Subtyping of obesity and type 2 diabetes using genetic discordance: a phenome-wide comparative analysis

Abstract Obesity and type 2 diabetes are causally related, yet the underlying mechanisms are poorly defined. We aimed to subclassify obesity into etiological subtypes using a genetic-driven agnostic approach. Genetic instruments for body mass index and type 2 diabetes were classified into profiles that convey highly concordant (48 SNPs) and discordant (19 SNPs) diabetogenic effects. We annotated association signals for these instruments across clinical and molecular phenotypic layers, spanning > 5,000 traits. Using a combination of machine-learning techniques, key differences were identified in fat distribution, liver metabolism, blood pressure, lipids and branched-chain amino-acids, as well as the sulfotransferase HS6ST2 and several bacterial taxa belonging to the Bacteroidetes and Firmicutes phyla. We reproduced the clinical differences with polygenic scores specific to each profile in an independent cohort, which revealed differential cardiovascular mortality by obesity subtype. Mendelian randomization analyses revealed prominent roles of waist-to-hip ratio, concentration and cholesterol content of HDL particles, and blood pressure in the causal processes leading to diabetes in obesity; through these analyses, we prioritized 17 genes from the discordant signature that convey protection against type 2 diabetes in obesity.