Polycystic ovary syndrome (PCOS) is a complex disorder with heterogenous phenotypes and unclear etiology. A recent phenotypic clustering study identified metabolic and reproductive subtypes of PCOS. We hypothesize that the heterogeneity of PCOS manifestations reflects different mechanistic pathways and can be identified using a genetic approach. We applied k-means clustering to categorize the genome-wide significant PCOS variants into clusters based on their associations with selected quantitative traits that likely reflect PCOS etiological pathways. We evaluated the association of each cluster with PCOS-related traits and disease outcomes. We then applied Mendelian randomization to estimate the causal effects between the traits and PCOS. Three categories of variants were identified: adiposity, insulin resistant, and reproductive. Significant associations were observed for variants in the adiposity cluster with body mass index (BMI), waist circumference and breast cancer, and variants in the insulin-resistant cluster with fasting insulin, glucose values, and homeostatic model assessment of insulin resistance (HOMA-IR). Sex hormone binding globulin (SHBG) has strong association with all three clusters. Mendelian randomization suggested a causal role of BMI and SHBG on PCOS. No causal associations were observed for PCOS on disease outcomes.
Keywords: Mendelian randomization; adiposity; causality; clustering; genetic heterogeneity; insulin resistance; polycystic ovary syndrome; sex hormone binding globulin.