Dissecting Genetic and Environmental Determinants of Plasma Molecular Signatures and Their Link to Type 2 Diabetes Risk

Abstract

Background: Type 2 diabetes (T2D) is a heterogeneous disease shaped by both genetic, environmental, cultural, and socioeconomic factors, with well-documented disparities in incidence across populations. The molecular pathways underlying these disparities, however, remain poorly understood. Plasma metabolites and proteins integrate both genetic and environmental influences on type 2 diabetes (T2D) risk, providing insight into disease mechanisms. We aimed to quantify the variance in these molecular profiles explained by environmental and genetic ancestry domains and to apply causal inference approaches to identify environmentally and genetic ancestry influenced pathways contributing to T2D risk.

Methods: We analyzed plasma proteomic and metabolomic profiles from 3,360 MESA participants (51.6% female), and in 1,333 participants from the Women's Health Initiative. To characterize the sources of variance in plasma proteomic and metabolomic profiles, we performed variance decomposition partitioning into four domains: biological (age, sex, BMI), genetic ancestry (principal components), lifestyle (smoking, alcohol intake, diet), and social determinants (self-reported race and ethnicity, income, education). To assess causal pathways towards T2D risk, we applied two-sample Mendelian Randomization to disentangle environmental and genetic contributors to T2D risk.

Results: The largest share of variance in proteomic and metabolomic profiles was explained by biological and lifestyle factors, while race and ethnicity and genetic ancestry accounted for smaller but non-redundant contributions. Genetic ancestry was primarily associated with lipid and apolipoprotein variation, whereas race and ethnicity and socioeconomic factors were associated with immune and inflammatory signatures. Environmentally influenced metabolites (e.g., diacylglycerols, phosphatidylethanolamines, lysophosphatidylcholines) and vascular-inflammatory proteins were consistently linked to higher T2D risk, while genetic ancestry influenced triglycerides and IGFBP3 reflected inherited risk pathways. Mediation analyses showed that selected lipids and proteins (e.g., IGFBP2, HGF, SSC4D) explained 10-25% of racial/ethnic disparities in T2D. Mendelian randomization identified causal roles for seven lipid species and IGFBP3 in T2D risk.

Conclusions: Our results reveal both genetic and non-genetic sources of variation in proteomic and metabolomic profiles, uncovering environmental and genetic pathways contributing to T2D risk. These findings advance precision medicine by identifying modifiable molecular mediators of disparities and potential causal targets for prevention.

Keywords: Biomarkers; Diabetes Mellitus; Metabolomics; Population Groups; Precision Medicine; Proteomics; Type 2.

Figure 1.

Study design. T2D Type 2 diabetes, MESA Multi-ethnic study of atherosclerosis, WHI Women’s Health Initiative. LC/MS Liquid Chromatography Mass Spectrometry. Figure created with BioRender

Figure 2.

(A) Total variance explained (R²) by the full linear regression model—covariate domains include biological (age, sex, BMI), lifestyle (smoking, alcohol, physical activity, diet), genetic ancestry, self-reported race/ethnicity, and socioeconomic status (SES—across metabolites and proteins in MESA. Normalized variance explained per covariate domain across individual metabolites (B) and proteins (C). Normalized variance explained per covariate domain across individual metabolites (B) and proteins (C). For each molecule, R² values were normalized to sum to 1 (100%), representing the relative contribution of each covariate domain. Bars are ordered by decreasing total variance explained.

Figure 3

(A) Mean variance explained (R²) in metabolites and proteins by covariate domains in MESA, partitioned into unique and shared contributions. Bars represent the average proportion of variance in metabolite and protein levels explained by each domain (B). Relationship between Statistical Significance and Unique Variance Explained. The plot visualizes the contribution of various covariate domains to the variance in metabolites and proteins. The x-axis represents the mean unique variance explained (Unique R², %) by each covariate domain. The y-axis shows the statistical significance of each covariate domain, quantified ACAT method.

Figure 4.

Top ten metabolites (A) and proteins (B) with the largest unique variance explained by each covariate domain, estimated using linear regression models (selected at P < 1×10⁻⁴) in MESA.

Figure 5 |. Independent and shared contributions of race/ethnicity and genetic ancestry to molecular variance in MESA

A, Variance explained (R²) for metabolites partitioned into shared variance, unique variance explained by genetic principal components (gPCs), and unique variance explained by self-reported race/ethnicity. B, Equivalent decomposition for proteins. Each bar represents one molecular feature. C, Mean variance explained (ΔR²) across metabolites and proteins under alternative covariate models. Error bars denote standard errors. D, Unique variance explained by self-reported race/ethnicity versus genetic principal components (gPCs) for metabolites. Each point represents one metabolite, with colour indicating the difference in variance explained (ΔR² = Race – gPCs). Labeled metabolites highlight those with the largest discrepancies. E, Unique variance explained by race/ethnicity versus gPCs for proteins, with discordant molecules highlighted. Variance components were estimated using linear regression models, with unique and shared variance derived from block-level decomposition.

Figure 6.. Environmentally and genetically shaped molecular signatures and causal pathways to type 2 diabetes.

(A) Prospective associations of metabolites (left) and proteins (right) with incident type 2 diabetes (T2D) in meta-analyses across MESA and WHI. Molecules were classified as predominantly environmentally influenced (pink) or genetically influenced (blue) based on variance decomposition. Points represent hazard ratios (HRs), and horizontal lines indicate 95% confidence intervals. (B) Causal mediation analysis of environmentally influenced molecules in the relationship between race/ethnicity and T2D risk in MESA. Significant mediators (P < 10⁻⁸) are shown, with circle size denoting the proportion of T2D risk mediated. (C) Two-sample Mendelian randomization of genetically influenced metabolites and proteins, testing their causal effects on T2D. Points represent causal effect estimates (HRs), and horizontal lines denote 95% confidence intervals across three methods: inverse variance weighted (IVW), weighted median, and MR-Egger.