Bioinformatic translation of cell line response data to patient selection biomarkers: learnings from taxane data

AACR Annual Meeting-- Apr 14-18, 2007; Los Angeles, CA 4424 Background: Recent work has indicated that genome-wide DNA and RNA data from cell lines combined with in-vitro drug sensitivity may predict clinical response to chemotherapeutic drugs. However, translating genome-wide discovery into clinically applicable tests remains challenging. We extend prior work with taxanes to understand which biological processes and gene-expression parameters are likely to translate from cell line to tumors. Methods: Gene expression signatures for docataxel and paclitaxel were derived from the NCI 60 panel of cancer cell lines and have been shown to be predictive of outcome to neoadjuvant therapy of breast cancer. In parallel, gene expression signatures have been derived from the clinical samples. These signatures involve dozens to hundreds of genes and are amenable to biological interpretation. An Ontology analysis was performed on both the cell line and patient-derived signatures and results were compared to determine common biological processes, and to determine their relationship to the expression of predefined markers such as p53, ER, and HER2 status. In addition, relative RNA expression across histologies was considered to determine potential applicability to a given histology. Results: Several biological processes were found common to both cell line derived and patient-derived predictive profiles, including regulation of cell growth, regulation of transcription, and protein ubiquitination. This information was combined with tissue non-specific expression patterns to define common predictive biomarkers, and with ER and HER2 expression patterns to define biomarkers predictive only in breast cancer. Conclusion: Genome-wide cell line data can be predictive of patients’ response to therapy. The development of clinically applicable biomarkers from such data will require consideration of (1) relevant biological processes, (2) histology specific and non-specific attributes, and (3) the ability to be measured in paraffin-embedded tissue. Our analysis of taxane sensitivity provides a testable hypothesis for the utility of this approach.