Abstract
Despite advances in blood-based screening tests for colorectal cancer (CRC), most existing assays focus on DNA-based biomarkers, which predominantly reflect tumor-derived fragments released at later disease stages. In contrast, whole-blood transcriptomic profiling can capture systemic immune responses and tumor-host interactions, offering a complementary strategy for earlier disease detection. However, clinically validated whole-blood transcriptomic signatures remain limited. Here, we investigated a whole-blood RNA-based biomarker discovery strategy by integrating multi-cohort transcriptomic resources. Public GEO datasets (GSE164191 and GSE11545) were harmonized and analyzed, yielding 956 differentially expressed genes (DEGs). Multi-layer biological filtering incorporating PPI networks, transcription factors, CRC-related GWAS variants, whole-blood eQTL signals, DigSeE, and CoReCG disease associations refined these to 375 high-confidence transcripts (WB-PADs). In parallel, RNA-seq analysis of a Korean cohort (10 CRC vs. 10 controls) identified 217 DEGs (WB-K). Cross-dataset convergence highlighted seven overlapping transcripts, and five candidates (DLG5, CD177, SH2D1B, NQO2, and KRT73) were selected for validation. RT-qPCR in an independent clinical cohort (106 CRC and 123 healthy controls) confirmed four transcripts with significant discriminatory ability. A multivariable logistic regression model derived from the five-transcript signature achieved an AUC of 0.952 (95% CI 0.884-1.000), with sensitivities of 0.889 and 0.667 at fixed specificities of 90% and 95%, respectively, demonstrating strong applicability for screening-relevant thresholds. Notably, the model retained high accuracy in early-stage CRC (Stage I-II: AUC 0.929, 95% CI 0.868-0.989). Overall, this study provides a robust analytic framework for reproducible whole-blood RNA biomarker discovery and establishes a multi-gene signature with promising translational potential for minimally invasive and early CRC detection.