Stage-dependent upregulation of tumor-derived MDK is associated with CAF state remodeling and an ECM-rich immunosuppressive microenvironment in colorectal cancer

Abstract Background Cancer-associated fibroblasts (CAFs) exhibit substantial heterogeneity in colorectal cancer (CRC), yet the upstream tumor-derived signals associated with their state remodeling during disease progression remain incompletely defined. Through integrative analysis of multi-cohort single-cell transcriptomics, proteomics, and bulk RNA sequencing data, we identified tumor-derived midkine (MDK) as a stage-associated candidate signal linked to CAF remodeling in CRC. Methods MDK cell-type specificity and stage-associated expression, together with CAF composition and stage-dependent remodeling, were evaluated in two large colorectal cancer single-cell RNA-sequencing cohorts, with the principal findings cross-validated between cohorts and further integrated with bulk transcriptomic, proteomic, and immunohistochemical datasets. Existing malignant-cell-to-CAF CellChat results were re-ranked using a joint breadth–strength framework to identify broadly activated and high-probability signaling pathways. CAF trajectories were reconstructed independently in both cohorts using Monocle 2 and rooted according to the same biological criterion, namely enrichment of the normal-like Fib-MFAP5 population. BayesPrism was used to deconvolve CAF subtype abundance from TCGA colorectal cancer bulk transcriptomes and to evaluate its association with overall survival. Results MDK was consistently enriched in malignant epithelial cells across both single-cell cohorts. Within the same-platform bulk RNA cohort, median log2-normalized MDK expression increased from 7.93 in adenoma to 8.19 in carcinoma in situ and 9.02 in adenocarcinoma, supporting a significant ordered increase during colorectal cancer progression (Jonckheere–Terpstra $$P=0.008$$ ; 10,000-permutation $$P=0.007$$ ). Fib-WNT5A abundance likewise showed a stage-associated increase and was positioned significantly later than the normal-like Fib-MFAP5 population in patient-level pseudotime analyses of both the discovery and validation cohorts. Fib-WNT5A exhibited strong TGF-β and extracellular matrix programs, and integration of subtype-associated ECM signatures with bulk proteomics linked this population to collagen-enriched matrix remodeling. Joint breadth–strength analysis identified MK signaling as the top-ranked broadly activated malignant-cell-to-CAF pathway. CellChat further predicted COL1A1/COL1A2–CD44 interactions between Fib-WNT5A and multiple immune populations, supporting an association between Fib-WNT5A expansion and an ECM-rich, immune-restrictive microenvironment. Patient-level pseudo-bulk analyses were used to evaluate BayesPrism recovery of major CAF states. In TCGA CRC, higher Fib-WNT5A abundance was associated with poorer overall survival in the prespecified median-based univariable analysis (HR = 1.64, 95% CI 1.09–2.49; log-rank P = 0.018), whereas higher Fib-MFAP5 abundance was associated with better survival (HR = 0.54, 95% CI 0.35–0.83; log-rank P = 0.0045). Conclusions Collectively, the data support a progression-linked association among tumor-cell-enriched MDK, CAF state remodeling, an ECM-high Fib-WNT5A program, and predicted collagen-mediated immune interactions in CRC. Functional perturbation studies are required to establish the direction and causality of the proposed MDK-CAF axis.