Identification of biomarkers for ovarian cancer using strong anion-exchange ProteinChips: potential use in diagnosis and prognosis

Abstract

One hundred eighty-four serum samples from patients with ovarian cancer (n = 109), patients with benign tumors (n = 19), and healthy donors (n = 56) were analyzed on strong anion-exchange surfaces using surface-enhanced laser desorption/ionization time-of-flight mass spectrometry technology. Univariate and multivariate statistical analyses applied to protein-profiling data obtained from 140 training serum samples identified three biomarker protein panels. The first panel of five candidate protein biomarkers, termed the screening biomarker panel, effectively diagnosed benign and malignant ovarian neoplasia [95.7% sensitivity, 82.6% specificity, 89.2% accuracy, and receiver operating characteristic (ROC) area under the curve of 0.94]. The other two panels, consisting of five and four candidate protein biomarkers each, effectively distinguished between benign and malignant ovarian neoplasia and were therefore referred to as validation biomarker panel I (81.5% sensitivity, 94.9% specificity, 88.2% accuracy, and ROC = 0.94) and validation biomarker panel II (72.8% sensitivity, 94.9% specificity, 83.9% accuracy, and ROC = 0.90). The three ovarian cancer biomarker protein panels correctly diagnosed 41 of the 44 blinded test samples: 21 of 22 malignant ovarian neoplasias [10 of 11 early-stage ovarian cancer (I/II) and 11 of 11 advanced-stage ovarian cancer (III/IV)], 6 of 6 low malignant potential, 5 of the 6 benign tumors, and 9 of 10 normal patient samples. In conclusion, we have discovered three ovarian cancer biomarker protein panels that, when used together, effectively distinguished serum samples from healthy controls and patients with either benign or malignant ovarian neoplasia.

Fig. 1.

Differentially expressed ovarian cancer-associated serum proteins. (A) Detection of differentially expressed ovarian cancer-associated serum proteins. A representative pseudogel view of SELDI-TOF-MS analysis of serum samples, processed on a SAX2 chip surface, shows relative abundance of potential ovarian cancer markers. The six spectral protein profiles at the top represent serum from healthy individuals, three benign samples are represented in the middle section, and the six spectral profiles on the bottom represent serum from patients with ovarian cancer. (B) Representative spectral overlay of serum from a healthy (solid line) vs. diseased (dashed line) individual. The overlay shows that there is a decrease of protein 13.9kDa in serum from ovarian cancer patients, whereas concurrently there is an increase of the proteins 15.1kDa and 15.9kDa. Numbers in the mass spectra represent the observed mass of the marker in that particular sample.

Fig. 2.

ROC curves and plot of sensitivity and specificity for the ovarian cancer screening panel. ROC curve analysis was based on 140 patients to compare the diagnostic performance of five neoplasia biomarkers making up the screening panel (4.4kDa, 15.9kDa, 18.9kDa, 23.0kDa, and 30.1kDa), which were identified by the sas multivariate analysis program. The area under the ROC curve is 0.94 (A), and the weighted sum of the marker intensities (score/index) is plotted as a function of sensitivity and specificity (B).

Fig. 3.

ROC curves and plot of sensitivity and specificity for the ovarian cancer validation panels I and II. Five malignant neoplasia biomarkers making up the validation panel I (3.1kDa, 13.9kDa, 21.0kDa, 79.0kDa, and 106.7kDa) and four markers representing the validation panel II (5.1kDa, 16.9kDa, 28.0kDa, and 93.0kDa) were identified by the sas multivariate analysis program. The areas under the ROC curves are 0.94 (A) and 0.90 (C). The weighted sum of the marker intensities (score/index) is plotted as a function of sensitivity and specificity (B and D).