Translational biomarkers for cancer drug response: Cell line panel derived drug response signature predicts patients' survival in a lung cancer clinic trial.

Discovery of candidate drug sensitivity predictive genomic models using pre-clinical data would significantly advance the ability to timely validate such models in clinical trials and utilize them for selecting appropriate treatments for patients. Here we report a case study using in vitro cell line viability data to build drug sensitivity models for Erlotinib or Sorafenib. Each individual model was subsequently tested on corresponding treatment arms using patient response data generated in the BATTLE clinical trial. Erlotinib and Sorafenib were selected for the case study as they have clear mechanism of actions, and publicly available molecular data sets coupled with patient outcome to treatment are available (the MD Anderson BATTLE clinic trial). A 240 tumor cell line panel (Oncopanel from Eurofins) was used to identify cell lines that were sensitive or resistant to Erlotinib or Sorafenib treatment, respectively. For each drug, a cell proliferation assay was conducted using 10 doses of Erlotinib or Sorafinib (3-fold dilution) and IC50 values were generated across the panel. For model building on drug sensitivity, baseline gene expression was used as the independent variable and IC50 as the dependent variable. A 6-step Partial Least Squares Regression (PLSR) model workflow was developed using OncoPanel data, which includes multiple steps of data reduction, feature selection, a special splitting strategy to capture consistent features across the dataset, selection of least-overlapping top models, calculation of consensus genes weights followed by selection of the core signature gene set, and ontology enrichment filtering to obtain the pathway-based PLSR model. This specially designed modeling approach was aimed to capture consensus information in the training dataset. When testing the cell line derived predictive models on BATTLE patients’ data, Erlotinib and Sorafenib models achieved overall accuracy of 84% and 79% respectively, and these models are specific to the corresponding drugs. In addition, the model derived signature genes reflect each drug9s known mechanism of action. When using the Erlotinib predictive model to stratify BATTLE patients, the median PFS for the Erlotinib-sensitive patient group was 3.84 month compared to Erlotinib-resistant patients with a PFS of 1.84 month. Similarly, the Sorafenib model identified a marker-sensitive group with a 4.53 month PFS vs. a marker-resistant group with a PFS of 1.87 months. Importantly, the signatures were drug-specific - the Erlotinib predictive model failed to separate marker-sensitive vs. marker-insensitive groups for Sorafenib treatment arm and vise versa. Taking together, the current case study demonstrates that in vitro cell line viability screening studies can be used to derive drug sensitive models predictive of patients’ response to treatment. Citation Information: Mol Cancer Ther 2013;12(11 Suppl):A37. Citation Format: Bin Li, Hyunjin Shin, Georgy Gulbekyan, Olga Pustovalova, Yuri Nikolsky, Andrew Hope, Marina Bessarabova, Matthew Schu, William Trepicchio. Translational biomarkers for cancer drug response: Cell line panel derived drug response signature predicts patients’ survival in a lung cancer clinic trial. [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2013 Oct 19-23; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2013;12(11 Suppl):Abstract nr A37.