Whole-genome analysis identifies novel molecular subtypes and driver structural variants in Japanese ovarian cancer

Ovarian cancer is a heterogeneous disease comprising a number of histotypes that are associated with distinct clinical characteristics such as therapeutic efficacy and likelihood of relapse. Since ovarian cancer exhibits heterogeneity even within a same histotype, it is critical to develop a complementary classification framework. Whereas for endometrial cancer, a molecular subtyping scheme based on genomic features was developed by The Cancer Genome Atlas (TCGA) and the scheme has been currently integrated into the system of 2020 WHO classification of Female Genital Tract tumour, large-scale genomic studies of ovarian cancer have predominantly focused on high-grade serous (HGS), a major histotype in Caucasian populations. The remaining histotypes exemplified by clear cell carcinoma, which is uniquely prevalent in the East Asian population, have not been covered by these studies, and therefore genomic features that can be utilized for molecular subtyping have not been well characterized for non-HGS ovarian cancer. To address this gap, we performed whole-genome sequencing of 1,412 Japanese ovarian cancers across six major histotypes: clear cell, high-grade serous, endometrioid, mucinous, carcinosarcoma, and sex cord-stromal tumors. Integrative whole-genome analysis identified five molecular subtypes-POLE, MSI, CNH-A, CNH-B, and CNL-with distinct genomic and clinicopathological features. CNH-A represented a conventional copy-number-high subtype characterized by frequent BRCA1 deletions and genomic features of homologous recombination deficiency, whereas CNH-B tumors showed frequent ATM alterations and copy-number variations (CNVs) with retained heterozygosity. Pathogenic germline variants in hereditary cancer genes exhibited histotype-specific patterns, and tumors with biallelic alterations in homologous recombination genes showed high HRD scores, accompanied by distinct somatic CNVs and structural variations (SVs). Furthermore, across 1,349 evaluable cases, we identified 101 genomic regions harboring candidate driver SVs, including 11 known driver regions and 20 fragile sites, as well as 70 previously unreported candidate driver SV regions. Together, this whole-genome analysis defines molecular subtypes characterized by distinct copy-number architectures and identifies recurrent driver SVs, providing a basis for refined molecular classification and future precision oncology.