Genetic and phenotypic discordance in cardiometabolic diseases

Cardiometabolic conditions such as obesity and type 2 diabetes (T2D) are among the first causes of death globally, and the number of people affected is rapidly increasing. Both conditions are intricately connected and are associated with many life-threatening cardiovascular disease (CVD). However, both obesity and T2D are highly heterogeneous, such that not all individuals with these conditions develop complications, while others are at disproportionatel higher risk. The purpose of this thesis is to improve our understanding of the heterogeneity in cardiometabolic conditions through the application of genetic analyses and machine learning techniques to identify profiles at disproportionately higher or lower risk of complications, providing insights into the mechanisms that give rise to these profiles and their potential clinical implications. In Paper I, I used cross-trait genetic analysis to derive genetically determined obesity profiles that are associated with higher body mass index (BMI) but are either concordantly associated with higher risk of T2D or discordantly associated with T2D protection. Through a comprehensive phenome-wide comparative analysis of these profiles, we highlighted adipose distribution, vascular function and extracelullar matrix remodelling as key mechanisms uncoupling obesity from its diabetogenic risk, and prioritise 17 genes with potential discordant effects in obesity. In Paper II, I reformulated the cross-trait genetic approach from Paper I to construct genetically determined diabetic profiles that are either concordantly associated with higher risk of CVD or discordantly associated with CVD protection. The phenome-wide comparison yields VLDL metabolism as a key mechanism detaching T2D from CVD and 8 loci with cardioprotective effects in T2D that include known targets of current medications, such as statins and GLP-1 receptor agonists. Additionally, I showed through polygenic score analyses (PRS) that quantifying genetic discordance in the general population can aid in CVD risk prediction, improving predictions in 5% of cases and 3% of non-cases. In Paper III, I designed a probabilistic, graph-based clustering ensemble algorithm to identify subgroups of individuals whose levels of common biomarkers of CVD risk deviate from the expected given their body size. We deploy this approach on four large independent cohorts, finding that biomarker deviate substantially from the BMI expectation in ~20% of the general population, and tend to form 5 distinct profiles with specific BMI-biomarker discordance patterns. Considering these discordant profiles can improve CVD prediction with benefits comparable to lipid fraction quantification. In Paper IV, I contributed in an investigation of overall, sex-specific and nonlinear causal effects of BMI on multiple outcomes, including T2D and CVD. We found consistent positive effects of BMI on T2D across sexes, but sex-differential effects on CAD. Evidence of nonlinear effects of BMI were found for lipids and glycaemia. Understanding the mechanisms by which some individuals are at disproportionately higher or lower risk to develop the complications generally associated with obesity and type 2 diabetes can help design better and more precise interventions.