Stratifying high-risk prediabetes clusters using blood-based epigenetic markers

BACKGROUND: Previously, we identified six prediabetes clusters, three at moderate and three at high-risk for type 2 diabetes and/or complications. While this novel classification could enable earlier and improved disease prevention, it relies on intensive clinical phenotyping. Here, we developed a machine learning workflow to identify blood-based epigenetic markers to distinguish between prediabetes clusters. METHODS: DNA methylation was profiled in blood cells of different cohorts including individuals that belong to clusters 2 (low-risk), 3, 5, and 6 (each high-risk) and data was subjected to a machine learning workflow. RESULTS: In a discovery cohort (n = 187), we identified 1,557 CpG sites as predictors for clusters 2, 3, 5, and 6. These CpGs were sufficient to distinguish between individuals belonging to the high-risk clusters 3, 5 and 6 in an independent replication cohort (n = 146) with an accuracy of 92%. Between 300 and 339 CpG sites were specific for each cluster and the corresponding genes linked to TGF-β receptor and calcium signaling (cluster 3), MAPK cascade and ECM organization (cluster 5), and Wnt/SMAD signaling (cluster 6), mirroring the metabolic deterioration observed in each cluster. CONCLUSIONS: Without the need for complex clinical measurements, the identified blood-based epigenetic signatures may improve the detection of individuals at high-risk of developing diabetes and complications and point to the potential molecular mechanism responsible for the heterogeneity in prediabetes. These markers highlight the potential of the blood epigenome as an effective proxy for predicting future complications and make extensive clinical assessments obsolete, enabling the identification of clusters in larger populations.