Discussion Cytotoxic chemotherapy prolongs the median survival of metastatic gastric cancer patients from 3�C5 to 9�C11 months compared with best supportive care, with a response rate of 40�C50%.18, 19, 20, 21 Combination CF constitutes the backbone for chemotherapy regimens commonly BET bromodomain inhibitor used for gastric cancers.19, 22 We also reported that CF in combination with low-dose docetaxel is active for metastatic gastric cancer with tolerable toxicity profile.18 The ability to predict the primary resistance of common solid tumors to cytotoxic chemotherapy is currently lacking, but would significantly improve patient care by identifying those who would best be treated by alternative strategies. This study has identified a three-gene predictor that distinguishes gastric cancer patients likely to receive a therapeutic benefit from CF from those who will not.
Most previous studies attempting to identify predictors of chemoresistance in gastric cancer have examined only individual genes such as TS or ERCC1.23, 24 High-throughput DNA microarray analyses to identify gene expression signatures predictive of chemotherapy or chemoradiotherapy resistance in gastrointestinal cancer patients have been limited by the small number of samples,2, 3 heterogeneous treatment4 or were not prospectively designed.5 In contrast to these previous studies, our study uses high-throughput genomic approaches, is prospective with a large, pre-defined number of training set patients, separate validation cohorts and survival data during an extended follow-up period.
Although previously reported TS and ERCC1 tend to be associated with poor prognosis of our patients, the association was not significant enough for them to be considered for our predictive model (P=0.073 and 0.076, for TS and ERCC1, respectively). Notably, the outcome discrimination predicted by the classifier was statistically significant on two validation groups, including the only available published microarray data set from chemotherapy-treated gastric cancer patients.4 Although the sample size of our validation set is relatively small, it is nonetheless large enough to show that our three-gene predictor provides a statistically significant discrimination of patient outcome in multivariate survival analyses. The study design we employed is consistent with an allocation of two-thirds to one-third training-to-test set sample allocation as recommended by statisticians.
25 We combined Batimastat analyses of gene expression changes identified by expression profiling with the identification of DNA copy number changes using array CGH to develop a predictor composed of a much smaller number of critical genes that potentially could be of clinical utility. We identified MYC, EGFR and FGFR2 in regions of amplification, as well as in the gene expression signature related to clinical outcome after CF therapy, suggesting that these genes might be functionally involved in determining resistance.