Abstract PL05-04: Gene signatures for targeted and cytotoxic agents can guide new anticancer drug development

Abstract

The identification of biomarkers or gene signatures predicting tumor response to novel targeted or cytotoxic therapy has the potential to individualize and improve the effectiveness of anticancer therapy. The FDA strongly recommends developing such companion diagnostic tests. Clinical relevance of predictive markers has, for example, been demonstrated for EGFR inhibitors like cetuximab or pamituzumab, which are primarily effective in tumors with wild type k‐ras status. Sets of genes ‐ the so‐called gene signatures ‐ have probably even a greater potential for prediction of tumor response than a single marker gene. Our hypothesis was that the correlation of drug response with gene expression would allow the identification of gene signatures that can predict the drug response of individual tumors. The expression profiles of thousands of genes are easily accessible with gene expression arrays, provided fresh or frozen tumor samples are available. However, in the clinic it is difficult to discriminate between groups of responding and resistant tumors as would be required for the derivation of gene signatures. Therefore, we used patient‐derived tumor models established subcutaneously in nude mice to determine the in vivo sensitivity towards registered and developmental compounds. Patient tumors established subcutaneously in serial passage in nude mice were characterized for their sensitivity towards two targeted and 10 standard cytotoxic anticancer agents. The latter included the alkylating agents cyclophosphamide, ifosfamide, mitomycin‐C and cisplatin; the antimetabolites 5‐FU and gemcitabine; the topoisomerase II inhibitors adriamycin and etoposide as well as the tubulin binders paclitaxel and docetaxel. The mean number of tumors treated with any of the various drugs was 54 (range 31–78). The tumor xenografts9 gene expression profiles were determined using the Affymetrix HG‐U133 plus 2.0 mRNA expression array representing ∼38,500 human genes. For all 12 tested agents, predictive gene signatures were identified and subsequently verified using the leave‐one‐out cross‐validation (LOOCV) technique. For three of the drugs, signatures were further validated using an independent set of previously untested tumors. Tumors were considered as responsive if the drugs effected a tumor volume inhibition to less than 11–41% of vehicle control tumors (T/C %). The median cut‐off for all drugs was a T/C of 25%. Using these criteria, on average one third of the test tumors were classified as sensitive (responders) and two thirds were resistant (non‐responders). The bioinformatic analysis yielded predictive gene signatures consisting of 20–129 genes (mean for the 12 drugs: 87 genes). On average, the response rate of predicted responders (83%) was 2.45 fold higher than that of all test tumors (random testing, 34%). This increase of response rates, following signature‐guided testing, was consistent for all agents. Conversely, 94% of the predicted non‐responders (range: 84–100%) proved to be non‐responders in the nude mouse. The function of the majority of genes (59%) making up the predictive gene signatures was unknown. Genes with known function are implicated in cell proliferation, apoptosis, DNA repair, cell cycle, metabolism and transcription. In summary, the predictive gene signatures presented here for 12 cytotoxic agents have the potential to substantially increase tumor response rates compared to empirical drug treatment. Similarly, the VEGF antibody Bevacizumab (BV) was also evaluated in 72 tumors (colon, NSC‐lung, breast and renal cancers). BV was administered iv once weekly for 3 weeks. Antitumor activity was evaluated as the minimum T/C value or after 28 days. Using a T/C of We have also developed a gene signature for Cetuximab (CTX) consisting of 21 genes. Using a T/C of For both marketed drugs and drugs in development, gene signatures have great potential to increase the response rates compared to empirical drug treatment. However, clinical application of the gene signatures needs validation in a clinical setting. In a first clinical feasibility study with 72 patients with different solid tumors that we have initiated, the tumor response to BV and CTX was predicted to be 26 tumors (36%) for BV and 13 (18%) for CTX, the latter mainly colon and gastric cancers. Gene signatures have a great potential to increase the response rates compared to empirical drug treatment for marketed drugs and also for drugs in development. However, the gene signatures need to be further validated in clinical studies. Citation Information: Mol Cancer Ther 2009;8(12 Suppl):PL05-04.