MO71-1 Systematic quantification of ovarian tumour heterogeneity using an in silico tissue microarray approach

High-grade serous ovarian carcinoma (HGSOC) has a markedly high mortality-to-incidence ratio, however clinically useful predictive biomarkers are limited to BRCA1/2, HRD and prior platinum sensitivity. HGSOC is a complex ecosystem characterized by spatiotemporal clonal evolution and multiscale tumor heterogeneity (TH). Integrative biomarkers hold promise, but robust characterization of relations between single cell genomic assays, cellular phenotypes and imaging features will require data-driven tumor sampling. Here we quantify histological TH of HGSOC within and across lesions, and propose a computationally derived sampling strategy that optimally captures TH metrics. We have developed a virtual TMA (vTMA) sampling method using digital H& E whole-slide images (WSI) from HGSOC FFPE blocks from debulking surgery. After cell segmentation and random forest tumor/stroma/immune cell classification, our R pipeline uses single cell-level outputs to perform vTMA core sampling and pathomic calculations at WSI and vTMA-levels. We characterised pathomic features of WSIs and vTMAs from 236 unique primary and metastatic blocks from ten treatment-naïve HGSOC patients. Using spatial cell coordinates (median 1.6×106cell detections/WSI, range 1.1×105-4.2×106), we quantified TH in terms of cell type abundance and densities, Shannon and Batty entropies and geospatial metrics (nearest-neighbor/Clark-Evans/Morisita-Horn/Moran indices, Getis-Ord hotspots) and characterised how these vary across different configurations of vTMAs, WSIs, lesions and patients. Our reproducible computational pipeline allows systematic characterisation of intra- and interTH based on standard WSIs of any cancer type, and can guide targeted TMA core sampling. Our approach enables evidence-based selection of tissue regions for emerging pathomic and spatial biology analyses, and represents an important step toward bridging the gap between radiomic and genomic data.