Abstract 2331: Uncovering gene co-regulation networks in spatial transcriptomics data

Abstract Single-cell spatial transcriptomics measures thousands of genes in situ over potentially millions of cells. These rich datasets pose the best problem in genomics: how can an analyst possibly uncover all the biological insights contained within them? Careful analysis can uncover intricate biological relationships, but only if the analyst knows where to look. One useful set of “places to look” is genes sharing spatial correlations, i. e. that tend to be expressed in the same regions as each other. When genes are expressed together in space, it suggests causality, either by one gene acting on another, or by some latent variable impacting all the genes. These causal relationships have only become detectable en masse with the advent of spatial transcriptomics data, making them rich in potential for previously unknown findings. Unfortunately, spatial correlation has largely fallen short of its promise for one reason: genes expressed by the same cell type are usually highly spatially correlated. These trivial correlations overwhelm results driven by interesting biology. Here we present an algorithm designed to discover spatial correlations arising from causal relationships and not from the cell type landscape. We compute genes’ conditional correlations over space, looking at the residuals of spatial expression after accounting for the cell type landscape. In a 6, 000-plex Spatial Molecular Imaging (SMI) analysis of colon cancer, we find that approximately 80% of the strongest spatial correlations can be attributed to cell type; the remaining 20% suggest underlying causal relationships. 96 distinct modules comprising up to 46 spatially correlated genes were discovered. The expression within these modules is driven by between 1 and 19 different cell types. Notable modules include those related to IFNG signaling, antigen presentation, B-cell chemotaxis and activation, and microenvironmental remodeling. Further, our study delves into the spatial correlations among ligand-receptor pairs, revealing spatial co-regulation in a limited subset of the 555 ligand-receptor pairs examined in the 6, 000-plex panel. This approach enables us to discern spatial correlations that hint at biological causality, thereby beginning to delineate the "spatial interactome" of cancer. Leveraging a growing collection of 6, 000-plex CosMxTM SMI cancer datasets that span various cancer types including breast, liver, kidney, lung, color, skin, and pancreatic cancers, we are systematically documenting the spatial dependencies of genes. This includes identifying correlations that are consistent across different tumors, as well as those unique to specific tumors or tumor types. Early observations have highlighted modules of chemokines (CCL3/CCL4/CCL3L, CXCL1/CXCL2/CXCL3), genes indicative of CAF phenotype (FN1 and collagens), and those involved in cell adhesion and collagen formation. For research use only. Not for use in diagnostic procedures. Citation Format: Patrick Danaher, Dan McGuire, Michael Patrick, David Kroeppler, Joachim Schmid, Joseph M. Beechem. Uncovering gene co-regulation networks in spatial transcriptomics data 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 (6Suppl): Abstract nr 2331.