Metabolic variation reflects dietary intake in a multi-ethnic Asian population

Dietary biomarkers reflecting habitual diet are explored largely in European and American populations. However, the "food metabolome" is highly complex, with its composition varying to region and culture. Here, by assessing 1,055 plasma metabolites and 169 foods/beverages in 8,391 comprehensively phenotyped individuals from the multi-ethnic Asian HELIOS cohort (69% Chinese, 12% Malay, 19% South Asian), we report novel observations for ethnic-relevant and common foods. Using machine-learning feature selection approach, we developed dietary multi-biomarker panels (3-39 metabolites each) for key foods and beverages in respective training sets. These panels comprised distinct and shared metabolite networks, and captured variances in intake prediction models in test sets better than single biomarkers. Composite metabolite scores, derived from the biomarker panels, associated significantly and more strongly with clinical phenotypes (HOMA-IR, type 2 diabetes, BMI, fat mass index, carotid intima-media thickness and hypertension), compared to self-reported intakes. Lastly, in 235 individuals that returned for a repeat visit (averaged 322 days apart), diet-metabolite relationships were robust over time, with predicted intakes, derived from biomarker panels and metabolite scores, showing better reproducibility than self-reported intakes. Altogether, our findings show new insights into multi-ethnic diet-related metabolic variations and new opportunity to link exposure to health outcomes in Asian populations.