Development and validation of a plasma proteomics signature for earlier diagnosis of ovarian cancer using prospectively collected blood samples

Background: There is currently no ovarian cancer biomarker appropriate for screening which is partially due to using retrospective clinical samples obtained at the time of diagnosis for biomarker discovery. Thus, we sought to discover novel plasma proteomic biomarkers for ovarian cancer early detection using prospectively collected blood samples. Method: We evaluated 10,778 plasma proteins measured using the SomaScan v5.0 assay in blood drawn at least three years prior to ovarian cancer diagnosis and matched controls in the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO; n=98, training dataset) and the Nurses' Health Studies (NHS; n=99, replication dataset). We used a conditional logistic regression to identify individual proteins associated with ovarian cancer in the two datasets separately. We also compared plasma proteins for early-stage and late-stage ovarian cancer in blood collected at diagnosis in the PreOperative Pelvic Mass Study to age-matched population-based controls (PreOp; n=134). Then we used Elastic Net to develop a proteomic-based score to discriminate ovarian cancer cases from controls in PLCO, compared to a model with CA125 alone, and validated the proteomic-based score performance in NHS by calculating the area under the receiver operating characteristic curve (AUC) and 95% confidence interval (CI). Results: Plasma proteins associated with ovarian cancer in blood samples collected prospectively differed from those associated with blood samples collected at diagnosis of early-stage disease compared to controls. There were 99 proteins associated with ovarian cancer diagnosed at least 3 years from blood collection (p<0.05) in PLCO, where 2 proteins, RCN3 and OBP2B, replicated in NHS (p<0.05). Of these 99 proteins, majority were not associated with ovarian cancer in PreOp and only three proteins overlapped (i.e., SERPINF2, ASAH2, BAGE3). In PLCO, adding a proteomic-based score comprised of 56 proteins to a model with CA125 alone significantly (p=0.02) improved discriminating ovarian cancer cases from controls with an AUC (95%CI) from 0.65(0.50-0.80) to 0.86(0.67,1.00). In NHS, proteomic-based score resulted in an AUC of 0.60(0.49-0.72) with marginal significance. Conclusion: Our results revealed plasma proteomic profiles differ between prospectively collected blood samples at least 3 years prior to diagnosis and blood samples collected at time of diagnosis regardless of stage. We developed a proteomics-based score that improved upon CA-125 alone, although application to an independent cohort did not demonstrate a strong improvement. However, differences between studies (e.g., menopausal status and hormone therapy use) may explain this variation.