Abstract B097: Image analysis-guided spatial transcriptomics improves recognition of biological features important for outcome in high grade serous carcinoma
Anna R. Laury,Shuyu Zheng,Niina Aho,Robin Fallegger,Satu M. Hänninen,Julio Saez Rodriguez,Jovan Tanevski,Omar Youssef,Jing Tang,Olli Carpén
DOI: https://doi.org/10.1158/1538-7445.ovarian23-b097
IF: 11.2
2024-03-04
Cancer Research
Abstract:Abstract High-grade serous ovarian carcinoma (HGSC) is characterized by late presentation, frequent recurrence, and treatment resistance. Our understanding of this disease is limited, partly due to considerable tumor heterogeneity, which is reflected in the variable tumor morphology. Beyond the diagnosis itself, no particular histologic features have proven sensitive or specific enough for use in clinical practice. Image analysis based deep learning models are now able to extract new information from standard H&E pathology slides, though biological interpretation of the regions detected by artificial intelligence (AI) remains a challenge. In this work, we interrogate the molecular features of tissue regions identified as being highly associated with patient outcome by a deep-learning based AI classification model. We determined the spatial transcriptomic profiles of these regions in 16 patients and compared them to the profiles of background tumor regions not selected by the AI model. We used multiple methods to characterize the regions types, including: differential gene expression, UMAP distances, machine learning-based outcome prediction, pathway enrichment, and spatial organization analyses. We find that regions identified by the AI-based model are significantly different from background tumor regions, and that their transcriptomic profiles are more predictive of patient outcome. We determine the differential signaling pathways and validate a representative transcript using RNA in-situ hybridization. We show that AI-detected regions in poor prognosis tumors have upregulated JUN expression, and are characterized by tumor cell subpopulations with a stress-associated transcriptomic signature. In contrast, spatial pattern analyses show that AI-detected regions in tumors with better outcome are typified by macrophage and plasma cell depleted DNA repair-signature tumor cell subpopulations. The synergy of spatial transcriptomic and image analysis data in our study offers a promising strategy to enhance clinical translation by establishing a more interpretable AI model. Citation Format: Anna R. Laury, Shuyu Zheng, Niina Aho, Robin Fallegger, Satu M. Hänninen, Julio Saez Rodriguez, Jovan Tanevski, Omar Youssef, Jing Tang, Olli Carpén. Image analysis-guided spatial transcriptomics improves recognition of biological features important for outcome in high grade serous carcinoma [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 B097.
oncology