13, 14, 15 Overexpression of ERCC2 (P=0.007 in our data) is associated with cisplatin resistance in lung cancer cell lines.13 Silencing of hHR23A (P=0.022 in our data) decreases the nuclear DRP1 level and cisplatin resistance in lung adenocarcinoma cells.14 Disruption of the Fanconi anemia�CBRCA pathway promotion information is reported in cisplatin-sensitive ovarian tumors.15 Thus, this gene ontology analysis supports the clinical relevance of these DNA repair canonical pathways, which were shown to be associated with in vitro cisplatin resistance. Ingenuity Pathway Analysis functional categories enriched in poor prognosis signature were: protein synthesis, DNA replication/recombination/repair and cancer (Supplementary Table 2).

The protein synthesis category includes ribosomal subunit mRNAs (RPL13, RPL18, RPL24, RPL30, RPL38, RPL5, RPL7, RPL7A, RPL8, RPS2, RPS5) and eukaryotic translation initiation factors (EIF1, EIF2B2, EIF2B4, EIF2S1, EIF3B, EIF3C, EIF3D, EIF3E, EIF3F, EIF3H, EIF3I, EIF4A1, EIF4A3, EIF4B, EIF4EBP1, EIF5, EIF5B). This result suggests that the most prominent feature of poor prognosis signature is increased protein synthesis, presumably resulting from activation of oncogenes, such as EGFR, FGFR2 and MYC (Supplementary Table 2). MYC-induced transcriptional activation of protein synthesis-related genes is previously shown by a microarray report that the majority of genes responsive to MYC overexpression are involved in macromolecular synthesis, protein turnover and metabolism, including 30 ribosomal protein genes.

16 Infinitesimal perturbation analysis canonical pathways enriched in 648 genes in good prognosis signature were antigen presentation pathway, B-cell development and interleukin-15 production. Enriched functional categories were gastrointestinal disease, inflammatory disease and genetic disorder. Development of the three-gene predictor Although such a gene ontology analysis of the whole signature provides some insight into clinically relevant mechanisms for chemotherapy resistance, this large number of genes is not readily amenable to clinical application. Therefore, we wished to narrow down 917 genes in the whole poor prognosis signature to the smaller number of genes, which may have driven the expression of majority of genes in the signature. Focusing on such ��driver gene’ candidates would also minimize the chance of including false-positive discovery in a genomic predictor.

For this purpose, a second tier of genomic analysis was performed to identify genes that could be functionally important in gastric cancer cells. Genomic DNA from samples available from the training set patients was analyzed by array CGH to identify gene amplifications. Age, sex and overall survival were similar between the 30 patients (31.3%) whose samples Cilengitide were analyzed by array CGH and the other patients in the training set. Using very conservative criteria (average tumor/normal log2 ratio >2.