Abstract
BACKGROUND: The growing field of metabolomics has opened new venues for identifying biomarkers of type 2 diabetes (T2D) and predicting its consequences, such as coronary heart disease (CHD). Despite their large size, Middle Eastern populations are underrepresented in omics research. In this study, we aim at investigating metabolomics profiles of T2D stratified by the CHD comorbidity for Middle Eastern population, such as Qatari population. METHODS: In this cross-sectional study, we used a total of 641 metabolites from a large cohort of 3,679 Qatari adults from the Qatar BioBank (QBB; 272 T2D and 2,438 non-T2D individuals) and Qatar Cardiovascular Biorepository (QCBio; all CHD patients; 488 T2D and 481 non-T2D individuals). Univariate and pathway enrichment analyses were performed to identify metabolites associated with T2D in the absence or presence of CHD. Machine learning (ML) models, and metabolite risk scores were developed to assess the predictive power of the different combinations of T2D and CHD. RESULTS: Many metabolites were significantly associated with T2D in both the QBB and QCBio cohorts. Among these, we observed 1,5-anhydroglucitol (1,5-AG) (P = 1.33 × 10(-68) [-5.20, -4.16] in QBB vs 9.82 × 10(-33) [-2.51, -1.80] in QCBio), glucose (P = 7.14 ×10(-57) [4.09, 5.23] in QBB vs. 3.26 × 10(-29) [1.41, 2.00] in QCBio), and mannose (P = 2.61 × 10(-54) [2.68, 3.45] in QBB vs. 1.01 × 10(-27) [1.45, 2.09] in QCBio). Other metabolites were significantly associated with T2D only in one cohort, e.g., gamma-glutamylglutamine (P = 1.79 × 10(-20) and β = -2.61 in QBB vs. P = 5.12 × 10(-1) and β = 0.10 in QCBio). The enriched pathways (FDR P< 0.05), common to both cohorts, included galactose metabolism and valine leucine, and isoleucine biosynthesis and degradation. Few pathways were significantly associated with T2D in only one cohort: fructose and mannose, and Pantothenate and CoA biosynthesis metabolisms were significant in the QCBio cohort, whereas Arginine biosynthesis, and Alanine, aspartate and glutamate metabolisms were significant in the QBB cohort. ML models performed well in predicting T2D with high accuracy (>80% in both QBB and QCBio). The metabolite risk score (MRS) developed in the QCBio and tested in the QBB while adjusting for hemoglobin A1C yielded an odds ratio (OR) of 21.18 for the top quintile vs. the remaining quintiles. CONCLUSIONS: Metabolomic profiling has the potential for the early detection of metabolic alterations that precede clinical symptoms of T2D and CHD in the presence of T2D. Risk scores showed great performance in predicting T2D and CHD, but longitudinal data are required to provide evidence for disease risk. Early detection allows timely interventions and improved management strategies for both T2D and CHD patients.