Abstract 3562: Comparison of spatial co-localization measures for studying the tumor immune microenvironment with application to ovarian cancer

Abstract

Abstract Introduction: The spatial contexture of the tumor immune microenvironment (TIME) has been associated with survival among cancer patients, including women with high-grade serous ovarian cancer (HGSOC). However, an assessment of statistical approaches to quantify the spatial characteristics of the TIME has not been conducted. Moreover, it is unknown which statistical approach is most sensitive in identifying spatial co-localization of two cell types in the TIME. Methods: Using the R package scSpatialSIM that simulates spatial single-cell protein data (i.e., multiplex immunofluorescence data), we completed a simulation study to compare four methods for assessing co-localization of cell types in the TIME (Ripley’s K, Nearest Neighbor G, Dixon’s segregation statistic, and a spatial interaction variable). The simulation study varied the cell abundance (e.g., low vs high), the co-localization (e.g., co-localization, random, segregation), and the variation in the marked point process. In addition to the simulated data, we assessed the four methods in large epidemiological studies (Nurses’ Health Study [NHSI/NHSII], New England Case Control Study [NECC], African American Cancer Epidemiology Study [AACES], and North Carolina Ovarian Cancer Study [NCOCS]) including 756 women with HGSOC. Multiplex immunofluorescence was conducted using AKOYA Biosciences OPALTM platform to measure cytotoxic T cells (CTLS) and regulatory T cells (Tregs). Cox proportional hazards regression models were fit to estimate the association of spatial co-localization of CTLS and Tregs with overall survival. Results: In the simulation study, we found that Ripley’s K was the most powerful approach for assessing co-localization for a variety of simulation scenarios. Surprisingly, the other approaches (Nearest Neighbor G, Dixon’s Segregation statistic, and spatial interaction statistic) were unable to detect the simulated co-localization of two cell types in most simulation scenarios. In the analysis of women with HGSOC in the epidemiological cohorts, we were able to detect significant co-localization of CTLS and Tregs in 40.2%, 12.2%, 21.2%, and 17.0% of samples using Ripley’s K, Nearest Neighbor G, Dixon’s statistic, and the interaction variable methods, respectively. However, the level of co-localization was not associated with survival. Conclusions: We found that Ripley’s K was the most powerful at detecting simulated co-localization of two cell types in the TIME. Similarly, Ripley’s K identified the highest level of significant spatial co-localization of CTLS and Tregs in the application to HGSOC. However, we did not observe an association of co-localization of CTLS and Tregs with overall survival. Future work is needed to develop powerful approaches for quantifying the spatial contexture of the TIME, along with assessment of the statistical properties of these methods. Citation Format: Brooke L. Fridley, Alex Soupir, Julia Wrobel, Christelle Colin-Leitzinger, Mary K. Townsend, Andrew B. Lawson, Kathryn L. Terry, Joellen M. Schildkraut, Shelley S. Tworoger, Lauren C. Peres. Comparison of spatial co-localization measures for studying the tumor immune microenvironment with application to ovarian cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 3562.