Predictive Primary Metabolomics Signatures in Early to Mid-pregnancy for Risk of Gestational Diabetes

Gestational diabetes (GDM) predisposes 7-10% of pregnant women in the U.S. to a myriad of adverse health sequelae. Early prediction is critical. By leveraging metabolomics and clinical data, we aimed to develop and validate predictive metabolic signatures in early to mid-pregnancy for GDM. In a nested matched case-controls study of 91 women with GDM and 180 non-GDM controls in the Pregnancy Environment and Lifestyle Study (PETALS) cohort (discovery set), 157 annotated primary metabolites were measured by gas chromatography/time-of-flight mass spectrometry using fasting serum at gestational weeks (GW) 10-13 and 16-19. Machine learning algorithm using 10-fold cross validated Lasso regression identified predictive metabolomics signatures for GDM risk in the discovery set. We validated the signature at GW 10-13 in a random PETALS subsample (42 GDM, 372 non-GDM; validation set 1) and a matched case-control study within the GLOW trial of overweight/obese women (35 GDM, 70 non-GDM; validation set 2) and the signature at GW 16-19 in validation set 1. Upregulated aromatic and acidic amino acids and purinones at both GW 10-13 and 16-19 and additionally hexoses at GW 16-19 were significantly associated with GDM risk (FDR-adjusted P <0.05). A 33-metabolite signature of organic acids, lipids, and organooxygen compounds at GW 10-13 revealed incremental predictive ability beyond conventional risk factors (CRF) including age, pre-pregnancy BMI, preexisting hypertension, previous GDM, diabetes family history, and fasting glucose (discovery set AUC: 0.871 vs. 0.742; validation set 1: 0.869 vs. 0.731; validation set 2: 0.972 vs. 0.742; all P <0.01). Similar predictive ability was found for 13 metabolites at GW 17-19 beyond CRF (discovery AUC: 0.838 vs. 0.732; validation 1: 0.830 vs. 0.774; both P <0.05). Our findings suggest potentially important roles of increased amino acids, purinones, and hexoses in the pathophysiology of GDM and a predictive ability of selected primary metabolites beyond CRF. Disclosure Y. Zhu: None. A. Ngo: None. S. Fan: None. D. Barupal: None. C. Quesenberry: None. O. Fiehn: None. A. Ferrara: None. Funding National Institutes of Health (K12HD052163); National Institute of Diabetes and Digestive and Kidney Diseases (K01DK120807)