Transcriptional Circuitry in HGSOC: A Dynamic Three-State Model Informed by a Living Biobank of Purified Tumour Fractions
High-grade serous ovarian cancer (HGSOC) is a heterogeneous disease, but efforts to define transcriptional subtypes using bulk RNA sequencing have been confounded by the presence of non-malignant cells. As a result, it remains unclear whether tumour-cell-intrinsic states exist, and whether these represent stable disease subtypes or are dynamically remodelled during disease progression and treatment. Here, we address this question using a living biobank of patient-derived ovarian cancer models (OCMs) cultured as purified tumour-cell populations under uniform conditions. RNA sequencing followed by unsupervised non-negative matrix factorisation (NMF) revealed a robust, hierarchical architecture comprising three core tumour-cell-intrinsic subtypes: the Alpha cluster, marked by cell-cycle deregulation and E2F-driven replication stress; the Beta cluster, defined by tumour-cell-intrinsic immune mimicry and inflammatory signalling; and the Gamma cluster, characterised by epithelial identity, extracellular matrix engagement, and metabolic adaptation. At higher clustering resolution, a fourth cluster, Delta, emerged as a Gamma sub-lineage distinguished by a vesicle-oriented, neuronal-like secretory programme. By projecting cluster labels onto a subset of matched longitudinal OCMs using non-negative least squares, we show that while some tumours retain stable subtype identities, others display transcriptional plasticity, including transitions from epithelial-like Gamma states to more proliferative or secretory phenotypes. Together, these findings define the core architecture and dynamic potential of tumour-cell-intrinsic transcriptional states within HGSOC, thereby bridging legacy bulk classifications with emerging single-cell insights, establishing a framework for more precise patient stratification.