Abstract
Activating mutations in the EGF receptor (EGFR) are associated with clinical responsiveness to EGFR tyrosine kinase inhibitors (TKI), such as erlotinib and gefitinib. However, resistance eventually arises, often due to a second EGFR mutation, most commonly T790M. Through a genome-wide siRNA screen in a human lung cancer cell line and analyses of murine mutant EGFR-driven lung adenocarcinomas, we found that erlotinib resistance was associated with reduced expression of neurofibromin, the RAS GTPase-activating protein encoded by the NF1 gene. Erlotinib failed to fully inhibit RAS–ERK signaling when neurofibromin levels were reduced. Treatment of neurofibromin-deficient lung cancers with a MAP–ERK kinase (MEK) inhibitor restored sensitivity to erlotinib. Low levels of NF1 expression were associated with primary and acquired resistance of lung adenocarcinomas to EGFR TKIs in patients. These findings identify a subgroup of patients with EGFR-mutant lung adenocarcinoma who might benefit from combination therapy with EGFR and MEK inhibitors.
Significance: The emergence of resistance to EGFR TKIs is a major clinical challenge in the treatment of lung adenocarcinomas driven by mutations in EGFR. This study suggests that, in a subset of patients, resistance is caused by reduced neurofibromin expression, and that in these cases there may be clinical benefit to combining EGFR TKIs with MEK inhibitors. Cancer Discov; 4(5); 606–19. ©2014 AACR.
See related commentary by Maertens and Cichowski, p. 519
This article is highlighted in the In This Issue feature, p. 495
Introduction
Lung cancer is the most frequently diagnosed cancer and a leading cause of cancer-related mortality worldwide, accounting for nearly 1.4 million deaths per year (1, 2). Lung adenocarcinoma is the most common histologic subtype of lung cancer, with 10% to 40% displaying activating mutations in the EGF receptor gene (EGFR), occurring most frequently in never-smokers and in East Asian populations (3, 4). The presence of activating EGFR mutations strongly correlates with clinical responsiveness to the EGFR tyrosine kinase inhibitors (TKI) erlotinib and gefitinib. However, although these drugs are initially very effective, resistance eventually arises in almost all patients, resulting in a modest overall survival benefit (3).
Several groups have investigated the mechanisms that underlie resistance to EGFR TKIs. One of the first resistance mechanisms identified in tumors in treated patients was a secondary mutation (T790M) in the EGFR gene (3, 5, 6), which enhances the affinity of EGFR for ATP and reduces binding of the inhibitor (5, 7–9) and accounts for about 50% to 60% of resistant cases (5, 6, 10–12).
Additional mechanisms of resistance identified in experimental settings include activation of insulin—like growth factor-I receptor (IGF-IR), amplification of MET, HER2, or MAPK1, upregulation of the AXL receptor or its ligand, or activating mutations in PIK3CA (13–18). The majority of these genetic alterations have been confirmed in human EGFR-mutant TKI-resistant lung tumor samples, varying in frequency from 5% to 20%. Histologic transformation involved in resistance has also been observed in clinical samples, most prominently the conversion of EGFR inhibitor–sensitive lung adenocarcinomas to drug-resistant small cell lung cancer (SCLC), described in about 5% of cases of acquired resistance to EGFR TKI (10, 12, 19). Less frequently, epithelial-to-mesenchymal transition, potentially related to loss of MED12 or upregulation of AXL, has been reported to result in a broad treatment resistance, including resistance to EGFR TKIs (10, 18, 20, 21). Importantly, the mechanism of acquired resistance is still unknown for about one third of TKI-resistant lung adenocarcinomas (10, 20). In addition, it is evident that multiple mechanisms may contribute to resistance within one tumor (10, 15, 18, 22). Understanding the heterogeneity of molecular mechanisms involved in the evolution of resistance is therefore necessary to optimize the treatment of individual patients with mutant EGFR-driven tumors.
With the aim of improving therapy for EGFR-mutant TKI-resistant lung cancer, we have set out to identify previously unknown mechanisms of resistance to TKIs in this disease. We first performed an in vitro systematic genome-wide analysis to screen for genes whose silencing by siRNA confers resistance to the EGFR inhibitor erlotinib in a human lung cancer cell line that is sensitive to this drug due to the presence of an activating mutation in EGFR. We also took an in vivo approach using a mutant EGFR-driven mouse lung cancer model, analyzing gene expression in tumors associated with the acquisition of resistance to erlotinib. As described previously, these mice develop lung adenocarcinomas that are initially responsive to EGFR TKIs, but develop resistance following repeated cycles of treatment (23). Analysis of the erlotinib-resistant mouse tumors revealed the T790M mutation in about 20% of cases and occasional amplification of Met (23), suggesting that the erlotinib-resistant mouse tumors recapitulate the molecular changes identified in human lung tumors that acquire resistance to EGFR TKIs.
One gene emerging from these two approaches is the negative regulator of RAS proteins, NF1. We show that reduced expression of neurofibromin, the RAS GTPase-activating protein (GAP) product of this gene, is associated with decreased sensitivity of human lung cancer cells to EGFR inhibitory drugs, due presumably to enhanced RAS signaling. Treatment of EGFR-mutant lung cancer cells expressing low levels of neurofibromin with inhibitors of MAP–ERK kinase (MEK), a RAS effector pathway component, restores their sensitivity to EGFR inhibitors. Moreover, the majority of erlotinib-resistant EGFR-mutant mouse lung tumors that do not express the T790M mutation respond to cotreatment with a MEK inhibitor. Finally, we observed reduced NF1 expression in two independent datasets of paired pre- and posttreatment lung adenocarcinomas that acquired resistance to EGFR TKI treatment. We also found that low levels of NF1 expression in pretreatment clinical specimens correlate with poor overall survival in patients with EGFR-mutant lung cancer treated with EGFR TKIs.
Collectively, our data identified low neurofibromin expression as a novel mechanism by which tumor cells are intrinsically less sensitive or acquire resistance to EGFR TKIs and provide a rationale for using drugs targeting MEK in combination with EGFR inhibitors as a therapeutic approach for the treatment of T790M-negative EGFR TKI–resistant lung cancer.
Results
Genome-Wide siRNA Screen Identifies Determinants of Erlotinib Resistance
To identify novel determinants of resistance to the EGFR TKI erlotinib, we performed a genome-wide RNA interference screen examining cell viability in the absence or presence of the drug. We transfected the EGFR-mutant human lung adenocarcinoma–derived PC9 cell line that is exquisitely sensitive to EGFR TKIs with a library of siRNA pools targeting approximately 21,000 unique human transcripts. Forty-eight hours after transfection, culture medium was replaced, and cells were incubated for an additional 72 hours, in the presence or absence of erlotinib (Fig. 1A). The experiment was performed in triplicate for both conditions. We used an erlotinib concentration slightly above the IC50 determined for PC9 cells, favoring identification of siRNAs that confer resistance to the drug, but still allowing detection of siRNAs that enhance killing in the same screen.
The effect of the individual siRNA pools on cell survival was analyzed in drug-treated versus untreated conditions (Supplementary Table S1), and we selected siRNAs that showed a substantial differential effect (residual Z score ≥ 2.0 or ≤ −2.0) without killing the untreated cells (control Z score ≥ −2.0; the Z score corresponds to how many SDs away from the mean of the population an individual siRNA lies; see Supplementary Experimental Procedures). Using these criteria, we identified 242 siRNA pools, of which 212 enhanced and 30 decreased cell survival in the presence of erlotinib (Supplementary Table S2). To test the reproducibility of our findings, we repeated the genome-wide siRNA screen in an independent experiment. Ranking the siRNA pools based on their residual Z scores from this repeat screen revealed that 106 of the previously identified 242 siRNA pools were in the top 5% of the repeat screen list. Almost all of these 106 siRNA pools enhanced survival in the presence of erlotinib (Supplementary Fig. S1 and Supplementary Table S2). The validated 106 siRNA pools were then taken forward to a deconvolution screen, where the four individual siRNA oligos targeting each gene were analyzed separately.
We established the influence of each individual siRNA-induced silencing on erlotinib sensitivity by determining the sensitivity index that takes into account the individual effects of erlotinib and siRNA-induced knockdown on cell viability (24). Using a cutoff value of ≥ 0.10 for sensitizing siRNAs and ≤ − 0.10 for desensitizing siRNAs, for 23 siRNA pools a reproducible effect of at least two of four of their individual deconvoluted siRNAs could be found (Fig. 1B).
Nf1 Downregulation in a Mouse Model of Erlotinib-Resistant EGFR-Mutant Lung Cancer
One of these genes, NF1, stood out, as its gene product neurofibromin has a recognized negative regulatory role in signaling downstream of EGFR due to its function as a RAS GAP (25, 26), suggesting a possible mechanistic rationale for its association with acquisition of resistance to EGFR inhibitory drugs. To assess the possible relevance of these in vitro results on resistance to EGFR inhibitory drugs in vivo, we made use of an inducible mouse model of EGFR-driven lung cancer (23, 27). In this model, expression of mutant EGFR leads to the development of lung adenocarcinomas that are sensitive to erlotinib treatment. Long-term intermittent treatment of these mice with erlotinib, however, leads to the outgrowth of resistant tumors. To assess whether the genes identified in our siRNA screen showed altered expression between untreated (erlotinib-sensitive) tumors and erlotinib-resistant tumors, we compared the expression levels of Nf1 in erlotinib-resistant tumors and corresponding adjacent normal lung using quantitative real-time PCR (qRT-PCR) analysis. These experiments revealed a decrease in Nf1 mRNA levels compared with normal lung in 10 of 18 erlotinib-resistant tumors, of which seven showed a more than 2-fold decrease (Fig. 2A). Comparable results were obtained with two additional qRT-PCR assays using different primer sets (Supplementary Fig. S2). Interestingly, tumors bearing the EGFRT790M gatekeeper mutation, Kras mutations, or Met amplification did not show decreased Nf1 expression, suggesting that loss of neurofibromin could be selected for by EGFR TKIs in the absence of other mechanisms of resistance. Gene expression profiling of erlotinib-sensitive and erlotinib-resistant tumors from these mice failed to show significant differential expression of any of the other genes emerging from the siRNA screen of PC9 cells listed in Fig. 1B.
Because Nf1 is a known tumor suppressor, neurofibromin expression might be decreased during tumor progression independent of erlotinib treatment. To directly address this possibility, one would ideally compare erlotinib-sensitive and erlotinib-resistant tumors derived from the same animal. Regrettably, such material is not available. Instead, we analyzed the relative amounts of Nf1 mRNA in EGFR-induced lung tumors and adjacent normal lung from untreated and erlotinib-treated animals. This analysis showed no significant differences in Nf1 expression between normal lung samples and untreated tumors, whereas erlotinib-resistant tumors with an unknown resistance mechanism did express significantly lower levels of Nf1 compared with untreated (erlotinib-sensitive) tumors (Fig. 2B). In an independent set of tumors, we evaluated NF1 protein expression in erlotinib-treated tumors versus untreated tumors and confirmed lower NF1 protein expression in a subset of erlotinib-resistant tumors relative to those that had not been exposed to the drug (Fig. 2C).
Overall, these observations with mouse lung tumors confirm our data with human lung cancer cell lines, showing that low Nf1 expression is associated with erlotinib resistance in EGFR-driven lung tumors that lack known resistance mechanisms such as the EGFRT790M mutation or Met amplification.
Reduced NF1 Expression Confers Resistance of Lung Cancer Cell Lines to Erlotinib
To validate the possible role of neurofibromin in erlotinib resistance, we introduced into PC9 cells two individual short hairpin RNAs (shRNA; #1 and #2) targeting different non-overlapping regions of the NF1 coding sequence, which are distinct from the previously used siRNA sequences. A nonsilencing scrambled shRNA (shSC) was used as a control throughout the study. Both NF1-targeting shRNA constructs decreased sensitivity to erlotinib, with a 26-fold increase in the drug concentration required for an absolute survival inhibition of 50% (IC50) for shNF1#1 and a 56-fold increase for shNF1#2 (Fig. 3A) as determined by CellTiter Blue measurement. The shRNA constructs efficiently suppressed neurofibromin mRNA and protein expression, with shNF1#2 having stronger effects. While neurofibromin silencing conferred resistance of PC9 cells to erlotinib or another EGFR kinase inhibitor, gefitinib, the response to chemotherapeutic agents, such as cisplatin or docetaxel, was not affected (Fig. 3B and Supplementary Fig. S3A).
To test longer-term effects of NF1 silencing, we performed colony formation assays in which cells were cultured in the presence of erlotinib for 10 days. NF1 silencing substantially enhanced survival in these assays (Fig. 3C). Prolonged treatment for 4 weeks in a competition assay, in which unlabeled parental PC9 cells were mixed in a ratio of 100:1 with GFP-labeled cells that expressed an shRNA targeting NF1 or a nonsilencing control shRNA, showed substantial outgrowth of the shNF1 cells (Fig. 3D). The percentage of GFP-positive cells remained approximately 1% in the absence of erlotinib, suggesting that NF1 silencing has little or no effect on the basal proliferation of PC9 cells. Selective outgrowth of NF1 knockdown cells was also seen in long-term assays with other lung adenocarcinoma cell lines harboring activating EGFR mutations that are sensitive to erlotinib treatment (Fig. 3E and Supplementary Figs. S3B–S3D). Thus, our data provide evidence to suggest that NF1 silencing reduces the sensitivity of lung adenocarcinoma cells to erlotinib-induced cell death and/or growth arrest.
To exclude off-target effects of the shRNAs, cells were stably transfected with the GAP-related domain (GRD) of neurofibromin or a control empty vector before the shRNA infections. Although NF1 silencing desensitized the control cells to erlotinib treatment, it did not affect erlotinib sensitivity of the neurofibromin GRD-expressing cells, in which the GRD was not targeted by the shNF1 constructs (Supplementary Fig. S3E). In fact, expression of neurofibromin GRD slightly increased erlotinib sensitivity (Fig. 3F). These data confirm a role for neurofibromin in erlotinib response in lung adenocarcinoma cells. Moreover, the observation that the GRD of neurofibromin could restore sensitivity suggests that neurofibromin influences this sensitivity through its function as a negative regulator of RAS proteins, rather than through RAS-independent pathways.
NF1 Silencing Activates the MAPK Pathway in the Presence of Erlotinib
To investigate whether NF1 silencing promotes RAS activation in PC9 cells, we analyzed the amounts of active, GTP-bound RAS in the absence and presence of erlotinib. Although erlotinib reduced the amount of active RAS in both control and NF1-knockdown cells, cells retain substantially higher levels of active RAS in the presence of erlotinib when NF1 expression is reduced (Fig. 4A).
To determine whether this increased RAS activity affects downstream signaling pathways, we examined the phosphorylation status of several downstream signaling proteins (Fig. 4B). Erlotinib reduced the phosphorylation of EGFR and AKT similarly in all cells. Importantly, although erlotinib completely abolished ERK phosphorylation in the parental and control shRNA-infected cells, remaining phosphorylated ERK (pERK) could also still be detected in shNF1 cells (Fig. 4B and Supplementary Fig. S4A) at 1 μmol/L (Supplementary Fig. S4B), which is around the steady-state plasma concentration found in patients treated with erlotinib (28, 29).
Examination of cells expressing the GRD of NF1 revealed that ERK phosphorylation was also completely abolished in these cells in the presence of erlotinib (Fig. 4C). We confirmed an important role for the active MAPK pathway by studying PC9 cells expressing constitutively active forms of MEK (MEK-DD) or AKT [myristylated-AKT (myr-AKT); Supplementary Fig. S4C]. MEK-DD caused a strong decrease in erlotinib sensitivity, whereas the effect of myr-AKT in cell survival assays is more moderate (Fig. 4D and E). The ability of MEK-DD–expressing cells to resist the inhibitory effects of erlotinib can be reversed by treatment with the MEK inhibitor AZD-6244 (Supplementary Fig. S4D). Thus, neurofibromin downregulation increases RAS activity and attenuates the effect of erlotinib on the downstream MAPK pathway, thereby decreasing the sensitivity to EGFR inhibitory drugs.
Cells Expressing Reduced Neurofibromin Respond to Erlotinib in Combination with a MEK Inhibitor
Given the likely importance of the residual ERK phosphorylation for cell survival in the presence of erlotinib, we reasoned that EGFR-mutant cells expressing low neurofibromin levels could be sensitive to treatment with a MEK inhibitor combined with a TKI. Indeed, analyzing a dose–response curve in the presence or absence of erlotinib revealed that the NF1- knockdown cells are resistant to the MEK inhibitor AZD-6244 or erlotinib alone, but do respond to the combination of both inhibitors (Fig. 5A). As expected, the control shSC cells did respond to single-agent erlotinib, as evidenced by the decreased survival at the beginning of the experiment. Similar results were obtained with the MEK inhibitors CI-1040 and PD0325901 (Supplementary Fig. S5A). Although EGFR or MEK inhibition alone is insufficient, these drugs abolished ERK phosphorylation completely in the shNF1 cells when used in combination (Fig. 5B). Similar assays were performed with a clone of PC9 cells that had acquired erlotinib resistance in vitro following prolonged drug exposure that resulted in the emergence of cells with the EGFR gatekeeper mutation T790M (PC9T790M; Supplementary Fig. S5B). These cells could not be resensitized to erlotinib by the addition of a MEK inhibitor (Fig. 5C), and maintain low levels of pERK in the presence of both drugs, in contrast to parental PC9 cells (Fig. 5D).
Although the above studies revealed that MEK inhibition in combination with erlotinib resensitizes erlotinib-resistant shNF1-infected cells in vitro, we continued to examine the sensitizing effect of MEK inhibition to erlotinib in vivo using tumor xenografts. We established xenografts of PC9 cells stably infected with a control (shSC) or NF1-targeting (shNF1#2) shRNA. Once tumors were detectable, mice were treated with either erlotinib or AZD-6244, or a combination of both drugs. Combined erlotinib and AZD-6244 treatment for 30 days indeed effectively reduced the tumor growth of both the NF1-silenced and the control tumors. As expected, we did not see a response to erlotinib in the NF1 knockdown tumors, whereas the control tumors did clearly respond (Fig. 5E). In contrast, xenograft tumors of PC9T790M cells failed to respond to the combination treatment (Fig. 5F). These data suggest that modest doses of a MEK inhibitor may resensitize tumors with reduced neurofibromin expression to erlotinib in the absence of the T790M mutation.
Erlotinib-Resistant Mouse Lung Adenocarcinomas Respond to Combined EGFR and MEK Inhibition
Although the above findings suggest a possible treatment opportunity for lung adenocarcinoma cells that express reduced levels of neurofibromin, the effect of combination EGFR and MEK inhibition in established EGFR-driven lung adenocarcinomas that developed resistance to erlotinib remained unknown. We therefore used our tetracycline-inducible mouse model of EGFR-dependent lung cancer, driven by the EGFRL858R point mutation, to generate erlotinib-resistant tumors (23). While maintained on a diet containing doxycycline to ensure continued expression of the transgene, the mice were treated with erlotinib for three treatment rounds of 4 weeks each, followed by 4 weeks without drug treatment. Tumors were monitored using micro-CT at the start and end of each treatment period. In a few mice, the tumor response was relatively minor during the second round of erlotinib, and these mice continued on erlotinib (Fig. 6A, M10, 11, 12, 15).
As described previously, a diminished response during the third round of erlotinib treatment was commonly observed (23). When possible, we maintained the mice on erlotinib for another 4-week round of treatment after the third round, to confirm that the tumors were indeed growing and resistant to long-term erlotinib treatment. Subsequent to the emergence of erlotinib-resistant tumors, mice were treated with erlotinib and the MEK inhibitor GSK1120212, also known as trametinib, for another 4 weeks, or as control with either inhibitor alone (Fig. 6A). We scanned the animals before and after these final 4 weeks of treatment and quantified the tumor volumes (Fig. 6B and Supplementary Fig. S6). As seen in Fig. 6B and C, the combined treatment of EGFR and MEK inhibitors had a striking effect on most erlotinib-resistant mouse lung adenocarcinomas. On the contrary, erlotinib or MEK inhibitor alone failed to induce significant tumor regression, confirming the resistant nature of these tumors (Fig. 6C). Importantly, combined erlotinib and MEK inhibitor treatment was well tolerated, and mice did not show any sign of weight loss (data not shown).
To assess whether secondary mutations in EGFR were associated with a reduced response to the combined erlotinib and MEK inhibitor treatment, we generated cDNA from RNA that was extracted from individual tumor nodules, and sequenced part of the human EGFR transgene cDNA spanning the transgene (L858R) and T790M region, as described previously (23). We detected the T790M amino acid substitution in two tumors and relatively high Met expression in two other tumors; three of these tumors were treated with combined erlotinib and MEK inhibitor, and these tumors showed a minimal response (Fig. 6C and Supplementary Table S3). As expected, all tumors did express the L858R driver mutation (data not shown). In addition, we evaluated Nf1 mRNA expression in these tumors and found that tumors responding well to erlotinib combined with MEK inhibitor more often express relatively lower levels of Nf1, but some tumors with higher levels do also respond (Supplementary Table S3), indicating that these tumors might have found alternative mechanisms to activate MEK (16). We furthermore confirmed that all tumors expressed wild-type Kras, Hras, and Nras (codons 12, 13, and 61; data not shown).
Collectively, these results demonstrate that erlotinib combined with a licensed MEK inhibitor substantially affects T790M-negative erlotinib-resistant lung adenocarcinomas in mouse models.
Reduced NF1 Expression in Human Lung Adenocarcinoma Samples with Resistance to EGFR Inhibitors
To assess the clinical relevance of this resistance mechanism, we evaluated NF1 expression in 13 human EGFR-mutant lung adenocarcinoma samples that acquired resistance to EGFR TKI treatment compared with matched pretreatment samples (Fig. 7A and B). We were able to extract sufficient RNA from 10 sample pairs to perform NF1 mRNA expression analyses by qRT-PCR (Supplementary Table S4 provides tumor purity and treatment response for each sample presented in Fig. 7A). Four of the erlotinib-resistant samples showed a >2-fold decrease in NF1 expression compared with their matched pretreatment sample, with relatively stable expression in the remaining six pairs (Fig. 7A). Importantly, one of the erlotinib-resistant tumors with reduced NF1 did not harbor an EGFRT790M mutation nor amplified MET (sample pair Y10). In addition, using RNA-sequencing (RNA-seq) data from a separate set of three human lung adenocarcinomas treated with EGFR TKIs (erlotinib or gefitinib), we identified reduced NF1 expression in all three posttreatment tumors, with the strongest reduction in the two EGFR TKI–resistant samples that did not express T790M (V1 and V3) and failed to respond to a second-line treatment with afatinib (Fig. 7B). Experiments with the PC9 cell line confirmed that NF1 silencing strongly reduced sensitivity to afatinib as well (Supplementary Fig. S7A). Together, these two clinical datasets indicate that low NF1 might be driving erlotinib resistance in these tumors.
Some posttreatment samples showed reduced NF1 mRNA in the posttreatment tumor co-occurring with an EGFRT790M mutation (sample pairs Y1, Y6, and Y9). Similar observations have been published for other resistance mechanisms (10, 18, 22), which could indicate that multiple mechanisms may contribute to resistance to EGFR TKIs. To address whether heterogeneity in resistance mechanisms could explain our observations, we assessed the abundance of the T790M mutation by pyrosequencing and compared this with the abundance of the original TKI-sensitivity–conferring EGFR driver mutation. In the posttreatment tumor of Y1, 28% of the EGFR present in the posttreatment sample harbored the exon19 deletion, whereas only 7% harbored the T790M mutation. Because a minority of cells in the resistant tumor cells expressed T790M, one can speculate that heterogeneity in resistance might occur in this sample with T790M expression in part of the tumor cells and low NF1 in other cells. A similar heterogeneity is seen in the posttreatment sample of Y9. However, one sample (pair Y6) showed an almost similar abundance of the driver mutation and T790M in the posttreatment sample, 28% and 21%, respectively. We therefore hypothesized that low NF1 and expression of T790M might have additive effects on drug resistance. Indeed, long-term treatment of erlotinib-resistant PC9T790M cells infected with an shRNA targeting NF1 (shNF1) or a nonsilencing control shRNA (shSC) resulted in selective outgrowth of the shNF1 cells when cultured in the presence of erlotinib or afatinib (Supplementary Fig. S7B). These data suggest that NF1 silencing and T790M have additive effects on the resistance of lung adenocarcinoma cells to EGFR inhibitor-induced death. Combined, the RNA-seq and the qRT-PCR analyses on human samples confirm a clinical relevance of NF1 downregulation in acquired resistance to EGFR TKIs.
To evaluate whether NF1 might be involved in primary resistance as well, we examined NF1 expression in a cohort of 34 NSCLC samples taken at diagnosis from patients who were then treated with erlotinib as first (n = 5) or second (n = 29) line of therapy; the EGFR mutation status of these tumors is unknown. Using the median as cutoff, we found that low NF1 expression was strongly associated with decreased overall survival with a median survival time of 7.6 months [95% confidence interval (CI), 6.8–8.4] compared with 19.1 months (95% CI, 14.0–24.2) for patients with high NF1 expression in their tumor (Fig. 7C). A multivariate Cox regression analysis with NF1 expression, gender, and morphology as input variables confirmed NF1 expression as an independent prognostic factor with a relative risk of 4.1 (95% CI, 1.6–10.7; P = 0.004).
Overall, these clinical data suggest that the level of NF1 expression can determine the responsiveness to EGFR TKI and provide a rationale for testing MEK inhibitors in combination with EGFR TKIs in patients with EGFR-mutant lung cancer.
Discussion
We have used a functional genomic in vitro screening approach, together with a genetically modified mouse lung cancer model system, to investigate mechanisms of acquired resistance to erlotinib in EGFR-mutant lung adenocarcinoma, identifying NF1 as a gene whose loss of function is capable of causing EGFR TKI drug resistance in both settings. Our analyses of two independent sets of paired EGFR-mutant lung adenocarcinoma samples from patients treated with EGFR inhibitors confirm that downregulation of NF1 expression at the time of TKI resistance is a common occurrence in the clinic. Furthermore, NF1 levels may influence the initial response to TKIs, as low neurofibromin in pretreatment specimens is strongly associated with reduced overall survival for patients with EGFR-mutant lung adenocarcinomas treated with EGFR TKI.
Several studies have shown that neurofibromin expression and function can be altered at a number of levels in lung and other cancers. Sequence analysis of the NF1 gene in 188 lung adenocarcinomas revealed mutations in nearly 10% of tumors, most of which lacked coincident KRAS mutations (30). A detailed genomic analysis of human glioblastoma by the Cancer Genome Atlas Research Network showed heterozygous deletion of the NF1 gene resulting in reduced neurofibromin expression, but also low NF1 mRNA without evidence of genomic alterations (31). Our RNA-seq data did not reveal mutations in NF1 in the resistant samples of the 3 patients analyzed (data not shown). We have also sequenced the DNA coding region of NF1 in 10 additional paired samples from patients with mutant EGFR-expressing lung tumors before and after the acquisition of resistance to EGFR inhibitor that do not express EGFRT790M, and found no evidence for somatic mutations in these samples either (data not shown). In addition, four of these matched biopsies had sufficient DNA material to analyze the methylation state of the NF1 promoter region, but we found no evidence for differences (data not shown). In our study, 6 out of a total of 13 paired samples show significant downregulation of NF1 mRNA upon TKI resistance. Future studies should evaluate neurofibromin expression at protein level, as additional posttranscriptional regulation could occur as described in glioma cells (32, 33). We have screened multiple antibodies for immunohistochemistry, but none showed the specificity required to detect neurofibromin in human tissue (data not shown).
Some of the human lung cancer samples analyzed in our study showed reduced NF1 expression co-occurring with the T790M mutation, suggesting that multiple mechanisms may contribute to resistance to EGFR TKIs. Such heterogeneity in mechanisms of resistance is consistent with other studies that have described T790M co-occurring with other resistance mechanisms (10, 18, 22). The abundance of transcripts containing the T790M mutation in comparison with those containing only the EGFR driver mutation in these resistant tumors suggests that heterogeneity in resistance mechanisms can be either in different cells of a tumor (Fig. 6, samples Y1 and Y9) or in an additive manner within the same cells (Fig. 6, sample Y6; Supplementary Fig. S7).
We provide several lines of evidence that reduced neurofibromin levels in lung adenocarcinoma cells affect drug response through activating RAS and the downstream ERK–MAPK pathway. First, we show increased levels of GTP-bound RAS in cells treated with erlotinib in which NF1 expression has been reduced by RNA interference. Second, pERK could be detected in NF1-silenced cells in the presence of erlotinib. Although reduced, such low levels of continued flux through ERK seemed sufficient to maintain cell survival in the presence of erlotinib, as inhibition of the upstream MEK kinase with a small-molecule inhibitor abolished ERK phosphorylation completely and resulted in cell death. Furthermore, xenograft tumors of shNF1 cells failed to grow when the animals were treated with erlotinib in combination with a MEK inhibitor, whereas either drug on its own had no effect. In line with our findings, the MAPK pathway is directly activated by the point mutation of KRAS in about 20% of human lung tumors (34), which is associated with primary resistance to drugs targeting EGFR in colon and lung cancer (35, 36). In addition, a recently published study identified MAPK1 amplification in 5% of erlotinib-resistant samples (16), also indicating a prominent role for active MAPK signaling as a mediator of EGFR TKI resistance. Research on acquired resistance to EGFR inhibitory drugs in lung cancer has mainly focused on the PI3K–AKT pathway, activated by MET amplification or PTEN loss of function (14, 15, 37). In line with these studies, the introduction of an active AKT construct into PC9 cells indeed reduced erlotinib sensitivity (Fig. 4D). An active MEK construct showed, however, a more pronounced effect on drug resistance, providing evidence that the MAPK pathway can play a crucial role in resistance to EGFR TKIs in lung adenocarcinomas.
Importantly, our in vivo studies show that more than half (59%) of the EGFRL858R-driven mouse tumors that acquired resistance to erlotinib did respond to combination treatment with EGFR and MEK inhibitory drugs. At the moment, treatment options for tumors with acquired resistance to EGFR inhibitors are limited (3, 10, 20). Our experiments provide a molecular basis for the combination of EGFR and MEK inhibitors for the treatment of patients with mutant EGFR-driven tumors that have acquired resistance to TKI but lack the T790M gatekeeper mutation. A recent clinical trial showed improved progression-free survival in patients with BRAF-mutant melanoma when a BRAF inhibitor is combined with the MEK inhibitor trametinib (38), which we used in our studies in mouse models. Interestingly, resistance to BRAF kinase inhibitors is often associated with reactivation of the MAPK pathway (39–41), and two recent studies found NF1 mutations to be associated with reduced response to BRAF inhibition in BRAF-mutant melanoma (42, 43). Given the effect of EGFR inhibitors on EGFR-mutant lung cancer, the combination of MEK and EGFR inhibition in the setting of EGFR-mutant disease may be very effective. A combination of MEK and EGFR inhibitors is currently being evaluated in clinical trials for lung cancer (NCI-10-C-0218 and NCT01192165). In these trials, increased toxicity has been reported in patients who received both inhibitors (28, 29). Nevertheless, it is expected that novel EGFR inhibitors that specifically inhibit the mutant forms of EGFR while sparing the wild-type protein (44) will reduce such toxicity.
Collectively, the present study provides a molecular basis for the combination of EGFR TKI with clinically available MEK inhibitors for the treatment of patients with mutant EGFR-driven lung adenocarcinomas that fail to respond to EGFR TKIs due to either primary or acquired resistance.
Methods
Cell Culture
H3255, HCC827, and HCC4006 cells were purchased from the American Type Culture Collection Cell Biology collection, and PC9 cells were kindly provided by Jeff Settleman (Massachusetts General Hospital Cancer Center and Harvard Medical School). Cell lines were authenticated by the Cancer Research UK Central Cell Services facility using short-tandem repeat profiling. Cells were cultured in RPMI, supplemented with 10% FBS, 100 μg/mL streptomycin, and 100 U/mL penicillin at 37°C and 10% CO2. PC9T790M cells were generated by continued culturing of PC9 cells with 1 μmol/L erlotinib. Surviving clones were picked and cultured in the presence of 1 μmol/L erlotinib for 3 months, and thereafter without erlotinib.
Genome-Wide siRNA Screen
The genome-wide siRNA library (21,121 siRNAs) was purchased from Dharmacon as the siGENOME SMARTpool collection. siRNA pools targeting PLK1 or UBB1 were used as positive controls for the transfection, and siGENOME RISC-Free Control siRNA, siGENOME Non-Targeting siRNA Pool #2 (scrambled), and siRNA pools targeting Luciferase GL3 were used as negative controls.
PC9 cells were seeded in 384-well plates (500 cells per well) and reverse transfected with 37.5 nmol/L siRNA using DharmaFECT transfection reagent #2. After 48 hours, the culture medium was replaced by RPMI with 30 nmol/L erlotinib (drug screen) or without erlotinib (control screen). As a control for drug treatment, we left two columns of the control siRNAs untreated in the erlotinib-treated plates. Cells were incubated for 72 hours before being fixed with 80% ethanol, stained with 1 μg/mL of DAPI (Roche) for 1.5 hours, and stored in PBS at 4°C. The number of cells in each well was quantified using an Acumen Explorer microplate cytometer (TTP LabTech). We performed triplicate experiments for both conditions: control and drug screens. We performed an independent genome-wide screen using the same conditions to validate the reproducibility of the hits.
The follow-up deconvolution siRNA screen was performed using the Dharmacon collection siGENOME set of four individual siRNAs targeting a single gene. The deconvolution siRNA screen was performed using the same conditions as in the genome-wide screen.
Details of data analysis of the genome-wide and deconvolution siRNA screens are given in Supplementary Experimental Procedures.
qRT-PCR Paired Human NSCLC Samples
Samples for qRT-PCR analysis were obtained from patients with EGFR-mutant lung adenocarcinoma who developed acquired resistance to erlotinib under Human Investigations Protocol #111000928 (Yale Cancer Center, New Haven, CT). Pre- and posttreatment specimens were reviewed by a pathologist to ensure adequate tumor content. Tumor areas were circled, and manual microdissection was performed to enrich for tumor content in downstream applications. Pyrosequencing was used to determine the abundance of the EGFR L858R and exon 19 DEL mutations as well as the T790M mutation using the EGFR Pyro Kit (Qiagen).
Total RNA was isolated from formalin-fixed paraffin-embedded (FFPE) tissue (5 μm × 5 μm slides) using the FFPE RNeasy Kit (Qiagen). qRT-PCR assays were performed in triplicate with 4.5-ng RNA input per reaction, using the TaqMan RNA-to-Ct 1 Step Kit, and TaqMan primers to amplify NF1 and the housekeeping genes ACTB, ESD, and POLR2A, described to be relatively stably expressed in human NSCLC (45). NF1 mRNA levels were analyzed using the comparative CT method and normalized to the average of the three housekeeping genes for individual samples, before the fold change relative to the pretreatment sample was determined.
RNA-seq Analysis of Human NSCLC Samples
RNA sequencing was performed as previously described (21), using RNA extracted from FFPE material, with at least 50% tumor content (areas were circled and manual microdissection was performed to enrich for tumor content), from paired pre- and posttreatment biopsies of 3 patients with NSCLC. These patients were treated with either erlotinib or gefitinib followed by afatinib at the Department of Pulmonary Diseases, VU University Medical Center (Amsterdam, the Netherlands). Tumor biopsies were obtained as part of routine medical care after informed consent had been obtained according to the local ethical committee regulations. Tumor EGFR mutational status was determined as previously described (46).
Overall Survival Analysis of NSCLC Patients Treated with EGFR TKI
The samples analyzed were retrieved from the diagnostic biobank of the Pathology Service from the Hospital Universitario Marques de Valdecilla (Santander, Spain). For this study, primary tumors were obtained from 34 patients with stage IV non–small cell lung cancer (NSCLC) at the time of diagnosis. None of the patients received chemotherapy or radiotherapy before sampling, and patients received EGFR TKI as primary (n = 5) or secondary (n = 29) treatment upon diagnosis. The presence of an EGFR mutation was not an inclusion criterion, and EGFR mutation status is therefore not available. The study was approved by the local ethics committee.
Total RNA was extracted using the FFPE RNeasy Kit (Qiagen). qRT-PCR assays were performed in triplicate using the TaqMan RNA-to-Ct 1 Step Kit and TaqMan primers to amplify NF1 and the housekeeping gene ESD. NF1 mRNA levels were normalized to ESD. Patients were divided into two groups, low and high NF1, using the median NF1 expression as the cutoff value. Kaplan–Meier analysis was carried out using SPSS Statistics for the probability of survival (overall survival) in both groups starting from the time of diagnosis.
Disclosure of Potential Conflicts of Interest
K. Politi has ownership interest (including patents) in Molecular MD and is a consultant/advisory board member of Takeda. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: E.C. de Bruin, S. Gettinger, D.A.M. Heideman, J. Gómez-Román, A. García-Castaño, Y. Gong, M. Ladanyi, H. Varmus, K. Politi, J. Downward
Development of methodology: E.C. de Bruin, M. Jiang
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.C. de Bruin, C. Cowell, M. Jiang, M.A. Melnick, S. Gettinger, Z. Walther, A. Wurtz, G.J. Heynen, R. Bernards, E.F. Smit, K. Politi
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.C. de Bruin, C. Cowell, R.E. Saunders, S. Gettinger, Z. Walther, G.J. Heynen, K. Politi
Writing, review, and/or revision of the manuscript: E.C. de Bruin, M.A. Melnick, S. Gettinger, G.J. Heynen, D.A.M. Heideman, J. Gómez-Román, A. García-Castaño, Y. Gong, M. Ladanyi, H. Varmus, E.F. Smit, K. Politi, J. Downward
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.C. de Bruin, P.H. Warne
Study supervision: E.C. de Bruin, D.A.M. Heideman, J. Gómez-Román, A. García-Castaño, Y. Gong, M. Ladanyi, H. Varmus, K. Politi, J. Downward
Acknowledgments
The authors thank Rachael Instrell and Michael Howell from the LRI High Throughput Screening laboratory for their support in performing the screen; Stuart Horswell from the Bioinformatics and Biostatistics department at the London Research Institute for analyzing the xenograft tumor measurements; Francois Lassailly from the In Vivo Imaging department for assistance with the micro-CT imaging; Francesco Mauri from the Department of Histopathology at Hammersmith Hospital, London, for evaluating NF1 antibodies for immunohistochemistry; Aldona K. Fonfara from the Division of Molecular Carcinogenesis at the Netherlands Cancer Institute for help with RNA extractions; the members of the Netherlands Cancer Institute Genomics Core Facility for the RNA-seq processing and analysis; and Don Nguyen from the Department of Pathology and Yale Cancer Center for analyzing mouse gene expression data. The authors are grateful for the support given by the Equipment Park, Biological Resources, and the FACS Laboratory at the London Research Institute.
Grant Support
E.C. de Bruin was financially supported by a KWF (Dutch Cancer Society) fellowship and has received funding from the European Commission's Seventh Framework Programme (FP7/2007-2013) under the grant agreement Lungtarget (project 259770). This work was supported by grants R00 CA131488 (to K. Politi), R01 CA120247 (to K. Politi), and P01 CA129243 (to M. Ladanyi) from the National Cancer Institute, NIH; a Pilot grant from the Section of Medical Oncology, Yale University School of Medicine (to S. Gettinger and K. Politi); and was financially supported by Cancer Research UK.