OR02-04 Resolving PCOS Heterogeneity: Subtype-Specific Genetic Risk Scores Reveal Evidence for Two Distinct Disorders

Abstract Disclosure: K. Brewer: None. K. van der Ham: None. L.M. Moolhuijsen: None. R.K. Sisk: None. Y. Louwers: None. J.A. Visser: None. J.S. Laven: None. C. Li: None. A.E. Dunaif: None. Polycystic ovary syndrome (PCOS) is a complex trait reflecting the interaction of susceptibility genes and environmental factors. A variable constellation of hormonal and metabolic features characterizes PCOS. Increasing evidence suggests PCOS heterogeneity reflects distinct etiological mechanisms. However, the current diagnostic criteria, based on expert opinion, do not capture this heterogeneity because the genetic architecture of resulting phenotypes is similar. In contrast, using machine learning approaches, we have identified subtypes of PCOS with unique genetic architectures rooted in distinct reproductive and metabolic causal pathways. PCOS subtypes are associated with additional traits that align with these pathways and are present in regionally and ethnically diverse PCOS cohorts. The existence of these PCOS subtypes has been confirmed using other machine-learning approaches. The generalizability of our clustering algorithm is limited by the availability of precisely measured phenotypic traits that are not consistently monitored in clinical practice. Accordingly, we tested the hypothesis that we could identify reproductive or metabolic PCOS cases using subtype-specific genetic risk scores (GRS). Two European ancestry PCOS cohorts diagnosed by NIH criteria were analyzed: Rotterdam, Netherlands (615 cases, 4431 controls) and a US cohort (915 cases, 3065 controls). Hierarchical clustering in the Netherlands cohort identified reproductive (n=198) and metabolic (n=417) subtypes (van der Ham et al., JCEM, 2024). Subtype-specific genome-wide association study (GWAS) summary statistics were used to calculate GRS in the US cohort using PRS-CS. Logistic regression assessed PCOS risk within GRS (GRS-R for reproductive and GRS-M for metabolic), adjusting for age, BMI, and genetic principal components. Phenotypic traits within top GRS deciles were compared using ANCOVA. Both GRS-R and GRS-M accurately identified PCOS cases in the US cohort (GRS-R OR=3.35 [2.98–3.80], P=1.1E-04; GRS-M OR=5.85 [4.94–7.00], P=<2E-16). PCOS cases with high GRS-M had significantly higher fasting insulin, BMI, fasting glucose, and triglycerides, while those with high GRS-R had significantly higher SHBG, LH, FSH, and HDL (P<0.001). PCOS cases with high GRS in both subtypes (n=37) consistently displayed trait similarity to GRS-M (P>0.05) and only resembled GRS-R for FSH (P>0.05). Testosterone, DHEAS, AMH, LDL, and cholesterol levels did not differ. These findings demonstrate that subtype-specific GRS generated in one PCOS cohort can accurately identify PCOS reproductive and metabolic subtype cases in an independent cohort. This approach provides a method for investigating clinically relevant parameters and additional phenotypes associated with PCOS subtypes in electronic health record-linked biobanks, advancing the translation of PCOS subtyping into clinical practice. Presentation: Saturday, July 12, 2025