Cancer-associated fibroblasts (CAFs) significantly influence tumor progression and therapeutic resistance in colorectal cancer (CRC). However, the distributions and functions of CAF subpopulations vary across the four consensus molecular subtypes (CMSs) of CRC. This study performed single-cell RNA and bulk RNA sequencing and revealed that myofibroblast-like CAFs (myCAFs), tumor-like CAFs (tCAFs), inflammatory CAFs (iCAFs), CXCL14+CAFs, and MT+CAFs are notably enriched in CMS4 compared with other CMSs of CRC. Multiplex immunohistochemistry was used to validate the distribution of CAF subtypes in patients with different CMSs. Prognosis-related CAF subtypes were identified, leading to the selection of four key genes (COL3A1, COL1A2, GEM, and TMEM47). Through machine learning, we developed a CAF poor-prognosis gene (CAFPRG) model to predict outcomes of patients with CMS4. High levels of CAFPRGs were identified as independent poor-risk factors for prognosis (p < 0.001). Tumors with elevated CAFPRGs exhibited increased infiltration of immune-suppressive cells and resistance to chemotherapy. The expression of these key genes was confirmed to be significantly higher in CAFs than in normal fibroblasts (NFs). Therefore, CAFPRGs may be valuable for precisely predicting patient survival and may present potential therapeutic opportunities for CMS4 CRC.
Keywords: CAF heterogeneity; Colorectal cancer; Consensus molecular subtypes; Machine learning; scRNA-seq.
© 2024. The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.