Significant improvements in the outcome of non–small cell lung carcinoma (NSCLC) have been reported in patients treated with the epidermal growth factor receptor (EGFR) inhibitor, erlotinib. To discover biomarkers for the enrichment of patients who might benefit from treatment, a pharmacogenomic approach was used to identify gene signatures that may predict erlotinib activity using in vitro model systems. Erlotinib sensitivity in a panel of 42 NSCLC cell lines was determined by EGFR-mediated proliferative potential, EGFR mutations, and/or EGFR gene amplification, thus supporting an underlying biological mechanism of receptor activation. A strong multigene signature indicative of an epithelial to mesenchymal transition (EMT) was identified as a determinant of insensitivity to erlotinib through both supervised and unsupervised gene expression approaches. This observation was further supported by expression analysis of classic EMT marker proteins, including E-cadherin and vimentin. To investigate the clinical relevance of these findings, we examined expression of the epithelial marker E-cadherin by immunohistochemistry on primary tumor samples from subjects enrolled in a randomized NSCLC clinical trial in which erlotinib in combination with chemotherapy previously failed to show clinical activity. The majority (75%) of the 87 subjects tested showed strong E-cadherin staining and exhibited a significantly longer time to progression (hazard ratio, 0.37; log rank P = 0.0028) and a nonsignificant trend toward longer survival with erlotinib plus chemotherapy treatment versus chemotherapy alone. These data support a potential role for EMT as a determinant of EGFR activity in NSCLC tumor cells and E-cadherin expression as a novel biomarker predicting clinical activity of the EGFR inhibitor erlotinib in NSCLC patients.

Activation of the epidermal growth factor receptor (EGFR) via mutation, amplification, and/or overexpression of EGFR ligands has been implicated in a wide variety of epithelial cancers (1, 2). In non–small cell lung carcinoma (NSCLC), EGFR is expressed in the majority of tumor cells and selective blockade of the EGFR via small molecule tyrosine kinase inhibitors has shown clear activity in preclinical NSCLC models (3). Erlotinib and gefitinib, small molecule inhibitors of EGFR, have been extensively studied in NSCLC patients, either alone or in combination with chemotherapy. Only treatment with erlotinib (Tarceva, a quinazoline derivative EGFR tyrosine kinase inhibitor) in second/third line as a monotherapy has been shown to improve the survival of patients with advanced NSCLC (4). Although there have been multiple attempts to prove the clinical benefit of EGFR small molecule tyrosine kinase inhibitors in the first-line setting in combination with chemotherapy, none of these have been successful (5). One potential explanation could be that no patient selection strategies have been used in these trials to identify tumors with an activated EGFR pathway, which may have a better chance of responding to treatment. This may be particularly relevant when three drug combinations are being evaluated because the incremental benefit is likely to be small and, therefore, difficult to show without very large clinical trials. Thus, the discovery of molecular biomarkers that identify the subsets of NSCLC patients who are most likely to benefit from therapy targeted to EGFR has become an important area of investigation.

The status of the EGFR gene and/or protein in comparison with clinical activity of EGFR tyrosine kinase inhibitors has been a focus of recent research. Several publications have reported EGFR mutations as predictors of response (tumor shrinkage) after treatment with gefitinib or erlotinib (68). The prevalence of these mutations in Caucasians, however, is low (<15% in Caucasians; ∼30-40% in Asians) and is unlikely to explain the majority of the survival benefit patients receive from treatment with erlotinib (4, 9). In a recent randomized, placebo-controlled phase III study evaluating erlotinib in NSCLC patients following failure of first-line or second-line chemotherapy, tumor-specific increase in copy number of the EGFR gene was shown to correlate with increased survival in NSCLC patients comparing erlotinib treatment with placebo (10). In the same study, a weaker correlation of EGFR protein expression with survival benefit was determined by immunohistochemical staining. Taken together, it is reasonable to assume that there are at least three mechanisms for EGFR activation supported by correlative clinical data: mutations in the EGFR kinase domain, increase in copy number of the EGFR gene, and overexpression of the EGFR protein. In addition, other EGFR activating mechanisms, such as overexpression of EGFR ligands (11) or heterodimerization with overexpressed coreceptors, cannot be excluded (12, 13), but have yet to be correlated with clinical benefit from EGFR inhibition.

Here, we identify a biological signature that would allow for the detection of erlotinib-sensitive NSCLC tumor cells with an activated EGFR pathway, independent from the mechanism leading to activation. We find that a gene signature indicative of an epithelial versus mesenchymal phenotype is not only predictive of erlotinib-mediated growth inhibition in NSCLC tumor cell lines but may also serve as a potential biomarker in predicting erlotinib clinical activity in NSCLC patients.

Cell culture experiments. Cell lines were obtained from either the American Type Culture Collection (Manassas, VA) or the National Cancer Institute Division of Cancer Treatment and Diagnosis tumor repository (Frederick, MD). All cell lines were maintained in RPMI 1640 supplemented with 10% fetal bovine serum (Sigma, St. Louis, MO) and 2 mmol/L l-glutamine. For erlotinib IC50 determination, cells were plated in triplicate at 5 × 103 per well in 96-well plates in RPMI containing 0.5% fetal bovine serum (assay medium) overnight before stimulation with assay medium containing 3 nmol/L transforming growth factor-α (TGF-α). Erlotinib (Genentech, Inc., South San Francisco, CA) was added at multiple concentrations (10 μmol/L maximal) and 72 hours later, cell viability was measured using the Celltiter-Glo Luminescent Cell Viability Assay (Promega, Madison, WI). The concentration of erlotinib resulting in the 50% inhibition of cell viability was calculated from a four-variable curve analysis and was determined from a minimum of three experiments. For assessing ligand-driven growth potential, cells were cultured for 72 hours in assay medium containing various concentrations of TGF-α (0.1-30 nmol/L) before determining cell viability as above. Stimulation indices were determined by dividing the average cell viability units of TGF-α-stimulated cultures by the average cell viability units of unstimulated cultures and are reported as the maximal stimulation index, independent of the concentration of TGF-α. For caspase activity determination, cells (1.2 × 104) were stimulated in assay medium containing 3 nmol/L TGF-α in the presence of a dose response of the EGFR inhibitors erlotinib or C225 (Calbiochem, La Jolla, CA). Caspase-3/7 activity was measured using the Apo-ONE Homogeneous Caspase-3/7 Assay (Promega) at either 24 hours (erlotinib) or 48 hours (C225) following treatment. Results are presented as a percentage of caspase activity relative to control (no drug treatment) cultures.

EGFR and KRAS gene status. For assessing sequence variations, PCR amplifications of genomic DNA isolated from NSCLC cell lines were carried out using nested primers to amplify EGFR exons 18 to 21 and KRAS exon 2, before the direct sequencing of products, as previously described (8). Genomic amplifications of the EGFR gene were determined by fluorescence in situ hybridization. NSCLC cell line tissue microarrays were constructed from formalin-fixed, paraffin-embedded cell pellets as previously described (14). Fluorescence in situ hybridization analysis was carried out on 3 μm sections using the LSI EGFR Spectrum Orange/CEP 7 Spectrum Green probe set (Vysis, Downers Grove, IL), as instructed in the protocol of the manufacturer. Fluorescence in situ hybridization results were manually scored in a blinded fashion and amplification was defined as an EGFR/CEP7 ratio >2.0, as described (15).

Quantitative expression analyses of HER receptors/ligands. HER family member receptor and ligand transcript expression levels were assessed by quantitative reverse transcription-PCR using standard Taqman techniques. Transcript levels were normalized to the housekeeping gene β-glucuronidase and results are expressed as normalized expression values(2−ΔCt). The primer/probe sets for β-glucuronidase are forward, 5′-TGGTTGGAGAGCTCATTTGGA-3′; reverse, 5′-GCACTCTCGTCGGTGACTGTT-3′; and probe, 5′(VIC)-TTTGCCGATTTCATGACT-(MGBNFQ)-3′. Primer/probe sets for the HER receptor family members and EGFR ligands were purchased from Applied Biosystems (Foster City, CA). The determination of EGFR cell surface expression levels was carried out by flow cytometry. Cells (1 × 106) were directly stained with the EGFR-FITC conjugated antibody (P54416F, Biodesign, Saco, ME) or a murine isotype-FITC conjugated control (USB), washed, resuspended in 1 mL buffer containing 1 μg/mL propidium iodide, and live cells were subsequently analyzed for mean shift in fluorescence on a FACSCalibur (Becton Dickinson, Franklin Lakes, NJ). Heat maps to visualize quantitative mRNA expression and mean shift in fluorescence values were generated using the software Cluster (16). Student's t test was used to calculate statistically significant differences in gene/protein expression between erlotinib “sensitive” and “insensitive” groups.

Microarray studies. Basal gene expression analysis of NSCLC cell lines was carried out using RNA extracted from subconfluent cell cultures on the Affymetrix (Santa Clara, CA) microarray platform (HGU133Plus_2.0 chips). Preparation of complementary RNA, array hybridizations, and subsequent data analysis were carried out using the protocol of the manufacturer as described (17).

For unsupervised hierarchical clustering analysis, gene expression values were subjected to a variation filter to exclude genes with minimal variation across the samples being analyzed. This was done by excluding the maximum and minimum value for each probe set and then testing for a fold change (max/min > 10) and absolute variation (max-min > 500) over samples, and excluding genes not obeying both conditions. In addition, a single probe was selected to represent a gene, although probe sets with no Unigene identifier were assumed to represent different genes. Data preprocessing involved log transforming and median centering gene expression values and average linkage clustering was carried out using Cluster and TreeView software (16).

For the supervised identification of marker genes exhibiting differential expression between sensitive and insensitive cell lines, cell lines were classified as sensitive or insensitive based on a 0.85 SD window around the median erlotinib IC50: cell lines exhibiting IC50 < 2.0 μmol/L were classified as sensitive and IC50 >8.0 μmol/L as insensitive. Cell lines with intermediate erlotinib sensitivities (2-8 μmol/L IC50) were excluded from the analysis. Gene expression values were subjected to a variation filter to exclude probe sets with minimal variation across the samples, and a signal to noise metric was subsequently used to identify marker probe sets using GenePattern software (http://www.broad.mit.edu/cancer/software/genepattern/).

To examine the expression of marker genes in primary tumors, gene expression profiles for 80 NSCLC tumors exhibiting ≥50% tumor content were extracted from the commercially available database, BioExpress (GeneLogic, Gaithersburg MD). A single probe set was selected for each gene, although probe sets with no Entrez gene identifier assigned were assumed to represent different genes. Only probe sets represented in both primary tumor and cell line data sets were considered. These included probe sets for 35 genes represented in the erlotinib-sensitive class and for 34 genes represented in the erlotinib-insensitive class. The average Pearson correlation coefficient was calculated for (a) all pairs of sensitive genes, (b) all pairs of insensitive genes, and (c) all pairs of genes in which one was from the sensitive set and one was from the insensitive set. None of the equivalent average correlations for 100,000 random sets of 35 and 34 genes met or exceeded those for the real sets.

Western blot analysis. NSCLC cell lines were lysed for 1 hour in Cell Extraction Buffer (Biosource, Camarillo, CA) supplemented with protease and phosphatase inhibitors and precleared by centrifugation before protein quantification using the BCA Protein Assay kit (Pierce, Rockford, IL). Proteins (20 μg) resolved by SDS-PAGE were electrophoretically transferred to polyvinylidene difluoride membrane and Western blots were carried out using standard techniques. The following primary antibodies were used in this study, E-cadherin (clone 67A4, Santa Cruz Biotechnology, Santa Cruz, CA), γ-catenin (clone H-1, Santa Cruz Biotechnology), N-cadherin (H-63, Santa Cruz Biotechnology), vimentin (clone RV202, Becton Dickinson), glyceraldehyde-3-phosphate dehydrogenase (clone 9484, Abcam, Inc., Cambridge, MA).

Immunohistochemistry. Immunohistochemistry for E-cadherin was done on freshly cut 5-μm sections of tumor samples assembled on a tissue microarray. Following deparaffinization, antigen retrieval was done by pretreating with Target Retrieval Solution at 110°C for 20 minutes (DakoCytomation, Carpinteria CA). The pretreated sections were then incubated with primary mouse monoclonal IgG2 antibody against E-cadherin (clone 36, PharMingen) at a concentration of 1 μg/mL for 60 minutes at ambient temperature. Primary antibody bound to the sections was detected using biotinylated horse anti-mouse IgG and visualized using the avidin-biotin peroxidase complex technique (Vectastain ABC Elite, Vector Laboratories) and diaminobenzidine as chromagen. The intensity of E-cadherin membranous/cytoplasmic expression was scored as follows: 0, absent; 1, weak; 2, moderate; 3, strong. Positive staining was prospectively defined as intensity scores 2 and 3, and negative staining as intensity scores 0 and 1.

Patient samples. TRIBUTE, a phase III randomized, placebo-controlled trial conducted in the United States sponsored by Genentech enrolled 1,079 chemotherapy-naïve patients with locally advanced or metastatic (stage IIIB or IV) NSCLC to compare the survival of patients who received erlotinib administered concurrently with a regimen of carboplatin and paclitaxel (n = 539) to patients who received carboplatin and paclitaxel alone (n = 540). The primary efficacy end point was duration of survival and the secondary efficacy end points were time to progression, objective response rate as defined by RECIST, duration of response, and time to symptomatic progression. The erlotinib-containing arm did not show any advantage for survival, objective response rate, time to progression, and duration of response over carboplatin and paclitaxel alone in the overall study population (18).

The TRIBUTE clinical protocol was approved by Independent Review Boards for each participating site and by the U.S Food and Drug Administration. All patients gave written informed consent for participation in the clinical study. Patients were given the option of providing an additional written informed consent to allow release of their archival tumor samples for research purposes. Only a subset of samples from patients who provided the second, fully optional written informed consent for tumor tissue research was included in this study. A tissue microarray was constructed, which contained triplicate 0.6-mm-diameter tumor tissue cores taken from donor specimens embedded in paraffin blocks. One hundred twenty-seven patients provided blocks containing tumor in sufficient quantity for microarray construction. Some cases suffered tissue loss or lack of tumor cell representation in the tissue microarray sections, such that E-cadherin immunohistochemistry scores were generated for 87 patients.

Statistical analyses. Demographic variables, best tumor response rate, time to progression, and overall survival were summarized by E-cadherin membranous/cytoplasmic (M/C) staining status and treatment received. Comparisons of best tumor response rate between the two treatment groups were made using the Fisher's exact test. The Kaplan-Meier estimated curves for time to progression and overall survival were presented by E-cadherin M/C staining status. Median time to progression and overall survival were estimated using the Kaplan-Meier approach. The log-rank test was used to compare time to progression and overall survival between two treatment groups. Hazard ratios were estimated using Cox regression.

Characterizing epidermal growth factor receptor-mediated growth potential and erlotinib sensitivity in non–small cell lung carcinoma cell lines. The growth inhibitory effect of erlotinib was investigated in a panel of 42 NSCLC-derived cell lines under low serum tissue culture conditions with added EGFR ligand (TGF-α, 3 nmol/L) to specifically activate EGFR. The degree of sensitivity to erlotinib was determined from a dose-response curve and the IC50 was assessed from multiple experiments (Fig. 1A, left axis). In addition, ligand-driven growth potential of each cell line was assessed at multiple concentrations of TGF-α, and the maximum degree of TGF-α-induced growth (independent of ligand concentration) was determined (Fig. 1A, right axis). NSCLC cell lines exhibited nearly a 2-log range in erlotinib IC50 (0.3 to ≥10 μmol/L), with the cell lines most responsive to the mitogenic effects of EGFR ligand also representing a majority of the erlotinib-sensitive lines. A few exceptions were noted, including the positive-control epidermal cell line, A431, which exhibits a submicromolar IC50, as reported previously (19). For subsequent analyses, cell lines were classified as sensitive or insensitive based on a 0.85 SD window around the median IC50, resulting in cell lines exhibiting IC50 < 2.0 μmol/L defined as sensitive and IC50 >8.0 μmol/L as insensitive. Cell lines exhibiting 2 to 8 μmol/L IC50 values were consequently defined to exhibit intermediate sensitivity.

Fig. 1.

Characterization of a panel of NSCLC cell lines for erlotinib sensitivity and EGFR pathway–related variables. A, inhibition of TGF-α-driven growth by erlotinib was determined in a panel of NSCLC cell lines and is reported as a median IC50 from a minimum of three experiments (left Y axis). In addition, TGF-α-driven growth potential was assessed from a dose-response of ligand and is presented as a maximal stimulation index (right Y axis). The histopathologic subtype, EGFR gene status, Kras gene status, and ability of erlotinib to induce caspase-3/7 activity in the cell lines is depicted below the graph. Apoptosis is defined as a dose-dependent induction in caspase-3/7 activity. EGFR gene amplifications were determined by fluorescence in situ hybridization analysis. A431 cells represent an epidermis-derived control cell line that expresses activated EGFR. Red, NSCLC cell lines exhibiting IC50 < 2 μmol/L; gray, 2 to 8 μmol/L; blue, >8 μmol/L. SI, stimulation index; A (blue), amplification; M, mutation; S, squamous; L, large cell, A (yellow), adenocarcinoma; B, bronchioalveolar; N, not defined; nd, not done. B, induction of caspase activity in representative NSCLC cell lines as determined from a dose response of erlotinib (•, μmol/L, left Y axis) or C225 (, μg/mL, right Y axis) in the presence of TGF-α-activated EGFR. C, heat map representing either transcript expression or cell surface protein expression levels of select HER family receptors and ligands, as determined by either Taqman quantitative PCR or fluorescence-activated cell sorting (FACS) analysis. Cell lines are sorted based on ascending IC50 and P values were determined between sensitive (<2 μmol/L IC50) and insensitive (>8 μmol/L IC50) NSCLC cell lines. Red, high expression; Green, low expression.

Fig. 1.

Characterization of a panel of NSCLC cell lines for erlotinib sensitivity and EGFR pathway–related variables. A, inhibition of TGF-α-driven growth by erlotinib was determined in a panel of NSCLC cell lines and is reported as a median IC50 from a minimum of three experiments (left Y axis). In addition, TGF-α-driven growth potential was assessed from a dose-response of ligand and is presented as a maximal stimulation index (right Y axis). The histopathologic subtype, EGFR gene status, Kras gene status, and ability of erlotinib to induce caspase-3/7 activity in the cell lines is depicted below the graph. Apoptosis is defined as a dose-dependent induction in caspase-3/7 activity. EGFR gene amplifications were determined by fluorescence in situ hybridization analysis. A431 cells represent an epidermis-derived control cell line that expresses activated EGFR. Red, NSCLC cell lines exhibiting IC50 < 2 μmol/L; gray, 2 to 8 μmol/L; blue, >8 μmol/L. SI, stimulation index; A (blue), amplification; M, mutation; S, squamous; L, large cell, A (yellow), adenocarcinoma; B, bronchioalveolar; N, not defined; nd, not done. B, induction of caspase activity in representative NSCLC cell lines as determined from a dose response of erlotinib (•, μmol/L, left Y axis) or C225 (, μg/mL, right Y axis) in the presence of TGF-α-activated EGFR. C, heat map representing either transcript expression or cell surface protein expression levels of select HER family receptors and ligands, as determined by either Taqman quantitative PCR or fluorescence-activated cell sorting (FACS) analysis. Cell lines are sorted based on ascending IC50 and P values were determined between sensitive (<2 μmol/L IC50) and insensitive (>8 μmol/L IC50) NSCLC cell lines. Red, high expression; Green, low expression.

Close modal

To address the induction of apoptosis as a potential mechanism for erlotinib activity in the NSCLC cell line panel, caspase-3/7 activity was measured in response to erlotinib or the anti-EGFR antibody, C225, as a specificity control (Fig. 1B, summarized in Fig. 1A). Although erlotinib did not induce caspase activity in all erlotinib-sensitive lines (e.g., H838), we did detect the dose-dependent induction of caspase activity in several sensitive NSCLC cell lines. The induction of apoptosis by erlotinib was likely specific to EGFR inhibition because cell lines exhibiting erlotinib-induced caspase activity always showed a similar effect with C225, albeit often at a somewhat lower level. In addition, one cell line (H1975) was identified that exhibited a dose-dependent induction of apoptosis with C225, but no apoptotic activity with erlotinib at any dose examined. We did not detect caspase induction in any of the tested cell lines that were insensitive to erlotinib. Thus, a subset of sensitive NSCLC cell lines can undergo apoptosis upon treatment with erlotinib.

Characterizing determinants of epidermal growth factor receptor antagonist activity in non–small cell lung carcinoma cell lines. Previously reported determinants of EGFR antagonist activity have included EGFR activating mutations, EGFR gene amplification, EGFR overexpression, KRAS status, and histopathologic subtypes (20). To determine whether such markers were associated with erlotinib sensitivity, we examined EGFR and KRAS status within the NSCLC cell line panel (Fig. 1A, bottom). EGFR activating mutations affecting the tyrosine kinase domain were identified in two NSCLC cell lines as reported previously, including an active site deletion in the sensitive cell line H1650 (del.E746-A750) and a double mutation in the erlotinib-insensitive cell line H1975 (L858R, T790M; ref. 21). A novel genomic amplification in the EGFR gene was identified in H1838 cells by fluorescence in situ hybridization and the previously reported amplification in the control A431 cell line was confirmed (22). Both EGFR amplified cell lines were sensitive to the growth inhibitory effects of erlotinib; however, they exhibited minimal mitogenic response to exogenous ligand. KRAS mutations were identified in 31% of the NSCLC cell lines, which, interestingly, clustered primarily in cell lines of intermediate sensitivity (2-8 μmol/L IC50) to erlotinib. Two cell lines (Hop18 and H2122) with KRAS mutation showed exquisite sensitivity to erlotinib (IC50 < 1 μmol/L), with Hop18 also showing induction of apoptosis after erlotinib or C225 treatment. In addition, no specific histopathologic subtype of NSCLC cell line showed a striking correlation with sensitivity to erlotinib.

Expression levels of all HER family receptors and several relevant EGFR ligands were examined by quantitative reverse transcription-PCR and EGFR surface receptor status was established by fluorescence-activated cell sorting analysis (Fig. 1C). EGFR expression as assessed by quantitative reverse transcription-PCR and by cell surface protein expression showed higher levels in some erlotinib-sensitive lines, but showed no statistically significant correlation with sensitivity to erlotinib (P = 0.06; t test sensitive versus insensitive). Upon further analysis of HER2 and HER3, we found that HER3 transcript levels were elevated in erlotinib-sensitive lines relative to insensitive lines (P = 0.04), consistent with a previous report using a smaller panel of cell lines (13). Erlotinib-sensitive lines exhibit overall elevated levels of EGFR ligand transcripts, including β-cellulin (P = 0.03), TGF-α (P = 0.0002), and amphiregulin (P = 0.01). Thus, sensitive cell lines exhibit either ligand-driven growth potential, elevated basal expression of select EGFR ligand transcripts, and/or EGFR mutations/amplifications.

Identification of gene signatures correlating with erlotinib sensitivity. Basal mRNA gene expression profiles were generated from the entire panel of 42 NSCLC cell lines using Affymetrix oligo microarrays. Using the 4,562 most variably expressed genes, unsupervised hierarchical clustering classified all NSCLC cell lines into two subclasses, with a majority of the erlotinib-sensitive (<2 μmol/L IC50) cell lines clustering within one of the two main branches (Fig. 2A). Only 2 of 16 sensitive cell lines (Sk-Mes-1 and H838) clustered with the insensitive/intermediate sensitive cell lines and, conversely, 2 of 14 insensitive lines (H1975 and H1781) and 2 of 11 intermediate sensitive lines (H647 and H441) clustered on the same branch as the erlotinib-sensitive lines. Similar results were obtained if cell lines were clustered using all genes on the Affymetrix oligo microarray (data not shown). To identify more accurately genes differentially expressed between the NSCLC cell lines classified as erlotinib sensitive or insensitive, a supervised gene identification analysis was carried out using the 19,592 most variably expressed probe sets. The 100 probe sets with the highest discriminatory ability between the two classes, sensitive and insensitive, were identified by querying the top 50 probe sets overexpressed in the sensitive and insensitive cell lines, respectively (Fig. 2B; Table 1). Many of the 50 probe sets exhibiting elevated expression within erlotinib-sensitive cell lines detected epithelial-related genes, such as E-cadherin, plakophilin 3, and stratifin1. Conversely, primarily mesenchymal and/or morphogenesis-related genes were detected within the 50-probe erlotinib insensitive marker set. Upon inspection, a gene ontology analysis of these 50 marker probe sets revealed that 24% (12 of 50) represented morphogenesis-related biological processes. Noteworthy genes within this data set included the mesenchymal markers vimentin, TCF8 (ZEB-1), and epimorphin, which have been shown to play important roles in epithelial-mesenchymal interactions (2325).

Fig. 2.

A multigene mRNA expression signature indicative of EMT differentially classifies erlotinib-sensitive versus erlotinib-insensitive cell lines. A, two-dimensional, nonsupervised hierarchical cluster analysis of NSCLC cell line mRNA expression profiles. The 4,562 most differentially expressed genes were selected for this analysis. Red, cell lines exhibiting IC50 < 2 μmol/L; gray, 2 to 8 μmol/L; blue, >8 μmol/L. B, heat map representing the top 100 marker genes identified from a supervised analysis of the most sensitive (<2 μmol/L IC50) or insensitive (>8 μmol/L IC50) NSCLC lines. The complete gene list is reported in Table 1; however, several notable genes are shown. Red lines, genes determined to classify as morphogenesis-related biological processes from a gene ontology analysis. C, two-dimensional hierarchical clustering of all NSCLC cell lines using seven classic EMT marker genes. Red, cell lines exhibiting IC50 < 2 μmol/L; gray, 2 to 8 μmol/L; blue, >8 μmol/L.

Fig. 2.

A multigene mRNA expression signature indicative of EMT differentially classifies erlotinib-sensitive versus erlotinib-insensitive cell lines. A, two-dimensional, nonsupervised hierarchical cluster analysis of NSCLC cell line mRNA expression profiles. The 4,562 most differentially expressed genes were selected for this analysis. Red, cell lines exhibiting IC50 < 2 μmol/L; gray, 2 to 8 μmol/L; blue, >8 μmol/L. B, heat map representing the top 100 marker genes identified from a supervised analysis of the most sensitive (<2 μmol/L IC50) or insensitive (>8 μmol/L IC50) NSCLC lines. The complete gene list is reported in Table 1; however, several notable genes are shown. Red lines, genes determined to classify as morphogenesis-related biological processes from a gene ontology analysis. C, two-dimensional hierarchical clustering of all NSCLC cell lines using seven classic EMT marker genes. Red, cell lines exhibiting IC50 < 2 μmol/L; gray, 2 to 8 μmol/L; blue, >8 μmol/L.

Close modal
Table 1.

Top 100 classifier probe sets predicting erlotinib sensitivity in NSCLC cell lines

Class: sensitive
Class: insensitive
S/N scorePUniGene IDCommon gene nameS/N scorePUniGene IDCommon gene name
1.52 5.9e−07 Hs.348553 Immortalization-up-regulated protein 1.2e−03 Hs.435800 Vimentin 
1.4 1.5e−05 Hs.95549 Hypothetical protein FLJ20273 0.97 9.5e−04 Hs.250687 TRP-1 protein 
1.4 3.2e−06 Hs.95549 Hypothetical protein FLJ20273 0.93 1.0e−02 Hs.232068 Transcription factor 8 (TCF8) 
1.37 1.0e−06 Hs.439180 Chromosome 19 open reading frame 21 0.92 2.1e−04 Hs.508741 Neuronal protein 3.1 
1.3 2.9e−05 Hs.278611 Polypeptide N-acetylgalactosaminyltransferase 3 0.92 6.7e−04 Hs.250687 TRP-1 protein 
1.3 3.0e−06 Hs.339 P2Y purinoceptor 2 0.92 7.9e−04 — Hypothetical protein LOC284267 
1.26 3.7e−05 Hs.95549 Hypothetical protein FLJ20273 0.89 3.9e−05 Hs.433394 Tubulin α-3 
1.24 2.8e−06 Hs.415770 Connexin 31 0.87 3.7e−04 Hs.99865 Epimorphin 
1.24 1.8e−05 Hs.148074 Plakophilin 3 0.87 2.0e−05 Hs.32374 Deltex-3 
1.24 1.4e−05 Hs.380684 Phosphatidate cytidylyltransferase 1 0.86 5.6e−03 Hs.164467 Transcriptional repressor AP-2rep 
1.23 3.2e−05 Hs.76550 MAL2 protein 0.85 1.7e−03 Hs.416385 Insulin-induced gene 1 protein 
1.2 2.5e−05 Hs.380684 Phosphatidate cytidylyltransferase 1 0.84 7.6e−03 Hs.22451 FLJ10357 protein 
1.19 1.0e−04 Hs.23582 Tumor-associated calcium signal transducer 2 0.84 6.9e−04 Hs.437379 Up-regulated in liver cancer 1 
1.17 1.6e−04 Hs.194657 E-cadherin 0.84 1.5e−02 Hs.301711 Hypothetical protein LOC387882 
1.17 4.5e−05 Hs.31439 SPINT2 0.83 1.1e−03 Hs.40368 Sigma-adaptin 1B 
1.16 5.0e−05 Hs.254605 Microtubule-associated protein 7 0.82 1.3e−02 Hs.9999 Hematopoietic neural membrane protein 
1.13 2.1e−05 Hs.55279 Maspin 0.81 3.4e−04 Hs.508741 Neuronal protein 3.1 
1.11 2.9e−04 Hs.184510 Stratifin 0.81 1.4e−03 Hs.511758 Centrosomal colon cancer autoantigen protein 
1.09 1.1e−04 Hs.380684 Phosphatidate cytidylyltransferase 1 0.81 6.4e−04 Hs.40368 Sigma-adaptin 1B 
1.08 1.1e−05 Hs.148074 Plakophilin 3 0.8 2.4e−04 Hs.387856 TPR repeat containing protein KIAA1043 
1.08 1.1e−06 Hs.11669 Laminin α-5 chain 0.8 1.7e−02 Hs.354740 Calcium-activated potassium channel αsubunit 1 
1.08 6.6e−04 Hs.184510 Stratifin 0.79 5.4e−04 Hs.40368 σ-Adaptin 1B 
1.07 1.7e−05 Hs.415770 Connexin 31 0.78 2.1e−02 Hs.130865 Dickkopf-3 
1.06 3.5e−04 Hs.21420 Serine/threonine-protein kinase PAK 6 0.77 1.1e−03 Hs.250687 TRP-1 protein 
1.04 1.8e−04 Hs.376874 Potassium channel subfamily K member 1 0.77 2.2e−03 Hs.380138 N-chimaerin 
1.04 1.1e−04 Hs.440411 Transcription factor AP-2γ 0.77 1.5e−02 Hs.232068 Transcription factor 8 (TCF8) 
1.03 3.9e−05 Hs.193490 Hypothetical protein FLJ31952 0.77 4.4e−03 Hs.109672 ST6GALNAC6 
1.03 1.6e−04 Hs.152423 Hypothetical protein FLJ21274 0.77 1.2e−03 Hs.306298 ZNF682 
1.02 2.4e−05 Hs.193490 Hypothetical protein FLJ31952 0.76 8.3e−03 Hs.150406 Zinc finger protein 37 homologue 
1.01 5.2e−05 Hs.165904 Epsin-3 0.76 1.8e−02 Hs.387385 SMURF2 
1.1e−04 Hs.513684 Unknown EST 0.76 3.1e−03 Hs.356109 DKFZp762F237 
0.99 2.3e−04 Hs.51133 BRK 0.75 1.1e−02 Hs.25220 Glycosyltransferase-like protein LARGE 
0.99 7.1e−05 Hs.67776 FLJ46072 protein 0.75 5.5e−03 Hs.438669 Insulin receptor 
0.99 3.4e−04 Hs.396798 β-1,4-galactosyltransferase 1 0.74 3.7e−04 Hs.40368 σ-Adaptin 1B 
0.99 3.7e−05 Hs.95549 Hypothetical protein FLJ20273 0.74 4.3e−03 Hs.29595 PRA1 family protein 2 
0.99 2.6e−04 Hs.298079 KLC2L protein 0.74 4.6e−02 Hs.270435 Hypothetical protein FLJ12985 
0.98 4.7e−04 Hs.21420 Serine/threonine-protein kinase PAK 6 0.74 6.7e−05 Hs.285673 Protein C14orf159 
0.98 1.5e−03 Hs.56937 ST14 protein 0.73 9.2e−02 Hs.76118 Ubiquitin carboxyl-terminal hydrolase isozyme L1 
0.98 9.6e−05 Hs.26765 LOC58489 protein 0.73 5.0e−05 Hs.25132 KIAA0470 protein 
0.98 7.2e−04 Hs.417795 Dermokine-β 0.73 9.3e−03 Hs.232068 Transcription factor 8 (TCF8) 
0.97 4.4e−04 Hs.194657 E-cadherin 0.72 1.1e−04 Hs.512712 Tubulin, β 
0.97 1.7e−03 Hs.233950 HAI-1 0.72 3.0e−02 Hs.164267 Dual-specificity tyrosine-phosphorylation regulated kinase 3 
0.97 7.2e−05 Hs.377416 Engulfment and cell motility protein 3 0.72 4.2e−02 Hs.37196 GPR protein 
0.97 1.1e−03 Hs.528844 Unknown EST 0.71 7.9e−04 Hs.60177 Zinc finger protein DZIP1 
0.96 6.6e−04 Hs.75746 Aldehyde dehydrogenase 1A3 0.71 1.2e−02 Hs.22451 FLJ10357 protein 
0.96 4.5e−04 Hs.178695 Mitogen activated protein kinase p38 δ 0.7 9.9e−03 Hs.148258 Protein C16orf45 
0.96 5.4e−04 Hs.24743 FLJ20171 protein 0.7 4.2e−02 Hs.50150 Unknown EST 
0.95 2.8e−05 Hs.427719 PARX protein 0.7 1.5e−02 Hs.370049 Hypothetical protein FLJ26472 
0.94 5.0e−04 — Claudin-7 0.69 7.8e−03 Hs.412117 Annexin A6 
0.94 7.4e−04 Hs.18894 Adaptor-related protein complex 1, μ2 subunit 0.69 7.5e−03 Hs.298258 Unknown EST 
Class: sensitive
Class: insensitive
S/N scorePUniGene IDCommon gene nameS/N scorePUniGene IDCommon gene name
1.52 5.9e−07 Hs.348553 Immortalization-up-regulated protein 1.2e−03 Hs.435800 Vimentin 
1.4 1.5e−05 Hs.95549 Hypothetical protein FLJ20273 0.97 9.5e−04 Hs.250687 TRP-1 protein 
1.4 3.2e−06 Hs.95549 Hypothetical protein FLJ20273 0.93 1.0e−02 Hs.232068 Transcription factor 8 (TCF8) 
1.37 1.0e−06 Hs.439180 Chromosome 19 open reading frame 21 0.92 2.1e−04 Hs.508741 Neuronal protein 3.1 
1.3 2.9e−05 Hs.278611 Polypeptide N-acetylgalactosaminyltransferase 3 0.92 6.7e−04 Hs.250687 TRP-1 protein 
1.3 3.0e−06 Hs.339 P2Y purinoceptor 2 0.92 7.9e−04 — Hypothetical protein LOC284267 
1.26 3.7e−05 Hs.95549 Hypothetical protein FLJ20273 0.89 3.9e−05 Hs.433394 Tubulin α-3 
1.24 2.8e−06 Hs.415770 Connexin 31 0.87 3.7e−04 Hs.99865 Epimorphin 
1.24 1.8e−05 Hs.148074 Plakophilin 3 0.87 2.0e−05 Hs.32374 Deltex-3 
1.24 1.4e−05 Hs.380684 Phosphatidate cytidylyltransferase 1 0.86 5.6e−03 Hs.164467 Transcriptional repressor AP-2rep 
1.23 3.2e−05 Hs.76550 MAL2 protein 0.85 1.7e−03 Hs.416385 Insulin-induced gene 1 protein 
1.2 2.5e−05 Hs.380684 Phosphatidate cytidylyltransferase 1 0.84 7.6e−03 Hs.22451 FLJ10357 protein 
1.19 1.0e−04 Hs.23582 Tumor-associated calcium signal transducer 2 0.84 6.9e−04 Hs.437379 Up-regulated in liver cancer 1 
1.17 1.6e−04 Hs.194657 E-cadherin 0.84 1.5e−02 Hs.301711 Hypothetical protein LOC387882 
1.17 4.5e−05 Hs.31439 SPINT2 0.83 1.1e−03 Hs.40368 Sigma-adaptin 1B 
1.16 5.0e−05 Hs.254605 Microtubule-associated protein 7 0.82 1.3e−02 Hs.9999 Hematopoietic neural membrane protein 
1.13 2.1e−05 Hs.55279 Maspin 0.81 3.4e−04 Hs.508741 Neuronal protein 3.1 
1.11 2.9e−04 Hs.184510 Stratifin 0.81 1.4e−03 Hs.511758 Centrosomal colon cancer autoantigen protein 
1.09 1.1e−04 Hs.380684 Phosphatidate cytidylyltransferase 1 0.81 6.4e−04 Hs.40368 Sigma-adaptin 1B 
1.08 1.1e−05 Hs.148074 Plakophilin 3 0.8 2.4e−04 Hs.387856 TPR repeat containing protein KIAA1043 
1.08 1.1e−06 Hs.11669 Laminin α-5 chain 0.8 1.7e−02 Hs.354740 Calcium-activated potassium channel αsubunit 1 
1.08 6.6e−04 Hs.184510 Stratifin 0.79 5.4e−04 Hs.40368 σ-Adaptin 1B 
1.07 1.7e−05 Hs.415770 Connexin 31 0.78 2.1e−02 Hs.130865 Dickkopf-3 
1.06 3.5e−04 Hs.21420 Serine/threonine-protein kinase PAK 6 0.77 1.1e−03 Hs.250687 TRP-1 protein 
1.04 1.8e−04 Hs.376874 Potassium channel subfamily K member 1 0.77 2.2e−03 Hs.380138 N-chimaerin 
1.04 1.1e−04 Hs.440411 Transcription factor AP-2γ 0.77 1.5e−02 Hs.232068 Transcription factor 8 (TCF8) 
1.03 3.9e−05 Hs.193490 Hypothetical protein FLJ31952 0.77 4.4e−03 Hs.109672 ST6GALNAC6 
1.03 1.6e−04 Hs.152423 Hypothetical protein FLJ21274 0.77 1.2e−03 Hs.306298 ZNF682 
1.02 2.4e−05 Hs.193490 Hypothetical protein FLJ31952 0.76 8.3e−03 Hs.150406 Zinc finger protein 37 homologue 
1.01 5.2e−05 Hs.165904 Epsin-3 0.76 1.8e−02 Hs.387385 SMURF2 
1.1e−04 Hs.513684 Unknown EST 0.76 3.1e−03 Hs.356109 DKFZp762F237 
0.99 2.3e−04 Hs.51133 BRK 0.75 1.1e−02 Hs.25220 Glycosyltransferase-like protein LARGE 
0.99 7.1e−05 Hs.67776 FLJ46072 protein 0.75 5.5e−03 Hs.438669 Insulin receptor 
0.99 3.4e−04 Hs.396798 β-1,4-galactosyltransferase 1 0.74 3.7e−04 Hs.40368 σ-Adaptin 1B 
0.99 3.7e−05 Hs.95549 Hypothetical protein FLJ20273 0.74 4.3e−03 Hs.29595 PRA1 family protein 2 
0.99 2.6e−04 Hs.298079 KLC2L protein 0.74 4.6e−02 Hs.270435 Hypothetical protein FLJ12985 
0.98 4.7e−04 Hs.21420 Serine/threonine-protein kinase PAK 6 0.74 6.7e−05 Hs.285673 Protein C14orf159 
0.98 1.5e−03 Hs.56937 ST14 protein 0.73 9.2e−02 Hs.76118 Ubiquitin carboxyl-terminal hydrolase isozyme L1 
0.98 9.6e−05 Hs.26765 LOC58489 protein 0.73 5.0e−05 Hs.25132 KIAA0470 protein 
0.98 7.2e−04 Hs.417795 Dermokine-β 0.73 9.3e−03 Hs.232068 Transcription factor 8 (TCF8) 
0.97 4.4e−04 Hs.194657 E-cadherin 0.72 1.1e−04 Hs.512712 Tubulin, β 
0.97 1.7e−03 Hs.233950 HAI-1 0.72 3.0e−02 Hs.164267 Dual-specificity tyrosine-phosphorylation regulated kinase 3 
0.97 7.2e−05 Hs.377416 Engulfment and cell motility protein 3 0.72 4.2e−02 Hs.37196 GPR protein 
0.97 1.1e−03 Hs.528844 Unknown EST 0.71 7.9e−04 Hs.60177 Zinc finger protein DZIP1 
0.96 6.6e−04 Hs.75746 Aldehyde dehydrogenase 1A3 0.71 1.2e−02 Hs.22451 FLJ10357 protein 
0.96 4.5e−04 Hs.178695 Mitogen activated protein kinase p38 δ 0.7 9.9e−03 Hs.148258 Protein C16orf45 
0.96 5.4e−04 Hs.24743 FLJ20171 protein 0.7 4.2e−02 Hs.50150 Unknown EST 
0.95 2.8e−05 Hs.427719 PARX protein 0.7 1.5e−02 Hs.370049 Hypothetical protein FLJ26472 
0.94 5.0e−04 — Claudin-7 0.69 7.8e−03 Hs.412117 Annexin A6 
0.94 7.4e−04 Hs.18894 Adaptor-related protein complex 1, μ2 subunit 0.69 7.5e−03 Hs.298258 Unknown EST 

To confirm that the coregulated expression of the above-described marker genes is also reflected in the biology of tumors and is not specific to cell lines, the correlations between the transcriptional expression levels of these genes using Affymetrix oligo microarray data for 80 primary NSCLC tumor samples were analyzed (data not shown). There was a significant positive correlation regarding expression levels of both genes being up-regulated within the erlotinib-sensitive set and between genes showing higher expression within the erlotinib insensitive set (P < 1e−5 in each case). There was also a significant negative correlation between pairs of genes in which one was from the sensitive set and one was from the insensitive set (P < 1e−5). This suggests that the sets of genes associated with erlotinib sensitivity and insensitivity are coordinately and oppositely regulated in NSCLC tumors.

A two-dimensional, unsupervised hierarchical clustering approach was used to classify all NSCLC lines using several well-established epithelial to mesenchymal transition (EMT) marker genes; five of which were not in our top 100 gene list (23, 26) to determine if a potential EMT gene signature could differentially classify NSCLC cell lines based on erlotinib sensitivity. Using the complete panel of cell lines, these EMT marker genes clustered in two primary branches; with the epithelial genes E-cadherin, γ-catenin, and α-catenin forming one cluster and the mesenchymal genes, N-cadherin, vimentin, FGFR1, and TGF-β forming the second cluster (Fig. 2C). With a few exceptions, the majority of sensitive and insensitive cell lines correctly fell within epithelial or mesenchymal branches, respectively. Taken together, these data support an epithelial to mesenchymal gene signature defining two distinct subsets of NSCLC cell lines and this gene signature associated with the sensitivity/insensitivity of the cell lines to treatment with erlotinib.

Expression of EMT protein markers in non–small cell lung carcinoma cell lines and primary tumor specimens. To evaluate further the potential role of EMT markers in defining erlotinib sensitivity, the protein expression levels of a few epithelial or mesenchymal-related proteins in the NSCLC cell lines were assessed (Fig. 3A). As was observed in the mRNA expression analyses, expression of the epithelial proteins E-cadherin and γ-catenin was high within the erlotinib-sensitive cell lines by Western blot analysis. Conversely, expression of the mesenchymal marker vimentin (and to some degree, N-cadherin) was detected primarily in the insensitive cell lines. An unsupervised hierarchical analysis of the semiquantitatively derived protein expression values, further shows that the majority of erlotinib-sensitive cell lines are represented within the epithelial subgroup (Fig. 3B).

Fig. 3.

Expression levels of EMT-related epithelial proteins differentially classify erlotinib-sensitive versus erlotinib-insensitive NSCLC cell lines. A, Western blot analysis of select EMT-related proteins in a panel of NSCLC cell lines, sorted based on ascending erlotinib IC50. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH), control immunoblot. B, two-dimensional hierarchical clustering using protein expression values for the four EMT proteins analyzed. Protein expression values were determined by densitometry from immunoblots. Red, cell lines exhibiting IC50 < 2 μmol/L; gray, 2 to 8 μmol/L; blue, >8 μmol/L. C, representative E-cadherin immunohistochemical staining of NSCLC cell lines and primary tumor sections. Staining intensity was determined on a 0 to 3 scale, with positive (+) E-cadherin staining being defined as a staining intensity of 2 to 3+ and negative (−) E-cadherin staining as 0 to 1+. Results of E-cadherin immunohistochemical analysis in NSCLC cell lines are annotated above the Western blots in (A).

Fig. 3.

Expression levels of EMT-related epithelial proteins differentially classify erlotinib-sensitive versus erlotinib-insensitive NSCLC cell lines. A, Western blot analysis of select EMT-related proteins in a panel of NSCLC cell lines, sorted based on ascending erlotinib IC50. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH), control immunoblot. B, two-dimensional hierarchical clustering using protein expression values for the four EMT proteins analyzed. Protein expression values were determined by densitometry from immunoblots. Red, cell lines exhibiting IC50 < 2 μmol/L; gray, 2 to 8 μmol/L; blue, >8 μmol/L. C, representative E-cadherin immunohistochemical staining of NSCLC cell lines and primary tumor sections. Staining intensity was determined on a 0 to 3 scale, with positive (+) E-cadherin staining being defined as a staining intensity of 2 to 3+ and negative (−) E-cadherin staining as 0 to 1+. Results of E-cadherin immunohistochemical analysis in NSCLC cell lines are annotated above the Western blots in (A).

Close modal

To evaluate the potential for an epithelial versus mesenchymal phenotype to be predictive of erlotinib activity in patients, immunohistochemistry for the epithelial marker, E-cadherin, was done on formalin-fixed, paraffin-embedded tissues. Of the EMT marker proteins, E-cadherin was chosen based on the fact that it exhibited the best correlation with sensitivity in cell lines (P < 0.0002; Table 1) and robust immunohistochemical assays have been described (27, 28). E-cadherin expression was initially evaluated on a tissue microarray representing the panel of NSCLC cell lines and exhibited strong membranous and cytoplasmic staining in subsets of NSCLC cell lines, with expression levels by immunohistochemistry correlating well with expression as determined by Western blotting (Fig. 3A and C). Nuclear staining of E-cadherin, which was not correlated with membranous/cytoplasmic staining, was observed in several cell lines. Because some of the cell lines with nuclear staining had no detectable expression by mRNA or Western analysis, the nuclear staining is likely to be artifactual and was not assessed in subsequent analyses. Applying similar immunohistochemical staining criteria, robust membraneous/cytoplasmic expression of E-cadherin was observed in epithelial tumor cells in primary NSCLC tissue specimens (Fig. 3C) and was, therefore, used to evaluate E-cadherin expression in clinical specimens. Tumors that exhibited E-cadherin immunoreactivity generally showed fairly consistent staining in >75% of the tumor cells, so only the intensity of membranous/cytoplasmic staining was used to score E-cadherin expression in further analyses. Again, nuclear staining was observed in some cases that did not correlate with membranous/cytoplasmic staining. Although membranous/cytoplasmic staining was limited to epithelial tumor cells, nuclear staining was also seen in stromal cells, consistent with artifactual staining.

Relationship between E-cadherin staining and outcome in TRIBUTE. To evaluate the potential relationship between E-cadherin expression and clinical activity of erlotinib, E-cadherin expression was evaluated in biopsies from chemotherapy-naïve patients with locally advanced or metastatic NSCLC from a randomized, placebo-controlled trial (TRIBUTE; ref. 18). Patients received either erlotinib administered concurrently with chemotherapy or chemotherapy alone. Baseline demographics and clinicopathologic characteristics are summarized in Table 2. For this study, we were able to analyze E-cadherin expression by immunohistochemistry on tumors from patients for whom adequate biopsy material was available for the construction of a tissue microarray. Immunohistochemical data for E-cadherin expression was obtained from 87 of 1,079 (8%) patients who participated in TRIBUTE. The clinical variables examined were balanced between those patients for whom E-cadherin expression could be determined versus all patients, with the exception of smoking history. There was a lower percentage of never smokers in patients with known E-cadherin expression (4 of 87 = 4.6% versus 116 of 1,078 = 10.8%). Within the E-cadherin status-known subgroup, patient characteristics were generally balanced between the two treatment arms.

Table 2.

Relationship between E-cadherin immunoreactivity and clinicopathologic variables/clinical response

E-cadherin positive
E-cadherin negative
All patients
Chemo (n = 37)E + chemo (n = 28)Chemo (n = 14)E + chemo (n = 8)Chemo (n = 540)*E + chemo (n = 539)
Demographics and baseline characteristics       
Age (y)       
    Mean ± SD 62 ± 10 63 ± 14 70 ± 9 63 ± 8 63 ± 10 63 ± 10 
Gender (%)       
    Female 13 (35) 13 (46) 4 (29) 6 (75) 207 (38) 217 (40) 
Ethnicity (%)       
    White 35 (95) 26 (93) 12 (86) 8 (100) 482 (89) 452 (84) 
    Asian 1 (3) 0 (0) 1 (7) 0 (0) 13 (2) 21 (4) 
    Others 1 (3) 2 (7) 1 (7) 0 (0) 44 (8) 66 (12) 
Smoking history (%)       
    Never smokers 0 (0) 2 (7) 1 (7) 1 (13) 44 (8) 72 (13) 
ECOG performance status (%)       
    0 12 (32) 8 (29) 6 (43) 2 (25) 195 (36) 186 (35) 
    1 25 (68) 20 (71) 8 (57) 6 (75) 342 (64) 353 (66) 
    2 0 (0) 0 (0) 0 (0) 0 (0) 1 (0.2) 0 (0) 
Cancer stage (%)       
    IIIB 7 (19) 3 (11) 1 (7) 0 (0) 96 (18) 84 (16) 
    IV 30 (81) 25 (89) 13 (93) 8 (100) 443 (82) 455 (84) 
Histology (%)       
    Adenocarcinoma 23 (62) 17 (61) 8 (57) 5 (63) 331 (61) 323 (60) 
    Squamous cell carcinoma 8 (22) 7 (25) 3 (21) 1 (13) 87 (16) 98 (18) 
    Others 6 (16) 4 (14) 3 (21) 2 (25) 121 (22) 118 (22) 
Months since initial NSCLC diagnosis       
    Median 1.7 1.2 1.7 2.3 1.1 1.1 
    Range 0-44 0-132 0-35 0-62 0-209 0-248 
       
Best response, time to progression, and overall survival       
Best response       
    Response rate 5 (13.5) 5 (17.9) 6 (42.9) 2 (25.0) 104 (19.3) 116 (21.5) 
    95% CI for response rate (2.5-24.5) (3.7-32.0) (16.9-68.8) (0.0-55.0) (15.9-22.6) (18.1-25.0) 
    P (Fisher's exact)  0.7341  0.6494  0.4059 
Time to progression       
    Median time (wk) 19.3 34.0 30.0 19.1 22.3 23.7 
    HR  0.37  1.63  0.90 
    95% CI for HR  (0.19-0.73)  (0.50-5.33)  (0.78-1.05) 
    P (log rank)  0.0028  0.3976  0.1697 
Overall survival       
    Median time (mo) 13.6 15.1 11.3 9.7 10.5 10.6 
    HR  0.82  1.82  1.00 
    95% CI for HR  (0.43, 1.58)  (0.60, 5.49)  (0.86, 1.16) 
    P (log rank)  0.5604  0.2823  0.9517 
E-cadherin positive
E-cadherin negative
All patients
Chemo (n = 37)E + chemo (n = 28)Chemo (n = 14)E + chemo (n = 8)Chemo (n = 540)*E + chemo (n = 539)
Demographics and baseline characteristics       
Age (y)       
    Mean ± SD 62 ± 10 63 ± 14 70 ± 9 63 ± 8 63 ± 10 63 ± 10 
Gender (%)       
    Female 13 (35) 13 (46) 4 (29) 6 (75) 207 (38) 217 (40) 
Ethnicity (%)       
    White 35 (95) 26 (93) 12 (86) 8 (100) 482 (89) 452 (84) 
    Asian 1 (3) 0 (0) 1 (7) 0 (0) 13 (2) 21 (4) 
    Others 1 (3) 2 (7) 1 (7) 0 (0) 44 (8) 66 (12) 
Smoking history (%)       
    Never smokers 0 (0) 2 (7) 1 (7) 1 (13) 44 (8) 72 (13) 
ECOG performance status (%)       
    0 12 (32) 8 (29) 6 (43) 2 (25) 195 (36) 186 (35) 
    1 25 (68) 20 (71) 8 (57) 6 (75) 342 (64) 353 (66) 
    2 0 (0) 0 (0) 0 (0) 0 (0) 1 (0.2) 0 (0) 
Cancer stage (%)       
    IIIB 7 (19) 3 (11) 1 (7) 0 (0) 96 (18) 84 (16) 
    IV 30 (81) 25 (89) 13 (93) 8 (100) 443 (82) 455 (84) 
Histology (%)       
    Adenocarcinoma 23 (62) 17 (61) 8 (57) 5 (63) 331 (61) 323 (60) 
    Squamous cell carcinoma 8 (22) 7 (25) 3 (21) 1 (13) 87 (16) 98 (18) 
    Others 6 (16) 4 (14) 3 (21) 2 (25) 121 (22) 118 (22) 
Months since initial NSCLC diagnosis       
    Median 1.7 1.2 1.7 2.3 1.1 1.1 
    Range 0-44 0-132 0-35 0-62 0-209 0-248 
       
Best response, time to progression, and overall survival       
Best response       
    Response rate 5 (13.5) 5 (17.9) 6 (42.9) 2 (25.0) 104 (19.3) 116 (21.5) 
    95% CI for response rate (2.5-24.5) (3.7-32.0) (16.9-68.8) (0.0-55.0) (15.9-22.6) (18.1-25.0) 
    P (Fisher's exact)  0.7341  0.6494  0.4059 
Time to progression       
    Median time (wk) 19.3 34.0 30.0 19.1 22.3 23.7 
    HR  0.37  1.63  0.90 
    95% CI for HR  (0.19-0.73)  (0.50-5.33)  (0.78-1.05) 
    P (log rank)  0.0028  0.3976  0.1697 
Overall survival       
    Median time (mo) 13.6 15.1 11.3 9.7 10.5 10.6 
    HR  0.82  1.82  1.00 
    95% CI for HR  (0.43, 1.58)  (0.60, 5.49)  (0.86, 1.16) 
    P (log rank)  0.5604  0.2823  0.9517 

Abbreviations: E, erlotinib; chemo, chemotherapy; ECOG, Eastern Cooperative Oncology Group.

*

One patient did not have any baseline information collected. This patient was included in the overall survival analysis only.

Best tumor response rate in all patients as defined by RECIST criteria was 22% for the erlotinib- and chemotherapy-treated group and 19% for the chemotherapy alone group. Within the E-cadherin-positive staining intensity subgroup, the response rates were 18% and 14%, respectively, and for the E-cadherin-negative staining subgroup, response rates were 25% and 43%, respectively. None of the differences in response rate reached statistical significance.

Data from various sources point to the importance of stable disease as an important and meaningful outcome of treatment with erlotinib (29). Therefore, time to progression and overall survival are likely to be a more relevant outcome measures in the context of this study. For all patients treated with chemotherapy and erlotinib versus those treated with chemotherapy alone, time to progression was 23.7 versus 22.3 weeks [hazard ratio (HR), 0.90; log rank P = 0.17]. However, within the E-cadherin-positive staining subgroup, time to progression for those receiving erlotinib and chemotherapy versus chemotherapy alone was 34 versus 19.3 weeks (HR, 0.37; log-rank P = 0.003; Fig. 4A). Conversely for the E-cadherin-negative staining group, time to progression was 19.1 weeks for erlotinib and chemotherapy and 30 weeks for chemotherapy alone (HR, 1.63), although this difference was not statistically significant (log rank P = 0.40; Fig. 4B). An analysis with adjustment for baseline characteristics was not done due to small sample size.

Fig. 4.

Kaplan-Meier estimated time to progression (A and B) and estimated overall survival (C and D) for NSCLC patients treated with erlotinib + chemotherapy versus chemotherapy alone by E-cadherin positive (A and C) or negative (B and D) expression.

Fig. 4.

Kaplan-Meier estimated time to progression (A and B) and estimated overall survival (C and D) for NSCLC patients treated with erlotinib + chemotherapy versus chemotherapy alone by E-cadherin positive (A and C) or negative (B and D) expression.

Close modal

Overall survival was not significantly different between the erlotinib and chemotherapy versus chemotherapy alone–treated groups in all patients (HR, 1.0). Within the positive staining subgroup, treatment with erlotinib was associated with a longer median survival time (15.1 versus 13.6 months; HR, 0.82), although this was not statistically significant (log rank P = 0.56; Fig. 4C). Within the negative staining group, treatment with chemotherapy alone was associated with a longer median survival time (11.3 versus 9.7 months; HR, 1.82), although not significantly (P = 0.28; Fig. 4D). In addition, E-cadherin-positive staining was associated with a longer median survival time irrespective of treatment (13.6 months versus 10.6 months), indicating that it may be a prognostic marker, as previously reported (27, 28), although this difference was not significant (P = 0.49).

A pharmacogenomic approach using in vitro model systems was taken to identify predictive markers of erlotinib activity. Identification of clinical markers with significant predictive value requires the model system to be representative of the target pathway activity and the identified markers to be translatable to the clinical setting. We evaluated erlotinib activity under a culture condition that would allow us to assess both ligand-dependent and ligand-independent mechanisms of EGFR activity in a panel of NSCLC cell lines. The results strongly suggest that an underlying EGFR-driven biology is predictive of erlotinib activity as shown by the finding that erlotinib sensitivity could be predicted by either ligand-driven growth potential, ligand expression, and/or EGFR-activating mutations/amplifications. Specifically, NSCLC cell lines exhibiting the greatest sensitivity to erlotinib either harbored genetically activated EGFR, specifically a kinase domain deletion (H1650) and a genomic amplification (H1838), or exhibited an intrinsic ability to proliferate in response to EGFR activation by ligand. Two of the highly sensitive cell lines, H1838 and A431 (positive control) with genomic amplification of EGFR resulting in overexpression of mRNA and protein, exhibited no mitogenic activity, suggesting a ligand-independent growth mechanism. Furthermore, nearly all of the erlotinib-sensitive cell lines produced elevated basal levels of EGFR ligands, including β-cellulin, TGF-α, and amphiregulin, which may indicate a potential autocrine dependence for these cell lines. In contrast, insensitive lines exhibited minimal ligand-dependent growth potential, minimal basal expression of EGFR ligands, and no EGFR-activating genetic alterations. The one exception, H1975, harbors two mutations in the EGFR kinase domain, one of which (T970M) has been previously reported to confer resistance to EGFR small molecule antagonists (21, 30). This is further shown by our finding that erlotinib fails to induce apoptosis in H1975 cells, although a dose-dependent induction of apoptosis was observed with an anti-EGFR antibody antagonist (C225; Fig. 1B). Taken together, the data supports both ligand-dependent and ligand-independent growth mechanisms underlying erlotinib sensitivity in this cell line panel.

Mechanisms leading to erlotinib-dependent growth inhibition remain to be elucidated; however, EGFR mutations and amplifications have been reported to sensitize cell lines to antagonist-induced apoptosis (19, 31, 32). We found that in addition to EGFR mutant (H1650) or amplified (A431 and H1838) cell lines, a dose-dependent induction in caspase activity was similarly observed upon EGFR inhibition in several EGFR wild-type lines. This induction in caspase activity was specific to EGFR inhibition, as the EGFR inhibiting antibody, C225, induced apoptosis in the same cell lines as erlotinib. Also, the absolute degree of caspase induction in a few EGFR wild-type cell lines was comparable with that observed in the EGFR mutant cell line. This included an EGFR wild-type cell line expressing a KRAS-activating mutation (Hop18). These findings would suggest that tumors bearing wild-type, nonamplified EGFR could also be equally responsive to erlotinib inhibition. This is in contrast to previous reports in which similar inductions of apoptosis were not observed in EGFR wild-type cells (19, 31). One explanation for this discrepancy could be that we have evaluated an apoptotic mechanism of action in the context of receptor activation with ligand, as opposed to high serum conditions only. Interestingly, we find that several erlotinib-sensitive cell lines, including the EGFR mutant H1650 line, are insensitive (IC50 ≥ 10 μmol/L) to erlotinib-mediated growth inhibition when tested under high serum conditions without exogenous ligand (data not shown). Thus, it is possible that ligand activation of the EGFR is required to sensitize subsets of NSCLC cell lines to erlotinib-induced apoptosis.

Further characterization of these cell lines revealed additional associations with previously reported determinants of EGFR antagonist activity. Specifically, we detected a significant correlation of erlotinib sensitivity with HER3 expression levels. A similar correlation of HER3 expression levels and gefitinib activity was previously reported in smaller panels of NSCLC cell lines (13). However, we do not find a significant association with the level of HER2 overexpression in erlotinib-sensitive cell lines, contrary to previous reports (12, 33). Such a discrepancy could be due to many factors, including differences in tissue types (33), assessment of endogenous versus ectopically expressed HER2 (12), and/or assessment of mRNA compared with protein levels. Similarly, we did not detect a statistically significant correlation of EGFR expression levels with erlotinib sensitivity, although expression was elevated in some of the sensitive cell lines. Finally, KRAS mutations have been reported to associate with a lack of clinical activity to EGFR antagonists (8, 34). In our study, a majority of the KRAS mutant cell lines exhibit intermediate or low erlotinib sensitivity (n = 9); however, four KRAS mutant lines were sensitive to erlotinib. Two of these cell lines exhibited IC50 values <1 μmol/L and one of these, Hop18, underwent apoptosis after erlotinib and C225 treatment. Thus, our findings do not support a major role of KRAS mutations in conferring resistance to inhibition of EGFR in NSCLC tumor cells.

A predominant subclassification of NSCLC cell lines can be based on epithelial or mesenchymal characteristics. During tumor progression, the morphologic transformation of a tumor from an epithelial to mesenchymal phenotype has been associated with increased tumor aggressiveness and metastasis (23, 26, 35, 36). In particular, the loss of E-cadherin expression represents a hallmark of EMT, as the subsequent disassembly of adherens junctions upon E-cadherin loss results in the acquisition of a more motile and invasive tumor. Clinically, this has been significantly correlated with poor prognosis in NSCLC patients (27, 28, 37). The loss of various epithelial markers, highlighted by E-cadherin, also serves as a strong determinant of erlotinib insensitivity in the NSCLC cell lines tested in our study. Conversely, genes characteristic of a mesenchymal phenotype (i.e., vimentin), and genes specifically associated with EMT (i.e., TGF-β, TCF8, and epimorphin), are overexpressed in cell lines that are insensitive to erlotinib-mediated growth inhibition. For example, the E-cadherin transcriptional repressor TCF8 (ZEB1) is up-regulated in erlotinib-insensitive cell lines and, as shown previously, can maintain E-cadherin repression in lung cancer cell lines (38). Most importantly, these genes capable of subclassifying erlotinib-sensitive cell lines were also coregulated in primary NSCLC tumors, suggesting that this EMT-associated signature identified in vitro may also play a role in defining EGFR antagonist activity in primary neoplasms. It remains to be determined how the EGFR pathway and EGFR-dependent growth is mechanistically linked to EMT. Previous reports have implicated a “cross-talk” between the EGFR and E-cadherin, specifically, suggesting that a critical link exists (3941). Also, whether a general HER family pathway activity is associated with an epithelial signature is unclear. It is noteworthy that one of the erlotinib-insensitive cell lines (H1781) exhibiting an epithelial signature harbored a previously reported activating mutation in HER2 (42). However, regardless of the mechanistic basis, our data suggests that the underlying epithelial or mesenchymal phenotypic characteristics of a tumor may serve as a determinant of erlotinib activity.

The hypothesis that E-cadherin expression could serve as a predictor of erlotinib-mediated clinical activity was evaluated by performing E-cadherin immunohistochemistry on treatment-naïve biopsies from patients enrolled in TRIBUTE, a randomized, placebo-controlled clinical trial. In this trial, patients with locally advanced or metastatic NSCLC received either erlotinib administered concurrently with chemotherapy or chemotherapy alone but showed no clinical benefit to erlotinib in combination with chemotherapy (18). However, we observed a trend toward better outcome in all measures of clinical benefit (response, time to progression, and overall survival) upon comparison of the E-cadherin-positive group treated with erlotinib + chemotherapy versus chemotherapy alone. Although the difference for response and overall survival between both arms was not significant, the HR and median time difference for time to progression reached statistical significance (P = 0.003). The nonsignificant trend for improved overall survival in the E-cadherin-positive groupings is not necessarily at odds. This scenario is commonly seen in oncology trials and may be due to the effect of subsequent therapy at the time of progression, particularly because erlotinib treatment would have been stopped at this time. Interestingly, analysis of the small group of E-cadherin-negative patients in our study revealed a worsened response rate, time to progression, and overall survival within the patient group treated with erlotinib. Although these findings were not significant and therefore have to be treated with caution, this could potentially indicate that a patient with a mesenchymal subtype tumor may fare better with chemotherapy treatment alone.

Previous molecular subgroup analyses from TRIBUTE revealed that patients with EGFR mutations have a higher response rate, but no significant differences in time to progression and overall survival comparing treatment with chemotherapy alone to treatment with added erlotinib (8). In addition, a KRAS-mutated subgroup of patients seemed to derive a negative effect similar to that observed in our E-cadherin-negative group (8). Due to small sample sizes, we were not able to assess a possible correlation between E-cadherin expression and EGFR or KRAS status. Finally, it is important to bear in mind that the retrospective subgroup analyses described here should be viewed cautiously. Given the small numbers of patients within each subgroup, imbalances in patient characteristics that affect outcomes cannot be excluded. Furthermore, statistical adjustments for such imbalances are methodologically difficult due to the small numbers.

In conclusion, an EMT expression signature correlates with sensitivity of NSCLC cell lines to erlotinib, suggesting that patients with epithelial-type NSCLC tumors could have a better outcome after treatment with an erlotinib-containing regimen. Consistent with the preclinical hypothesis, this was supported by retrospective subgroup analyses in TRIBUTE. Patients with positive expression of the epithelial marker E-cadherin showed a significantly better time to progression when comparing erlotinib plus chemotherapy treatment to chemotherapy alone. As this clinical study was limited to the retrospective evaluation of a relatively small subset of patients, further evaluation in future studies is warranted.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Note: R.L. Yauch, T. Januario, and D. Eberhard contributed equally to this work.

We thank Patti Tobin for excellent technical assistance; Audrey Goddard, Mark Sliwkowski, and Paul Waring for editorial review of the manuscript, the investigators who provided tissue blocks and/or unstained sections from one or more of their patients enrolled on the TRIBUTE clinical trial, without their enthusiasm and hardwork this study would not have been possible, and most importantly, the patients who participated in the TRIBUTE clinical trial, and their families, particularly those who consented to the use of their archival tumor material for research.

1
Yarden Y, Sliwkowski MX. Untangling the ErbB signalling network.
Nat Rev Mol Cell Biol
2001
;
2
:
127
–37.
2
Hynes NE, Lane HA. ERBB receptors and cancer: the complexity of targeted inhibitors.
Nat Rev Cancer
2005
;
5
:
341
–54.
3
Akita RW, Sliwkowski MX. Preclinical studies with Erlotinib (Tarceva).
Semin Oncol
2003
;
30
:
15
–24.
4
Shepherd FA, Pereira J, Ciuleanu TE, et al. A randomized placebo-controlled trial of erlotinib in patients with advanced non-small cell lung cancer (NSCLC) following failure of 1st line or 2nd line chemotherapy. A National Cancer Institute of Canada Clinical Trials Group (NCIC CTG) trial.
J Clin Oncol
2004
;
22
:
7022
. Meeting Abstracts.
5
Pao W, Miller VA. Epidermal growth factor receptor mutations, small-molecule kinase inhibitors, and non-small-cell lung cancer: current knowledge and future directions.
J Clin Oncol
2005
;
23
:
2556
–68.
6
Lynch TJ, Bell DW, Sordella R, et al. Activating mutations in the epidermal growth factor receptor underlying responsiveness of non-small-cell lung cancer to gefitinib.
N Engl J Med
2004
;
350
:
2129
–39.
7
Paez JG, Janne PA, Lee JC, et al. EGFR mutations in lung cancer: correlation with clinical response to gefitinib therapy.
Science
2004
;
304
:
1497
–500.
8
Eberhard DA, Johnson BE, Amler LC, et al. Mutations in the epidermal growth factor receptor and in KRAS are predictive and prognostic indicators in patients with non-small cell lung cancer treated with chemotherapy alone and in combination with erlotinib.
J Clin Oncol
2005
;
23
:
5900
–9.
9
Giaccone G. Epidermal growth factor receptor inhibitors in the treatment of non-small-cell lung cancer.
J Clin Oncol
2005
;
23
:
3235
–42.
10
Tsao MS, Sakurada A, Lorimer I, et al. Molecular analysis of the epidermal growth factor receptor (EGFR) gene and protein expression in patients treated with erlotinib in National Cancer Institute of Canada Clinical Trials Group (NCIC CTG) trial BR.21.
J Clin Oncol
2005
;
23
:
7007
. Meeting Abstracts.
11
Salomon DS, Brandt R, Ciardiello F, Normanno N. Epidermal growth factor-related peptides and their receptors in human malignancies.
Crit Rev Oncol Hematol
1995
;
19
:
183
–232.
12
Hirata A, Hosoi F, Miyagawa M, et al. HER2 overexpression increases sensitivity to gefitinib, an epidermal growth factor receptor tyrosine kinase inhibitor, through inhibition of HER2/HER3 heterodimer formation in lung cancer cells.
Cancer Res
2005
;
65
:
4253
–60.
13
Engelman JA, Janne PA, Mermel C, et al. ErbB-3 mediates phosphoinositide 3-kinase activity in gefitinib-sensitive non-small cell lung cancer cell lines.
Proc Natl Acad Sci U S A
2005
;
102
:
3788
–93.
14
Kononen J, Bubendorf L, Kallioniemi A, et al. Tissue microarrays for high-throughput molecular profiling of tumor specimens.
Nat Med
1998
;
4
:
844
–7.
15
Corzo C, Tusquets I, Salido M, et al. Characterization of HER1 (c-erbB1) status in locally advanced breast cancer using fluorescence in situ hybridization and immunohistochemistry.
Tumour Biol
2005
;
26
:
25
–30.
16
Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysis and display of genome-wide expression patterns.
Proc Natl Acad Sci U S A
1998
;
95
:
14863
–8.
17
Hoffman EP, Awad T, Palma J, et al. Expression profiling-best practices for data generation and interpretation in clinical trials.
Nat Rev Genet
2004
;
5
:
229
–37.
18
Herbst RS, Prager D, Hermann R, et al. TRIBUTE—a phase III trial of erlotinib HCl (OSI-774) combined with carboplatin and paclitaxel (CP) chemotherapy in advanced non-small cell lung cancer (NSCLC).
J Clin Oncol
2004
;
22
:
7011
. Meeting Abstracts.
19
Janmaat ML, Kruyt FA, Rodriguez JA, Giaccone G. Response to epidermal growth factor receptor inhibitors in non-small cell lung cancer cells: limited antiproliferative effects and absence of apoptosis associated with persistent activity of extracellular signal-regulated kinase or Akt kinase pathways.
Clin Cancer Res
2003
;
9
:
2316
–26.
20
Hirsch FR, Witta S. Biomarkers for prediction of sensitivity to EGFR inhibitors in non-small cell lung cancer.
Curr Opin Oncol
2005
;
17
:
118
–22.
21
Pao W, Miller VA, Politi KA, et al. Acquired resistance of lung adenocarcinomas to gefitinib or erlotinib is associated with a second mutation in the EGFR kinase domain.
PLoS Med
2005
;
2
:
e73
.
22
Merlino GT, Xu YH, Ishii S, et al. Amplification and enhanced expression of the epidermal growth factor receptor gene in A431 human carcinoma cells.
Science
1984
;
224
:
417
–9.
23
Kang Y, Massague J. Epithelial-mesenchymal transitions: twist in development and metastasis.
Cell
2004
;
118
:
277
–9.
24
Guaita S, Puig I, Franci C, et al. Snail induction of epithelial to mesenchymal transition in tumor cells is accompanied by MUC1 repression and ZEB1 expression.
J Biol Chem
2002
;
277
:
39209
–16.
25
Hirai Y, Takebe K, Takashina M, Kobayashi S, Takeichi M. Epimorphin: a mesenchymal protein essential for epithelial morphogenesis.
Cell
1992
;
69
:
471
–81.
26
Hazan RB, Qiao R, Keren R, Badano I, Suyama K. Cadherin switch in tumor progression.
Ann N Y Acad Sci
2004
;
1014
:
155
–63.
27
Deeb G, Wang J, Ramnath N, et al. Altered E-cadherin and epidermal growth factor receptor expressions are associated with patient survival in lung cancer: a study utilizing high-density tissue microarray and immunohistochemistry.
Mod Pathol
2004
;
17
:
430
–9.
28
Bremnes RM, Veve R, Gabrielson E, et al. High-throughput tissue microarray analysis used to evaluate biology and prognostic significance of the E-cadherin pathway in non-small-cell lung cancer.
J Clin Oncol
2002
;
20
:
2417
–28.
29
Janne PA, Engelman JA, Johnson BE. Epidermal growth factor receptor mutations in non-small-cell lung cancer: implications for treatment and tumor biology.
J Clin Oncol
2005
;
23
:
3227
–34.
30
Kobayashi S, Boggon TJ, Dayaram T, et al. EGFR mutation and resistance of non-small-cell lung cancer to gefitinib.
N Engl J Med
2005
;
352
:
786
–92.
31
Tracy S, Mukohara T, Hansen M, Meyerson M, Johnson BE, Janne PA. Gefitinib induces apoptosis in the EGFRL858R non-small-cell lung cancer cell line H3255.
Cancer Res
2004
;
64
:
7241
–4.
32
Sordella R, Bell DW, Haber DA, Settleman J. Gefitinib-sensitizing EGFR mutations in lung cancer activate anti-apoptotic pathways.
Science
2004
;
305
:
1163
–7.
33
Moasser MM, Basso A, Averbuch SD, Rosen N. The tyrosine kinase inhibitor ZD1839 (“Iressa”) inhibits HER2-driven signaling and suppresses the growth of HER2-overexpressing tumor cells.
Cancer Res
2001
;
61
:
7184
–8.
34
Pao W, Wang TY, Riely GJ, et al. KRAS mutations and primary resistance of lung adenocarcinomas to gefitinib or erlotinib.
PLoS Med
2005
;
2
:
e17
.
35
Thompson EW, Newgreen DF. Carcinoma invasion and metastasis: a role for epithelial-mesenchymal transition?
Cancer Res
2005
;
65
:
5991
–5.
36
Tarin D. The fallacy of epithelial mesenchymal transition in neoplasia.
Cancer Res
2005
;
65
:
5996
–6001.
37
Kase S, Sugio K, Yamazaki K, Okamoto T, Yano T, Sugimachi K. Expression of E-cadherin and β-catenin in human non-small cell lung cancer and the clinical significance.
Clin Cancer Res
2000
;
6
:
4789
–96.
38
Ohira T, Gemmill RM, Ferguson K, et al. WNT7a induces E-cadherin in lung cancer cells.
Proc Natl Acad Sci U S A
2003
;
100
:
10429
–34.
39
Andl CD, Rustgi AK. No one-way street: cross-talk between e-cadherin and receptor tyrosine kinase (RTK) signaling: a mechanism to regulate RTK activity.
Cancer Biol Ther
2005
;
4
:
28
–31.
40
Fedor-Chaiken M, Hein PW, Stewart JC, Brackenbury R, Kinch MS. E-cadherin binding modulates EGF receptor activation.
Cell Commun Adhes
2003
;
10
:
105
–18.
41
Qian X, Karpova T, Sheppard AM, McNally J, Lowy DR. E-cadherin-mediated adhesion inhibits ligand-dependent activation of diverse receptor tyrosine kinases.
EMBO J
2004
;
23
:
1739
–48.
42
Shigematsu H, Takahashi T, Nomura M, et al. Somatic mutations of the HER2 kinase domain in lung adenocarcinomas.
Cancer Res
2005
;
65
:
1642
–6.