Spatial transcriptome mapping identifies endothelial anergy and CD4⁺ T cell stress as key drivers of immune escape in colorectal cancer liver metastasis

Background Colorectal liver metastasis (CRLM) remains poorly responsive to immunotherapy, yet the mechanisms underlying its immune-resistant microenvironment are not fully understood. To address this, we generated a single-cell spatial transcriptomic atlas from four replacement-only, microsatellite-stable CRLM treated with neoadjuvant FOLFOX chemotherapy, representing the most common clinical scenario. Our dataset spans intratumoral, peritumoral, and adjacent normal liver, comprising 1,930,656 cells. In parallel, we performed thirteen whole-slide multiplexed immunohistochemistry to dissect the spatial cellular and molecular interactions shaping the immune landscape in CRLM. Results We identified a progressive shift in endothelial phenotypes across regions: from a non-inflamed state in the normal liver, to a highly inflamed peritumoral endothelium characterized by TNFalpha, NFκB, STING and inflammasome signature enrichment, and finally to an anergic, angiogenic endothelial population within CRLM. This anergic endothelial state was defined by reduced expression of adhesion molecules required for T cell extravasation, including SELE, SELP, ICAM1 and VCAM1, in contrast to the more immune-permissive peritumoral endothelium. In line with these endothelial differences, the peritumoral region contained a high density of T cells, most of which were TCR-reactive yet displayed features of exhaustion. The intratumoral region contained far fewer T cells ( p <0.01), and those present were predominantly CD4⁺ cells adopting a stress-response phenotype, characterized by an upregulation of genes related to integrated stress response, unfolded protein response, and heat-shock protein pathways. These intratumoral CD4⁺ T cells lacked cytokine expression, showed minimal TCR activity, and localized to hypoxic niches. Their transcriptional profile reflected a hypoxia- and PGE2-driven stress programme rather than classical exhaustion, suggesting that CD4⁺ T cell dysfunction in CRLM is driven by stress-induced pathways. Fibroblasts also demonstrated marked regional variation. In CRLM, fibroblasts secreted neutrophil-attracting chemokines CXCL1 and CXCL8, and produced ECM-remodeling factors such as LOXL2 and TNC. This cancer-associated fibroblast population showed strong spatial correlation with ‘glial-like' cells, which were found almost exclusively within CRLM ( p <0.0001). These glial-like cells expressed neural ligands, including NGF and NPY, supporting a glial-fibroblast signaling axis that may reinforce fibroblast activation and matrix deposition. Conclusions Together, these findings reveal key mechanisms driving immune exclusion in CRLM, highlighting endothelial anergy, intratumoral CD4⁺ T cells in a stress-response state, and a glial-fibroblast axis as central features of its immunosuppressive landscape.