Comprehensive serum profiling for the discovery of epithelial ovarian cancer biomarkers

Abstract

FDA-cleared ovarian cancer biomarkers are limited to CA-125 and HE4 for monitoring and recurrence and OVA1, a multivariate panel consisting of CA-125 and four additional biomarkers, for referring patients to a specialist. Due to relatively poor performance of these tests, more accurate and broadly applicable biomarkers are needed. We evaluated the dysregulation of 259 candidate cancer markers in serum samples from 499 patients. Sera were collected prospectively at 11 monitored sites under a single well-defined protocol. All stages of ovarian cancer and common benign gynecological conditions were represented. To ensure consistency and comparability of biomarker comparisons, all measurements were performed on a single platform, at a single site, using a panel of rigorously calibrated, qualified, high-throughput, multiplexed immunoassays and all analyses were conducted using the same software. Each marker was evaluated independently for its ability to differentiate ovarian cancer from benign conditions. A total of 175 markers were dysregulated in the cancer samples. HE4 (AUC=0.933) and CA-125 (AUC=0.907) were the most informative biomarkers, followed by IL-2 receptor α, α1-antitrypsin, C-reactive protein, YKL-40, cellular fibronectin, CA-72-4 and prostasin (AUC>0.800). To improve the discrimination between cancer and benign conditions, a simple multivariate combination of markers was explored using logistic regression. When combined into a single panel, the nine most informative individual biomarkers yielded an AUC value of 0.950, significantly higher than obtained when combining the markers in the OVA1 panel (AUC 0.912). Additionally, at a threshold sensitivity of 90%, the combination of the top 9 markers gave 88.9% specificity compared to 63.4% specificity for the OVA1 markers. Although a blinded validation study has not yet been performed, these results indicate that alternative biomarker combinations might lead to significant improvements in the detection of ovarian cancer.

Figure 1. Receiver Operating Characteristic curves for the nine most informative biomarkers with area underneath the curve values greater than 0.800.

Figure 2. Serum level distributions broken out by International Federation Of Gynecology And Obstetrics (FIGO) ovarian cancer stage for the nine most informative biomarkers with area underneath the curve values greater than 0.800.

The box boundaries represent the 25th and 75th percentiles, and the bar within the box represents the median value. The minimum and maximum values are represented by the extremes of the whiskers. ***, P-value<0.001; **, P-value 0.001-0.01; *, P-value 0.01–0.05; ns, P-value>0.05. Abbreviations: OvCa, ovarian cancer; Non-OvCa, non-ovarian cancer; CA, cancer antigen; IL, interleukin.

Figure 3. Serum level distributions broken out by subtype of ovarian cancer stage for the nine most informative biomarkers with area underneath the curve values greater than 0.800.

The box boundaries represent the 25th and 75th percentiles, and the bar within the box represents the median value. The minimum and maximum values are represented by the extremes of the whiskers. ***, P-value<0.001; **, P-value 0.001–0.01; *, P-value 0.01-0.05; ns, P-value>0.05. Abbreviations: OvCa, ovarian cancer; Non-OvCa, non-ovarian cancer; CA, cancer antigen; IL, interleukin.

Figure 4. Correlation matrix for biomarkers with area underneath the curve values greater than 0.600.

Hierarchical clustering was implemented with Pearson's correlation coefficients as the distance measure. Intense red and blue colors, signify strong positive and negative correlation, respectively.