Metabolic remodeling in early- and late-onset colorectal cancer supports an age-informed detection model

The incidence of early-onset colorectal cancer (EOCRC) has increased in recent decades; however, most metabolomic studies have focused on all age groups. Consequently, it remains uncertain how metabolic alterations drive EOCRC progression and whether biomarker signatures identified in older patients are valid in younger populations. We performed untargeted plasma metabolomic profiling of 791 participants, including those with colorectal cancer (CRC), colorectal adenoma (CRA), and healthy controls. Individuals were stratified by age (<50 and ≥50 years) to explore metabolic remodeling during EOCRC and late-onset CRC (LOCRC) development. A multistep feature selection process and 12 machine learning algorithms were used to establish an age-informed CRC detection model. CRC progression was associated with the suppression of caffeine and histidine metabolism and the enhancement of amino acid metabolism, particularly branched-chain amino acid, glycine-serine, and arginine-proline pathways. EOCRC showed increased phenylalanine, butanoate, and lipoic acid metabolism, whereas LOCRC was enriched in purine and fatty acid biosynthesis. A 7-metabolite panel (caffeine, ergothioneine, hypoxanthine 1, L-valyl-L-alanine, L-seryl-L-aspartic acid, cholic acid, and cycloleucine) was trained and achieved excellent detection values in all age groups (AUC > 0.95). SHapley Additive exPlanations (SHAP) analysis highlighted metabolite-specific contributions, supporting biological interpretability. These findings delineate age-related metabolic remodeling in CRC and demonstrate that integrating plasma metabolomics with machine learning supports a robust and interpretable non-invasive detection approach. The proposed age-informed model may inform future age-tailored CRC detection strategies amid the increasing burden of early-onset cancers.