MAPK Pathway Alterations Correlate with Poor Survival and Drive Resistance to Therapy in Patients with Lung Cancers Driven by ROS1 Fusions

This article is featured in Highlights of This Issue, p. 2769

ROS1 is a proto-oncogene that encodes a receptor tyrosine kinase involved in cell growth and differentiation, and belongs to the insulin receptor subfamily (1). ROS1 gene rearrangements resulting from the fusion of the ROS1 tyrosine kinase domain with various gene partners including CD74, EZR, SDC4, and SLC34A2, causes constitutive activation of downstream pathways and malignant transformation, and occur in 1%–2% of lung adenocarcinomas (2–4). Although signaling activated by ROS1 fusion proteins involve multiple growth and survival pathways, the detailed mechanisms by which ROS1 fusions promote an oncogenic phenotype remain unclear (5–7). The presence of a ROS1 fusion confers susceptibility of tumors to ROS1 tyrosine kinase inhibitors (TKI) such as crizotinib, cabozantinib, and lorlatinib (8). However, several on-target mutations such as G2032R and D2033N kinase domain mutations have already been reported as mechanisms of acquired resistance (AR) to crizotinib and cabozantinib, respectively (9–11). To overcome the resistance to crizotinib caused by on-target mutations, additional ROS1-targeted drugs including lorlatinib, a potent, oral, third-generation TKI directed at ALK and ROS1, is currently being investigated in clinical trials. In a phase I study of patients with ALK or ROS1 fusion–positive lung adenocarcinoma, lorlatinib showed overall response rate of 46% for ALK-rearranged patients (19 of 41 including 19 patients who had received prior ALK TKI) and 50% for ROS1-rearranged patients (6 of 12 including 7 crizotinib-pretreated patients), and was effective for patients harboring ALK G1202R mutation, which analogous to ROS1 G2032R (12). On the other hand, little is known about specific off-target mechanisms of resistance to any ROS1 TKI.

In this study, we analyzed our clinical and genomic data for concurrent ROS1 and other pathway alterations that could potentially impinge upon response of tumor cells to therapy. We found several RAS–MAPK pathway alterations that were present at diagnosis or acquired after resistance to therapy emerged. Using isogenic cells genetically engineered to express a ROS1 fusion, we demonstrate that a novel in frame MEK1 or MEKK1 deletion and knockdown of NF1 block sensitivity to ROS1 TKIs. A combination of lorlatinib and selumetinib was effective in blocking growth of allograft tumors. Our results suggest that genetic alterations in the MAPK pathway represent novel molecular mechanisms mediating either primary or acquired resistance to ROS1 TKIs in patients with ROS1-rearranged lung adenocarcinoma.

In accordance with the Belmont report and following Institutional Review Board/Privacy Board at Memorial Sloan Kettering (MSK) for retrospective review of records and waiver of consent, we identified all patients with lung cancer who had ROS1 rearrangements with/without concurrent MAPK pathway alterations identified by targeted next-generation sequencing. MSK-IMPACT (13) was the primary platform, but results from other methods such as FoundationOne (14) were also included if performed. MSK-Fusion, a custom targeted RNA sequencing (RNA-seq) panel, was used to confirm fusion status in cases with sufficient tissue (15). The frequency of ROS1 rearrangements with/without concurrent MAPK pathway among all patients was queried. To assess the impact of concurrent alterations in advanced disease, patient records with stage IV disease were reviewed to collect demographic, clinical outcome, and molecular data. Overall survival (OS) was defined as date of metastatic diagnosis to date of last follow as of March 15, 2019 or date of death. Fisher exact and log-rank tests were used to assess relationships between patients with/without MAPK alterations and Kaplan–Meier methodology was used to generate overall survival curves.

To generate a fusion linking EZR exons 1- 9 with ROS1 exons 34–43, two gRNAs were designed to target EZR intron 9 and three gRNAs were designed to target ROS1 intron 33 (http://crispr.mit.edu/). The sequence of gRNAs is shown in Supplementary Table S2. gRNAs were cloned into pSpCas9(BB)-2A-GFP (px458; Addgene, plasmid #48138) as described previously (16, 17). To test the efficacy of fusion generation, HEK-293T cells were transfected with each possible pair of gRNAs (one targeting EZR and one targeting ROS1) using Fugene HD reagent and 72 hours later, a pool of transfected cells was harvested for RNA extraction, cDNA synthesis, and RT-PCR to detect the fusion mRNA. Primer sequences and PCR conditions are provided in Supplementary Table S3. After gRNA validation, px458-gRNAs were transfected into HBECp53 cells using FuGENE HD (Promega). After 48 hours, GFP-positive cells were isolated by FACS and seeded at a density of one cell/well into 96-well plates. Three days after plating, the growth media was changed from complete KSFM to DMEM/F12 supplemented with 5% FBS. Clones were serially moved to larger plates and then subjected to RT-PCR for fusion detection. The resulting HBECp53 cell line with an EZR/ROS1 fusion is referred to as HBEC-ER cells, with numbers to indicate individual, distinct clones.

Six-week-old female NOD/SCID gamma mice (Envigo) were used for in vivo study. All mice were cared for in accordance with guidelines approved by the Memorial Sloan Kettering Cancer Center Institutional Animal Care and Use Committee and Research Animal Resource Center. Each cell line (5 × 10⁶ cells) was mixed with Matrigel (50%) and injected into the subcutaneous flank of mice. Tumor volume was calculated using the empirical formula: V = length × width² × 0.52. When tumors reached approximately 100 mm³ (7 days after injection), mice were randomly assigned to 6 groups (n = 5 for each group) that received either vehicle control, 25 mg/kg crizotinib (once daily), 3 mg/kg lorlatinib (once daily), 40 mg/kg selumetinib (once daily), or a combination of 25 mg/kg crizotinib or 3 mg/kg lorlatinib plus 40 mg/kg selumetinib (once daily). Crizotinib and lorlatinib were prepared in 2% DMSO, 30% PEG 300, and water. Selumetinib was prepared in 4% DMSO, 30% PEG 300, 5% Tween 80, and water. Vehicles and drugs were administered orally as a suspension by gavage once daily, on a 4-day-on and 3-day-off-schedule. Mice were observed daily throughout the treatment period for signs of morbidity/mortality. Data were analyzed by Student t test for significance.

HEK-293T cells were directly cultured on 1-well glass slide (Lab-Tek II Chamber Slide System, Thermo Fisher Scientific) for 24 hours, and then were transfected with cDNAs encoding MEK1 wild-type or MEK1delE41_L54. Forty-eight hours after transfection, cells were fixed and permeabilized by methanol at −20°C for 15 minutes. Nonspecific binding was blocked using PBST (PBS with 0.1% Tween 20) supplemented with 1% BSA. Samples were incubated with an anti-ERK1/2 mAb in a humidified chamber at 4°C overnight. After washing by PBST, the slides were counterstained with Hoechst 33342 (Thermo Fisher Scientific). Fluorescent microscopy was used for visualization of the antigen–antibody complexes.

To better understand the transcriptional response evoked by ROS1 fusion protein, we downloaded RNA-seq data, mutation profiles, and clinical information of 510 lung adenocarcinoma cases profiled by The Cancer Genome Atlas via the cBioPortal (http://www.cbioportal.org/). ROS1 fusion status for those cases were determined using the cBioPortal, as a result, seven of 510 had ROS1 fusion. To reduce the statistical bias that may exist between ROS1 fusion–positive (n = 7) and -negative (n = 503) groups, a propensity score (PS)–matched analysis was performed. The PS was calculated by logistic regression based on available factors that potentially confound the association between ROS1 fusion–positive and -negative groups. Six factors that were selected in the PS calculation were age, sex, histology, clinical stage, tumor size, and mutation status in RAS pathway. According to a PS-matched analysis, seven cases were extracted from each group. Gene set enrichment analysis (GSEA) was then performed against the Oncogenic Pathways (18). The GSEA software (GSEA ver. 3.0) downloaded from the GSEA Website (http://software.broadinstitute.org/gsea/index.jsp).

Among 5,470 patients with non–small cell lung cancer who underwent next-generation sequencing via MSK-IMPACT, we identified 75 (1%) patients with ROS1 fusions. Overall, 8 (11%) of the 75 patients and 7 (13%) of 53 patients with metastatic disease had concurrent MAPK pathway alterations in the same tumor specimen. The initial clinical characteristics of the patients stratified by presence or absence of concurrent MAPK alterations were generally similar (Fig. 1A). There were no significant differences in the two groups by age, sex, or smoking status. The most common ROS1 fusion partner in samples with a concurrent MAPK alteration was CD74 (n = 3) followed by SLC34A2 (n = 2). In cases with sufficient tissue, targeted RNA-seq (n = 3) confirmed the presence of the ROS1 rearrangement. MAPK pathway gene alterations included 2 MAP3K1 mutations, 1 MAP2K1 mutation, 1 MAP2K4 mutation, 1 KRAS mutation, and 1 BRAF mutation (Fig. 1B; Supplementary Table S1). In addition, we identified two neurofibromin 1 (NF1) truncating mutations. Among the patients with metastatic disease, 5 patients (ID#1-5) had concurrent MAPK alterations and ROS1 fusions detected at the time of metastatic diagnosis and 2 patients (ID#6, 7) had MAPK alterations detected at the time of resistance to ROS1 TKIs.

The patients received several ROS1-directed therapies, including crizotinib (n = 3), entrectinib (n = 3), lorlatinib (n = 2), TPX-0005 (n = 1), and cabozantinib (n = 1), sequentially before and after standard cytotoxic chemotherapy and immunotherapy (Supplementary Table S1). Patients with advanced stage lung adenocarcinoma with de novo concurrent MAPK alterations at diagnosis typically had minimal or very short time to discontinuation on ROS1 TKIs (Supplementary Table S1). When compared with other patients with ROS1-rearranged advanced lung adenocarcinoma, patients with concurrent MAPK alterations had shorter overall survival despite similar initial clinical characteristics (Fig. 1C). Two patients had two novel MAPK alterations (deletions of several nucleotides in MAP3K1 and MAP2K1 deletions) acquired at the time of resistance to TKIs. Case summaries of the 2 patients with acquired MAPK alterations are provided (Supplementary Case Report). MAP3K1 encodes the protein MEKK1, which has an N-terminal ubiquitin ligase domain and a C-terminal serine/threonine kinase domain, and can activate the ERK and JNK pathways (19). MAP2K1 encodes MEK1, a well-characterized upstream ERK kinase. The schema of the MAPK pathway is shown in Fig. 1D.

To examine the influence of mutations in the MAPK pathway on sensitivity of ROS1 fusion–positive cells to ROS1 TKIs, we established stable isogenic NIH-3T3 cells expressing CD74/ROS1 or SLC34A2/ROS1 fusions using retroviral plasmids harboring the respective chimeric cDNAs. In addition, to faithfully model a ROS1 fusion in lung adenocarcinoma, we also generated an endogenous EZR/ROS1 fusion in HBECp53 cells using CRISPR/Cas9 genome editing system as we have previously done for BRAF fusions (17). A schematic diagram of CRISPR/Cas9 genome engineering is illustrated in Fig. 2A. gRNAs were transfected into HEK-293Ts and then the presence of EZR/ROS1 fusion was examined by RT-PCR in the transfected population. The validation of gRNAs revealed that the pool of cells containing EZR gRNA #2 and ROS1 gRNA #3 generated more detectable fusion mRNA (Supplementary Fig. S1A). We hence used this pairing of gRNAs for the transfection. Three clones of HBECp53 cells that were positive for EZR/ROS1 fusion at the mRNA and protein level were successfully isolated (Fig. 2B and C). Sanger sequencing of the cDNA using primers that cover the fusion junction confirmed that the translocation resulted from the pairing of EZR exon 9 with ROS1 exon 34 (Fig. 2D). Among the three clones isolated, the clone with the highest EZR/ROS1 fusion expression was used for further analysis (HBEC-ER1). Subsequently, we examined the sensitivity of growth and phosphorylation of downstream signaling proteins to ROS1 TKIs in established isogenic ROS1 fusion–positive cells (3T3-CD74/ROS1, 3T3-SLC34A2/ROS1, and HBEC-ER1) to ROS1 TKIs. As shown in Fig. 2E and Supplementary Fig. S1B, growth of cells with a ROS1 fusion was more sensitive to crizotinib and cabozantinib compared with the isogenic control lines. Western blot analysis showed that phosphorylation of ROS1, MEK1/2 and ERK1/2 were inhibited by crizotinib in both 3T3-SLC34A2/ROS1 and HBEC-ER1 cells in a dose-dependent manner (Fig. 2F). On the other hand, phosphorylation of AKT, p70 S6 kinase, and S6 were less inhibited by crizotinib in comparison with ERK or MEK1/2, suggesting that the RAS–MEK–ERK signaling axis is more dependent on ROS1 for activation than the PI3K–AKT pathway (Fig. 2F; Supplementary Fig. S1C). These results prompted us to further explore the importance of MAPK pathway in growth and survival of ROS1 fusion–positive cells.

To assess the changes in downstream signaling caused by expression of ROS1 fusion, the phosphorylation status of key signaling elements in the MAPK pathway and the PI3K/AKT pathway were examined by Western blot analysis. As shown in Fig. 3A and Supplementary Fig. S2A, phosphorylation of MEK1/2 and ERK1/2 were highly induced in ROS1 fusion–positive cells, compared with that of AKT or 4EBP1. S6 phosphorylation more closely mirrored activation of ERK1/2 than AKT in the ROS1 fusion–positive cells. p70 S6 Kinase phosphorylation on the other hand, seems to be either cell line-specific or ROS1 fusion-specific.

To gain insight into the relationship between ROS1 fusion proteins and MAPK pathway, we performed coimmunoprecipitation analysis to look at the association of upstream MAPK activators GRB2 and SOS1 with ROS1. GRB2 is well known to activate MAPK pathway via binding to SOS1, leading to cell proliferation and mitogenesis (20, 21). We found that ROS1 fusion proteins coimmunoprecipitated with GRB2 and SOS1 (Fig. 3B; Supplementary Fig. S2B). In contrast, we could not detect any association between the adaptor protein GAB1 (which links PI3K to growth factor receptors) or the p85 regulatory subunit of PI3K, key upstream mediators of AKT signaling. Knockdown of GRB2 by siRNA caused a reduction in MEK1/2 and ERK1/2 phosphorylation, and inhibited growth of HBEC-ER1 cells (Fig. 3C and D). Furthermore, GSEA analysis revealed that gene sets related to the activation of KRAS signaling were upregulated in ROS1 fusion–positive lung adenocarcinoma samples, compared with ROS1 fusion–negative lung adenocarcinoma (NOM P = 0.00; Fig. 3E). These results indicate that a direct interaction between ROS1 fusion protein with the GRB2–SOS1 complex plays an important role in the activation of MAPK pathway and the tumorigenic properties of ROS1 fusions. Next, to determine the sensitivity of ROS1 fusion–positive cells to MEK and RAF inhibitors, we examined the growth of cells in the presence of trametinib (MEK1/2 inhibitor), selumetinib (MEK1 inhibitor), cobimetinib (MEK1 inhibitor), and a pan-RAF inhibitor, LY3009120. ROS1 fusion–positive cells showed increased sensitivity to the three MEK inhibitors and the pan-RAF inhibitor (Fig. 3F; Supplementary Fig. S2C). Taken together, these results indicate that introduction of ROS1 fusion proteins induces the preferential activation of MAPK pathway via the binding to GRB2–SOS1 complex, and this pathway is essential for the growth of ROS1-driven cancer cells.

To explore whether the novel in-frame deletions MEK1delE41_L54 and MEKK1delH907_C916 that were acquired after ROS1 TKIs induce drug resistance in ROS1 fusion–positive lung adenocarcinoma, we generated isogenic cell line models by transduction with lentiviral vectors. MEK1 wild-type and MEK1delE41_L54 were overexpressed in 3T3-SLC34A2/ROS1 cells, and MEKK1 wild-type and MEKK1delH907_C916 del were overexpressed in 3T3-CD74/ROS1 cells. In addition, HBEC-ER1 cells established by CRISPR/Cas9 genome editing, which maintain physiologic protein expression, were used to examine the effect of expression of MEK1delE41_L54 and MEKK1delH907_C916 on sensitivity to ROS1-TKI. Successful introduction of MEK1 delE41_L54 or MEKK1delH907_C916 del was confirmed by Sanger sequencing (Supplementary Fig. S3A). Western blot analysis revealed that expression of MEK1delE41_L54 or MEKK1delH907_C916 induced remarkable phosphorylation of ERK, suggesting that these mutations are activating (Fig. 4A; Supplementary Fig. S3B). In addition, the expression of BIM, a proapoptotic protein, was more reduced in cells expressing both ROS1 fusion protein and MEK1delE41_L54, compared with cells expressing ROS1 fusion alone (Supplementary Fig. S3C). We next examined the effect that these mutant MAPKs have on the sensitivity of ROS1 fusion–positive cells to ROS1 TKIs that had been used for each patient: the patient with MEK1delE41_L54 was treated with crizotinib, entrectinib, and lorlatinib, and the patient with MEKK1delH907_C916 was treated with crizotinib and cabozantinib. Cell viability assay revealed that ROS1 fusion–positive cells expressing MEK1delE41_L54 or MEKK1delH907_C916 were less sensitive to ROS1 TKIs, compared with control cells expressing an empty vector (Fig. 4B and C; Supplementary Fig. S3D–S3G). Expression of wild-type MEK1 or MEKK1 did not affect sensitivity to ROS1 inhibitors. To better understand how MEK1delE41_L54- and MEKK1delH907_C916–mediated TKI resistance, we examined their effects on ROS1 fusion and other downstream signaling proteins. Whereas phosphorylation of ROS1 was inhibited by crizotinib in a dose-dependent fashion in cells expressing either wild-type or the corresponding in-frame deletion, phosphorylation of ERK remained insensitive to crizotinib treatment only in MEK1delE41_L54- or MEKK1delH907_C916–expressing cells (Fig. 4D). Similar results were obtained with the other ROS1 TKIs: entrectinib and lorlatinib for MEK1delE41_L54 and cabozantinib for MEKK1delH907_C916 (Supplementary Fig. S3H–S3J). As for the patient with MEK1delE41_L54, PIK3CA E545K mutation was also detected in the post-TKI sample. To assess whether this mutation affects the drug sensitivity, we overexpressed PIK3CA wild-type and E545K by retroviral transduction using 3T3-SLC34A2/ROS1 and HBEC-ER1 cells. As shown in Supplementary Fig. S3K–S3M, the expression of PIK3CA E545K mutation induced the phosphorylation of AKT, but did not alter sensitivity of growth of ROS1 fusion–positive cells to crizotinib and cabozantinib.

As indicated in Supplementary Table S1, 2 patients had a ROS1 fusion and a concurrent NF1 truncating mutation. One of these 2 cases (case #2) also showed an intragenic deletion of NF1 exon 17. However, whether this intragenic deletion occurred in cis or in trans with the nonsense mutation cannot be determined. We could not identify a second hit in the remaining two cases, although one of these had low tumor content. Studies of NF1 truncating mutations in lung adenocarcinoma have generally not documented biallelic events in most cases, even when the presence of the NF1 truncating mutation appeared biologically significant (22), possibly due to technical limitations in detecting single copy loss or small intragenic deletions. Likewise, NF1-mutant lung cancer cell lines show complete loss of NF1 protein expression even when no second hit is obvious (23). Finally, even in the absence of a second hit, it is also possible that NF1 haploinsufficiency may been significant, as it has been shown that reduced NF1 expression can contribute to EGFR TKI resistance (24). On the basis of these data and considerations, and given that NF1 is a well-characterized negative regulator of the RAS/MEK/ERK pathway and the pathway is suggested to be highly activated in ROS1 fusion–positive lung adenocarcinoma, we sought to examine the effect of NF1 loss on ROS1 fusion–positive cells. To model NF1 loss-of-function mutation in lung adenocarcinoma with a ROS1 fusion, we performed knockdown of NF1 using siRNAs in HBEC-ER1 cells. Knockdown of NF1 was confirmed at mRNA and protein level by qRT-PCR and Western blot analysis (Fig. 4E; Supplementary Fig. S3N). As expected, ERK phosphorylation was increased in NF1-knockdown cells. We next examined growth of cells in the presence of crizotinib and cabozantinib. As shown in Fig. 4F and Supplementary Fig. S3O, knockdown of NF1 conferred resistance to crizotinib and cabozantinib in HBEC-ER1 cells. Collectively, our results indicate the possibility that the activation of MAPK pathway by multiple mutations is responsible for the resistance to ROS1 TKIs.

Because little is known about the MEK1delE41_L54 mutation, we assessed its transforming ability in Ba/F3 cells. Ba/F3 cells require IL3 for growth and survival; however, if they are transformed by introduction of an oncogene, they become IL3-independent. We expressed MEK1delE41_L54 or the wild-type MEK1 and cultured the cells in growth media without IL3. Whereas Ba/F3 cells expressing MEK1 wild-type were not able to survive without IL3, Ba/F3 cells expressing MEK1delE41_L54 can grow without IL3, suggesting that Ba/F3 cells were transformed by the MEK1delE41_L54 transduction (Fig. 5A). To investigate the functional role of MEK1delE41_L54, we generated NIH-3T3 and HBECp53 cell line models and examined the effects of expression on growth and phosphorylation status of key signaling molecules. As shown in Supplementary Fig. S4A, induction of MEK1delE41_L54 promoted cell growth in NIH-3T3 cell line models. In Western blot analysis, ERK was highly phosphorylated in 3T3 and HBEC cells expressing MEK1delE41_L54. However, the degree of MEK1 phosphorylation was lower than that of ERK (Fig. 5B). Of note, phosphorylation of BRAF, which is upstream of MEK1, was suppressed in cells expressing MEK1delE41_L54. We then performed knockdown of BRAF by siRNA and examined the phosphorylation of MEK1 and ERK. In HBECp53 cells expressing MEK1 wild-type, knockdown of BRAF suppressed the phosphorylation of MEK1 and ERK (Fig. 5C). In contrast, despite knockdown of BRAF, the phosphorylation of MEK1 and ERK were sustained at the same level as the negative control in HBECp53 cells expressing MEK1delE41_L54. To better elucidate this phenomenon, we transiently transfected MEK1 wild-type or MEK1delE41_L54 into HEK-293T cells, and the lysates were subjected to coimmunoprecipitation Western blot analysis. As shown in Fig. 5D, whereas the binding of MEK1 to BRAF was observed in MEK1 wild-type–expressing cells, a physical association between MEK1 and BRAF was not observed in cells expressing MEK1delE41_L54. These results indicate that the E41_L54 in-frame deletion in MEK1 resulted in a constitutively active kinase. We then examined the changes in drug sensitivity to MEK and RAF inhibitors caused by expression of MEK1delE41_L54. Introduction of MEK1delE41_L54 increased the sensitivity to MEK inhibitors in Ba/F3 and HBECp53 cell line models (Fig. 5E; Supplementary Fig. S4B and S4C). In contrast, LY3009120, a pan-RAF inhibitor, did not affect growth of MEK1delE41_L54–expressing cells, reinforcing that MEK1delE41_L54 can activate downstream pathways independently of RAF signaling.

To further characterize MEK1delE41_L54, we subsequently performed comparisons with MEK1 harboring several well-known activating mutations (F53L, Q56P, and K57N). MEK1 wild-type, MEK1delE41_L54, MEK1-F53L, MEK1-Q56P, and MEK1-K57N were transiently transfected into HEK-293T cells, and the phosphorylation status of ERK was compared. We found that ERK is highly phosphorylated in MEKdelE41_L54–transfected cells in comparison with MEK1-F53L-, MEK1-Q56P-, and MEK1-K57N–transfected cells, suggesting the more potent activating ability of MEK1delE41_L54 (Fig. 5F). The E41-L54 deletion overlaps the region encoding the nuclear export signal (NES, residues 32–51) of MEK1 (Supplementary Fig. S4D). Therefore, we examined the influence of this deletion on the subcellular localization of MEK1. Wild-type MEK1 or MEK1 mutants (E41_L54 del, F53L, Q56P, and K57N) tagged with GFP were transiently transfected into HEK-293T cells, and GFP-labeled cells were observed by fluorescence microscope. The results revealed that in MEK1 wild-type or other mutations except for E41_L54 del–transfected cells, MEK1 is largely excluded from the nucleus, whereas in cells expressing MEK1delE41_L54, MEK1 remained in the nucleus without being exported (Fig. 5G). For further confirmation, we compared expression of MEK1 in the nuclear fraction by Western blot analysis. Consequently, a remarkable enrichment of MEK1 was observed in MEK1delE41_L54–expressing cells (Supplementary Fig. S4E). In contrast to MEK1 localization, ERK remained largely cytoplasmic in MEKdelE41_L54–expressing cells, suggesting the acceleration of ERK dynamics (Fig. 5H). Finally, we searched for MEK1 deletion that overlaps the NES region across all cancers in MSK-IMPACT. As a result, we identified two patients (melanoma and pancreatic cancer) with a similar deletion (E41_F53), suggesting that in-frame deletion of the NES region is a recurrent mutation.

To overcome the resistance to ROS1 TKIs caused by the activation of the MAPK pathway, we first examined the efficacy of MEK inhibitors. As shown in Fig. 6A and Supplementary Fig. S5A and S5B, the induction of MEK1delE41_L54 or MEKK1delH907_C916 increased sensitivity to MEK inhibitors in comparison with cells expressing the wild-type proteins. In contrast, cells expressing MEK1delE41_L54 showed decreased sensitivity to LY3009120. Consistent with this result, Western blot analysis revealed that phosphorylation of ERK was sustained even at the high concentration of 1 μmol/L LY3009120 in MEK1delE41_L54–expressing cells (Fig. 6B). To better elucidate how MEK inhibitors affect growth, we examined their effect on phosphorylation of EGFR, HER2, ROS1, and other downstream signaling proteins. Time-course experiments showed that treatment with selumetinib completely suppressed phosphorylation of ERK by 48 hours, but reactivation of ERK was observed in the cells treated with trametinib (Fig. 6C). Although the feedback activation of RTKs and AKT was unclear in Fig. 6C, it is well known that MEK inhibition causes the activation of several upstream receptor tyrosine kinases by relieving physiologic feedback suppression (25–27). Therefore, we tested the hypothesis that coinhibition of MEK and ROS1 could be an efficient therapeutic strategy to overcome resistance to ROS1 TKIs through MAPK activation. We examined the combined effect of ROS1 TKIs and MEK inhibitors in cell viability assays, and synergism was evaluated using the Chou–Talalay Method (28). In HBEC-ER1 and 3T3-SLC34A2/ROS1 isogenic models, combination of selumetinib and crizotinib showed better synergistic effect than trametinib and crizotinib (Fig. 6D; Supplementary Fig. S5C). We also confirmed by Western blot analysis that phosphorylation of both ERK and AKT was effectively suppressed by the combination of crizotinib and selumetinib (Fig. 6E). In addition, expression of cleaved PARP was prominently induced in cells expressing ROS1 fusion and MEK1E41_L54del when treated with a combination of crizotinib and selumetinib (Supplementary Fig. S5D). To validate these in vitro findings, we performed in vivo experiments testing the efficacy of the combined inhibition of MEK and ROS1. We first evaluated the in vivo tumorigenicity of MEK1 wild-type and MEK1delE41_L54. Whereas allografts of 3T3-SLC34A2/ROS1-MEK1E41_L54del grew significantly faster than 3T3-SLC34A2/ROS1-empty or MEK1 wild-type allografts, the expression of MEK1 wild-type had no effect on tumorigenesis (Supplementary Fig. S5E). Next, mice bearing 3T3-SLC34A2/ROS1-Vector control or 3T3-SLC34A2/ROS1-MEK1delE41_L54 allografts were treated with either crizotinib alone (25 mg/kg once daily), lorlatinib alone (3 mg/kg), selumetinib alone (40 mg/kg once daily), a combination of crizotinib and selumetinib, or a combination of lorlatinib and selumetinib. The changes in volume of individual tumors upon treatment are shown in Fig. 6F and Supplementary Fig. S5F. Crizotinib or lorlatinib treatment suppressed growth of 3T3-SLC34A2/ROS1-EV allograft tumors, but was less effective at modulating growth of 3T3-SLC34A2/ROS1-MEK1delE41_L54 allograft tumors, indicating that MEK1delE41_L54 can confer resistance to ROS1 TKIs in vivo. Consistent with our in vitro data, the combination of crizotinib or lorlatinib with selumetinib successfully inhibited phosphorylation of ERK in 3T3-SLC34A2/ROS1-MEK1delE41_L54 allograft tumors (Supplementary Fig. S5G). In addition, in agreement with our observations that the MEK–ERK pathway is preferentially activated by ROS1 fusions, allograft tumors with ROS1 fusion alone were more sensitive to a combination of selumetinib and lorlatinib, than to a monotherapy with lorlatinib (Supplementary Fig. S5H). The drug combination was well tolerated during the treatment course with no significant changes in animal weight (Supplementary Fig. S5I).

In this study, we observed that 11% of patients with a ROS1 fusion had a concurrent MAPK pathway alteration and that this correlated with poor survival. Two patients acquired novel activating mutations in the MAPK pathway (MEK1delE41_L54 and MEKK1delH907_C916) following a ROS1 TKI, and we demonstrated that aberrant activation of MEK–ERK pathway caused by these mutations can confer resistance to ROS1 TKIs. We also found that loss-of-function mutations in the RASGAP NF1, which enhances RAS signaling, thereby activating downstream MEK/ERK pathway, were present in 2 patients prior to ROS1 TKI. Deletion of NF1 in cells with ROS1 fusion reduced sensitivity to ROS1 TKIs, suggesting that loss-of-function NF1 mutations may also contribute to resistance to ROS1 TKIs. Several studies support the notion that loss of NF1 plays an important role in resistance to kinase inhibitors in cancers with genetic alterations causing upstream activation of MAPK pathway (24, 29, 30). Because we find that expression of a ROS1 fusion preferentially activates MAPK signaling over the AKT pathway, loss of NF1 is a candidate mechanism to be examined as part of the landscape of off-target alterations that can mediate resistance to ROS1-targeted TKI therapy. To overcome MAPK pathway–mediated resistance, we tested the efficacy of combined inhibition of ROS1 and MEK with several inhibitors and found that the combination therapy successfully suppressed the growth of ROS1 fusion–positive cells in vitro and in vivo. As for the optimal types of MEK inhibitors for combination therapy, our results suggest the possibility that selumetinib may be a better partner to be used with ROS1 TKIs because the more commonly used MEK1 inhibitor trametinib activates ERK signaling, likely helping to counteract inhibition of ROS1. Further investigation of the optimal types of MEK and ROS1 inhibitors for combination therapy should be explored.

When considering a therapeutic strategy to overcome drug resistance, on-target resistance mechanism and off-target resistance mechanism, such as activation of bypass signaling should be distinguished. For the former, a more potent and structurally modified inhibitor targeting the oncogene itself can be used, whereas for the latter, combination therapy with targeted agents is likely to be more effective (11). In ROS1-rearranged lung adenocarcinoma, multiple on-target mechanisms for the resistance to crizotinib have been reported (9, 10, 31, 32). However, there are few reports of drug resistance caused by off-target mechanisms in ROS1-rearranged lung adenocarcinoma. Dziadziuszko and colleagues have reported that an activating KIT receptor mutation caused crizotinib resistance in a ROS1-rearranged lung adenocarcinoma patient and this was overcome by combined inhibition of ROS1 and KIT (33). All other studies describing off-target mechanisms of drug resistance to ROS1 TKIs were conducted in cells in culture. Song and colleagues reported that upregulation of the EGFR pathway conferred resistance to crizotinib in the HCC78 cell line (harbors an SLC34A2/ROS1 fusion; ref. 32) and several studies have suggested the possibility that activation of RAS signaling pathway can render HCC78 cells resistant to ROS1-TKIs (7, 34). Here we reported 2 cases of acquired resistance to ROS1 TKI caused by recurrent activation of MAPK pathway in ROS1-rearranged LUAD and demonstrate that other co-occurring MAPK pathway alterations such as loss-of-function NF1 mutations can be consequential. Moreover, several studies have demonstrated that the activation of MAPK pathway was one of the most important off-target mechanisms in the setting of resistance to EGFR TKIs (35–37). Considering these findings, the activation of MAPK pathway should be noted as one of the crucial off-target mechanisms in ROS1-rearranged lung adenocarcinoma. Considering these findings, activation of MAPK pathway should be noted as one of the crucial off-target mechanisms in ROS1-rearranged lung adenocarcinoma. In contrast to the evidence for MAPK pathway alteration as driving resistance to therapy, a few studies have reported on PI3K–AKT and JAK–STAT3 pathway gene mutations as mechanisms of TKI resistance in ALK-rearranged and ROS1-rearranged lung adenocarcinoma (11, 38–40). This may imply a less dominant role of these pathways in ROS1- or ALK-mediated proliferation. Indeed, our results show that the PIK3CA E545K mutation has no effect on sensitivity to ROS1 TKIs. Taken together, our results indicate the importance of MEK/ERK signaling in primary or acquired resistance in ROS1-rearranged lung adenocarcinoma.

Several studies have explored the specific signaling pathways that are activated by ROS1 fusion protein; Jun and colleagues reported that CD74/ROS1, but not FIG/ROS1, induced phosphorylation of E-Syt1, resulting in more invasive properties (41). Neel and colleagues showed that ROS1 fusion proteins exhibit differential activation of MAPK signaling according to the subcellular localization of fusion proteins (42). These studies suggested the possibility that downstream signaling of ROS1 fusion may differ depending on the fusion partners. In this study, we demonstrated that ROS1 fusion proteins can interact with the GRB2–SOS1 complex by coimmunoprecipitation studies. This phenomenon was observed with three fusions (CD74, SLC34A2, and EZR), suggesting that binding of ROS1 fusion protein to GRB2 is a common phenomenon that does not depend on the 5′ fusion partner. The tyrosine phosphatase, PTPN111 (SHP2) is associated with ROS1 via phosphorylated Y2274, and it has been shown to recruit GRB2 and SOS1, and activate the MAPK pathway (43, 44). Although we did not investigate SHP2 in our study, it is possible that the ROS1–GRB2–SOS1 complex includes SHP2. More than 10 ROS1 fusion partner genes have been reported in lung adenocarcinoma and the detailed mechanisms of ROS1–GRB2 interaction are still unclear. Further investigation is important to clarify these potentially important biochemical interactions.

Our results suggest that MEK1_E41_L54 in-frame deletion is a novel oncogenic driver mutation. Gao and colleges previously performed a broad survey of MEK1 mutations and defined three subsets of MEK1 mutations based on RAF dependency (RAF-dependent, RAF-regulated, and RAF-independent groups; ref. 45). Among them, the RAF-independent group was shown to possess RAF-independent kinase activity and to be insensitive to allosteric MEK inhibitors. However, in cells expressing MEK1delE41_L54, RAF-independent ERK activation and increased sensitivity to allosteric MEK inhibitors were simultaneously observed. One rationale for these contrasting results may be explained by changes in ERK and MEK1 subcellular localization. In general, it is known that MEK1/2 shuttles between the cytoplasm and nucleus, and relocalization of nuclear MEK1/2 to the cytoplasm is regulated by the NES. MEK1/2 in the cytoplasm act as a cytoplasmic anchor of ERK and in the nucleus, it regulates nuclear export of dephosphorylated ERK (46, 47). We demonstrated that deletion of a segment overlapping the NES region leads to retention of MEK1 in the nucleus and promotes export of dephosphorylated ERK to the cytoplasm. Rapid relocalization of deactivated ERK to the cytoplasm may accelerate the activation of the MAPK cascade, thereby leading to higher phosphorylation level of ERK in comparison with other MEK1 mutations. Although there are some unknowns, such as how MEK redistribute to the cytoplasm, critical changes in the intracellular dynamics of MEK and ERK are pivotal characteristics of mutations occurring in NES region, and such specific MEK1 mutations may need to be considered as a new subset.

In conclusion, our results indicate that the activation of MAPK pathway via the interaction with GRB2–SOS1 complex plays a key role in the tumorigenic properties of ROS1-rearranged LUAD. Activation of the MAPK pathway by multiple genetic alterations can lead to high proliferative ability in the presence of a ROS1 TKI manifested clinically as drug resistance. Combined inhibition of ROS1 and MEK can be a promising therapeutic strategy that should be explored clinically in patients that have a ROS1 fusion and a MAPK pathway alteration.

M. Offin reports receiving speakers bureau honoraria from PharmaMar, Novartis, and Targeted Oncology. A. Drilon reports receiving speakers bureau honoraria from Ignyta/Roche/Genentech, Loxo/Bayer/Lilly, Takeda/Ariad/Millenium, TP Therapeutics, AstraZeneca, Pfizer, Blueprint, Helsinn, Beigene, BergenBio, Hengrui, Exelixis, Tyra, Verastem, MOREHealth, and Abbvie, and reports receiving other remuneration from GlaxoSmithKline, Teva, Taiho, PharmaMar, WoltersKluwer, Puma, Merus, and Boehringer Ingelheim. N. Rosen is a paid consultant for Tarveda, AstraZeneca, Ribon, Chugai, Beigene, MapCURE, Zai Labs, and Boehringer Ingelheim; reports receiving commercial research grants from Chugai; and holds ownership interest (including patents) in Beigene, Zai Labs, Kura, and Fortress. M.G. Kris is a paid consultant for AstraZeneca, Regeneron, Pfizer, and Daiichi-Sankyo. D. B. Solit is a paid advisory board member for Loxo Oncology, Pfizer, Lilly Oncology, QED Therapeutics, Illumina, and Vivideon Therapeutics. G. J. Riely reports receiving commercial research grants from Pfizer, Novartis, Merck, Mirati, Roche, and Takeda, and is an unpaid consultant/advisory board member for Daiichi and Takeda. R. Somwar reports receiving commercial research grants from Helsinn Health Care, Loxo Oncology, Merus, and 14Ner Oncology. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Sato, E. Siau, K. Suzawa, M. Offin, A. Drilon, M.G. Kris, R. Somwar

Development of methodology: H. Sato, E. Siau, D. Kubota, D.B. Solit, R. Somwar

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Sato, A.J. Schoenfeld, E. Siau, Y.C. Lu, D. Kubota, A.J.W. Lui, B. Qeriqi, M. Mattar, M.G. Kris, D.B. Solit, E. de Stanchina, G.J. Riely, M. Ladanyi, R. Somwar

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Sato, A.J. Schoenfeld, E. Siau, D. Kubota, M. Offin, M. Sakaguchi, A. Drilon, D.B. Solit, M.A. Davare, M. Ladanyi, R. Somwar

Writing, review, and/or revision of the manuscript: H. Sato, A.J. Schoenfeld, E. Siau, M. Offin, S. Toyooka, A. Drilon, N. Rosen, M.G. Kris, D.B. Solit, M.A. Davare, G.J. Riely, M. Ladanyi, R. Somwar

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Tai, D. Kubota, M. Offin, A. Drilon, M.G. Kris, D.B. Solit, G.J. Riely

Study supervision: M. Offin, M. Sakaguchi, M. Ladanyi, R. Somwar

This work was supported by NIH grants U54 OD020355, P30 CA008748, and P01 CA 129243.

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.