Longitudinal metabolomics integrated with machine learning identifies novel biomarkers of gestational diabetes mellitus.

BACKGROUND:Evidence from longitudinal studies is crucial to enhance our understanding of the role of metabolites in the progression of gestational diabetes mellitus (GDM). Herein, a longitudinal untargeted metabolomic study was conducted to reveal the metabolomic profiles and biomarkers associated with the progression of GDM, and characterize the changing patterns of metabolites. METHODS:We collected serum samples at three trimesters from 30 patients with GDM and 30 healthy Chinese pregnant women with pre-pregnancy BMI, age, and parity matched, and untargeted metabolomic analysis was performed, followed by machine learning approaches that integrated bootstrap and LASSO. Cluster analysis was conducted to elucidate the patterns of metabolite changes. Pathway analyses were conducted to gain insights into the underlying pathways involved. RESULTS:A total of 32 metabolites, mainly belonging to amino acid and its derivatives, were significantly associated with GDM across three trimesters, and were clustered into three distinct patterns. Metabolites belonging to phosphatidylcholines, lysophosphatidylcholines, lysophosphatidic acids, and lysophosphatidylethanolamines were consistently upregulated, and 2,3-Dihydroxypropyl dihydrogen phosphate was downregulated in GDM group. Amino acid-related, glycerophospholipid, and vitamin B6 metabolism were enriched in multiple trimesters. The levels of allantoic acid, which was positively correlated with blood glucose, was consistently higher in GDM patients and exhibited good discriminatory ability for GDM in the early and mid-pregnancy. CONCLUSION:We identified and characterized distinct patterns of metabolites associated with GDM throughout pregnancy, and found that allantoic acid was a potential biomarker for early diagnosis of GDM.