Gene profiling of lung cancers with EGFR mutations

Abstract

Response to gefitinib/Iressa has recently been found to be associated with somatic mutation of EGFR in lung cancers. This important discovery should lead to a better selection of patients who will most likely benefit from gefitinib treatment. One way of identifying these patients is to use a gene profiling approach, whereby testing the expression levels of a small set of genes can predict whether a patient will harbor an EGFR mutation and thus would potentially respond well to the drug. To this aim, 17 primary non-small cell lung cancers from the University of Michigan, of which half (8) contained an EGFR mutation, were analyzed with microarrays (Affymetrix GeneChips HG-U133Plus2 which contains more than 54,000 elements). Using these as a training set, a leave-one-out supervised classification approach identified 100 genes whose expression levels formed a good class predictor. These genes were then applied to a test set of 36 primary non-small cell lung cancers originated from Hong Kong and analyzed with Affymetrix GeneChips HG-U133A and HG-U133B (∼45,000 elements altogether) and whose EGFR mutation status had previously been determined (15 wild-type and 21 mutated). Expression levels of the 100 class predictor genes in those 36 tumors could correctly classify them with a 75% accuracy (P = 0.0006) as either EGFR mutated or wild-type. Specifically, 17 of the 21 EGFR mutants were correctly predicted (81%, P = 0.0007) as well as 10 of the 15 EGFR wild-types (67%, P = 0.06). This shows that a small set of genes can be used to screen patients likely to be responsive to gefitinib and serves as an alternative diagnostic tool to direct EGFR sequencing. Further, this prediction method worked even though the training and test sets were from different populations and were analyzed with different GeneChip platforms. Additionally, we analyzed 44 lung cancer cell lines from Dallas with Affymetrix HG-U133A and HG-U133B, of which 8 were found to be mutated in EGFR. These and the previous two sets of lung cancers were screened for differentially regulated genes in EGFR mutant versus wild-type tumor samples. Using a cutoff of 2-fold difference and filtering for named genes with a T-test p-value of less than 0.05, we identified 93, 220, and 537 genes upregulated in the EGFR-mutated subgroups from the Michigan, Hong Kong, and Dallas sets respectively, and 80, 211, and 657 genes downregulated in the same subgroups. Forty-two genes were found upregulated in at least two tumor sets (including EGFR which was upregulated in all three sets), and 55 genes were downregulated in two or more tumor sets. Study of these genes should prove valuable both for EGFR signaling research and the discovery of new drug targets. An independent analysis of the Hong Kong tumor set with additional cases is presented in the abstract by CL Lam et al. (Supported by Lung Cancer SPORE P50CA70907 & Gilson Longenbaugh Foundation)