Background: Circulating plasma microparticles (PMPs) are extracellular vesicles released from most types of cells in the human body including tumour cells, during cell differentiation, activation or apoptosis. They transport bioactive molecules (proteins/RNAs) from the parental cell to other parts of body, reflecting the molecular and functional characteristics of their parental cells. For this reason, PMPs have been considered as emerging biomarkers for disease prognosis including cancers. This study aims to identify prognostic protein biomarkers from PMPs for early detection of colorectal cancer (CRC).
Methods: One hundred clinical plasmas (n=20 per CRC AJCC stages I-IV and healthy controls) were employed. Mass spectrometry-based shotgun proteomics was used to generate a CRC specific PMP protein library. SWATH-MS analyses were performed to quantify protein expression levels in all samples. The data were analysed using standard statistics (differential expression by ANOVA, p<0.05; fold-change>1.5) and further examined using two separate machine learning methodologies: one combining randomization and neural networks, and a second based on random forests and logistic regression.
Results: A total of 1,393, high-quality, proteins were identified with 1% FDR level. Of these, Uniprot Gene Ontology analysis showed 735 proteins were PMPs/exosome related proteins. Identified proteins were cross-checked with a top 100 extracellular vesicle protein markers (http://www.exocarta.org/) and 91 proteins were common, validating the methodology employed.
Statistical analysis revealed 9 candidates (e.g., PSB2, RLA2) exhibited differential expression across all CRC stages compared to healthy controls. Furthermore, by comparing protein expression levels between CRC stages I-III patients, whose tumour recurred or not within 5 years post-surgery, 19 protein candidates (e.g., EXTL2, NSF) were identified as differentially expressed. The predictive potential of a subset of these candidates was reinforced by both machine learning approaches above.
Conclusions: Our analysis revealed known prognostic candidates and suggested novel potential prognostic biomarkers. Validation with a larger cohort is required to translate these into clinical practice.