Hierarchical Clustering Identifies Reproducible Polycystic Ovary Syndrome Subtypes with Ancestry-Specific Genetic Associations
Abstract Background Complex diseases often represent the phenotypic convergence of multiple genetic etiologies under one diagnosis. Resolving this heterogeneity can increase power for genetic discovery and delineate causal pathways. PCOS diagnostic criteria are based on expert opinion, and the resulting phenotypes appeared to share similar genetic architecture. However, hierarchical clustering of quantitative traits has resolved discrete reproductive and metabolic subtypes with distinct genetic architectures; cases without stable cluster membership form a background subtype. Because the subtypes are defined from phenotypic traits alone, finding that they carry distinct genetic signals provides an orthogonal confirmation of clustering validity that internal metrics and trait-correlated data cannot. Here we test whether these subtypes are reproducible across ancestries, identify susceptibility loci, and validate the clustering method. Methods We performed hierarchical clustering of eight reproductive and metabolic traits in 1,449 European-ancestry and 418 Korean PCOS cases. We conducted overall and subtype-specific genome-wide association studies in the European cohort (991 cases, 3,877 controls) and the Korean cohort (417 cases, 926 controls), followed by trans-ancestry meta-analysis. We compared hierarchical and k-means clustering in 874 European cases using internal stability and separation metrics, with subtype-specific association as orthogonal validation. Results Hierarchical clustering reproduced the reproductive and metabolic subtypes in both cohorts. Before subtyping, the European cohort confirmed two established loci and identified one novel signal, whereas the Korean cohort yielded no genome-wide significant associations. Subtype-specific analysis identified additional loci that were only nominally significant before subtyping. Trans-ancestry meta-analysis identified two robust, subtype-restricted signals with concordant cross-ancestry effects. Remaining subtype-specific loci had minor allele counts below the pre-specified threshold and are reported as provisional signals requiring replication. Hierarchical clustering with eight traits produced more stable subtyping with subtype-specific genetic associations; k-means yielded slightly greater geometric separation but no associated loci. Conclusions Phenotypically defined PCOS subtypes carry distinct genetic architectures that replicate across ancestries. Trans-ancestry meta-analysis identifies shared, subtype-restricted risk with concordant effects. Because the subtypes are defined solely on phenotypic traits, their distinct genetic associations provide orthogonal validation, supporting hierarchical clustering as a generalizable strategy for resolving complex trait heterogeneity.
Authors
- Andrea Elizabeth Dunaif (ORCID: https://orcid.org/0000-0002-8175-4677)
- Hyejin Lee
- Chun Li (ORCID: https://orcid.org/0000-0002-8819-2443)
- Kelly Brewer (ORCID: https://orcid.org/0000-0002-7086-4571)
- Ryan Sisk
Institutions
- University of Southern California (US)
- Ewha Womans University (KR)
- Icahn School of Medicine at Mount Sinai (US)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-16
- DOI
- https://doi.org/10.21203/rs.3.rs-11676895/v1
- Primary Topic
- Ovarian function and disorders
- Type
- preprint