Abstract Background & objective: High grade serous ovarian cancer (HGSC) is the most lethal type of ovarian cancer malignancies. Around 50% of HGSCs are DNA-repair homologous recombination deficient (HRD) and 15% have CCNE1 gene amplification. Previous results have shown that tumor microenvironment (TME) can impact clinical prognosis in HGSC. In this study our objective was to characterize the interplay of the TME with HRD and CCNE1 amplification in HGSC and its association with clinical outcomes. Materials and methods: We collected TME spatial protein expression, genomics, and clinical data from 250 HGSC patients. We determined BRCA1/2 germline and somatic mutation status using targeted sequencing and promoter hyper-methylation. For the tumors without BRCA1/2 mutations, we performed shallow whole genome sequencing, and estimated the CCNE1 amplification and HRD status using shallowHRD and BRCAness copy number profiles. Immune cell infiltration was assessed also using IHC for CD8, CD20, CD68, and CD103. The single-cell spatial landscapes were analyzed for 1000 tumor cores from the tumor center and tumor periphery using cyclic immunofluorescence (tCycIF) imaging for 34 different protein markers. We categorized and spatially analyzed the single cells from the TME to distinct cellular subpopulations using the TRIBUS and scimap software. Results: In our study of the TME using highly multiplexed tissue imaging we identified a total of 4.8 million single cells of them 1.6 million were immune cells belonging to eight biological subpopulations. The immune cell proportion detected in each sample showed high concordance with the IHC estimation by a pathologist. Interestingly, the CD8+, CD20+, CD11c+, and CD15+ high immune cell infiltrations were higher in BRCA1/2 mutated as compared to the CCNE1 amplified tumors. High immune infiltration was associated with prolonged overall survival only in samples debulked before chemotherapy but not in samples debulked after chemotherapy. Using consensus clustering for cell protein expression and cell morphology, we identified five distinct tumoral and stromal cell clusters which showed significant heterogeneity among the tumor genotypes and association with clinical outcomes. Then, after clustering of the spatial single-cell proximity data, we identified 13 distinct recurrent cellular neighborhoods (RCNs). The two RCNs related to the tumor-stroma and tumor-immune interfaces exhibited the highest differences between HRD and CCNE1 samples. Finally, by using random forest machine learning classifiers of the cells and its neighborhoods, and using an approach to minimize overfitting, we were able to detect the features more representative in HRD samples, for example, MHC-I expression in cancer cells and in stromal cells. Conclusion: The integration of multi-omics data revealed distinct TMEs associated with HRD and clinical outcomes; this has the potential to be used to discover biomarkers for precision oncology in HGSCs. Citation Format: Fernando Perez-Villatoro, Lilian van Wagensveld, Aleksandra Shabanova, Siamak Hajizadeh, Julia Casado, Ella Anttila, Maaike A. van der Aa, Roy F.P.M. Kruitwagen, Gabe S. Sonke, Koen K. Van de Vijver, Peter K. Sorger, Hugo M. Horlings, Anniina Färkkilä. A spatially resolved single-cell tumor microenvironment of clinicomolecular subtypes of high-grade serous ovarian cancer [abstract]. In: Proceedings of the AACR Special Conference on Ovarian Cancer; 2023 Oct 5-7; Boston, Massachusetts. Philadelphia (PA): AACR; Cancer Res 2024;84(5 Suppl_2):Abstract nr B073.
Read full abstract