Whole blood-derived miRNA profiles as potential new tools for ovarian cancer screening

Abstract

Background: Screening is an unsolved problem for ovarian cancer (OvCA). As late detection is equivalent to poor prognosis, we analysed whether OvCA patients show diagnostically meaningful microRNA (miRNA) patterns in blood cells.

Methods: Blood-borne whole miRNome profiles from 24 patients with OvCA and 15 age- and sex-matched healthy controls were biostatistically evaluated.

Results: Student's t-test revealed 147 significantly deregulated miRNAs before and 4 after Benjamini-Hochberg adjustment. Although these included miRNAs already linked to OvCA (e.g., miR-16, miR-155), others had never before been connected to specific diseases. A bioinformatically calculated miRNA profile allowed for discrimination between blood samples of OvCA patients and healthy controls with an accuracy of >76%. When only cancers of the serous subtype were considered and compared with an extended control group (n=39), accuracy, specificity and sensitivity all increased to >85%.

Conclusion: Our proof-of-principle study strengthens the hypothesis that neoplastic diseases generate characteristic miRNA fingerprints in blood cells. Still, the obtained OvCA-associated miRNA pattern is not yet sensitive and specific enough to permit the monitoring of disease progression or even preventive screening. Microarray-based miRNA profiling from peripheral blood could thus be combined with other markers to improve the notoriously difficult but important screening for OvCA.

Figure 1

Deregulation of miR-30c-1^* and miR-181a^* in OvCA samples compared with healthy controls. (A) Shown are the intensities of expression for miR-30c-1^* (upper panel) and miR-181a^* (middle panel) in blood samples from healthy donors (dark grey, n=15) or OvCA patients (light grey, n=24). P values (OvCA vs control) were 0.01 for miR-30c-1^* and 0.04 for miR-181a^*, as calculated by Student's unpaired two-tailed parametric t-test followed by the Benjamini–Hochberg adjustment for multiple comparisons. (B) Receiver operating characteristics (ROC) were generated to show how the sensitivity of OvCA detection and the rate of false positives vary with the discrimination threshold for single miRNAs. Shown is the ROC curve for miR-342-3p. AUC denotes the area under the curve, which is equal to the probability that a classification based on miRNA-342-3p will rank a randomly chosen positive sample higher than a randomly chosen negative one.

Figure 2

Classification of samples from OvCA patients or healthy controls. (A) This classification plot that is based on 60 miRNAs was computed using a radial basis function SVM as described in (Keller et al, 2006, 2009a). The black boxes showing the accuracy (‘acc’), specificity (‘spec’) and sensitivity (‘sens’) for classification of all OvCA and control samples (n=24 for OvCA, n=15 for controls) and were calculated through 100 repetitions of 10-fold cross-validation. The grey boxes show the results obtained when the same mathematical operation is performed in permutation tests (‘random’) in which the class labels (OvCA vs control) have been assigned randomly before the values are computed. This is used to validate the classification procedure. The ordinate shows the proportion of samples that were classified correctly to their group. (B) Serous OvCA (n=20) were compared with an extended group of healthy controls (n=39) as in (A), using 100 repetitions and 40 miRNAs.