The discovery of novel predictive biomarkers and early-stage pathophysiology for the transition from gestational diabetes to type 2 diabetes

Diabetologia. 2019 Apr;62(4):687-703. doi: 10.1007/s00125-018-4800-2. Epub 2019 Jan 15.


Aims/hypothesis: Gestational diabetes mellitus (GDM) affects up to 20% of pregnancies, and almost half of the women affected progress to type 2 diabetes later in life, making GDM the most significant risk factor for the development of future type 2 diabetes. An accurate prediction of future type 2 diabetes risk in the early postpartum period after GDM would allow for timely interventions to prevent or delay type 2 diabetes. In addition, new targets for interventions may be revealed by understanding the underlying pathophysiology of the transition from GDM to type 2 diabetes. The aim of this study is to identify both a predictive signature and early-stage pathophysiology of the transition from GDM to type 2 diabetes.

Methods: We used a well-characterised prospective cohort of women with a history of GDM pregnancy, all of whom were enrolled at 6-9 weeks postpartum (baseline), were confirmed not to have diabetes via 2 h 75 g OGTT and tested anually for type 2 diabetes on an ongoing basis (2 years of follow-up). A large-scale targeted lipidomic study was implemented to analyse ~1100 lipid metabolites in baseline plasma samples using a nested pair-matched case-control design, with 55 incident cases matched to 85 non-case control participants. The relationships between the concentrations of baseline plasma lipids and respective follow-up status (either type 2 diabetes or no type 2 diabetes) were employed to discover both a predictive signature and the underlying pathophysiology of the transition from GDM to type 2 diabetes. In addition, the underlying pathophysiology was examined in vivo and in vitro.

Results: Machine learning optimisation in a decision tree format revealed a seven-lipid metabolite type 2 diabetes predictive signature with a discriminating power (AUC) of 0.92 (87% sensitivity, 93% specificity and 91% accuracy). The signature was highly robust as it includes 45-fold cross-validation under a high confidence threshold (1.0) and binary output, which together minimise the chance of data overfitting and bias selection. Concurrent analysis of differentially expressed lipid metabolite pathways uncovered the upregulation of α-linolenic/linoleic acid metabolism (false discovery rate [FDR] 0.002) and fatty acid biosynthesis (FDR 0.005) and the downregulation of sphingolipid metabolism (FDR 0.009) as being strongly associated with the risk of developing future type 2 diabetes. Focusing specifically on sphingolipids, the downregulation of sphingolipid metabolism using the pharmacological inhibitors fumonisin B1 (FB1) and myriocin in mouse islets and Min6 K8 cells (a pancreatic beta-cell like cell line) significantly impaired glucose-stimulated insulin secretion but had no significant impact on whole-body glucose homeostasis or insulin sensitivity.

Conclusions/interpretation: We reveal a novel predictive signature and associate reduced sphingolipids with the pathophysiology of transition from GDM to type 2 diabetes. Attenuating sphingolipid metabolism in islets impairs glucose-stimulated insulin secretion.

Keywords: Gestational diabetes mellitus; Glucose-stimulated insulin secretion; Lipidomic study; Machine learning; Multiple logistic regression; Pathophysiology; Predictive biomarker; Prospective cohort; Sphingolipid metabolism; Type 2 diabetes.

Publication types

  • Research Support, N.I.H., Extramural
  • Research Support, Non-U.S. Gov't

MeSH terms

  • Adult
  • Animals
  • Area Under Curve
  • Asian Americans
  • Biomarkers / blood*
  • Case-Control Studies
  • Decision Trees
  • Diabetes Mellitus, Type 2 / blood*
  • Diabetes Mellitus, Type 2 / diagnosis*
  • Diabetes Mellitus, Type 2 / ethnology
  • Diabetes, Gestational / blood*
  • Diabetes, Gestational / ethnology
  • Disease Progression
  • Female
  • Glucose Tolerance Test
  • Hispanic Americans
  • Humans
  • Islets of Langerhans / metabolism
  • Machine Learning
  • Male
  • Mice
  • Mice, Inbred C57BL
  • Postpartum Period
  • Pregnancy
  • Prospective Studies
  • Risk Factors
  • Sphingolipids / metabolism
  • United States


  • Biomarkers
  • Sphingolipids