Single-cell RNA sequencing reveals specific cell subtypes that influence survival and determine molecular subtype classification of ovarian cancers

High-grade serous tubo-ovarian cancer (HGSTOC) is characterised by extensive inter- and intratumour heterogeneity, resulting in persistent therapeutic resistance and poor disease outcome. Molecular subtype classification based on bulk RNA sequencing facilitates a more accurate characterisation of this heterogeneity, but the lack of strong prognostic or predictive correlations with these subtypes currently hinders their clinical implementation. Stromal admixture profoundly affects the prognostic impact of the molecular subtypes, but the contribution of stromal cells to each subtype has poorly been characterised. Increasing the transcriptomic resolution of the molecular subtypes based on single-cell RNA sequencing (scRNA-seq) may provide insights in the prognostic and predictive relevance of these subtypes.

Researchers at KU Leuven performed scRNA-seq of 18,403 cells unbiasedly collected from 7 treatment-naive HGSTOC tumours. For each phenotypic cluster of tumour or stromal cells, they identified specific transcriptomic markers. The researchers explored which phenotypic clusters correlated with overall survival based on expression of these transcriptomic markers in microarray data of 1467 tumours. By evaluating molecular subtype signatures in single cells, they assessed to what extent a phenotypic cluster of tumour or stromal cells contributes to each molecular subtype.

The researchers identified 11 cancer and 32 stromal cell phenotypes in HGSTOC tumours. Of these, the relative frequency of myofibroblasts, TGF-β-driven cancer-associated fibroblasts, mesothelial cells and lymphatic endothelial cells predicted poor outcome, while plasma cells correlated with more favourable outcome. Moreover, they identified a clear cell-like transcriptomic signature in cancer cells, which correlated with worse overall survival in HGSTOC patients. Stromal cell phenotypes differed substantially between molecular subtypes. For instance, the mesenchymal, immunoreactive and differentiated signatures were characterised by specific fibroblast, immune cell and myofibroblast/mesothelial cell phenotypes, respectively. Cell phenotypes correlating with poor outcome were enriched in molecular subtypes associated with poor outcome.

scRNA-seq-based tumour microenvironment analysis of
18,403 single cells from 7 treatment-naive HGSTOC patients

A Schematic overview of the sampling site (ovary, omentum or peritoneum) and tissue type (normal or tumour tissue) of the 12 biopsies from seven treatment-naive patients as well as the analysis workflow. B t-SNE representation of all single cells colour-coded for their assigned major cell type (left) and for the expression of three marker genes used for this annotation as indicated on the top row. Marker genes: B cell (CD79A, IGHG3, IGKC), dendritic cells (CD1C, CD1A, CLEC9A), endothelial cells (CLDN5, PECAM1, VWF), fibroblasts (COL1A1, COL1A2, BGN), myeloid cells (CD68, LYZ, AIF1), ovarian stroma cells (STAR, FOXL2, DLK1), T cells (CD3D, CD3E, TRAG), epithelial cancer cells (EPCAM, PAX8, CD24). Dendritic cells remained co-clustered with myeloid cells in the first clustering step, but separated from myeloid cells before further subclustering based on established marker genes (CLEC9A, CD1C, CD1A). C Barplot showing for each of the 32 stromal and 11 cancer subclusters (from left to right) the number of cells, tissue type (normal or tumour), their distribution across the 7 patients, their distribution across sampling sites (ovary, omentum and peritoneum) and their correlation to the copy number alteration (CNA) profile of patient 1 using low-coverage whole-genome sequencing. D t-SNE visualisation of dendritic cell (up) and myeloid cells (down) subclusters containing tissue-specific cells enriched in the omentum, defined as Langerhans-like dendritic cells (DC_CD207) and lipid-associated macrophages (M_MMP9). E CNA profiles of fibroblasts compared to those from the tumour subclusters and monocyte subclusters using inferCNV confirming the CN stable profile of all fibroblast subclusters

These researchers used scRNA-seq to identify stromal cell phenotypes predicting overall survival in HGSTOC patients. These stromal features explain the association of the molecular subtypes with outcome but also the latter’s weakness of clinical implementation. Stratifying patients based on marker genes specific for these phenotypes represents a promising approach to predict prognosis or response to therapy.