Highly accurate detection of early-stage colorectal cancer using tumor and immune extracellular vesicles biomarkers.

262 Background: Colorectal cancer (CRC) is one of the most common cancers worldwide, and early detection is critical for successful treatment and improved survival rates. However, precancerous and early-stage CRC present significant diagnostic challenges due to the small size of lesions and the very low expression of tumor-specific biomarkers in the bloodstream. Current genomics-based diagnostic methods struggle to detect these early-stage and precancerous lesions, making it imperative to develop more sensitive and specific approaches. Circulating extracellular vesicles (EVs) are emerging as a promising solution for early-stage CRC detection. Since EVs are produced by tumor cells as well as tumor microenvironment and host immune response cells, EVs offer the means to detect and monitor small early-stage tumors through both direct detection of tumor-associated biomarkers as well as tumor specific host response and tumor microenvironment biomarkers. Methods: To identify novel early-stage CRC specific biomarkers, we performed proteomics analysis on EVs purified from patient plasma using size exclusion chromatography and a proprietary buffer system that enhances EV and corona protein recovery. TrueDiscovery Data-independent acquisition (DIA) mass spectrometry (MS) analysis was conducted on 24 pre-cancer/stage 0 and 25 stage 1 CRC patients and 75 normal patient plasma samples. An in-house developed machine learning pipeline was used to identify differentially expressed proteins and model candidate multiplexes to identify those with extremely high diagnostic accuracy (> 0.99). Results: An average of ~2,500 proteins were identified per sample and included in bioinformatics analysis. Comparative analysis between control patient EVs and either pre-cancerous/Stage 0, or Stage 1 colon cancer EVs using an in-house Machine Learning pipeline identified 336 and 493 differentially expressed proteins for each group (FDR adjusted p-value < 0.001). Of these, the best 17 pre/stage 0 proteins and 53 stage 1 proteins were trained and tested using Support Vector Machine to assess all potential 3-plexes in order to identify those with mean diagnostic accuracy > 99%. This yielded 5 3-plexes in precancer/stage 0 and 34 3-plexes in stage 1 with near perfect diagnostic accuracy. These plexes are currently being assessed in single analyte and multiplex immunoassays with the goal of translating candidate 3-plexes to the clinical. Conclusions: Key biomarkers from CRC patient EVs indicate the potential to detect and diagnose early-stage and low tumor burden cancers using our methods. Interestingly, the most accurate 3-plexes consist of proteins from immune, inflammatory and metabolic processes, suggesting that for the detection of early stage cancer it may be critical to include both tumor and host specific biomarkers.