Abstract

Abstract Introduction: Combination drug treatment in cancer therapy aims to improve response rate and decrease the development of drug resistance. The discovery of novel drug combinations is constrained by the cost and effort of carrying out large unbiased screens and is hampered by poor translation towards the clinic. We investigate if computational models combined with screening can more efficiently reveal synergistic combinations and improve translation towards the clinic. The models are based on three different biological views of synergy. First, compounds that exploit similar vulnerabilities in cancer cells frequently show synergy in a particular genetic context (targeted synergy, model 1) [1,2]. Secondly, drugs with non-overlapping resistance mechanisms often yield clinically relevant drug combinations (model 2) [3]. Finally, drugs that mimic synthetic lethal interactions will often act synergistically (model 3) [4]. Experimental procedures Oncolines™ is a panel of 102 genetically characterized cell lines from diverse tumor origins, on which proliferation assays are run in parallel in nine-point dose-response curves. In this cell panel we profiled 162 different cancer therapeutics, including many standard of care, chemotherapeutic agents, epigenetic modulators, and approved and pre-clinical drugs such as CDK4/6, ALK and PI3K-inhibitors [5]. Cancer hotspot mutations and gene expression data were downloaded from the DepMap and CCLE databases. ANOVA and Pearson correlations were used to analyze the single agent response and genetic data in the statistical software R, to determine gene mutation (model 1) and resistance markers (model 2). For model 3, clinically relevant synthetic lethal pairs were combined with gene expression data [4]. The methods were used to predict results from published synergy screens, e.g. DREAM [6] and results from an independent experiment (NTRC SynergyScreen™) in which fixed concentrations of the poly-ADP ribose polymerase (PARP) inhibitor niraparib and the BET bromodomain inhibitor JQ1, were combined with nine-point dose-response curves of 150 anticancer agents. Synergies were quantified using curve-shift and CI-index. Results: Using a cancer cell line IC50 profile of a compound (such as Oncolines™), all three computational models of synergy prediction can be applied. Method 1 is successful in predicting combinations that augment a targeted effect, such as combining BRAF and MEK inhibitors in BRAF-mutant cancer [1]. Method 2 does not distinguish between additive or synergistic combinations but can identify clinically relevant combinations [3]. For method 3, a tool was developed that computationally assesses if drug pairs can pharmacologically mimic clinically validated synthetic lethal interactions [4]. This can predict DREAM data with a high AUC of 0.67 in a ROC curve. The tool successfully predicts synergy between niraparib and the SUMO inhibitor 2-D08, and niraparib and BCL2 inhibitors, which we observed in our screen. Conclusions: Computational tools for synergy have predictive value and can be useful to prioritize libraries for empirical combination screening.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call