Abstract A103: Improved cell phenotyping of ovarian cancer tumor microenvironment

Abstract

Abstract Introduction: Multiplexed imaging at single-cell resolution is becoming widely used to decipher the role of the cellular microenvironment in cancer and other complex diseases. To identify spatial patterns of single cells on a tissue, accurate cell-type phenotyping is a crucial step. We present Tribus, an interactive, semi-supervised classifier that avoids thresholds and manual labeling, with user-friendly visualization of the results and automated quality control reports. Materials and Methods: We developed Tribus, an open-source Python 3 software utilizing self-organizing maps for data clustering with an integrated Napari plug-in for fast visualization. It assigns the cell phenotypes based on user-defined label description tables and median marker expressions, following a hierarchical structure to minimize overspill bias. It has a modular architecture and can be run via command line, Jupyter Notebook, or the Napari interface. For exploring the Tribus performance, we used 1) manually expert annotated datasets with different tissue origins 2) dataset with cluster-based annotation 3) previously unpublished dataset. To ensure usability, we utilized datasets generated by using different multiplexed imaging technologies, and of variable size including large tissue microarrays and whole-slide images. Results: Using Tribus we accurately annotated the cell phenotypes in an expert-labeled Co-detection by indexing (CODEX) dataset, consisting of 8 regions of tissue samples from a single donor, including both healthy colon and small bowel sections. Tribus analysis of a high-grade ovarian cancer (HGSC) tumor microarray dataset generated by tCyCIF revealed a subset of lymphoid cells previously mislabelled using a cluster annotation approach. Finally, we utilized Tribus for deep cell phenotyping on a large whole-slide image tCyCIF dataset consisting of over a million cells from five matched HGSC samples collected before and after neoadjuvant chemotherapy (NACT). In the paired samples we successfully identified cell subtypes crucial for immune surveillance. Our analysis accurately differentiated M1 and M2 macrophage subpopulations. In addition, the analyzed post-NACT samples showed a higher proportion of tumor-infiltrating CD8+ T-cells, interestingly displaying occasional spatial clusters by the tumor borders. Therefore, Tribus has shown a great potential to study the spatial patterns and cell-cell interactions of relevant immune cells in response to neoadjuvant chemotherapy treatment. Conclusion: Overall, Tribus is an accurate, fast and efficient way to analyze different sizes of datasets across different technologies, resulting in accelerated analysis of the spatial tumor microenvironment (TME). The accurate cell-type phenotyping with Tribus uncovered high-resolution cellular phenotypes and their spatial patterns in the TME in ovarian cancer after neoadjuvant chemotherapy. Citation Format: Ziqi Kang, Teodora Farago, Angela Szabo, Inga-Maria Launonen, Fernando Perez, Ella Anttila, Kevin Elias, Julia Casado, Peter Sorger, Anniina Färkkilä. Improved cell phenotyping of ovarian cancer tumor microenvironment [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 A103.