Exosome Proteomics and Machine Learning Identify Novel Biomarkers of Primary Graft Dysfunction

Abstract

Purpose The etiology of primary graft dysfunction (PGD) is unclear and prediction of PGD remains an area of active research. Exosomes are secreted microvesicles found in numerous body fluids and have protein and RNA content with biomarker potential. Here we investigate pre-transplant serum exosomes using proteomics to identify novel recipient biomarkers of PGD prior to heart transplant. Methods Pre-transplant serum samples were collected prospectively at Columbia University from June 2014 to December 2015 (n=16). Pre-transplant serum samples were assembled retrospectively from HLA laboratories of Cedars Sinai Medical Center (n=44) and Pitie Salpetriere (n=29). Exosomes were purified with the Total Exosome Isolation Kit (Life Technologies) and proteins were labelled with TMT10plex isobaric reagent. We performed LC-MS/MS with an Orbitrap Fusion Tribrid mass spectrometer. Proteome Discoverer software was used to search the acquired MS/MS data against Uniprot human protein database and generate TMT ratios. Conditional logistic regressions with bootstrapping modeled the association of each protein to PGD status controlling for LVAD status and batch. Logistic Regression (LR), Random Forest (RF), Support Vector Machines (SVM), and Gradient Boosting Classifier (GBC) models, with regularization, and 10-fold cross validated recursive feature elimination (RFE) were used to select the most predictive proteins in the dataset. Results 131 proteins were found in the prospective cohort showing a differential protein expression signature (FDR Conclusion Exosome proteomics of serum samples from both prospective and retrospective cohorts followed by machine learning based analysis identified potential biomarkers for PGD within the recipient prior to transplant.