Abstract / Summary
Abstract Although myofibroblastic cancer-associated fibroblasts (myCAFs) drive colorectal cancer (CRC) progression, the transcriptional dynamics underlying their metastatic maturation remain poorly understood. Here, we integrated five single-cell RNA sequencing datasets and two bulk transcriptomic cohorts encompassing normal colon, primary tumors (pCRC), and metastatic CRC (mCRC), combining pseudotime trajectory inference, consensus molecular subtyping (CMS), and ensemble machine learning. Findings were experimentally validated by qRT-PCR across 185 clinical specimens and four CRC cell lines. Trajectory modeling revealed a progressive myCAF continuum toward an extracellular matrix (ECM)-remodeling phenotype enriched in metastases ( P < 0.05). This myCAF maturation was associated with stroma-rich CMS4 enrichment, increasing from 14.7% in pCRC to 72.9% in liver and 90.0% in peritoneal metastases ( P < 2.2 × 10 − 16 ). Ensemble machine learning prioritized CAVIN1 and SULF1 as top discriminative features (AUCs > 0.90, P < 0.0001), exhibiting synchronized dynamics with canonical myCAF markers ( r = 0.79, P = 2.48 × 10 − 5 ). Experimental validation confirmed their fibroblast-specific expression, significant metastatic upregulation, and robust discriminatory capacity in distinguishing metastatic from pCRC (AUCs ≥ 0.80, P < 0.0001). Delineating the CAVIN1 / SULF1 -associated myCAF maturation continuum provides mechanistic insights into stroma-driven progression and reveals candidate biomarkers for metastatic risk stratification in CRC.