Background: Colorectal cancer (CRC) is the third leading cause of cancer-related deaths worldwide, highlighting the critical need for reliable diagnostic markers to improve prognosis. Current screening methods are often invasive or have limited sensitivity. Extracellular vesicles have emerged as key players in tumor progression. These stable vesicles carry a molecular cargo that mimics the fingerprint of their parent tumor cells, including tumor-specific miRNAs. This unique property makes them ideal non-invasive biomarkers for early cancer diagnosis. In this study, we performed a comprehensive analysis to identify EV-miRNAs as potential biomarkers for the early diagnosis and treatment targets for CRC. Methods: We integrated differential expression analysis, machine learning (ML), Mendelian randomization (MR), and clinical validation to identify EV-derived miRNAs associated with CRC diagnosis. Publicly available EV-miRNA dataset GSE188627 were analyzed to identify differentially expressed miRNAs (DEmiRs). Within each fold of five-fold nested cross-validation, differential expression analysis, LASSO, and SVM-RFE were applied for feature selection, followed by training and evaluation of four machine learning methods (LASSO, RF, SVM-BFE and XGBoost) on the corresponding test set. Two-sample MR was conducted using miRNA-expression quantitative trait loci (eQTL) from the Rotterdam Study and CRC GWAS summary statistics from IEU OpenGWAS to investigate the causal relationship between miRNA and CRC. Results: After rigorous filtering, 570 EV-related miRNAs from 102 CRC patients and 90 controls were analyzed from GSE188627. Each outer training fold yielded a candidate feature set by integrating differential expression and feature selection. Hsa-miR-125a-5p, hsa-miR-194-5p, and hsa-miR-32-5p were shared across all five feature sets. Among four machine learning models, the XGBoost model had mean Accuracy=0.859, sensitivity=0.833, specificity=0.889, PPV=0.896, NPV=0.833, F1=0.860, MCC=0.726, and AUC=0.940 (95% CI: 0.852–0.996). Mendelian randomization further supported a protective role of miR-125a-5p (OR= 0.87, 95% CI: 0.79–0.95, P=0.002), which was validated by its lower expression in CRC patients, underscoring its potential as a diagnostic biomarker. No causal relationship was found between hsa-miR-194-5p (OR=0.99, 95% CI: 0.91–1.07, P=0.73) or hsa-miR-32-5p (OR=1.04, 95% CI: 0.99–1.09, P=0.14) and CRC. Functional network analysis implicated both miRNAs in key CRC-related pathways and gene targets (e.g., BCL2, TP53). Conclusion: Our integrative multi-step approach identifies hsa-miR-125a-5p as a potential diagnostic biomarker in CRC. These findings offer mechanistic and translational insights into the role of EV-miRNAs and warrant further validation in larger, multi-ethnic cohorts.
Xinlong Shi (Thu,) studied this question.