Abstract 5442: Identifying cell-cell interactions across spatial scales in spatial transcriptomics.

Abstract Introduction: Single-cell spatial transcriptomics provides rich data for the gene expression profiles at cellular resolutions. This level of data offers the ability to estimate interactions between cells and associate those interactions with clinical information such as survival or response to immunotherapy. Previously, we had shown that COL4A1 and ITGAV are significantly more spatially enriched in primary clear cell renal cell carcinoma (ccRCC) tumors after exposure to immunotherapy (IO), specifically showing high expression in malignant cells and fibroblasts/myofibroblasts. We assessed this interaction by using K=3, but spatial proximity is also an important consideration to explore. Methods: To address true spatial context, we developed a functional data analysis approach to profile cell-cell interactions at varying spatial scales. We used 14 stromal compartment FOVs (8x IO naïve, 6x IO exposed) from Soupir et. al. (2024). Bivariate Moran’s I was calculated for COL4A1 and ITGAV using row standardized weights from Gaussian transformed distances. The bandwidth of the kernel was varied from 0 to 250 to calculate Moran’s I as a function of bandwidth, I(h), using 100 permutations to determine complete spatial randomness (CSR). CSR was subtracted from I(h) (Degree of I(h)) to make values comparable across samples. We used functional principal component (FPC) analysis to analyze the full Degree of I(h) curves. FPC scores were used to compare the IO naïve and IO exposed tumors. Results from our approach were compared to SpatialDM, another approach based on Moran’s I, with h=75. Results: FOVs from tumors exposed to IO showed a distinct, positive Degree of I(h) curve with a peak at a bandwidth between 25-50 while FOVs from IO naïve tumors were each unique in either shape or sign of Degree of I(h). Calculating (FPCs) from all Degree of I(h) curves showed that FPC1 describes the overall strength of the spatial relationship (positive scores indicate overall elevated and negative scores indicate overall decreased interaction) while FPC2 describes whether the interaction occurs at a near/far scale. Plotting FPC2 vs FPC1, FPC1 perfectly separates IO naïve from IO exposed FOVs (Wilcox Test p=0.00067, where IO exposed FOVs have a positive FPC1 score and FPC2 scores around 0 (no change in near/far). SpatialDM’s global I didn’t show significant differences between IO exposures (Wilcox Test p=0.1079) Conclusion: The application of our approach to ccRCC indicates that COL4A1 and ITGAV in these stroma FOVs from IO exposed primary ccRCC tumors are more strongly spatially related across spatial scales than stroma FOVs from IO naïve primary ccRCC tumors. Our approach also showed a significantly stronger spatial association than SpatialDM. Further research is needed to better understand the underlying cause of this shift, which may lead to new drug targets. Citation Format: Alex C. Soupir, Mitchell T. Hayes, Brandon J. Manley, Lauren Cole Peres, Julia Wrobel, BROOKE FRIDLEY. Identifying cell-cell interactions across spatial scales in spatial transcriptomics abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 5442.