Practical biomarkers and robust multiplex models for the prediction of response to promising first-line chemotherapy: A theranostic study in metastatic ovarian cancer patients with residual peritoneal tumors

Abstract

Abstract Background: In advanced or metastatic ovarian cancer patients, the therapeutic impact of molecular targeted agents and immunotherapy is limited, and current chemotherapeutic algorithms are still far from personalized medicine. We recently demonstrated that intraperitoneal carboplatin with dose-dense paclitaxel (ddTCip) therapy is a promising front-line chemotherapy even in patients with residual peritoneal tumors, which led us to this theranostic study for biomarker discovery to realize precision medicine (ID: UMIN000001713 on Feb 16th, 2009). Methods: We first validated previously suggested markers (41 genes and 3 predictive models for therapeutic efficacy and 31 polymorphisms for toxicity), sought out more active effective biomarkers through genome-wide transcriptome and genotyping analyses, and then developed multiplex statistical prediction models for progression-free survival (PFS) and toxicity. Multiple regression analysis following the forward stepwise method and the classification and regression tree (CART) algorithm were mainly employed to develop multiplex prediction models. Results: The association analyses with PFS in 76 patients followed by the validation study using data sets in 189 patients published in The Cancer Genome Atlas revealed that SPINK1 expression could be a possible predictive biomarker of ddTCip efficacy even when used alone, and multiple regression analyses provided a potent efficacy prediction model using expression data of 5 genes. SPINK1 appeared to be a critical resistant determinant of ddTCip therapy, which indicates the potential of SPINK1 as a novel therapeutic target. For toxicity prediction, ABCB1 rs1045642 and ERCC1 rs11615 polymorphisms appeared to be closely associated with grade 2-4 hematologic toxicity and peripheral neuropathy, respectively. We further successfully composed robust multiplex prediction models for adverse events - CART models using a total of 4 genotype combinations and further powerful multiple regression models using 15 polymorphisms on 12 genes-. Conclusions: We newly proposed SPINK1 expression as a powerful predictive biomarker of the efficacy of ddTCip therapy and confirmed the predictive values of ABCB1 and/or ERCC1 polymorphisms for toxicity. The multiplex prediction models composed herein were also found to work well for the prediction of therapeutic response. These findings may raise the potential to realize precision medicine in the essential treatment for metastatic ovarian cancer patients.