Diagnostic and Prognostic Gene Signatures Linking Endoplasmic Reticulum Stress and Metabolic Reprogramming in Colon Cancer

Objective: Globally, Colon Cancer ranks among the leading cancers in incidence and mortality, and existing diagnostic and therapeutic options have considerable limitations, highlighting the need for novel biomarkers and therapeutic targets. This study aimed to integrate multi-cohort bulk and single-cell transcriptomic data to identify genes jointly associated with endoplasmic reticulum stress (ERS) and metabolic reprogramming in colon cancer, develop diagnostic and prognostic models, and characterize their immune and cell-type-specific contexts. Methods: TCGA-COAD RNA-sequencing data were retrieved usingTCGAbiolinks and merged with GTEx data to build the GTEx-TCGA-COAD cohort (290 A and 349 control samples). GSE17536 (177 samples) served as an external prognostic cohort, and GSE200997 was analyzed for single-cell RNA-sequencing. ERS-related genes (ERSRGs, n = 1,536) and metabolic reprogramming-related genes (MRRGs, n = 843) were compiled from GeneCards and published studies. Limma, WGCNA, clusterProfiler, and STRING wereused for differential expression,co-expression network,enrichment, and protein–protein interaction (PPI) analyses, respectively. An AI-assisted feature-selection strategy integrating random forest, support vector machine, and LASSO was used to identify key genes. A logistic regression model was evaluated using receiver operating characteristic, calibration, and decision curve analyses. A LASSO-Cox risk model was evaluated using Kaplan–Meier survival analysis. Immune infiltration, pathway enrichment, protein structure prediction, single-cell analysis, and virtual knockout were performed using CIBERSORT, Gene set enrichment analysis, AlphaFold 3, Seurat, and scTenifoldKnk, respectively. Results: Differential expression analysis identified7,934 DEGs, with 154 ERS- and metabolic reprogramming-related differentially expressed genes (ERSMRRDEGs) identified through intersection with ERSRGs and MRRGs. WGCNA identified15 co-expression modules, yielding21 genes through further intersection. GO and KEGG analyses identified pathways related to transcription-factor regulation, nuclear import, membrane rafts, ER lumen, ubiquitin protein ligase binding, and disease-associated pathways. AI-assisted feature selection identifiedseven key genes (CLU, TIMP1, CTNNB1, CDKN2A, GDF15, VDAC1, and TMEM33). The seven-gene logistic regression model achieved an area under the curve >0.9 in the GTEx-TCGA-COAD cohort. The prognostic risk score based on TIMP1, CDKN2A, and GDF15 significantly stratified overall survival in both GTEx-TCGA-COAD and GSE17536 (both p < 0.05). Immune analysis identified 19 differentially abundant cell types (p < 0.05); GDF15correlated positively with M0 macrophages (r = 0.704, p < 0.05) and negatively with resting mast cells (r = −0.719, p < 0.05). Single-cell analysis revealed that epithelial cells predominantly expressed key genes. AlphaFold 3 predicted high-confidence structural domains for TIMP1 and CDKN2Aand a medium-confidence domain for GDF15. Virtual knockout linked GDF15 perturbation to Translation Factors (FDR = 0.015) and TIMP1 perturbation to translational elongation (FDR = 0.0006) and cytokine signalingin the immune system (FDR = 0.017). Conclusion: This integrative transcriptomic study identified candidate ERS- and metabolic reprogramming-related biomarkers for Colon Cancer classification and prognostic stratification and generated cell-type-specific and regulatory hypotheses for GDF15 and TIMP1. Further experimental and independent clinical validation is required before clinical or therapeutic application.