Abstract
We present a novel computational approach, Prognostic Subgroup Finder (PSFinder), to predict outcome of platinum and taxane combination therapy in high-grade serous ovarian cancer (HGS-OvCa) at treatment-naive stage. PSFinder fuses transcriptomics and clinical data to identify subgroups that have significant survival association (if such exist in the data) stratified by co-expressed markers. Thus, the results from PSFinder are readily interpretable, which is in sharp contrast to the existing methods. PSFinder uses an iterative rule-based approach to search for co-expressed genes that divide the samples into groups with significant association to survival data. The iterative process starts from all the samples in the cohort and proceeds to identify subsets of samples until the subsets cannot be further divided.
PSFinder was applied to HGS-OvCa samples from patients treated with the platinum-taxane therapy available at The Cancer Genome Atlas (TCGA). We identified 61 transcripts (32 genes) that define three subgroups with outcome differences (Kaplan-Meier, log-rank test p = 0.007). The results were validated in eight independent data sets, including a prospectively collected ovarian cancer cohort.
HGS-OvCa patients with dysfunctional DNA repair genes BRCA1 and/or BRCA2 have increased overall survival due to better response to cytotoxic chemotherapy. Multivariate Cox regression analysis using age, grade, stage, residual disease, BRCA1/2 mutation status, and PSFinder identified subgroups shows that the strongest predictors of HGS-OvCa patient outcome are the BRCA1/2 mutation status (p = 0.0003) and the PSFinder identified prognostic types (p = 0.02). Integration of BRCA1/2 mutation data with the PSFinder signature produced markedly improved prediction of HGS-OvCa patients who benefit from platinum and taxane treatment: None of the patients carrying BRCA1/2 mutation and predicted to be good responder by PSFinder signature died during the 5-year follow-up period, whereas 61% of the patients without BRCA1/2 mutation and predicted to be poor responders by PSFinder died within five years after diagnosis. The group of exceptional responders included approximately 8% of the patients in the TCGA cohort (Chen, et al. Cancer Research 2015).
As basal-like breast cancer (BL-BrCa) and HGS-OvCa share several molecular commonalities, etiology and similar therapeutic opportunities, we applied the PSFinder HGS-OvCa predictor (32 genes) to BL-BrCa data from TCGA. Interestingly, the PSFinder signature was able to identify BL-BrCa patients that benefit from platinum-treatment (Kaplan-Meier, log-rank p = 0.017; unpublished). Additionally, we applied PSFinder signatures to other cancers, such as endometrium, melanoma, cervix, colorectal cancer and glioblastoma (unpublished).
In conclusion, this contribution provides a crucially needed method for precision medicine in ovarian cancer. The use of PSFinder predicted prognostic subsets and BRCA1/2 mutation status allows identification of HGS-OvCa patients who truly benefit from platinum and taxane combination therapy and patients who require alternative treatment strategies. Extending the HGS-OvCa derived signature to other cancers further demonstrates the usefulness of the signature as well as the PSFinder approach. Identifying patients having similar molecular landscape and response to therapy regardless of tumor histology facilitates identification of subjects, such as exceptional or refractory responders, to basket trials.
Citation Format: Ping Chen, Kaisa Huhtinen, Katja Kaipio, Piia Mikkonen, Viljami Aittomaki, Rony Lindell, Johanna Hynninen, Annika Auranen, Seija Grenman, Rainer Lehtonen, Olli Carpén, Sampsa Hautaniemi. Novel integrative approach to identify therapy sensitive and insensitive ovarian cancer patients. [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 B47.