ABSTRACT
Background
Insulin resistance is a complex metabolic disorder and is often associated with type 2 diabetes (T2D).
Objectives
The aim of this study was to test whether baseline metabolites can additionally improve the prediction of insulin resistance beyond classical risk factors. Furthermore, we examined whether a multimetabolite model predicting insulin resistance in nondiabetics can also predict incident T2D.
Methods
We used a case-cohort study nested within the Prevención con Dieta Mediterránea (PREDIMED) trial in subsets of 700, 500, and 256 participants without T2D at baseline and 1 and 3 y. Fasting plasma metabolites were semiquantitatively profiled with liquid chromatography–tandem mass spectrometry. We assessed associations between metabolite concentrations and the homeostasis model of insulin resistance (HOMA-IR) through the use of elastic net regression analysis. We subsequently examined associations between the baseline HOMA-IR–related multimetabolite model and T2D incidence through the use of weighted Cox proportional hazard models.
Results
We identified a set of baseline metabolites associated with HOMA-IR. One-year changes in metabolites were also significantly associated with HOMA-IR. The area under the curve was significantly greater for the model containing the classical risk factors and metabolites together compared with classical risk factors alone at baseline [0.81 (95% CI: 0.79, 0.84) compared with 0.69 (95% CI: 0.66, 0.73)] and during a 1-y period [0.69 (95% CI: 0.66, 0.72) compared with 0.57 (95% CI: 0.53, 0.62)]. The variance in HOMA-IR explained by the combination of metabolites and classical risk factors was also higher in all time periods. The estimated HRs for incident T2D in the multimetabolite score (model 3) predicting high HOMA-IR (median value or higher) or HOMA-IR (continuous) at baseline were 2.00 (95% CI: 1.58, 2.55) and 2.24 (95% CI: 1.72, 2.90), respectively, after adjustment for T2D risk factors.
Conclusions
The multimetabolite model identified in our study notably improved the predictive ability for HOMA-IR beyond classical risk factors and significantly predicted the risk of T2D.
Keywords: metabolomics, insulin resistance, type 2 diabetes, prediction, PREDIMED
Introduction
Insulin resistance is a complex metabolic disorder characterized by the reduced responsiveness of certain tissues, mainly muscle, liver, and adipose tissue, to insulin signaling (1). Insulin resistance is a risk factor for type 2 diabetes (T2D) (2) due to mechanisms in which metabolic abnormalities play a leading role. Metabolites comprising amino acids, sugars and glycolysis intermediates, ketone bodies, methyl donors, and lipid and phospholipid constituents have been associated with insulin resistance (3, 4).
For example, several plasma amino acids, including branched-chain amino acids (BCAAs) and aromatic amino acids, alanine, proline, and glutamate, have been associated with insulin resistance (5, 6), implying a dysregulation of their metabolism. Concerning lipid classes, Rhee et al. (7) found significant positive correlations between plasma triacylglycerols (TGs) of a relatively lower acyl carbon number and lower double bond content and the HOMA-IR. Other studies (8) reported further relations between various plasma lipid species [TGs, phosphatidylcholines, sphingomyelins (SMs), lysophosphatidylcholines (LPCs), and lysophosphatidylethanolamines] and HOMA-IR. Furthermore, a deregulation of fatty acid oxidation and mitochondrial function due to insulin resistance may lead to higher concentrations of circulating acylcarnitine species (9).
Previous findings (10) have shown the complex relations between metabolism and insulin resistance, showing variance due to genotype, sex, or dietary patterns, and indicating that systemic metabolic analysis is essential in order to gain a deeper understanding of the metabolic perturbations associated with diabetes development. In addition, examining the predictive ability of metabolites involved in different pathways, beyond classical risk factors, may lead to individual biomarkers or sets of biomarkers that may improve the prediction of diabetes risk (11).
Using metabolite profiling, we examined associations between plasma concentrations of metabolites with HOMA-IR at baseline, as well as after 1 and 3 y of follow-up, in participants in the Prevención con Dieta Mediterránea (Prevention of Disease with Mediterranean Diet; PREDIMED) study. We further assessed whether changes in metabolite concentrations were associated with changes in HOMA-IR after 1 y of follow-up. In addition, we investigated the predictive performance of metabolites for insulin resistance index beyond classical risk factors. Finally, in order to ascertain the role of metabolic abnormalities associated with an insulin-resistant state in further T2D development, we investigated whether metabolites predicting HOMA-IR at baseline could also be predictors of T2D incidence after a median 3.8-y follow-up.
Methods
Study design and participants
This is a substudy of a case-cohort study nested within the PREDIMED trial (ISRCTN35739639). A detailed description of the PREDIMED trial can be found elsewhere (12, 13), and more details in relation to the case-cohort study and substudy are included in Supplemental Figure 1. For the present study, 700 out of 892 subjects from the case-cohort study, without T2D at study inception and with available blood samples, were included. Of these, 500 and 256 participants out of the 700 had available samples and did not develop T2D after 1 and 3 y of follow-up, respectively, and were included in the 1- and 3-y analyses. The American Diabetes Association criteria (14), namely 2 confirmations of fasting plasma glucose ≥7.0 mmol/L or 2-h plasma glucose ≥11.1 mmol/L after a 75-g oral-glucose load, were used to ascertain T2D. The institutional review boards of the recruitment centers approved the study protocol, and participants provided written informed consent.
Metabolomics
Fasting (≥8 h) plasma samples were collected in EDTA-coated tubes, at baseline and after 1 y of follow-up and stored at −80°C. A detailed description of the methods used for metabolomic analyses can be found in the Supplemental Methods(15). Metabolites tested for association with HOMA-IR are listed in Supplemental Table 1.
Biochemical measurements
Fasting blood glucose and insulin concentrations were assessed at baseline and at the end of 1 and 3 y of follow-up. Glucose was measured by an enzymatic method that involved converting glucose to 6-phosphogluconate (ADVIA Chemistry System). The intra- and interassay coefficients were 1.2 and 1.6, respectively. Insulin concentrations were measured by an immunoenzymometric assay (ADVIA Centaur XPT) with intra- and interassay CVs of 3.7 and 4.4, respectively. Insulin resistance was estimated with the following equation (16): HOMA-IR = [fasting insulin (μIU/mL) × fasting glucose (mmol/L)]/22.5.
Assessment of other variables
At baseline, a 47-item questionnaire about lifestyle variables, smoking status, medical history, and medication use was administered. Physical activity was assessed through the use of a validated Spanish version of the Minnesota Leisure-Time Physical Activity Questionnaire (17). BMI was calculated as weight divided by height squared (kg/m2). An anthropometric tape was used to measure waist circumference (WC), which was measured midway between the lowest rib and the iliac crest. To assess the degree of adherence to a Mediterranean diet, a 14-item validated questionnaire was filled in for each participant (18). Participants’ TG, total-cholesterol, HDL-cholesterol, and LDL-cholesterol concentrations were measured from fasting plasma at baseline.
Statistical analysis
Baseline characteristics of study participants are described as means ± SDs for quantitative variables and percentages or numbers for categorical variables. Missing values of individual metabolites were imputed (in those metabolites with <20% of missing values) through the use of the random forest imputation approach (“missForest” R package). The concentrations of 134 metabolites were normalized and scaled to multiples of 1 SD with the rank-based inverse normal transformation. Due to the high dimensionality and collinear nature of the data, logistic regression with elastic net penalty implemented in the “glmnet” R package (α = 0.5) was used to build a predictive model for high HOMA-IR (median value or higher) in 3 different time periods (baseline and 1 and 3 y). We also tested the associations between 1-y changes in metabolite concentrations and 1-y changes in HOMA-IR (binary variable: median or lower and higher than median value in changes). We performed 10-fold cross-validation to find the optimal value of the tuning parameter (λ) that yielded the minimum mean-squared error (minMSE) (19). The values minMSE and minMSE + 1 SE were calculated using argument s = “lambda.min” in the cv.glmnet function (“glmnet” R package) because this gives the minimum mean cross-validated error. The predictive model scores were computed as the weighted sum of all metabolites with weights equal to the regression coefficients from the predictive models. To estimate the prediction accuracy we split the data into 90% and 10% sets. Within the 90% set, we used the same elastic net procedure we used to build the model. Another 10-fold cross-validation was used to tune the model parameters. Then, we used the outer 10% set to evaluate the model built at the previous step. This procedure guarantees that the outer 10% set is completely separated from the model-building procedure, so the predicting accuracy estimated in this step is unbiased. We then repeated all these steps 10 times and averaged their prediction accuracy. Because each of them is an unbiased estimate of prediction accuracy, the average is also unbiased. We assessed the predictive performance of the multimetabolite score together with a model consisting of only classical risk factors (sex, age, BMI, WC, physical activity, and smoking) for high HOMA-IR. To predict 1- and 3-y changes in HOMA-IR, we also added baseline HOMA-IR to the model. Logistic regression analysis was performed, and the derived coefficients were used to build the model consisting of only classical risk factors or classical risk factors together with multimetabolite score. The AUC was used to assess the discriminating power of the prediction models. We used a nonparametric method to compare the AUCs of these models (20). The ORs of high HOMA-IR per 1-SD increment of models consisting of classical risk factors alone or together with the multimetabolite score were calculated by logistic regression models. To test the robustness of our results we computed the multimetabolite model at baseline and then tested its performance at each of the time points (1 and 3 y) using the whole samples. To address issues such as statistical power we conducted a sensitivity analysis. We treated HOMA-IR as a continuous variable (log-transformed prior to analyses), and linear regression with elastic net penalty was performed in order to examine associations with metabolites. The variance of HOMA-IR explained by the metabolites was estimated from the adjusted R2 by including all selected metabolites along with classical factors in the model. The additional variance explained was the increase in adjusted R2 over the variance explained by classical risk factors alone. These analyses were performed with the R statistical package version 3.1.1 (R Development Core Team, 2011; http://cran.r-project.org). We used Cox proportional hazard models, with Barlow weights (inverse probability weights to account for the overrepresentation of cases) to estimate HRs and their 95% CIs for risk of T2D. Person-time of follow-up was calculated as the interval between the baseline date and date of T2D event, death, or date of the last participant contact, whichever came first. A multivariable-adjusted Cox regression model was fitted as follows: multivariable model, adjusted for age (years), sex (male or female), BMI (kg/m2), WC (centimeters), intervention group, baseline fasting glucose (milligrams per deciliter) (adding a quadratic term to account for the departure from linearity), smoking (never, current, or former), leisure-time physical activity (metabolic equivalent tasks in minutes per day), baseline dyslipidemia (yes or no), and hypertension (yes or no). We stratified the models according to recruitment center. Some imperfections in the randomization procedures affecting a small subset of participants in the PREDIMED trial have been recently reported (12). For this reason, we conducted the analyses on the associations of baseline models' scores with the risk of T2D with the use of a robust variance to account for clusters (clinics and members of the same household). In these analyses, multimetabolite scores predicting either high HOMA-IR or continuous HOMA-IR at baseline using the whole sample were analyzed as continuous variables (1-SD increment in their scores). These analyses were performed with Stata 13.1 (StataCorp). A 2-sided P value <0.05 was considered significant.
Results
Participants’ characteristics
Participants’ characteristics are summarized in Table 1. The mean ± SD age of participants at baseline was 66.5 ± 5.8 y, and the mean ± SD BMI was 29.9 ± 3.4. Compared with the low–HOMA-IR group (less than the median), those participants with a high HOMA-IR (median or higher) also had higher BMI and WC, in addition to higher fasting glucose and TGs and lower HDL cholesterol (Table 1).
TABLE 1.
Characteristic | Total (n = 700) | Low HOMA-IR (less than the median) | High HOMA-IR (median or higher) |
---|---|---|---|
HOMA-IR | 3.9 ± 2.4 | 2.2 ± 0.6 | 5.5 ± 2.32 |
Age, y | 66.5 ± 5.8 | 66.4 ± 6.0 | 66.6 ± 5.5 |
BMI, kg/m2 | 29.9 ± 3.4 | 28.8 ± 3.0 | 31.1 ± 3.53 |
Waist circumference, cm | 100.0 ± 10.5 | 96.6 ± 10.4 | 103.5 ± 9.53 |
Physical activity, MET-min/d | 243.4 ± 237.9 | 249.8 ± 227.4 | 237.1 ± 248.2 |
Smoking, % | |||
Never | 60.0 | 59.7 | 60.3 |
Former | 22.4 | 21.4 | 23.4 |
Current | 17.6 | 18.9 | 16.3 |
Fasting blood glucose, mg/dL | 104.0 ± 16.5 | 96.8 ± 13.0 | 111.3 ± 16.62 |
Total cholesterol, mg/dL | 224.8 ± 38.0 | 226.2 ± 39.1 | 223.4 ± 36.9 |
HDL cholesterol, mg/dL | 55.4 ± 13.1 | 58.5 ± 13.2 | 52.2 ± 12.32 |
LDL cholesterol, mg/dL | 142.8 ± 32.3 | 144.7 ± 34.0 | 140.8 ± 2.2 |
Triacylglycerols, mg/dL | 135.4 ± 68.2 | 118.8 ± 65.5 | 152.7 ± 66.72 |
Values are means ± SDs unless otherwise indicated. MET-min, metabolic equivalent minutes.
P < 0.001 (Mann-Whitney U test).
P < 0.001 (Student's t test).
Associations of baseline metabolites and 1-y changes with HOMA-IR (binary)
Supplemental Figure 1A–D shows both the number of metabolites in the models and minMSE as functions of λ. The figure also shows the location of minMSE and minMSE + 1 SE. Table 2 shows selected metabolites ranked from the highest to the lowest elastic net positive regression coefficients for high HOMA-IR at baseline and 1 and 3 y. Eighteen metabolites were positively associated with baseline HOMA-IR, 12 metabolites with 1-y HOMA-IR, and 14 metabolites with 3-y HOMA-IR. The highest positive regression coefficient predicting baseline HOMA-IR was obtained for methyladenosine, whereas for 1- and 3-y HOMA-IR the strongest predictor was isoleucine. Positive associations were also found between 1-y changes in the concentrations of 9 metabolites and 1-y changes in HOMA-IR and the highest regression coefficient was observed for N-acetyl-tryptophan. Several metabolites, including isoleucine and proline, were selected in all 3 time periods: methyladenosine, alanine, C4 carnitine, cyclohexylamine, xanthine, and N-acetyl-leucine were selected at baseline and 1 y, whereas dimethylglycine was selected at baseline and 3 y. Hypoxanthine and thiamin were selected at both 1 and 3 y. Metabolites ranked from the highest to the lowest elastic net negative regression coefficients for high HOMA-IR at baseline and 1 and 3 y are presented in Table 3. Twenty-four metabolites were selected for baseline HOMA-IR, 17 metabolites for 1-y HOMA-IR, and 14 metabolites for 3-y HOMA-IR. The highest negative regression coefficient for baseline HOMA-IR was obtained for glycine, whereas for 1 y it was C18:1 LPC, and for 3 y it was guanidoacetic acid. Negative associations were also noted between 1-y changes in 13 metabolites and 1-y changes in HOMA-IR. The highest regression coefficient was found for C18:1 LPC. Several metabolites, including glycine, C16:0 SM, C18:1 LPC, amino-isobutyric acid, C10:2 carnitine, and betaine, were selected in all 3 time periods, whereas symmetric dimethylarginine, arginine, acetylglycine, C9 carnitine, and C16:0 lysophosphatidylethanolamine were selected at baseline and 1 y.
TABLE 2.
Baseline HOMA-IR | HOMA-IR at 1 y | HOMA-IR at 3 y | 1-y changes in HOMA-IR |
---|---|---|---|
Methyladenosine | Isoleucine | Isoleucine | N-Acetyl-tryptophan |
Alanine | Methyladenosine | Dimethylglycine | Methylguanosine |
Valine | Alanine | N-Carbamoyl-β-alanine | Valine |
C14:0 LPC | C4 Carnitine | Proline | C4 Carnitine |
C4 Carnitine | Hypoxanthine | Pseudouridine | Methylhistidine |
C3-DC-CH3 Carnitine | C3 Carnitine | Hypoxanthine | Alanine |
C5 Carnitine | N-Acetyl-leucine | Glutamate | Tyrosine |
C14:0 SM | 5-Hydroxyindoleacetic acid | Uric acid | C7 Carnitine |
Cyclohexylamine | Cyclohexylamine | Thiamin | C26 Carnitine |
Hydroxyproline | Thiamin | Adenosine | |
Isoleucine | Xanthine | Sphinganine | |
N-Acetyl-arginine | Proline | Serotonin | |
Xanthine | C5 carnitine | ||
Proline | Cortisol | ||
Dimethylglycine | |||
N-Acetyl-leucine | |||
3-Hydroxyanthranilic acid | |||
C2 Carnitine |
Cutoff values of median of HOMA-IR are ≥3.32 at baseline, ≥3.42 at 1 y, and ≥3.38 at 3 y. DC-CH3, decanoylcarnitine-methyl ; LPC, lysophosphatidylcholine; SM, sphingomyelin.
TABLE 3.
Baseline HOMA-IR | HOMA-IR at 1 y | HOMA-IR at 3 y | 1-y changes in HOMA-IR |
---|---|---|---|
Glycine | C18:1 LPC | Guanidoacetic acid | C18:1 LPC |
Asparagine | Glycine | Methylhistidine | β-Amino-isobutyric acid |
C16:0 SM | C10:2 Carnitine | C18:1 LPC | C16:1 LPC Plasmalogen |
C18:1 LPC | C16:0 SM | C16:0 SM | C18:2 Carnitine |
β-Amino-isobutyric acid | SDMA | Betaine | C20:4 LPC |
Choline | Betaine | Glycine | C20:4 Carnitine |
Histidine | Arginine | C10:2 Carnitine | Ornithine |
C20:4 LPC | Guanidoacetic acid | β-Amino-isobutyric acid | Asparagine |
SDMA | β-Amino-isobutyric acid | Acetylcholine | Glycine |
Inosine | Acetylglycine | C18 Carnitine | Methylnicotinamide |
Arginine | C9 Carnitine | C18:1-OH Carnitine | SDMA |
C10:2 Carnitine | C18:2 Carnitine | C22:6 LPC | Cotinine |
C10 Carnitine | Phosphocholine | Dehydroxycarnitine | Kynurenic acid |
Acetylglycine | Methoxytyramine | Acetaminophen | |
Ectoine | Threonine | ||
Betaine | l-NMMA | ||
Niacinamide | C16:0 LPE | ||
C20:4 carnitine | |||
Trimethylbenzene | |||
TMAO | |||
Cotinine | |||
C22:6 LPC | |||
C9 Carnitine | |||
C16:0 LPE |
Cutoff values of median of HOMA-IR are ≥3.32 at baseline, ≥3.42 at 1 y, and ≥3.38 at 3 y. LPC, lysophosphatidylcholine; LPE, lysophosphatidylethanolamine; l-NMMA, NG-monomethyl-l-arginine; SDMA, symmetric dimethylarginine; SM, sphingomyelin; TMAO, trimethylamine N-oxide.
Prediction of high HOMA-IR
The predictive models were constructed with the selected metabolites with the use of the elastic net model and classical risk factors. The ORs associated with a 1-SD increase in the classical model score were 2.30 (95% CI: 1.93, 2.75) for baseline HOMA-IR, 5.54 (95% CI: 4.19, 7.35) for 1-y HOMA-IR, and 3.97 (95% CI: 2.79, 5.63) for 3-y HOMA-IR; for 1-y changes in HOMA-IR the OR was 1.55 (95% CI: 1.28, 1.87). The respective ORs associated with a 1-SD increase in the full model including the set of metabolites identified (classical risk factors + metabolites) were 5.44 (95% CI: 4.30, 6.88) for baseline HOMA-IR, 6.90 (95% CI: 5.10, 9.35) for 1-y HOMA-IR, and 6.25 (95% CI: 4.16, 9.39) for 3-y HOMA-IR; for 1-y changes in HOMA-IR the OR was 2.99 (95% CI: 2.38, 3.76) (Table 4).
TABLE 4.
HOMA-IR | Model 1 | Model 2 | Model 3 |
---|---|---|---|
Baseline | 2.30 (1.93, 2.75) | 5.44 (4.30, 6.88) | 1.85 (1.57, 2.13) |
1 y | 5.54 (4.19, 7.35) | 6.90 (5.10, 9.35) | 1.76 (1.45, 2.08) |
3 y | 3.97 (2.79, 5.63) | 6.25 (4.16, 9.39) | 2.38 (1.80, 2.96) |
1-y change | 1.55 (1.28, 1.87) | 2.99 (2.38, 3.76) | 1.29 (1.02, 1.56) |
Values are ORs (95% CIs). Model 1: sex, age, BMI, waist circumference, physical activity, and smoking, with further inclusion of baseline HOMA-IR for 1 and 3 y and for 1-y changes. Model 2: model 1 + multimetabolite score. Model 3: multimetabolite score.
To explore the predictive ability of the aforementioned models, AUC analyses were performed (Table 5). The AUC was 0.69 (95% CI: 0.66, 0.73) in the classical model and significantly improved to 0.81 (95% CI: 0.79, 0.84) in the full model for high HOMA-IR at baseline (P = 0.001). Significant improvements in AUCs were also found in the full model as compared with the classical model for 1-y changes in HOMA-IR [0.69 (95% CI: 0.66, 0.72) compared with 0.57 (95% CI: 0.53, 0.62)], but not for 1- and 3-y values. When the baseline model was tested for its performance at 1 and 3 y, no significant improvements in AUCs were observed in the full model compared with the classical model (Supplemental Table 2).
TABLE 5.
HOMA-IR | Model 1 | Model 2 | Model 3 |
---|---|---|---|
Baseline | 0.69 (0.66, 0.73) | 0.81 (0.79, 0.84)* | 0.81 (0.78, 0.84)* |
1 y | 0.84 (0.80, 0.88) | 0.83 (0.82, 0.85) | 0.75 (0.71, 0.78)*,# |
3 y | 0.81 (0.75, 0.86) | 0.81 (0.76, 0.85) | 0.72 (0.70, 0.74)*,# |
1-y change | 0.57 (0.53, 0.62) | 0.69 (0.66, 0.72)* | 0.66 (0.62, 0.69)*,# |
Model 1: sex, age, BMI, waist circumference, physical activity, and smoking, with further inclusion of baseline HOMA-IR for 1 and 3 y and for 1-y changes. Model 2: model 1 + multimetabolite score. Model 3: multimetabolite score. *,#P < 0.05: *model 2 compared with model 1 or model 3 compared with model 1; #model 3 compared with model 2.
Sensitivity analysis: associations of baseline metabolites and 1-y changes with HOMA-IR (continuous)
When we treated HOMA-IR as a continuous variable, several metabolites were selected from the elastic net regression with positive and negative regression coefficients (Supplemental Tables 3 and 4). These results of sensitivity analysis on HOMA-IR were consistent with those of the primary analysis in relation to selected metabolites. In addition, the model (model 2) comprising metabolites in combination with the classical risk factors explained 52%, 62%, 50%, and 25% of variance in HOMA-IR for baseline and 1 and 3 y and for 1-y changes, respectively, and added an additional 31%, 6%, 7%, and 23%, respectively, to the variance explained by the model consisting of classical risk factors alone (model 1) (Table 6).
TABLE 6.
HOMA-IR | Model 1 | Model 2 |
---|---|---|
Baseline | 0.206 | 0.516 |
1 y | 0.559 | 0.622 |
3 y | 0.427 | 0.504 |
1-y change | 0.025 | 0.246 |
Continuous log-transformed values are shown. Model 1: sex, age, BMI, waist circumference, physical activity (log), and smoking, with further inclusion of baseline HOMA-IR (log) for 1 and 3 y and for 1-y changes. Model 2: model 1 + multimetabolite score.
Associations of baseline multimetabolite score with risk of T2D
A total of 152 incident cases and 548 control participants were included in this analysis (median follow-up: 3.8 y). For every 1-SD increase in the identified baseline multimetabolite score (model 3) predicting high HOMA-IR (median value or higher) or HOMA-IR treated as continuous, the risk of T2D significantly increased, with HRs of 2.00 (95% CI: 1.58, 2.55) and 2.24 (95% CI: 1.72, 2.90), respectively, independently of classical T2D risk factors.
Discussion
Novel markers may help to elucidate aspects of metabolic dysfunction contributing to diabetes risk and improve the early identification of this condition (10). Because T2D has a progressive onset, metabolite profiles at multiple time points before T2D diagnosis would be useful to identify new, early diagnostic biomarkers of T2D (21). In the present study, we identified specific metabolites associated with HOMA-IR that improved insulin resistance prediction beyond classical risk factors at baseline but not at 1 and 3 y. One reason could be due to suboptimal statistical power in the follow-up time points, which included a smaller number of subjects. The larger sample at baseline may have enabled the elastic model to select more metabolites than at the other time points, and thus the prediction probably was improved at baseline. With regard to 1-y changes, the set of metabolites also improved the prediction of changes in HOMA-IR. Interestingly, when HOMA-IR was treated as a continuous variable, the variance explained by the full model significantly improved at all time points. One explanation for the observed fluctuations in the selected metabolites when HOMA-IR was modeled at different time points could be differences in sample size and composition at each time point. On the other hand, these metabolites may be influenced by physiologic or body adaptation changes unrelated to insulin resistance. Furthermore, a set of metabolites found to be associated with HOMA-IR at baseline further predicted T2D incidence independently of classical risk factors.
The metabolites positively selected when we treated HOMA-IR as a continuous variable included many amino acids and derivatives, such as proline, alanine, isoleucine, valine, acetylglycine, and dimethylglycine; tryptophan derivatives such as N-acetyl-tryptophan, 3-hydroxyanthranilic acid, N-acetyl-leucine, tyrosine, etc.; some lipids such as carnitines (C3-DC-CH3, C4, C4-OH, C5, C14, C18, C26), SMs (C18:0, C14:0), etc.; and purine and pyrimidine derivatives such as methyladenosine, methylguanosine, methylcytosine, cytosine, uric acid, pseudouridine, etc. These metabolites were identical or similar to those reported as associated with or predicting HOMA-IR in other different populations (10).
Using a dichotomous (binary) approach, a set of only 18 metabolites was positively associated with baseline HOMA-IR, 12 metabolites with 1-y HOMA-IR, and 14 metabolites with 3-y HOMA-IR. Most of these metabolites were identified through the use of the continuous approach and have been previously reported to be associated with insulin resistance. These, more specifically, included the amino acids and derivatives isoleucine, N-acetyl-tryptophan, alanine, dimethylglycine, valine, proline, N-carbamoyl-β-alanine, glutamate, tyrosine, 5-hydroxyindole-acetic acid, hydroxyproline, N-acetyl-leucine, dimethylglycine, N-acetyl-leucine, 3-hydroxyanthranilic acid, N-acetyl-arginine, and methylhistidine; lipids, carnitines (C2, C3-DC-CH3, C4, C5, C7, C26), 1 SM (C14:0), 1 LPC (C14:0), and sphinganine; purine and pyrimidine derivatives methyladenosine, methylguanosine, hypoxanthine, pseudouridine, hypoxanthine, uric acid, adenosine, and xanthine; vitamin B-1 (thiamin); the neurotransmitter serotonin; the hormone cortisol; and cyclohexylamine.
The highest positive regression coefficient for baseline HOMA-IR was obtained for methyladenosine, whereas for 1 and 3 y, the highest coefficient was seen for isoleucine. Methyladenosine has been associated with urinary system cancers and uremia (22) and was reported to decrease (along with alanine, proline, trans-cinnamic acid, tyrosine, and BCAAs) with ≥10% weight loss in obese individuals after a 1-y weight-loss program (23). Previous reports in the literature regarding purine catabolism, uric acid, and insulin resistance, especially in obese patients (24), support the role of purine metabolites in the pathophysiology of insulin resistance and T2D. Elevation in the BCAAs valine, leucine, and isoleucine, and in aromatic amino acids, has been previously observed, especially in obesity-related insulin resistance. Insulin is a regulator of branched-chain α-keto-acid dehydrogenase complex, a rate-limiting enzyme of BCAA catabolism; a reduced enzyme activity of the branched-chain α-keto-acid dehydrogenase complex in obesity and diabetes has been reported (25). With regard to isoleucine, its concentration was reported, along with glutamic acid, leucine, and tyrosine, to have the strongest correlation with insulin resistance, independent of sex or obesity status, and similar to traditional markers of insulin resistance (e.g., glucose, TGs) (26). N-Acetyl-tryptophan had the highest regression coefficient found between 1-y changes in the concentrations of metabolites and 1-y changes in HOMA-IR. This metabolite is produced by the liver and was also shown to increase in the serum of patients with chronic renal failure (27).
Short-chain acylcarnitines (isobutyrylcarnitine or C4, isovalerylcarnitine, and propionylcarnitine or C2) are derived from the metabolism of BCAAs and other amino acids (28); their positive correlation with HOMA-IR is not unexpected because this most likely also reflects the increased BCAA catabolism. In contrast, our finding concerning SM C14:0 is not in accordance with the results of Hanamatsu et al. (29), who reported that the concentrations of SM C14:0 and C16:0 did not correlate with obesity, insulin resistance, liver function, and lipid metabolism. Concerning the role of LPC C14:0, Rauschert et al. (30) indicated a significant association between LPC C14:0 concentration and HOMA-IR in normal-weight but not in obese or overweight young women. Finally, concerning sphinganine, which is produced through ceramide metabolism (31), plasma concentrations remain unchanged in prediabetic animals, but significantly increased when insulin sensitivity decreases and diabetes is established. In humans, a derivative, 1-deoxysphinganine, has been reported to predict T2D incidence (32).
Moderate thiamin deficiency is reported in 17–79% of diabetic patients (33). In our study, thiamin was positively associated with HOMA-IR at 1- and 3-y follow-ups; differences in the diet, as well as in the genetic background (33), of the populations involved may explain the discrepancy. Serotonin (5-hydroxytryptamine) is synthesized from tryptophan and is contained mainly in circulating platelets (34); serotonin- or tryptophan-rich foods (avocados, bananas, plums, walnuts, etc.) do not contribute significantly to serum concentrations (35); the reason for its selection for HOMA-IR association in our model remains unclear. Cortisol is also known to be significantly positively correlated with glucose HOMA-IR (36), whereas cyclohexylamine possibly indicates the consumption of cyclamates (artificial sweeteners) (37).
With regard to metabolites found to be negatively associated with HOMA-IR, the higher consistency in relation to these associations across different time points was shown for glycine, β-amino-isobutyric acid, betaine, and LPC C18:1, followed by symmetric dimethylarginine (SDMA) and arginine. Epidemiologic data have shown that low plasma concentrations of glycine are associated with increased HOMA-IR (38). A previous experimental study in mouse models found that BAIBA (β-aminoisobutyric acid) attenuates insulin resistance, suppresses inflammation, and induces fatty acid oxidation (39). Plasma betaine concentrations were also reported to be lower in insulin-resistant humans and were positively associated with insulin sensitivity (40). Lower concentrations of plasma LPC C18:1 were also associated with increased HOMA-IR scores (41). Reduced SDMA concentrations have been reported in insulin-resistant subjects (42), and a strong negative relation between SDMA and insulin resistance has been described in Caucasians (43).
The recent approaches to precision nutrition from different perspectives highlight the need for an integrative framework that takes into account innovative tools and methods in this field (44). In due course, integrated approaches, including monitoring specific metabolites and groups of metabolites, have the potential to guide necessary dietary behaviors in an individualized or a group-based manner.
Our study has several limitations. Beyond the influence of the genetic background, HOMA-IR can be considered an “imprecise” method for assessing insulin resistance. However, individuals from opposing HOMA-IR index categories had clear differences in the concentrations of TGs and HDL cholesterol, supporting the idea that these individuals had true differences in insulin sensitivity. Furthermore, the observation that metabolite scores produced by our analysis and predicting high HOMA-IR (median value or higher) or HOMA-IR treated as continuous at baseline also predict the risk of T2D indicates a consistency in our approach. Another limitation is that participants were elderly individuals at high cardiometabolic risk and from the Mediterranean region, which may limit the generalizability of our findings to other age groups or populations.
In conclusion, the present study showed that specific metabolites significantly improved the predictive ability of an insulin resistance index beyond classical risk factors at baseline and during a 1-y period in an elderly population at high cardiovascular risk. Although most of the selected metabolites at baseline remained predictors during the follow-up, others failed to act as predictors, implying that further external validation of these potential biomarkers is required. Selected metabolites for insulin resistance at baseline additionally predicted the risk of T2D, indicating consistency in our approach. Future studies may extend this approach by establishing a small set of metabolites that could be clinically assessed and lead to a practical metabolomics-based HOMA-IR and T2D test. The identification of causal relations (45) among the metabolites selected in this study due to their association with HOMA-IR and T2D is a direction for future research.
Supplementary Material
ACKNOWLEDGEMENTS
We thank George A Fragkiadakis, Department of Nutrition and Dietetics, Technological Education Institute of Crete, Greece, for his intellectual contributions to this manuscript.
The authors’ responsibilities were as follows—FBH, JS-S, and MM-G: designed the research; CP, MB, YZ, MR-C, EY, MG-F, CR, DC, DW RE, ER, M Fitó, M Fiol, LL, MAM-G, and JS-S: conducted the research; DC, RE, M Fitó, FA, M Fiol, JL, LS-M, MAM-G, and JS-S: were the coordinators of subject recruitment at the outpatient clinics; CP and PH-A: analyzed the data; CP, MB, FBH, and JS-S: interpreted statistical analysis and data; CD, AD, and CBC: acquired and processed metabolomics data; CP: drafted the manuscript; FBH and JS-S: supervised the study; CP, MB, and JS-S: had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis; and all authors: revised the manuscript for important intellectual content and read and approved the final manuscript. None of the authors declared a conflict of interest.
Notes
Supported by grants R01-DK-102896 and F31-DK114938 from the NIH. The Prevención con Dieta Mediterránea (PREDIMED) trial was supported by the official funding agency for biomedical research of the Spanish government, the Instituto de Salud Carlos III, through grants provided to research networks specifically developed for the trial [grant RTIC G03/140 (to RE); grant RTIC RD 06/0045 (to MAM-G)], and through the Centro de Investigación Biomédica en Red de Fisiopatología de la Obesidad y Nutrición and by grants from Centro Nacional de Investigaciones Cardiovasculares (grant CNIC 06/2007), the Fondo de Investigación Sanitaria Fondo Europeo de Desarrollo Regional (grants PI04-2239, PI 05/2584, CP06/00100, PI07/0240, PI07/1138, PI07/0954, PI07/0473, PI10/01407, PI10/02658, PI11/01647, P11/02505, and PI13/00462), the Ministerio de Ciencia e Innovación (grants AGL-2009-13906-C02, SAF2016-80532-R, and AGL2010-22319-C03), the Fundación Mapfre 2010, Consejería de Salud de la Junta de Andalucía (grant PI0105/2007), the Public Health Division of the Department of Health of the Autonomous Government of Catalonia, Generalitat Valenciana (grants ACOMP06109, GVA-COMP2010-181, GVACOMP2011-151, CS2010-AP-111, PROMETEO17/2017, and CS2011-AP-042), and the Regional Government of Navarra (grant P27/2011). CP was supported by a postdoctoral fellowship granted by the Autonomous Government of Catalonia (PERIS 2016-2020 Incorporació de Científics I Tecnòlegs, SLT002/0016/00428). DW was supported by a postdoctoral fellowship granted by the American Heart Association (16POST31100031). MG-F was supported by the European Foundation for the Study of Diabetes/Lilly through the Institut d'Investigacions Sanitàries Pere i Virgili (IISPV).
Supplemental Figure 1, Supplemental Methods, and Supplemental Tables 1–4 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/ajcn/.
CP and MB contributed equally to this work.
Data availability: the data sets generated and analyzed during the current study are not publicly available due to national data regulations and for ethical reasons, including the possibility that some information might compromise research participants’ consent because our participants only gave their consent for the use of their data by the original team of investigators. However, these data can be requested by signing a data-sharing agreement as approved by the relevant research ethics committees and the steering committee of the PREDIMED study.
Abbreviations used: BCAA, branched-chain amino acid; LPC, lysophosphatidylcholine; minMSE, minimum mean-squared error; PREDIMED, Prevención con Dieta Mediterránea; SDMA, symmetric dimethylarginine; SM, sphingomyelin; TG, triacylglycerol; T2D, type 2 diabetes; WC, waist circumference.
References
- 1. Wilcox G. Insulin and insulin resistance. Clin Biochem Rev. 2005;26(2):19–39. [PMC free article] [PubMed] [Google Scholar]
- 2. Kahn SE, Hull RL, Utzschneider KM. Mechanisms linking obesity to insulin resistance and type 2 diabetes. Nature. 2006;444(7121):840–6. [DOI] [PubMed] [Google Scholar]
- 3. Würtz P, Mäkinen VP, Soininen P, Kangas AJ, Tukiainen T, Kettunen J, Savolainen MJ, Tammelin T, Viikari JS, Rönnemaa T et al.. Metabolic signatures of insulin resistance in 7,098 young adults. Diabetes. 2012;61(6):1372–80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Cheng S, Rhee EP, Larson MG, Lewis GD, McCabe EL, Shen D, Palma MJ, Roberts LD, Dejam A, Souza AL et al.. Metabolite profiling identifies pathways associated with metabolic risk in humans. Circulation. 2012;125(18):2222–31. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Tai ES, Tan ML, Stevens RD, Low YL, Muehlbauer MJ, Goh DL, Ilkayeva OR, Wenner BR, Bain JR, Lee JJ et al.. Insulin resistance is associated with a metabolic profile of altered protein metabolism in Chinese and Asian-Indian men. Diabetologia. 2010;53(4):757–67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Würtz P, Soininen P, Kangas AJ, Rönnemaa T, Lehtimäki T, Kähönen M, Viikari JS, Raitakari OT, Ala-Korpela M. Branched-chain and aromatic amino acids are predictors of insulin resistance in young adults. Diabetes Care. 2013;36(3):648–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Rhee EP, Cheng S, Larson MG, Walford GA, Lewis GD, McCabe E, Yang E, Farrell L, Fox CS, O'Donnell CJ et al.. Lipid profiling identifies a triacylglycerol signature of insulin resistance and improves diabetes prediction in humans. J Clin Invest. 2011;121(4):1402–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Kopprasch S, Dheban S, Schuhmann K, Xu A, Schulte KM, Simeonovic CJ, Schwarz PE, Bornstein SR, Shevchenko A, Graessler J. Detection of independent associations of plasma lipidomic parameters with insulin sensitivity indices using data mining methodology. PLoS One. 2016;11(10):e0164173. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Mai M, Tönjes A, Kovacs P, Stumvoll M, Fiedler GM, Leichtle AB. Serum levels of acylcarnitines are altered in prediabetic conditions. PLoS One. 2013;8(12):e82459. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Zhao X, Han Q, Liu Y, Sun C, Gang X, Wang G. The relationship between branched-chain amino acid related metabolomic signature and insulin resistance: a systematic review. J Diabetes Res. 2016;2016:2794591. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Newgard CB. Metabolomics and metabolic diseases: where do we stand?. Cell Metab. 2017;25(1):43–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Estruch R, Ros E, Salas-Salvadó J, Covas MI, Corella D, Arós F, Gómez-Gracia E, Ruiz-Gutiérrez V, Fiol M, Lapetra J et al.. Primary prevention of cardiovascular disease with a Mediterranean diet supplemented with extra-virgin olive oil or nuts. N Engl J Med. 2018;378(25):e34. [DOI] [PubMed] [Google Scholar]
- 13. Martínez-González MÁ, Corella D, Salas-Salvadó J, Ros E, Covas MI, Fiol M, Wärnberg J, Arós F, Ruíz-Gutiérrez V, Lamuela-Raventós RM et al.. Cohort profile: design and methods of the PREDIMED study. Int J Epidemiol. 2012;41(2):377–85. [DOI] [PubMed] [Google Scholar]
- 14. American Diabetes Association. Diagnosis and classification of diabetes mellitus. Diabetes Care. 2008;31:S55–S6. [DOI] [PubMed] [Google Scholar]
- 15. Mascanfroni ID, Takenaka MC, Yeste A, Patel B, Wu Y, Kenison JE, Siddiqui S, Basso AS, Otterbein LE, Pardoll DM et al.. Metabolic control of type 1 regulatory T cell differentiation by AHR and HIF1-alpha. Nat Med. 2015;21(6):638–46. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia. 1985;28(7):412–9. [DOI] [PubMed] [Google Scholar]
- 17. Elosua R, Marrugat J, Molina L, Pons S, Pujol E. Validation of the Minnesota Leisure Time Physical Activity Questionnaire in Spanish men. The MARATHOM investigators. Am J Epidemiol. 1994;139(12):1197–209. [DOI] [PubMed] [Google Scholar]
- 18. Schröder H, Fitó M, Estruch R, Martínez-González MA, Corella D, Salas-Salvadó J, Lamuela-Raventós R, Ros E, Salaverría I, Fiol M et al.. A short screener is valid for assessing Mediterranean diet adherence among older Spanish men and women. J Nutr. 2011;141(6):1140–5. [DOI] [PubMed] [Google Scholar]
- 19. Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models via coordinate descent. J Stat Softw. 2010;33(1):1–22. [PMC free article] [PubMed] [Google Scholar]
- 20. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837–45. [PubMed] [Google Scholar]
- 21. Carter TC, Rein D, Padberg I, Peter E, Rennefahrt U, David DE, McManus V, Stefanski E, Martin S, Schatz P et al.. Validation of a metabolite panel for early diagnosis of type 2 diabetes. Metabolism. 2016;65(9):1399–408. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Ishiwata S, Itoh K, Yamaguchi T, Ishida N, Mizugaki M. Comparison of serum and urinary levels of modified nucleoside, 1-methyladenosine, in cancer patients using a monoclonal antibody-based inhibition ELISA. Tohoku J Exp Med. 1995;176(1):61–8. [DOI] [PubMed] [Google Scholar]
- 23. Geidenstam N, Al-Majdoub M, Ekman M, Spégel P, Ridderstråle M. Metabolite profiling of obese individuals before and after a one year weight loss program. Int J Obes (Lond). 2017;41(9):1369–78. [DOI] [PubMed] [Google Scholar]
- 24. Elizalde-Barrera CI, Estrada-García T, Lozano-Nuevo JJ, Garro-Almendaro AK, López-Saucedo C, Rubio-Guerra AF. Serum uric acid levels are associated with homeostasis model assessment in obese nondiabetic patients: HOMA and uric acid. Ther Adv Endocrinol Metab. 2017;8(10):141–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Yamada C, Kondo M, Kishimoto N, Shibata T, Nagai Y, Imanishi T, Oroguchi T, Ishii N, Nishizaki Y. Association between insulin resistance and plasma amino acid profile in non-diabetic Japanese subjects. J Diabetes Investig. 2015;6(4):408–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Seibert R, Abbasi F, Hantash FM, Caulfield MP, Reaven G, Kim SH. Relationship between insulin resistance and amino acids in women and men. Physiol Rep. 2015;3(5):e12392. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Saito A, Niwa T, Maeda K, Kobayashi K, Yamamoto Y, Ohta K. Tryptophan and indolic tryptophan metabolites in chronic renal failure. Am J Clin Nutr. 1980;33(7):1402–6. [DOI] [PubMed] [Google Scholar]
- 28. Roe DS, Roe CR, Brivet M, Sweetman L. Evidence for a short-chain carnitine-acylcarnitine translocase in mitochondria specifically related to the metabolism of branched-chain amino acids. Mol Genet Metab. 2000;69(1):69–75. [DOI] [PubMed] [Google Scholar]
- 29. Hanamatsu H, Ohnishi S, Sakai S, Yuyama K, Mitsutake S, Takeda H, Hashino S, Igarashi Y. Altered levels of serum sphingomyelin and ceramide containing distinct acyl chains in young obese adults. Nutr Diabetes. 2014;4:e141. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Rauschert S, Uhl O, Koletzko B, Kirchberg F, Mori TA, Huang RC, Beilin LJ, Hellmuth C, Oddy WH. Lipidomics reveals associations of phospholipids with obesity and insulin resistance in young adults. J Clin Endocrinol Metab. 2016;101(3):871–9. [DOI] [PubMed] [Google Scholar]
- 31. Brozinick JT, Hawkins E, Hoang Bui H, Kuo MS, Tan B, Kievit P, Grove K. Plasma sphingolipids are biomarkers of metabolic syndrome in non-human primates maintained on a Western-style diet. Int J Obes (Lond). 2013;37(8):1064–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Mwinyi J, Boström A, Fehrer I, Othman A, Waeber G, Marti-Soler H, Vollenweider P, Marques-Vidal P, Schiöth HB, von Eckardstein A et al.. Plasma 1-deoxysphingolipids are early predictors of incident type 2 diabetes mellitus. PLoS One. 2017;12(5):e0175776. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Francis NK, Pawar HS, Mitra A, Mitra A. Assessment of insulin sensitivity and its convalescence with dietary rehabilitation in undernourished rural West Bengal population. J Clin Diagn Res. 2017;11(5):LC29–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Kema IP, de Vries EG, Muskiet FA. Clinical chemistry of serotonin and metabolites. J Chromatogr B Biomed Sci Appl. 2000;747(1-2):33–48. [DOI] [PubMed] [Google Scholar]
- 35. Kema IP, Schellings AM, Meiborg G, Hoppenbrouwers CJ, Muskiet FA. Influence of a serotonin- and dopamine-rich diet on platelet serotonin content and urinary excretion of biogenic amines and their metabolites. Clin Chem. 1992;38(9):1730–6. [PubMed] [Google Scholar]
- 36. Yan YX, Xiao HB, Wang SS, Zhao J, He Y, Wang W, Dong J. Investigation of the relationship between chronic stress and insulin resistance in a Chinese population. J Epidemiol. 2016;26(7):355–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Renwick AG, Thompson JP, O'Shaughnessy M, Walter EJ. The metabolism of cyclamate to cyclohexylamine in humans during long-term administration. Toxicol Appl Pharmacol. 2004;196(3):367–80. [DOI] [PubMed] [Google Scholar]
- 38. Magnusson M, Wang TJ, Clish C, Engström G, Nilsson P, Gerszten RE, Melander O. Dimethylglycine deficiency and the development of diabetes. Diabetes. 2015;64(8):3010–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Jung TW, Hwang HJ, Hong HC, Yoo HJ, Baik SH, Choi KM. BAIBA attenuates insulin resistance and inflammation induced by palmitate or a high fat diet via an AMPK-PPARδ-dependent pathway in mice. Diabetologia. 2015;58(9):2096–105. [DOI] [PubMed] [Google Scholar]
- 40. Ejaz A, Martinez-Guino L, Goldfine AB, Ribas-Aulinas F, De Nigris V, Ribó S, Gonzalez-Franquesa A, Garcia-Roves PM, Li E, Dreyfuss JM et al.. Dietary betaine supplementation increases Fgf21 levels to improve glucose homeostasis and reduce hepatic lipid accumulation in mice. Diabetes. 2016;65(4):902–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Wallace M, Morris C, O'Grada CM, Ryan M, Dillon ET, Coleman E, Gibney ER, Gibney MJ, Roche HM, Brennan L. Relationship between the lipidome, inflammatory markers and insulin resistance. Mol Biosyst. 2014;10(6):1586–95. [DOI] [PubMed] [Google Scholar]
- 42. Zsuga J, Török J, Magyar MT, Valikovics A, Gesztelyi R, Lenkei A, Csiba L, Kéki S, Zsuga M, Bereczki D. Dimethylarginines at the crossroad of insulin resistance and atherosclerosis. Metabolism. 2007;56(3):394–9. [DOI] [PubMed] [Google Scholar]
- 43. Schutte AE, Schutte R, Huisman HW, van Rooyen JM, Fourie CM, Malan L, Malan NT, Schwedhelm E, Strimbeanu S, Anderssohn M et al.. Dimethylarginines: their vascular and metabolic roles in Africans and Caucasians. Eur J Endocrinol. 2010;162(3):525–33. [DOI] [PubMed] [Google Scholar]
- 44. Wang DD, Hu FB. Precision nutrition for prevention and management of type 2 diabetes. Lancet Diabetes Endocrinol. 2018;6(5):416–26. [DOI] [PubMed] [Google Scholar]
- 45. Yazdani A, Yazdani A, Samiei A, Boerwinkle E. Identification, analysis, and interpretation of a human serum metabolomics causal network in an observational study. J Biomed Inform. 2016;63:337–43. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.