Abstract

BackgroundColorectal cancer still causes many cancer deaths, and patient outcomes differ a lot. The tumor microenvironment can shape tumor growth and treatment response. Autophagy is a cell recycling process linked to tumor survival and immune control, but its cell-type pattern in colorectal cancer tissue and its clinical meaning are not clear. We mapped autophagy activity across cell types, defined autophagy-based subtypes, and tested their value for risk stratification.MethodsSingle-cell RNA sequencing data from colorectal cancer tumor tissues were integrated to evaluate autophagy activity across malignant, stromal, and immune cell populations. Subsequently, autophagy-related genes were curated from multiple public databases and analyzed in bulk transcriptomic cohorts obtained from The Cancer Genome Atlas and The Gene Expression Omnibus. Additionally, unsupervised consensus clustering was applied to define autophagy-based molecular subtypes. Multiple machine learning algorithms were evaluated to construct an autophagy-related prognostic model, and feature importance was assessed using explainable modeling approaches. Furthermore, immune microenvironment characteristics, genomic alterations, and stemness features were systematically analyzed. Key genes were further validated in clinical colorectal cancer specimens and through in vitro functional experiments.ResultsAutophagy activity varied across the tumor microenvironment. Malignant and stromal cells showed higher autophagy levels than immune cells. High-autophagy tumors showed stronger tumor–stroma interactions, weaker immune communication, and an immune-suppressive microenvironment. Using 47 prognostic autophagy-related genes, we identified four subtypes with different overall survival, genomic alteration patterns, stemness signatures, and immune landscapes. The prognostic model performed well across independent cohorts. GOLGA2 (Golgi autoantigen, golgin subfamily A member 2) was the top risk gene and was upreg