Clinical analysis of EV-Fingerprint to predict grade group 3 and above prostate cancer and avoid prostate biopsy.

BACKGROUND: There is an unmet clinical need for minimally invasive diagnostic tests to improve the detection of grade group (GG) ≥3 prostate cancer relative to prostate antigen-specific risk calculators. We determined the accuracy of the blood-based extracellular vesicle (EV) biomarker assay (EV Fingerprint test) at the point of a prostate biopsy decision to predict GG ≥3 from GG ≤2 and avoid unnecessary biopsies. METHODS: This study analyzed 415 men referred to urology clinics and scheduled for a prostate biopsy, were recruited to the APCaRI 01 prospective cohort study. The EV machine learning analysis platform was used to generate predictive EV models from microflow data. Logistic regression was then used to analyze the combined EV models and patient clinical data and generate the patients' risk score for GG ≥3 prostate cancer. RESULTS: The EV-Fingerprint test was evaluated using the area under the curve (AUC) in discrimination of GG ≥3 from GG ≤2 and benign disease on initial biopsy. EV-Fingerprint identified GG ≥3 cancer patients with high accuracy (0.81 AUC) at 95% sensitivity and 97% negative predictive value. Using a 7.85% probability cutoff, 95% of men with GG ≥3 would have been recommended a biopsy while avoiding 144 unnecessary biopsies (35%) and missing four GG ≥3 cancers (5%). Conversely, a 5% cutoff would have avoided 31 unnecessary biopsies (7%), missing no GG ≥3 cancers (0%). CONCLUSIONS: EV-Fingerprint accurately predicted GG ≥3 prostate cancer and would have significantly reduced unnecessary prostate biopsies.

Building predictive disease models using extracellular vesicle microscale flow cytometry and machine learning.

Extracellular vesicles (EVs) are highly abundant in human biofluids, containing a repertoire of macromolecules and biomarkers representative of the tissue of origin. EVs released by tumours can communicate key signals both locally and to distant sites to promote growth and survival or impact invasive and metastatic progression. Microscale flow cytometry of circulating EVs is an emerging technology that is a promising alternative to biopsy for disease diagnosis. However, biofluid-derived EVs are highly heterogeneous in size and composition, making their analysis complex. To address this, we developed a machine learning approach combined with EV microscale cytometry using tissue- and disease-specific biomarkers to generate predictive models. We demonstrate the utility of this novel extracellular vesicle machine learning analysis platform (EVMAP) to predict disease from patient samples by developing a blood test to identify high-grade prostate cancer and validate its performance in a prospective 215 patient cohort. Models generated using the EVMAP approach significantly improved the prediction of high-risk prostate cancer, highlighting the clinical utility of this diagnostic platform for improved cancer prediction from a blood test.

Quantitative in vivo whole genome motility screen reveals novel therapeutic targets to block cancer metastasis.

Metastasis is the most lethal aspect of cancer, yet current therapeutic strategies do not target its key rate-limiting steps. We have previously shown that the entry of cancer cells into the blood stream, or intravasation, is highly dependent upon in vivo cancer cell motility, making it an attractive therapeutic target. To systemically identify genes required for tumor cell motility in an in vivo tumor microenvironment, we established a novel quantitative in vivo screening platform based on intravital imaging of human cancer metastasis in ex ovo avian embryos. Utilizing this platform to screen a genome-wide shRNA library, we identified a panel of novel genes whose function is required for productive cancer cell motility in vivo, and whose expression is closely associated with metastatic risk in human cancers. The RNAi-mediated inhibition of these gene targets resulted in a nearly total (>99.5%) block of spontaneous cancer metastasis in vivo.