight and stimulated with 100 nM insulin for 2 hours. Extracellular lactate was measured with the Lactate Pro 2 analyzer (Axon Lab, Stuttgart, Germany). ## 3H-2DG uptake assay in vitro Differentiated adipocytes were starved for serum for 5 hours and then incubated in Krebs Ringer Phosphate Hepes (KRPH) buffer composed of 0.6 mM Na2HPO4 (Fluka Chemie, Buchs, Switzerland), 0.4 mM NaH2PO4 (Fluka Chemie, Buchs, Switzerland), 120 mM NaCl (Sigma-Aldrich, St. Louis, Missouri), 6 mM KCl (Merck), 1 mM CaCl2 (Merck, Burlington, Massachusetts), 1.2 mM MgSO4 (Merck, Burlington, Massachusetts), 12.5 mM HEPES (Thermo Fisher Scientific, Waltham, Massachusetts), $0.2\%$ fatty acid-free BSA (Sigma-Aldrich, St. Louis, Missouri) pH7.4 with or without 100 nM insulin (Sigma-Aldrich, St. Louis, Missouri) for 20 min. Cells were incubated with cold 50 µM 2DG (Sigma-Aldrich, St. Louis, Missouri) containing 0.25 µCi 3H-2-deoxyglucose (2DG, Perkin Elmer, Waltham, Massachusetts) for 5 min and washed three times with cold PBS (Sigma-Aldrich, St. Louis, Missouri). Cells were lysed in the tissue lysis buffer and cleared by centrifugation at 14,000 g for 10 min. Incorporated 3H-2DG was measured with a scintillation counter (Perkin Elmer, Waltham, Massachusetts). ## Insulin tolerance test, glucose tolerance test, pyruvate tolerance test For the insulin and glucose tolerance tests, mice were fasted for 6 hours and insulin Humalog (i.p. 0.75 or 0.5 U/kg body weight, Lilly, Indianapolis, Indiana) or glucose (2 g/kg body weight, Sigma-Aldrich, St. Louis, Missouri) was given, respectively. For the pyruvate tolerance test, mice were fasted for 15 hours and pyruvate (2 g/kg body weight, Sigma-Aldrich, St. Louis, Missouri) was administered. Blood glucose was measured with a blood glucose meter (Accu-Check, Roche Diabetes Care, Indiana polis, Indiana). ## Hyperinsulinemic-euglycemic clamp Hyperinsulinemic-euglycemic clamp was performed as previously described (Smith et al., 2018). In brief, mice were fasted for 6 hours and anesthetized by i.p. injection of 6.25 mg/kg acetylpromazine (Fatro, Bologna, Italy), 6.25 mg/kg midazolam (Sintetica, Val-de-Travers, Switzerland) and 0.31 mg/kg fentanyl (Mepha, Aesch, Switzerland). An infusion needle were placed into the tail vein and 3H-glucose (Parkin Elmer, Waltham, Massachusetts) was infused for 60 min to achieve steady-state levels. Thereafter, the hyperinsulinemic clamp started with a bolus dose (3.3mU) and a constant infusion of insulin (0.09 mU/min) and 3H-glucose. A variable infusion of $12.5\%$ D-glucose (Sigma-Aldrich, St. Louis, Missouri) was used to maintain euglycemia. Blood glucose was measured with a a blood glucose meter (Accu-Check, Roche Diabetes Care, Indiana polis, Indiana) every 5–10 min and the glucose infusion rate was adjusted to maintain euglycemia. Blood samples were taken to determine steady-state levels of [3H]-glucose. After 90 min from the start of the insulin clamp, 14C-2-Deoxyglucose (Parkin Elmer, Waltham, Massachusetts) was i.p. administered to assess tissue-specific glucose uptake. Mice were euthanized by cervical dislocation and the organs were removed and frozen. 3H-glucose and 14C-2-DG phosphate counts in plasma and tissues were measured by a scintillation counter. ## Hematoxylin and eosin staining Tissues were fixed in $4\%$ formalin (Leica Biosystems, Wetzlar, Germany), embedded in paraffin (Leica Biosystems, Wetzlar, German), and sliced into 3-µm-thick section. Tissue sections were stained with Hematoxylin (Sigma-Aldrich, St. Louis, Missouri) and eosin (Waldeck, Münster, Germany), and imaged by Axio Scan. Z1 slidescanner (Zeiss, Oberkochen, Germany). ## In vivo lipogenesis Ad libitum-fed mice were i.p. injected with 1 mCi 3H-H2O (American Radiolabeled Chemicals, St. Louis, Missouri) and sacrificed with i.p.160 mg/kg ketamine (Streuli Pharma, Uznach, Switzerland) and 24 mg/kg xylazine (Streuli Pharma, Uznach, Switzerland). For triglyceride extraction, plasma samples were mixed with 1 mL of 2-propanol (Merck, Burlington, Massachusetts) /n-hexane (Merck, Burlington, Massachusetts) /1 N H2SO4 (Sigma-Aldrich, St. Louis, Missouri) (4:1:1) and incubated for 30 min. ddH2O and n-hexane were added and the resulting n-hexane phase was collected, dried, and counted by a scintillation counter. ## Ex vivo lipogenesis vWAT explants were washed and incubated with low glucose DMEM (Sigma-Aldrich, St. Louis, Missouri) supplemented with $2\%$ fatty acid-free BSA (Sigma-Aldrich, St. Louis, Missouri) and 20 mM HEPES (Thermo Fisher Scientific, Waltham, Massachusetts). Explants were further washed with a buffer containing 10 mM HEPES (Thermo Fisher Scientific, Waltham, Massachusetts), 116 mM NaCl (Sigma-Aldrich, St. Louis, Missouri), 4 mM KCl (Merck, Burlington, Massachusetts), 1.8 mM CaCl2 (Merck, Burlington, Massachusetts), 1 mM MgCl2 (Fluka Chemie, Buchs, Switzerland), 4.5 mM D-glucose (Sigma-Aldrich, St. Louis, Missouri), and $2.5\%$ fatty acid-free BSA (Sigma-Aldrich, St. Louis, Missouri). 2 µCi of 3H-H2O (Perkin Elmer) was added in the absence or presence of 100 nM insulin (Sigma-Aldrich, St. Louis, Missouri). After 4.5 hours, explants were washed with cold PBS (Sigma-Aldrich, St. Louis, Missouri) three times and frozen in liquid nitrogen. Triglycerides were extracted as described above and $\frac{1}{3}$ of the n-hexane phase was used for triglyceride measurement. For fatty acid extraction, the remaining hexane phase was deacylated with ethanol (Merck, Burlington, Massachusetts):water:saturated KOH (Merck, Burlington, Massachusetts) (20:1:1) at 80 °C for 1 hour, neutralized with formic acid (Sigma-Aldrich, St. Louis, Missouri). Fatty acid was extracted with n-hexane and dried. The incorporation of 3H was counted by a scintillation counter and normalized to tissue weight. ## Ex vivo NEFA release vWAT explants were washed and incubated with low glucose DMEM (Sigma-Aldrich, St. Louis, Missouri) supplemented with $2\%$ fatty acid-free BSA (Sigma-Aldrich, St. Louis, Missouri) and 20 mM HEPES (Thermo Fisher Scientific, Waltham, Massachusetts) for 2 hours. Explants were washed twice with Krebs-Ringer phosphate buffer (0.6 mM Na2HPO4 (Fluka Chemie, Buchs, Switzerland), 0.4 mM NaH2PO4 (Merck, Burlington, Massachusetts), 120 mM NaCl (Sigma-Aldrich, St. Louis, Missouri), 6 mM KCl (Merck, Burlington, Massachusetts), 1 mM CaCl2 (Merck, Burlington, Massachusetts), 1.2 mM MgSO4 (Merck, Burlington, Massachusetts), 70 mM HEPES (Thermo Fisher Scientific, Waltham, Massachusetts), 5 mM glucose (Sigma-Aldrich, St. Louis, Missouri), $2\%$ fatty acid-free BSA (Sigma-Aldrich, St. Louis, Missouri), pH7,4). Explants were treated with DMSO or 10 µM isoproterenol (Sigma-Aldrich, St. Louis, Missouri) in the absence or presence of 100 nM insulin (Sigma-Aldrich, St. Louis, Missouri) for 2 hours. Explants were transferred to chloroform (Sigma-Aldrich, St. Louis, Missouri): methanol (Sigma-Aldrich, St. Louis, Missouri):$100\%$ acetic acid (Merck, Burlington, Massachusetts) (200:100:3) and incubated at 37 °C for 1 hour, and protein concentration was determined by BCA assay (Pierce). Non-esterified fatty acid levels in conditioned media were determined by a colorimetric assay kit (Sigma-Aldrich, St. Louis, Missouri), and normalized with protein amounts in the explants. ## AHA-incorporation vWAT from ND- or HFD-fed mice were immediately incubated in low glucose DMEM (Sigma-Aldrich, St. Louis, Missouri) containing 50 µM azidohomoalanine (AHA, Thermo Fisher Scientific, Waltham, Massachusetts) for 30 min. The tissues were frozen, pulverized, lysed by bio-rupture and clarified by centrifugation at 15,000 g for 15 min twice. 100 µg of protein was used for CLICK reaction (Thermo Fisher Scientific, Waltham, Massachusetts) chemistry following the manufacturer’s instructions. Biotinylated protein was pulled-down after mixing with streptavidin magnetic beads (Thermo Fisher Scientific, Waltham, Massachusetts) for 2 hours at 4 °C. The pulled-down protein was digested with trypsin (Promega, Madison, Wisconsin). The digested peptides were acidified using $5\%$ TFA (Thermo Fisher Scientific, Waltham, Massachusetts) and desalted using C18 columns (Waters, Milford, Massachusetts). The eluted peptides were dried and analyzed by mass spectrometry. ## Metabolites measurement Plasma insulin levels were measured by ultrasensitive mouse insulin ELISA kit (Crystal Chem, Downers Grove, Illinois) according to the manufacturer’s instructions. Plasma Leptin levels were measured by mouse Leptin ELISA kit (Crystal Chem, Downers Grove, Illinois) according to the manufacturer’s instructions. Plasma Adiponectin levels were measured by mouse Adiponectin ELISA kit (Crystal Chem, Downers Grove, Illinois) according to the manufacturer’s instructions. Hepatic triglyceride levels were measured using a triglyceride assay kit (Abcam, Cambridge, United Kingdom) according to the manufacturer’s instructions. Plasma triglyceride and cholesterol levels were measured by a biochemical analyzer (Cobas c III analyser, Roche Diagnostics, Indianapolis, Indiana). Plasma NEFA levels were measured by colorimetric NEFA (Sigma-Aldrich, St. Louis, Missouri). ## Body composition measurement Body composition was measured by nuclear magnetic resonance imaging (Echo Medical Systems, Houston, Texas). ## Study approval All animal experiments were performed in accordance with federal guidelines for animal experimentation and were approved by the Kantonales Veterinäramt of the Kanton Basel-Stadt (#31986–3040) or KU Leuven animal ethical committee (#$\frac{206}{2020}$). For human biopsies, the study protocol was approved by the Ethikkomission Nordwest- und Zentralschweiz (EKNZ, BASEC 2016–01040). ## Statistics Sample size was chosen according to our previous studies and published reports in which similar experimental procedures were described. The investigators were not blinded to the treatment groups except for the hyperinsulinemic-euglycemic clamp study. All data are shown as the mean ± SEM. Sample numbers are indicated in each figure legend. For mouse experiments, n represents the number of animals, and for cell culture experiments, N indicates the number of independent experiments. To determine the statistical significance between 2 groups, an unpaired two-tailed Student’s t test, Mann-Whitney test, or multiple t test was performed. For more than three groups, one-way ANOVA was performed. For ITT, GTT, PTT, glucose infusion rate, weigh curve data, two-way ANOVA was performed. All statistical analyses were performed using GraphPad Prism 9 (GraphPad Software, San Diego, California). A p value of less than 0.05 was considered statistically significant. ## Funding Information This paper was supported by the following grants: ## Data availability All gel images and numerical data are uploaded as source data. ## References 1. Abel ED, Peroni O, Kim JK, Kim YB, Boss O, Hadro E, Minnemann T, Shulman GI, Kahn BB. Adipose-selective targeting of the GLUT4 gene impairs insulin action in muscle and liver. Nature (2001) 409 729-733. DOI: 10.1038/35055575 2. Ahrné E, Glatter T, Viganò C, Schubert CV, Nigg EA, Schmidt A. Evaluation and improvement of quantification accuracy in isobaric mass tag-based protein quantification experiments. Journal of Proteome Research (2016) 15 2537-2547. DOI: 10.1021/acs.jproteome.6b00066 3. Beg M, Abdullah N, Thowfeik FS, Altorki NK, McGraw TE. Distinct akt phosphorylation states are required for insulin regulated glut4 and glut1-mediated glucose uptake. eLife (2017) 6. DOI: 10.7554/eLife.26896 4. Blaszkiewicz M, Willows JW, Dubois AL, Waible S, DiBello K, Lyons LL, Johnson CP, Paradie E, Banks N, Motyl K, Michael M, Harrison B, Townsend KL. Neuropathy and neural plasticity in the subcutaneous white adipose depot. PLOS ONE (2019) 14. DOI: 10.1371/journal.pone.0221766 5. Brown MS, Goldstein JL. Selective versus total insulin resistance: a pathogenic paradox. Cell Metabolism (2008) 7 95-96. DOI: 10.1016/j.cmet.2007.12.009 6. Cybulski N, Polak P, Auwerx J, Rüegg MA, Hall MN. Mtor complex 2 in adipose tissue negatively controls whole-body growth. PNAS (2009) 106 9902-9907. DOI: 10.1073/pnas.0811321106 7. Dentin R, Tomas-Cobos L, Foufelle F, Leopold J, Girard J, Postic C, Ferré P. Glucose 6-phosphate, rather than xylulose 5-phosphate, is required for the activation of ChREBP in response to glucose in the liver. Journal of Hepatology (2012) 56 199-209. DOI: 10.1016/j.jhep.2011.07.019 8. DeWaal D, Nogueira V, Terry AR, Patra KC, Jeon SM, Guzman G, Au J, Long CP, Antoniewicz MR, Hay N. Hexokinase-2 depletion inhibits glycolysis and induces oxidative phosphorylation in hepatocellular carcinoma and sensitizes to metformin. Nature Communications (2018) 9. DOI: 10.1038/s41467-017-02733-4 9. Ducluzeau PH, Perretti N, Laville M, Andreelli F, Vega N, Riou JP, Vidal H. Regulation by insulin of gene expression in human skeletal muscle and adipose tissue: evidence for specific defects in type 2 diabetes. Diabetes (2001) 50 1134-1142. DOI: 10.2337/diabetes.50.5.1134 10. Elia I, Broekaert D, Christen S, Boon R, Radaelli E, Orth MF, Verfaillie C, Grünewald TGP, Fendt SM. Proline metabolism supports metastasis formation and could be inhibited to selectively target metastasizing cancer cells. Nature Communications (2017) 8. DOI: 10.1038/ncomms15267 11. Fishman RB, Dark J. Sensory innervation of white adipose tissue. The American Journal of Physiology (1987) 253 R942-R944. DOI: 10.1152/ajpregu.1987.253.6.R942 12. Frei IC, Weissenberger D, Ritz D, Heusermann W, Colombi M, Shimobayashi M, Hall MN. Adipose mtorc2 is essential for sensory innervation in white adipose tissue and whole-body energy homeostasis. Molecular Metabolism (2022) 65. DOI: 10.1016/j.molmet.2022.101580 13. Gottlob K, Majewski N, Kennedy S, Kandel E, Robey RB, Hay N. Inhibition of early apoptotic events by akt/PKB is dependent on the first committed step of glycolysis and mitochondrial hexokinase. Genes & Development (2001) 15 1406-1418. DOI: 10.1101/gad.889901 14. Heikkinen S, Pietilä M, Halmekytö M, Suppola S, Pirinen E, Deeb SS, Jänne J, Laakso M. Hexokinase II-deficient mice: prenatal death of homozygotes without disturbances in glucose tolerance in heterozygotes. The Journal of Biological Chemistry (1999) 274 22517-22523. DOI: 10.1074/jbc.274.32.22517 15. Herman MA, Peroni OD, Villoria J, Schön MR, Abumrad NA, Blüher M, Klein S, Kahn BB. A novel chrebp isoform in adipose tissue regulates systemic glucose metabolism. Nature (2012) 484 333-338. DOI: 10.1038/nature10986 16. Horton JD, Goldstein JL, Brown MS. SREBPs: activators of the complete program of cholesterol and fatty acid synthesis in the liver. The Journal of Clinical Investigation (2002) 109 1125-1131. DOI: 10.1172/JCI15593 17. Iizuka K, Bruick RK, Liang G, Horton JD, Uyeda K. Deficiency of carbohydrate response element-binding protein (chrebp) reduces lipogenesis as well as glycolysis. PNAS (2004) 101 7281-7286. DOI: 10.1073/pnas.0401516101 18. Jackson RA, Roshania RD, Hawa MI, Sim BM, DiSilvio L. Impact of glucose ingestion on hepatic and peripheral glucose metabolism in man: an analysis based on simultaneous use of the forearm and double isotope techniques. The Journal of Clinical Endocrinology and Metabolism (1986) 63 541-549. DOI: 10.1210/jcem-63-3-541 19. Jiang ZY, Zhou QL, Coleman KA, Chouinard M, Boese Q, Czech MP. Insulin signaling through akt/protein kinase B analyzed by small interfering RNA-mediated gene silencing. PNAS (2003) 100 7569-7574. DOI: 10.1073/pnas.1332633100 20. Kabashima T, Kawaguchi T, Wadzinski BE, Uyeda K. Xylulose 5-phosphate mediates glucose-induced lipogenesis by xylulose 5-phosphate-activated protein phosphatase in rat liver. PNAS (2003) 100 5107-5112. DOI: 10.1073/pnas.0730817100 21. Karczewski KJ, Francioli LC, Tiao G, Cummings BB, Alföldi J, Wang Q, Collins RL, Laricchia KM, Ganna A, Birnbaum DP, Gauthier LD, Brand H, Solomonson M, Watts NA, Rhodes D, Singer-Berk M, England EM, Seaby EG, Kosmicki JA, Walters RK, Tashman K, Farjoun Y, Banks E, Poterba T, Wang A, Seed C, Whiffin N, Chong JX, Samocha KE, Pierce-Hoffman E, Zappala Z, O’Donnell-Luria AH, Minikel EV, Weisburd B, Lek M, Ware JS, Vittal C, Armean IM, Bergelson L, Cibulskis K, Connolly KM, Covarrubias M, Donnelly S, Ferriera S, Gabriel S, Gentry J, Gupta N, Jeandet T, Kaplan D, Llanwarne C, Munshi R, Novod S, Petrillo N, Roazen D, Ruano-Rubio V, Saltzman A, Schleicher M, Soto J, Tibbetts K, Tolonen C, Wade G, Talkowski ME, Neale BM, Daly MJ, MacArthur DG. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature (2020) 581 434-443. DOI: 10.1038/s41586-020-2308-7 22. Katz EB, Stenbit AE, Hatton K, DePinho R, Charron MJ. Cardiac and adipose tissue abnormalities but not diabetes in mice deficient in GLUT4. Nature (1995) 377 151-155. DOI: 10.1038/377151a0 23. Kawaguchi T, Osatomi K, Yamashita H, Kabashima T, Uyeda K. Mechanism for fatty acid “ sparing ” effect on glucose-induced transcription: regulation of carbohydrate-responsive element-binding protein by AMP-activated protein kinase. The Journal of Biological Chemistry (2002) 277 3829-3835. DOI: 10.1074/jbc.M107895200 24. Kim M-S, Krawczyk SA, Doridot L, Fowler AJ, Wang JX, Trauger SA, Noh H-L, Kang HJ, Meissen JK, Blatnik M, Kim JK, Lai M, Herman MA. Chrebp regulates fructose-induced glucose production independently of insulin signaling. The Journal of Clinical Investigation (2016) 126 4372-4386. DOI: 10.1172/JCI81993 25. Kowalski GM, Bruce CR. The regulation of glucose metabolism: implications and considerations for the assessment of glucose homeostasis in rodents. American Journal of Physiology. Endocrinology and Metabolism (2014) 307 E859-E871. DOI: 10.1152/ajpendo.00165.2014 26. Kreier F, Kap YS, Mettenleiter TC, van Heijningen C, van der Vliet J, Kalsbeek A, Sauerwein HP, Fliers E, Romijn JA, Buijs RM. Tracing from fat tissue, liver, and pancreas: a neuroanatomical framework for the role of the brain in type 2 diabetes. Endocrinology (2006) 147 1140-1147. DOI: 10.1210/en.2005-0667 27. Kumar A, Lawrence JC, Jung DY, Ko HJ, Keller SR, Kim JK, Magnuson MA, Harris TE. Fat cell-specific ablation of Rictor in mice impairs insulin-regulated fat cell and whole-body glucose and lipid metabolism. Diabetes (2010) 59 1397-1406. DOI: 10.2337/db09-1061 28. Li MV, Chen W, Harmancey RN, Nuotio-Antar AM, Imamura M, Saha P, Taegtmeyer H, Chan L. Glucose-6-Phosphate mediates activation of the carbohydrate responsive binding protein (ChREBP). Biochemical and Biophysical Research Communications (2010) 395 395-400. DOI: 10.1016/j.bbrc.2010.04.028 29. Makwana K, Chodavarapu H, Morones N, Chi J, Barr W, Novinbakht E, Wang Y, Nguyen PT, Jovanovic P, Cohen P, Riera CE. Sensory neurons expressing calcitonin gene-related peptide α regulate adaptive thermogenesis and diet-induced obesity. Molecular Metabolism (2021) 45. DOI: 10.1016/j.molmet.2021.101161 30. Miyamoto S, Murphy AN, Brown JH. Akt mediates mitochondrial protection in cardiomyocytes through phosphorylation of mitochondrial hexokinase-II. Cell Death and Differentiation (2008) 15 521-529. DOI: 10.1038/sj.cdd.4402285 31. Muñoz S, Franckhauser S, Elias I, Ferré T, Hidalgo A, Monteys AM, Molas M, Cerdán S, Pujol A, Ruberte J, Bosch F. Chronically increased glucose uptake by adipose tissue leads to lactate production and improved insulin sensitivity rather than obesity in the mouse. Diabetologia (2010) 53 2417-2430. DOI: 10.1007/s00125-010-1840-7 32. Nawaz MH, Ferreira JC, Nedyalkova L, Zhu H, Carrasco-López C, Kirmizialtin S, Rabeh WM. The catalytic inactivation of the N-half of human hexokinase 2 and structural and biochemical characterization of its mitochondrial conformation. Bioscience Reports (2018) 38. DOI: 10.1042/BSR20171666 33. Niijima A, Fukuda A. Release of glucose from perfused liver preparation in response to stimulation of the splanchnic nerves in the toad. The Japanese Journal of Physiology (1973) 23 497-508. DOI: 10.2170/jjphysiol.23.497 34. Ortega-Prieto P, Postic C. Carbohydrate sensing through the transcription factor chrebp. Frontiers in Genetics (2019) 10. DOI: 10.3389/fgene.2019.00472 35. Patra KC, Wang Q, Bhaskar PT, Miller L, Wang Z, Wheaton W, Chandel N, Laakso M, Muller WJ, Allen EL, Jha AK, Smolen GA, Clasquin MF, Robey B, Hay N. Hexokinase 2 is required for tumor initiation and maintenance and its systemic deletion is therapeutic in mouse models of cancer. Cancer Cell (2013) 24 213-228. DOI: 10.1016/j.ccr.2013.06.014 36. Perry RJ, Camporez JPG, Kursawe R, Titchenell PM, Zhang D, Perry CJ, Jurczak MJ, Abudukadier A, Han MS, Zhang XM, Ruan HB, Yang X, Caprio S, Kaech SM, Sul HS, Birnbaum MJ, Davis RJ, Cline GW, Petersen KF, Shulman GI. Hepatic acetyl coa links adipose tissue inflammation to hepatic insulin resistance and type 2 diabetes. Cell (2015) 160 745-758. DOI: 10.1016/j.cell.2015.01.012 37. Ranalletta M, Jiang H, Li J, Tsao TS, Stenbit AE, Yokoyama M, Katz EB, Charron MJ. Altered hepatic and muscle substrate utilization provoked by GLUT4 ablation. Diabetes (2005) 54 935-943. DOI: 10.2337/diabetes.54.4.935 38. Riddle MR, Aspiras AC, Gaudenz K, Peuß R, Sung JY, Martineau B, Peavey M, Box AC, Tabin JA, McGaugh S, Borowsky R, Tabin CJ, Rohner N. Insulin resistance in cavefish as an adaptation to a nutrient-limited environment. Nature (2018) 555 647-651. DOI: 10.1038/nature26136 39. Roden M, Shulman GI. The integrative biology of type 2 diabetes. Nature (2019) 576 51-60. DOI: 10.1038/s41586-019-1797-8 40. Sakaguchi M, Fujisaka S, Cai W, Winnay JN, Konishi M, O’Neill BT, Li M, García-Martín R, Takahashi H, Hu J, Kulkarni RN, Kahn CR. Adipocyte dynamics and reversible metabolic syndrome in mice with an inducible adipocyte-specific deletion of the insulin receptor. Cell Metabolism (2017) 25 448-462. DOI: 10.1016/j.cmet.2016.12.008 41. Sassmann A, Offermanns S, Wettschureck N. Tamoxifen-inducible cre-mediated recombination in adipocytes. Genesis (2010) 48 618-625. DOI: 10.1002/dvg.20665 42. Shearin AL, Monks BR, Seale P, Birnbaum MJ. Lack of Akt in adipocytes causes severe lipodystrophy. Molecular Metabolism (2016) 5 472-479. DOI: 10.1016/j.molmet.2016.05.006 43. Shepherd PR, Gnudi L, Tozzo E, Yang H, Leach F, Kahn BB. Adipose cell hyperplasia and enhanced glucose disposal in transgenic mice overexpressing GLUT4 selectively in adipose tissue. The Journal of Biological Chemistry (1993) 268 22243-22246. PMID: 8226728 44. Shimazu T, Fukuda A. Increased activities of glycogenolytic enzymes in liver after splanchnic-nerve stimulation. Science (1965) 150 1607-1608. DOI: 10.1126/science.150.3703.1607 45. Shimobayashi M, Albert V, Woelnerhanssen B, Frei IC, Weissenberger D, Meyer-Gerspach AC, Clement N, Moes S, Colombi M, Meier JA, Swierczynska MM, Jenö P, Beglinger C, Peterli R, Hall MN. Insulin resistance causes inflammation in adipose tissue. The Journal of Clinical Investigation (2018) 128 1538-1550. DOI: 10.1172/JCI96139 46. Smith MA, Katsouri L, Virtue S, Choudhury AI, Vidal-Puig A, Ashford MLJ, Withers DJ. Calcium channel CaV2.3 subunits regulate hepatic glucose production by modulating leptin-induced excitation of arcuate pro-opiomelanocortin neurons. Cell Reports (2018) 25 278-287. DOI: 10.1016/j.celrep.2018.09.024 47. Song CK, Schwartz GJ, Bartness TJ. Anterograde transneuronal viral tract tracing reveals central sensory circuits from white adipose tissue. American Journal of Physiology. Regulatory, Integrative and Comparative Physiology (2009) 296 R501-R511. DOI: 10.1152/ajpregu.90786.2008 48. Stewart SA, Dykxhoorn DM, Palliser D, Mizuno H, Yu EY, An DS, Sabatini DM, Chen ISY, Hahn WC, Sharp PA, Weinberg RA, Novina CD. Lentivirus-delivered stable gene silencing by rnai in primary cells. RNA (2003) 9 493-501. DOI: 10.1261/rna.2192803 49. Sukonina V, Ma H, Zhang W, Bartesaghi S, Subhash S, Heglind M, Foyn H, Betz MJ, Nilsson D, Lidell ME, Naumann J, Haufs-Brusberg S, Palmgren H, Mondal T, Beg M, Jedrychowski MP, Taskén K, Pfeifer A, Peng X-R, Kanduri C, Enerbäck S. Foxk1 and FOXK2 regulate aerobic glycolysis. Nature (2019) 566 279-283. DOI: 10.1038/s41586-019-0900-5 50. Tan S-X, Fisher-Wellman KH, Fazakerley DJ, Ng Y, Pant H, Li J, Meoli CC, Coster ACF, Stöckli J, James DE. Selective insulin resistance in adipocytes. The Journal of Biological Chemistry (2015) 290 11337-11348. DOI: 10.1074/jbc.M114.623686 51. Tang Y, Wallace M, Sanchez-Gurmaches J, Hsiao WY, Li H, Lee PL, Vernia S, Metallo CM, Guertin DA. Adipose tissue mtorc2 regulates chrebp-driven de novo lipogenesis and hepatic glucose metabolism. Nature Communications (2016) 7. DOI: 10.1038/ncomms11365 52. Titchenell PM, Quinn WJ, Lu M, Chu Q, Lu W, Li C, Chen H, Monks BR, Chen J, Rabinowitz JD, Birnbaum MJ. Direct hepatocyte insulin signaling is required for lipogenesis but is dispensable for the suppression of glucose production. Cell Metabolism (2016) 23 1154-1166. DOI: 10.1016/j.cmet.2016.04.022 53. van Gorsel M, Elia I, Fendt SM. 13C tracer analysis and metabolomics in 3D cultured cancer cells. Methods in Molecular Biology (2019) 1862 53-66. DOI: 10.1007/978-1-4939-8769-6_4 54. Vazirani RP, Verma A, Sadacca LA, Buckman MS, Picatoste B, Beg M, Torsitano C, Bruno JH, Patel RT, Simonyte K, Camporez JP, Moreira G, Falcone DJ, Accili D, Elemento O, Shulman GI, Kahn BB, McGraw TE. Disruption of adipose rab10-dependent insulin signaling causes hepatic insulin resistance. Diabetes (2016) 65 1577-1589. DOI: 10.2337/db15-1128 55. Vijayakumar A, Aryal P, Wen J, Syed I, Vazirani RP, Moraes-Vieira PM, Camporez JP, Gallop MR, Perry RJ, Peroni OD, Shulman GI, Saghatelian A, McGraw TE, Kahn BB. Absence of carbohydrate response element binding protein in adipocytes causes systemic insulin resistance and impairs glucose transport. Cell Reports (2017) 21 1021-1035. DOI: 10.1016/j.celrep.2017.09.091 56. Wang Y, Yang F, Gritsenko MA, Wang Y, Clauss T, Liu T, Shen Y, Monroe ME, Lopez-Ferrer D, Reno T, Moore RJ, Klemke RL, Camp DG, Smith RD. Reversed-phase chromatography with multiple fraction concatenation strategy for proteome profiling of human MCF10A cells. Proteomics (2011) 11 2019-2026. DOI: 10.1002/pmic.201000722 57. Wang Y, Leung VH, Zhang Y, Nudell VS, Loud M, Servin-Vences MR, Yang D, Wang K, Moya-Garzon MD, Li VL, Long JZ, Patapoutian A, Ye L. The role of somatosensory innervation of adipose tissues. Nature (2022) 609 569-574. DOI: 10.1038/s41586-022-05137-7 58. Wasserman DH. Four grams of glucose. American Journal of Physiology. Endocrinology and Metabolism (2009) 296 E11-E21. DOI: 10.1152/ajpendo.90563.2008 59. Zebisch K, Voigt V, Wabitsch M, Brandsch M. Protocol for effective differentiation of 3T3-L1 cells to adipocytes. Analytical Biochemistry (2012) 425 88-90. DOI: 10.1016/j.ab.2012.03.005
--- title: Effectiveness of the application of an educational program based on the Health Belief Model (HBM) in Adopting Preventive Behaviors from Self-Medication among Women in Iran. A Randomized Controlled Trial* authors: - Ehsan Movahed - Monireh Rezaee Moradali - Mohammad Saeed Jadgal - Morad Ali Zareipour - Mina Tasouji Azari journal: Investigacion y Educacion en Enfermeria year: 2023 pmcid: PMC10017128 doi: 10.17533/udea.iee.v40n3e11 license: CC BY 4.0 --- # Effectiveness of the application of an educational program based on the Health Belief Model (HBM) in Adopting Preventive Behaviors from Self-Medication among Women in Iran. A Randomized Controlled Trial* ## Abstract ### Objective. To evaluate the effectiveness of the application of an educational program based on the Health Belief Model (HBM) in Adopting Preventive Behaviors from Self-Medication among Women in Iran. ### Methods. Interventional study with pre and post phases. 200 women referring to the health centers of Urmia were selected by simple random sampling, divided into two groups of treatment and control. Data collection instruments were researcher-devised questionnaire including the questionnaire of Knowledge of Self-medication, the Questionnaire of Preventive Behaviors from Self-medication, and the questionnaire of Health Belief Model. The questionnaires were assessed for expert validity and then, were checked for reliability. The educational intervention was conducted for the treatment group during four weeks four 45-minute sessions. ### Results. The average scores of knowledge, perceived susceptibility, perceived severity, perceived benefits, perceived barriers, cues to action, self-efficiency, and post-intervention performance in have increased in treatment group, comparing to the control group, All findings were statistically significant ($p \leq 0.05$). Furthermore, social media, doctors, and disbelief in self-medication were more effective in increasing awareness and encouraging to have proper medication, also, the highest self-medication was in taking pain-relievers, cold tablets and antibiotics, which showed significant decrease in treatment group after the intervention. ### Conclusion. The educational program based on Health Belief Model was effective in reducing the self-medication among the studied women. Furthermore, it is recommended to use social media and doctors to improve the awareness and motivation among people. Thus, applying the educational programs and plans according to the Health Belief Model can be influential in reducing the self-medication. ## Objetivo. Evaluar la eficacia de la aplicación de un programa educativo basado en el Modelo de Creencias sobre la Salud (MCS) en la adopción de conductas preventivas de la automedicación entre las mujeres de Irán. Avaliar a eficácia da aplicação de um programa educativo baseado no Modelo de Crenças em Saúde (HCM) na adoção de comportamentos preventivos de automedicação entre mulheres no Irã. ## Métodos. Estudio de intervención con evaluación pre y post. Se seleccionaron 200 mujeres que acudieron a los centros de salud de Urmia, a quienes se asignaron a los dos grupos de estudio (tratamiento y control) mediante un muestreo aleatorio simple. Para la recolección de la información se utilizaron los cuestionarios sobre: Conocimientos acerca de la automedicación, conductas preventivas de la automedicación y el modelo de creencias sobre la salud. Se evaluó la validez de los cuestionarios por parte de los expertos y luego se comprobó su confiabilidad. La intervención educativa se llevó a cabo para el grupo de tratamiento durante cuatro semanas con 1 sesión semanal de 45 minutos de duración. Estudo de intervenção com pré e pós avaliação. Duzentas mulheres que frequentavam os centros de saúde de Urmia foram selecionadas e alocadas nos dois grupos de estudo (tratamento e controle) por meio de amostragem aleatória simples. Para a coleta de informações, foram utilizados os questionários sobre: Conhecimento sobre automedicação, comportamentos preventivos de automedicação e o modelo de crenças sobre saúde. A validade dos questionários foi avaliada pelos especialistas e, em seguida, verificada sua confiabilidade. A intervenção educativa foi realizada para o grupo de tratamento durante quatro semanas com 1 sessão semanal com duração de 45 minutos. ## Resultados. Las puntuaciones medias de los conocimientos, la susceptibilidad percibida, la gravedad percibida, los beneficios percibidos, las barreras percibidas, las señales para la acción, la autoeficacia y el rendimiento posterior a la intervención aumentaron en el grupo de tratamiento en comparación con el grupo de control, y todos los resultados fueron estadísticamente significativos ($p \leq 0.05$). Además, los medios de comunicación social fueron eficaces para aumentar la concienciación y animar a tener una medicación adecuada. La mayor automedicación fue en la toma de analgésicos, pastillas para el resfriado y antibióticos, que mostró una disminución significativa en el grupo de tratamiento después de la intervención. Os escores médios de conhecimento, suscetibilidade percebida, gravidade percebida, benefícios percebidos, barreiras percebidas, pistas para ação, autoeficácia e desempenho pós-intervenção aumentaram no grupo de tratamento em comparação com o grupo de intervenção. controle, e todos os resultados foram estatisticamente significativos ($p \leq 0.05$). Além disso, as mídias sociais foram eficazes na conscientização e no incentivo à medicação adequada. A maior automedicação foi em uso de analgésicos, antissépticos e antibióticos, que apresentou diminuição significativa no grupo de tratamento após a intervenção. ## Conclusión. El programa educativo basado en el Modelo de Creencias de Salud fue eficaz para reducir la automedicación entre las mujeres estudiadas. Además, se recomienda utilizar los medios de comunicación social para mejorar la concienciación y la motivación de las personas. ## Conclusão. O programa educativo baseado no Modelo de Crenças em Saúde foi eficaz na redução da automedicação entre as mulheres estudadas. Além disso, recomenda-se o uso das mídias sociais para melhorar a conscientização e a motivação das pessoas. ## Introduction Healthy human is regarded as the basis of sustainable development, in which the role of medication has proven to be primary, effective and determinant.[1] As the most common form of self-care, self-medication is defined as taking medication without a doctor's prescription, using medications prescribed for other family members, refusing to take the original prescribed medication, and overusing over-the-counter under-medication medications.[2,3] The consequences of Self-medication may include complexities like disturbance in drug market, very high expenses of medicine budgeting of the government, the delay in treatment of a sever disease, development of drug resistance, no optimal treatment, poisoning, unwanted consequences and eventually may leading to death.[1] These days, according to Panchal et al., arbitrary drug use and self-medication in general, is considered as one of the biggest social, economic and health problems of different societies including Iran.[4] In the World Health Day Slogan in 2011 was declared the resistance to anti-bacterial drugs as a global threat. During a study on Italian families, $69\%$ had arbitrary drug use at least once [6], and in the study done by Pavan, $5\%$ of people had experienced arbitrary drug use.[5] The amount and range of self-medication is different in different cities of the country, so that it is reported as $94\%$ in Ahvaz,[7] $63\%$ in Tabriz,[8] $86\%$ in Isfahan,[9] $54\%$ in Arak,[10] and $83\%$ in Yazd.[11] In the meantime, paying attention to the population of women has great importance, as their experiencing some critical periods like pregnancy and breast feeding, more being in contact with family members, and being regarded as role models and examples for other family members. Different studies show that women show particular tendency to have self-medication, so arbitrarily and frequently use drugs to cure problems like dysmenorrhea, to eliminate symptoms of menopause, menstrual disorders, mood disorders, as well as the problems occur in pregnancy and breast feeding, all can bring about self-medication among women.[12] The Health Belief Model (HBM) was one of the first behavior change models to explain health decision-making and the consequences of behavior, which was proposed by social psychologists in the 50s to explain people's desire to adopt preventive behaviors. After making corrections and adding new structures to this model, it was used to identify people's behavior in the field of disease prevention, screening and control.[13] The constructs of the HBM consist of perceived susceptibility, perceived severity, perceived benefits, perceived barriers, self-efficacy, and cues to action. According to this model, a person's decision to adopt a behavior depends on the person's perception and how much they consider themselves to be at risk and prone to disease (perceived susceptibility); then understand the depth of clinical, medical, and social consequences (perceived severity); With the cues and stimuli they receives (cues to action), they believe in the benefits and applicability of preventive behavior (perceived benefits) and finds the factors that prevent the behavior to be low-cost (perceived barriers) and sees himself as capable of performing the preventive behavior (perceived self-efficacy) to finally choose the correct behavior.[13] In this vein, the findings of the studies on applying the HBM model aiming at increasing the physical activity, adopting preventive behaviors from Alzheimer, prove the health promoting self-care behaviors manifesting their efficiency.[14,15] Regarding the ever-increasing widespread occurrence of self-medication in communities, and the direct role of the individual in drug selection and use in order to have a longer life with a fairly healthy and active living, it seems urgent to determine the effective factors on. Therefore, the purpose of the present study is to evaluate the effectiveness of the application of an educational program based on the Health Belief Model (HBM) in Adopting Preventive Behaviors from Self-Medication among Women in Iran. ## Methods The present study is an interventional and semi-experimental one, conducted among women in Urmia in 1398 [2018] with the purpose of applying Health Belief Model (HBM) in adopting preventive behaviors from self-medication of women. Having regarded the previously conducted studies, the prevalence of self-medication was estimated to be $36\%$,[16] which has been calculated with α=$5\%$, $95\%$ confidence level, and $d = 6$%, on the sample size of 200 people, of which taking 100 for each groups of control and intervention. In the formula, $$p \leq 36$$%, $q = 64$%, $d = 6$% and $z = 1.96.$ The sampling was the stratified sampling way performed by referring to the health centers of the city. The samples were selected randomly from women referred to 10 defined health centers, then every one of the centers were placed in the control and intervention groups (5 control centers and 5 intervention centers). The sampling in the clinics was done randomly according to the household codes. The way of sampling in each clinic was simple and random as well, according to the code of the household available at the centers, as these samples were invited to a meeting on a definite day to the health center with the objective of getting acquainted, getting informed of the purpose of the study, as well as receiving the written informed agreement and consent of participation in the study. Inclusion criteria included consent to participate in the study, having a health record at the health center, and not having a specific disease. Exclusion criteria were women who were unable to cooperate. The whole population of people taking part in the present study accounted for 200 people, of whom 18 people didn’t fill in the questionnaire, thus more women were added to the study to have 200 accomplished questionnaires at the end. The data collection instruments include researcher-devised questionnaire including the questionnaire of knowledge of self-medication, the questionnaire of preventive behaviors from self-medication, and the questionnaire of Health Belief Model. The items relevant to each indicated consequences were selected according to the literature review, and to meet the validity and reliability of the questionnaire, the approaches of content validity and Alpha-Cronbach Test were implemented respectively. To measure the validity of the questionnaire, it was sent to 10 experts in the health education and gynecologists, then the necessary modifications were applied on the basis of their comments. The validity was measured to be above $80\%$. To assess the reliability of the questionnaire, it was responded by 30 women, then by using Cronbach Alpha Test, the reliability coefficient of the questions of Knowledge was set as $0.77\%$, Perceived Sensitivity $0.75\%$, Perceived Severity $0.82\%$, Perceived Barriers $0.86\%$, Cues to Action $0.81\%$, Perceived Benefits $0.84\%$, and Self-efficiency as $0.76\%$. Assessing Knowledge was done in the form of 12 questions with Yes/No/Don’t Know options, as the respond of ‘Yes’ receiving 2 points, ‘Don’t Know’ getting 1 point, and ‘No’ having zero. The scores of Knowledge Questionnaire varied from zero to 24. Next, the questions and scores of the Health Belief Model include all questions on the basis of Likert scale with three options of “Agree, No Idea, and Disagree”. Because of the more complex understanding of the area being assessed, the ‘Perceived Sensitivity’ had 5 questions with maximum and minimum of 5-15 points, ‘Perceived Severity’ had 6 questions with 6-18 points, ‘Perceived Barriers” had 5 questions 5-15 points, ‘Perceived Benefits’ having 7 questions with 7-21 points, ‘Self-efficiency’ having 10 questions 10-30 points, ‘Cues for Action’ having 7 questions with 7-21 points. To measure the preventive behaviors from self-medication, 10 questions were applying as well, with the options of ‘Always Yes’, ‘Sometimes Yes’, and ‘No’, as the response of ‘Always Yes’ receiving 2 points, ‘Sometimes Yes’ with 1 point and ‘No’ having zero point. The scores of the questionnaire of the preventive behaviors from self-medication varied from zero to 20. The expected intervention was performed according to the Health Belief Model for the intervention group, including 4 educational sessions for 45 minutes. As the number of the participants in the control group was 100 people, the educational classes were divided into 20 people classes, having 4 educational sessions on the basis of Health Belief Model. The implemented educational methods were lecturing, asking and answering questions, and group discussion, while in order to assist women’s better understanding, to prevent misunderstanding, and to involve their visual learning as its critical importance, other educational equipment and materials like posters, educational pamphlets, booklets and whiteboard were implemented as well. The educational content and materials were prepared on the basis of educational goals, the needs-analysis conducted earlier, and regarding the valid books and pamphlets available, as well as the pharmaceutical monthly magazine titled “Razi”, and counselling the pharmaceutical specialists. The first session consisted the history of self-medication, and improving the knowledge of people on self-medication, the next sessions were on the basis of Health Belief Model, including education to raise the severity, self-efficiency, benefits and barriers on the issue, together with external and internal cues to action regarding self-medication or arbitrary use of drugs. It is worthy of noticing that the presented materials were written in a pamphlet and educational manuals submitted to the participants, then were reviewed and summarized the following session briefly. Table1. Educationalcontent on the basis of Health Belief Model regarding self-medicationSessions(each 45-min)The title of educational subjectEducational contentSessions(each 45-min)Session 1KnowledgeKnowing about self-medication, self-medication among women, prevalence of self-medication among women, the causes and factors of self-medication, reasons and factors leading to self-medication, preventing ways for self-medicationSession 1Session2Perceived susceptibilityStating the prevalence frequency of self-medication among women, created physiologic changes in women, increasing the possibility of drug-resistance, feeling the menace of being exposure to disease, feeling the risk of self-medication, feeling the need to modify medicationSession2 Perceived severityStating the consequences of self-medication in the physical, psychological, social, and economic aspects, not being able to do the assigned responsibilities after self-medication, cost of medication and hospitalization afterwards, the way of occurrence long-term and short-term complications, highlighting the severity of the consequences of self-medication Session 3Perceived benefitsIdentifying the benefits of not doing self-medication, expressing the positive effects of preventing self-medication, on each mentioned consequence of self-medication (reducing the medication expenses, preventing the harm and damage, preventing having some disabilities, not being dependent on others, ability to participate in social events, preventing from staying home, the ability to do every day and recreational activitiesSession 3 Perceived barriersTrying to persuade the reduction in perceived medication expenses and not self-medication, lectures, discussion and exchange comments on perceived expenses of self-medication with the group members, providing some educational solutions to minimize the perceived expenses, counceling and discussing with the heads of the families on the perceived expenses, persuading them to have enough time to visit the doctor and medication according to the doctor’s prescription Session 4Perceived self-efficacyDefining the meaning of the self-efficiency and its significance in preventing self-medication, verbal encouragement to promote the feeling of efficiency, using the ways to enhance self-efficiency including simplifying the behavior, others’ experience on modifying the self-medication, providing ways to control time and stress and its significance in promoting the sense of efficiency and preventing self-medicationSession 4 Cues to actionActing according to the advice of the health experts to prevent self-medication, listening to the advice of the husband and relatives to prevent self-medication, listening to the advice of the peers to prevent self-medication. In this session, it was asked the people attending in the session with the experience of self-medication to express their experience and consequences of self-medication. After the educational intervention, the phone numbers of the participants or their relatives were recorded to being followed up. They were followed up for three months via phone calls, and eventually, after three months, the questionnaires were distributed and the data for both groups were collected again. The Independent T-Test was used to compare the research units on the basis of demographic data of both intervention and control groups. Furthermore, regarding the inference data, statistical T-Test, Paired T-Test, or their non-parametric equivalents like Mann-Whitney Test and Wilcoxon were used to compare the control and intervention groups before and after intervention. ## Results In this study, 200 individuals were examined in two groups of intervention (100 people), and control (100 people). The mean age and the standard deviation of the age of the participants of the present study in the control and intervention groups were 27.45±12.36 and 26.51±11.46 respectively. Using Independent T-Test showed there is no significant difference between the control and intervention groups in terms of age, marital status, education, career, health insurance, and financial status (Table 2). Table 2Demographic features of women in the Control and Intervention groupsVariableIntervention group n (%)Control group n (%) p-valueAge 0.220-2430 [30]32 [32] 25-3937 [37]35 [35] 30-3423 [23]20 [20] Above3510 [10]13 [13] Marital Status 0.3Married71 [71]70 [70] Unmarried 11 [11]14 [14] Widowed 18 [18]16 [16] Education 0.4Illiterate elementary22 [22]19 [19] Guidance40 [40]45 [45] Diploma and above38 [38]36 [36] Career 0.2Housewife 74 [74]70 [70] Working26 [26]30 [30] Insurance 0.1Yes81 [81]83 [83] No19 [19]17 [17] Financial Status 0.08Weak 22 [22]28 [28] Average54 [54]52 [52] Good 24 [24]20 [20] The findings of Wilcoxon test show that the mean scores of knowledge, perceived sensitivity, perceived severity, perceived self-efficiency, perceived benefits, perceived barriers, cues to action and action are statistically meaningful in the intervention group after the intervention ($p \leq 0.05$). As the mean scores of these variables has been increased while the results of this test haven’t shown any significant differences in the control group before and after the intervention ($p \leq 0.05$). The results of the Mann-Whitney proved the lack of any significant difference between control and intervention groups before the intervention, while the difference was meaningful (Table 3). Table 3Comparing mean score and Standard Deviation of the variables being studied in Control and Intervention groupsStructure ModelGroupPre-interventionPost-intervention p_value * M±SDM±SD KnowledgeIntervention15.82±3.7020.94±2.60p<0.01 Control14.53±3.2415.34±3.410.03 p_value0.57p<0.01 Perceived susceptibilityIntervention8.65±3.3913.95±3.11p<0.01 Control8.52±3.529.02±3.580.09 p_value0.83p<0.001 Perceived severityIntervention9.52±3.3114.2±2.77p<0.001 Control9.48±2.8210.04±3.220.08 p_value0.37P<0.001 Perceived self-efficacyIntervention15.48±3.4822.18±3.53p<0.001 Control15.1±3.5115.84±3.510.2 p_value0.16p<0.001 Perceived benefitsIntervention12.29±3.3517.56±2.83p<0.001 Control12.40±3.2413.04±3.790.06 p_value0.12p<0.001 Perceived barriersIntervention11.39±3.478.66±3.73p<0.001 Control11.54±3.3510.89±3.870.14 p_value0.18p<0.001 Cues to actionIntervention14.49±3.5719.16±3.23p<0.001 Control15.44±3.5516.09±3.470.05 p_value0.09p<0.001 Action on self-medicationIntervention18.52±3.9914.71±2.31p<0.001 Control15.76±3.414.34±2.90.01 p_value0.3p<0.001 () Wilcoxon test, () Mann-Whitney In the present study, regarding the variable of “External cues to action”, for both groups of control and intervention, social media and the physician had the biggest role in receiving the self-medication of women. On the other hand, “interior cues to action” which encourages the individual to take medicine properly, and non-belief into self-medication ($47\%$), had the highest role in both groups of control and intervention (Table 4). Table 4The frequency distribution of internal and external cues to actionExterior cues to actionIntervention Control n%n%RV and radio24243131Book and pamphlet45454040Physician60606565Family and relatives49494545Other mothers referring to health centers 131399Social media65657171Interior cues to action Fear of the consequences of self- medication 37373232Disbelief in self-medication47474646Favorable general condition 31312828Feeling more healthy in self-medication42423535 The findings of the women’s performance on taking different medicines in the control and medication groups, before and after the intervention have been shown in percentage. The findings revealed that most drugs women taken through self-medication before the treatment were pain-relievers, cold tablets, and antibiotics, which have been reduced significantly in the self-medication of intervention group after the educational intervention (Table 5). Table 5the frequency distribution of the self-medication in Intervention and Control groupsType of the medicationIntervention group Control group Before AfterBefore After n (%)n (%) n (%) n (%)Pain relievers61 [61]27 [27]62 [62]60 [60]Cold tablets53 [53]27 [27]55 [55]50 [50]Antibiotics 49 [49]22 [22]48 [48]39 [39]Folic Acid36 [36]16 [16]37 [37]34 [34]Acetaminophen 37 [37]31 [31]38 [38]34 [34]Iron tablet29 [29]23 [23]28 [28]26 [26]Multi-vitamins18 [18]8 [8]20 [20]19 [19]Herbal medicines28 [28]11 [11]27 [27]26 [26]Antihistamine 16 [16]12 [12]14 [14]13 [13]Antacid 13 [13]9 [9]12 [12]11 [11]Sleeping pill10 [10]8 [8]11 [11]9 [9]Anti-nausea pill12 [12]9 [9]13 [13]12 [12]Blood pressure pill9 [9]6 [6]10 [10]9 [9]Antipyretic pill8 [8]6 [6]9 [9]8 [8] ## Discussion The findings revealed that all components of Health Belief Model had positive meaningful changes after the intervention, being proper indicator of self-medication among women. Furthermore, the social media, physicians and non-belief in self-medication had the biggest role in increasing the knowledge and encouraging to proper use of drugs. Eventually, the findings revealed significant decrease in self-medication of intervention group. In the present study, the knowledge of women has been increased by educating the intervention group. The reason of the difference can be the knowledge of women has been increased by educating the intervention group, which has been in line with what Masoudi Alavi and colleagues,[18] Beijani and colleagues,[19] Xiaosheng Lei.[20] However, there has seen no significant difference in the knowledge scores of the students after the intervention in the study done by Movahed and colleagues.[1] The reason of the difference can be related to the held educational sessions on arbitrary use of the drugs by using different media such as poster, pamphlet, speeches, and slides. Also, the gender and age groups can be the reasons of incompatibility of studies. Therefore, it seems education would be beneficial in modifying people behavior on self-medication. The perceived sensitivity of women has been increased after the intervention, which is in line with the studies one by Moghadam,[20] Nikbakht,[21] and Kouhpaye[22] as well. Observing the meaningful significant difference between both groups after the educational intervention in several studies can be prove the importance and effect of the educational intervention on improving the perceived severity of pregnant women in the intervention group, as most mothers, after the intervention, believed that they might had experienced self-medication as well. After the intervention, the mean score of the perceived severity on the consequences of the arbitrary use of drugs has increased significantly in the intervention group. This growing perceived severity has been claimed in other studies like the study done on high school boy students in Manojan,[1] and the study done by Niksadat and colleagues,[21] Beijani and colleagues,[18] as well. In the present study, warning on the serious and sever consequences of the self-medication and drawing people’s attention on loss of health and high treatment expenses have been the two key factors in improving the level of perceived severity of the sample being studied. In the present study, showing the videos of people suffering from consequences of self-medication, and other scenarios prepared to highlight the seriousness of these consequences, and drawing the attention of the participants to the health loss, occurrence of other diseases, and high treatment expenses result in improving the level of the perceived severity of the sample of the participants in this regard. In the present study, after the educational intervention, the mean score of the perceived benefits has increased for the intervention group, in line with other studies.[18,23] However, it was not in line with the study done by Movahhed and colleagues,[1] Bakhtiar and colleagues,[24] Torshizi and colleagues,[25] The reason of the difference may lie in the fact that there should be adopted proper educational medium in education, regarding the cultural and social differences of the cases being studied. If people noticed the fact that proper use of the medication can reduce the side effects and accelerate the recovery, their perception on the medicine would increase and they would take them properly according to the instructions, while this would never come true but by using the proper educational methods and various media to express the benefits of appropriate manner. In the present study, the mean score of the perceived barriers has been reduced after the intervention. Moreover, Bakhtiar and colleagues name the perceived barriers as a strong predictor of the self-medication,[25] while Vahedian-Shahroodi and colleagues stated that the perceived barriers can also predict the behaviors relevant to the Calcium intake. In the study done by Shaghaghi and colleagues, the health care costs, lack of adequate time to refer to the doctor, no accessibility of doctors are regarded as the basic barriers of proper use of the medication. Additionally, the present study was similar to the previous ones, necessitates the importance of planning to decrease the barriers.[26] It seems that the perceived barriers are one of the main components of the Health Belief Model, of which the importance has been elaborated in previous studies, and the proper behavior accelerated by its reduction. In the present study, the social net workings, the doctors and non-belief in the self-medication as the cues for action had the biggest role in improving the knowledge and encouraging to proper use of the drugs. Also, the study of Patrica[27] was in line with this study, introduced doctors and books as the most important exterior cues to action, but the fear of the consequences of the drugs was regarded as the interior cue to action, which was in contrast with the findings of the present study. However, in the study done by Jalilian and colleagues,[28] previous medication, similar prescription and improved symptoms were the greatest reasons of the self-medication among the participants. In the study of Xiaosheng Lei and colleagues the findings were incompatible with the findings of this study, as the relatives’ and friends’ recommendations, Internet, papers and magazines were regarded as the cues to action.[19] In the study done by Movahhed and colleagues,[1] more than half of the participants get the medication information from the doctors, read the drug labels, and a few of them introduced TV, magazines and friends as the reference of getting information on appropriate use of drugs. Furthermore, it is in contrast with the study done in Pakistan, through which, $48\%$ of the participants proposed family as the main source of information on the medication.[27] In the study of Gharouni and colleagues[29] it is revealed that $60\%$ of the patients didn’t read the medicine brochures at all. It is recommended that the doctors have been implemented as a strong leverage in the educational programs, family and friends should be taken as the appropriate guides and supports as well. Regarding the interior cues to action, it is also highlighted the attention to the perceived threats and the harms caused by the arbitrary usage of the drugs. The findings of the present study revealed that the self-efficiency has increased after the educational intervention, which is in line with the similar studies.[21,30] Paying the special attention to the self-efficiency in the previous studies indicates its significance in the process of education. In the present study, findings revealed that the women performance in terms of arbitrary use of medication has been decreased. The similar results were also shown in the studies done by Shamsi,[24] and Hosseini [32] and Izadirad[33] by a reduction in self-medication. The results of the present study showed that the most drugs arbitrarily used by women before the educational self-medication were pain-relievers, antibiotics and cold tablets. However, in the studied of Jalilian and colleagues,[30] the painkillers, antibiotics, anti-cough and adult tablets were the mostly-used drugs in self-medication. In the study of Bakhtiar,[24] the diseases of headache ($77.3\%$), pain relievers ($76.5\%$), and the cold ($62.1\%$) had the highest amount of the self-medication. The findings of the present study showed that amount of using the antibiotics and cold tablets are high in the self-medication, thus it is recommended to have monitoring the drugstores to prevent from selling the drugs not prescribed and to educate their proper usage according to the prescription. Additionally, the satisfactory and effective results would be achieved by paying attention to the reasons of the self-medication and taking them in the educational programs in terms of arbitrary drug usage. ## Conclusion The conclusion of this study is that educating on the basis of the Health Belief Model was effective in improving the performance of women referring to the health centers in terms of prevention from the arbitrary use of the drugs. Therefore, it is recommended to have educational intervention by adopting the models of health education, especially, Health Belief Model in preventing and reducing the arbitrary use of the drugs, leading to the improvement in the health behavior of self-medication. Considering the positive effect of training based on the HBM model in preventing self-treatment, the special role of nurses in training and promoting self-treatment literacy based on the model seems necessary. Using this method, nurses can be effective in reducing adverse outcomes and improving women's health. It is recommended to use this educational model as a part of nurses' activities to reduce the problems of hospitalized and treated women. Strengths and limitations. Implementing the educational model and the type of the study can be regarded as the strengths of the present study. However, the study had limitations including the self-report instrument and limited place, so it ought to be conducted in other settings and places. Additionally, it is recommended to do the research study having the interviews as the data collection instrument. Lack of facilities and supplementary materials, as well as the coordination procedures were regarded as the complexities of the study. ## References 1. Movahed E, Arefi Z. The effect of health belief model-based training (HBM) on self-medication among the male high school students. Iran J. Health Educ. Health Promot.* (2014) 2 65-72 2. Rezaei Jaberee S, Aghamolaei T, Mohseni S, Eslami H, Hassani L. Adopting Self-Medication Prevention Behaviors According to Health Belief Model Constructs. Hormozgan Med. J. (2020) 24 3. Moradali Zareipour ZMR, Jafari Farzaneh, Ghaderzadh Sheiyda, Moradali Monireh Rezaee. Determinants of Self-Medication Prevention in Women based on the Health Belief Model in Urmia City, Iran.. Int. J. Pharm. Res (2020) 12 908-914 4. Javadzade H, Mahmoodi M, Sharifirad G, Fakhraee M, Reisi M. Investigation of psychological factors based on health belief model and health literacy on adult Self-Medication in Bushehr province.. J. Health Lit (2020) 5 39-49 5. Panchal PJ, Pandya AS, Parmar MR. Knowledge, attitude and practice of self medication among under graduate MBBS students at tertiary care teaching hospital. J. Health Sci. Res. (2015) 5 192-197 6. Garofalo L, Di Giuseppe G, Angelillo IF. Self-medication practices among parents in Italy. BioMed Res. Int (2015) 580650-580650. PMID: 25688359 7. Behroozpour A, Shams M, Mousavi M, Ostovar R. Reducing self-medication in Iranian women based on health belief model: A brief report. Shiraz E. Med. J (2021) 22 8. Jeihooni AK, Jormand H, Ansari M, Harsini PA, Rakhshani T. The effect of educational intervention based on health belief model and social support on testicular self-examination in sample of Iranian men. BMC Cancer (2021) 685-685. PMID: 34112094 9. Sharifirad G, Pirzadeh A, Azadbakht L. Knowledge and practice in association with self-medication of nutrient supplements, herbal and chemical pills among women based on Health Belief Model. J. Res. Med. Sci (2011) 16 852-852. PMID: 22091318 10. Tajik R, Shamsi M, Beygee AM. Survey Prevalence of Self Medication and Factors Effected in Woman’s Arak City. Sci. J. Hamadan Nurs. Midwifery Fac (2008) 16 29-39 11. Chuwa BB, Njau LA, Msigwa KI, Shao E. Prevalence and factors associated with self medication with antibiotics among University students in Moshi Kilimanjaro Tanzania. Afr. Health Sci (2021) 21 633-639. PMID: 34795717 12. Sharifi-Rad G, Mohebi S, Motalebi M.. Prevalence of self and modifiable factors affecting the health belief model in elderly. Gonabad. J. Health Sys. Res (2011) 7 13. Hosseinalipour SA, Mohammadbeigi A, Rahbar A, Mohebi S. The Impact of Educational Intervention Based on Extended Health Belief Model with Social Support on Promoting Self-care Behaviors in Patients with Smear Positive Pulmonary TB. Qom Univ. Med. Sci. J (2021) 15 312-321 14. Ghanbary M K, Shamsi M, Khorsandi M, Farazi A, Ranjbaran M, Eshrati B. Effect of training with teaching methods designed based on health belief model on knowledge and self-efficacy in nurses on the disciplines standard precautions in hospitals. J. Hum. Health (2015) 1 51-51 15. Khavoshi N, Tol A, Shojaeizade D, Shamshiri A. Effect of educational intervention on the lifestyle of elderly people referred to clinical centers of Eslamshahr, Iran: application of health belief model. J. Nurs. Educ (2015) 3 19-28 16. Shamsi M TR, Mohammad Beigi A. Study of Arbitrary Drug Use among mothers referring to health centers of Arak. Scientific J. Hamadan Univ. Med. Sci (2008) 16 29-39 17. Masoudi Alavi N, Izadi F, Ebadi A, Hajbagheri A. Metabolism. Self treatment experience in diabetes mellitus type 2. Iran. J. Endocrin. Metab (2019) 10 581-588 18. Bijani M, Haghshenas A, Ghasemi A. Evaluation of the effect of education based on health belief model on self-therapy and self-medication in students at fasa medical sciences dormitories. Int. J. Pharm. Res (2019) 11 1732-1739 19. Lei X, Jiang H, Liu C, Ferrier A, Mugavin J. Self-medication practice and associated factors among residents in Wuhan, China. Int. J. Environ. Res Public Health (2018) 15 68-68. PMID: 29300318 20. Moghadam SMK. Effect of education based on health belief model to prevent the arbitrary use of the drug in women referring to Health Centers sabzevar city. Health Syst. Res (2017) 9 1876-1888 21. Niksadat N, Solhi M, Shojaezadeh D, Gohari M. Investigating the effect of education based on health belief model on improving the preventive behaviors of self-medication in the women under the supervision of health institutions of zone 3 of Tehran. Razi J Med. Sci (2013) 20 48-59 22. Kouhpayeh A, Jeihooni AK, Kashfi SH, Bahmandoost M. Effect of an educational intervention based on the model of health beliefs in self-medication of Iranian mothers. Invest. Educ. Enferm (2017) 35 59-68. PMID: 29767924 23. Ghodsi H, Mokhtari Lake N, Asiri S, Kazem Nezhad Leili E. Prevalence and correlates of cigarette smoking among male students of Guilan‎ University of Medical Sciences. J. Holistic. Nurs. Midwifery (2018) 22 38-43 24. Shamsi M, Hidarnia A, Niknami S, Rafiee M, Zareban I, Karimy M. The effect of educational program on increasing oral health behavior among pregnant women: Applying health belief model. Health Educ. Health Promot (2013) 1 21-36 25. Torshizi L, Anoosheh M, Ghofranipour F, Ahmadi F, Houshyar-rad A. The effect of education based on Health Belief Model on preventive factors of osteoporosis among postmenopausal women. Iran. J. Nurs (2019) 22 71-82 26. Shaghaghi A, Asadi M, Allahverdipour H. Predictors of self-medication behavior: a systematic review. Iran J Public Health (2014) 43 136-136. PMID: 26060736 27. Neafsey P, Jarrin O, Luciano S, Coffman MJ. Self medication practice in spanish speaking older adultsin Hartford, Conneticut. Hisp. Health Care Inrt (2007) 5 169-178 28. Jalilian F, Mehdi Hazavehei S, Vahidinia AA, Jalilian M, Moghimbeig A. Prevalence and related factors for choosing self-medication among pharmacies visitors based on health belief model in Hamadan Province, west of Iran. J. Res. Health Sci. (2013) 13 81-85. PMID: 23772020 29. Gharouni K, Ardalan A, Araban M, Ebrahimzadeh F, Bakhtiar K, Almasian M. Application of Freire’s adult education model in modifying the psychological constructs of health belief model in self-medication behaviors of older adults: a randomized controlled trial. BMC Public Health. (2020) 20 1350-1350. PMID: 32887596 30. Zareipour MA, Mahmoodi H, Valizadeh R, Ghorooji MG, Moradali MR, Zare F. Impact of an educational intervention based on the BASNEF model on skin cancer preventive behavior of college students. Asian Pac. J. Cancer Prev. (2018) 19 31. Heidarnia AJ. Factors influencing self-medication among elderly urban centers in Zarandieh based on Health Belief Model. J. Arak Univ. 2 Me. Sci (2016) 14 70-78 32. Hosseini FS, Joveini H, Jahromi VK, Sharifi N. Prevention of Self-medication in Women of Reproductive Age Based on a Health Belief Model: A Quasi-experimental Study. J. Educ. Community Health (2022) 9 18-25 33. Izadirad H, Niknami S, Zareban I, Hidarnia A.. Improving prenatal care in pregnant women in Iranshahr, Iran: Applying Health Belief Model. Women Health (2018) 58 1167-1178. PMID: 29111919
--- title: 'Efficacy of Motivational Interviewing and Brief Interventions on tobacco use among healthy adults: A systematic review of randomized controlled trials' authors: - Rajesh Kumar - Maya Sahu - Tamar Rodney journal: Investigacion y Educacion en Enfermeria year: 2023 pmcid: PMC10017134 doi: 10.17533/udea.iee.v40n3e03 license: CC BY 4.0 --- # Efficacy of Motivational Interviewing and Brief Interventions on tobacco use among healthy adults: A systematic review of randomized controlled trials ## Abstract ### Objective. To assess the effectiveness of a brief intervention and motivational interviewing in reducing the use of different tobacco-related products in adults ### Methods. For this systematic review, PubMed, Web of Science, and PsychINFO databases were electronically searched for randomized controlled trials on the effect of a brief intervention and / or motivational interview on tobacco reduction among healthy adults published between January 1, 2011 to January 1, 2021. Data from eligible studies were extracted and analyzed. CONSORT guidelines were used to assess the quality of the studies by two reviewers for the included studies. The titles and abstracts of the search results were screened and reviewed by two independent reviewers for eligibility criteria per the inclusion and exclusion criteria. Cochrane review criteria were used to assess the risk of bias in included studies. ### Results. A total of 12 studies were included in the final data extraction of 1406 studies. The brief intervention and motivational interviewing showed varied effects on tobacco use reduction among adults at different follow-ups. Seven of the 12 studies ($58.3\%$) reported a beneficial impact on reducing tobacco use. Pieces of evidence on biochemical estimation on tobacco reduction are limited compared to self-reports, and varied results on quitting and tobacco cessation with different follow-ups. ### Conclusion. The current evidence supports the effectiveness of a brief intervention and motivational interviewing to quit tobacco use. Still, it suggests using more biochemical markers as outcome measures to reach an intervention-specific decision. While more initiatives to train nurses in providing non-pharmacological nursing interventions, including brief interventions, are recommended to help people quit smoking. ## Objetivo. Evaluar la eficacia de una intervención breve y de la entrevista motivacional para reducir el consumo de diferentes productos relacionados con el tabaco en adultos. ## Métodos. Para esta revisión sistemática, se buscaron en las bases de datos PubMed, Web of Science y PsychINFO ensayos controlados aleatorizados sobre el efecto de una intervención breve y/o una entrevista motivacional en la reducción del consumo de tabaco entre adultos sanos, que hubieran sido publicados entre el 1 de enero de 2011 y el 1 de enero de 2021. Los títulos y los resúmenes de los artículos incluidos fueron evaluados por dos revisores independientes para determinar los criterios de elegibilidad, se analizó la calidad de los estudios con la guía CONSORT y se utilizaron los criterios de Cochrane para evaluar el riesgo de sesgo. ## Resultados. Se incluyeron un total de 12 de los 1406 estudios que arrojó la búsqueda. La intervención breve y la entrevista motivacional mostraron efectos variados en la reducción del consumo de tabaco entre los adultos en diferentes seguimientos. Siete de los 12 estudios ($58.3\%$) informaron de un impacto beneficioso en la reducción del consumo de tabaco. La utilización de indicadores bioquímicos de la reducción del consumo de tabaco fueron limitados en comparación con los autoinformes. Los resultados sobre el abandono y la cesación del tabaco fueron variados con diferentes seguimientos. ## Conclusión. La evidencia apoyó la efectividad de una intervención breve y de la entrevista motivacional para la cesación del consumo de tabaco. Sin embargo, se sugiere realizar más estudios con marcadores bioquímicos como medidas de resultado para llegar a una decisión específica de la intervención. Se recomienda formar a los enfermeros en la realización de intervenciones de enfermería no farmacológicas, incluidas las intervenciones breves, para ayudar a las personas a dejar de fumar. ## Objetivo . Avaliar a eficácia de uma intervenção breve e entrevista motivacional na redução do uso de diferentes produtos relacionados ao tabaco em adultos. ## Métodos . Para esta revisão sistemática, se buscou nas bases de PubMed, Web of Science e PsychINFO ensaios controlados aleatórios sobre o efeito de uma breve intervenção e/ou entrevista motivacional na redução do uso de tabaco entre adultos saudáveis, publicados entre 1º de janeiro de 2011 e 1º de janeiro de 2021. Os títulos e resumos dos artigos incluídos foram avaliados por dois revisores independentes para critérios de elegibilidade, a qualidade do estudo foi avaliada usando a diretriz CONSORT e os critérios Cochrane foram usados para avaliar o risco de viés. ## Resultados . Um total de 12 dos 1.406 estudos retornados pela busca foram incluídos. Intervenção breve e entrevista motivacional mostraram efeitos mistos na redução do uso de tabaco entre adultos em diferentes acompanhamentos. Sete dos 12 estudos ($58.3\%$) relataram um impacto benéfico na redução do uso de tabaco. O uso de indicadores bioquímicos de redução do uso de tabaco foi limitado em relação ao autorrelato. Os resultados sobre parar de fumar e parar de fumar foram variados com diferentes seguimentos. ## Conclusão . As evidências apoiaram a eficácia de uma intervenção breve e entrevista motivacional para a cessação do uso do tabaco. No entanto, mais estudos com marcadores bioquímicos como medidas de resultados são sugeridos para chegar a uma decisão de intervenção específica. Recomenda-se que os enfermeiros sejam treinados na execução de intervenções de enfermagem não farmacológicas, incluindo intervenções breves, para ajudar as pessoas a parar de fumar. ## Introduction Tobacco in any form is harmful and affects millions of lives every year.[1] In 2017, 8 million lives were lost due to smoking-related diseases.[2] Tobacco-related deaths are rising even after a decline in tobacco use trends because of the chronic nature of conditions.[3] In 2000, around $33.3\%$ of the global population over 15 years old were current tobacco users.[3] The negative consequences of tobacco use are well known and extend beyond individuals and countries regarding increasing health care expenditure and loss of productive life.[4] The tobacco consumption trend was three times higher in males than females in 2000, which was increased to four times in 2015 and is projected to be five times by 2025.[1,3] Notably, the detrimental effects of tobacco use gravely affected lower socio-economic populations with higher smoking prevalence.[5] However, tobacco use practices are varied and influenced by the locally available tobacco products in the different regions worldwide.[6] Smoking is one of the modifiable risk factors for many life-threatening health problems, including respiratory and cardiovascular health and genitourinary problems.[7] It has been estimated that $50\%$ of smokers who start smoking in adolescence die due to tobacco-related health problems.[8] Thus, an effective measure to control tobacco addiction is paramount. Implementing a wide range of interventions and strengthening tobacco control policy, including taxation, ban on tobacco use in public places, restriction on advertising of tobacco products, and creating smoke-free zones in educational institutions, brought a substantial decline in tobacco use in recent decades.[4] In addition to government initiatives to curb tobacco use, many pharmacological and non-pharmacological approaches are also involved in reducing tobacco-associated mortality and the burden of diseases.[6,9] Earlier studies reported that using a combination of pharmacologic and non-pharmacologic intervention is highly effective in reducing tobacco use. [ 10-12] However, non-pharmacological interventions have advantages over pharmacological interventions, including no side effects, long-term behavior changes,[13] knowing the real health hazards of long-term tobacco use, and cost-effective to show higher compliance.[11,12,14] Non-pharmacologic interventions for tobacco cessation include telephone counseling, individual and group counseling, health care provider interventions, exercise programs, and self-help programs.[12] Brief intervention or motivational interview is a brief yet realistic strategy offered to those who have a low motivation to quit.[15] Brief intervention is goal-directed but non-directive communication designed to improve motivation for change in quit behavior by eliciting feedback to plan for change.[12,16-20] The terms brief intervention (BI) and motivational interview (MI) are used with a common principle of active engagement of the client in the process of reduced use and teaching alternative coping skills.[21] These interventions are based on the philosophy that the client holds a key role in showing commitment and successful recovery.[22] Brief intervention sometimes follows the principles of the motivational interview to motivate the specific behavior of an individual to reduce or quit substance use.[23] However, these interventions are substantially modified in the delivery approach, format, and content in earlier published work.[12] Brief intervention primarily focuses on present concerns and stressors rather than exploring the historical antecedents of an individual and is conducted by a trained therapist.[20,24] Earlier work on the efficacy of brief intervention reported evidence that brief intervention increases the motivation to quit short-term use.[18,25] However, the evidence on long-term effects of brief interventions is equivocal, with no reduction of tobacco use at three months while higher self-reported abstinence at 1-year post-brief intervention.[26] Conversely, the brief intervention was found to be effective in improving quit rates, prolonging abstinence, and improving self-reported continuous abstinence among smokers at six months[27] and 1-year post-intervention[28] in other work. Still, there is a lack of consistent evidence on brief interventions to reduce use or quit tobacco use among the adult population. Nurses are an essential attribute of the health care system and play a vital role in delivering various interventions. It is natural to expect that nurses with adequate knowledge and skills in the brief intervention will do more to help their patients quit smoking. This meta-analysis will highlight the need for encouragement and opportunities to nurses to receive training on smoking cessation interventions. In addition, this will be insightful for the nurses to understand the significance of a non-pharmacological intervention to quit smoking. Towards this end, training nurses in the brief intervention using motivational interviews may be helpful to smokers and their families. Consequently, this systematic review aims to assess the effectiveness of the brief intervention in reducing tobacco use among adults. ## Methods A literature review was conducted with online databases PubMed, Web of Science, and PsychINFO. A literature search was completed using Boolean operators and truncations for the following key terms: [1] "Brief Intervention, [2] OR Screening and Brief Intervention” "tobacco products” AND [3] “Tobacco OR "tobacco products,” (MESH terms are also included in the search). The problem/disease was tobacco use among adults in the experimental group. The primary outcomes of interest were cessation in tobacco use, motivation/readiness to quit, reduction in tobacco quantity, days, abstinence days, quit attempts, and point prevalence measured by self-reported methods or biochemical verification at different intervals. Selection criteria and data extraction. The inclusion criteria for the studies included in this review were as follows: [1] the content of the article mainly focused on the provision of brief intervention and/or motivational interview for tobacco use reduction or cessation; [2] the participants were current smokers and adults; [3] the articles were published in peer-reviewed journals within the last ten years; [4] the study method reflected a randomized control trial (RCT). Articles were excluded if they focused primarily on other pharmacologic interventions, included any other substance use, were not designed as an RCT, or had mixed interventions. The search strategy was based on the population, intervention, control, and outcomes (PICO) approach with a PICO question, ‘does motivational interviewing and brief interventions helpful in reducing tobacco use in healthy adults?’; where P- Healthy tobacco users, I- Motivational Interview and/or Brief Intervention, C- Usual care or on other interventions and O- Smoking cessation.[29] A total of 1406 articles were included for a title and abstract review; at least two team members discussed discrepancies. 77 articles met the inclusion criteria for a full-text review, and 12 articles were selected for data extraction. See the PRISMA framework (Figure 1) that guided the review process.[30] Figure 1PRISMA Flow Diagram Bias assessment. Cochrane review criteria were used to assess the risk of bias in included studies in the review (Table 1).[31] All studies were evaluated on six evidence-based domains: allocation concealment, random sequence generation, participants and personnel blinding, outcome blinding, incomplete outcome data, and selective reporting.[31] Allocation concealment refers to concealing the information on the randomization process to the subjects. Random sequence generation occurs when study participants are not aware of the random sequence generation process. Blinding of participants and personnel refers to when participants and team members do not know the intervention or control condition to which subjects are assigned. Blinding of outcomes assessment refers to whether outcome measurement could have been changed by prior intervention knowledge to participants or team members delivered in work. Selective reporting refers to presenting only findings of interest. An incomplete outcome does not consider attrition while submitting the result.[31] For each study, these components are shown in ‘high risk,’ ‘low risk,’ or ‘unclear’ as written in the published version of the manuscript to decide on bias assessment. In data extraction, two authors assessed each study for bias. The authors discuss the risk bias criteria of the study using a checklist and conclude. The discrepancies were resolved after a discussion with the third author Table 1. Table 1Quality assessment of the included studiesSourcesRandom sequence allocationAllocation concealmentBlinding of participants / personnelBlinding of outcomes assessmentComplete outcomes dataAvoidance of selective reportingQuality of studyCatley et al. ✔✔XX✔✔ModerateMujika et al. ✔✔✔X✔✔HighVirtanen et al. ✔✔XX✔✔ModerateCook et al. ✔? ✔X✔✔ModerateSteinberg et al.✔???✔✔Moderate Ho et al. ✔✔XX✔✔ModerateCabriale et al. ✔??? ✔✔LowKrigel et al.✔✔??✔✔ModerateMeyer et al.✔? ✔✔✔✔HighSchane et al. ✔??? ✔✔LowLeavens ELS et al.✔?X?✔✔LowCabriales et al.✔?XX✔✔Low ## Results The electronic search produced a total of 3162 articles. 1406 articles were found suitable after removing duplicate records. Abstracts of all articles were reviewed independently by two reviewers. A total of 1262 articles were excluded after careful scrutiny of abstracts. Full-text articles were retrieved for 79, and after reviewing these articles independently, 67 articles were further excluded for a specific reason. After applying the eligibility criteria, 12 articles were included in the present review. The PRISMA flow diagram (Figure 1) summarizes the study selection and scrutiny process used for the articles. A summary of the selected studies summarized by year of publication, author, setting, type of study, sampling techniques, sample size, eligibility criteria (inclusion and exclusion), intervention, outcomes, strengths and limitations, and any other specific notes to the study. Study characteristics. Of the 12 included studies, eight were conducted in the United States, one in Sweden, one in Hong Kong, one in Germany, and one in Spain. All studies used a randomized controlled trials design with one or another trial feature, including allocation concealment and blinding. Of the 12 studies, 3 studies used brief intervention or brief advise,[30,32,35,38,41] 6 studies used motivational interviews [15,30,33,34,37,40], and one study used brief counseling on harm to self and harm to others [39] and quit immediately award model based on brief intervention approach. Seven of the 12 studies ($58.3\%$) reported a beneficial effect of brief advice or motivational interview on reducing tobacco use (Table 2). Motivational Interviewing (MI). The concept and use of motivational interviewing as an intervention is not new in substance use,[42] smoking reduction,[43] chronic lifestyle disease,[44] health behavior,[45] medication adherence,[46,47] oral health in adolescents,[48] and chronic pain management.[49] The concept was published by Miller & Rollnick and presented as a therapeutic effort to strengthen personal motivation and commitment to a specific goal by eliciting and exploring the individual’s reason for a change in behavior with compassion and acceptance.[16] Motivational interviewing (MI) is a patient-centered, directive therapeutic style to improve readiness to change behavior by resolving the ambivalence.[43] MI was found to be an effective method in a series of addictive behaviors.[50] Some research[33] among healthy adult smokers tested multiple interventions revealed a promising effect of motivational interviewing on smoking reduction. However, the study concluded[50] that motivational interviewing and other interventions will produce the most consistent and marked reduction in smoking. A contrasting study[15] used motivational interviewing over health education and brief advice but did not report any change in quit attempts at 6 months. However, the same study reported increased cessation of medication use, motivation, and confidence to quit compared to brief advice, which further indicates the effectiveness of MI in behavior changes to quit smoking. In a study[34] at a Northeastern US State, daily smokers attended brief motivational interviewing and significantly reduced cigarette use. Likewise, motivational interviewing effectively improved quitting smoking among nurses over brief advice in a study conducted in Spain.[30] However, in another work[37] on college tobacco smokers, the use of motivational interviewing over health education (HE) showed no significant reduction in motivation to quit, abstinence, and quit attempts. Likewise, the consistent findings are presented in earlier studies[15,51] that reported no significant advantage of MI on smoking cessation compared to alternative interventions. In a recent work conducted in the Midwest United States, a brief motivational interview showed no improvement in reducing water pipe use[40]; however, MI was found to improve awareness of risk perceptions, commitment, and confidence to quit waterpipe (WP) smoking. Furthermore, in a recent meta-analysis, MI reported a modest yet significant beneficial increase in quitting rates in a group that utilized motivational interviewing. Further, findings revealed that long-term motivational interviewing by a primary physician or counselor is more effective in quitting tobacco. However, there is no specific evidence on the duration and number of MI sessions on quitting the behavior. Another meta-analysis[52] reported a greater likelihood of abstinence behavior in the experimental arm comprising adults and adolescents when compared to the comparison group. Still, only a few older interventions and meta-analyses demonstrate the effectiveness of motivational interviewing in smoking cessation. There is evidence that motivational interviewing is less effective in low-motivation patients.[18,53] However, the conclusive evidence to prove the quality and fidelity of MI implementation remains contentious concerning its effectiveness in smoking reduction. Brief Intervention. Brief intervention or advice for harmful substance use has been practiced for many years. [ 54] It aims to identify the current and potential problems with substance use and motivate people to change high-risk behavior.[55] Brief intervention is a personalized, supportive and non-judgmental approach to treatment.[55] It is* also defined as a verbal ‘stop smoking’ message loaded with harmful effects of tobacco use.[56] Brief intervention can be used in various methodologies, including unstructured counseling and feedback to formal structured treatment.[57-59] World Health Organization uses education, simple advice, and brief counseling as alternative types of brief interventions for high-risk individuals with alcohol use disorders.[60] Brief intervention also uses screening and referral services and is therefore called screening, brief intervention, and referral to treatment (SBIRT).[61] Brief therapy can help motivate an individual to change his high-risk behavior at a different stage of behavior change.[62] The stage of change model proposed by Prochaska & DiClemente, helps clinicians tailor a brief intervention to the stage of behavior change and the client's needs.[63] Brief interventions for tobacco use disorders aim to enhance motivation for change and provide evidence-based resources to reduce usage or complete cessation of tobacco products. The 5A’s approach (Ask, Advise, Assess, Assist, & Arrange) is an evidence-based approach that helps tobacco users in different settings with motivational strategies in a systematic fashion.[64] In addition, FLAGS-Feedback, Listen, Advice, Goals, Strategies and ‘FRAMES’-Feedback, Responsibility, Advice, Menu of options, Empathy, and Self-efficacy, are other frameworks used to deliver brief interventions.[65] The brief intervention is effective in many ways, including cost-effectiveness in terms of time and money,[66] increased abstinence rate and days,[35,67] and early days of discharge, and regular follow-ups [68]. Similarly, a more intensive planned brief advice (>20 minutes) may augment the effect on quit rate and 6-months abstinence compared to minimal brief advice.[69] Additionally, the use of brief components in AWARD [Ask, Warn, Advice, Refer, Do-It Again) model, and cut down to quit: [CDTQ]), reported a higher quit rate in the former group. [ 35] Furthermore, brief advice in combination with tailored practice was highly effective on 7-days point prevalence and 7-days and 6-months abstinence rate among adult smokers. [ 38] Brief counseling also reported a significant reduction in quit rate, abstinence phenomenon, improved motivation, and self-efficacy in a regular follow-up in a group of nondaily smokers.[36,39] Conversely, brief therapy showed no significant changes in abstinence rate among adults who underwent immediate and delayed intervention at the family health clinic U.S.-Mexico border,[41] and hence, the efficacy of brief therapy has been questioned in recent years.[70] Further, brief treatment can be helpful for varied kinds of the population, including adolescents, older smokers, smokers with mental illness and co-morbidities, alcohol users, and pregnant women across different racial and ethnic groups.[66,70] However, current or former tobacco smokers who were willing or unwilling to make quit attempts are the most eligible groups to attend the brief intervention.[66] ## Discussion The use of tobacco has innumerable adverse effects on health. The present review aimed to assess the effectiveness of a brief intervention in reducing tobacco use among adults. The review findings indicate that brief intervention alone or combined with Motivational Interviews or Health Education was effective, supported by previous results.[15,52] In contrast, an earlier systematic review documented that motivational interviewing was modestly successful in promoting smoking cessation compared with usual care or brief advice.[25] Conversely, motivation to quit was higher after Brief Advice than MI.[71] Another recent systematic review conducted with 37 studies reported insufficient evidence to show whether MI helps people stop smoking compared with no intervention, as an addition to other types of behavioral support, or compared with different kinds of behavioral support for smoking cessation.[72] Modality and intensity of interventions with follow-up and primary outcomes were also determining factors for the effectiveness of the studies. In the current review, the intervention modality varied in face‐to‐face sessions or a combination of face‐to‐face and telephone sessions. Initial sessions were conducted face-to-face, and the follow-up was done over the telephone for most of the study, which is usual with much other previous work.[72] Brief intervention provided through telephone has great significance in the present scenario. Amid the COVID-19 pandemic, when individuals have restricted movement or limited resources available, virtual or phone delivered brief intervention can play a significant role in helping the adults quit smoking or reduce tobacco use. A previous study has documented moderate‐certainty evidence of proactive telephone counseling in increasing the quit rates in smokers who seek help from quitlines.[73] The included studies had intervention sessions as little as one brief session[37] to four sessions based on Motivational Interviews. [ 30] Prior literature suggests that multiple sessions might increase the likelihood of quitting over single-session treatment, but positive outcomes were reported in both cases.[25] However, there is no specific evidence on the duration and number of MI sessions on quitting the behavior.[72]The current review found that the included studies had a follow-up of the intervention ranging from 3 months to 12 months. However, face-to-face or telephone counseling follow-up did not show a significant effect of an intervention. However, reduction of smoking behavior or abstinence was not sustained over time. These findings were supported by a previous work where smoking abstinence averaged $10\%$ at 1 month and around $2\%$ at 3, 6, and 12 months.[71] At present, evidence is unclear on the optimal number of follow-up calls.[25,43] The primary outcomes of the studies were smoking abstinence, reduction in smoking rates, and an increase in motivation to quit. However, outcomes other than cessation may be essential to assess when determining the effects of brief interventions for tobacco use. Hence, different outcomes were self-efficacy, motivation, and changes in depression over the studies. Biological tests to confirm tobacco abstinence provided more reliable findings than self-reported abstinence. Intervention programs on Smoking cessation, such as brief advice, motivational interviews, or the 5A approach (Ask, Advise, Assess, Assist, and Arrange), are effective among specific populations or specialized clinical settings.[45,74] Professional support and cessation interventions or medications significantly increase the chance of successfully quitting.[3] A systematic review and meta-synthesis explored smokers' perspectives regarding smoking cessation and reported that lack of motivation to quit was one of the significant issues they felt for tobacco cessation.[75] Nonetheless, these non-pharmacological interventions had shown efficacy similar to the pharmacological intervention[74] with additional benefits of cost-effectiveness, competency of the provider, and accessibility to the treatment center. Tobacco-related deaths and disabilities are increasing around the globe because of the continued use of different kinds of tobacco products. Many earlier studies confirmed the beneficial effect of a brief intervention based on motivational principles to reduce tobacco use. Nurses' role is precise in tobacco cessation to endorse the International Council of Nurses statement to integrate tobacco use prevention and cessation as part of their regular nursing practice.[76] This systematic review indicates the potential benefits of brief intervention, which can be a breakthrough for nurses in tobacco reduction around the globe. However, nursing policymakers should incorporate smoking cessation interventions as a part of standard practice for all the patients. Hence, brief intervention or motivational interviews provide promising results in cessation or reduction of tobacco use which needs to be further supported by evidence. The present review should be appraised under its many limitations and strengths. Among its strengths is that it provides coverage of randomized controlled trials that included brief intervention and motivational interviewing on smoking and other tobacco use among adults. This review included samples of those with clinical and non-clinical samples using tobacco. The major strength of this review lies in the inclusion of RCT studies that give a clear description of participants' characteristics, methodology, and implemented intervention. Secondly, the risk of bias assessment showed that most studies had low to moderate risk. This review highlights several opportunities for future research, such as brief intervention or motivational interview combined with other adjuncts to improve outcomes and further research integration of these interventions with combination therapies of psychotherapeutic and pharmacological interventions. In terms of limitations, the heterogenicity of the selected studies did not allow to reach a specific conclusion. Studies included in this review used different brief intervention and motivational interview forms, making it challenging to synthesize the results and suggest a potential use of these interventions in day-to-day practice. Heterogeneity in population also made it challenging to generalize the findings across all people around the globe. Further, studies involved in the review only investigated tobacco cessation among healthy adults may confer unique limitations on the generalizability of results. The authors suggest interpreting and using review findings cautiously due to variations in treatment fidelity and the inclusion of a limited number of studies. ## Conclusion. Over time there have been changes in treatment modalities for tobacco cessation. Preference for non-pharmacological intervention over pharmacological has led the researchers to find supportive evidence. The present review highlights the effectiveness of a brief intervention and motivational interviewing in reducing tobacco use among adults. It also demonstrates that the effects are far-reaching. However, it remains inconclusive which intervention is more effective than the other. Future longitudinal studies or RCTs with direct comparison of different interventions may further refine the evidence-based practice on tobacco cessation among adults. Table 2Characteristics of included studies in the review Reference 15: Catley D, Goggin K, Harris KJ, Richter KP, Williams K, Patten C, et al. A randomized trial of motivational interviewing: Cessation induction among smokers with low desire to quit. Am. J. Prev. Med. 2016; 50[5]:573-83.Population and sample size: Setting: Midwestern city, Kansas, USA. Sample: Adult smokers. Sample size: 255. Age (Mean, SD): 45.8 [SD = 10.9]). Design: Single site, parallel-group RCT design. Randomization: Computer-generated random assignment, Imbalanced allocation (2:2:1) for three interventionsInclusion criteria: Adult age 18 years & currently smoking one or more cigarettes per day, able to speak English, have stable reachability, no intention to get pregnant in the next 6 months, not using any medication for smoking cessation, have no cessation plan in the next 7 days and confirm tobacco use on CO≥7 ppm. Exclusion criteria: N/AIntervention and comparators: Motivational interview (MI, $$n = 102$$) Versus Health education (HE, $$n = 102$$) Versus Brief advise (BA, $$n = 52$$)Primary outcomes: The health education group significantly shows a higher abstinence rate at 6-month follow-up, Motivational interviews and health education groups showed a more significant increase in reduced medication use, motivation, and confidence to quit over the brief advice group, Health advice was relatively found better to improve motivation than motivational interviewing. Others: Strengths: Biochemical verification of 7-day smoking point prevalence by saliva testing, use of intensity match comparison design to test the exact effect of MI over health education. Limitations: Self-reported measures to test motivation, desire to quit, quit attempts, and point prevalence, the study was limited to willing to quit smokers, and findings may not be generalizable to unmotivated smokers. Any other Notes: Follow-up for all three interventions at 3 months and 6 months. Missing data handling using appropriate measures to avoid bias in the study. Reference 30: Mujika A, Forbes A, Canga N, de Irala J, Serrano I, Gascó P, et al. Motivational interviewing as a smoking cessation strategy with nurses: an exploratory randomised controlled trial. Int. J. Nurs. Stud. 2014; 51[8]:1074-82.Population and sample size: Setting: Clinical Universidad de Navarra (CUN) in Pamplona (Navarra), teaching hospital, North Spain. Sample: Nurses. Sample size: 30. Age (Mean, SD): 40.15[SD = 9.45]). Design: Two groups parallel experimental design. Randomization: Computer generated random allocation method, and seal the opaque envelope for location concealment. Inclusion criteria: Nurses who smoke and are ready to participate in the study and nurses work in the hospital irrespective of thinking of quit or not.] Exclusion criteria: N/A.Intervention and comparators: Motivational interview ($$n = 15$$)/ brief advices ($$n = 15$$)Primary outcomes: More nurses in the intervention arm had quit smoking with an absolute difference of $33.3\%$ $95\%$ CI (2.6-58.2). Progress in the stage of changes was more significant in nurses who attended a motivational interview. Others: Strengths: Biochemical verification of urine cotinine level for recent smoking detection and Micro+Smokerlyzer use for expired Carbon Monoxide (CO) detection for enrollment of the subjects. Detection of self-report of abstinence by biochemically urine cotinine measurement. Intention-to-treat analysis to control bias. Limitations: Use of self-reported measures to report nicotine dependence, desire, and readiness to quit. Very low small size to study the effectiveness of the intervention. No follow-up to measure smoking cessation. No sample size analysis; small sample size. Any other Notes: Collection of data at baseline, end of the intervention, and 3 months after the intervention to cross-check adherence. High satisfaction with the acceptability and feasibility of the intervention indicates the genuine interest of the participants. Use of one-to-one sessions with each participant. Reference 32. Virtanen SE, Zeebari Z, Rohyo I, Galanti MR. Evaluation of a brief counseling for tobacco cessation in dental clinics among Swedish smokers and snus users. A cluster randomized controlled trial (the FRITT study). Prev. Med. 2015; 70:26-32.Population and sample size: Setting: Gavleborg and Sodermanland county, Sweden. Sample: Patients currently using tobacco daily. Sample size: 467. Age (Mean, SD): 45.57 [SD = 14.91]). Design: Randomized Cluster design. Randomization: Setting randomization with a 1:1 computer-generated random number. Inclusion criteria: Patient’s age 18-75 years, Daily tobacco users since last 1 year, able to converse in the Swedish language. Exclusion criteria: Patients with acute dental illness, severe psychiatric disease, alcohol problems, or use illicit drugs and are currently involved in other cessation programs. Intervention and comparators: Brief advice based on 5A’s principles ($$n = 225$$) Versus usual care ($$n = 242$$).Primary outcomes: Reduction of tobacco consumption & changes in the expected direction for all outcomes were more frequent in the intervention arm. Others: Strengths: The study used brief advice as per standard 5 A’s approach. Selection of big sample size to make the findings generalizable to a similar population. Limitations: Lack of randomization for patients, use of computer randomized random sequence for only clinics used; lack of blindness and self-report data; failure to screen all eligible patients at some clinics. Any other Notes: Sub-groups analysis to differentiate the impact of the intervention on snus and smoke users; Demonstration of counseling using interactive teaching techniques; Follow-ups visits after 6- months. Reference 33: Cook JW, Collins LM, Fiore MC, Smith SS, Fraser D, Bolt DM, et al. Comparative effectiveness of motivation phase intervention components for use with smokers unwilling to quit: a factorial screening experiment. Addiction. 2016; 111[1]:117-28.Population and sample size: Setting: Southern Wisconsin, USA. Sample: Adult smokers. Sample size: 517. Age (Mean, SD): 47.0 ([SD = 14.4]). Design: Balanced four-factor randomized factorial design. Randomization: Stratified permuted, computer-generated block randomization (block size 16) based on gender and clinic. Inclusion criteria: Adult aged ≥18 years; smoked ≥ 5 cigarettes/day for the previous 6months, adult not interested in quitting in the next 30 days but willing to cut down, able to read, write and speak the English language, agreed to complete assessment, planned to remains in the area for next 6 months, not currently using Bupropion and Varenicline, consented to use only study smoking medication during the study if reported current NRT use; nonmedical contraindications to Nicotine Replacement Therapy (NRT) use, women of potential childbearing agree to use birth control pills. Exclusion criteria: N/A.Intervention and comparators: Motivational interviewing vs. none x Nicotine patch vs. none, x Nicotine gum vs. none x Behavioral reduction vs. no intervention ($$n = 253$$) or usual care ($$n = 264$$).Primary outcomes: Smoking reduction was higher in nicotine gum combined with behavioral reduction counseling group and behavioral reduction counseling combined with motivational interviewing. Others: Strengths: Use factorial design to test multiple interventions compared to usual care and stratified permuted random sampling. Follow-ups at 12- and 26-weeks following study enrollment. Limitations: Self-reported response for outcomes measures and limited blinding for staff and participants. Any other Notes: Use of phase base model of smoking intervention, the use of multiple treatment strategies using factorial design will help to test multiple hypotheses at one time. Reference 34: Steinberg ML, Rosen RL, Versella M V, Borges A, Leyro TM. A Pilot Randomized Clinical Trial of Brief Interventions to Encourage Quit Attempts in Smokers From Socioeconomic Disadvantage. Nicotine Tob. Res. 2020; 22[9]:1500-8.Population and sample size: Setting: Local community soup kitchen, Northeastern US State. Sample: Daily smokers. Sample size: 64. Age (Mean, SD): (Mage= 47.4 years [SD = 10.7]). Design: Pilot Randomized Clinical Trial. Randomization: Block randomization. Inclusion criteria: Patient’s age 19-65 years, daily tobacco users, able to read and speak the English language, and Carbon Monoxide (CO) reading greater than 5 ppm. Exclusion criteria: Patients on U.S FDA approved smoking cessation aids, patients with severe psychiatric disease, alcohol problems, illicit drug use, and are currently involved in other cessation programs, patients on antipsychotics medications, self-reported current medical problems potential concern to nicotine replacement, pending legal issues with the potential to result in incarceration and women should be on effective birth control and could not be nursing or pregnant or planning to become pregnant in the next 2 months. Intervention and comparators: Brief (e.g., 30 m) Motivational Interviewing [19], Nicotine Replacement Therapy (NRT) ($$n = 19$$), or a Referral-Only intervention ($$n = 20$$).Primary outcomes: $40\%$ of the sample reported making a serious quit attempt at follow-up, significant self-reported reduction in smoking and more use of NRT and lozenge in NRT group at 6 months’ follow-up. Others: Strengths: Unique population (socio-economically disadvantaged smokers), follow-up (30 days) the cases to measure self-reported quit rate/attempt and comparison of three interventions simultaneously in one design. Limitation: Study included a small sample size ($$n = 57$$).Any other Notes: Follow-up at 1 month, unique population; socio-economically disadvantaged smokers, use of Post hoc analysis to find financial strain as a significant moderator of the effect of the intervention on smoking behaviorReference 35: Ho KY, Li WHC, Wang MP, Lam KKW, Lam TH, Chan SSC. Comparison of two approaches in achieving smoking abstinence among patients in an outpatient clinic: A Phase 2 randomized controlled trial. Patient Educ. Couns. 2018; 101[5]:885-93. Population and sample size: Setting: Hong Kong -outpatient clinic. Sample: Chinese smokers- medical follow-up. Sample size: 100. Age (Mean, SD): (Mage= 55.6 years [SD = NA]). Design: A Phase 2 RCT. Randomization: Computer generated Inclusion criteria: 18- years or older and smoked at least five cigarettes per day. Exclusion criteria: Unstable medical conditions, poor cognitive function, mental illness, currently participating in other smoking cessation programs or services. Intervention and comparators: (Quit immediately: [QI]- received a booklet about smoking cessation and brief intervention using the AWARD [ask, Warn, Advice, Refer, Do-It Again) model, and cut down to quit: [CDTQ]), to quit progressively. Primary outcomes: QI group had a significantly higher self-reported quit rate than those in the CDTQ group at the 6-monthfollow-up ($18.0\%$ vs. $4.0\%$, adjusted OR = 0.190, $95\%$ CI = 0.039-0.929). Not significant at the 12-month follow-up ($12.0\%$ vs. $4.0\%$, adjusted OR = 0.306, $95\%$ CI = 0.059-1.594).Others: Strengths: 4 follow-ups (1,3,6,12 months) to measures outcomes, use of allocation concealment to blind randomization and intention-to-treat analysis to control bias in the analysis. Limitations: A pilot approach to select all subjects from the same setting may infuse participant selection bias and only 6 and 12 months follow up with 73 % retention rate. Any other Notes: 50 years and over half had received education at the lower secondary school level or below CDTQ methods are relatively more complicated than QI methods, which require an understanding of smoking education strategies and close monitoring of the number of cigarettes consumed and reduced. Reference 36: Cabriales JA, Suro Maldonado B, Cooper T V. Smoking transitions in a sample of Hispanic daily light and intermittent smokers. Addict Behav. 2016; 62:42-6. Population and sample size: Setting: Health clinic, hospital, or university on the U.S/México border. Sample: Hispanic (DLS/ITS) daily light (DLS;<=10 cigarettes per day) and intermittent (ITS; nondaily) smokers. Sample size: 190, a subset of 390 follow-up samples. Age (Mean, SD): (Mage= 38.6 years [SD =15.1]) Design: Randomized controlled trial. Randomization: Randomly assigned to either an immediate or delayed intervention group at baseline using an online random number generatorInclusion criteria: Age of at least 18 years and smoking between one cigarette a month to 10 cigarettes per day (CPD).Exclusion criteria: N/AIntervention and comparators: Immediate brief cessation intervention versus delayed intervention (control) group. Primary outcomes: Smoking categories to control group (DLS/ITS) remains stable, with no significant group difference. DLS group at both points showed higher nicotine dependence levels$.8.95\%$ went from daily light smokers (DLS) to quitting, and $5.26\%$ went from intermittent smokers to quitting at 3-month follow-up. Others: Strengths: Specific population; Hispanic, an underrepresented population in smoking cessation studies, use of multi-component intervention in one study. The first study to discuss light and intermittent smoking to compare efficacy of brief smoking cessation intervention. 3- month follow-up to measure to measures outcomes in both groups. Limitations: High attrition rate ($48\%$); “contact-information mobility” - challenges to maintain communication with participants; participant work schedules; prioritization of “personal and family safety” over health-related behaviors; “the study was brief and perhaps not intensive enough to cause cessation.” The self-report method at baseline and follow-up for smoking status rather than biochemical process. Any other Notes: All-Hispanic, predominantly Mexican/Mexican American community sample potentially limits generalizability. Reference 37: Krigel SW, Grobe JE, Goggin K, Harris KJ, Moreno JL, Catley D. Motivational interviewing and the decisional balance procedure for cessation induction in smokers not intending to quit. Addict Behav. 2017; 64:171-8.Population and sample size: Setting: Urban University using the psychology department research pool, USA. Sample: University students. Sample size: 82 Age (Mean, SD): (Mage= 26.9 years [SD =9.6]) Design: Not Specified [Random assignment of the subjects in two groups]. Randomization: Computer-generated random number assignment in a sealed envelope. Inclusion criteria: Smoking at least one cigarette during the last 7 days, having no intentions to quit in the next 30 days, age at least 18, college enrollment, and reachability via phone & email. Exclusion criteria: N/A.Intervention and comparators: Motivational Interviewing using only the decisional balance component (MIDB)/ health education around smoking cessation (HE).Primary outcomes: Both groups showed significant reductions in smoking rates and increases in motivation to quit, quit attempts, and self-reported abstinence, with no significant group differences. Others: Strengths: Cost & time efficient interventions, use of intention-to-treat analysis and maximum-likelihood estimation to accommodate missing data. Limitations: Population of interest is a small/limited group; “college students who were generally light smokers”. The use of a small sample size may hinder generalizability. Outcomes measures were self-reported without control group with no biochemical verification of abstinence. Any other Notes: Recruitment materials made no mention of quitting smoking, and participants were informed they would receive up to $20 for study completion. Only one session of MIDB or HE was performed per participant. Each session was, on average <20 minutes. Reference 38: Meyer C, Ulbricht S, Gross B, Kästel L, Wittrien S, Klein G, et al. Adoption, reach and effectiveness of computer-based, practitioner delivered and combined smoking interventions in general medical practices: a three-arm cluster randomized trial. Drug Alcohol Depend. 2012; 121[1-2]:124-32.Population and sample size: Setting: Northern Eastern, Germany. Sample: Adult smoker patients. Sample size: 263. Age (Mean, SD): 41.17 years [SD = 15.2]). Design: Three-arm clustered randomized controlled design. Randomization: Cluster randomization of the medical practices ($$n = 151$$).Inclusion criteria: Patients aged more than 18 years or older reported any tobacco smoking use in the last 6 months. Exclusion criteria: Practices registered for another facility apart from general practice. Intervention and comparators: Brief advice (practice $$n = 50$$; patients $$n = 618$$)/Tailored letter (practice $$n = 50$$; patients $$n = 1484$$) / Combination (practice $$n = 51$$; patients $$n = 1113$$).Primary outcomes: The seven-day point prevalence was higher in the combination group compared to brief advice or tailored intervention. The rate of 6- month prolonged was higher in the combination group than the brief advice and tailored letters group. 7-days and 6-month prolonged abstinence were statistically significant between the combination group and the other two groups. Tailored letters group shows significantly higher abstinence within past 7-days at 12-month follow-up in contract to combination and brief advice. The number of abstinent patients was significantly higher in a tailored letter or combination group followed by brief advice. Others: Strengths: Recruiting a large sample size for a three-arm clustered randomized design. Use of advanced imputations to find best results for ‘missing at random’ cases. Limitations: Self-reported abstinence and lost to follow-up of one-quarter of patients at 12-months. Any other Notes: 12 months’ follow-ups for all registered patients. Comparison of three interventions in different arms at a time to determine the efficacy of three different interventions. Reference 39: Schane RE, Prochaska JJ, Glantz SA. Counseling nondaily smokers about secondhand smoke as a cessation message: a pilot randomized trial. Nicotine Tob. Res. 2013; 15[2]:334-42.Population and sample size: Setting: San Francisco Bay Area, U.S. Sample: Nondaily smokers. Sample size: 52 Age (Mean, SD): 32.66 years [SD = 11.11]). Design: A randomized pilot trial. Randomization: Random sequence created by SAG using the random number generator in Minitab 14.Inclusion criteria: Respondents smoked at least 100 cigarettes in their lifetime, smoked at least once in the past seven days but not every day, age 18 years or older and speak the English language. Exclusion criteria: Participants had an exhaled carbon monoxide (CO) exceeding 10ppm. Intervention and comparators: Brief counseling on Harm to Self-group (HTS, $$n = 26$$) provided information on tobacco use and its risk on developing different medical conditions along with chemical ingredients of tobacco by a nurse/Harm to Others (HTO, $$n = 26$$) informed about tobacco use and its risk on friends and family members similar to the HTS group. Primary outcomes: A significant difference in abstinence between harm to others (HTO) ($36.8\%$) and harm to self (HTS) ($9.5\%$) groups. A significant change in contemplation ladder score between participants who completed follow-ups than who lost to follow-up. Trying to reduce or quit smoking is higher in the HTO group (not significant, $$p \leq 0.607$$). Comparable smoking reduction at 3 months follows in both groups. No difference in intervention acceptability in both the groups. Improved motivation and self-efficacy from baseline to 3-month follow-up in both groups. Others: Strengths: Bio confirmed tobacco abstinence at the 3- month follow-up. Limitations: The sample size was small for testing efficacy and limited to self-reported smoking cessation at 3-month follow-up. Any other Notes: 3-month follow-up for smoking cessation. Bio confirmed tobacco abstinence at the 3 months and use of urinary cotinine test to cross-check the abstinence. Reference 40: Leavens ELS, Meier E, Tackett AP, Miller MB, Tahirkheli NN, Brett EI, et al. The impact of a brief cessation induction intervention for waterpipe tobacco smoking: A pilot randomized clinical trial. Addict Behav. 2018; 78:94-100.Population and sample size: Setting: Water pipe (WP) lounges in urban and suburban areas in the Midwest U.S. Sample: Water pipe smokers. Sample size: 109. Age (Mean, SD): 21.1 [SD = 5.08]). Design: Pilot randomized control trial. Randomization: Cluster randomization (block of 4).Inclusion criteria: Participant age ≥18 years. Exclusion criteria: N/A.Intervention and comparators: Brief motivational interview ($$n = 53$$) /No intervention ($$n = 55$$).Primary outcomes: No Significant difference in WP (number of days WP used and number of WP used). Increase awareness on risk perceptions, commitment to quit, and confidence to quit WP smoking. Others: Strengths: Cluster randomization to avoid bias in sample selection. Carbon monoxide exposure detection by eCO (exhaled carbon monoxide) detector. Multiple outcome measurement. Limitations: No eCO detection at 3 months’ follow-up. Any other Notes: Use of eCO detector at baseline, immediately before entering to lounge and post-session gave more reliable findings. Follow-up survey at 3 months of post-session. Reference 41: Cabriales JA, Cooper T V., Salgado-Garcia F, Naylor N, Gonzalez E. A randomized trial of a brief smoking cessation intervention in a light and intermittent Hispanic sample. Exp. Clin. Psychopharmacol. 2012; 20[5]:410-9.Population and sample size: Setting: StopLite smoking cessation intervention at a family health clinical (primarily) or university on the U.S. Mexico border. Sample: Hispanic smokers. Sample size: 214. Age (Mean, SD): 38.62 years [SD = 15.08]). Design: Pretest-posttest randomized control-group design with replacement of control group with delayed intervention. Randomization: Online random number generator. Inclusion criteria: Hispanic at least 18 years of age and smoking between one cigarette a month to 10 cigarettes per day. Exclusion criteria: Non-HispanicIntervention and comparators: Carbon Monoxide (CO) feedback, ME, trigger management, and HE (Immediate versus delayed intervention group).Primary outcomes: No significant differences in abstinence rates between the immediate and delayed intervention conditions. Significant increases in motivation to quit in the immediate intervention compared to the delayed intervention group. Others: Strengths: 3-month follow-up by telephone, mail, or in person. Participants in a delayed intervention (control group) received the brief intervention after the end of the study. Limitations: Self-reported nicotine status as outcome measures and limited to the Hispanic population only. Any other Notes: The brief intervention included self-efficacy, motivational enhancement, trigger management, and health education components. Non-eligible participants were offered QuitLine & Quintet resources. ## References 1. WHO Global Report: Mortality Attributable to Tobacco. Who Global Report (2012.0) p. 392-p. 392 2. Who Report on the Global Tobacco Epidemic. ASHE-ERIC Higher Education Report (2013.0) Vol. 23 p. iii-p. vii 3. WHO global report on trends in prevalence of tobacco use 2000-2025. World Health Organization (2019.0) 4. Smoking prevalence and attributable disease burden in 195 countries and territories, 1990-2015: a systematic analysis from the Global Burden of Disease Study 2015. Lancet (2017.0) 389 1885-1906. PMID: 28390697 5. Parsons VL, Moriarity C, Jonas K, Moore TF, Davis KE, Tompkins L. Design and estimation for the national health interview survey, 2006-2015. Vital Health Stat 2 (2014.0) 165 1-53 6. Haokip HR, Kumar DR, Singh Rawat DV, Kumar Sharma DS. Efficacy of standard nicotine replacement therapy (NRT) versus video-assisted nurse-led NRT on tobacco cessation: A randomized controlled pilot trial. Clin. Epidemiol. Glob. Health (2021.0) 9 141-146 7. Fagerström K. The epidemiology of smoking: health consequences and benefits of cessation. Drugs (2002.0) 62 1-9 8. Tobacco Fact Sheet No.339.. World Health Organization (2013.0) 9. Hatsukami D, Jensen J, Allen S, Grillo M, Bliss R. Effects of behavioral and pharmacological treatment on smokeless tobacco users. J. Consult. Clin. Psychol (1996.0) 64 153-161. PMID: 8907095 10. Giulietti F, Filipponi A, Rosettani G, Giordano P, Iacoacci C, Spannella F. Pharmacological Approach to Smoking Cessation: An Updated Review for Daily Clinical Practice. High blood Press. Cardiovasc. Prev (2020.0) 27 349-362. PMID: 32578165 11. Cornuz J, Willi C. Nonpharmacological smoking cessation interventions in clinical practice. Eur. Respir. Rev (2008.0) 17 187-191 12. Niaura R. Nonpharmacologic therapy for smoking cessation: characteristics and efficacy of current approaches. Am. J. Med (2008.0) 121 11-19 13. Laland KN R, Choe JC. Encyclopedia of Animal Behavior (2010.0) 380-386. DOI: 10.1016/B978-0-12-813251-7.00057-2 14. Cornuz J, Gilbert A, Pinget C, McDonald P, Slama K, Salto E. Cost-effectiveness of pharmacotherapies for nicotine dependence in primary care settings: a multinational comparison. Tob. Control (2006.0) 15 152-159. PMID: 16728744 15. Catley D, Goggin K, Harris KJ, Richter KP, Williams K, Patten C. A randomized trial of motivational interviewing: Cessation induction among smokers with low desire to quit. Am. J. Prev. Med (2016.0) 50 573-583. PMID: 26711164 16. Miller WR RS. Motivational Interviewing: Helping People Change (2013.0) 17. Lancaster T, Stead LF. Individual behavioural counselling for smoking cessation. Cochrane Database Syst. Rev (2017.0) 3 18. Hettema JE, Hendricks PS. Motivational interviewing for smoking cessation: a meta-analytic review. J. Consult. Clin. Psychol (2010.0) 78 868-884. PMID: 21114344 19. Rice VH, Heath L, Livingstone-Banks J, Hartmann-Boyce J. Nursing interventions for smoking cessation. Cochrane Database Syst. Rev (2017.0) 12 20. 20 World Health Organization (WHO) The ASSIST-linked brief intervention for hazardous and harmful substance use. Manual for use in primary care2012cited 2021 Jul 2Available from: https://www.who.int/publications/i/item/the-assist-linked-brief-intervention-for-hazardous-and-harmful-substance-use. The ASSIST-linked brief intervention for hazardous and harmful substance use. Manual for use in primary care (2012.0) 21. Heather N ST. The essential and book of treatment and prevention of alcohol patients (2004.0) 22. Huffstetler AN. Brief Interventions and Motivational Interviewing 23. Tucker JS, D’Amico EJ, Ewing BA, Miles JN V, Pedersen ER. A group-based motivational interviewing brief intervention to reduce substance use and sexual risk behavior among homeless young adults. J. Subst. Abuse Treat (2017.0) 76 20-27. PMID: 28340904 24. 24 Center for Substance Abuse Treatment Brief Interventions and Brief Therapies for Substance AbuseRockville (MD)Substance Abuse and Mental Health Services Administration (US)1999Report No. 99-3353. Brief Interventions and Brief Therapies for Substance Abuse (1999.0) 25. Lai DT, Cahill K, Qin Y, Tang JL. Motivational interviewing for smoking cessation.. Cochrane Database Syst. Rev (2010.0) 1 26. Colby SM, Nargiso J, Tevyaw TO, Barnett NP, Metrik J, Lewander W. Enhanced motivational interviewing versus brief advice for adolescent smoking cessation: results from a randomized clinical trial. Addict. Behav (2012.0) 37 817-823. PMID: 22472523 27. Cabezas C, Advani M, Puente D, Rodriguez-Blanco T, Martin C. Effectiveness of a stepped primary care smoking cessation intervention: Cluster randomized clinical trial (ISTAPS study). Addiction (2011.0) 106 1696-1706. PMID: 21561497 28. Lindqvist H, Forsberg LG, Forsberg L, Rosendahl I, Enebrink P, Helgason AR. Motivational interviewing in an ordinary clinical setting: a controlled clinical trial at the Swedish National Tobacco Quitline. Addict. Behav (2013.0) 38 2321-2324. PMID: 23584193 29. Eriksen MB, Frandsen TF. The impact of patient, intervention, comparison, outcome (PICO) as a search strategy tool on literature search quality: a systematic review. J. Med. Libr. Assoc (2018.0) 106 420-431. PMID: 30271283 30. ujika A, Forbes A, Canga N, de Irala J, Serrano I, Gascó P. Motivational interviewing as a smoking cessation strategy with nurses: an exploratory randomised controlled trial. Int. J. Nurs. Stud (2014.0) 51 1074-1082. PMID: 24433609 31. 31 The Cochrane Collaboration Cochrane Risk of Bias Tools for Randomized Trials2011cited 2022 May 25Available from: https://methods.cochrane.org/bias/resources/cochrane-risk-bias-tool. Cochrane Risk of Bias Tools for Randomized Trials (2011.0) 32. Virtanen SE, Zeebari Z, Rohyo I, Galanti MR. Evaluation of a brief counseling for tobacco cessation in dental clinics among Swedish smokers and snus users. A cluster randomized controlled trial (the FRITT study).. Prev. Med. (Baltim) (2015.0) 70 26-32 33. Cook JW, Collins LM, Fiore MC, Smith SS, Fraser D, Bolt DM. Comparative effectiveness of motivation phase intervention components for use with smokers unwilling to quit: a factorial screening experiment. Addiction (2016.0) 111 117-128. PMID: 26582140 34. Steinberg ML, Rosen RL, Versella M V, Borges A, Leyro TM. A Pilot Randomized Clinical Trial of Brief Interventions to Encourage Quit Attempts in Smokers from Socioeconomic Disadvantage. Nicotine Tob. Res (2020.0) 22 1500-1508. PMID: 32161942 35. Ho KY, Li WHC, Wang MP, Lam KKW, Lam TH, Chan SSC. Comparison of two approaches in achieving smoking abstinence among patients in an outpatient clinic: A Phase 2 randomized controlled trial. Patient Educ. Couns (2018.0) 101 885-893. PMID: 29439844 36. Cabriales JA, Suro Maldonado B, Cooper T V. Smoking transitions in a sample of Hispanic daily light and intermittent smokers.. Addict. Behav (2016.0) 62 42-46. PMID: 27310033 37. Krigel SW, Grobe JE, Goggin K, Harris KJ, Moreno JL, Catley D. Motivational interviewing and the decisional balance procedure for cessation induction in smokers not intending to quit. Addict Behav (2017.0) 64 171-178. PMID: 27619008 38. Meyer C, Ulbricht S, Gross B, Kästel L, Wittrien S, Klein G. Adoption, reach and effectiveness of computer-based, practitioner delivered and combined smoking interventions in general medical practices: a three-arm cluster randomized trial. Drug Alcohol Depend (2012.0) 121 124-132. PMID: 21924563 39. Schane RE, Prochaska JJ, Glantz SA. Counseling nondaily smokers about secondhand smoke as a cessation message: a pilot randomized trial. Nicotine Tob. Res (2013.0) 15 334-342. PMID: 22592447 40. Leavens ELS, Meier E, Tackett AP, Miller MB, Tahirkheli NN, Brett EI. The impact of a brief cessation induction intervention for waterpipe tobacco smoking: A pilot randomized clinical trial. Addict Behav (2018.0) 78 94-100. PMID: 29128712 41. Cabriales JA, Cooper T V., Salgado-Garcia F, Naylor N, Gonzalez E. A randomized trial of a brief smoking cessation intervention in a light and intermittent Hispanic sample. Exp. Clin. Psychopharmacol (2012.0) 20 410-419. PMID: 22731733 42. Sayegh CS, Huey SJ, Zara EJ, Jhaveri K. Follow-up treatment effects of contingency management and motivational interviewing on substance use: A meta-analysis. Psychol. Addict. Behav (2017.0) 31 403-414. PMID: 28437121 43. Lindson-Hawley N, Thompson TP, Begh R. Motivational interviewing for smoking cessation. Cochrane Database Syst. Rev (2015.0) 3 44. O’Halloran PD, Blackstock F, Shields N, Holland A, Iles R, Kingsley M. Motivational interviewing to increase physical activity in people with chronic health conditions: a systematic review and meta-analysis. Clin. Rehabil (2014.0) 28 1159-1171. PMID: 24942478 45. Frost H, Campbell P, Maxwell M, O’Carroll RE, Dombrowski SU, Williams B. Effectiveness of Motivational Interviewing on adult behaviour change in health and social care settings: A systematic review of reviews. PLoS One (2018.0) 13 46. Binning J, Woodburn J, Bus SA, Barn R. Motivational interviewing to improve adherence behaviours for the prevention of diabetic foot ulceration. Diabetes Metab. Res. Rev (2019.0) 35 47. Palacio A, Garay D, Langer B, Taylor J, Wood BA, Tamariz L. Motivational Interviewing Improves Medication Adherence: a Systematic Review and Meta-analysis. J Gen Intern Med (2016.0) 8 929-940 48. Wu L, Gao X, Lo ECM, Ho SMY, McGrath C, Wong MCM. Motivational Interviewing to Promote Oral Health in Adolescents. J Adolesc Heal Off Publ Soc Adolesc Med (2017.0) 61 378-384 49. Alperstein D, Sharpe L. The efficacy of motivational interviewing in adults with chronic pain: A meta-analysis and systematic review. J. Pain (2016.0) 17 393-403. PMID: 26639413 50. Madson MB, Loignon AC, Lane C. Training in motivational interviewing: a systematic review. J. Subst. Abuse Treat (2009.0) 36 101-109. PMID: 18657936 51. Davis MF, Shapiro D, Windsor R, Whalen P, Rhode R, Miller HS. Motivational interviewing versus prescriptive advice for smokers who are not ready to quit. Patient Educ. Couns (2011.0) 83 129-133. PMID: 20627440 52. Heckman CJ, Egleston BL, Hofmann MT. Efficacy of motivational interviewing for smoking cessation: a systematic review and meta-analysis. Tob. Control (2010.0) 19 410-46.. PMID: 20675688 53. Apodaca TR, Longabaugh R. Mechanisms of change in motivational interviewing: A review and preliminary evaluation of the evidence.. Addiction (2009.0) 104 705-715. PMID: 19413785 54. McCambridge J, Cunningham JA. The early history of ideas on brief interventions for alcohol. Addiction (2014.0) 109 538-546. PMID: 24354855 55. 55 World Health Organization Screening and brief interventions for substance use problems2012cited 2022 May 25Available from: https://www.who.int/activities/screening-and-brief-interventions-for-substance-use-problems. Screening and brief interventions for substance use problems (2012.0) 56. Collins SE, Witkiewitz K, Kirouac M, Marlatt GA. Preventing Relapse Following Smoking Cessation. Curr. Cardiovasc. Risk Rep (2010.0) 4 421-428. PMID: 26550097 57. 57 Center for Substance Abuse Treatment Rockville (MD)Substance Abuse and Mental Health Services Administration (US)1999Treatment Improvement Protocol (TIP) Series, No. 34.https://www.ncbi.nlm.nih.gov/books/NBK64942/. Treatment Improvement Protocol (TIP) Series, No. 34. (1999.0) 58. Kristenson H, Ohlin H, Hultén-Nosslin MB, Trell E, Hood B. Identification and intervention of heavy drinking in middle-aged men: results and follow-up of 24-60 months of long-term study with randomized controls. Alcohol Clin. Exp. Res (1983.0) 7 203-209. PMID: 6135365 59. Chick J, Lloyd G, Crombie E. Counselling problem drinkers in medical wards: a controlled study. BMJ (1985.0) 290 965-967. PMID: 2858246 60. Babor T, Higgins-Biddle JC. Brief intervention for hazardous and harmful drinking: A manual for use in primary care (2001.0) 61. Sarkar S, Pakhre A, Murthy P, Bhuyan D. Brief Interventions for Substance Use Disorders. Indian J. Psychiatry (2020.0) 62 S290-S298. PMID: 32055071 62. 62 World Health Organization (WHO) The Alcohol, Smoking and Substance Involvement Screening Test (ASSIST): manual for use in primary careGenevaWHO2010cited 2021 Aug 21Available from: https://apps.who.int/iris/handle/10665/44320. The Alcohol, Smoking and Substance Involvement Screening Test (ASSIST): manual for use in primary care (2010.0) 63. Prochaska JO, DiClemente CC. Self change processes, self efficacy and decisional balance across five stages of smoking cessation. Prog. Clin. Biol. Res (1984.0) 156 131-140. PMID: 6473420 64. 64 Royal Australian College of General Practitioners Smoking, nutrition, alcohol, physical activity (SNAP): a population health guide to behavioural risk factors in general practice2nd edMelbourne2015cited 2022 May 25Available from: https://www.racgp.org.au/your- practice/guidelines/snap. Smoking, nutrition, alcohol, physical activity (SNAP): a population health guide to behavioural risk factors in general practice (2015.0) 65. Haber P, Lintzeris N, Proude E LO. Guidelines for the treatment of alcohol problems (2009.0) 66. A clinical practice guideline for treating tobacco use and dependence: 2008 update. A U.S. Public Health Service report. Am. J. Prev. Med (2008.0) 35 158-176 67. Carr AB, Ebbert JO. Interventions for tobacco cessation in the dental setting. A systematic review. Community Dent. Health (2007.0) 24 70-74. PMID: 17615820 68. 68 Maryland: Tobacco Control Resource Center Brief Interventionscited 2021 Aug 21Available from: https://marylandtcrc.org/cessation-programs/brief-interventions. Brief Interventions 69. Stead LF, Bergson G, Lancaster T. Physician advice for smoking cessation. Cochrane database Syst. Rev (2008.0) 2 70. Rodgers C. Brief interventions for alcohol and other drug use. Aust. Prescr (2018.0) 41 117-121. PMID: 30116080 71. Rohsenow DJ, Martin RA, Monti PM, Colby SM, Day AM, Abrams DB. Motivational Interviewing versus Brief Advice for Cigarette Smokers in Residential Alcohol Treatment. J. Subst. Abuse Treat (2014.0) 46 346-346. PMID: 24210533 72. Lindson N, Thompson TP, Ferrey A, Lambert JD, Aveyard P, Group CTA. Motivational interviewing for smoking cessation.. Cochrane Database Syst. Rev (2019.0) 7 73. Matkin W, Ordóñez‐Mena JM, Hartmann‐Boyce J, Group CTA. Telephone counselling for smoking cessation. Cochrane Database Syst. Rev (2019.0) 5 74. Chen D, Wu LT. Smoking cessation interventions for adults aged 50 or older: A systematic review and meta-analysis. Drug Alcohol Depend (2015.0) 154 14-24. PMID: 26094185 75. Manolios E, Sibeoni J, Teixeira M, Révah-Levy A, Verneuil L, Jovic L. When primary care providers and smokers meet: a systematic review and metasynthesis. Prim. Care Respir. Med. (2021.0) 31 1-8 76. Tobacco use and health ICN Position. . International Council of Nurses (2012.0)

Subsets and Splits

No community queries yet

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