Detection of early-stage colorectal cancer using multiplexed extracellular vesicle proteomic biomarkers.

3616 Background: Colorectal cancer (CRC) is very common worldwide, and detection of early-stage cancer is critical for improved survival rates. Advanced adenoma (AA) and early-stage CRC present challenges due to their small size and very low plasma expression of tumor-specific biomarkers. Genomics-based diagnostics struggle to detect these lesions, making it imperative to develop more sensitive approaches. Extracellular vesicles (EVs) are emerging as a promising solution for early-stage CRC detection. EVs are produced by tumor, tumor microenvironment and host cells, thus, unbiased analysis of all plasma EVs offers an expanded set of cancer specific biomarkers, and may aid in detection of small early-stage tumors. Methods: EVs were purified from patient plasma using size exclusion chromatography and a proprietary buffer system that enhances EV and corona recovery. TrueDiscovery Data-independent acquisition mass spectrometry (MS) analysis was conducted on EVs purified from 24 advanced adenoma (AA)/stage 0 dysplasia, 25 Stage 1 CRC and 75 normal patient plasma samples. An in-house Machine Learning (ML) pipeline was developed to identify differentially expressed proteins and to define protein multiplexes with extremely high Sens/Spec (> 0.99). Results: Around 2,500 QC’d proteins were ID'd per sample and ML identified 336 (AA/Stage 0) and 493 (Stage 1) differentially expressed proteins (DEP: absolute Log2FC>0.5; q-value <0.001; AUC>0.5). The DEPs from AA/Stage 0 and Stage 1 were trained and tested using ML and SMV to identified dozens of MS based multiplexes for each stage with near perfect diagnostic accuracy (~98-100%). These biomarkers were enriched with metabolic, immune and inflammatory proteins in line with our unbiased approach. MS protein multiplexes were then translated to ELISAs to build simple, high throughput assays. ELISA analysis of multiple biomarkers was performed in the training cohort, ML/SMV was used to define multiplex ELISA-based classifiers for Stage 1 CRC (Sen=0.95/Spec=0.96). This Stage 1 CRC training assay was locked and tested on a blinded cohort of Stage 1 CRC (30 control, 30 cancer). In this blinded cohort, we targeted maximal specificity yielding Sen=0.80, Spec=0.97 and AUC=0.95. These data represent significantly higher Sen/Spec than current genomic based liquid biopsy assays. Training and validation of AA and Stage 0 ELISAs is underway and will be presented. Conclusions: Blinded validation of EV proteomic biomarkers yielded extremely high Sen/Spec early-stage CRC. The unbiased EV analysis underscores the need to assess cancer specific biomarkers from TME and host response and highlights the value and opportunity of expanding liquid biopsy analysis beyond tumor specific genomics. Taken together this platform provides an exceptional opportunity to increase early-stage cancer detection which should result in better patient outcomes.