Mechanisms of Resistance to Crizotinib in Patients with ALK Gene Rearranged Non–Small Cell Lung Cancer

Non–small cell lung cancer (NSCLC) is increasingly recognized as a heterogeneous set of diseases at the molecular level and these differences can drive therapeutic decision making (1–3). Transforming rearrangements of the anaplastic lymphoma kinase (ALK) gene were initially identified in anaplastic large cell lymphoma (4). In 2007, an ALK gene rearrangement (ALK⁺) in which the 5′ end of the echinoderm microtubule-associated protein-like 4 (EML4) gene was fused to the 3′ portion of ALK was identified in NSCLC (5).

Recently, crizotinib has gained U.S. Food and Drug Administration approval for the treatment of ALK⁺ NSCLC. Crizotinib approval was based, in part, on data from the phase I clinical trial which showed an overall response rate of 57% and a probability of progression-free survival at 6 months of 72% (6). Unfortunately, some ALK⁺ patients will not derive any benefit from crizotinib (intrinsic resistance), whereas patients who initially derive benefit later develop resistance (acquired resistance).

Several groups have explored molecular mechanisms of acquired resistance in a comparable clinical scenario involving resistance to epidermal growth factor receptor (EGFR)-tyrosine kinase inhibitors (TKI) in EGFR mutant NSCLC (7–9). The most common mechanism noted in these patients involves a change in the target of the drug, specifically a second EGFR mutation, T790M, that alters the binding kinetics of the reversible TKIs to the target molecule (10). In addition, clinical examples of second oncogenic drivers within the same cell harboring the EGFR mutation, notably MET gene amplification, bypassing the block caused by the TKI have also been reported (8, 11).

Isolated case reports have recently identified mutations in ALK⁺ NSCLC that occur after crizotinib therapy (12, 13). However, the frequencies with which these and any other mechanisms of resistance occur remain unknown. Here, we report on the clinical and molecular details of 14 patients with ALK⁺ NSCLC with intrinsic or acquired resistance to crizotinib.

Patients with ALK⁺ NSCLC, who were treated on a phase I study of crizotinib, were considered for rebiopsy upon progression on crizotinib therapy (6). Patients were consented for tissue banking under the University of Colorado Lung Cancer SPORE tissue banking protocol and all research tests were conducted under Institutional Review Board approval. Formalin-fixed, paraffin-embedded (FFPE), fresh-frozen and/or tissue for the initiation of cell lines were collected from each patient as deemed safe and feasible by the interventional radiologist/pulmonologist. Diagnostic sampling was prioritized over research sampling in each case. No sample size was prespecified and the analysis conducted was descriptive.

ALK positivity was ascertained by FISH using ALK break-apart probes [Vysis LSI ALK (2p23) Dual Color, Break Apart Rearrangement Probe; Abbott Molecular]. The FISH assays and analyses were conducted as described previously with minor modifications (14). Using the ALK break-apart probe, 3′ (red) and 5′ (green) signals physically separated by ≥2 signal diameters were considered split. Specimens were considered positive for ALK rearrangement if more than 15% of tumor cells showed split signals or single red signals. Copy number of ALK rearrangement was based on determination of the mean of split and isolated red signals per tumor cell (15). At least 100 tumor cells were analyzed per specimen.

RNA from either FFPE or frozen tissue was processed using the RecoverAll Total Nucleic Acid Isolation Kit from Ambion. Reverse transcriptase PCR (RT-PCR) was carried out using the SuperScript III One-Step RT-PCR System with Platinum Taq DNA Polymerase from Invitrogen with primers to EML4-ALK. PCR products were either resolved on an agarose gel or analyzed with a Bioanalyzer from Agilent Technology, excised, and purified using Wizard SV Gel and PCR Clean Up Kit from Promega, then sequenced as described later. Primers used for RT-PCR or multiplexed RT-PCR of the EML4-ALK gene fusion transcript are listed in Supplementary Table S1 or have been previously described (16, 17).

Genomic DNA was isolated from manually microdissected FFPE tumor samples using the QIAamp FFPE DNA Isolation Kit from Qiagen. Samples were PCR amplified using custom primer sets for exon 2 of KRAS, exons 19 to 21 of EGFR, or exons 21 to 25 of ALK (see Supplementary Table S1 for primers) and directly sequenced using the ABI Big Dye Thermocycle Sequencing Kit and analyzed on an ABI 3730 DNA Sequencer (16). Mutation analysis was assisted by the use of Mutation Surveyor software v3.97-4.0.0 from SoftGenetics. The reference sequence for ALK sequence comparison used was NM_004304.4, for EML4 NM_019063.3, for EGFR NM_005228.3, and for KRAS NM_004985.3. We have used the colloquial single letter amino acid substitution for mutations described here rather than the official nomenclature for simplicity (e.g., L1196M rather than p.Leu1196Met).

The SNaPshot assay for evaluation of multiple oncogenic mutations in APC, AKT1, BRAF, CTNNB1, EGFR, FLT3, JAK2, KIT, KRAS, MAP2K1 (MEK1), NOTCH1, NRAS, PIK3CA, PTEN, and TP53 was conducted by amplification using 13 multiplexed PCR reactions followed by single nucleotide base extension reactions. The products were separated by capillary electrophoresis and analyzed using GeneMapper 4.0 as has been previously described (18). To confirm patient identity across multiple samples in select cases, short tandem repeat analysis was conducted on extracted genomic DNA using the AmpFlSTR Identifiler PCR Amplification Kit (Applied Biosystems).

Primary cell lines were derived from patients by placing fresh tumor tissue into sterile tissue culture dishes and culturing in RPMI supplemented with 10% FBS. A new cell line, CUTO-1 (Colorado University Thoracic Oncology), derived from a tumor biopsy sample of patient #10 is described here. Ba/F3 cells (kindly provided by James DeGregori) were cultured in RPMI supplemented with 10% FBS and 1 ng/mL of interleukin (IL)-3 from R&D Systems. 293T human embryonic kidney cells and NIH3T3 mouse fibroblast cells were obtained from American Type Culture Collection and grown in Dulbecco's Modified Eagle's Media supplemented with 5% FBS. NCI-H3122 and NCI-H2228 (kindly provided by John D. Minna and Adi F. Gazdar, University of Texas Southwestern, Dallas, TX) were grown in RPMI with 10% FBS.

Mouse monoclonal against total ALK (#3791), rabbit polyclonal against phosphorylated ALK Tyr 1278/1282/1283 (#3983), mouse monoclonal against AKT (#2920), mouse monoclonal against total ERK p42/44 (#9107), rabbit polyclonal against phosphorylated ERK Thr202/Tyr204 (#9101), mouse monoclonal against total STAT3 (#9139), and rabbit polyclonal against phosphorylated STAT3 Tyr705 (#9145) were purchased from Cell Signaling Technology. Rabbit polyclonal antibodies against phosphorylated AKT Ser473 (sc-135651) and mouse monoclonal antibodies against α-tubulin (#sc-8035) were purchased from Santa Cruz Biotechnology. IRDye 680LT- and 800CW-conjugated goat anti-mouse secondary antibodies were purchased from LI-COR Biotechnology. PF-02341066 (crizotinib) was kindly provided by Pfizer and dissolved in dimethyl sulfoxide for experiments.

The EML4-ALK gene fusion containing EML4 exons 1–6 variant and ALK exons 20–29 (E6a;A20) was cloned by RT-PCR from mRNA isolated from the cell line, H2228, and inserted into the lentiviral expression plasmid, pCDH-MCS1-EF1-puromycin from Systems Biosciences. Point mutations were introduced into the EML4-ALK fusion gene using the QuikChange II XL Site-Directed Mutagenesis Kit from Agilent Technology (see Supplementary Table S1 for primers). Lentiviral transduction of Ba/F3, NIH3T3, and H3122 was carried out as previously described (19).

Immunoblotting was conducted as previously described with minor modifications (20). Briefly, cells were lysed in modified radioimmunoprecipitation assay (RIPA) buffer supplemented with Halt Protease and Phosphatase Inhibitor Cocktail purchased from Thermo Scientific. Total protein was separated by SDS-PAGE, transferred to nitrocellulose, and stained with the indicated primary antibodies. Protein detection was achieved by imaging with an Odyssey Imager and Odyssey Version 3.0 image analysis software from LI-COR Biotechnology.

Proliferation of Ba/F3 cells was measured using the CellTiter 96 Aqueous Proliferation Assay from Promega according to the manufacturer's instructions. Briefly, cells were seeded into 96-well plates 24 hours before drug treatment, and proliferation was measured 72 hours after treatment. The absorbance at 490 nm was measured in 96-well plates using a Microplate Reader from Molecular Devices. The IC₅₀ values were calculated using Prism v5.02 from GraphPad Software.

To measure anchorage-independent growth and its inhibition by crizotinib, NIH3T3 cells expressing nonmutated or mutated cDNAs encoding EML4-ALK (E6a;A20) were suspended in media containing 0.4% agar and plated in 6-well plates containing media and 0.5% agar per well. Wells were fed every 3 days with media with the indicated concentrations of crizotinib and incubated for 14 days. Colonies were stained for 24 hours with nitroblue tetrazolium and photographs were taken for colony quantification with MetaMorph Software from Molecular Devices.

Fourteen ALK⁺ patients underwent 15 rebiopsy procedures after radiologic evidence of disease progression/lack of response on crizotinib (Fig. 1B and Supplementary Fig. S1A and Table S1). Typically, diagnostic biopsies were used as the pre-crizotinib (baseline) sample for comparison purposes within this study. Two patients were biopsied after disease progression at the first evaluation on crizotinib (intrinsic resistance). Twelve patients underwent biopsy after initial benefit then progression (acquired resistance) after median time on crizotinib of 8.9 months (range, 3.8–21.1 months). One patient underwent 2 separate biopsy procedures postinitiation of crizotinib, one after a lack of response (stable disease), followed by biopsy of a separate lesion after disease progression per Response Evaluation Criteria in Solid Tumors (RECIST; version 1.0).

Three patients (#1–3) failed to yield evaluable material from their biopsy. Frozen tumor tissue was collected from 4 patients and FFPE was collected on 11 patients. All 11 patients with evaluable tissue underwent repeat ALK FISH testing (Table 1). Attempts to propagate a cell line occurred for 8 patients. Currently, only one cell line, from patient #10, is propagating at a rate permitting evaluation.

Acquired mutations clustering around the kinase domain ATP-binding site are a well-recognized mechanism of acquired resistance to TKIs (7, 21, 22). We therefore conducted direct sequencing of ALK exons 21–25, encoding the kinase domain. Four patients (#4–7) were found to have point mutations in the kinase domain of ALK (Table 2). None of these patients had sufficient tissue in their pre-crizotinib biopsy to determine if these mutations were detectable before crizotinib therapy. Patients #4 and #5 were found to have the previously described mutation encoding the L1196M substitution (data not shown). Two patients (#6 and 7) showed the presence of a novel mutation that encodes a G1269A substitution (Table 2, Supplementary Fig. S2A, and data not shown).

A multiplexed RT-PCR assay for EML4-ALK was carried out when possible to determine the variant of the ALK gene fusion and we successfully confirmed the presence of EML4-ALK in 4 patients. Interestingly, patient #7 showed a new variant fusing exon 6 of EML4 to exon 19 of ALK (E6;A19; Table 2, Supplementary Fig. S3). An EML4-ALK–specific RT-PCR was carried out on mRNA to selectively amplify and sequence the expressed transcripts of EML4-ALK (Supplementary Fig. S3A).

The amino acid substitutions encoded by the observed mutations were mapped onto the crystal structure of the ALK domain (Fig. 1A and B; ref. 23). All of the mutations detected in this series were found in, or near, the long narrow groove that comprises the binding pocket for both ATP and crizotinib. The L1196M substitution has been previously described in a patient with crizotinib resistance and is homologous to the gatekeeper mutations identified in BCR-ABL (T315I) and EGFR (T790M; refs. 7, 12, 21).

The Gly1269 residue is positioned at the end of the narrow ATP-binding pocket of ALK (Fig. 1A and B). Substitution of Gly1269 with the larger Ala residue would be expected to reduce binding of crizotinib due to steric hindrance. To determine whether the G1269A substitution produced resistance to crizotinib, the mutation encoding this substitution was generated in a cDNA encoding the E6a;A20 variant of EML4-ALK. Lentiviral vectors encoding wild-type EML4-ALK (E6a;E20) or the same cDNA encoding G1269A, C1156Y, or L1196M substitutions or empty vector were introduced into Ba/F3 cells. Proliferation assays in the presence of increasing doses of crizotinib were conducted (Fig. 1C). The G1269A mutation induced crizotinib resistance that was intermediate between the previously identified mutations, C1156Y and L1196M. Similar results were obtained when these constructs were introduced into the NIH3T3 cell line and colony formation was measured in soft agar (Supplementary Fig. S4). Consistent with the observed cell line resistance, ALK phosphorylation and phosphorylation of downstream effectors, extracellular signal-regulated kinase (ERK), STAT3, and AKT were preserved at higher doses of crizotinib in the G1269A mutant compared with wild-type EML4-ALK (Fig. 1D and E).

Preclinical and clinical evidence suggest that the gatekeeper mutation T790M in EGFR mutant NSCLC may provide a growth disadvantage compared with NSCLC with an EGFR mutation lacking T790M (24, 25). We therefore examined the relative fitness of the G1269A, C1156Y, and L1196M mutations compared with wild-type EML4-ALK (variant E6a;A20) in the Ba/F3 cell system (Fig. 1F). We did not detect a growth disadvantage for any of the resistant mutations compared with wild type. Indeed, the EML4-ALK constructs harboring resistance mutations induced increased proliferation compared with nonmutated EML4-ALK in the absence of crizotinib.

A gain in ALK gene fusion copy number has recently been implicated as a mechanism of resistance to crizotinib in vitro (26). In addition to standard FISH analysis for ALK, we measured the number of copies per cell of the ALK gene rearrangement before and after crizotinib treatment (15). Copy number gain (CNG) was defined as more than 2-fold increase in the mean of the rearranged gene per cell in the posttreatment specimen compared with the pretreatment specimen. Two patients showed a marked increase in abnormal signal copy number (#7 at 5-fold and #8 at >4-fold), consistent with CNG of the ALK gene fusion (Table 1, Fig. 2A and B). The CNG in these 2 patients was due both to more copies of the ALK rearrangement per cell and more cells displaying the rearrangement pattern.

Patient #9 underwent 2 separate biopsy procedures each of different lesions. The first was conducted after 61 days on crizotinib after the initial restaging scans showed stable disease, which was considered unusual given the responses seen in the majority of ALK-positive patients treated with crizotinib (6). This biopsy showed a lack of an ALK gene rearrangement by FISH (Fig. 2C). Further evaluation by direct sequencing showed the presence of an EGFR exon 21 mutation encoding the L858R substitution (Table 2, Supplementary Fig. S2B) that was not present in the initial transbronchial biopsy used to establish the ALK⁺ diagnosis (data not shown). The EGFR mutation on this specimen was confirmed by SNaPshot analysis (data not shown). Interestingly, a second biopsy conducted on a progressing liver lesion after 113 days on crizotinib did show an ALK gene rearrangement, but no evidence of an ALK domain mutation, an EGFR mutation, or of any other abnormal oncogenes as assessed by SNaPshot (Tables 1 and 2, data not shown). To confirm that all samples were indeed from the same patient, the original diagnostic biopsy and both rebiopsy samples were fingerprinted using short tandem repeat analysis, confirming the common genetic origin of the samples (data not shown).

Of note, patient #1, who underwent a post-crizotinib biopsy but had no evaluable material (Supplementary Table S1), showed the presence of an EGFR exon 20 mutation (S768I) in the pre-crizotinib biopsy sample in addition to the presence of an ALK gene rearrangement by FISH analysis. This patient received erlotinib therapy for more than 9 months before discontinuing it and beginning crizotinib.

Patient #10 received only 27 days of crizotinib before disease progression was evident (Supplementary Table S1). Testing of the rebiopsy sample showed a persistent ALK⁺ FISH test but no further molecular testing was carried out because the remainder of the tumor sample was used to initiate a cell line (Supplementary Fig. S5). Initial analysis of this cell line showed marked resistance to crizotinib in comparison to 2 known EML4-ALK⁺ NSCLC cell lines (Fig. 3A). FISH analysis of this cell line later showed no evidence of an ALK gene rearrangement and repeated RT-PCR analysis failed to show evidence of an EML4-ALK gene transcript (data not shown). SNaPshot analysis of the cell line showed the presence of a KRAS G12C mutation and this was confirmed by direct sequencing (Supplementary Fig. S3C). Given the short duration until progression, we asked whether this mutation was detectable in the pre-crizotinib sample. Direct sequencing of the microdissected pre-crizotinib biopsy, which had not been previously analyzed for KRAS, showed the presence of the KRAS G12C mutation (Supplementary Fig. S3C).

Patient #11 experienced a partial response to treatment, before progression in the liver after 7 months, when a biopsy was carried out and showed persistence of an ALK gene rearrangement as well as a KRAS mutation encoding the G12V substitution (Tables 1 and 2; Supplementary Fig. S3D). Analysis of the diagnostic biopsy showed no evidence of a preexisting mutation in EGFR or KRAS (Supplementary Table S1).

In patient #11, we were unable to determine whether the ALK gene rearrangement and KRAS mutation occurred in the same or different tumor cells; whereas in the case of patient #10, given that an ALK-negative, KRAS-positive cell line could be generated from a lesion in whom both abnormalities were present in the biopsy this means that ALK- and KRAS-positive cells must have existed as separate subclones within the same tumor. To further explore whether acquisition of a KRAS mutation could serve as a direct mechanism of acquired resistance to crizotinib in ALK-positive cells, we asked whether expression of a mutant KRAS G12V could elicit crizotinib resistance in an EML4-ALK–positive cell line (H3122), which is normally sensitive to crizotinib. KRAS G12V was selected because it was the form observed in patient #11, where the coexistence in the same cells could not be formally evaluated. Mutant KRAS G12V or empty vector was introduced into H3122, and cell proliferation was measured after exposure to increasing doses of crizotinib (Fig. 3B). The IC₅₀ value of H3122 expressing KRAS G12V was not significantly different from H3122 harboring the empty vector.

All post-crizotinib biopsy samples with evaluable tumor tissue underwent repeat ALK FISH testing. In patient #10, an ALK gene fusion was not observed in the cells that were propagated from the rebiopsied supraclavicular fossa lesion, which were later shown to be KRAS mutant. In patient #9, an ALK gene fusion was not observed in the rebiopsied supraclavicular fossa lesion, which was later shown to harbor and EGFR mutation. Patient #12 also lacked an ALK gene fusion as detected by FISH in the rebiopsy of the infraclavicular lymph node (Table 1, Fig. 2D). RT-PCR of the post-crizotinib sample from patient #12 using a multiplexed assay to detect different EML4-ALK variants also failed to show the presence of an ALK gene fusion (data not shown). Unlike with patients #10 and #11, no other abnormality was detected in the evaluated genes in this patient using the SNaPshot assay.

Patients #13 and #14 showed the presence of an ALK gene rearrangement by FISH analysis after progression on crizotinib (Table 1). Patient #13 showed no evidence of ALK gene CNG or loss and no evidence of an ALK domain mutation. Patient #14 showed no evidence of ALK gene CNG or loss, no evidence of an ALK domain mutation, and no evidence of an EGFR or KRAS mutation (Tables 1 and 2).

Here, we describe the molecular mechanisms of resistance in a large series of patients with ALK⁺ NSCLC with progression on crizotinib therapy. Previous studies have described the development of in vitro resistance mechanism using cell lines grown in the presence of crizotinib (26–29). Selected clinical cases of crizotinib resistance have been described, but without a denominator to estimate the frequency of the given abnormalities (12, 13). In this study, we successfully obtained molecular data on 11 of 14 rebiopsied patients with ALK⁺ NSCLC with progression on crizotinib. A specific potential resistance mechanism was identified in 9 of these cases (Fig. 4A).

Precedent exists for the emergence of kinase domain mutations as mechanisms of resistance, most notably T315I in BCR-ABL and T790M in EGFR (7, 21, 22). We identified 4 patients (36%) with acquired resistance mutations in ALK. Two patients had the previously described L1196M mutation, in the classical gatekeeper position homologous to T315I and T790M (12). L1196M levels were low in one patient but were confirmed by RFLP analysis (data not shown). Two patients showed a novel G1269A mutation. One of these patients also harbored a novel ALK gene fusion involving exon 6 of EML4 and exon 19 of ALK (Supplementary Fig. S3). All previously published ALK gene fusions involve exon 20 of ALK. Mutations at position G1269 have previously been identified using an in vitro mutagenesis screen (29). In vitro studies with G1269A show persistent ALK phosphorylation and downstream effector phosphorylation at higher doses of crizotinib than in the wild type. Decreased growth inhibition from crizotinib was also observed with G1269A relative to the wild type. Interrogation of the ALK crystal structure bound to crizotinib reveals G1269 to be critically situated in the ATP-binding pocket (Fig. 1A and B). Crizotinib binding in this pocket is unlikely to be able to tolerate larger amino acid substitutions in either G1269 or L1196. Replacement of the Leu side chain with a longer thioether side chain (L1196M) has been shown to significantly compromise the interaction with crizotinib (12). Similarly, a bulkier Ala in place of Gly1269 would preclude proper binding of the halogenated aromatic ring of crizotinib (Fig. 1A and B).

The observation that 4 of 10 resistant samples had ALK kinase domain mutation suggests that resistance mutations in ALK⁺ NSCLC will likely be a common mechanism of resistance to crizotinib, similar to that of EGFR-TKI resistance (9). However, unlike EGFR-TKI resistance in NSCLC, there seems to be a greater diversity of resistance mutations as 4 different mutations have now been identified in patients with ALK⁺ NSCLC (12, 13). This is reminiscent of patients with chronic myelogenous leukemia (CML) treated with imatinib, in which multiple different mutations in the ABL kinase domain have been identified (30). One can speculate that the kinase domain of EGFR may be more constrained given the presence of an existing activating mutation and may only be able to tolerate a very limited set of mutations that inhibit drug binding but still allow constitutive activation. In contrast, EML4-ALK, like BCR-ABL, has a native kinase domain structure, which may be able to tolerate a wider spectrum of resistance mutations that inhibit crizotinib binding but still allow constitutive activation. Consistent with this hypothesis is the in vitro data showing that the ALK resistance mutations studied here exhibited enhanced growth compared with the nonmutated EML4-ALK in an isogenic cell line. In contrast, EGFR T790M mutations confer impaired growth compared with cells carrying EGFR-activating mutations alone (25).

The existence of multiple different resistance mutations in ALK may have important clinical implications. First, testing for resistance mutations will require direct sequencing of multiple exons or a multiplexed assay looking for different specific mutations as data emerge. Second, the possibility of different mutations existing in the same patient has to be considered, increasing the difficulty of detecting each mutation. This is in addition to multiple other mechanisms that could affect detection including allelic dilution or a mixture of mutant and nonmutant cells in the biopsy sample (31, 32). Consequently, some resistance mutations may be missed and could conceivably be present in either of the patients in our study who retained ALK but appeared to have an unknown mechanism of resistance, or in the patient with CNG alone as their apparent mechanism of resistance. CNG alone has been described as a potential mechanism of resistance in vitro; however, this appeared to be a precursor state to the development of a resistance mutation (26). Finally, based on our own and others preclinical work different mutations seem to have different sensitivity rank orderings to crizotinib and this ordering may differ between ALK inhibitors (Fig. 1C and Supplementary Fig. S4; refs. 27, 29). Therefore, the optimal exposure of any ALK inhibitor in the TKI-naive setting may differ from that in the acquired resistance setting. In addition, beyond dosing and toxicity issues, different ALK inhibitors may have to be prioritized depending on the specific mutation involved.

Seven of 11 patients in this study did not have ALK domain mutations, which led us to investigate alternate oncogenes as contributors to resistance. Prior studies have shown the coexistence of both an EGFR-activating mutation or a KRAS mutation and an ALK gene rearrangement in the same tumor sample (13, 33–35). Here we describe 2 cases with a KRAS mutation (one of which was retrospectively detected in the pre-crizotinib specimen) and one patient with an EGFR-activating mutation at the time of resistance. An additional patient had a coexistent EGFR S768I mutation and an ALK gene rearrangement in the pre-crizotinib tumor sample, but no evaluable tissue after progression on crizotinib. The presence of EGFR and ALK, or KRAS and ALK in 2 of the pre-crizotinib tumor samples is consistent with results from a large cohort of ALK⁺ patients showing that 3 of 38 (8%) patients with ALK⁺ NSCLC also showed the presence of a KRAS or EGFR mutation (36).

The presence of multiple oncogenes in a tumor sample raises the question of whether these occur in the same tumor cells or different tumor cells and how these different oncogenic drivers arise (Fig. 4B). In EGFR mutant NSCLC with acquired resistance to EGFR-TKIs via MET gene amplification, in vitro data suggest that MET gene amplification occurs as a secondary event in the same tumor cell as depicted in model #1 (Acquisition of Second Oncogenic Driver; ref. 11). This can be detected at low levels in the biopsies taken before EGFR-TKI treatment (37). Treatment with an EGFR-TKI selects for those clones with an EGFR-activating mutation and MET amplification. We suggest a different model by which an alternate oncogene provides resistance to crizotinib in ALK⁺ NSCLC. In model #2 (Emergence of a Separate Oncogenic Driver), EGFR or KRAS mutations exist in separate subclonal populations that lack an ALK gene rearrangement. The presence of EGFR/ALK or KRAS/ALK double-positive results in biopsies before crizotinib treatment cannot distinguish model #1 from model #2. However, the outgrowth of a KRAS mutant, ALK-negative cell line from patient #10 confirms that model #2 can occur clinically. Similarly, although we did not have cell line data from patient #11 who manifested both ALK and KRAS positivity after progression, the fact that the introduction of the KRAS mutation seen in patient #11 into an ALK-positive cell line did not seem to alter their sensitivity to crizotinib in vitro argues against model #1, at least with regard to KRAS and ALK. In contrast to KRAS-activating mutations, introduction of an EGFR-activating mutation into H3122 cells in vitro is sufficient to induce crizotinib resistance, suggesting that in some situations, the acquisition of a second driver within the same ALK-positive cells could act as a mechanism of resistance clinically (model #1; ref. 13). However, as an EGFR-activating mutation was noted in progressing lesions without evidence of a persistent ALK gene rearrangement in patient #9, clinically, even for EGFR, model #2 can also exist. We also cannot exclude the presence of 2 separate primary cancers in this patient. Formally, whether separate oncogenic driver subclones arise completely independently, whether they share a common progenitor that lacks either driver and these drivers are developed independently as later events, or whether both drivers coexist within the same cell, and then one or other is lost in subclonal evolution is unclear.

Emergence of an ALK gene fusion–negative tumor was observed in one patient where another oncogenic driver was not identified. Given the limited tumor sample and lack of a cell line in this patient, we were unable to query the presence of other oncogenic drivers beyond selected alleles of the genes evaluated in the SNaPshot panel. However, given that the emergence of an ALK-negative tumor was associated with definite evidence of a separate oncogenic driver in both patients #9 and #10, the assumption is that some other as yet unidentified oncogenic driver is present in these cells for them to persist. It should also be noted that in cases of the emergence of an ALK FISH–negative tumor that the percentage of positive cells was not zero, consistent with background noise in the break-apart FISH assay as described previously (38).

ALK CNG was observed in 2 patients. One patient had ALK CNG in conjunction with an ALK mutation and one had CNG without another detectable oncogene or mutation being present. The percentage of cells positive for a rearrangement was greater after CNG consistent with previous findings (15). A cell line that was partially crizotinib resistant in vitro apparently because of ALK gene fusion amplification alone was recently described (26). Increase of BCR-ABL copy number in CML serves as a precedent for this mechanism of resistance (39). In contrast, EGFR CNG has been more associated with EGFR-TKI sensitivity rather than resistance (40, 41).

We have described a series of patients with ALK-positive NSCLC with intrinsic or acquired resistance to crizotinib. Multiple different mechanisms seem to occur (Fig. 4). We would predict that in patients in whom ALK remains the dominant driver of their cancers (those with kinase domain mutations and ALK CNG), different therapies primarily directed toward the ALK protein, either TKIs or HSP90 inhibitors, may be beneficial (30, 42). However, when separate or second drivers occur—drug combinations or broader based treatments such as cytotoxic chemotherapies may be required (13, 43). If understanding the heterogeneity present in NSCLC is important enough to direct patients to the correct initial therapy, then it is becoming clear that rebiopsying and reanalyzing cancers as they stop responding may be equally important as resistance is not occurring through a single mechanism (36). Fully understanding the basis and frequency of the different mechanisms of resistance to crizotinib that are emerging will help us to continue to exploit personalized medicine approaches when considering how to overcome crizotinib resistance in patients with ALK⁺ NSCLC in the future.

R.C. Doebele has research grants from Pfizer, Eli Lilly, and ImClone. R.C. Doebele and M. Varella-Garcia have received speaker's fees from Abbott Molecular. R.C. Doebele, M. Varella-Garcia, D.R. Camidge, K.L. Kondo, A.J. Weickhardt, and D.L. Aisner have served as consultants for Pfizer. M. Varella-Garcia has served as a consultant for Abbott Molecular. D.R. Camidge has served as a consultant/advisory board member for Chugai, Ariad, and Eli Lilly. D.L. Aisner has served as a consultant for GlaxoSmithKline. No potential conflicts of interest were disclosed by the other authors.

The authors thank Barbara A. Helfrich and Christopher Korch for technical assistance and Delee A. Maxson for administrative assistance on the manuscript.

This research was supported by the University of Colorado Lung Cancer SPORE grant (P50CA058187) to R.C. Doebele, M. Varella-Garcia, and L.E. Heasley, by a research grant from Eli Lilly & Co. to R.C. Doebele and D.R. Camidge, and by funds from the Boettcher Foundation's Webb-Waring Biomedical Research Program to R.C. Doebele.

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.