Extracellular Vesicle Gene Expression Enables Sensitive Detection of Colorectal Neoplasia

Extracellular vesicles (EVs), including exosomes, are emerging as promising carriers of disease-specific biomarkers due to their molecular cargo reflective of cellular origin. While cell-free DNA (cfDNA) methylation assays have recently been developed for colorectal cancer (CRC) screening and perform well for cancer detection, they show limited sensitivity for advanced adenomas (AA), a key precursor in the CRC pathway. To address this gap, we sought to determine whether other blood-based analytes, specifically EV-derived long and small RNA transcriptomes and EV proteomes, could improve detection of AA and early-stage CRC. In a prospective cohort, we performed a head-to-head comparison of EV transcriptomics, EV proteomics, and their combinations against cfDNA methylation, all measured from the same patient cohort, enabling a direct performance benchmark and identification of the most promising modality for larger-scale CRC screening studies.

METHODS

We prospectively collected pre-colonoscopy plasma samples from 220 participants across three clinical sites. EVs were isolated and profiled using long RNA-seq, small RNA-seq, and Olink-based proteomics. cfDNA was analyzed for methylation patterns. Analyses were conducted according to a statistical analysis plan pre-specified before unblinding. Machine learning models were developed under nested cross-validation to evaluate sensitivity for detecting AA and CRC at a fixed specificity of 91%, consistent with clinical screening benchmarks.

RESULTS

EV-derived gene expression on long RNA demonstrated the highest sensitivity for detecting AA as well as CRC: 54.8% (95% CI, 26.4%–75.6%) for AA and 94.1% (95% CI, 79.2%–100%) for CRC. This outperformed cfDNA methylation (33.6% [95% CI, 9.7%–60.3%] for AA, 81.3% [95% CI, 64.6%– 93.9%] for CRC) and other EV-based modalities. In addition, for small RNA the sensitivities were 42.6% (95% CI, 35.9%-47.4%) for AA, and 78.9% (95% CI, 69.2%-83.0%) for CRC, while for proteomics the sensitivities were 30.0% (95% CI, 13.9%-40.0%) for AA, and 64.2% (95% CI, 43.6%-87.2%) for CRC. Transcriptomic profiles revealed progressive enrichment of hallmarks of cancer pathways, including apoptosis and epithelial-mesenchymal transition, across disease stages. Multiomic integration did not improve performance beyond EV transcriptomics alone.

CONCLUSIONS

By directly comparing multiple EV-based and cfDNA analytes within the same patient cohort, we found that EV transcriptomics delivers the strongest diagnostic performance for both AA and CRC. This rigorous benchmarking approach allows clear prioritization of the most promising modality guiding the design of larger validation studies and accelerating development of next-generation, blood-based CRC screening tools.