Sequential ALK Inhibitors Can Select for Lorlatinib-Resistant Compound ALK Mutations in ALK-Positive Lung Cancer

Chromosomal rearrangements of the ALK gene define a distinct molecular subset of non–small cell lung cancer (NSCLC; refs. 1, 2). ALK rearrangements lead to expression of constitutively activated ALK fusion proteins that function as potent oncogenic drivers. Since the discovery of ALK rearrangements in NSCLC one decade ago, numerous ALK inhibitors have been developed for the treatment of patients with advanced ALK-rearranged (i.e., ALK-positive) NSCLC (3). Until recently, the first-generation ALK inhibitor crizotinib was the standard therapy for newly diagnosed ALK-positive NSCLC, inducing responses in over 70% of patients with a median duration of response of approximately 11 months (4). For patients relapsing on crizotinib, more potent second-generation ALK inhibitors, such as ceritinib, alectinib, and brigatinib, have become standard treatments, reinducing responses in the majority of crizotinib-resistant patients (5–8). Several recent randomized trials have demonstrated that second-generation ALK inhibitors may be more effective than crizotinib as first-line therapy (9–11). However, regardless of when patients receive second-generation ALK inhibitors, resistance almost always develops, leading to clinical relapse.

Multiple different molecular mechanisms can cause resistance to second-generation ALK inhibitors (12–15). In the largest study to date of clinically resistant specimens, one half of cases resistant to a second-generation ALK inhibitor harbored ALK resistance mutations (i.e., on-target resistance; ref. 12). The most common ALK mutation emerging on all second-generation inhibitors was the solvent front ALK^G1202R substitution, accounting for approximately one half of on-target resistance. Other less common ALK resistance mutations that were identified included ALK^I1171 mutations with alectinib and ALK^F1174 mutations with ceritinib. Importantly, based on analysis of patient-derived cell lines, those cancers harboring ALK resistance mutations remained ALK-dependent and sensitive to the pan-inhibitory, third-generation ALK inhibitor lorlatinib. In contrast, those cancers without identifiable ALK resistance mutations were ALK-independent and lorlatinib-insensitive, with resistance likely mediated by off-target mechanisms such as bypass signaling or lineage changes (12).

Consistent with these and other preclinical studies (16, 17), lorlatinib has demonstrated clinical activity in resistant ALK-positive patients previously treated with two or more ALK inhibitors, including a second-generation inhibitor (18). In phase I testing, lorlatinib demonstrated a confirmed response rate of 42% and a median progression-free survival of 9.2 months in the subset of ALK-positive patients who had failed two or more ALK inhibitors. On the basis of repeat biopsies taken prior to lorlatinib, the presence of an ALK resistance mutation such as ALK^G1202R predicted for clinical response to lorlatinib (18). The promising antitumor activity seen with lorlatinib has led to FDA breakthrough therapy designation and has solidified the sequential treatment approach using first- and/or second-generation ALK inhibitors followed by lorlatinib.

As with other ALK inhibitors, acquired resistance to lorlatinib develops in essentially all patients. We previously reported a case of a patient with advanced ALK-positive lung cancer who had been treated with sequential crizotinib, ceritinib, and lorlatinib (19). Molecular analysis of this patient's lorlatinib-resistant specimen revealed a novel compound resistance mutation, ALK^{C1156Y/L1198F}, with both mutations residing on the same allele of the ALK fusion gene. ALK^{C1156Y/L1198F} conferred resistance to lorlatinib but paradoxically resensitized the cancer to crizotinib. Indeed, the patient experienced a dramatic re-response when rechallenged with crizotinib (19). Outside of this single case report, no other mechanisms of resistance to lorlatinib have been described, hindering our ability to develop effective therapeutic strategies for patients who relapse on lorlatinib.

Here, we report the results of preclinical studies aimed at discovering clinically relevant mechanisms of resistance to lorlatinib. Utilizing in vitro cell-based accelerated mutagenesis screens, we have identified numerous compound (but not single) ALK mutations that confer high-level resistance to lorlatinib. By analyzing a series of repeat biopsies taken from lorlatinib-resistant patients, we demonstrate that these compound ALK mutations develop in a stepwise fashion in patients treated with sequential ALK inhibitors. Our results identify novel compound ALK mutants and highlight the importance of therapeutic strategies aimed at preventing the emergence of highly refractory compound mutations.

To predict ALK mutations conferring resistance to lorlatinib, we employed accelerated N-ethyl-N-nitrosourea (ENU) mutagenesis screening of Ba/F3 models of ALK-positive cancer. We utilized established Ba/F3 cell lines expressing either nonmutant EML4–ALK to model ALK inhibitor–naïve disease, or mutant EML4–ALK harboring a single ALK resistance mutation to model resistant disease after first- or second-generation ALK inhibitor. As previously reported (12), these Ba/F3 models exhibit different sensitivities to different ALK inhibitors, with lorlatinib retaining significant potency against all models, including EML4–ALK^G1202R (Fig. 1A).

To determine whether any single ALK mutations are capable of conferring resistance to lorlatinib, Ba/F3 cells expressing nonmutant EML4–ALK [either variant 1 (E13; A20) or variant 3a (E6a; A20), which represent the two most common EML4–ALK variants in NSCLC] were chemically mutagenized with ENU, a potent inducer of point mutations (20). After treatment with ENU, mutagenized cells were cultured in the presence of a range of crizotinib or lorlatinib concentrations (100 nmol/L–1,000 nmol/L) that simulate clinical drug exposures (Fig. 1B). Emerging resistant clones were isolated and DNA was sequenced to identify ALK kinase domain mutations. As 100 nmol/L of crizotinib was insufficient to prevent growth of nonmutagenized cells, this dose of crizotinib was excluded from our analysis; long-term cell proliferation assays confirmed that all other concentrations of crizotinib and lorlatinib prevented the outgrowth of nonmutagenized cells (Supplementary Fig. S1), and results are summarized below.

As shown in Fig. 1C, we observed numerous resistant clones emerging after treatment with 300 to 600 nmol/L of crizotinib. These crizotinib-resistant clones harbored a variety of single ALK kinase domain point mutations, including the majority of crizotinib resistance mutations identified in clinical specimens (12, 21–23). In contrast, no resistant clones emerged after treatment with 300 to 600 nmol/L of lorlatinib (Fig. 1C). This range of drug concentrations is comparable with the plasma exposures achieved in vivo, as most patients treated at standard-dose lorlatinib have unbound drug levels exceeding 369 nmol/L (18). Of note, a small number of clones did emerge after treatment with 100 nmol/L of lorlatinib and were found to harbor predominantly ALK^I1171N or, less commonly, ALK^L1196M mutations. Of note, Ba/F3 cells expressing these two particular ALK mutants were slightly less sensitive to lorlatinib compared with other ALK mutants, as shown by the relatively higher IC₅₀s in cellular assays (Fig. 1A). However, these mutants were fully suppressed at higher (and clinically achievable) lorlatinib concentrations (Fig. 1C). These results are consistent with previous studies that failed to identify a single ALK mutation sufficient to confer high level resistance to lorlatinib and suggest that lorlatinib treatment may be able to suppress the emergence of all single ALK resistance mutations.

Many patients will have disease progression on a second-generation ALK tyrosine kinase inhibitor (TKI) prior to lorlatinib, and about half of these cancers are likely to have on-target ALK resistance mutations (12). To recapitulate this clinical setting, we performed ENU mutagenesis screening of EML4–ALK-expressing Ba/F3 cells each harboring one of the most common ALK resistance mutations observed after failure of first- and second-generation ALK inhibitors (C1156Y, F1174C, L1196M, G1202R, and G1269A). Because each single EML4–ALK mutant exhibits a different sensitivity to lorlatinib (Fig. 1A), we first determined the range of lorlatinib concentrations sufficient to prevent clonal outgrowth using long-term cell proliferation assays (Fig. 2A; Supplementary Fig. S2). After ENU mutagenesis, Ba/F3 cells were cultured in the presence of lorlatinib at concentrations that prevented the outgrowth of each single mutant. The minimum concentrations of lorlatinib used in the screen were 50 nmol/L for C1156Y and F1174C, 100 nmol/L for G1269A, and 300 nmol/L for L1196M and G1202R.

For each single ALK-mutant model, 12 to 49 lorlatinib-resistant clones were identified, each harboring a compound ALK mutation, that is, two mutations on the same allele (Fig. 2B; Supplementary Fig. S3). The exact compound mutations and the lorlatinib concentrations at which they emerged are shown in Fig. 2B. Mutagenesis of C1156Y or L1196M Ba/F3 cells yielded lorlatinib-resistant clones harboring a multitude of different compound ALK mutations. Among the nine different compound ALK^C1156Y mutations, one was ALK^{C1156Y/L1198F}, which we previously identified in a lorlatinib-resistant patient (19). Similarly, the L1196M model yielded eight different compound ALK^L1196M mutations, two of which included L1198 mutations. The remaining three ALK mutant models, F1174C, G1269A, and G1202R, yielded a smaller spectrum of compound ALK mutations (Fig. 2B). For example, mutagenesis of G1202R Ba/F3 cells yielded lorlatinib-resistant clones harboring primarily the ALK^{G1202R/L1196M} compound mutation. This compound mutation was the only one to emerge at the highest concentration of lorlatinib (1,000 nmol/L), suggesting it may represent a highly recalcitrant lorlatinib-resistant mutation. This mutation was also subsequently identified in a lorlatinib-resistant patient (see below). Overall, these results reveal a broad spectrum of compound ALK mutations that can mediate on-target resistance to lorlatinib.

In a parallel approach to identifying on-target mechanisms of resistance to lorlatinib, we treated sensitive H3122 cells with increasing concentrations of lorlatinib for over 4 months until resistance emerged (Supplementary Fig. S4A). Three independent resistant cell lines (H3122 LR-A, LR-B, and LR-C) were ultimately derived and maintained in 1 μmol/L lorlatinib. All three cell lines were resistant to lorlatinib in cell viability assays (Supplementary Fig. S4B), and none harbored a secondary mutation in the ALK tyrosine kinase domain. Of note, in similar studies of crizotinib resistance, we previously identified the ALK^L1196M gatekeeper mutation in H3122 cells made resistant to crizotinib in vitro (24).

To model resistance to lorlatinib in vivo, we generated subcutaneous tumors from the sensitive EML4–ALK variant 1 cell line MGH006. Tumor-bearing mice were treated with lorlatinib by oral gavage, leading to tumor regression for more than 50 days, as reported previously (17). With continued lorlatinib treatment, three of six tumors showed regrowth consistent with the emergence of resistance (Supplementary Fig. S4C). Three cell lines (MGH006 LR-B1, G3, and J2) were derived from the resistant tumors. As with the in vitro generated models, all three cell lines were resistant to lorlatinib in cell culture (Supplementary Fig. S4D), and none harbored an ALK resistance mutation. Taken together, these results support the notion that no single ALK mutations confer resistance to lorlatinib.

To confirm that the compound ALK mutations identified by ENU mutagenesis screening confer resistance to lorlatinib, we independently generated Ba/F3 cell lines expressing three of the identified compound mutations, ALK^{G1202R/L1196M}, ALK^{G1202R/L1198F}, and ALK^{L1196M/L1198F}. Cells were treated with crizotinib, ceritinib, alectinib, brigatinib, or lorlatinib, and cell viability was determined after 48 hours. Compared with ALK^G1202R or ALK^L1196M single mutants, the compound ALK^{G1202R/L1196M} mutant conferred high-level resistance to lorlatinib (IC₅₀ 1,116 nmol/L vs. 37 or 18 nmol/L with G1202R or L1196M, respectively; Fig. 3A). The ALK^{G1202R/L1196M} double mutant was also resistant to all first- and second-generation ALK inhibitors tested (Fig. 3B). In contrast, ALK^L1198F-containing compound mutants were resistant to lorlatinib and second-generation ALK inhibitors, but exhibited sensitivity to crizotinib, as shown previously (Supplementary Fig. S5A and S5B; ref. 19). For all of these studies, similar results were obtained using either variant 1 or variant 3 of EML4–ALK (Supplementary Fig. S5A and S5B).

We next assessed biochemical inhibition of compound ALK mutants by examining ALK phosphorylation across the different Ba/F3 models treated with lorlatinib. Consistent with the results of cell viability assays, lorlatinib treatment suppressed ALK phosphorylation of nonmutant ALK and single ALK mutants, but failed to suppress ALK phosphorylation of the ALK^{G1202R/L1196M}, ALK^{G1202R/L1198F}, and ALK^{L1196M/L1198F} compound mutants (Fig. 3C; Supplementary Fig. S6). Thus, the compound ALK mutants identified by ENU mutagenesis screening maintain ALK activation in the presence of lorlatinib.

To validate the findings made in Ba/F3 cells, we engineered the sensitive EML4–ALK v1 lung cancer cell line H3122 to overexpress EML4–ALK^{G1202R/L1196M}. Consistent with the Ba/F3 data, H3122 cells overexpressing ALK^{G1202R/L1196M} were highly resistant to lorlatinib compared with H3122 cells overexpressing ALK^G1202R (IC₅₀ 2,253 nmol/L vs. 46 nmol/L, respectively; Supplementary Fig. S7A). In addition, lorlatinib treatment failed to suppress ALK phosphorylation of the compound ALK mutant, but was able to suppress ALK phosphorylation of both the nonmutant and single ALK^G1202R-mutant H3122 cells (Supplementary Fig. S7B).

To identify clinical mechanisms of resistance to lorlatinib, we performed repeat biopsies of resistant tumors in 20 patients with ALK-positive lung cancer relapsing on lorlatinib. As shown in Table 1, these patients may have had primary (or intrinsic) resistance to lorlatinib, or have developed resistance after an initial response to lorlatinib (i.e., acquired resistance). Nineteen of the 20 patients had received two or more ALK inhibitors, including crizotinib and at least one second-generation ALK inhibitor; the remaining patient had received the second-generation ALK inhibitor brigatinib. Among the 20 cases, 11 had paired pre- and post-lorlatinib specimens (Supplementary Table S1). Pre-lorlatinib specimens were obtained on the ALK inhibitor just prior to lorlatinib, whereas post-lorlatinib specimens were obtained at the time of relapse on lorlatinib. Clinical history is summarized in Supplementary Table S2.

All lorlatinib-resistant specimens underwent standard histopathology and molecular profiling using either the MGH SNaPshot next-generation sequencing (NGS) assay (25) or the FoundationOne platform (Table 1; Supplementary Table S3). All samples showed NSCLC histology with no evidence of small cell transformation. As shown in Table 1, 7 of the 8 (88%) patients with primary resistance to lorlatinib had no detectable mutations within the ALK tyrosine kinase domain. In the 4 patients who had paired pre-lorlatinib specimens, none harbored ALK resistance mutations, suggesting the presence of ALK-independent mechanisms of resistance. In contrast, among the 12 patients with acquired resistance to lorlatinib, six (50%) developed compound ALK resistance mutations, including four with a double ALK mutation and two with a triple ALK mutation. In four cases with paired pre- and post-lorlatinib specimens, the pre-lorlatinib specimen harbored a single or double ALK resistance mutation present in the compound mutant post-lorlatinib specimen, suggesting a stepwise accumulation of ALK resistance mutations during lorlatinib therapy. Interestingly, in 2 patients with acquired resistance, their pre-lorlatinib tumors harbored the lorlatinib-sensitive ALK resistance mutations I1171N and G1202R, which were subsequently “lost” or no longer detectable post-lorlatinib.

In addition to ALK, a variety of co-occurring mutations were detected in lorlatinib-resistant specimens, with the most common being TP53 mutations in 10 of the 20 cases (Fig. 4). Among the co-occurring mutations identified in samples lacking ALK mutations, several were potential drivers of ALK-independent resistance, including a MAP3K1 mutation in MGH098 and an activating NRAS^G12D mutation in MGH9107. Of note, the NRAS^G12D mutation was most likely acquired, as it was not detected in the patient's paired pre-lorlatinib specimen (Supplementary Table S3).

In one patient (MGH065), we identified a single ALK resistance mutation, ALK^G1269A, at the time of lorlatinib relapse. A different ALK mutation, ALK^L1196M, was identified in this patient's pre-lorlatinib sample. ALK^G1269A is predicted to be sensitive to lorlatinib in biochemical and cellular studies (12), raising the possibility that an off-target mechanism(s) of resistance driving ALK-independent growth developed in a G1269A-mutant clone. To test this hypothesis, we established a cell line from the lorlatinib-resistant biopsy specimen (MGH065-3H). This cell line was resistant to crizotinib, ceritinib, and lorlatinib in cell growth assays (Supplementary Fig. S8A and S8B). On the basis of immunoblotting experiments, ALK phosphorylation was potently suppressed by ceritinib and lorlatinib (Supplementary Fig. S8C), consistent with an ALK-independent mechanism(s) of resistance.

On the basis of in vitro mutagenesis screening, preclinical modeling of acquired resistance, and molecular analysis of resistant patient samples, compound but not single ALK resistance mutations can cause resistance to lorlatinib. To determine the evolutionary origin of compound ALK resistance mutations in patients treated with sequential ALK inhibitors, we performed whole-exome sequencing (WES) on serial biopsy samples from 3 patients. The first patient, MGH953, was selected for further analysis as her lorlatinib-resistant tumor harbored the compound ALK^{G1202R/L1196M} mutation also identified by in vitro mutagenesis screening. This patient had received treatment with sequential first-, second-, and third-generation ALK inhibitors. Although she had derived clinical benefit from each inhibitor, she developed resistance to each drug within 5 to 8 months (Fig. 5A). She underwent biopsy and molecular profiling at the time of each relapse, with no ALK mutation detected at diagnosis or post-crizotinib, ALK^G1202R identified in a post-alectinib cell line derived from a malignant pleural effusion, and both ALK^G1202R and ALK^L1196M mutations detected in cis in a post-lorlatinib malignant pleural effusion (Table 1; Fig. 5A).

We next performed WES of this patient's tumor and normal samples. We detected the compound ALK^{G1202R/L1196M} mutations in cis in the lorlatinib-resistant sample, and the single ALK^G1202R mutation in the alectinib-resistant sample. The ALK^L1196M mutation was not detectable in any samples except for the lorlatinib-resistant specimen. The ALK^G1202R mutation was detectable in both the alectinib- and lorlatinib-resistant samples, but not the crizotinib-resistant specimen. Clonal analysis revealed a dominant clone in the pretreatment specimen characterized by a set of 40 truncal mutations (Fig. 5B). This clone gave rise to a crizotinib-resistant subclone harboring an additional 25 mutations, and subsequently an alectinib-resistant subclone harboring 21 additional mutations, including ALK^G1202R (Fig. 5B). With chronic exposure to lorlatinib, this ALK^G1202R subclone eventually evolved to acquire an additional 127 mutations, including ALK^L1196M. This double ALK^{G1202R/L1196M}-mutant subclone most likely led to the patient's relapse on lorlatinib.

The second patient, MGH086, had also been treated with sequential first-, second-, and third-generation ALK inhibitors and underwent multiple repeat biopsies throughout his course of disease (Fig. 5C). We previously reported this patient's clonal evolution of resistance to sequential crizotinib followed by brigatinib (12). Clonal analysis demonstrated that an ALK^E1210K-mutant clone emerged on crizotinib and that under the selective pressure of brigatinib, this clone subsequently gave rise to two brigatinib-resistant compound ALK^E1210K-mutant subclones, ALK^{E1210K/S1206C} and ALK^{E1210K/D1203N}. These subclones were identified in excisional biopsies of recurrent left axillary disease taken approximately 14 and 21 months after starting on brigatinib, respectively (12). The patient continued on brigatinib and underwent three more excisional biopsies of the same site before switching to lorlatinib, initially in combination with a checkpoint inhibitor (Supplementary Table S2). He had a significant response to lorlatinib lasting over 15 months, at which time he underwent repeat biopsy of a lorlatinib-resistant subcutaneous metastasis. We performed WES on the last three brigatinib-resistant lesions and the one lorlatinib-resistant lesion. Clonal analysis of all sequenced specimens demonstrated that the parental ALK^E1210K clone, which first developed on crizotinib, gave rise to three distinct brigatinib-resistant ALK^E1210K double mutants (Fig. 5D). The dominant subclone ALK^{E1210K/D1203N}, which was detected in three of five brigatinib-resistant samples, initially responded to lorlatinib; however, after chronic exposure to lorlatinib, this subclone acquired a third ALK mutation at residue G1269, likely leading to failure of lorlatinib (Fig. 5D).

Finally, patient MGH987 was treated with four sequential ALK TKIs, crizotinib, alectinib, ceritinib, and lorlatinib, and had relatively brief responses to each TKI (Fig. 5E). Repeat biopsy taken when the patient was relapsing on alectinib demonstrated a known alectinib-resistant mutation, ALK^I1171N. On the basis of this finding, the patient was switched to ceritinib and achieved another clinical response lasting over 7 months. When he relapsed on ceritinib, no biopsy was performed and the patient was transitioned to lorlatinib. He had clinical improvement on lorlatinib, but after 3 months, restaging scans demonstrated worsening disease. Repeat biopsy of the same site and SNaPshot NGS revealed a compound ALK^{I1171N/L1198F} mutation. WES and clonal analysis demonstrated that the ALK^I1171N-mutant clone gave rise to the compound ALK^{I1171N/L1198F}-mutant subclone (Fig. 5F), which is known to confer resistance to lorlatinib (19). Thus, in this patient as well as the prior 2 patients, sequential ALK-targeted therapies fostered the stepwise accumulation of ALK resistance mutations, leading to lorlatinib-resistant double and triple mutants. Although ALK^L1198F-containing double mutants may be sensitive to crizotinib, other compound mutants, particularly those containing ALK^G1202R, are likely to be resistant to all available ALK TKIs.

We selected the compound ALK^{G1202R/L1196M} mutant for more detailed biochemical and structural studies because (1) ALK^G1202R is the most common ALK resistance mutation emerging after failure of second-generation ALK TKIs and (ii) ALK^{G1202R/L1196M} was discovered in our ENU mutagenesis screen and subsequently identified in a lorlatinib-resistant patient. The ALK^L1196M gatekeeper mutation was one of the first crizotinib-resistant mutations identified and was proposed to mediate resistance through steric interference with crizotinib binding (26). Lorlatinib is active against ALK^L1196M, but with some decrement in potency in cellular assays (Fig. 1A) and loss of binding by an order of magnitude compared with nonmutant ALK, as measured by the inhibitory constant (K_i; ref. 16). To determine how the co-occurrence of L1196M and G1202R leads to lorlatinib resistance, we first compared the cocrystal structures of nonmutant ALK and ALK^L1196M bound to lorlatinib (Fig. 6A). Aside from the ALK^L1196M substitution and movement in the ATP-phosphate binding loop (P-loop) leading to a maximal separation of 3.1 Å at residue 1126, the structures of the nonmutant ALK and ALK^L1196M kinase domains were highly similar. Binding of the ligand lorlatinib was also almost identical (Fig. 6A). Thus, the decreased binding of lorlatinib to ALK^L1196M may be due to subtle overall structural changes or motion of the protein–ligand complex, rather than a clash with the methionine at position 1196. Furthermore, in kinetic assays, the enzymatic activity of the L1196M mutant (k_cat/K_M,substrate, a measure of catalytic efficiency) was increased approximately 3- to 5-fold compared with nonmutant ALK (Supplementary Table S4). Taken together, these results suggest that the decreased potency of lorlatinib against ALK^1196M may reflect both increased enzyme activity and decreased drug binding.

We next performed molecular dynamics (MD) simulations on the single ALK^G1202R mutant and the ALK^{G1202R/L1196M} double mutant with lorlatinib bound in the ATP pocket. MD simulations suggested that the orientation of the polar portion of the side chain of ALK^G1202R was directed into solvent due to interactions with residue E1210, at times forming 0, 1, or 2 hydrogen bonds. On average, the simulations revealed about 1 hydrogen bond between E1210 and G1202R. Initially, we suspected that the increased bulk of the ALK^G1202R side chain might perturb the ligand position and consequently reduce the quality of hinge interactions to the 2-aminopyridine of lorlatinib. However, there was no difference in the average hydrogen bonding partner distances and angles between lorlatinib and nonmutant ALK, ALK^G1202R, and ALK^{G1202R/L1196M} in the simulations (Supplementary Fig. S9A). Instead, MD simulations suggested that introduction of the arginine at position 1202 increases steric bulk near the pyrazole of the ligand, which destabilizes both the ligand and P-loop above. The net effect of the G1202R mutation is that the pyrazole of lorlatinib moves “up” approximately 0.5 Å above where it is normally seen in the nonmutant ALK or ALK^L1196M structures (Fig. 6B). With G1202R, the P-loop opens wider by a similar amount (0.5 Å) near the pyrazole, and in combination with L1196M, the P-loop opens 0.7 Å near the pyrazole and about 0.4 Å in the inner portion of the P-loop near V1130 (Fig. 6B; Supplementary Fig. S9B). This increased distance between the protein and ligand would be expected to weaken favorable CH-π interactions and van der Waals contacts, and may incur additional ligand strain. The need to adopt this wider conformation may also increase strain in the P-loop or favor dissociation of the ligand. Of note, based on MD simulations, the ROS1 resistance mutation G2032R, which is analogous to ALK^G1202R, may confer resistance to ROS1 inhibitors through a similar mechanism involving P-loop conformational changes leading to shorter residence time (27).

Taken together, these results suggest that ALK^G1202R destabilizes lorlatinib binding due to steric and conformational effects induced by the arginine substitution. Although this leads to a decrease in potency of lorlatinib (Fig. 1A), exposures in the clinic are still sufficient to adequately inhibit the single ALK^G1202R mutant (18). However, when combined with L1196M, which both reduces binding affinity of lorlatinib and enhances the enzymatic activity of ALK, the additive effects of both mutations significantly impair lorlatinib's potency, fueling the emergence of resistance.

The current therapeutic paradigm for patients with advanced ALK-positive NSCLC is to treat with sequential ALK targeted therapies, often moving from first- to second- to third-generation ALK inhibitors (3). Although this treatment approach has dramatically improved clinical outcomes, acquired resistance invariably develops and leads to clinical relapse. In this report, we have focused on resistance to the third-generation ALK inhibitor lorlatinib (16), which is widely anticipated to become a standard therapy for ALK-positive patients after failure of one or more prior ALK inhibitors. Using in vitro cell-based accelerated mutagenesis screening, we identified multiple different compound ALK mutations that confer resistance to lorlatinib, but no single ALK mutations capable of causing high-level lorlatinib resistance. Importantly, in lorlatinib-resistant tumor specimens from patients, on-target resistance mechanisms consisted of only compound and not single ALK mutations. Furthermore, two of the compound ALK mutations identified in our mutagenesis screen, ALK^{C1156Y/L1198F} and ALK^{G1202R/L1196M}, were also identified in patients, validating the utility of this screen in discovering clinically relevant mechanisms of resistance. Consistent with previous work (19), we show that resistance to ALK inhibition is a dynamic and clonal process, with a founder single ALK mutant clone in the pre-lorlatinib tumor giving rise to a lorlatinib-resistant compound ALK-mutant subclone.

The evolution of recurrent on-target resistance mutations in ALK-positive NSCLC is highly reminiscent of other oncogene-addicted cancers. For example, in EGFR-mutant, T790M-positive NSCLC, the third-generation EGFR inhibitor osimertinib is initially highly effective (28), but tumors eventually develop acquired resistance. In approximately 30% of cases, resistance is due to acquisition of the EGFR^C797S mutation, usually in cis with T790M (29). Like ALK^{G1202R/L1196M}, the EGFR mutant containing T790M/C797S is resistant to all available kinase inhibitors. However, in contrast to EGFR-mutant NSCLC in which T790M is essentially the only on-target resistance mutation observed after failure of first- and second-generation inhibitors, numerous (>10) different single ALK kinase domain mutations can emerge in tumors resistant to first- and second-generation ALK inhibitors (12). This multitude of single ALK mutations serves as the substrate for the diverse array of compound ALK mutations driving resistance to lorlatinib. A similar stepwise accumulation of on-target resistance mutations leading to multiple different compound resistance mutations has been observed in chronic myelogenous leukemia (CML) treated with sequential ABL TKIs (30, 31). Thus, resistance in ALK-positive lung cancer may more closely mirror resistance in BCR–ABL-driven CML rather than EGFR-mutant NSCLC.

Our studies on resistance to lorlatinib have several important implications for the field of ALK-positive lung cancer. First, with the increasing availability of second- and third-generation ALK inhibitors, the vast majority of ALK-positive patients will be treated with sequential ALK inhibitors, and approximately 35% will develop compound ALK resistance mutations on lorlatinib. On the basis of in vitro mutagenesis screening, we identified 24 different compound ALK mutations associated with resistance to lorlatinib, two of which were also discovered in patients. Comprehensive cataloging of compound ALK mutations that emerge in the clinic is needed to determine the most common pairings and identify the most promising targets for drug development. As the solvent front ALK^G1202R mutation is the most common kinase domain mutation seen after second-generation ALK inhibitors (12), G1202R-containing compound mutations may become the most common on-target resistance mechanism in patients relapsing after sequential second- and third-generation inhibitors. The compound G1202R/L1196M mutant is refractory to all known ALK inhibitors, and based on structural modeling studies, we predict that other G1202R-containing compound mutations may be similarly recalcitrant. However, not all compound ALK mutations are necessarily refractory to currently available ALK inhibitors. We previously showed that the compound ALK^{C1156Y/L1198F} mutant is resistant to next-generation inhibitors but sensitive to crizotinib (19). In our mutagenesis screening, we identified this and other L1198-containing compound mutations, all of which are predicted to be sensitive to crizotinib. Thus, in patients treated with sequential ALK inhibitors, repeat biopsies at the time of resistance are critical to identify refractory versus resensitizing compound mutations and to select the most effective therapeutic strategies.

Second, as the sequential treatment approach fosters the stepwise accumulation of ALK resistance mutations, culminating in a potentially refractory compound mutation, a more effective therapeutic strategy may be to prevent the early development of single ALK mutations by using pan-inhibitory molecules such as lorlatinib up front. Consistent with previous preclinical studies (17), mutagenesis screening using cell lines expressing nonmutant EML4–ALK failed to identify any single ALK kinase domain mutations capable of conferring high-level resistance to lorlatinib. However, by using cell lines already harboring single ALK resistance mutations, we discovered numerous different compound ALK mutants conferring variable degrees of resistance to lorlatinib. Similarly, among the 20 clinical specimens obtained from lorlatinib-resistant patients, seven harbored compound resistance mutations, including double and triple ALK mutations. Of note, one lorlatinib-resistant case did harbor a single ALK resistance mutation, G1269A. However, analysis of the cell line derived from this case showed that ALK was effectively inhibited by lorlatinib and suggests that alternative, or ALK-independent, signaling pathways were likely driving resistance in this case. Taken together, our results suggest that compound ALK mutations, but not single ALK mutations, may be necessary for cancers to become resistant to lorlatinib. As compound mutants are less likely to emerge in a treatment-naïve, nonmutant ALK molecule, we speculate that up-front treatment with lorlatinib could completely suppress or at least significantly delay on-target resistance, leading to more durable clinical benefit than the current sequential treatment approach.

Third, regardless of the line in which it is used, a broadly potent ALK inhibitor such as lorlatinib may ultimately foster the development of ALK-independent resistance mechanisms. Although roughly 50% of patients relapsing on a second-generation ALK inhibitor may harbor an ALK resistance mutation and remain ALK dependent, the remaining 50% of patients do not harbor ALK mutations (12). These patients' cancers are likely driven by off-target, or ALK-independent, mechanisms of resistance, such as bypass signaling or lineage changes. To date, a variety of different bypass signaling pathways have been reported as mediating resistance to ALK inhibitors, including EGFR, MET, c-KIT, SRC, RAS/MAPK, and SHP2, among others (12, 15, 23, 32, 33). Lineage changes ranging from epithelial-to-mesenchymal transition to small-cell transformation have also been reported in resistant ALK-positive tumors (12, 34, 35). Among the 20 lorlatinib-resistant biopsies we performed, 12 (60%) had no detectable ALK mutations and likely harbored ALK-independent mechanisms of growth and survival. Of these 12 cases, two patients, MGH9092 and MGH040, harbored ALK resistance mutations prior to lorlatinib, exhibited durable responses to lorlatinib, and then “lost” the ALK mutations when they became resistant to lorlatinib (Table 1), suggesting elimination of on-target resistance and outgrowth of an ALK-independent clone. As we optimize therapeutic approaches to overcome and even prevent on-target resistance, cancers will almost certainly develop a diverse array of off-target mechanisms of resistance. Improved understanding of these off-target resistance mechanisms will be critical to developing combination strategies that effectively suppress both on- and off-target resistance.

Over the last decade, numerous experimental systems have been explored to model acquired resistance to targeted therapies in vitro. In this study, we utilized cell-based accelerated mutagenesis screens to predict ALK mutations conferring resistance to lorlatinib. This methodology provides an efficient system for generating many different ALK point mutations and has been used successfully to identify the most common on-target resistance mutation in ROS1-rearranged lung cancer, ROS1^G2032R, analogous to ALK^G1202R (36–38). Similar mutagenesis screens have also been used to define the resistance profiles of imatinib and other ABL inhibitors in CML (39). However, there are several limitations of ENU mutagenesis screening. ENU mutagenesis is an artificial means of inducing point mutations, and ENU is known to lead to specific patterns of single-nucleotide substitutions. In this mammalian cell system, GC to AT, AT to GC, and AT to TA substitutions preferentially occur (40, 41). This mutational bias could lead to overrepresentation of mutations that are unlikely to develop in vivo, or alternatively failure to induce all clinically relevant resistance mutations. Under the conditions of the screen, ENU mutagenesis would also not capture more complicated genetic alterations of ALK such as the triple-compound ALK mutations identified in patients MGH086 and MGH087 (Table 1).

In summary, we have shown through in vitro cell-based mutagenesis screening and molecular analysis of patient samples that treatment with sequential first-, second-, and third-generation ALK inhibitors fosters the development of diverse compound ALK mutations, some of which are highly refractory to all available ALK inhibitors. It is tempting to speculate that up-front treatment with the third-generation inhibitor lorlatinib may be able to prevent the emergence of single and subsequently compound ALK mutations, potentially improving clinical outcomes. This hypothesis will be formally addressed in an ongoing randomized phase III trial comparing lorlatinib with crizotinib as first-line therapy in advanced ALK-positive lung cancer (NCT03052608). However, our data also suggest that highly potent target inhibition with lorlatinib could ultimately select for ALK-independent resistance mechanisms, which may be difficult to overcome once established. Thus, optimal first-line therapy may require development of lorlatinib-based combinations to prevent both ALK-dependent and ALK-independent resistance.

Ba/F3 immortalized murine bone marrow–derived pro-B cells were obtained from the RIKEN BRC Cell Bank (RIKEN BioResource Center) in 2010 and cultured in DMEM with 10% FBS with (parental) or without (EML4–ALK) IL3 (0.5 ng/mL). cDNAs encoding EML4–ALK variant 1 (E13; A20) and variant 3a (E6a; A20) containing different point mutations were cloned into retroviral expression vectors, and Ba/F3 cells were infected with the virus as described previously (24). After retroviral infection, Ba/F3 cells were selected in puromycin (0.7 μg/mL) for 2 weeks. IL3 was withdrawn from the culture medium for more than 2 weeks before experiments. Patient-derived cell lines were established as described previously (23, 32). MGH006 was developed in 2010 from a malignant pleural effusion from a TKI-naïve, ALK-positive patient, and MGH065-3H was developed in 2015 from a lorlatinib-resistant lymph-node biopsy. H3122 was provided by the Center for Molecular Therapeutics at Massachusetts General Hospital (MGH; Boston, MA) in 2010. H3122 and H3122-derived resistant lines were cultured in RPMI-1640 supplemented with 10% FBS. MGH006, MGH006-resistant cell lines, and MGH065-3H were cultured in DMEM with 10% FBS. Cell lines were sequenced to confirm the presence of ALK rearrangement and ALK mutations. Additional authentication was performed by SNP fingerprinting in 2017. The expression vectors of EML4–ALK were transfected with Lipofectamine 3000 (Invitrogen) according to the manufacturer's protocol. Lorlatinib was purchased from Selleck Chemicals; for the H3122 and MGH006 experiments, lorlatinib was provided by Pfizer. Crizotinib and ceritinib were purchased from Selleck Chemicals. Alectinib and brigatinib were purchased from MedChem Express.

The ENU mutagenesis screen protocol was based on procedures published by Bradeen and colleagues and O'Hare and colleagues (39, 42). The cells were exposed to a final concentration of 100 μg/mL ENU (Sigma) for 16 hours and were collected and washed with PBS. After a 24-hour incubation in normal media, the cells were incubated with various concentrations of lorlatinib or crizotinib, in 96-well plates (1 × 10⁶ – 3 × 10⁶ cells/well). Plates were visually inspected for media color change and for cell growth throughout the 4-week experiment. The contents of wells exhibiting cell growth were digested with proteinase K, and ALK kinase domain was PCR-amplified from gDNA with Fast Start PCR Master (Roche). The PCR fragments were sequenced bidirectionally by Sanger sequencing. Each figure represents the combined results of two independent experiments.

Ba/F3 cells (2,000) were plated in triplicate into 96-well plates. Forty-eight hours after drug treatment, cells were incubated with CellTiter-Glo (Promega), and luminescence was measured with a SpectraMax M5 Multi-Mode Microplate Reader (Molecular Devices LLC). GraphPad Prism (GraphPad Software) was used to graphically display data and determine IC₅₀ values by a nonlinear regression model utilizing a four-parameter analytic method.

A total of 2 × 10⁶ Ba/F3 cells were treated in 6-well plates for 6 hours with lorlatinib. Cell protein lysates were prepared as described previously (24). Phospho-ALK (Y1282/1283), ALK, and β-actin antibodies were obtained from Cell Signaling Technology.

All animal studies were conducted in accordance with the guidelines as published in the Guide for the Care and Use of Laboratory Animals. Experiments and were approved by the Institutional Animal Care and Use Committee of MGH. Female Nu/Nu mice aged 6 to 8 weeks were obtained from Charles River Laboratories. Mice were maintained in laminar flow units in sterile filter-top cages with Alpha-Dri bedding.

Patients with ALK-positive NSCLC and disease progression on an ALK inhibitor underwent repeat biopsies of lorlatinib-resistant tumors between November 2014 and October 2017. Standard histopathology was performed to confirm the presence of malignancy. Electronic medical records were retrospectively reviewed to obtain clinical data and treatment histories. All patients provided signed informed consent under an Institutional Review Board–approved protocol. This study was conducted in accordance with the Belmont Report and the U.S. Common Rule.

All post-lorlatinib biopsies were analyzed for ALK resistance mutations using either the MGH SNaPshot NGS platform (25) or the commercially available FoundationOne platform. The MGH SNaPshot platform uses anchored multiplex PCR to detect single-nucleotide variants (SNV) and insertions/deletions within 39 cancer-related genes (version 1) or 91 cancer-related genes (version 2) including ALK (exons 22, 23, and 25 with version 1 or exon 21–23 and 25 with version 2), and to detect copy-number variants in 92–94 cancer-related genes (version 2). This assay can detect SNV and indel variants at ≥5% allelic frequency in target regions with sufficient read coverage.

For WES, genomic DNA was extracted from formalin-fixed paraffin-embedded samples. Whole-exome capture libraries were constructed from 100 ng of extracted tumor and normal DNA following shearing, end repair, phosphorylation, and ligation to barcoded sequencing adapters. Ligated DNA was size-selected for lengths between 200 and 350 bp and subjected to exonic hybrid capture using SureSelect v2 Exome bait (Agilent). Samples were multiplexed and sequenced on multiple Illumina HiSeq flowcells (paired-end 76 bp reads) to average depth of coverage of 41–250× and 90–107× for tumor and normals, respectively.

Massively parallel sequencing data were processed using two consecutive pipelines as described previously (19).

These are described in detail in Supplementary Methods.

J.J. Lin has received speakers bureau honoraria from Chugai and is a consultant/advisory board member for Boehringer Ingelheim. B.J. Burke has ownership interest (including patents) in Pfizer. J.F. Gainor is a consultant/advisory board member for Ariad/Takeda, Array, Bristol-Myers Squibb, Genentech/Roche, Incyte, Loxo, Merck, Novartis, Pfizer, and Theravance. R.S. Heist is a consultant/advisory board member for Boehringer Ingelheim. A.J. Iafrate has ownership interest (including patents) in ArcherDx. C.H. Benes reports receiving a commercial research grant from Novartis and commercial research support from Amgen. J.A. Engelman has ownership interest (including patents) in Novartis. A.N. Hata reports receiving commercial research grants from Novartis and Relay Therapeutics. A.T. Shaw is a consultant/advisory board member for Ariad/Takeda, Blueprint Medicines, Daiichi Sankyo, EMD Serono, Foundation Medicine, Genentech/Roche, Ignyta, KSQ Therapeutics, Loxo, Natera, Novartis, Pfizer, and Taiho. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Yoda, L. Dardaei, H. Hubbeling, J.A. Engelman, A.N. Hata, A.T. Shaw

Development of methodology: S. Yoda, K. Prutisto-Chang, S. Timofeevski, J.K. Lennerz, A.N. Hata

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Yoda, J.J. Lin, B.J. Burke, L. Friboulet, K. Prutisto-Chang, I. Dagogo-Jack, S. Timofeevski, J.F. Gainor, L.A. Ferris, A.K. Riley, K.E. Kattermann, D. Timonina, R.S. Heist, A.J. Iafrate, J.K. Lennerz, M. Mino-Kenudson, A.N. Hata, A.T. Shaw

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Yoda, J.J. Lin, M.S. Lawrence, B.J. Burke, L. Friboulet, A. Langenbucher, L. Dardaei, K. Prutisto-Chang, S. Timofeevski, J.F. Gainor, R.S. Heist, A.J. Iafrate, C.H. Benes, J.K. Lennerz, M. Mino-Kenudson, J.A. Engelman, T.W. Johnson, A.N. Hata, A.T. Shaw

Writing, review, and/or revision of the manuscript: S. Yoda, J.J. Lin, B.J. Burke, L. Friboulet, I. Dagogo-Jack, S. Timofeevski, H. Hubbeling, J.F. Gainor, L.A. Ferris, R.S. Heist, A.J. Iafrate, J.K. Lennerz, M. Mino-Kenudson, J.A. Engelman, T.W. Johnson, A.N. Hata, A.T. Shaw

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Yoda, J.J. Lin, B.J. Burke, H. Hubbeling, L.A. Ferris, A.J. Iafrate, C.H. Benes, J.K. Lennerz, A.T. Shaw

Study supervision: J.A. Engelman, A.N. Hata, A.T. Shaw

This work was supported by grants from the NCI (R01CA164273, to A.T. Shaw), by the National Foundation for Cancer Research (to A.T. Shaw), by an AACR–AstraZeneca Fellowship in Lung Cancer Research (17-40-12-DARD, to L. Dardaei), by Be a Piece of the Solution, by the Evan Spirito Foundation, and by the Targeting a Cure for Lung Cancer Research Fund at MGH.

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.