SpaceMarkers 2.0: A framework for spatially aware cell-cell communication in spatial transcriptomic data

Background: One key limitation of single cell RNA-sequencing (scRNA-seq) is the loss of spatial resolution due to the dissociation of cells from their tissue. Despite this limitation, current computational approaches infer cell-cell interactions from scRNA-seq and often generalize findings to the tissue context. SpaceMarkers was developed to infer cell-cell communication directly in-situ using 10x Visium. SpaceMarkers 2.0 is extended to leverage both Visium and VisiumHD datasets, as illustrated using samples from colorectal cancer (CRC) patient tumor and normal adjacent tissue. Method: SpaceMarkers uses spatially varying features as proxies for cellular activity and applies kernel-based smoothing to represent the ‘influence' from cell populations in their vicinity. We hypothesize that two cell states interact in regions where their cellular weights and influence overlap. A Gaussian mixture model is used to separate background from enriched values to quantify ‘hotspots' within a given tissue. Previously, SpaceMarkers inferred molecular changes due to cell-cell communication for individual genes in the transcriptome. Here, we added the ability to incorporate prior knowledge of ligand-receptors (LR) in pairs of interacting cell populations by computing ligand overexpression within a cell population under the influence of a spatially overlapping population. We also estimate receptor scores from statistical tests to identify cell type-specificity for each receptor. SpaceMarkers calculates an aggregated LR score based on the geometric mean of the ligand and receptor score enriched in the tissue-defined interaction region. We extend this by computing per-sample and grouped condition hotspot overlaps as well as applying rank statistics to identify molecular programs between cancer and normal tissue. Results: SpaceMarkers 2.0 identified spatial hotspots where normal samples were enriched for plasma to epithelial cell signaling programs while tumor samples were enriched for stromal and plasma cell interactions. At the plasma-stromal boundary we observed a tumor specific enrichment of fibroblast growth factor (FGF) ligand signaling from plasma cells to stromal cells, with FGF23 emerging as a dominant ligand engaging FGFR4 and FGFR3 on stromal cells. Conclusion: In CRC Visium and Visium HD datasets, this framework delineates tumor specific interaction boundaries and resolves which cell states drive LR axes in-situ across patients. Beyond CRC, the SpaceMarkers framework can be applied to additional solid tumors, enabling comparisons in cohorts of samples in matched conditions or longitudinal sampling of carcinogenesis. Overall, SpaceMarkers 2.0 proposes a central framework for leveraging spatial information and multi-sample modeling for interpreting cellular communication in the tumor microenvironment.