Abstract
BACKGROUND: Radiotherapy is a fundamental component of rectal cancer treatment, yet patient responses remain highly heterogeneous due to the lack of reliable biomarkers supported by genomic variation evidence. Integrating multi-cohort transcriptomic data with machine-learning approaches enables systematic identification of genes with both predictive and therapeutic relevance. This study aimed to develop a robust model for predicting radiotherapy response and to functionally characterize CES1 as a key regulator of radiosensitivity and a potential therapeutic target. METHODS: Three independent GEO cohorts were standardized and integrated, followed by a comprehensive machine-learning pipeline incorporating LASSO, Elastic Net, Random Forest, XGBoost, and information gain. A consensus-ranked five-gene model was constructed using nested cross-validation. CES1, identified as the top-ranked contributor to model performance, was selected for biological validation. To connect transcriptomic findings with genetic variation, copy number alteration (GISTIC2) and somatic mutation analyses were performed in TCGA-READ. Functional assays-including quantitative PCR, CCK-8 viability assays, colony formation, wound-healing migration assays, Annexin V/PI flow cytometry, and rescue by CES1 overexpression-were performed in HT-29 and SW480 cells to evaluate its mechanistic role in radiotherapy response. RESULTS: The machine-learning model demonstrated high discriminative accuracy across datasets and consistently highlighted CES1 as a dominant contributor to radiosensitivity prediction. CES1 expression increased after clinical chemoradiotherapy and showed dose-dependent induction following irradiation in vitro. CES1 knockdown significantly reduced radiation-induced apoptosis, enhanced clonogenic survival, and promoted migratory capacity, collectively indicating a radioresistant and more aggressive phenotype. Restoration of CES1 expression in CES1-silenced cells reversed radioresistance and re-established irradiation sensitivity. Genomic analysis in TCGA-READ further demonstrated that CES1 expression was positively associated with copy number status, whereas coding-sequence mutations in CES1 were infrequent, suggesting dysregulation primarily through copy-number and transcriptional mechanisms. CONCLUSION: This integrative computational and experimental study identifies CES1 as a predictive biomarker and copy number-linked regulator of radiosensitivity in rectal cancer. Modulation of CES1 directly alters cellular responses to irradiation, supporting its role as a mechanistically interpretable biomarker for response stratification. These findings align with the emerging concept that integrating genetic variation profiling with functional validation can accelerate variant-to-biomarker translation in precision oncology.