Biological age as estimated by baseline circulating metabolites is associated with incident diabetes and mortality.

Abstract

It is unclear how metabolomic assessment of biological aging performs in non-White populations and whether such an approach can predict future mortality. We aimed to evaluate the application of serum metabolomics combined with machine learning methodologies to predict incident diabetes and mortality in a Thai population. We analyzed serum samples and mortality data over 11 years from among 454 participants with no previous history of diabetes and with a fasting plasma glucose ≥85th percentile (5.4 mmol/L) but <7 mmol/L. Untargeted serum metabolomics were assessed using liquid chromatography/mass spectrometry. A deep artificial neural network was used to predict biological age based on serum metabolite profiles and chronological age. The mean age of participants was 40.5 ± 6.4 years, and 70.8% were men. We found a significant positive correlation between metabolomic age and chronological age (r = 0.71, P < 0.001). After 5 years, 61 of 404 participants with available glycated hemoglobin status (15.1%) progressed to diabetes. Chronological age was associated with incident diabetes but was not significant (P = 0.08), after adjusting for BMI and sex. Metabolomic age was significantly related to incident diabetes after controlling for BMI and sex (P < 0.05). Over the 11-year follow-up, 10 participants died owing to non-accidental causes. When metabolomic age and chronological age were included together in the model, metabolomic age (but not chronological age) was associated with mortality, independent of age, sex, and BMI. Among all identifiable metabolites, beta-D-mannosylphosphodecaprenyl and phosphatidylserines were the five leading metabolites associated with mortality. We concluded that serum metabolomic profile was associated with incident diabetes as well as mortality over our 11-year study period, which may render it potentially useful in assessing biological aging in humans.