https://orcid.org/0000-0003-3753-6717
Figures
Abstract
Introduction
Lifestyle improvements are key modifiable risk factors for Type 2 diabetes mellitus (DM) however specific influences of biologically active dietary metabolites remain unclear. Our objective was to compare non-targeted plasma metabolomic profiles of women with versus without confirmed incident DM. We focused on three lipid classes (fatty acyls, prenol lipids, polyketides).
Materials and methods
Fifty DM cases and 100 individually matched control participants (80% with human immunodeficiency virus [HIV]) were enrolled in a case-control study nested within the Women’s Interagency HIV Study. Stored blood samples (1–2 years prior to DM diagnosis among cases; at the corresponding timepoint among matched controls) were assayed in triplicate for metabolomics. Time-of-flight liquid chromatography mass spectrometry with dual electrospray ionization modes was utilized. We considered 743 metabolomic features in a two-stage feature selection approach with conditional logistic regression models that accounted for matching strata.
Citation: Yu EA, Alemán JO, Hoover DR, Shi Q, Verano M, Anastos K, et al. (2022) Plasma metabolomic analysis indicates flavonoids and sorbic acid are associated with incident diabetes: A nested case-control study among Women’s Interagency HIV Study participants. PLoS ONE 17(7): e0271207. https://doi.org/10.1371/journal.pone.0271207
Editor: Anandh Babu Pon Velayutham, University of Utah, UNITED STATES
Received: November 13, 2021; Accepted: June 26, 2022; Published: July 8, 2022
Copyright: © 2022 Yu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper and its Supporting information files.
Funding: The WIHS is funded by the National Institute of Allergy and Infectious Diseases (UO1-AI-35004, UO1-AI-31834, UO1-AI-34994, UO1-AI-34989, UO1-AI-34993, and UO1-AI-42590) and by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (UO1-HD-32632). The study is co-funded by the National Cancer Institute, the National Institute on Drug Abuse, and the National Institute on Deafness and Other Communication Disorders. Funding is also provided by the National Center for Research Resources (UCSF-CTSI Grant Number UL1 RR024131). This manuscript was also partially supported by NIDDK K08-DK117064 (J.O.A.). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. There was no additional external funding received for this study.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Diabetes mellitus (DM) is associated with an increasingly heavy burden of disease globally [1,2], including among people with human immunodeficiency virus (HIV) [3,4]. Over the last three decades, the number of people with DM more than doubled from 211 million in 1990 to 476 million in 2017 [1]. This increase largely reflects the growing number of people with Type 2 diabetes mellitus (T2DM), which also accounts for most DM cases [1]. A major obstacle to reducing T2DM incidence, prevalence, and mortality is increasing the effectiveness of prevention strategies, including through an improved understanding of modifiable risk factors [5] in diverse phenotypic subgroups.
Lifestyle modifications, including healthier dietary patterns with more fruits and vegetables and fewer processed foods, are key prevention recommendations for reducing the risk of T2DM [2]. Despite a large literature regarding specific diets [6] and nutrients [7] in association with diabetes outcomes, findings across some previous studies are inconsistent [8]. It remains a challenge to account for the extensive inter- and intra-individual heterogeneity in consumption patterns, nutritional requirements, dietary responses (e.g., nutrient absorption) [9] as well as the roles of non-nutrients and other dietary components [10]. Evaluation of dietary interventions, particularly long-term adherence, is a major obstacle. Circulating biomarkers of dietary intake could circumvent these issues and potentially serve as improved metrics of specific biologically-active metabolites and earlier predictors of long-term metabolic health [11–13].
Metabolomics can provide high-throughput, comprehensive, and relatively non-biased examination of low molecular weight metabolites [14]. Metabolomic data have the potential to characterize overall dietary intake and to identify earlier, modifiable dietary risk factors for DM [14]. Branched-chain amino acids and sphingolipids have been extensively evaluated in the context of insulin resistance and DM [15,16]. In a recent study among Women’s Interagency HIV Study (WIHS) participants, cholesteryl esters, diacylglycerols, lysophosphatidylcholines, phosphatidylcholines, and phosphatidylethanolamines were associated with diabetes risk [17].
This individually matched nested case-control study compared non-targeted plasma metabolomic profiles among women with versus without confirmed, incident DM. We evaluated lipids and lipid classes that represent potential dietary modifiable risk factors of DM. Specifically, our focus was on three classes of lipids (fatty acyls, prenol lipids, polyketides) [18].
Materials and methods
Study participants
WIHS was a multicenter prospective cohort study among U.S. women with HIV and women without HIV who had similar risk behaviors as HIV-seropositive women [19,20]. WIHS merged with the Multicenter AIDS Cohort Study (MACS) in 2019 to form the MACS/WIHS Combined Cohort Study [21]. In WIHS, HIV-seronegative women were enrolled based upon having similar risk behaviors as HIV-seropositive women [19,20]. This study included data collected from 3,772 women enrolled at six WIHS consortia (Bronx/Manhattan, NY; Brooklyn, NY; Los Angeles/Southern California/Hawaii; San Francisco/Bay Area, CA; Chicago, IL; Washington, DC) [19]. This nested-case control study included 50 cases and 100 matched controls in the final analytic dataset (S1 Fig).
Data collection
As part of the parent cohort study, participants completed in study visits every six months from October 2000 to April 2008. At baseline and at each semi-annual follow-up visit, women completed questionnaires regarding self-reported sociodemographics, behavioral risk and lifestyle factors. During study visits, trained study staff conducted interviews of medical history including antiretroviral treatment history, and performed physical examinations (e.g., anthropometry) and phlebotomy.
Case (incident diabetes mellitus) and control definitions
We defined women as cases with incident, confirmed DM if they met any of the following criteria: a) ≥ two fasting blood glucose (FBG) ≥126 mg/dL; b) one FBG ≥ 126 mg/dL and one random blood glucose (RBG) ≥ 200 mg/dL; c) one FBG ≥ 126 mg/dL and self-reported DM medications (S1 Table). For each case, the index visit (visit 0) was the visit of DM diagnosis. If participants had two FBG measurements, visit 0 was considered the first date of DM presentation (i.e., first of two DM measurements). All FBG concentrations prior to the index visit were <126 mg/dL. Semiannual visits immediately preceding visit 0 were denoted by the corresponding negative study visit number (e.g., -1 for six months prior, -2 for 12 months prior). We assayed a single stored plasma sample from a study visit between one to two years before the index visit of each case.
We matched every DM case to two controls based on blood glucose, HIV serostatus, use of antiretroviral therapy, race and ethnicity, age ± 15 years, and availability of stored blood sample. To control for metabolic parameters potentially associated with impaired fasting glucose, the first control (“FBG-matched control”) was matched on the case’s FBG ± 10 mg/dL at the same calendar period visit that their corresponding case had an available stored plasma sample. The second control (“normoglycemic control”) had all prior longitudinal glucose values <100 mg/dL and was selected without matching by FBG at the same visit as their corresponding case; this control had a plasma sample available at the same calendar period visit as the case.
Glucose assays
Fasting blood samples were assayed for glucose concentrations by hexokinase assay (Olympus 5200, 5400 and AU600 automated instruments; Olympus America, Inc., Melville, NY), as previously detailed [22].
Metabolomic profiling
Plasma samples were collected in sodium citrate (CPT) vacutainers, centrifuged, and stored at -80°C until thawed for non-targeted metabolomic assays. Plasma samples were randomly sorted by matching strata (DM case, FBG-matched and normoglycemic control) into three sets. Samples in each set were assayed for metabolomic data in a separate run; these three batches are subsequently referred to as WIHS1-3. All sample processing and metabolomic assays were conducted by laboratory technicians blinded to the case or control status of each sample. Initial sample processing to extract metabolites followed the same protocol, which has been previously detailed [23]. Standard operating procedures and quality assurance/quality control of metabolomic assays have also been described [24].
Liquid chromatography-mass spectrometry.
Plasma samples were assayed in triplicate for metabolomic profiles by time-of-flight liquid chromatography mass spectrometry (LC-MS; Model 6250; Agilent Technologies, Santa Clara, CA) with dual electrospray ionization (ESI) modes [24]. Analytes were separated by C18-based reverse phase column (2 mm x 150 mm Zorbax SB Aq 3.5 um column) in positive and negative ESI modes, which enables greater coverage of features [25]. LC parameters included: autosampler temperature 4°C, 5 μL injection volume, column temperature 55°C, and flow rate 0.4 ml/L. The linear gradient was 2–98% of 0.2% (v/v) acetic acid in water (solvent A) to 0.2% (v/v) acetic acid in methanol over 15 min, followed by 2 min hold of solvent B and 5 min post-time. ESI settings included: capillary voltage (Vcap) at 4000 V for positive ion mode and 3500 V for negative ion mode, fragmentor voltage at 135 V, liquid nebulizer at 45 psi, N2 drying gas at 12 L/min and 250°C. Data were acquired by Agilent MassHunter Qual Workstation Data Acquisition software with the following settings: rate 2.5 spectra/s, centroid mode, and mass scan range 15–2250 [26].
Metabolomic data extraction and preprocessing.
Each metabolomic feature was defined by a unique mass-to-charge ratio (m/z) and retention time (RT) combination; relative abundance of feature ion intensities were reported as peak areas. An internal reference standard mix included six standard masses ranging from 112.985587 to 1633.949753; this was utilized for mass axis calibration, error assessments and corrections. Major pre-processing steps included: feature detection and extraction; correlation (co-varying ions within each chromatogram); accounting for adducts, isomers, and fragments.
In terms of data-filtering, metabolomic features with ion counts in >80% across participant samples in each data subset (by assay batch [WIHS1-3] and ESI mode [+, -]) were retained for analysis [27]. Missing relative abundance values (e.g. ≤1) were set to the limit of detection (LOD)/2. All feature ion counts were log2 normalized prior to analysis.
Statistical and bioinformatic analysis
Analysis was conducted utilizing R (version 4.0.3; R Foundation for Statistical Computing; Vienna, Austria), including MetaboAnalystR [28], and SAS (version 9.4; SAS Institute Inc.; Cary, NC, US). Statistical significance was based on two-sided hypothesis tests, and α < 0.05. We initially screened metabolomic features with feature-by-feature unadjusted regressions (Stage 0); since this was a screening criterion, features remained eligible with a p<0.05 that was not false discovery rate adjusted. Subsequently, eligible features were evaluated in feature-by-feature adjusted regressions with metabolomic data (Stage 1); false discovery rate (FDR) adjusted q-value <0.05 was considered significant (S2 Fig). We used a complete-case approach for all key variables aside from metabolomic data (S1 Fig).
Descriptive analysis and visualizations.
Continuous and categorical variables were summarized as medians (interquartile ranges [IQR]) or N’s (percentages). Metabolomic features (i.e., log2 relative abundance) were compared across subgroups by non-parametric test statistics (e.g. Kruskal-Wallis). Log2-normalized feature relative abundances and clinical indicators were evaluated by Spearman rank-order correlation coefficients. We visually compared differences of log2-normalized feature relative abundances between the three case-control groups via unsupervised dimensionality reduction (principal components analysis [PCA]), supervised discriminant analysis approaches (e.g. partial least squares discriminant analysis [PLS-DA], orthogonal PLS-DA [OPLS-DA]), and hierarchical clustering in heatmaps. Heatmaps were based on calculated Euclidean distances as the similarity index with Ward’s linkage as the agglomeration method (clustering based on minimizing sum of squares between any two clusters). We considered permutation test statistics for PLS-DA due to potential overfitting issues.
Metabolomic feature selection approach.
We utilized a two-stage metabolomic feature selection approach to evaluate the associations between features and case-control status in each data subset (by assay batch [WIHS1-3] and ESI mode [+, -]; (S2 Fig). All conditional logistic models considered a binary categorization of DM cases versus both controls as the primary dependent variable of interest and accounted for matching strata, which reflect individual-matching by blood glucose (FBG-matched, normoglycemic), HIV serostatus, use of antiretroviral therapy, race and ethnicity, age ± 15 years, and availability of stored blood sample. In Stage 0 screening, unadjusted conditional logistic regressions models assessed the associations between case-control status and log2 feature relative abundance. Metabolomic features differing across groups (p<0.05) were considered eligible for Stage 1 regression models.
In Stage 1, multivariable conditional logistic regressions evaluated associations between case-control status and log2 feature relative abundance while accounting for the matching strata and additional covariates. The model equation was:
(1)
where p = probability of DM case study group, and z = stratum indicator variables (Eq (1)). Metabolomic features were considered associated with the study group (DM cases vs controls) across groups based on β1 (FDR-adjusted q<0.05). We only reported Stage 1 results from three lipid classes of interest (fatty acyls, prenol lipids, polyketides), in light of recent lipidomics studies focusing on other lipids classes.
Feature annotations.
The putative chemical compound identities of metabolomic features were annotated by comparison with lipids curated from METLIN [29]. Annotations were based on monoisotopic accurate mass match (within ± 10−5). Selected feature annotations were subsequently manually cross-referenced with Lipid Maps [30] and Human Metabolome Database reference database information [31]. We evaluated feature annotation confidence according to the multi-level system proposed by the Schymanski et al [32], which was based on the Metabolomics Standards Initiative (MSI) scoring [33]. Annotations of selected metabolomic features (from adjusted regressions) were considered Levels 2 or 3 [33].
Ethical conduct of research
The Institutional Review Boards (IRBs) at each WIHS site approved of the study protocol and consent forms (IRB approval numbers: Georgetown University #1993–077, Johns Hopkins University H.34.97.05.19.A2, Montefiore Medical Center #03-07-174, Rush University #13–184, State University of New York Downstate Health Sciences University #266921–64, University of California, San Francisco #21–33925, University of Southern California # HS-21-00496). All study participants provided written informed consent in English or Spanish prior to voluntary enrollment and data collection.
Results
One-hundred and fifty women met the inclusion and exclusion criteria and were included in the final analytic dataset. Among these participants, 50 had DM, 50 were FBG-matched controls, and 50 were normoglycemic controls (S1 Fig). Ages ranged from 19 to 62 years at the index study visit; across the three case-control groups, median age ranged from 42 (IQR 36, 48) to 43 (IQR 38, 48; Table 1). In all case-control groups, 80.0% of women had HIV infection (Table 1). Comparing women with HIV infection across the three case-control groups, CD4 cell counts (p = 0.93) and the proportions of women with HIV RNA <400 copies/mL (p = 0.79) were similar (Table 1). Percentages of women on combination antiretroviral therapy (cART), protease inhibitors, stavudine, zidovudine were similar across the three subgroups (all p>0.05; Table 1). Family history of DM was highest among women with DM (61.0%), compared to those in the control subgroups (FBG-matched 28.6%; normoglycemic 43.2%; p = 0.01; Table 1). Median BMI (p = 0.02) and waist circumference (p<0.01) differed across the 3 subgroups (Table 1). Women with DM had the highest median BMI (29.7 kg/m2 [IQR 27.6, 36.5]) and waist circumference (97.4 cm [90.1, 106.5]), compared to the control subgroups (Table 1).
Comparing relative abundance of metabolomic features by diabetes case and controls status
After data-filtering, 743 metabolomic features remained (S1 and S3 Figs). Stratifying by the six data subsets (based on assay batch [WIHS1-3] and ESI mode [+, -]), the number of remaining metabolomic features ranged between 23 and 273 (S1 and S3 Figs). Considering these metabolomic features in a hierarchical clustering heatmap, the similarity indices (Euclidean distances) appeared distinct across the three case-control groups (WIHS1 participants, positive ESI mode; Fig 1A). Visualizing metabolomic features in each data subset, unsupervised (PCA) and supervised (OPLS-DA) approaches showed similar clustering across the three case-control groups (S4 and S5 Figs). Fig 1B shows the first three components from PLS-DA of metabolomic features among WIHS1 participants (positive ESI mode; permutation test statistic p>0.05).
A: Hierarchical clustering heatmap was based on calculated Euclidean distances as the similarity index with Ward’s linkage as the agglomeration method (clustering based on minimizing sum of squares between any two clusters). Log2-normalized relative abundance of metabolomic features are represented in rows; study groups of participants are indicated in columns. DM cases are indicated in red (n = 17), FBG-matched controls in green (n = 17), and normoglycemic controls in blue (n = 17). B: Supervised dimensionality reduction was conducted by PLS-DA, in order to visualize clustering across metabolomic features. Study groups are represented as Δ (DM cases), + (FBG-matched controls), and X (normoglycemic controls). Abbreviations: DM, diabetes mellitus; ESI, electrospray ionization; FBG, fasting blood glucose; PLS-DA, partial least squares discriminant analysis; WIHS, Women’s Interagency HIV Study.
Table 2 summarizes associations between metabolomic features and case-control status (DM cases versus controls), based on unadjusted logistic regressions (Stage 0) with conditional likelihood, stratified by data subset. In WIHS1, three metabolomic features (0 in positive ESI mode; 3 in negative ESI mode) were associated with case-control status (all p<0.05). In WIHS2, seven metabolomic features (2 in positive ESI mode; 5 in negative ESI mode) were associated with case-control status (all p<0.05). In WIHS3, 14 metabolomic features (13 in positive ESI mode; 1 in negative ESI mode) were associated with case-control status (all p<0.05).
Adjusted associations between metabolomic features and diabetogenic subgroups.
In conditional multivariable logistic regressions (Stage 1), 7 metabolomic features were respectively associated with case-control status, accounting for matching strata, BMI, and age (all FDR-adjusted q<0.05; Table 2). Per unit increase, two fatty acyls, 6-methyloctan-3one (adjusted odds ratio [aOR] 1.5 [95% CI 1.0, 2.1]; q = 0.04) and sorbic acid (aOR 2.8 [95% CI 1.1, 7.2]; q = 0.04) were associated with elevated odds of diabetes (Table 3). Per unit increases, four polyketides were respectively associated with odds of diabetes, specifically including heteroflavanone C (aOR 0.1 [95% CI <0.1, 0.8); q = 0.04), rotenonic acid (aOR 0.1 [95% CI <0.1, 0.8); q = 0.04), louisfieserone A (0.2 [95% CI <0.1, 0.8); q = 0.04), and (E)-4-nitrostilbene (aOR 1.5 [95% CI 1.0, 2.4]; q = 0.04; Table 3). Podocarpic acid was associated with increased odds of diabetes (aOR 7.1 [95% CI 1.5, 33.4]; q = 0.02; Table 3). Relative abundance of podocarpic acid was compared by case-control status (Fig 2). Data subsets (assay batch [WIHS1-3], ESI mode [+, -]) are specified in Tables 2 and 3.
a Data subset (e.g. WIHS1 +) specified in Table 3. Abbreviations: DM, diabetes mellitus; FBG, fasting blood glucose.
Discussion
A total of 743 metabolomic features were observed among participants with DM and their controls matched by blood glucose (FBG-matched, normoglycemic), HIV serostatus, use of antiretroviral therapy, race and ethnicity, age ± 15 years, and availability of stored blood sample. Overall, seven features were significantly associated with odds of DM incidence, accounting for matching strata and after FDR adjustment (all q<0.05). Three flavonoids were associated with lower odds of DM incidence, and sorbic acid was associated with greater odds of DM incidence. Our results indicate the need for confirmation of flavonoids, sorbic acid, and their related metabolites via targeted validation with absolute quantitation and mechanistic studies to elucidate their potential respective influences on DM risk.
Protective effects of flavonoids in diabetes
Phytochemicals synthesized by plants and ubiquitous in the human diet, including many flavonoids [34], are hypothesized to be protective against insulin resistance [35] and DM [36], as well as modulate glucose metabolism [37,38]. Our finding that three flavonoids were associated with lower odds of DM is consistent with the directionality of associations found in previous studies [36,39], though our exposure assessment was based on circulating metabolites which differs from dietary intake in other studies. In a meta-analysis including 284,806 participants, dietary intake of total flavonoids was associated with lower risk of T2DM [36]. High dietary intake of flavonoids [39] and adherence to plant-based dietary patterns [40] have also been associated with reduced T2DM risk. Prior studies have suggested potential mechanisms to explain this association, including the ability of some individual flavonoids to inhibit oxidative stress [41] and glycogen phosphorylase, which is a primary enzymatic regulator of glucose and glycogen homeostasis [37]. More broadly, polyphenols have been found to affect glucose and insulin metabolism [42], as well as inhibit glycation and advanced glycation end products production [43].
Previous studies have reported mixed associations, including null results, between diabetogenic indicators and dietary supplementation of isoflavones [44,45]. We found that a circulating isoflavan (isosativan) was associated with greater odds of DM, which contrasts with the null or protective associations observed in other observational studies of dietary isoflavonoid intake on DM-related biomarkers [35,45,46]. These inconsistent findings are potentially explained by the unclear mechanisms linking isoflavonoids and DM, which could include mediators and covariates that need to be accounted for (e.g., extensive heterogeneity of DM pathophysiology, observed pleiotropic influences and differing bioavailabilities of isoflavonoids) [34,35,45].
Elucidating sorbic acid in diabetes
Sorbic acid, or sorbate, is a common synthetic food preservative and metabolite of potassium sorbate, which is a food and pharmaceutical additive [47]. Our finding that sorbate was associated with greater odds of DM is consistent with preliminary evidence of potential explanatory mechanisms [47,48]. Potassium sorbate is completely absorbed after oral ingestion and has cytotoxic and genotoxic influences, which could contribute to elevated risk of a diabetogenic state [47]. Preliminary mechanistic evidence has also shown sorbate to be linked with dysregulated hepatic fatty acid metabolism [48]. Sorbate has also been hypothesized to be an upstream substrate of AGEs [47], which upregulate inflammation and oxidative stress [49] and potentially function as endocrine disrupting chemicals [50]. Future directions of research could examine the: specific metabolic pathways by which sorbic acid and other sorbate additives (e.g., calcium sorbate, potassium sorbate) and other food additives might affect long-term risk of DM incidence, as well as influences of frequency, quantity, timing, and types of sorbates consumed over the human life course on metabolic health.
Strengths and limitations
A major strength of this study was the nested case-control design within a large ongoing prospective cohort study with standardized protocols [19,20]. Specifically, the study design included the confirmation of each participant with incident DM diagnosis after the measurement of metabolomic features; selection of two individually matched controls based on clinical and sociodemographic criteria; and comparison of stored blood samples collected at the same earlier study visit within each matching stratum. The broad consideration of metabolomic features from non-targeted profiling provided a relatively non-biased perspective. This approach was advantageous given limited prior literature regarding the specific lipid classes of interest in context of DM. Furthermore, the inclusion of only women was a strength in light of sex-based differences in metabolism and DM [51]. Simply controlling for biological sex as a variable in regression models does not preclude residual confounding from other related factors (e.g., sex hormone differences), since the etiology of many observed sex-linked differences remains incompletely understood [51].
Several limitations should be noted in interpreting results, particularly the modest sample size, inability to determine causal inferences, and single timepoint evaluation of metabolomic data. In the final analysis, we categorized the two control groups into one group, given the sample size per metabolomic assay batch (WIHS1-3). Further validation of metabolites with authentic reference standards and absolute quantification (plasma concentrations) are needed, in order to confirm feature annotations with higher confidence (e.g., Level 1 [32]) and to facilitate comparisons with other populations. We were not able to consider other covariates, such as inflammation, socioeconomic factors, and ART type, and inter-individual variability of gut microbiota [52,53], that potentially influence our associations of interest; future studies should consider these additional covariates. For example, commensal bacteria have been hypothesized to metabolize dietary flavonoids [54] and to be modulated by polyphenols [55] which may subsequently affect metabolic health. Since HIV status was a matching criterion for selecting controls, this study was not designed to evaluate the role of HIV as a comorbidity. However, some flavonoids have antioxidant functions [34] and a recent study demonstrated that two flavonoid glycosides can activate Vδ1+ T cells to suppress HIV-1 [56], emphasizing the need for future studies to consider the associations of individual flavonoids with DM, HIV, and other comorbidities.
Conclusions
In summary, seven plasma metabolomic features differed among women with DM incidence, compared to their matched controls. Three flavonoids were associated with lower odds of DM incidence. Sorbic acid, a common food preservative, was associated with greater odds of DM. Further studies are needed to validate and delineate the underlying mechanisms of flavonoids and food additives as potential modifiable dietary factors associated with DM, which could improve DM prevention efforts.
Supporting information
S1 Fig. Inclusion and exclusion criteria for WIHS study participants, and data filtering of metabolomic features.
https://doi.org/10.1371/journal.pone.0271207.s001
(TIF)
S3 Fig. Proportions of feature peak areas observed across participants, stratified by metabolomic assay batch (WIHS1-3) and analytical column (+, - ESI).
In each of the six data subsets, the final analytic subset of participants was considered those individuals in complete matching strata. Features were included below if remaining after data filtering (observed among ≥80% of participant samples).
https://doi.org/10.1371/journal.pone.0271207.s003
(TIF)
S4 Fig. Unsupervised clustering (PCA) of metabolomic features in each data subset (WIHS sets 1–3, positive and negative ESI modes).
https://doi.org/10.1371/journal.pone.0271207.s004
(TIF)
S5 Fig. Supervised clustering (OPLS-DA) of metabolomic features in each data subset (WIHS sets 1–3, positive and negative ESI modes).
https://doi.org/10.1371/journal.pone.0271207.s005
(TIF)
Acknowledgments
We thank Kyu Rhee for his assistance with the study design, providing analytical instrumentation, and reviewing the manuscript. The authors gratefully acknowledge the contributions of the study participants and dedication of the staff at the MACS/WIHS Combined Cohort Study sites.
References
- 1. Lin X, Xu Y, Pan X, Xu J, Ding Y, Sun X, et al. Global, regional, and national burden and trend of diabetes in 195 countries and territories: an analysis from 1990 to 2025. Sci Rep. 2020;10(1):14790. pmid:32901098
- 2.
World Health Organization. Diabetes. Fact sheet. Geneva: World Health Organization; 2021.
- 3. American Diabetes Association. 2. Classification and Diagnosis of Diabetes: Standards of Medical Care in Diabetes—2021. Diabetes Care. 2021;44(Supplement 1):S15.
- 4. Monroe AK, Glesby MJ, Brown TT. Diagnosing and managing diabetes in HIV-infected patients: current concepts. Clin Infect Dis. 2015;60(3):453–62. pmid:25313249
- 5. Chan JCN, Lim L-L, Wareham NJ, Shaw JE, Orchard TJ, Zhang P, et al. The Lancet Commission on diabetes: using data to transform diabetes care and patient lives. Lancet. 2020;396(10267):2019–82. pmid:33189186
- 6. Sarsangi P, Salehi-Abargouei A, Ebrahimpour-Koujan S, Esmaillzadeh A. Association between adherence to the Mediterranean diet and risk of type 2 diabetes: An updated systematic review and dose-response meta-analysis of prospective cohort studies. Adv Nutr. 2022; nmac046. pmid:35472102
- 7. Zheng Y, Li Y, Qi Q, Hruby A, Manson JE, Willett WC, et al. Cumulative consumption of branched-chain amino acids and incidence of type 2 diabetes. Int J Epidemiol. 2016;45(5):1482–92. pmid:27413102
- 8. Mustafa ST, Hofer OJ, Harding JE, Wall CR, Crowther CA. Dietary recommendations for women with gestational diabetes mellitus: a systematic review of clinical practice guidelines. Nutr Rev. 2021;79(9):988–1021. pmid:33677540
- 9. Lampe JW, Navarro SL, Hullar MAJ, Shojaie A. Inter-individual differences in response to dietary intervention: integrating omics platforms towards personalised dietary recommendations. Proc Nutr Soc. 2013;72(2):207–18. pmid:23388096
- 10. Yates AA, Dwyer JT, Erdman JW Jr., King JC, Lyle BJ, Schneeman BO, et al. Perspective: Framework for developing recommended intakes of bioactive dietary substances. Adv Nutr. 2021;12(4):1087–99. pmid:33962461
- 11. Roberts LD, Koulman A, Griffin JL. Towards metabolic biomarkers of insulin resistance and type 2 diabetes: progress from the metabolome. Lancet Diabetes Endocrinol. 2014;2(1):65–75. pmid:24622670
- 12. Bhupathiraju SN, Hu FB. One (small) step towards precision nutrition by use of metabolomics. Lancet Diabetes Endocrinol. 2017;5(3):154–5. pmid:28089710
- 13. Rinschen MM, Ivanisevic J, Giera M, Siuzdak G. Identification of bioactive metabolites using activity metabolomics. Nat Rev Mol Cell Biol. 2019;20(6):353–67. pmid:30814649
- 14. Newgard CB. Metabolomics and metabolic diseases: Where do we stand? Cell Metab. 2017;25(1):43–56. pmid:28094011
- 15. Guasch-Ferré M, Hruby A, Toledo E, Clish CB, Martínez-González MA, Salas-Salvadó J, et al. Metabolomics in prediabetes and diabetes: A systematic review and meta-analysis. Diabetes Care. 2016;39(5):833. pmid:27208380
- 16. White PJ, Newgard CB. Branched-chain amino acids in disease. Science. 2019;363(6427):582. pmid:30733403
- 17. Zhang E, Chai JC, Deik AA, Hua S, Sharma A, Schneider MF, et al. Plasma lipidomic profiles and risk of diabetes: 2 prospective cohorts of HIV-infected and HIV-uninfected individuals. J Clin Endocrinol Metab. 2021;106(4):999–1010. pmid:33420793
- 18. O’Donnell VB, Dennis EA, Wakelam MJO, Subramaniam S. LIPID MAPS: Serving the next generation of lipid researchers with tools, resources, data, and training. Sci Signal. 2019;12(563).
- 19. Bacon MC, von Wyl V, Alden C, Sharp G, Robison E, Hessol N, et al. The Women’s Interagency HIV Study: an observational cohort brings clinical sciences to the bench. Clin Diagn Lab Immunol. 2005;12(9):1013–9. pmid:16148165
- 20. Barkan SE, Melnick SL, Preston-Martin S, Weber K, Kalish LA, Miotti P, et al. The Women’s Interagency HIV Study. WIHS Collaborative Study Group. Epidemiology. 1998;9(2):117–25. pmid:9504278
- 21. D’Souza G, Bhondoekhan F, Benning L, Margolick JB, Adedimeji AA, Adimora AA, et al. Characteristics of the MACS-WIHS combined cohort study: Opportunities for research on aging with HIV in the longest US observational study of HIV. Am J Epidemiol. 2021; 190(8):1457–1475. pmid:33675224
- 22. Glesby MJ, Hoover DR, Shi Q, Danoff A, Howard A, Tien P, et al. Glycated haemoglobin in diabetic women with and without HIV infection: data from the Women’s Interagency HIV Study. Antivir Ther. 2010;15(4):571–7. pmid:20587850
- 23. Want EJ, O’Maille G, Smith CA, Brandon TR, Uritboonthai W, Qin C, et al. Solvent-dependent metabolite distribution, clustering, and protein extraction for serum profiling with mass spectrometry. Anal Chem. 2006;78(3):743–52. pmid:16448047
- 24. Lakshmanan V, Rhee KY, Wang W, Yu Y, Khafizov K, Fiser A, et al. Metabolomic analysis of patient plasma yields evidence of plant-like α-linolenic acid metabolism in Plasmodium falciparum. J Infect Dis. 2012;206(2):238–48. pmid:22566569
- 25. Liigand P, Kaupmees K, Haav K, Liigand J, Leito I, Girod M, et al. Think negative: Finding the best electrospray ionization/MS mode for your analyte. Anal Chem. 2017;89(11):5665–8. pmid:28489356
- 26. Sana TR, Waddell K, Fischer SM. A sample extraction and chromatographic strategy for increasing LC/MS detection coverage of the erythrocyte metabolome. J Chromatogr B Analyt Technol Biomed Life Sci. 2008;871(2):314–21. pmid:18495560
- 27. Smilde AK, van der Werf MJ, Bijlsma S, van der Werff-van der Vat BJ, Jellema RH. Fusion of mass spectrometry-based metabolomics data. Anal Chem. 2005;77(20):6729–36. pmid:16223263
- 28. Chong J, Soufan O, Li C, Caraus I, Li S, Bourque G, et al. MetaboAnalyst 4.0: towards more transparent and integrative metabolomics analysis. Nucleic Acids Res. 2018;46(W1):W486–w94. pmid:29762782
- 29. Guijas C, Montenegro-Burke JR, Domingo-Almenara X, Palermo A, Warth B, Hermann G, et al. METLIN: A technology platform for identifying knowns and unknowns. Anal Chem. 2018;90(5):3156–64. pmid:29381867
- 30. Liebisch G, Fahy E, Aoki J, Dennis EA, Durand T, Ejsing CS, et al. Update on LIPID MAPS classification, nomenclature, and shorthand notation for MS-derived lipid structures. J Lipid Res. 2020;61(12):1539–55. pmid:33037133
- 31. Wishart DS, Jewison T, Guo AC, Wilson M, Knox C, Liu Y, et al. HMDB 3.0—The Human Metabolome Database in 2013. Nucleic Acids Res. 2013;41(Database issue):D801–7. pmid:23161693
- 32. Schymanski EL, Jeon J, Gulde R, Fenner K, Ruff M, Singer HP, et al. Identifying small molecules via high resolution mass spectrometry: Communicating confidence. Environ Sci Technol. 2014;48(4):2097–8. pmid:24476540
- 33. Sumner LW, Amberg A, Barrett D, Beale MH, Beger R, Daykin CA, et al. Proposed minimum reporting standards for chemical analysis Chemical Analysis Working Group (CAWG) Metabolomics Standards Initiative (MSI). Metabolomics. 2007;3(3):211–21. pmid:24039616
- 34. Manach C, Scalbert A, Morand C, Rémésy C, Jiménez L. Polyphenols: food sources and bioavailability. Am J Clin Nutr. 2004;79(5):727–47. pmid:15113710
- 35. Duru KC, Kovaleva EG, Danilova IG, van der Bijl P, Belousova AV. The potential beneficial role of isoflavones in type 2 diabetes mellitus. Nutr Res. 2018;59:1–15. pmid:30442228
- 36. Liu Y-J, Zhan J, Liu X-L, Wang Y, Ji J, He Q-Q. Dietary flavonoids intake and risk of type 2 diabetes: A meta-analysis of prospective cohort studies. Clin Nutr. 2014;33(1):59–63. pmid:23591151
- 37. Jakobs S, Fridrich D, Hofem S, Pahlke G, Eisenbrand G. Natural flavonoids are potent inhibitors of glycogen phosphorylase. Mol Nutr Food Res. 2006;50(1):52–7. pmid:16317787
- 38. Kerimi A, Jailani F, Williamson G. Modulation of cellular glucose metabolism in human HepG2 cells by combinations of structurally related flavonoids. Mol Nutr Food Res. 2015;59(5):894–906. pmid:25712349
- 39. Bondonno NP, Dalgaard F, Murray K, Davey RJ, Bondonno CP, Cassidy A, et al. Higher habitual flavonoid intakes are associated with a lower incidence of diabetes. J Nutr. 2021. pmid:34313759
- 40. Qian F, Liu G, Hu FB, Bhupathiraju SN, Sun Q. Association between plant-based dietary patterns and risk of type 2 diabetes: A systematic review and meta-analysis. JAMA Intern Med. 2019;179(10):1335–44. pmid:31329220
- 41. Huang S-M, Wu C-H, Yen G-C. Effects of flavonoids on the expression of the pro-inflammatory response in human monocytes induced by ligation of the receptor for AGEs. Mol Nutr Food Res. 2006;50(12):1129–39. pmid:17103373
- 42. Momtaz S, Salek-Maghsoudi A, Abdolghaffari AH, Jasemi E, Rezazadeh S, Hassani S, et al. Polyphenols targeting diabetes via the AMP-activated protein kinase pathway; future approach to drug discovery. Crit Rev Clin Lab Sci. 2019;56(7):472–92. pmid:31418340
- 43. Anwar S, Khan S, Almatroudi A, Khan AA, Alsahli MA, Almatroodi SA, et al. A review on mechanism of inhibition of advanced glycation end products formation by plant derived polyphenolic compounds. Mol Biol Rep. 2021;48(1):787–805. pmid:33389535
- 44. González S, Jayagopal V, Kilpatrick ES, Chapman T, Atkin SL. Effects of isoflavone dietary supplementation on cardiovascular risk factors in type 2 diabetes. Diabetes Care. 2007;30(7):1871. pmid:17468359
- 45. Cao H, Ou J, Chen L, Zhang Y, Szkudelski T, Delmas D, et al. Dietary polyphenols and type 2 diabetes: Human Study and Clinical Trial. Crit Rev Food Sci Nutr. 2019;59(20):3371–9. pmid:29993262
- 46. Rienks J, Barbaresko J, Oluwagbemigun K, Schmid M, Nöthlings U. Polyphenol exposure and risk of type 2 diabetes: dose-response meta-analyses and systematic review of prospective cohort studies. Am J Clin Nutr. 2018;108(1):49–61. pmid:29931039
- 47. Dehghan P, Mohammadi A, Mohammadzadeh-Aghdash H, Ezzati Nazhad Dolatabadi J. Pharmacokinetic and toxicological aspects of potassium sorbate food additive and its constituents. Trends Food Sci Technol. 2018;80:123–30.
- 48. Chia-Hui C, Sin-Ni H, Po-An H, Yu Ru K, Tzong-Shyuan L. Food preservative sorbic acid deregulates hepatic fatty acid metabolism. J Food Drug Anal. 2020;28(2):12–22.
- 49. Rungratanawanich W, Qu Y, Wang X, Essa MM, Song B-J. Advanced glycation end products (AGEs) and other adducts in aging-related diseases and alcohol-mediated tissue injury. Exp Mol Med. 2021;53(2):168–88. pmid:33568752
- 50. Ravichandran G, Lakshmanan DK, Raju K, Elangovan A, Nambirajan G, Devanesan AA, et al. Food advanced glycation end products as potential endocrine disruptors: An emerging threat to contemporary and future generation. Environ Int. 2019;123:486–500. pmid:30622074
- 51. Tramunt B, Smati S, Grandgeorge N, Lenfant F, Arnal J-F, Montagner A, et al. Sex differences in metabolic regulation and diabetes susceptibility. Diabetologia. 2020;63(3):453–61. pmid:31754750
- 52. Watanabe D, Murakami H, Ohno H, Tanisawa K, Konishi K, Tsunematsu Y, et al. Association between dietary intake and the prevalence of tumourigenic bacteria in the gut microbiota of middle-aged Japanese adults. Sci Rep. 2020;10(1):15221. pmid:32939005
- 53. Kim CH. Microbiota or short-chain fatty acids: which regulates diabetes? Cell Mol Immunol. 2018;15(2):88–91. pmid:28713163
- 54. Pei R, Liu X, Bolling B. Flavonoids and gut health. Curr Opin Biotechnol. 2020;61:153–9. pmid:31954357
- 55. Anhê FF, Choi BSY, Dyck JRB, Schertzer JD, Marette A. Host–microbe interplay in the cardiometabolic benefits of dietary polyphenols. Trends Endocrinol Metab. 2019;30(6):384–95. pmid:31076221
- 56. Yonekawa M, Shimizu M, Kaneko A, Matsumura J, Takahashi H. Suppression of R5-type of HIV-1 in CD4+ NKT cells by Vδ1+ T cells activated by flavonoid glycosides, hesperidin and linarin. Sci Rep. 2019;9(1):7506. pmid:31101837