The EGFR Exon 19 Mutant L747-A750>P Exhibits Distinct Sensitivity to Tyrosine Kinase Inhibitors in Lung Adenocarcinoma Free

Epidermal growth factor receptor (EGFR) mutations associated with clinical response to EGFR tyrosine kinase inhibitors (TKI) were discovered over a decade ago in non–small cell lung cancer (NSCLC), and five TKIs (erlotinib, gefitinib, afatinib, dacomitinib, and osimertinib) are currently approved by the FDA for the first-line treatment of EGFR-mutant NSCLC (1–4). Erlotinib and gefitinib, both first-generation TKIs, bind reversibly to the ATP-binding pocket of the receptor, whereas the second-generation TKIs afatinib and dacomitinib additionally react covalently with the side-chain of cysteine 797 (C797) in EGFR (5). The major mechanism of resistance to first- and second-generation TKIs is a secondary acquired T790M mutation. Osimertinib is a third-generation irreversible TKI which also reacts with C797 and can inhibit EGFR harboring the secondary T790M mutation. Osimertinib has been shown to be effective in the second line for patients with NSCLC harboring EGFR T790M mutations (6), and more recently was shown to improve progression-free survival (PFS) compared with first-generation TKIs in the first-line setting (7). Based on these findings, osimertinib now has FDA approval for first-line treatment of patients with metastatic NSCLC harboring EGFR mutations.

The most common EGFR alterations associated with TKI sensitivity are in-frame deletions in exon 19 and a point mutation in exon 21 (L858R). Together, these account for approximately 90% of all EGFR alterations (8). The most frequently observed EGFR exon 19 deletion leads to elimination of 5 amino acids (E746-A750) between the third β-strand of the EGFR tyrosine kinase domain and its key regulatory αC helix (9). However, a number of other exon 19 deletion mutations have also been observed in NSCLC between amino acids 745 and 753 (10). Many of these deletions start at leucine 747 and are often complex insertion–deletions (indels) leading to replacement of the deleted amino acids with a non-native residue, such as the L747-A750>P and L747-P753>S variants, where proline and serine respectively are introduced (10). Although it is well established that EGFR exon 19 deletion mutant tumors are sensitive to TKIs (11, 12), very little is known about potential differences in TKI sensitivity between individual EGFR exon 19 deletions. One recent study in Ba/F3 cells confirmed sensitivity of several EGFR exon 19 deletions and indels to first-, second-, and third-generation TKIs (13), but suggested subtle differences in TKI sensitivity of individual mutants that have not been explored in detail. The impact of these differences for patient responses has also not been examined, but recent data indicate that they may be clinically important (14). Recent data also suggest that the type of EGFR exon 19 deletion mutation present at baseline is associated with the emergence of a specific osimertinib resistance mutation (15). Here, we investigate differences in TKI sensitivity among the most common EGFR exon 19 deletion mutants. We identify one key variant (L747-A750>P) that is partly refractory to inhibition by erlotinib and osimertinib in engineered, patient-derived and established cell lines, but is strongly inhibited by afatinib. We also report analysis of a Yale patient cohort in which erlotinib-treated patients with tumors harboring the L747-A750>P mutation demonstrated significantly worse outcomes than those with tumors harboring other exon 19 deletions. Our cellular and clinical data thus underscore the importance of analyzing the specific exon 19 mutation present in lung cancers at the time of diagnosis.

Human lung adenocarcinoma cell lines (PC9, H1975, HCC827, and HCC4006) and the patient-derived xenograft (PDX)–derived cell line were cultured in RPMI + L-glutamine (Thermo Fisher Scientific), supplemented with 10% heat-inactivated bovine serum (Thermo Fisher Scientific) and 1% penicillin/streptomycin (Thermo Fisher Scientific). Human embryonic kidney 293T cells were cultured in DMEM + L-glutamine (Thermo Fisher Scientific), supplemented with 10% heat-inactivated bovine serum (Thermo Fisher Scientific) and 1% penicillin/streptomycin (Thermo Fisher Scientific). CHO cells were cultured in DMEM/F12 medium (Gibco), supplemented with 10% FBS and 1% penicillin/streptomycin. PC9, H1975, HCC827, HCC4006, CHO, and 293T cell lines were purchased from the ATCC. Cell lines were routinely tested for mycoplasma contamination. The lung adenocarcinoma cell lines were authenticated by testing for the presence of the expected EGFR mutation. The presence of the EGFR L747-A750>P alteration was confirmed in the PDX-derived cell line using a mutation-specific TaqMan assay. Note that 20 ng of DNA were used in the reaction using assay 12382 (EGFR c.2239_2248>C).

Plasmids carrying cDNAs encoding EGFR mutations (EGFR E746-A750, EGFR L747-S752, EGFR L858R+T790M) and wild-type EGFR were as described previously (3, 16). pcDNA3.1 (-) containing wild-type EGFR was used as a template to generate additional L747-A750>P and L747-P753>S EGFR-mutated plasmids using primers designed through the QuickChange Primer design tool by Agilent Technologies (http://www.genomics.agilent.com/primerDesignProgram.jsp). The following primers were used to introduce the L747-A750>P mutation: forward 5′TAAAATTCCCGTCGCTATCAAGGAGCCAACATCTCCGAAAGCCAACAAGG-3′ and reverse 5′CCTTGTTGGCTTTCGGAGATGTTGGCTCCTTGATAGCGACGGGAATTTTA-3′. The following primers were used to introduce the L747-P753>S mutation: forward 5′ATTCCCGTCGCTATCAAGGAATCGAAAGCCAACAAG-3′ and reverse 5′CTTGTTGGCTTTCGATTCCTTGATAGCGACGGGAAT-3′. Mutations were introduced in full-length EGFR using QuickChange II Site-Directed Mutagenesis (Agilent Technologies). All mutated clones were confirmed using direct Sanger sequencing. The constructs were transiently expressed in 293T human embryonic kidney cells. Briefly, 293T cells were transfected with the plasmids for 36 hours, cells were then serum-starved for 24 hours. After starvation, cells were treated with varying concentrations of erlotinib, afatinib, osimertinib, and dacomitinib for 1 hour and then harvested for immunoblot analysis. Erlotinib and afatinib (obtained from the Organic Synthesis Core Facility at MSKCC), osimertinib (AstraZeneca), and dacomitinib (Selleck) were dissolved in DMSO.

PCR products were sequenced by Sanger sequencing, and sequences and chromatograms were manually reviewed in the forward and reverse directions. The presence of EGFR exon 19 deletions and the EGFR T790M mutation in exon 20 was evaluated using the following primers: EGFR-2074F: 5′-cttacacccagtggagaagc-3′, EGFR-2502R: 5′-caccaagcgacggtcctcca-3′. The presence of the EGFR L858R mutation in exon 21 was confirmed with the following primers: EGFR-2445F: 5′-caactggtgtgtgcagatcg-3′, EGFR-3616R: 5′-cactgcttggtggcgcgac-3′.

293T cells were serum-starved for 24 hours and treated with indicated concentration of TKIs. HCC4006, PC9, H1975, and the YLR086PDX-derived cell lines were treated with TKI for 6 hours before Western blot analysis. Cells were lysed in ice-cold RIPA lysis buffer (50 mmol/L Tris, pH 8.0, 150 mmol/L NaCl, 5 mmol/L MgCl₂, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS) supplemented with protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific #78440). Equal amounts of total protein were separated by SDS-PAGE and probed as indicated. Signals were detected using either SuperSignal West Pico or Femto chemiluminescent substrates (Thermo Scientific). Antibodies for immunoblotting against EGFR (#2232), phospho-EGFR-Y1068 (#3777), and the secondary anti-rabbit HRP antibody (#7074) were from Cell Signaling Technology. The β-actin antibody (#47778) was from Santa Cruz Biotechnology. All antibodies were used at the dilutions suggested by manufacturer.

Cells were plated at 2,000 cells/well (PC9 and H1975), 8,000 cells/well (HCC4006 and HCC827), or 10,000 cells/well (YLR086) in a 96 well-plate. After 24 hours, cells were exposed to the relevant drugs in media containing serum for 72 hours. Cell viability was measured using 20 μL/well of CellTiter Blue Reagent (Promega, G8081).

Stable CHO cell lines expressing different EGFR exon 19 deletion mutations were obtained after approximately 2 months of G418 selection following electroporation with the relevant pcDNA 3.1(-) derivatives using Ingenio electroporation solution (Mirus Bio LLC) and maintained with 200 μg/mL G418. To analyze EGFR depletion following Hsp90 inhibition, cells were treated with 50 to 500 nmol/L geldanamycin for 24 hours, adding the drug from a 1 mmol/L stock in DMSO. After washing with ice-cold PBS buffer, cells were lysed by scraping in cell lysis buffer (Cell Signaling Technology) supplemented with Complete protease inhibitor (Roche) and PhosSTOP phosphatase inhibitor (Roche) on ice. Whole cell lysates were cleared by centrifugation at 14,000 rpm for 10 minutes at 4°C and analyzed by immunoblotting using antibodies against EGFR, Hsp90, and Grb2 as loading control. Anti-EGFR (D38B1: cat#4267) and Grb2 (cat#3972) antibodies were from Cell Signaling Technology, and Hsp90 monoclonal antibody (AC88:cat#ADI-SPA-830) was purchased from Enzo Life Science. All primary antibodies were diluted 1:1,000. Chemiluminescence detection using horseradish peroxidase–conjugated secondary antibodies was performed using SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Scientific) and a Kodak Image station 440CF (Kodak Scientific).

A starting model was generated using the crystal structure of the EGFR kinase domain [aa 695–1022: (PDB ID: 1M17)] in complex with erlotinib (17). This structure was initially subjected to the “protein preparation wizard” available in the Schrödinger Suite 2017 (18), to remove water molecules and fill in the missing amino acid residues as well as missing hydrogen atoms. Geometry minimization of the subsequent structure (verified by convergence), with and without erlotinib, yielded two corresponding structures that were then used as starting points for generating subsequent models for the wild-type kinase and exon 19 deletion mutants bound to the relevant inhibitors. Full geometry optimizations on the model systems were performed using the Optimized Potentials for Liquid Simulations (OPLS-2005) force field with simulated water (VSGB 2.0 solvation model) available under the Schrödinger Suite 2017 running on Macintosh platform (18, 19). To verify convergence and consistency of the optimizations, a number of examples were reoptimized from multiple starting points; energetic variations of 0.1 kcal/mol or less were found among these calculated structures.

Geometry-optimized structures were used as starting points for Molecular Mechanics/Generalized Born Surface Area (MM-GBSA) modeling studies in the Schrödinger Suite 2017 (18), considering noncovalent interactions only. No structural constrains were imposed on the binding pocket (extending to residues 5.0 Å beyond the ligands) so that geometry reoptimization was possible. MM-GBSA calculations can be used to estimate relative binding affinities for different ligands, although absolute values do not typically agree with experimental binding affinities.

Patients enrolled in Institutional Review Board (IRB)–approved protocols at Yale Cancer Center with a diagnosis of NSCLC whose tumors harbored EGFR exon 19 mutations (E746-A750, L747-P753>S, or L747-A750>P) were identified. Patients who received erlotinib as first-line therapy for advanced disease (as this was the TKI most commonly used during the time period observed), either alone or in combination with an agent not considered to have meaningful clinical activity, were included in this analysis. Patients were excluded if they received systemic therapy for early-stage disease in the year prior to starting erlotinib, had previously received an EGFR TKI at any time, had discontinued TKI treatment due to toxicity, or had insufficient data.

Clinical, pathologic, and molecular characteristics were retrospectively obtained from extraction of medical records. Duration of treatment (DOT) was defined as time from treatment initiation to discontinuation of erlotinib due to any reason other than toxicity. PFS was defined as time from treatment initiation to clinical disease progression (as determined by the treating oncologist) or death, and overall survival (OS) was defined as time from treatment initiation to death. Primary resistance was defined as progression of disease on first imaging after initiation of erlotinib.

PFS, DOT, and OS were compared using a log-rank test. Cox proportional hazard models were used to derive hazard ratios (HR) and 95% confidence intervals (CI) comparing exon 19 mutation groups. Kaplan–Meier estimates were used to construct survival curves and calculate PFS, DOT, and OS. Statistical analysis was performed with GraphPad Prism 7.0 software and R version 3.3.2.

It includes five figures, Supplementary figure legends, and two Supplementary tables.

To study the frequency of uncommon EGFR exon 19 deletion mutations in lung cancer, we first analyzed data from the Catalog of Somatic Mutations in Cancer (COSMIC) that includes information on 4,476 lung tumors with EGFR exon 19 deletions between residues K745 and I759 (http://cancer.sanger.ac.uk/cosmic/). As expected, in this cohort the majority of tumors had the classical EGFR exon 19 deletion mutation E746-A750 (69%). The remaining 31% represent a wide variety of EGFR exon 19 deletion mutations, including several complex indels. In decreasing order of frequency, uncommon deletions in exon 19 (reported in at least 40 samples) were: L747-P753>S (6%), L747-T751 (5%), L747-A750>P (4%), E746-S752>V (3%), L747-S752 (2%), E746-T751>A (1%), and L747-T751>P (1%; Supplementary Table S1). We compared the mutational frequencies reported in COSMIC with those in a Yale cohort of 86 patients with EGFR-mutant lung cancer enrolled on an IRB-approved protocol at our institution. We found that the frequency of EGFR exon 19 deletions was consistent between the two datasets. Of 86 patients in the Yale cohort, 51 (59%) had tumors with EGFR exon 19 deletions, and information regarding the specific exon 19 deletion was available for 44 of these. The common exon 19 deletion (E746-A750) accounted for 34 (77%) of the tumors, whereas 10 tumors (23%) had one of the uncommon deletions in EGFR exon 19 (Supplementary Table S2). The most frequently observed indels were L747-P753>S and L747-A750>P, found in 3 and 2 patients, respectively. Collectively, these are the most commonly observed infrequent EGFR exon 19 indels across both COSMIC and Yale data (Fig. 1A), so we proceeded to investigate their properties further. We also studied the L747-S752 infrequent deletion found in both cohorts.

To study the sensitivity of different EGFR exon 19 deletion mutants to first-generation TKIs, we generated expression constructs encoding E746-A750, L747-P753>S, L747-A750>P, L747-S752, and L858R+T790M (3, 16). The mutated receptors were transiently expressed in 293T cells, which have low levels of endogenous EGFR. Western blotting of cell lysates (Fig. 1B) showed strong constitutive autophosphorylation of the mutated (activated) EGFR variants—although this was reduced in the case of L747-S752, despite similar expression levels, as reported previously (3). Adding erlotinib at 1 μmol/L completely inhibited autophosphorylation of the canonical deletion mutant E746-A750 as well as the L747-P753>S and L747-S752 variants—reducing EGFR phosphorylation to undetectable levels in each case (Fig. 1B). Unexpectedly, however, 1 μmol/L erlotinib only partly reduced phosphorylation of the L747-A750>P variant (Fig. 1B), arguing that the ability of erlotinib to suppress phosphorylation differs depending on the specific exon 19 deletion.

Based on these results, we also compared the ability of the second-generation irreversible TKI afatinib to inhibit each of the deletion mutants. In contrast to erlotinib, 1 μmol/L afatinib suppressed phosphorylation completely in all exon 19 deletion mutants (Fig. 1B). Importantly, afatinib fully suppressed phosphorylation of EGFR in the L747-A750>P variant despite its reduced sensitivity to erlotinib (Fig. 1B). In further studies at different drug doses, we found that erlotinib could not inhibit this variant completely even when added at 10 μmol/L (Fig. 1C), whereas afatinib could completely inhibit L747-A750>P EGFR at concentrations of 0.1 μmol/L (Fig. 1C), and even at the clinically achievable dose of 0.01 μmol/L (ref. 20; Fig. 1D). By contrast, the L858R+T790M mutant, whose phosphorylation can be suppressed by afatinib in vitro, but not at clinically achievable doses in vivo, retained a high level of phosphorylation after treatment with 0.1 or 0.01 μmol/L afatinib (Fig. 1D). The other irreversible second-generation TKI, dacomitinib, was also able to suppress phosphorylation of L747-A750>P EGFR completely, similar to afatinib (Supplementary Fig. S1). Finally, we tested the sensitivity of the L747-A750>P and E746-A750 variants to the third-generation TKI osimertinib. The L747-A750>P variant retained substantial residual phosphorylation even at 10 μmol/L osimertinib (Fig. 1E), whereas the E746-A750 variant was completely inhibited at 1 μmol/L osimertinib.

To explore functional consequences of the L747-A750>P mutation further, we investigated in vitro TKI sensitivity of an established human lung adenocarcinoma cell line HCC4006 that carries this specific exon 19 deletion, specifically comparing it with two lung cancer cell lines that carry the canonical E746-A750 exon 19 deletion (PC9 and HCC827). Consistent with the transient transfection studies, HCC4006 cells showed increased sensitivity to afatinib compared with PC9 (HCC4006 vs. PC9 at 0.001 μmol/L, P = 0.0455) and HCC827 cells (HCC4006 vs. HCC827 at 0.001 μmol/L, P = 0.0094; Fig. 2A), with an IC₅₀ value of 0.0006 μmol/L—which is 5-fold smaller than seen for HCC827 cells (IC₅₀ = 0.004 μmol/L) and 3.3-fold smaller than for PC9 cells (IC₅₀ = 0.002 μmol/L). All three cell lines showed similar sensitivities to the first-generation TKI erlotinib (Fig. 2B). In order of reduced sensitivity, IC₅₀ values were 0.015 μmol/L (HCC827 cells), 0.027 μmol/L (PC9 cells), and 0.033 μmol/L (HCC4006 cells). The sensitivity of HCC4006 cells to the third-generation inhibitor osimertinib (IC₅₀ = 0.006 μmol/L) was also within the same range seen for cells with an E746-A750 deletion (Fig. 2C), which had IC₅₀ values of 0.006 μmol/L (HCC827 cells) and 0.009 μmol/L (PC9 cells). Figure 2D reports the ratio of IC₅₀ values for erlotinib/afatinib (left) and osimertinib/afatinib (right) as a measure of relative efficacy. Whereas afatinib is 14 times more effective at a given dose in PC9 cells and 4 times more effective in HCC827 cells, the ratio is increased to 53-fold in HCC4006 cells with the L747-A750>P mutation (Fig. 2D, left). Afatinib is also 10-fold more effective than osimertinib in HCC4006 cells, whereas it is only 6-fold more effective in PC9 cells and 1.3-fold more effective in HCC827 cells (Fig. 2D, right). Overall, as also shown by Western blotting analysis of phosphorylated EGFR levels in Fig. 2E, HCC4006 cells are most sensitive to afatinib compared with the other TKIs.

To further assess the increased afatinib sensitivity of EGFR harboring the L747-A750>P exon 19 alteration, we studied the case of a patient enrolled on the Yale rebiopsy protocol. One patient, YLR086, whose tumor harbored a L747-A750>P exon 19 alteration, did not have a durable response to erlotinib. The patient, a 60-year-old female never-smoker, was diagnosed in 2013 with stage IV adenocarcinoma of the lung. Molecular analysis of the diagnostic tumor biopsy specimen revealed the presence of the uncommon EGFR exon 19 deletion (L747-A750>P). The patient received first-line systemic therapy with erlotinib, resulting in significant regression of disease in the lung, but progression in the brain. She then received palliative radiotherapy to the hip and brain. Further progression was observed within 2 months, so treatment with chemotherapy (carboplatin and pemetrexed) was initiated, alongside continuation of erlotinib. Imaging after 3 cycles of chemotherapy demonstrated further regression of disease in the lung, but multiple new osseous sclerotic lesions consistent with either treatment response or progressive disease. Imaging after two additional cycles of chemotherapy revealed new liver lesions consistent with metastatic disease and progression of the osseous lesions. Biopsy of a liver mass confirmed the presence of the EGFR exon 19 L747-A750>P deletion mutation (Supplementary Fig. S2A). She received no further TKI therapy.

Further analysis of the liver specimen from this patient revealed that it lacked known potential mechanisms of resistance to TKIs like the EGFR T790M mutation, or amplification of ERBB2 or MET. We were able to generate a cell line derived from a PDX of the liver biopsy sample and confirmed the presence of the EGFR L747-A750>P alteration (Supplementary Fig. S2B). We then investigated erlotinib sensitivity of the YLR086 PDX–derived cell line. Erlotinib failed to completely suppress EGFR phosphorylation in YLR086 cells even at 5 μmol/L (Supplementary Fig. S2C), whereas 0.1 μmol/L erlotinib was sufficient for complete suppression in PC9 (EGFR E746-A750) lung adenocarcinoma cells. H1975 lung adenocarcinoma cells (which harbor EGFR L858R and T790M mutations) are shown as a (resistant) negative control (Supplementary Fig. S2C). The effect of erlotinib on cell viability was also substantially blunted in YLR086 PDX cells compared with PC9 cells, as shown in Supplementary Fig. S2D, with IC₅₀ increased by approximately 20-fold. Afatinib suppressed EGFR phosphorylation more effectively than erlotinib in YLR086 cells at concentrations down to 0.01 μmol/L (Supplementary Fig. S2E) and inhibited cell viability more effectively than erlotinib with an IC₅₀ of approximately 0.06 μmol/L (Supplementary Fig. S2F and S2G). YLR086 cells also exhibited significant resistance to osimertinib (Supplementary Fig. S2E, S2F, and S2H), which suppressed EGFR phosphorylation only when added at 1 μmol/L (Supplementary Fig. S2E) and had an IC₅₀ value in cell viability assays of approximately 0.2 μmol/L (Supplementary Fig. S2F and S2H). Therefore, although the YLR086 PDX–derived cells were more resistant to TKIs overall compared to the other EGFR-mutant cell lines studied, afatinib was the most effective TKI in this cell line. Altogether, our data indicate that the L747-A750>P mutation exhibits partial resistance to both erlotinib and osimertinib, but—importantly—not to afatinib.

We hypothesized that the reduced apparent sensitivity of L747-A750>P EGFR to erlotinib might reflect its destabilization by introduction of a rigid proline between the β3 strand and αC helix—contrasting with the serine introduced instead in the L747-P753>S variant. Erlotinib cycles on and off the EGFR kinase domain very rapidly, limiting its occupancy time in the active site (21)—which in turn defines efficacy. Destabilizing the kinase would reduce this occupancy time further by accelerating conformational fluctuations in the kinase. It was reported that several lung cancer mutations destabilize EGFR, as evidenced by increased dependence on Hsp90 for their stability (22–24). Such destabilization could explain why L747-A750>P EGFR shows reduced sensitivity to erlotinib (which rapidly cycles on/off) but not to afatinib (which binds covalently to EGFR). To test this possibility, we compared levels of different exon 19 deletion variants over a range of concentrations of the Hsp90 inhibitor geldanamycin as described (22). As shown in Supplementary Fig. S3, although several exon 19 deletion variants did increase sensitivity of EGFR levels to geldanamycin treatment compared with wild-type EGFR, the effect was not significantly greater for the L747-A750>P mutant. The effects were small, and the most destabilized variant appeared to be E746-A750 EGFR, which retains normal erlotinib sensitivity (Fig. 1B). IC₅₀ values for geldanamycin were 106 ± 20 nmol/L for wild-type EGFR degradation, 44 ± 6 nmol/L for E746-A750 EGFR, and 98 ± 38 nmol/L for L747-A750>P. These results argue that stability per se cannot explain the altered inhibition profile of this variant.

Because reduced EGFR stability does not appear to explain the reduced erlotinib sensitivity of L747-A750>P EGFR, we explored structural consequences of this deletion mutant using molecular dynamics (MD) simulations (see Materials and Methods). We performed simulations of the wild-type EGFR kinase domain plus the E746-A750 and L747-A750>P exon 19 deletion variants—both with and without bound TKIs (erlotinib, afatinib, and osimertinib). As shown in Fig. 3A, the most notable structural change in the deletion mutants is a downward displacement of the glycine-rich (β1/β2) loop by about 3 Å in the L747-A750>P mutant (cyan) and 3.5 Å in the E746-A750 (magenta) mutant compared with wild-type (gray). This change arises from steric clashes of the β1/β2 loop with the β3/αC loop that do not occur in wild-type EGFR. The different displacement in the two deletion mutants arises because of different β3/αC loop conformations, reflecting the effective insertion of “Glu-Pro” into this loop of L747-A750>P compared with E746-A750 (Fig. 3A). Associated with this change is a small displacement of several residues that interact with the bound erlotinib molecule as described by Stamos and colleagues (ref. 17; Fig. 3B). The L788 side-chain is moved away from erlotinib slightly (by ∼1 Å) in L747-A750>P compared with E746-A750—taking it out of van der Waal's contact with the drug. The F723 side-chain makes van der Waal's contact with erlotinib in the E746-A750/erlotinib complex, but is displaced by about 2 Å in the L747-A750>P/erlotinib complex so that this contact is lost. The K745 side-chain is also slightly shifted, but other interactions observed by Stamos and colleagues remain. We infer that this slight overall structural modification at the drug/ATP-binding site of the EGFR kinase domain in the L747-A750>P deletion mutant impairs erlotinib binding. The diminution of erlotinib-binding strength is relatively small, however, as evidenced by the fact that significant inhibition of L747-A750>P EGFR is still clearly seen in Fig. 1B (it is just not complete at 1 μmol/L).

Comparing Fig. 3C and D (and Supplementary Fig. S4) provides one suggestion as to why afatinib binding might be retained more effectively than erlotinib binding in L747-A750>P EGFR. The most notable difference is that the 3-ethynylaniline ring of erlotinib appears to be less intimately “packed” between the side-chains of K745, L788, and T790 in Fig. 3C than does the equivalently located 3-chloro-4-fluoroanilene ring of afatinib in Fig. 3D. Moreover, the position of afatinib's 3-chloro-4-fluoroanilene moiety is stabilized by halogen bonding with the T790 side-chain in both deletion variants. The tetrahydropyranyl ring unique to afatinib (bottom right in Fig. 3D, far left in Supplementary Fig. S4B) is also predicted to form a hydrogen bond with the K728 side-chain in EGFR, but only in models of the L747-A750>P variant (Supplementary Fig. S4B). This interaction is also seen in the crystal structure of the afatinib-bound EGFR kinase (25), but cannot form with erlotinib (17) or osimertinib (26). The resulting increased set of interactions made between the L747-A750>P variant and afatinib, alongside the more intimate aniline ring packing outlined above, argues that slight rearrangement of the drug-binding site in this variant has less negative impact on afatinib binding than on erlotinib binding—as we see experimentally. As a result, the noncovalent afatinib/EGFR complex presumably can last long enough with the L747-A750>P variant for covalent bonding to C797 to proceed efficiently, allowing effective inhibition to be retained. MM-GBSA calculations (see Materials and Methods) further suggested that L747-A750>P EGFR makes a more extensive set of noncovalent interactions with afatinib than with erlotinib (by ∼3 kcal/mol), whereas results for afatinib and erlotinib binding were very similar (within 1 kcal/mol) for wild-type EGFR and other exon 19 deletions.

In parallel, we considered osimertinib binding to the L747-A750>P EGFR variant. As shown in Fig. 3E, osimertinib does not utilize the cavity in the EGFR kinase domain into which the aniline rings of erlotinib and afatinib project, so alterations in this cavity cannot explain the reduced inhibition seen in Fig. 1E. However, movement in the L747-A750>P variant of the β1/β2 loop (Fig. 3A) will reposition residues L718, F723, and V726, which all make van der Waal's contacts with osimertinib to promote binding. Following the 3 Å displacement of the β1/β2 loop marked in Fig. 3A, these three side-chains will clash sterically with bound osimertinib (Fig. 3F) unless it also moves by an equivalent amount—which in turn would break other interactions and reduce binding and covalent interaction with EGFR. Just as moving the β1/β2 loop (including L718) toward the bound osimertinib (in L747-A750>P EGFR) reduces the effectiveness of this inhibitor (data in Fig. 1E) through steric clashes, so does a recent report suggest that “retracting” the L718 side-chain by mutating it to valine impairs osimertinib binding and function (27).

To correlate these in vitro and in silico findings with clinical observations, we investigated responses to erlotinib in patients with tumors carrying EGFR exon 19 deletions. We included patients with tumors harboring an EGFR exon 19 mutation (E746-A750, L747-P753>S, or L747-A750>P) who had received erlotinib as first-line therapy for advanced disease. Between June 2008 and May 2016, 32 patients met criteria for analysis: twenty-four patients harbored the common E746-A750 mutation, 4 had the L747-P753>S mutation, and 4 had the L747-A750>P mutation. Thirty of the 32 patients received erlotinib alone, and 2 received erlotinib in combination with hydroxychloroquine. At the time of analysis, all 32 patients had experienced progression of disease, 30 of 32 (94%) had discontinued erlotinib, and 22 (69%) had died.

Erlotinib-treated patients with tumors harboring the L747-A750>P mutation demonstrated significantly worse outcomes than those with tumors harboring E746-A750 or L747-P753>S mutations (Fig. 4; Supplementary Fig. S5). PFS was significantly shorter for erlotinib-treated patients with tumors with the L747-A750>P mutation than those with tumors with the E746-A750 mutation [median 4.1 months and 11.7 months, respectively; HR, 9.7 (95% CI, 2.5–37.1); P <0.0001; Fig. 4A]. Similarly, the duration of treatment (DOT) was shorter for patients with tumors harboring the L747-A750>P mutation than for those with tumors with the E746-A750 mutation [median 5.9 months and 14.8 months; HR, 10.5 (95% CI, 2.7–40.5); P <0.0001; Fig. 4B]. Overall survival (OS) was also significantly shorter for erlotinib-treated patients with L747-A750>P mutant tumors than those with tumors with the E746-A750 mutation [median 14.1 months and 47.0 months; HR, 10.2 (95% CI, 2.2–47.7); P <0.001; Fig. 4C]. There was no significant difference in outcomes between the E746-A750 and L747-P753>S groups (Supplementary Fig. S5). Importantly, 2 of 4 patients with tumors harboring the L747-A750>P mutation demonstrated primary resistance to erlotinib (i.e., progressive disease as the best response), whereas no patients harboring the E746-A750 or L747-P753>S mutation exhibited primary resistance.

The key finding described here, revealed by biochemical, signaling, and cell viability studies, is that individual EGFR exon 19 deletion mutations can differ in their sensitivity for individual EGFR-targeted TKIs in a potentially clinically significant way. Specifically, we find that the L747-A750>P variant exhibits reduced sensitivity to both the first-generation TKI erlotinib and the third-generation TKI osimertinib—yet retains high sensitivity to the second-generation TKIs afatinib and dacomitinib in cell lines. Importantly, these observations are likely to underlie our finding that patients with EGFR L747-A750>P mutant tumors who were treated with erlotinib have inferior clinical outcomes compared with patients whose tumors harbor the common E746-A750 mutation.

Indications that there are differences in TKI sensitivity among EGFR exon 19 deletion mutations have begun to emerge from several experimental and clinical studies. In vitro cellular studies of a range of patient-derived EGFR variants have shown different transforming potential and TKI sensitivities (13, 28). Moreover, patients with tumors harboring exon 19 deletions in which the LRE motif (between L747 and E749) was included in the deleted region were reported to have better clinical outcomes than those with mutations that do not eliminate this motif (29). Further, when EGFR exon 19 deletion mutant tumors were subdivided into those with the deletion beginning at E746 versus those beginning at L747, the former were found to have better PFS than the latter in one study (30)—although a different study did not reach this conclusion (31). A recent report also showed that the G724S osimertinib-resistance EGFR mutation emerges in the context of specific EGFR exon 19 deletion mutations (15). Together with our results, these data argue that differences between individual exon 19 deletions in EGFR should be examined in detail. The findings also underscore the fact that not all EGFR mutations are the same—highlighting the importance of mutation-specific EGFR-TKI selection. Indeed, it has been shown that the precise nature of deletions in the structural-equivalent (β3/αC loop) region of BRAF and ERBB2 affects drug selectivity (32), arguing that this may be a general property of therapeutically targetable oncogenic kinases.

In addition to our observations with the L747-A750>P variant, analyses of several previous clinical case studies suggest that variants in which L747 in the β3/αC loop of EGFR is replaced with a proline are associated with primary resistance to first-generation TKIs (33–35). Based on these clinical findings plus our experimental and modeling studies, it seems likely that (as in L747-A750>P) the presence of a proline at this position displaces the adjacent β1/β2 glycine-rich loop (Fig. 3A and B) and thus impairs drug binding. Interestingly, EGFR insertions in exon 19 have been described that also include a proline mutation at position 747 (36). Similar to our findings with the L747-A750>P mutation, exon 19 insertion mutations exhibit sensitivity to afatinib and partial sensitivity to erlotinib. In light of our studies, more in-depth analyses of these different mutations and their TKI sensitivity patterns in patients are warranted. Further, it is interesting that the L747-A750>P variant is most sensitive to a second-generation irreversible inhibitor (afatinib) and less sensitive to both a reversible inhibitor (erlotinib) and an irreversible inhibitor (osimertinib) in cell lines. Understanding the reduced sensitivity to erlotinib seems relatively straightforward—the altered binding site impairs kinase occupancy and thus inhibition. For the irreversible/covalent inhibitors, the difference between osimertinib and afatinib underscores the need for noncovalent interactions to drive kinase occupancy for a time sufficient to allow reaction of the inhibitor with the C797 side-chain. These findings have potential implications for drug design since they indicate that the types of covalent and noncovalent interactions formed by a drug with different mutant EGFR molecules need to be considered.

We demonstrated that patients with tumors with the L747-750>P mutation have inferior outcomes compared with those harboring the canonical E746-A750 mutation when treated with a first-generation EGFR TKI. Limitations of our study include the small sample size and the lack of clinical outcome data for second- and third-generation EGFR TKIs in patients harboring the L747-A750>P mutation. Ongoing and future investigations attempting to validate these findings in larger cohorts including patients with tumors with uncommon exon 19 mutations treated with second- and third-generation EGFR TKIs will be important. An interesting question that remains is whether osimertinib is optimally effective in the context of the EGFR T790M mutation (which is the most common resistance mechanism for both first- and second-generation EGFR TKIs) in tumors with uncommon exon 19 mutations.

In conclusion, our study demonstrates that the presence of the L747-A750>P deletion is associated with exquisite sensitivity to afatinib and reduced sensitivity to the first-generation TKI erlotinib and third-generation TKI osimertinib in preclinical models. Our study provides evidence that the precise type of EGFR exon 19 deletion mutation influences sensitivity to different EGFR-targeted TKIs. Our findings argue in favor of analyzing the specific exon 19 mutation at the time of diagnosis. Moreover, particularly in light of recent studies demonstrating increased PFS with osimertinib treatment over first-generation EGFR TKIs, our findings argue that there may be an important role for second-generation EGFR-targeted TKIs in patients with certain uncommon exon 19 deletions, although further studies in patients are warranted. Indeed, while all of these TKIs retain some activity against the L747-A750>P mutant (especially osimertinib), incomplete suppression of EGFR activity by first- and third-generation TKIs could affect the long-term clinical benefit of these therapies. Further investigation of these issues will help define optimal treatment strategies for patients with tumors harboring various EGFR exon 19 deletion mutations.

S.B. Goldberg reports receiving commercial research grants from AstraZeneca and is a consultant/advisory board member for AstraZeneca, Boehringer Ingelheim, Bristol-Myers Squibb, Eli Lilly, Amgen, and Spectrum. K. Politi reports receiving commercial research grants from AstraZeneca, Roche, Symphogen, and Kolltan; holds ownership interest (including patents) in MolecularMD/MSKCC; and is a consultant/advisory board member for AstraZeneca, Merck, Tocagen, Dynamo Therapeutics, Maverick Therapeutics, and NCCN. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Truini, Z. Walther, J.H. Park, S. Gettinger, D. Zelterman, M.A. Lemmon, S.B. Goldberg, K. Politi

Development of methodology: M. DeVeaux, S. Gettinger, D. Zelterman, S.B. Goldberg

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Truini, J.H. Starrett, T. Stewart, Z. Walther, A. Wurtz, D. Lu, J.H. Park, X. Song, S. Gettinger, S.B. Goldberg

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Truini, J.H. Starrett, T. Stewart, K. Ashtekar, Z. Walther, D. Lu, J.H. Park, M. DeVeaux, S. Gettinger, D. Zelterman, M.A. Lemmon, S.B. Goldberg, K. Politi

Writing, review, and/or revision of the manuscript: A. Truini, J.H. Starrett, T. Stewart, K. Ashtekar, D. Lu, J.H. Park, M. DeVeaux, X. Song, S. Gettinger, M.A. Lemmon, S.B. Goldberg, K. Politi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Politi

Study supervision: M.A. Lemmon, K. Politi

This work was supported by NIH/NCI grants P50 CA196530 (K. Politi, S.B. Goldberg, M.A. Lemmon, S. Gettinger, and D. Zelterman) and R01 CA198164 (M.A. Lemmon), T32 CA193200 (J.H. Starrett), an Italian Association for Cancer Research (AIRC) fellowship (A. Truini), and an Arnold and Mabel Beckman Foundation Postdoctoral fellowship (K. Ashtekar).

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.