Integrated single-cell and spatial transcriptomic analysis reveals mCAF-SPP1⁺ macrophage-T cell crosstalk shaping immunosuppressive niches in colorectal cancer liver metastasis.

BACKGROUND: Colorectal cancer liver metastasis (CRC-LM) carries poor prognosis and responds poorly to current immunotherapies. Liver metastases often display T-cell exclusion, but how stromal and myeloid circuits jointly shape T-cell states across primary and metastatic sites, and how to capture this in robust prognostic markers, remains unclear. METHODS: We integrated multi-cohort single-cell RNA sequencing datasets of primary colorectal cancer and CRC-LM to define tumour-associated macrophages (TAMs), cancer-associated fibroblasts (CAFs) and T-cell subsets, and used CellChat and NicheNet to map ligand–receptor networks and downstream programs. Visium spatial transcriptomics with cell2location deconvolution delineated stromal–immune architectures in CRC-LM. A stromal–myeloid gene signature derived from key populations was evaluated in TCGA-COAD/READ and GSE39582, and a GCC-targeted second-generation CAR-T model was used to test the impact of recombinant SPP1 and MIF on tumour killing. RESULTS: In this study, we found that CRC-LM were enriched for stress-response (TSTR) and exhausted (Tex) T cells and featured a recurrent mCAF–SPP1⁺ TAM–Tex/TSTR niche. Spatial analysis revealed a stereotyped three-layer organization at the tumour–stroma interface, with outer mCAF belts, inner SPP1⁺ TAM–rich zones and intervening TSTR/Tex bands. Communication analysis identified SPP1–CD44, FN1–CD44 and MIF/CXCL12 as dominant stromal–myeloid axes promoting T-cell retention, stress and exhaustion. In GCC-directed CAR-T assays, recombinant SPP1 and MIF impaired tumour clearance, and a three-gene stromal–myeloid signature (KLF2, ZBTB20, ARL4C) robustly stratified recurrence and disease-free survival independent of TNM stage. CONCLUSIONS: Integrating single-cell, spatial and functional data, we define a stromal–myeloid niche that underlies immune confinement in CRC-LM and highlight actionable ligand–receptor pathways. The three-gene stromal–myeloid signature provides a concise, mechanism-based tool for risk stratification and patient selection in colorectal cancer.