Multi-omics analysis reveals features improving outcomes and immunotherapy resistance in MSS colorectal cancer

In contrast to the MSI-H subtype, most microsatellite-stable (MSS) colorectal cancer (CRC) patients exhibit poor prognosis and limited responsiveness to immunotherapy, highlighting the urgent need for robust biomarkers to stratify prognosis and predict immunotherapy responsiveness. Here, 68 single-cell and 3,767 bulk transcriptome samples were integrated to identify and validate MSS patients with high immune infiltration. Multidimensional, prognosis-related, and immunological features specific to MSS CRC were characterized at cellular and genetic levels. To reduce algorithm selection bias, 93 artificial intelligence algorithms were applied to construct a machine learning-derived MSS prognostic-immune signature (MLDMPIS). MLDMPIS demonstrated robust prognostic and immunotherapy response prediction performance in 1,323 MSS patients and six independent immunotherapy cohorts, outperforming clinical indicators, published prognostic signatures, and immune response models. Mechanistically, CADM1 was identified as a key molecule associated with MLDMPIS. Functional analyses revealed that CADM1 promotes tumor stemness, thereby enhancing proliferation, invasion, and metastasis in MSS CRC.