Targeting molecular residual disease and immune escape in colorectal cancer recurrence: biomarker-guided immunotherapy and cell therapies.

Colorectal cancer (CRC) recurrence after curative-intent treatment remains a major clinical challenge. Conventional postoperative risk stratification relies on clinicopathological variables, whereas recurrence is increasingly understood as a dynamic biological process shaped by molecular residual disease (MRD), immune editing, tumor microenvironmental remodeling and therapy-resistant cellular states. Circulating tumor DNA (ctDNA) has emerged as a clinically relevant MRD biomarker for identifying patients at high risk of relapse and for refining adjuvant-treatment decisions. At the same time, immune biomarkers such as mismatch repair deficiency, microsatellite instability, tumor mutational burden, antigen-presentation status, T-cell exhaustion, myeloid suppression and spatial immune exclusion determine whether recurrent disease is susceptible to immunotherapy. This Mini Review discusses how MRD and immune escape can be integrated into a biomarker-guided framework for recurrent CRC. We summarize recent evidence supporting ctDNA-guided recurrence surveillance, immune checkpoint blockade in MSI-H/dMMR CRC, vaccine-based strategies targeting neoantigens or mutant KRAS, and engineered cell therapies including CAR-T, CAR-NK and TCR-engineered approaches. We also discuss how interpretable machine-learning models and SHAP-based feature attribution may help prioritize recurrence biomarkers and enrich clinical trials, provided that models undergo rigorous external validation. A biomarker-guided strategy linking MRD detection, immune profiling and cell-therapy target selection may provide a translational pathway toward more individualized immune intervention for recurrent CRC. In this framework, AI is positioned as an auditable decision-support layer that links ctDNA kinetics, immune biomarker domains, cell-therapy eligibility, and trial-enrichment decisions rather than as an autonomous treatment selector.