Abstract Epithelial ovarian cancer (EOC) is diagnosed in more than 70% of the cases in advanced FIGO stages which renders EOC the most deadly gynecologic cancer. Up to date there are no specific symptoms and no screening tests for general population or for patients at risk. The development of early diagnostic tools that will improve clinical outcome and differentiate between benign gynecological diseases and EOC are still an unmet need. In a multicenter study financed by the Seventh Framework Program of the European Commission (TRANSCAN) called ImPECT, three European EOC centers (Medical University of Vienna, Charité in Berlin, and the European Institute of Oncology in Milan) validated a “combined blood based gene expression and plasma protein abundance signature for diagnosis of epithelial ovarian cancer” (Pils et al. BMC Cancer, 13:178). In total, blood of 576 women were enrolled in this validation study, including healthy controls, patients with EOC, with breast cancer (to assess tumor specificity), with benign gynecologic diseases (to assess the potential for differential diagnosis), and with inflammatory diseases and diabetes (to assess non-malignant immune changes). A specific blood leukocyte fraction and plasma were isolated and RNA prepared. The expression of 23 genes (13 from the diagnostic signature, seven from an unpublished prognostic signature and three house-keeping) were measured by a multiplexed hybridization based method (Nanostring's nCounter Elements™) and the abundance of six proteins from plasma by a multiplexed bead-based technology (Luminex). Due to evidence for substantial fluctuations of gene expression values by RT-qPCR, especially of the reverse transcription (RT) reaction, we replaced the RT-qPCR method by a hybridization based method without RT. Also, one protein of the protein panel (IGF2) was replaced by another protein (HE4) for two reasons: measurement of IGF2 cannot be multiplexed with the other proteins and HE4 was shown to have a higher discriminative power compared to IGF2. In a first step the published discriminative models were validated with 96 healthy controls and 96 EOC patients. All models showed significant area under the receiver operating characteristic (ROC) curve (AUC) values (p<0.05): Protein models, 0.951 (three proteins) and 0.954 (five proteins), gene expression models, 0.653 (seven genes) and 0.605 (13 genes), and combined models, 0.709 (four proteins/five genes) and 0.773 (five proteins/13 genes). The low AUC values, especially for the combined models, are disillusioning but probably due to the different measurement methods. Therefore new models were built using all 20 gene expression and six protein abundance values using more robust machine learning approaches, namely penalized Support Vector Machines (SVMs). Indeed, much better discriminative models could be built, discriminating healthy controls from EOC patients: Five proteins, AUC 0.972 (CI95:0.952-0.993); 16 genes, AUC 0.848 (CI95:0.794-0.903), and four proteins/five genes (combined signature), AUC 0.981 (CI95:0.963-0.998). The last, combined, signature is significantly better compared to the first, only protein, signature (p=0.029), indicating an additional discriminative value for the gene expressions. To assess the technical reproducibility of the Nanostring method, each 84 samples were measured a second time and correlated to the first measurement: Interestingly, the median Pearson's r correlations over all 20 genes was only 0.689 (0.337-0.799) but the SVM model between controls and EOC patients was more robust: AUC of 0.811 compared to 0.848 (p=0.394). A clinical validation of these models with independent patient and control cohorts (healthy controls and patients with benign diseases) and the specificity compared to breast cancer and diseases with immune system involvement will be presented. Citation Format: Anna Bachmayr-Heyda, Stefanie Aust, Katharina Auer, Peter Borossay, Nina Pecha, Ugo Cavallaro, Ioana Braicu, Sarah Igidbashian, Costanza Savino, Franchi Dorella, Jalid Sehouli, Sonia Zaafrani, Florian Fitzal, Christian Singer, Michael Gnant, Robert Zeillinger, Dietmar Pils. Validation of a combined 13 gene six protein blood signature for earlier detection of ovarian cancer. [abstract]. In: Proceedings of the AACR Special Conference on Advances in Ovarian Cancer Research: Exploiting Vulnerabilities; Oct 17-20, 2015; Orlando, FL. Philadelphia (PA): AACR; Clin Cancer Res 2016;22(2 Suppl):Abstract nr B28.
Read full abstract