Abstract
Histological assessment of prostate cancer is the key diagnostic test and can predict disease outcome. This is however an invasive procedure that carries associated risks, hence non-invasive assays to support the diagnostic pathway are much needed. A key feature of disease progression, and subsequent poor prognosis, is the presence of an altered stroma. Here we explored the utility of prostate stromal cell-derived vesicles as indicators of an altered tumour environment. We compared vesicles from six donor-matched pairs of adjacent-normal versus disease-associated primary stromal cultures. We identified 19 differentially expressed transcripts that discriminate disease from normal stromal extracellular vesicles (EVs). EVs isolated from patient serum were investigated for these putative disease-discriminating mRNA. A set of transcripts including Caveolin-1 (CAV1), TMP2, THBS1, and CTGF were found to be successful in discriminating clinically insignificant (Gleason = 6) disease from clinically significant (Gleason > 8) prostate cancer. Furthermore, correlation between transcript expression and progression-free survival suggests that levels of these mRNA may predict disease outcome. Informed by a machine learning approach, combining measures of the five most informative EV-associated mRNAs with PSA was shown to significantly improve assay sensitivity and specificity. An in-silico model was produced, showcasing the superiority of this multi-modal liquid biopsy compared to needle biopsy for predicting disease progression. This proof of concept highlights the utility of serum EV analytics as a companion diagnostic test with prognostic utility, which may obviate the need for biopsy.