Tumour architecture shapes polarized epithelial states that predict survival in high-grade serous ovarian cancer

High-grade serous ovarian carcinoma (HGSC) is among the most genomically unstable cancers, yet how this diversity relates to malignant-cell states and how they influence treatment response remain unclear. Integrating 1.98 million single-cell transcriptomes across 13 studies, we identify a dominant axis of polarization in malignant secretory cells, from proliferative, progenitor-like (SecA) to quiescent, injury-like (SecB) states that recur across patients and coexist within genetic clones. Spatial profiling of eight whole tumors and tissue microarrays from 378 patients links this polarization to tumor architecture: SecB cells occupy avascular, hypoxic niches, where they adopt HIF/NF-{kappa}B survival programs, exclude lymphocytes, and co-organize with macrophages exhibiting similar glycolytic states. Their localization and enrichment in ascites link SecB to metastasis. SecB predicts poor survival across independent cohorts, but preferential benefit from bevacizumab in the ICON7 trial. Together, these findings reveal shared malignant phenotypes across genomically diverse tumors and highlight their potential to guide treatment selection.