Abstract A20: A systematic patient-derived xenograft based solution for pre-clinical biomarker discovery

Abstract

There is an acute need for biomarkers at every phases of drug development from selecting preclinical models in pharmacogenomic studies to enrollment of patients in clinical trials. However, their identification remains extremely challenging due to the limited availability of clinical samples. In contrast, standard tumor models such as cell lines are available but are genetically relatively far from patient tumors. Use patient-derived xenografts (PDX) for anticancer agent-testing is of increasing interest due to their closer similarity to patient tumors compared to cell lines. Over the last 30 years, we have established a collection of 400 PDX covering more than 30 different cancer types. PDX models have been extensively characterized using the microarray or next-generation sequencing technologies for gene expression, copy number variations and whole-exome mutations. Biomarker research is now possible using these data in combination with drug response data from in vivo or in vitro 2D or 3D assays routinely performed on-site with large panels of 100-200 PDX. We present here a fully integrated bioinformatics pipeline dedicated to biomarker discovery in which the complete molecular profiles of our PDX have been systematically tested for association with drug sensitivity. To identify the biomarkers associated with drug response, several statistical tests have been performed. Drug response data were treated either as continuous variables using the Spearman or Wilcoxon tests, or as categorical variables (with two groups of responders and non-responders) using the LIMMA, t-test or Fisher exact test. Given that high throughput data frequently leads to large biomarker lists, we used specific filters to narrow down the list of candidates by defining thresholds based on corrected p-values, by intersecting results from different tests, or by integrating the tumor type into the statistical tests. Since sensitivity to anticancer agents is often multi-factorial, we also used integrative approaches that combined gene mutations, copy number loss and lack of gene expression for association with drug response. Finally, significant biomarkers were visualized using clustering heatmaps and enrichment GO/pathway approaches to get more insight in their biological function. Using a selection of several datasets of PDX drug responses to chemotherapeutics and targeted therapies (targeting RTK/RAS/RAF and PI3K/MTOR pathways and using specific compounds such as Vemurafenib, Erlotinib or Cetuximab), we demonstrate the efficacy of our approach to retrieve biomarkers of known clinical utility. Using these datasets we also could address the questions of model panel sizes, molecular data type and tumor subtype representation, and show how more accurate biomarkers can be validated using an independent dataset of samples. The development of strategies for testing anticancer agents using PDX in mouse clinical trials, or high throughput in vitro 2D, 3D screening approaches coupled to a more systematic biomarker research should significantly contribute to early biomarker identification and facilitate drug development. Citation Format: Bruno Zeitouni, Anne-Lise Peille, Zakia Amalou, Thomas Metz, Heinz-Herbert Fiebig, Vincent Vuaroqueaux. A systematic patient-derived xenograft based solution for pre-clinical biomarker discovery. [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2015 Nov 5-9; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2015;14(12 Suppl 2):Abstract nr A20.