Acquired Resistance to the Mutant-Selective EGFR Inhibitor AZD9291 Is Associated with Increased Dependence on RAS Signaling in Preclinical Models Free

Tumors containing activating EGFR mutations (EGFRm; e.g., deletion in exon 19 or an L858R point mutation) account for about 20% of advanced non–small cell lung cancer (NSCLC; ref. 1). Although these mutations also sensitize EGFR to inhibition by the established tyrosine kinase inhibitor (TKI) therapies erlotinib and gefitinib (2), almost all tumors will develop acquired resistance to these TKIs within 9 to 15 months (3, 4). In approximately 60% of cases, this acquired resistance is associated with an additional T790M mutation in EGFR (5–7). As there are currently no treatments approved for patients with these tumors (8, 9), new EGFR TKIs such as AZD9291, WZ4002, and CO-1686 are being developed, which inhibit both EGFRm and T790M mutations in preclinical models (10–12). AZD9291 and CO-1686 have also shown promising phase 1 activity in patients with T790M-advanced NSCLC who have progressed while on prior therapy with an EGFR-TKI (10, 11). These new TKIs may also provide treatment options in the TKI-naive setting for patients with advanced EGFRm NSCLC. However, despite the potential improvements from therapy with these TKIs, experience with targeted agents suggests that resistance to these drugs may also emerge and potentially limit their effectiveness. Therefore, identification of resistance mechanisms is important to drive new therapeutic strategies for treating drug resistance in patients.

In vitro, EGFRm cells chronically exposed to escalating doses of gefitinib or erlotinib acquire clinically relevant resistance mechanisms (13, 14), and subsequent studies have identified a range of further potential resistance mechanisms (15–20). Although the clinical importance of many of these mechanisms remains to be determined, trying to predict acquired resistance, especially to new emerging agents such as AZD9291, is a critical area of research. To date, resistance mechanisms have typically been determined from single clonal lines selected from resistant populations of cancer cells, and therefore may represent only a small percentage of the original cancer cell population. Because human NSCLC samples are heterogeneous (21–23) and tumors are likely to derive acquired resistance through multiple mechanisms, we postulated that it may be better to take a population approach to understanding the diversity and interplay of resistance mechanisms. We studied multiple cell populations resistant to gefitinib, afatinib, WZ4002, or AZD9291 to identify predominant mechanisms of resistance and to investigate signaling pathways activated by various resistance mechanisms.

All AstraZeneca cell lines were authenticated by short-tandem repeat analysis (STR). PC9 cells (obtained 2011, STR tested May 2013) were from Akiko Hiraide, PreclinicalSciences R&D, AZ, Japan. NCI-H1975 (CRL-5908, obtained 2004, STR tested Nov 2012), NCI-H820 (HTB-181, obtained 2011, STR tested Jan 2013), and HCC827 (CRL-2868, obtained 2012, STR tested Oct 2012) cells were obtained from the ATCC. HCC-2279 (K72279, obtained 2013, STR tested Mar 2013) cells were obtained from KCLB. Cells were cultured in RPMI containing 10% FCS with 2 mmol/L glutamine, supplemented with EGFR inhibitor for resistant cell populations. Selumetinib, gefitinib, afatinib, WZ4002, BMS-536924, AZD5363, AZD2014, AZD8055, GDC-0941, AZD4547, AZD1152-HQPA, and AZD9291 were synthesized according to published methods. AZ_6592, AZ_0012, AZ_1902 and AZ_9424 in house compounds (AstraZeneca).

Phenotypic cell assays, immunoblotting, and RAS activation assays were carried out as previously described (10, 24) and detailed in Supplementary Methods. Cells were transfected using Lipofectamine RNAiMAX reagent, Invitrogen, FuGENE 6 Transfection Reagent, Promega or by electroporation, MaxCyte. siRNA and DNA constructs are detailed in Supplementary Methods.

SnaPshot mutation analysis was carried as previously described (25). Targeted and whole-exome sequencing were performed on MiSeq and HiSeq platforms, Illumina; Ion Torrent PGM platform, Life Technologies and by Sanger di-deoxy sequencing. Comparative genomic hybridization was performed using SurePrint G3 Human CGH microarrays, Agilent Technologies. Sequence data processing, mutation detection, and gene copy-number assessment were carried out as described in Supplementary Methods. Data are accessible in NCBI's Sequence Read Archive accession number SRP044079 and NCBI's Gene Expression Omnibus (GEO) accession number GSE59239.

In vivo experiments were carried out as previously described using both EGFR^L858R+T790M and EGFR^L858R transgenic models (10). Details are included in Supplementary Methods.

To carry out a broad investigation into acquired resistance to EGFR inhibitors, we generated in parallel multiple EGFRm (PC9; Ex19del. chosen as a validated cell line for modeling EGFR inhibitor resistance; ref. 26) and EGFRm/T790M (PC9 derivatives and NCI-H1975; L858R/T790M) cell populations with induced resistance to gefitinib, afatinib, WZ4002, or AZD9291, using either a dose-escalation method or by culturing the cells in a single dose of AZD9291 (Supplementary Table S1).

To investigate whether the survival of resistant populations was through activation of alternative signaling pathways that circumvent EGFR dependency, we used a diverse panel of small-molecule inhibitors representing key signaling pathways or emerging resistance mechanisms (Supplementary Table S2). The ability of each agent, in the presence of originating EGFR TKI, to inhibit cell growth was measured using an in vitro phenotypic assay, and IC₅₀ values determined following 72 hours treatment (Table 1; Supplementary Table S3). It was striking that 13 of 28 PC9-resistant populations and 2 of 4 NCI-H1975–resistant populations were greater than five times more sensitive to the MEK inhibitor selumetinib in combination with the originating EGFR inhibitor, when compared with the corresponding parental cells treated with selumetinib. We therefore focused subsequent studies on understanding mechanisms of selumetinib sensitivity in these populations.

To confirm that increased selumetinib sensitivity was related to the RAS–MAPK pathway inhibition, phosphorylation levels of ERK1/2 and MEK1/2 were analyzed by immunoblotting PC9 parental and resistant populations grown in the presence of EGFR inhibitor and treated with increasing concentrations of selumetinib. Resistance to selumetinib in the PC9, PC9 GR_4 and PC9 GR_5 cells was associated with a strong induction of phosphorylated MEK and weaker inhibition of ERK phosphorylation when compared with the effects of selumetinib in the sensitive cell populations, PC9 GR_2 and PC9 AZDR_4 (Fig. 1A). The dependency of EGFR resistant cell populations on RAS–MAPK activity was further analyzed using PC9 WZR_1 cells, which showed >5-fold increased sensitivity to selumetinib (Table 1). Consistent with PC9 GR_2 and AZDR_4 populations, WZR_1 cells maintained in presence of WZ4002 demonstrated expected inhibition of phosphorylated EGFR, and phosphorylated ERK (Fig. 1Bi) and growth inhibition (Fig. 1C) were highly sensitive to selumetinib treatment. In contrast, WZR_1 cells that had been cultured in the absence of WZ4002 displayed an EGFR and selumetinib ERK sensitivity profile similar to that seen in PC9 parental cells namely weak inhibition of ERK phosphorylation and strong induction of pMEK (Fig. 1A and 1Bii), with associated growth inhibition refractory to selumetinib (Fig. 1C). The strong increase in levels of phosphorylated MEK1/2 in response to selumetinib treatment in the resistant compared with the sensitive populations (Fig. 1A and 1Bi) suggests a difference in pathway activity upstream of MEK between sensitive and resistant populations in response to relief of negative feedback loops upon MEK inhibition.

To investigate the molecular drivers of EGFR TKI resistance, we analyzed DNA samples prepared from parental and a selection of 32 resistant populations (Supplementary Table S1) for the presence of gene mutations and/or copy-number changes across a panel of cancer associated genes using multiple assay platforms. The genetic modifications detected and associated allele frequencies for PC9 and NCI-H1975 populations are summarized in Table 1; Supplementary Fig. S1. Each mutation detected was confirmed across at least two different assay platforms (Supplementary Table S4).

Across the PC9 populations, 7 of 8 gefitinib- and 2 of 3 afatinib-resistant populations had detectable T790M mutations, whereas none of the populations resistant to the WZ4002 or AZD9291 had acquired a detectable T790M mutation (Table 1). The T790M gefitinib-resistant populations mostly showed sensitivity to AZD9291, with dose–response curves indicating almost all cells in populations PC9 GR_4, 6, and 7 were sensitive to AZD9291 (Supplementary Fig. S2Ai). However, less than 50% of cells in populations PC9 GR_1 (T790M, KRAS gain 5.43-fold), PC9 GR_3 (T790M), and PC9 GR_5 (T790M) were sensitive to growth inhibition by AZD9291 (Supplementary Fig. S2Aii), suggesting that these populations contained heterogeneous resistant mechanisms. The IC₅₀ values across all AZD9291-sensitive cells were similar (Supplementary Fig S2B). Although T790M was detected within the PC9 GR_8 (T790M, KRAS gain 7.06-fold) population, these cells showed no sensitivity to AZD9291 (Supplementary Fig. S2Aiii), suggesting that the entire population contained resistant clones. This observation of heterogenenous mechanisms of resistance to gefitinib within populations is consistent with clinical setting, supporting use of this population approach for understanding resistance dynamics.

Notably, PC9-resistant cell populations lacking detectable T790M frequently displayed increased sensitivity to selumetinib in combination with EGFR inhibition. In selumetinib-sensitive EGFRm/T790M populations with induced resistance to EGFR inhibitors, no additional EGFR mutations were detected. This suggests that RAS–MAPK signaling was important for driving resistance in the absence of EGFR signaling (Table 1).

Interestingly, a number of different NRAS alterations were observed in PC9 populations resistant to AZD9291, gefitinib and WZ4002, and NCI-H1975 cells resistant to AZD9291 (Table 1). Notably, we observed a novel noncanonical E63K mutation in NRAS in the only gefitinib-resistant T790M-negative PC9 population and in two AZD9291-resistant PC9 populations (Table 1; Supplementary Fig. S3). This novel NRAS mutation occurs within the functional domain at a sequence position parallel to somatic mutations observed in both KRAS (27) and HRAS (28). We also identified functionally activating NRAS G12V and G12R mutations in two different AZD9291-resistant PC9 populations (Table 1). This is the first identification of G12V NRAS in the context of NSCLC. In addition to gene mutations, copy-number gains of MAPK1, CRKL, NRAS, and KRAS were detected across the resistant populations (Table 1), with the gain of NRAS and KRAS resulting in increased protein levels (Supplementary Fig. S4). Of particular interest, KRAS gain was observed in the T790M PC9_GR_1 population that was partially sensitive to AZD9291 (Table 1), suggesting that KRAS contributes to the heterogeneous mechanisms of resistance to gefitinib within this population. Indeed four separate AZD9291-resistant populations of PC9 GR_1 cells were subsequently generated, and KRAS gain was retained within each resistant population (Table 1). Interestingly, although T790M was still present in populations, PC9 GR1_AZDR_1, 3 and 4 T790M was no longer detectable in PC9 GR1_AZDR_2 cells.

As NRAS mutations were detected in seven of the resistant populations, and frequently associated with selumetinib sensitivity, its role in resistance was further investigated. Basal levels of active NRAS were lower in parental PC9 cells compared with resistant PC9 populations in which E63K, G12V (Fig. 2A), E63K or Q61K (Supplementary Fig. S4) NRAS mutation had been detected. Treatment of parental PC9 cells with 160 nmol/L AZD9291 decreased levels of phosphorylated EGFR and active NRAS. In contrast, in mutant NRAS cells, a decrease in phosphorylated EGFR was not associated with corresponding decrease in active NRAS, suggesting that constitutive activation of NRAS disconnected from EGFR in these cells (Fig. 2A and Supplementary Fig. S5A). In transient exogenous expression in PC9 cells, WT and mutant NRAS variants were activating (Supplementary S5Bi–S5Bii), and prevented cell growth inhibition by either AZD9291 or gefitinib compared with control transfected cells (Fig. 2B). Increased resistance to growth inhibition by AZD9291 was also observed in additional parental EGFRm cell lines similarly trasfected with WT and activating mutant NRAS variants (Supplementary Fig. S5C). Knockdown of NRAS in cell populations with three separate siRNAs, but not KRAS, for 72 hours resulted in a significant decrease in phosphorylated ERK in the resistant populations harboring NRAS mutations, but to a lesser extent in the PC9 parental cells (Fig. 2C). Moreover, knockdown of NRAS but not KRAS was associated with inhibition of proliferation only in the NRAS-mutant populations (Fig. 2D). These data indicate that activating NRAS mutations, including the novel E63K NRAS, are sufficient to drive resistance to EGFR inhibition. Similarly, knockdown of NRAS in the presence of AZD9291 caused a significant decrease in cell growth of PC9 GR_6, AZDR_2 (NRAS gain 2.4-fold), and PC9 GR_6 AZDR 3 (NRAS gain 3.68-fold) populations (data not shown).

As KRAS gain was detected within eight resistant populations (Table 1), we determined whether this could similarly drive resistance. Knockdown of KRAS in PC9 parental cells had no effect on cell growth or levels of phosphorylated ERK (Fig. 2C), whereas KRAS knockdown in PC9 AR_1 (KRAS gain 24.6-fold), in the presence of afatinib, caused a significant decrease in both phosphorylated ERK levels after 48 hours (Fig. 3A) and proliferation over 72 hours (Fig. 3B). Interestingly, knockdown of KRAS in the PC9 GR_1 (T790M and KRAS gain 5.43-fold) population, had no effect on cell proliferation alone or when treated in combination with gefitinib (Fig. 3C). However, a significant decrease in cell growth was observed when KRAS knockdown was combined with AZD9291 treatment (Fig. 3C). Consistent with this, knockdown of KRAS in the presence of AZD9291 resulted in complete inhibition of phosphorylated ERK, but not in the presence of gefitinib (Fig. 3D). These observations suggest that KRAS and T790M are both important for driving resistance in the PC9 GR_1 population.

Interestingly, we noted that a 2.64-fold gain of KRAS, as detected in WZR_3 cells, was associated with increased sensitivity to selumetinib, but cell populations with KRAS gains of between 4.44- and 24.6-fold were insensitive to selumetinib (Table 1), suggesting a threshold effect of KRAS expression. Indeed, partial knockdown of KRAS for 48 hours in AR_1 cells (KRAS gain 24.6-fold) with afatinib resensitized them to selumetinib inhibition, as revealed by decreased phosphorylated ERK, FRA1, and p90RSK levels compared with cells similarly treated with control siRNA (Fig 3E). Moreover, partial knockdown of KRAS followed by treatment with selumetinib resulted in significantly greater inhibition of cell growth compared with cells treated with control siRNA (Fig. 3G). In contrast, no significant increase in inhibition of the MAPK pathway or cell growth was observed with partial knockdown of KRAS followed by selumetinib treatment in WZR_3 cells (KRAS gain 2.64-fold) cultured in the presence of WZ4002 (Fig. 3F and 3H). Interestingly, selumetinib inhibition of MEK1/2 in AR_1 cells resulted in enhanced pMEK1/2 levels compared with that observed in selumetinib-sensitive WZR_3 cells (Fig. 3E and F). Collectively, these data are consistent with previous reports in which KRAS amplification in colon cells drives high levels of pathway output and ERK-dependent negative feedback, relief of which, upon MEK inhibition, drives relative insensitivity to MEK inhibitors (29). Similarly, we observed that enhanced exogenous expression of wild-type NRAS in PC9 AR_6 cells (NRAS 4.23-fold gain) reduced the effectiveness of selumetinib treatment on phosphorylated ERK and growth inhibition compared with PC9 AR_6 cells treated with control DNA (data not shown).

Because the data had indicated that RAS–MAPK activation was a frequent mechanism of AZD9291 and other EGFR TKI resistance, we tested whether treatment with a combination of AZD9291 and selumetinib could delay or prevent the emergence of resistance in these settings.

PC9 (EGFRm) cells were treated with 160 nmol/L AZD9291 or 100nmol/L selumetinib alone or in combination. Selumetinib alone did not inhibit proliferation of PC9 cells (Fig. 4A). Whereas resistant PC9 cells began to emerge after 34 days treatment with AZD9291, no resistant cells were observed over a similar time with a combination of AZD9291 and selumetinib (Fig. 4A). To investigate the combination in the EGFRm/T790M setting, NCI-H1975 cells were treated with 160 nmol/L AZD9291 or 100 nmol/L selumetinib alone or in combination. Treatment with 100 nmol/L selumetinib alone did not inhibit proliferation of NCI-H1975 cells (Fig. 4B). Treatment with AZD9291 initially slowed the rate of proliferation; however, a small resistant population had emerged following 17 days of treatment (Fig. 4B). Treatment with a combination of AZD9291 and selumetinib significantly delayed outgrowth of resistant cells compared with AZD9291 alone (Fig. 4B). Similarly, a combination of AZD9291 with selumetinib prevented emergence of resistance in two other cell lines, HCC827 (EGFR Ex19del) and NCI-H820 (EGFR Ex19del/T790M⁺/MET^amp; Supplementary Fig. S6A and S6B).

To further explore the EGFRm/T790M setting, PC9 GR_1 cells (T790M and KRAS gain) were treated with a combination of AZD9291 and selumetinib. Following treatment an increase in proapoptotic markers, cleaved PARP and BIM, and a decrease in antiapoptotic BclxL was observed together with a more profound effect on phosphorylated ERK levels than either agent alone (Fig. 4C). Moreover, the combined effect of inhibition of ERK signaling and apoptotic pathway was associated with a greater decrease in cell number over 12 days compared with either inhibitor alone (Fig. 4D). Although combination of selumetinib and AZD9291 did not increase apoptotic markers in NCI-H1975 (Fig. 4E), a profound antiproliferative effect was observed following 12 days treatment with the combination compared with each agent alone (Fig. 4F). Overall these in vitro studies indicated that combining AZD9291 with selumetinib lead to more profound mechanistic and phenotypic inhibition of cells.

Finally, we tested the concept that MEK pathway inhibition could circumvent resistance to AZD9291 using in vivo mouse tumor models that develop lung adenocarcinomas driven by EGFR^L858R+T790M or EGFR^L858R (30), and are highly sensitive to inhibition by AZD9291 (10). Three animals with EGFR^L858R+T790M transgenic tumors were chronically treated with 5 mg/kg/d AZD9291 and showed initial tumor regression followed by progressive disease after 3 months treatment (Fig. 5A). Animals were subsequently treated with AZD9291 in combination with 5 mg/kg twice daily of selumetinib. Remarkably, resistant tumors in 3 of 6 animals showed a profound response to the combination, showing strong regression after 1 to 2 months of combination treatment (Fig. 5A). Tumor regression was not observed when tumor-bearing EGFR^L858R+T790M mice were treated with selumetinib alone for 1 to 2 weeks (Fig. 5B). Similarly, an animal bearing an EGFR^L858R tumor showed progression after 3 months of AZD9291 treatment, which regressed following combination of AZD9291 with selumetinib (Fig. 5C). These data provide compelling in vivo evidence to support RAS–MAPK signaling dependency as an important resistance mechanism to AZD9291.

We also explored whether the activity status of the RAS–MAPK pathway could identify tumors that would benefit from combination of EGFR TKI with selumetinib. Unexpectedly, immunoblotting from parental and resistant populations revealed little correlation between basal ERK phosphorylation levels and selumetinib sensitivity (Supplementary Fig. S4). Consistent with these data, analysis of phosphorylated and total protein levels using a Reverse Phase Protein Array, showed basal levels of phosphorylated ERK were not indicative of selumetinib sensitivity (Supplementary Tables S1 and S5). In conjunction with this, sensitivity to MEK inhibition was not consistently correlated with levels of other phosphorylated or total proteins known to be involved in RAS–MAPK signaling (Supplementary Table S5).

Significant advances in our understanding of acquired resistance to EGFR-targeted drugs in EGFRm NSCLC, including but not limited to identification of T790M, MET or HER2 amplification, and PIK3CA mutants (6), are helping toward the development of new rational treatment strategies to potentially prolong patient benefit, such as AZD9291 and CO-1686, which target T790M. However, a large proportion of EGFR inhibitor acquired resistance remains unexplained, and it is anticipated that cells will also find alternative mechanisms to circumvent inhibition by new agents such as AZD9291 and CO-1686.

We have used a novel approach by directly comparing resistance mechanisms across 32 populations with acquired resistance to different EGFR TKIs, and provide the first preclinical in-depth analysis of AZD9291 acquired resistance. We took a population approach to try to better emulate the heterogeneity of resistance that occurs in advanced tumors due to competing pressures on both selection of existing clones and gain of new alterations.

A key finding is identification of certain NRAS mutations or NRAS gain as the most frequently detected genetic modifications able to drive resistance to AZD9291. Although previous in vitro data have similarly identified an NRAS Q61K mutation in acquired resistance to gefitinib or erlotinib (24, 31), this is the first report of an NRAS-activating mutation conferring acquired resistance to other EGFR inhibitors such as AZD9291. Furthermore, this is the first report of the novel NRAS E63K mutation, and together with NRAS G12V, the first report of these NRAS mutations associated with EGFRm NSCLC. The high incidence of NRAS modifications was somewhat surprising in light of recent clinical data in which common NRAS mutations were not detected in lung cancers with acquired resistance to gefitinib or erlotinib (7, 24). However, genetic alterations in NRAS have been associated with resistance to EGFR agents in other disease settings such as colorectal cancer (32, 33), raising the hypothesis that NRAS aberrations may become important in lung cancer too. Copy-number changes were not analyzed in the previous studies; therefore, the clinical relevance of NRAS copy number gain in lung cancer remains unknown. Using more extensive DNA analysis, a role for NRAS activation in EGFR TKI–resistant EGFRm NSCLC may eventually emerge, and furthermore may only become more apparent as newer agents become established in clinical practice.

Despite the clinical prevalence of specific NRAS molecular alterations being unclear, it was notable how activation of RAS–MAPK signaling independent of EGFR activity was a common biologic theme, although the precise molecular nature driving resistance remains unclear for a number of populations. Others have reported alternative mechanisms of resistance to EGFR TKIs associated with increased dependency on RAS–MAPK signaling, including loss of NF1, CRKL amplification and EMT (31, 34–36). In addition, amplification of MAPK1 was reported as a resistance mechanism to WZ4002 (15) and has been observed in PC9 AZD9291-resistant populations in this study. Taken together, these studies suggest that activation of RAS–MAPK signaling independent of EGFR could be a frequent resistance mechanism for the TKIs currently in development, with multiple different aberrations converging on RAS–MAPK signaling. This mirrors other disease areas, where resistance mechanisms to EGFR targeting result in the RAS–MAPK pathway activation by various mechanisms, for example, mutations in KRAS, NRAS, and BRAF in colorectal cancer (32, 33) or overexpression of RAS family in head and neck cancer (37) are associated with cetuximab resistance. Moreover, data presented here and by others (15, 35) support use of MEK inhibitors such as selumetinib in combination with new EGFR TKIs to overcome such acquired resistance mechanisms or potentially in earlier treatment as part of prevention strategies. Interestingly, our data support that this combination may provide benefit in both T790M and EGFRm TKI-naïve settings.

In addition to increased sensitivity to selumetinib across a number of the resistant populations, we also observed increased sensitivity to the Aurora kinase B inhibitor AZD1152-HQPA, in combination with AZD9291, in all of the AZD9291-resistant NCI-H1975 populations compared with the parental cells (Supplementary Table S3). This is consistent with recent reports (38) and is worthy of further investigation.

Overall, the emerging preclinical evidence presented here supports a picture, whereby during chronic exposure to AZD9291, competing selection pressures are likely to influence that clones within a heterogeneous population ultimately become dominant. This could also involve T790M clones becoming less prevalent within a tumor as other resistance clones become more dominant. Moreover, our data and those of others provide a compelling rationale for combining inhibitors of the RAS–MAPK signaling such as selumetinib with AZD9291 across EGFRm settings in NSCLC, to tackle RAS–MAPK as a potentially important escape mechanism within such clones. A key challenge will be to develop effective patient selection strategies to identify those patients who may benefit from such a combination. Emerging data suggest multiple genetic and nongenetic alterations, including certain NRAS modifications reported here, could occur that converge on activating the RAS–MAPK pathway, and therefore it is possible that a broad biomarker platform will need to be established. It is important that measuring basal phosphorylation levels of ERK is unlikely to be sufficient to determine dependence on RAS–MAPK signaling or sensitivity to MEK inhibitors (39), thus more sophisticated predictive biomarker strategies will need to be developed. Future studies will determine how clinically prevalent these preclinical mechanisms will be, but these preclinical observations provide important insights to focus clinical translation efforts.

P.D. Smith has provided expert testimony for AstraZeneca PLC. K.S. Thress is a translational scientist at AstraZeneca. W. Pao reports receiving a commercial research grant from AstraZeneca and has ownership interest (including patents) in MolecularMD. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C.A. Eberlein, P.D. Smith, J.R. Dry, W. Pao, D.A.E. Cross

Development of methodology: C.A. Eberlein, D. Stetson, K.J. Al-Kadhimi, Z. Lai, A. Ahmed, L.E. Ratcliffe, C.H. Barnes, J.R. Dry, B. Dougherty, W. Pao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.A. Eberlein, D. Stetson, K.J. Al-Kadhimi, P. Spitzler, E. Ichihara, S.J. Ross, A. Ahmed, E.L. Christey O'Brien, J.R. Dry, G. Beran, K.S. Thress, B. Dougherty, W. Pao

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.A. Eberlein, D. Stetson, A.A. Markovets, K.J. Al-Kadhimi, Z. Lai, P.R. Fisher, S.J. Ross, M.J. Ahdesmaki, A. Ahmed, H. Brown, J.R. Dry, G. Beran, K.S. Thress, B. Dougherty, W. Pao, D.A.E. Cross

Writing, review, and/or revision of the manuscript: C.A. Eberlein, D. Stetson, A.A. Markovets, K.J. Al-Kadhimi, Z. Lai, P.R. Fisher, C.B. Meador, S.J. Ross, A. Ahmed, C.H. Barnes, P.D. Smith, J.R. Dry, G. Beran, K.S. Thress, B. Dougherty, W. Pao, D.A.E. Cross

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.A. Eberlein, K.J. Al-Kadhimi, P.R. Fisher, C.B. Meador, P. Spitzler, H. Brown, J.R. Dry, B. Dougherty, W. Pao

Study supervision: C.A. Eberlein, M.J. Ahdesmaki, J.R. Dry, W. Pao, D.A.E. Cross

Other (performed experiments with mice, MRI, measured tumor burden, harvested lungs): P. Spitzler

The authors thank Carl Barrett for helping support delivery of sequencing studies and Hedley Carr for technical advice and assistance. The authors also thank Gayle Marshall for support in generating resistance models. At Vanderbilt, the authors thank Helen Pan for assistance with molecular genotyping assays and Daniel Colvin and Fuxue Xin for magnetic resonance imaging assistance.

This work was supported by AstraZeneca. Some of this work was supported in part by grants from the V Foundation (W. Pao) and the NCI (R01-CA121210, P01-CA129243, and U54-CA143798; W. Pao). W. Pao received additional support from Vanderbilt's Specialized Program of Research Excellence in Lung Cancer grant (CA90949) and the VICC Cancer Center Core grant (P30-CA68485). Imaging support to W. Pao was provided by NCI P30 CA68485.

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.