Characterisation of the genetic discordance between body mass index and type 2 diabetes: a phenome-wide analysis: No 111

Background and aims: Obesity is on the rise globally, and is a leading risk factor for T2D. However, it is very heterogeneous, with varying degrees of T2D risk within the same levels of BMI. Better classification may lead to improve outcomes of current preventive and therapeutic strategies. Moreover, by elucidating the mechanisms uncoupling obesity from T2D risk, new possible therapeutic targets may emerge. Leveraging the vast amount of genetic data produced to date may contribute to reach these goals while overcoming the obstacles imposed by common assumptions, biases and confounders present in observational studies. Our aim is to compare the phenome-wide association patterns of BMI-increasing genetic profiles that either concordantly increase or discordantly decrease T2D risk. Materials and methods: Highly concordant and highly discordant SNPs between BMI and T2D were obtained from the latest GWAS for both conditions. Their standardized effect sizes (SES) across multiple traits in the phenome, metabolome, proteinome and transcriptome were retrieved from the online genomic repositories. After alignment to the BMI-increasing allele, these effects were organized into a SNP x Trait matrix. A hierarchical clustering technique, combining PCA and Random Forest algorithms was applied, retrieving the optimal number of clusters of traits, organized in order of importance, useful to distinguish a discordant from a concordant SNP. Posterior probabilities of colocalization with T2D were calculated for each gene using transcriptome results. Tissue, biological process, molecular mechanism and cellular component enrichments were evaluated. The predictive potential of GRSs informed by these findings were assessed in the UK Biobank dataset. Results: 121 SNPs were found to be significantly associated with BMI and T2D. 18 were discordant and 104 concordant. A total of 1372 variables were included in the analyses (Phenome = 546, Metabolome = 233, Proteinome = 593). The most important difference between discordant and concordant SNPs in the phenome matrix was found in a cluster of traits led by hypertension (Mean discordant SES = -1.59, Mean concordant SES = 2.56), highly correlated with two clusters led by coronary heart disease and overall health status, respectively. The second most important cluster was led by physical activity-adjusted WHR (Mean discordant SES = -2.69, Mean concordant SES = 0.24). The model obtained from the phenome matrix had the highest classification performance (Matthews Correlation Coefficient, MCC = 0.79). Metabolome results showed differences in polyunsaturated fatty acids and lipid contents in VLDL, but with lower performance (MCC = 0.67). The model from the proteinome matrix was unable to correctly classify SNPs (MCC = -0.03). Two genes (CCDC92 and DNAH10) showed the strongest association within the discordant set in adipose tissue, both involved in cilia formation. A GRS of these 121 SNPs with weights derived from the clusters with high classification performance was highly associated with T2D in both the general and obese populations in UK Biobank (p < 1x1016). Conclusion: The main difference between BMI-increasing genetic profiles that either discordantly decrease or concordantly increase T2D risk is found in hypertension risk and physical activity-adjusted WHR. These traits can be used to inform GRSs to better classify T2D risk in obesity. Molecular mechanisms behind the discordant profile appear to involve cilia formation in the adipose tissue.