A propionate metabolism-related gene signature predicts prognosis and immunological features in colorectal cancer.

Abstract

Propionate metabolism has emerged as a key metabolic axis shaping colorectal cancer (CRC) biology and the tumor microenvironment; however, propionate metabolism-related molecular stratification and its prognostic and therapeutic implications remain insufficiently characterized. Transcriptomic profiles of The Cancer Genome Atlas colon adenocarcinoma (TCGA-COAD) and rectum adenocarcinoma (TCGA-READ) were integrated to construct a CRC training cohort, and GSE38832 served as an external validation set. Propionate metabolism-related genes (PMRGs) were curated from GeneCards and intersected with differentially expressed genes to define differentially expressed PMRGs (DEPMRGs). Prognostic DEPMRGs were screened by univariate Cox regression and used for consensus clustering to identify propionate metabolism-related subtypes, followed by functional enrichment and immune landscape profiling (ESTIMATE, single-sample Gene Set Enrichment Analysis, and CIBERSORT). After adjustment for clinicopathological characteristics, Cox proportional hazards modeling was performed to verify whether the signature retained independent prognostic relevance. To facilitate individualized risk estimation, a nomogram was constructed from the significant variables, and its reliability and potential value in clinical decision-making were further evaluated through calibration analysis and decision curve analysis. Immunotherapy relevance was explored using Immunophenoscore and Tumor Immune Dysfunction and Exclusion, and anti-PD-L1 response was assessed in the IMvigor210 cohort. To characterize molecular differences across risk subgroups, the mutational landscape and tumor mutational burden were profiled with the maftools package, whereas potential therapeutic susceptibility was inferred from pRRophetic prediction and CellMiner-based drug response analyses. Application of consensus clustering to prognostic PMRG-related genes revealed two molecular classes in CRC, which were distinguished by different prognostic patterns and immune microenvironmental characteristics. A four-gene prognostic signature was then established, enabling classification of patients into low- and high-risk subsets with clearly separated survival curves in the training cohort; this predictive pattern was further reproduced in GSE38832. Comprehensive immune characterization demonstrated that the two risk groups differed substantially in immune-cell infiltration, checkpoint-gene expression, and immunotherapy-associated signatures, with the low-risk group showing a more favorable predicted response in the IMvigor210 cohort. Propionate metabolism-associated subtyping and a four-gene prognostic signature enable robust risk stratification in CRC and link metabolic programs to immune contexture and therapeutic vulnerability, providing a potential framework for prognosis assessment and individualized treatment.