EGFR Mediates Responses to Small-Molecule Drugs Targeting Oncogenic Fusion Kinases

Inhibition of oncogene fusions such as ALK, ROS1, RET, or NTRK1 blocks cancer cell proliferation and survival, consistent with the model of oncogene addiction (1–3). ALK⁺ and ROS1⁺ patients treated with the inhibitor crizotinib demonstrate remarkable objective response rates and progression-free survival times (4, 5). Early evidence also supports the efficacy of targeting TRK and RET in lung cancer patients bearing oncogenic forms of these RTKs (6–8). These oncogenes result from genomic rearrangements, which generate expression of a chimeric protein with a constitutively activated kinase domain (9, 10), herein referred as a fusion kinase.

Complete tumor responses are rare following oncogene inhibition with tyrosine kinase inhibitors, suggesting that a large population of tumor cells survive inhibition of the dominant oncogene (11, 12). Ultimately all patients will experience disease progression, most often from cellular resistance to the targeted therapies (13–15). The primary approach of most drug resistance studies has been to study tumor samples from progressing tumor lesions or by use of established cancer cell lines that have undergone long-term selective pressure under targeted therapy (13, 14). These strategies have been valuable in determining acquired resistance mechanisms that arise from the outgrowth of drug-resistant cancer cell clones. They have led to the development of next-generation tyrosine kinase inhibitors (TKI) that can overcome some mechanisms of acquired drug resistance, such as kinase domain mutations, but do not yield insight into how to improve initial treatment with upfront combination therapy, an alternate approach to combat the development of drug resistance (16).

Analysis of depth of response using RECIST criteria from a combined cohort of ALK⁺ patients treated with an ALK inhibitor showed an association between greater individual patient tumor response and longer survival (17). Thus, an alternate approach to drug resistance would be to investigate the mechanisms that underlie the incomplete tumor response observed in the majority of patients who respond to oncogene-targeted therapy. Indeed, early adaptive signaling mechanisms could permit survival of a substantial number of cancer cells following the initial insult of a kinase inhibitor, resulting in the residual tumor burden observed in the majority of patients treated with oncogene-targeted therapy, and ultimately allowing the outgrowth of drug-resistant cancer cell clones (18, 19). This particular study is focused on understanding signaling mechanisms used by cancer cells harboring oncogenic fusions that allow survival despite targeted inhibition.

EGFR is one of the most studied receptor tyrosine kinases (RTK) because it plays an essential role during embryonic development and adult homeostasis and is often aberrantly activated in cancer (20). EGFR is expressed at high levels or mutated in many epithelial malignancies including lung, glioma, hepatocellular carcinoma (HCC), breast, colorectal, head and neck squamous cell carcinoma (HNSCC), and ovarian cancers (20). Wild-type EGFR is an established therapeutic target in HNSCC, colorectal cancer, and squamous cell lung cancer, and thus plays an important role in cancer cell signaling in these tumors (21–23). EGFR is overexpressed in approximately 80% of NSCLC and is associated with a poor prognosis, but most current clinical trials in this disease focus on targeting only those tumors that harbor drug-sensitizing EGFR mutations, not the majority of patient tumors that express high levels of the wild-type receptor (24).

The studies presented here investigated the role of wild-type EGFR signaling in cancer cells that harbor oncogenic fusion kinases prior to the onset of acquired drug resistance in an effort to improve our understanding of early adaptive cancer cell signaling that mediates survival in the face of oncogene inhbition. We evaluated potential roles for EGFR signaling following treatment with a fusion kinase inhibitor (FKI), but also in the absence of any therapeutic stress in fusion oncogene–positive human-derived cancer cell lines, murine cancer models, and patient tumor samples. In this report, we describe multiple and distinct roles for EGFR signaling in drug-naïve fusion kinase–positive cancer cells, which indicate a novel role for EGFR as a cooperative signaling partner in these molecular subtypes.

Written informed consent was obtained from the patients prior to use of the patients' tumor samples. The consent form and protocol were reviewed and approved by the Colorado Multiple Institutional Review Board. Excess tumor tissue from patients was collected for derivation of tumor xenografts (COMIRB #10-1386) or for additional research studies (COMIRB #11-1621). Patient studies were conducted according to the Declaration of Helsinki, the Belmont Report, and the U.S. Common Rule.

H3122, H2228, HCC78, H1650, H520, and H1993 (all obtained in 2008-2010 from Dr. Paul A. Bunn, University of Colorado, Aurora, CO), CUTO-3 (derived in 2013 by our laboratory), CUTO-2 (derived in 2012 by our laboratory), DFCI-032 (obtained in 2014 from Dr. Pasi A. Jänne, Dana Farber Cancer Institute, Boston, MA), STE-1 (obtained in 2013 from Christine M. Lovly, Vanderbilt University, Nashville, TN), and KM12 (obtained in 2013 from Dr. Gail Eckhardt, University of Colorado, Denver, CO) cell lines have been described previously (1, 13, 15, 25–27). LC-2/ad cells were purchased from Sigma-Aldrich in 2013 and 293T purchased from ATCC in 2009. All cancer cell lines were authenticated using DNA fingerprinting analysis by short tandem repeat profiling, subjected to routine mycoplasma testing, and cultured in RPMI1640 supplemented with 1%–10% FBS. 293-T cells were maintained in DMEM supplemented with 5% FBS. Gefitinib, crizotinib, and TAE-684 were purchased from Selleck Chemicals. ARRY-470 was kindly provided by Array Biopharma, and foretinib (XL880) was kindly provided by GlaxoSmithKline. All EGF was purchased from R&D Systems. The following plasmids were previously described: pCDH-EML4-ALK, pCDH-SDC4-ROS1, pCDH-MPRIP-NTRK1, and pCDH-MPRIP-NTRK1-KN (1, 13, 15, 28). The FGFR1 cDNA was obtained from Origene, and EGFR was cloned from the EGFR-EGFP expression plasmid from the Sorkin Laboratory (Addgene, plasmid #32751) and both were cloned into the PLVX plasmid.

293-T cell transient transfections were performed using Mirus TransIT LT1 or TransIT-X2 reagents according to the manufacturer's protocol. Mission Non-Target shRNA Control #SHC002 and 2 EGFR shRNAs were ordered from the Functional Genomics Facility at the University of Colorado (Boulder, CO) TRCN0000010329 (sh#1) and TRCN0000039634 (sh#2). EGFR shRNAs were transfected into 293T cells and stably transduced into CUTO-3 and LC-2/ad cells as described previously (1).

Cell lysates were obtained using a modified RIPA lysis buffer and incubated rotating overnight at 4°C with each designated antibody or antibody–bead conjugate (mouse EGFR sepharose beads, Cell Signaling Technology, #8083). Lysate–antibody complexes were then incubated for 2 hours with PBS-washed protein-G–conjugated agarose beads at 4°C (Merck Millipore, catalog # 539207). Beads were washed through inversions and spun at 13,000 rpm for 30 seconds at 4°C three times with ice-cold PBS. Beads were resuspended in sample loading buffer, boiled, and stored at −80°C prior to immunoblot analysis. Antibodies were used for immunoblot analysis as described below.

HCC78 or H3122 cells were incubated with PureProteome streptavidin–conjugated magnetic beads (Millipore), following the specified drug or growth factor treatments, according to the manufacturer's protocol. Beads were washed three times using a magnet, and resuspended in sample loading buffer, boiled, and stored at −80°C prior to immunoblot analysis.

Immunoblot analysis was performed as described previously (1). Precast 4%–20% gradient gels (Bio-Rad) were used for EGFR and GRB2 immunoprecipitations. The following antibodies were used from Cell Signaling Technology: pTRK Y490 (9141 or 4619) and Y464/65 (C50F3), pEGFR Y1068 (D7A5) and Y1173 (53A5), total EGFR (2232L or XP D38B1), HA (C29F4 or 6E2), pALK Y1278/1282/1283 (3983), total ALK (D5F3, 3333, or 31F12), total ROS1 (D4D6), pRET Y905 (3221), total RET (C31B4), pERK1/2 XP T202/Y204, total ERK1/2 (3A7), pAKT S473 XP (D9E 4060 or 193H12), GRB2 (3972), and pan-keratin (C11). Total RET was also purchased from Abcam (ab134100), GAPDH (MAB374), and pTYR (4G10) from Millipore, total EGFR (610017) from BD Biosciences, total TRK (C-14) and tubulin antibodies from Santa Cruz Biotechnology, and FGFR1 antibody (#TA301021) was purchased from Origene.

All proliferation assays were performed in media supplemented with 5% FBS as described previously using CellTiter 96 MTS (Promega; ref. 1). Cells were seeded at 750, 1,000, or 2,000 cells per well and treated for 72 hours at the drug concentrations indicated. Each assay was performed in triplicate and using at least three independent biological replicates. Data were graphed and analyzed using GraphPad Prism software, with 50% inhibitory concentrations (IC₅₀) values calculated as half maximal response. The nine cell lines assayed by MTS were additionally pooled together for statistical analysis. All statistical analysis was performed using a paired Student t test.

Cells were seeded onto chamber slides at 2.5–5 × 10⁴ cells/well. Cells were treated with the indicated doses and times then fixed for 15 minutes by shaking at room temperature in 4% paraformaldehyde. Cells were rinsed twice in PBS, and then the Duolink In Situ PLA kit (Sigma Aldrich) in mouse/rabbit (with red detection) was used according to the manufacturer's protocol (catalog # DUO92002, 92004, 92008). Antibody concentrations were optimized prior to proximity ligation assay (PLA) experiments. FFPE tissue PLAs from human cell line xenografts, a patient-derived xenograft, or patient tumor samples were prepared as described in histology. In addition, FFPE samples were treated with 300 mmol/L glycine for 30 minutes prior to the blocking step, then the assay was performed according to the manufacturer's protocol. Cells were mounted using Duolink In Situ Mounting Medium (with DAPI) and cured, and sealed with clear nail polish prior to imaging. Images were taken on a Nikon standard inverted fluorescent microscope at 40×. The following antibodies were used in the PLA: ALK (D5F3), RET (C31B4), ROS1 (D4D6), EGFR (D38B1), and Trk (C17F1) from Cell Signaling Technology or (C-14, SC-11) from Santa Cruz Biotechnology; SHC1 (3F4, H00006464) from Novus Biologicals; Grb2 (26B04) from Invitrogen; Grb2 (610111) from BD Biosciences; EGFR (29.1, E2760) from Sigma-Aldrich; and EGFR (SC-120) and Grb7 (A-12) from Santa Cruz Biotechnology. The antibodies used in PLAs were tested for specificity using various control cell lines. A stage micrometer (reticles.com, KR-85-11) was imaged and the scale bar inserted into the images. Each scale bar shown represents 50 μm, unless stated otherwise.

At least three independent experiments and at least 300 nuclei were assayed in every quantification analysis. CellProfiler software was used to quantify these assays (29). The pipeline used in these studies to analyze images was: ColorToGray (to split fluorescent channels), EnhanceorSuppressFeatures (to enhance the red speckles), IdentifyPrimaryObjects (to specify the number of nuclei in the blue channel), and MeasureImageIntensity (of speckles in the red channel) for each image. Red fluorescence intensity was normalized to the number of nuclei in each image field. Each quantification graph is expressed as the mean ± SEM. All statistical analysis was performed using a paired Student t test.

NTRK1 FISH assays were performed as described previously (1).

Tumor specimens were fixed in 10% neutral buffered formalin, processed, paraffin-embedded, and cut at 4 μm. Slides were stained with hematoxylin and eosin and examined microscopically. For IHC staining, the slides were deparaffinized in xylene and rehydrated in graded concentrations of ethanol before antigen retrieval in EDTA pH 9 solution (Dako) in a pressure cooker at 121°F for 10 minutes. Next, the slides were cooled for 20 minutes before washing in 1× Wash Buffer (Dako). They were then treated with Dual Endogenous Enzyme Block (Dako) for 10 minutes, followed by Serum-Free Block (Dako) for 20 minutes. The samples were incubated with antibody for 60 minutes at room temperature followed by HRP-conjugated EnVision + Dual Link System (Dako). Staining was developed using DAB+ chromogen solution (Dako). Slides were counterstained with automatic hematoxylin (Dako), washed with wash buffer and distilled water, alcohol dehydrated, and mounted with resinous media. EGFR D38B1 and pEGFR D7A5 from Cell Signaling Technology were used. H-score analysis of EGFR IHC in patient samples was performed as described previously (30).

Animal care and procedures were approved by the Institutional Animal Care and Use Committee Office of the University of Colorado Anschutz Medical Campus (Aurora, CO). Forty nu:nu male mice were injected with 10⁶ H3122 cells in each flank subcutaneously. Mice were kindly provided a breeder colony run by Lynn Heasley and James DeGregori (University of Colorado, Aurora, CO). Once tumors reached 250 mm³, mice were randomly distributed into 4 groups (n = ∼20 tumors/group) and treated daily by oral gavage with the following: diluent, crizotinib 50 mg/kg, gefitinib 25 mg/kg, or both for 17 days. Tumor size was measured twice weekly using calipers and tumor volume was calculated using the formula: volume = (length × width²)/2. Gefitinib was purchased from Selleck Chemicals, and crizotinib was kindly gifted back to the Doebele laboratory from patients. Drugs were reconstituted in “diluent,” which was 1% polysorbate 80. Mice were sacrificed when tumor volume or appearance exceeded regulations, and tumor specimens were excised, fixed in neutral buffered formalin, paraffin-embedded, and slides cut at 4 μm. Each tumor was analyzed by the indicated PLA as pharmacodynamic markers. Final statistical comparisons of tumor volume change were calculated using ANOVA with GraphPad Prism software.

Animal care and procedures were approved by the Institutional Animal Care and Use Committee Office of the University of Colorado Anschutz Medical Campus (Aurora, CO). Propagation and maintenance of CULC001 and CULC002 xenografts was described previously (6, 31).

Previous studies from our laboratory have shown that EGFR mediated partial intrinsic resistance to targeted inhibition of ROS1 in a ROS1⁺ cell line, and EGFR mediated the acquired resistance in that cell line when treated chronically with a ROS1 FKI (13, 28). Here, we observed incomplete inhibition of critical downstream signaling pathways in drug-naïve gene fusion–positive cell lines where the fusion kinase was inhibited pharmacologically (Fig. 1A; Supplementary Fig. S1). We hypothesized that EGFR could account for the incomplete inhibition in these cell lines. In support of this hypothesis, the addition of the EGFR-specific tyrosine kinase inhibitor (TKI), gefitinib, increased inhibition of downstream signaling pathways (Fig. 1A) while EGF stimulation had the opposite effect by rescuing phosphorylation of important downstream signaling pathways such as ERK1/2 and AKT in lung cancer cell lines bearing TRKA, RET, ROS1, and ALK fusions (Supplementary Fig. S1).

We also asked whether EGFR activation could influence constitutive fusion kinase signaling in the absence of inhibition by an FKI. We measured the kinetics of EGFR, fusion kinase, and ERK1/2 phosphorylation following EGF stimulation. We observed rapid, maximal activation of EGFR at 5 minutes and increased phosphorylation of the fusion kinases above levels that occur from its constitutive activation, which also resulted in further amplification of downstream signaling (Fig. 1B; Supplementary Fig. S1). These data provide evidence that the activity and signaling of oncogenic fusion kinases can be modified by nononcogenic RTK signaling, such as EGFR, under normal constitutive signaling or FKI-treated conditions.

The ability of EGFR to regulate both fusion kinase phosphorylation and downstream signaling suggests an important role for this RTK in tumor cells. We therefore asked whether EGFR and its signaling regulated tumorigenic properties, such as cell proliferation in cancer cells harboring oncogenic fusions. We expanded upon our initial studies to include 9 different fusion kinase–positive cell lines (Supplementary Table 1) to determine the influence of EGFR inhibition or activation on cellular proliferation in the presence of a cognate FKI (Fig. 1C). Inhibition of EGFR with gefitinib significantly increased the sensitivity of the cell lines to their respective FKIs by an average of 12.7-fold (P = 0.01), whereas stimulation of EGFR with EGF significantly decreased the sensitivity of the cells by 5.9-fold (P = 0.004), demonstrating an important role for EGFR signaling in cancer cells that express fusion oncogenes (Fig. 1C; Supplementary Table S1). It is notable that EGFR inhibitors had an effect on proliferation even in the absence of exogenous EGF, consistent with the downstream signaling results in Fig. 1A. In addition, depletion of EGFR by RNAi enhanced FKI inhibition in TRK and RET models, further supporting the role of EGFR in proliferation of these cells (Supplementary Fig. S2).

After observing the effect of EGF stimulation on constitutive fusion kinase signaling, we tested whether EGFR activation could induce phosphorylation of the fusion kinase in the presence of an FKI. Although EGF-induced bypass signaling was anticipated, rephosphorylation of the drug-inhibited fusion kinase was not expected (Fig. 2A; Supplementary Fig. S1). This result suggested an additional, unexplored role for EGFR signaling in cancer cells. Because EGFR signaling can transactivate a fusion kinase in an inactive state (FKI-bound), we asked whether EGFR could also transactivate a different type of inactive state of the fusion kinase, a catalytically inactive version of the fusion kinase. To address this, we coexpressed EGFR and kinase-dead, HA-tagged myosin phosphatase Rho-interacting protein (MPRIP)-NTRK1 K544N fusion genes in 293T cells (1). EGF stimulation resulted in phosphorylation of the catalytically inactive RIP-TRKA protein, and this effect was abrogated with EGFR kinase inhibition (Fig. 2B). Similarly, EGFR-mediated transactivation was observed for the catalytically inactive mutant of CD74-NTRK1 (Supplementary Fig. S1E).

To further understand the mechanism of fusion kinase rephosphorylation, we tested whether EGF-induced activation of EGFR resulted in the dissociation of the FKI. To address this, we used a biotin-linked crizotinib or biotin-linked cabozantinib as a bait to pull down the drug and its target kinase, ALK or ROS1, respectively, using streptavidin-conjugated beads (32). Biotin-labeled cabozantinib pulled down ROS1, and biotin-labeled crizotinib pulled down ALK, and the biotinylated drug–fusion kinase complex was lost upon competition with the addition of excess, unlabeled drug (Fig. 2C). Importantly, stimulation of cells with EGF also disrupted the biotin–cabozantinib pull down of ROS1 or biotin–crizotinib pull down of ALK (Fig. 2C). These data suggest that EGFR activation by EGF destabilized FKI binding to ROS1 and ALK.

We have shown that the fusion kinase is rephosphorylated following EGFR activation, and that FKI binding was disrupted, but we wanted to ask whether the fusion kinase was reengaged in signaling. We first addressed this by using phosphotyrosine-specific antibodies, and were able to identify a common phosphorylated tyrosine residue(s) mapping to the conserved activation loop of the ALK, RET, and TRKA kinases (Supplementary Fig. S3A and S3B). Next, based on the rephosphorylation of tyrosines critical to kinase function, we tested whether this phosphorylation resulted in a rescue of signaling by the fusion kinase as measured by adaptor protein binding. Therefore, a PLA was employed to detect kinase–adaptor signaling complexes in cells (33). We designed assays to recognize complexes formed between each fusion kinase and a predicted, commonly used adaptor protein based on the conserved phosphorylated tyrosine(s) detected following inhibition and rescue with EGF (Supplementary Fig. S3A and S3B; refs. 34–36). The PLA is designed to detect 2 proteins within a distance as large as 40 nm (33, 37). We have previously shown active signaling TRK–SHC1 complexes using this approach, including the ability to disrupt these complexes by RNAi knockdown of one partner or by pharmacologic inhibition of the kinase (6). TRKA-SHC1 (CUTO-3 cells), RET-GRB7 (LC-2/ad), ROS1-GRB2 (HCC78), and ALK-GRB2 (H3122 and STE-1) kinase–adaptor PLAs revealed that each fusion was signaling through the adaptor at baseline as expected, and that these complexes were disrupted by a cognate FKI (Fig. 3A and C; Supplementary Fig. S3C). Importantly, the fusion kinase–adaptor signaling complexes were restored after EGF stimulation (in the presence of the FKI), and were only fully disrupted by combination treatment of the FKI and gefitinib when EGF was also present (Fig. 3A and C; Supplementary Fig. S3C). These data corroborate the rephosphorylation results observed in the immunoblot experiments (Fig. 2; Supplementary Fig. S1) and support a model where EGFR signaling is not only rephosphorylating and disrupting FKI binding, but reengaging the signaling function of the inhibited fusion kinase.

We next queried the role of EGFR signaling in these gene fusion–positive cell lines by implementing an EGFR-GRB2 PLA. This assay was validated using cell lines that demonstrate different levels of EGFR expression and activity (Supplementary Fig. S4A–S4C). We observed dose-dependent loss of EGFR signaling complexes in the presence of an EGFR inhibitor (Supplementary Fig. S4A and S4B). We observed two patterns of EGFR activation in the gene fusion–positive cell lines: in the first type, observed in ALK and TRKA fusion cell lines, EGFR signaling complexes were present in unperturbed cells (basal EGFR activation; Fig. 3B and C; Supplementary Fig. S3D). Alternatively, in ROS1 and RET fusion–kinase positive cell lines, an increase in EGFR signaling complexes was observed following inhibition of the fusion kinase (induced EGFR activation; Fig. 3B and C; Supplementary Fig. S3D). Thus, these assays provide additional evidence that EGFR can signal independently through GRB2 in these cells.

To evaluate the increase in EGFR-GRB2 signaling with FKI treatment observed in Fig. 3, we performed a time course using the ROS1 cell line HCC78 and the RET cell line LC-2/ad to delineate the kinetics of the increase in EGFR-GRB2 signaling following FKI treatment. The ALK and TRK cell lines were not included, because they display higher levels of EGFR signaling prior to FKI treatment. We observed a rapid increase in EGFR signaling complexes following addition of a ROS1 inhibitor, and this was readily reversible, reverting to baseline levels upon wash out of the ROS1 inhibitor (Fig. 4A, B, D, E). This change in GRB2 binding adaptor binding from ROS1 to EGFR was also observed using coimmunoprecipitations following FKI treatment in the HCC78 cells (Fig. 4C). This change in binding of GRB2 from ROS1 to EGFR was specific, and did not occur with the MET RTK, which is both expressed and phosphorylated in these cells (Supplementary Fig. S5A and S5B; ref. 13). Use of H1993 cells, which display MET gene amplification and are sensitive to MET inhibitors such as crizotinib, were utilized as a positive control to demonstrate that MET–GRB2 complexes could be detected using the PLA and that they were disrupted with MET inhibition (Supplementary Fig. S5C and S5D) (38). A similar result, albeit with slower kinetics, was seen in the RET⁺ LC-2/ad cells. Using both PLA and coimmunoprecipitation, both GRB2 and SHC1 adaptor proteins demonstrated a similar switch from RET to EGFR (Supplementary Fig. S6). Collectively, these experiments suggest that while phosphorylation is necessary to initiate signaling, it may not be a sufficient indicator of adaptor binding and the initiation of a signaling cascade. These results demonstrate the plasticity of signaling in cancer cells as indicated by the ability of the GRB2 or SHC1 adaptors to switch from one kinase to another following short-term kinase inhibition.

Given the rapid phosphorylation of the fusion kinase by EGFR and the ability of GRB2 to switch to EGFR, we asked whether EGFR could exist in a complex with the fusion kinase. All 4 fusion kinases coimmunoprecipitated with EGFR both in lung cancer cells, where both proteins are endogenously expressed, and also when ectopically coexpressed in 293T cells (Supplementary Fig. S7A–S7H). This interaction was sustained under various kinase-inhibitory and -stimulatory conditions, including combined inhibition and EGF stimulation, suggesting the interaction is independent of kinase activity (Supplementary Fig. S7A–S7H). Overexpression of another RTK, FGFR1, did not generate complexes with ROS1, indicating a degree of specificity for the EGFR–oncogenic fusion kinase interaction (Supplementary Fig. S7I and S7J). Complexes between each of the fusion kinases and EGFR were also observed endogenously in each cell line using PLA (Supplementary Fig. 7K), but not with cytokeratins, another abundant class of proteins in these cells (Supplementary Fig. S7L). EGFR–fusion kinase complex detection was specific to cells harboring the correct fusion kinase, demonstrating an absence of antibody cross-reactivity (Supplementary Fig. S4M). These data provide evidence for complexes between TRKA, ALK, RET, or ROS1 fusion kinases and EGFR.

To determine the role of EGFR in vivo, we sought to establish whether the combination of crizotinib and gefitinib induced a more substantial reduction in tumor volume compared with monotherapy crizotinib in an ALK⁺ xenograft mouse model. The combination of gefitinib and crizotinib resulted in a significant decrease in tumor growth over crizotinib monotherapy (Fig. 5A). In addition, pharmacodynamic analyses using PLA were conducted on tumors from each of the four treatment arms in this study to evaluate the target inhibition of ALK and EGFR at the termination of the study. There was a significant reduction of ALK-GRB2 and EGFR–GRB2 signaling complexes with combination therapy, compared with ALK monotherapy, consistent with the effects of combination therapy on tumor growth (Fig. 5B and C). Together, these data provide evidence for a functional role of EGFR in fusion kinase–driven cell proliferation as well as tumor maintenance.

We next asked whether EGFR activation could be detected in a patient-derived xenograft (PDX) mouse model derived from a NSCLC patient whose tumor harbored an NTRK1 fusion (Supplementary Fig. S8A; refs. 1, 6). Tumors from the PDX model are positive for the NTRK1 rearrangement by FISH and demonstrate both high levels of EGFR expression and phosphorylation by IHC (Supplementary Fig. S8B and S8C). FFPE tumor samples from the NTRK1⁺ PDX model were positive for TRKA-SHC1 and EGFR-GRB2 signaling complexes by PLA, as well as TRK–EGFR PLA complexes, whereas an NTRK1⁻ PDX model derived from a patient with an unknown oncogene was negative for each of these PLAs (Supplementary Fig. S8D–S8I).

EGFR is highly expressed in unselected NSCLC (24). We evaluated EGFR expression in a cohort of 26 ALK⁺ and ROS1⁺ patients who also had clinical EGFR IHC testing performed at our institution. Similar to unselected NSCLC, the majority of ALK or ROS1⁺ tumors had high levels of EGFR expression with 21 of 26 (81%) samples exhibiting an IHC score ≥200 (out of 300) demonstrating that EGFR is available for signaling contribution in most of these gene fusion–positive tumors (Supplementary Fig. S9A).

Finally, we asked whether fusion kinase and EGFR signaling complexes could be detected in patient tumor samples. We analyzed ROS1⁺, ALK⁺, and RET⁺ patient tumor samples using PLAs specific for either fusion kinase–adaptor or EGFR–GRB2 complexes (Fig. 5D and E; Supplementary Fig. S9). Minimal EGFR–GRB2 signaling was detected by PLA prior to treatment with a FKI in a ROS1⁺ patient, but a significant increase in EGFR–GRB2 signaling was detected at disease progression post-FKI treatment in that ROS1⁺ patient (Fig. 5D and E). Abundant levels of EGFR–GRB2 signaling complexes were also detected in resistant ALK⁺ or RET⁺ lung cancer patients (Supplementary Fig. S9). The fusion kinase–adaptor PLAs were also detected in each patient's resistant sample following or while still on treatment with an FKI. None of these tumor samples had evidence of a resistance mutation in the kinase domain to account for persistent fusion kinase activation. Thus, the persistence of fusion kinase signaling in the samples is consistent with our in vitro results, which demonstrated fusion kinase transactivation by EGFR (Fig. 5D; Supplementary Fig. S9B–S9F). A robust, distinct pattern of EGFR–GRB2 was also detected in a tumor sample from an ALK⁺ patient's brain metastasis surrounding stromal fibrovascular bundles (Supplementary Fig. S9F). The same patient samples were further tested for PLA specificity using mismatched oncogene PLAs (e.g., ALK-GRB2 PLA on RET fusion–positive sample) and no PLA cross-reactivity was observed (Supplementary Fig. S9G). The in situ patient tumor data support the in vitro and in vivo findings, providing further evidence for a role for EGFR–GRB2 signaling in fusion kinase–positive patients.

The work presented here demonstrates two novel mechanisms by which EGFR can rescue critical signaling in cancer cells harboring gene fusions treated with targeted inhibitors: reactivation of the drug-inhibited oncogene and dynamic adaptor protein-mediated signaling rewiring. These data suggest that wild-type EGFR, which is expressed at high levels on the majority of ALK⁺ human tumor samples, plays a critical role in cancer cells harboring gene fusions. While it is clear from clinical trials of crizotinib in ALK⁺ and ROS1⁺ patients that EGFR cannot sustain sufficient signaling in all patient tumors or in all cancer cells, the capability of an EGFR rescue described in this work may explain the heterogeneous or incomplete responses seen in the majority of patients (16). Notably, this work demonstrates a potential role for inhibiting wild-type EGFR using EGFR inhibitors in ALK⁺ or ROS1⁺ patients being treated with crizotinib or other ALK inhibitors, and potentially for RET⁺ and TRKA⁺ patients once targeted drugs are approved for these indications.

These studies demonstrate that EGFR can influence the activity of an oncogenic fusion kinase in cell lines derived from human tumors. Similar to the heterogeneity of responses to ALK or ROS1 inhibitors in NSCLC patients, the magnitude of EGFR contribution was also heterogeneous, but was present in each of the 9 models tested (Supplementary Table S2; refs. 11, 12). The ability of EGFR to transactivate the fusion kinase in the absence of drug perturbation suggests that this signaling partnership with EGFR exists to augment or supplement signaling from the dominant oncogene, in this case the fusion oncogene. Incomplete attenuation of MAPK and AKT signaling by FKI alone and more complete inhibition of these pathways by FKI and EGFR inhibition support this supplementary signaling role for EGFR. Coimmunoprecipitation and PLA experiments support a model where EGFR is in a complex with the fusion kinase. Further evidence for an interaction between EGFR and the fusion kinases come from the demonstration that EGFR activation leads to phosphorylation of the fusion kinase in the presence of a potent FKI. The biotinylated drug experiments used a novel approach to demonstrate that EGFR activation leads to disruption of drug binding to the fusion kinase, further supporting a model in which EGFR is complexed with the fusion kinase. Thus, our model of oncogene addiction allows for the modulation of the dominant oncogene by other signaling nodes (Fig. 6). Recent work by several groups demonstrate that WT EGFR can positively modulate mutant KRAS activity and is consistent with the model in which dominant oncogenes can be modulated by nononcogene signaling pathways (39, 40).

The RTK–adaptor PLAs used in this study enabled the detection of activated EGFR signaling complexes in cell lines and FFPE sections of PDX or patient tumor samples in the absence of any detectable activating EGFR mutations. The rapid activation of EGFR in ROS1 and RET cell line models following fusion kinase inhibition suggest a model of dynamic signaling. In this model, cellular reprogramming, which can require days to weeks of oncogene inhibition, is not required for the cell to maintain partial signaling through critical signaling pathways despite inhibition of the dominant oncogene. Different signaling pathways (e.g., ROS1 and EGFR) are not necessarily engaged simultaneously at baseline, but a dynamic rewiring of signaling may occur following perturbations such as drug inhibition of the dominant oncogene. Adaptor proteins may demonstrate binding plasticity and “switch” from the dominant oncogene(s) to EGFR under therapeutic stress (Figs. 3B and C and 4A; Supplementary Fig S6). However, removal of the inhibitor allows GRB2 to revert to ROS1 binding, suggesting a preference for the dominant fusion oncogene (Fig. 4D). The interplay of adaptor binding is likely subject to many other factors, including concentration and competition in its environment, and thus may not automatically occur following phosphorylation (41). The ROS1⁺ HCC78 cell line provides a strong example of cell signaling plasticity that occurs following loss of ROS1 signaling. The acute loss of ROS1 signaling after minutes of drug treatment allowed signaling via EGFR in this study (Figs. 3 and 4). Previous studies from our laboratory showed that ROS1 signalling suppression for several days was required for these cancer cells to adapt to signaling via an activating KIT mutation and a loss of the ROS1 fusion gene at the genomic level occurred following long-term drug selection with a ROS1 inhibitor, permitting a switch to WT EGFR dependence (13, 42). Our adaptive signaling model is supported by recent work using phosphoproteomic approaches, which demonstrated that GRB2 was bound to EGFR in unperturbed ALK⁺ lung cancer cells and that EGFR phosphorylation increased in response to crizotinib treatment (43). Collectively, these data demonstrate a need for the loss of the oncogene to occur to allow signaling rewiring. These data support an approach of targeting the fusion kinase and EGFR at the initiation of therapy to prevent this rewiring (Fig. 6).

Growth factor ligands have previously been shown to mediate resistance to targeted inhibition of dominant oncogenes in various cell types by activating bypass signaling pathways (13, 27, 44, 45). Importantly, EGF and other EGFR ligands can signal in an autocrine, paracrine, or juxtacrine manner and they are expressed or upregulated in lung cancer (13, 45, 46). Prior studies demonstrated a mechanism whereby ligand engagement leads to bypass of the dominant oncogene. This indirect rescue of downstream signaling by the EGF/EGFR axis (Fig. 1A and C; Supplementary Fig. S1) in this study is consistent with prior studies demonstrating this in ALK and ROS1⁺ cells (45); however, extending these findings to RET and TRKA human-derived cancer cell models, suggest that this is a common feature among fusion oncogene–driven cancer cells. As described above, our data suggest additional mechanisms by which growth factor ligands can mediate resistance to oncogene-targeted therapy (Fig. 6).

Other studies have shown a role for EGFR or other ERBB family members in mediating acquired resistance in ALK⁺ and ROS1⁺ cell lines (13, 47–50). The common dependence of EGFR and the ERBB family for acquired resistance likely reflect the critical role and intrinsic wiring toward EGFR in cells harboring oncogene fusions prior to treatment with oncogene-directed therapy. Our studies do not rule out a potential role of other HER family members, such as HER2 or HER3, as these RTKs are known to heterodimerize with EGFR or signal independently. However, gefitinib is far less potent against HER2 than EGFR and we used a dose of gefitinib that discriminates between these RTKs (51). Also, EGFR is a more attractive target as it is more widely expressed than HER2 in NSCLC patients (24, 52). We speculate that cells are likely to rely on preexisting signaling pathways for acquired resistance, rather than generating bypass signaling de novo from signaling programs that do not preexist. Although additional models will need to be identified and tested, our observations in a colorectal cancer cell line KM12 harboring the TPM3–NTRK1 fusion suggests that this role for EGFR in the context of fusion kinases might also apply to malignancies other than lung cancer (9, 53, 54). Previous work in other cancers demonstrating similar cross-talk between RTKs and oncogenes phenotypes may indicate additional relevance to our studies in other cancers (55–57).

Two clinical trials of an ALK inhibitor plus an EGFR inhibitor have been conducted. Both studies utilized crizotinib as a MET inhibitor with the intent of treating MET-driven acquired resistance in patients with EGFR-mutant lung cancer. A phase I study of erlotinib plus crizotinib was conducted; however, the ALK and ROS1 status were not collected in this trial to determine whether there was any activity in this population (58). A phase I trial of dacomitinib plus crizotinib was also conducted, but none of the patients were positive for an ALK rearrangement (59). Both studies required dose reduction of crizotinib, thus highlighting the potential for increased toxicities with combinations of TKIs and the need for alternative strategies for inhibiting wild-type EGFR and this work is currently ongoing in our laboratory.

In summary, these studies identify a significant contribution by EGFR signaling to the oncogenic program of four different fusion kinases and demonstrate that EGFR can enable an early, adaptive rescue of signaling in cancer cells whose dominant oncogenes are inhibited. Importantly, exogenous EGF is not necessary for all of the EGFR roles described here in maintaining critical signaling in cancer cells. Notably, all of the EGFR contributions could be abrogated with the EGFR TKI gefitinib. Ongoing work will determine other potential pharmacologic strategies to inhibit WT EGFR in the clinic. Further investigation of inhibition of EGFR in combination with oncogene-specific inhibitors is warranted to determine whether these combinations will deepen and/or prolong initial patient responses, and perhaps ultimately delay the onset of disease progression by acquired resistance.

A.T. Le has ownership interest (including patents) in Abbott Molecular. M. Varella-Garcia has ownership interest (including patents) in a patent to use EGFR tests for selecting patients for targeted therapy. D.L. Aisner reports receiving a commercial research grant from Genentech, has received speakers bureau honoraria from AstraZeneca and Clovis Oncology, and is a consultant/advisory board member for Inivata. R.C. Doebele reports receiving a commercial research grant from Loxo, Threshold, and Ignyta, is a consultant/advisory board member for Ariad, Pfizer, Guardant, Trovagane, AstraZeneca, and Igynta, and has ownership interest (including patents) in Abbott Molecular. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Vaishnavi, R.C. Doebele

Development of methodology: A. Vaishnavi, A.T. Le, M.J. Glogowska, M.A. Smith, S.L. Kako, K.D. Davies, M. Varella-Garcia, E.B. Haura, A. Jimeno, R.C. Doebele

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Schubert, L.A. Marek, S. Keysar, S.L. Kako, N.J. Sumi, K.D. Davies, M. Varella-Garcia, A. Jimeno, D.L. Aisner, R.C. Doebele

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Vaishnavi, L. Schubert, L.A. Marek, A.T. Le, S. Keysar, S.L. Kako, M. Varella-Garcia, R.C. Doebele

Writing, review, and/or revision of the manuscript: A. Vaishnavi, L. Schubert, U. Rix, L.A. Marek, M.A. Smith, S.L. Kako, K.D. Davies, M. Varella-Garcia, E.B. Haura, A. Jimeno, L.E. Heasley, D.L. Aisner, R.C. Doebele

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Vaishnavi, S.L. Kako, K.E. Ware, L.E. Heasley

Study supervision: A.T. Le, R.C. Doebele

Other (creation of critical reagents): U. Rix

The authors sincerely thank Adriana Estrada-Bernal, Chad G. Pearson, Ricardo Pineda, Harshani R. Lawrence, and Barbara A. Helfrich for their assistance with the studies pertaining to this article. We also thank Pasi A. Jänne from the Dana Farber Cancer Institute and Christine M. Lovly from Vanderbilt-Ingram Cancer Center for providing cell lines used in this study. We also would like to thank GlaxoSmithKline and Array Biopharma for providing foretinib (XL-880) and ARRY-470, respectively. We would also like to thank Patrick Chestnut and the Molecular Pathology Shared Resource for their assistance with collection of patient specimens for these studies.

This publication was supported by the NIH/NCI1 R01 CA193935, V Foundation for Cancer Research, NIH/NCIP50 CA058187 (University of Colorado Lung Cancer SPORE to P.A. Bunn, principal investigator), NIH/NCI5 K12 CA086913 to R.C. Doebele; NIH/NCI T32CA190216 to L. Schubert; NIH/NCI1 R01 CA181746 to U. Rix; NIH/NCIR21 CA181848 and Florida Biomedical Research Program 5BC07 to E.B. Haura; and NIH/NCI P30-CA076292 Cancer Center Support Grant to the H. Lee Moffitt Cancer Center Core Facilities.

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.