Mechanisms of Resistance to NTRK Inhibitors and Therapeutic Strategies in NTRK1-Rearranged Cancers

The neurotrophic receptor tyrosine kinase 1 (NTRK1) gene was first identified as a fusion oncogene, trkA (tropomyosin receptor kinase), in a colorectal cancer in 1986 (1). The NTRK1 gene belongs to a family of nerve growth factor receptor genes consisting of NTRK1, NTRK2, and NTRK3, encodes a single transmembrane receptor tyrosine kinase (RTK) protein (also called as Trk A, B, and C). NTRK1, 2, and 3 are receptors of nerve growth factor (NGF), brain-derived growth factor (BDNF), and neurotrophin 3 (NTF-3), respectively. They are mainly expressed in human neuronal tissues and play critical roles in the development and homeostasis of the nervous system. The high-affinity ligand of NTRK1, NGF (a member of the neurotrophins), leads to dimerization and activation of NTRK1 and its downstream signaling, which plays an important role in the survival, development, and function of neurons (2). In contrast, the NTRK1 fusion oncogenes, such as tropomyosin 3 (TPM3) and NTRK1, both located on chromosome 1, induce constitutive activation of NTRK1 tyrosine kinase mediated by constitutive expression by the promoter of the fusion partner gene and oligomerization using domains, such as the coiled-coil domain, in the fusion partner protein. Various fusion partner genes of NTRK1 (e.g., TPR, PPL, and LMNA) have been reported, and almost all the reported fusion partner genes possess oligomerization domains. NTRK1 fusion oncogenes have been identified in many types of cancers, including thyroid cancer (12%), glioblastoma (2.5%), lung cancer (1%), and others (3, 4). However, there are no clinically approved drugs for the treatment of NTRK-rearranged cancers. More than seven drug candidates have been developed, and some of them have been evaluated in clinical trials (5). Among them, entrectinib and LOXO-101 have been reported to induce marked tumor shrinkage in patients with NTRK-rearranged cancers (6–9). However, the emergence of tumors resistant to NTRK-TKIs is inevitable, as has been seen in EGFR mutant, ALK-, or ROS1-fusion positive lung cancer (10–13). A case of entrectinib resistance mediated by secondary mutations in NTRK1 (G595R and G667C mutations), and a case of entrectinib resistance by the G623R mutation in NTRK3 (corresponding to G595R in NTRK1) in mammary analogue secretory carcinoma (MASC) were recently reported (14, 15).

In this study, we screened and identified resistance mutations by N-ethyl-N-nitrosourea (ENU) random mutagenesis using TPM3-NTRK1-overexpresseing Ba/F3 cells and established the NTRK-TKI-resistant cells using colon cancer KM12 cells harboring TPM3-NTRK1. We also conducted a drug screening with FDA-approved inhibitors and other well-known TKIs against both KM-12 cells and wild-type or our identified resistant mutant TPM3-NTRK1 Ba/F3 cells. We identified several potential inhibitors, including cabozantinib (XL-184; FDA-approved for medullary thyroid carcinoma), nintedanib (BIBF-1120; FDA-approved for idiopathic pulmonary fibrosis), and ponatinib (AP24534; FDA-approved for T315I-positive chronic myeloid leukemia and T315I-positive Philadelphia chromosome positive acute lymphoblastic leukemia). From the establishment of cabozantinib- resistant KM12 cells and analysis using the resistant cells, we found that insulin growth factor receptor type 1 (IGF1R) mediated resistance to cabozantinib and other NTRK-TKIs and that combination therapy targeting IGF1R and NTRK1 completely reversed the resistance. We also found that a G595R mutation harbored by NTRK1-TKI-resistant cells emerged in monoclonal KM12 cells, and that a TKI-resistant minor population, called the “persister clone,” existed in both parental and single clone-derived KM12 cells, which could be eradicated by the combination of IGF1R with NTRK1 inhibitors. Our study suggests potential resistance mechanisms and therapeutic strategies to overcome the resistance.

Ba/F3, immortalized murine bone marrow-derived pro-B cells, were obtained from the RIKEN BRC Cell Bank (RIKEN BioResource Center) in 2012 and were cultured in DMEM (low glucose) containing 10% FBS with or without mIL-3. KM12 cells originally provided from NCI-60 in 1991 were cultured in DMEM (low glucose) containing 10% FBS. All cells were routinely tested and verified to be free of mycoplasma contamination. KM12 cell line was authenticated by short tandem repeat profiling analysis (BEX) in 2016, and all the cell lines were used within 6 month periods after thaw the original authenticated cell line stocks.

Inhibitors were purchased from each companies as shown below. Stock solutions of the inhibitors were prepared by dissolving them in 10 mmol/L DMSO or 10 mmol/L ethanol. 17-AAG (16), GDC0941, Gefitinib, Imatinib, Lapatinib, and Erlotinib were purchased from LC Laboratories. LOXO-101, AEW541, Alectinib, OSI906, Trametinib, cabozantinib, ceritinib, CH5183584, and PF-06463922 were from ActiveBiochem. Crizotinib, brigatinib (AP26113), vandetanib, and BGJ398 were purchased from Biochempartner. Nintedanib, Ponatinib, Sorafenib, Ponatinib, Sunitinib, Vemurafenib, E7080, and AZD9291 (Osimertinib) were from Selleck, and ASP3026, AZD6244 (17), BIBW2992 (Afatinib), and TAE684 (18) were from ChemieTek. PHA665752 and SB218078 were from Tocris Bioscience, Entrectinib was from MedChem Express, MS209 was from Sigma, AZD3463 (19) was from BioVision, CEP701 was from Calbiochem, Foretinib and PP242 were from AdooQ Bioscience, Rapamycin was from AG Scientific, and Everolimus was purchased from Chem Scene.

cDNA of TPM3-NTRK1 extracted from KM12 cells was cloned into a pLenti6.3 vector and then was induced into Ba/F3 cells (IL3 dependent), as previously reported (20). In brief, Ba/F3 cells were infected with lentivirus produced in 293FT cells using a lentiviral infection system (ViraPower packaging plasmid: Invitrogen) and pLenti6.3 viral plasmid. After viral infection, Ba/F3 cells were selected with blasticidin (7 μg/mL) for about 1 week. IL3 was then removed from the media, and the TPM3-NTRK1-dependently growing Ba/F3 cells were established.

A single clone (clone 28) was selected from the TPM3-NTRK1 Ba/F3 cells and was exposed to ENU (a DNA alkylating compound, 100 μg/mL, purchased from Sigma) for 24 hours at 1.0 × 10⁶ cells/mL. The mutated cells were then washed with PBS, and resistant cells were selected under various concentrations of cabozantinib (300, 1,000, 2,000 nmol/L), nintedanib (700, 1,000, 2,000 nmol/L), and foretinib (100, 300, 1,000 nmol/L) in 96-well plates (5.0 × 10⁴ cells/well) and cultured up to 4 weeks. The resistant cells were then expanded and assayed. Everything that came in contact with ENU was decontaminated with 1 M NaOH.

Single clones from KM12 colorectal cancer cells were obtained by limiting dilution. Parental or single clone-derived KM12 cells were treated with 1 μmol/L of cabozantinib or 600 nmol/L of Entrectinib, and multiple cabozantinib-resistant KM12 CR cell lines were established.

From 2,000 to 3,000 cells per well were seeded into a 96-well plate, and inhibitors were added at various concentrations. After 72 hours of incubation, CellTiter-Glo (Promega) was added, and the cells were incubated for 10 minutes at room temperature. The luminescence of the cells was then measured with a Tristar LB 941 microplate luminometer (Berthold Technologies). The data were graphically displayed with a GraphPad Prism version 5.0 (GraphPad Software). A nonlinear regression model determined the IC₅₀ with a sigmoidal dose response in GraphPad.

From 3.0 × 10⁵ to 1.0 × 10⁶ cells were treated with various concentrations of inhibitors and were collected after 3 hours. Equal amounts of lysates were electrophoresed and transferred to a membrane. Indicated proteins were detected using antibodies against phospho-NTRK1 (Tyr 674/675, C50F3), NTRK1 (C17F1), phospho-AKT (Ser 473, D9E), AKT (C67E7), phospho-ERK (Thr 202/204), ERK, phospho-IGF1R (Tyr 1135/1136, 19H7), IGF1R (111A9), phospho-S6 (Ser 240/244, D68F8), S6 (54D2) (Cell Signaling Technology), and β-actin (Sigma). Phospho-RTK array (R&D) was performed as indicated in the manufacturer's instruction using the cell lysates from the KM12 cell line and its resistant cells treated with or without the indicated inhibitors.

KM12 cells were cultured in soft agar (0.3%) layered on 0.6% soft agar, with or without the inhibitors (cabozantinib or Entrectinib, 0 and 300 nmol/L; LOXO-101, 1000 nmol/L; OSI906, 300 nmol/L; cabozantinib + OSI906, 300 nmol/L) and incubated for 7 to 14 days. After the incubation, the live cells were stained with 0.1% crystal violet. After washing out the excessive crystal violet, images of the live cells were obtained by the EPSON ES2200 scanner.

All mouse studies were conducted through Institutional Animal Care and Use Committee–approved animal protocols according to the institutional guidelines. KM12 cells (3 × 10⁶) were suspended in 50 μL of HBSS and subcutaneously implanted into Balb-c nu/nu mice (Charles River). Tumor growth was monitored twice weekly by bilateral caliper measurement, and tumor volume was calculated as 0.5 × length × width × width (mm³). When the average tumor volume reached approximately 200 mm³, the mice were randomized into vehicle and treatment groups using the restricted randomization such that the mean tumor size of each group was equivalent. The mice were treated once daily by oral gavage for the indicated period. Relative tumor volume was calculated by dividing by the tumor volume on day 0. The body weights of the mice were measured twice weekly. Implanted tumors were resected from the mice on day 3 of drug treatment and were examined by immunoblot. The sample size (minimum n = 5 per treatment group) was selected to ensure satisfactory inter-animal reproducibility. The Mann–Whitney U test was used for the statistical analysis of the mice experiments.

For the treatment of NRTK-rearranged cancers, multiple NTRK-TKIs have been developed and are being evaluated in clinical trials. Among the inhibitors, Entrectinib, and LOXO-101 have been reported to be effective in patients with NTRK fusion-positive cancers (6). KM12 colorectal cancer cells have been reported to harbor TPM3-NTRK1 fusion gene and addicted to the TPM3-NTRK1-mediated growth signaling (21). To study the mechanisms of resistance to NTRK-TKI, we established the TPM3-NTRK1-dependently growing Ba/F3 cell line model. IL3 dependently growing Ba/F3 cells were infected with the lentivirus expressing TPM3-NTRK1. The infected Ba/F3 cells were able to grow in the absence of IL3, as previously reported (21). Cabozantinib, entrectinib, and LOXO-101 were confirmed to inhibit TPM3-NTRK1 Ba/F3 cell growth. Through a focused kinase inhibitor screening using TPM3-NTRK1 Ba/F3 cells, we also found that nintedanib (22, 23) and foretinib (24) were potent NTRK1 inhibitors (Fig. 1A). Consistent with other studies, crizotinib also inhibited NTRK1 (Fig. 1A). These inhibitors, but not OSI-906 (25), an IGF1R inhibitor, were also effective at inhibiting KM12 cell growth (Fig. 1B). The calculated IC₅₀ of Ba/F3-TPM3-NTRK1 or KM12 cells to NTRK inhibitors were not largely different from the previous studies (Fig. 1A and B, right). Foretinib and entrectinib completely inhibited NTRK1 phosphorylation at 10 nmol/L in both cell lines. Cabozantinib and LOXO-101 inhibited NTRK1 phosphorylation at 100 nmol/L, whereas nintedanib only partially inhibited NTRK1 phosphorylation at 100 nmol/L (Fig. 1C).

Molecular targeting drugs such as TKIs often induce marked tumor shrinkage; however, the emergence of drug-resistant cancer is inevitable in most of the cases. To identify potential NTRK1 mutations that will occur in NTRK1-TKI resistance, we conducted ENU mutagenesis screening. Single clones were selected from TPM3-NTRK1 Ba/F3 cells by limiting dilution and expanded. All the clones were confirmed to express the NTRK1 fusion protein (Supplementary Fig. S1A). All of the picked-up mono clones showed similar sensitivity to NTRK inhibitors (Supplementary Fig. S1B–S1D), and clone 28, which had sensitivity similar to that of TPM3-NTRK1 Ba/F3 polyclonal cells, was selected to conduct ENU mutagenesis (Fig. 1A and Supplementary Fig. S1E). From the results of ENU mutagenesis screening selected by cabozantinib, nintedanib, or foretinib, we identified five resistant mutations in the NTRK1 kinase domain: L564H, G595L, G595R, F646I, and D679G (Fig. 2A). Except for D679G, which is located in the activation loop, all the mutations were located in the ATP-binding pocket (Fig. 2B). The G595R mutant, which was selected by cabozantinib, nintedanib, or foretinib, was the most common mutation. The L564H, G595L, F646I, and D679G mutations were only obtained through cabozantinib selection in the screening. There were also a total of 221 resistant cells without any mutations in the NTRK1 kinase domain (Fig. 2A). All the resistant cells with secondary mutations obtained from ENU were also resistant to foretinib and nintedanib as compared with the wild-type clone 28 cells (Fig. 2C and Supplementary Fig. S2). The NTRK1 phosphorylation level of all the mutants was maintained even when they were treated with 100 nmol/L cabozantinib (Fig. 2D). Notably, the G595L mutant did not decrease its NTRK1 phosphorylation level upon treatment with 1000 nmol/L cabozantinib.

Recently, two acquired resistance mutations to entrectinib in the catalytic domain of NTRK1, p.G595R, and p.G667C were found by the analysis of circulating tumor DNA (ctDNA), KM12 xenograft model, and xenopatient, which was established by directly transplanting tumor biopsy tissue of a patient with LMNA–NTRK1 fusion positive colorectal cancer to the mouse (14). Thus, we subsequently established BaF3 cells that had the identified NTRK1 mutations and the G667C mutation (Fig. 3A). We then examined the sensitivity of G667C or G595R mutant TPM3-NTRK1-harboring Ba/F3 cells to NTRK inhibitors in clinical trials (Entrectinib, LOXO-101, and Cabozantinib). The results showed that the G595R mutant was highly resistant to entrectinib, LOXO-101, and cabozantinib. In contrast, G667C mutants were highly resistant to entrectinib and LOXO-101, but very sensitive to cabozantinib, with approximately 10-fold lower IC₅₀ than that of TPM3-NTRK1 (wild-type) Ba/F3 cells (Fig. 3B–D). To identify the inhibitors that could overcome these resistant mutant NTRK1s, we performed a focused drug screening by treating the mutant, wild-type TPM3-NTRK1-expressing, or parental Ba/F3 with the 36 inhibitors consisting of various kinase inhibitors targeting tyrosine or serine/threonine kinases that are now available clinically or are being evaluated in clinical trials, referring to the report by Duong-Ly and colleagues that showed the potential to repurpose inhibitors against disease-associated or drug-resistant mutant kinases (26). Foretinib, ponatinib (27, 28), and nintedanib, in addition to cabozantinib, were identified as potential inhibitors of G667C mutants. However, no inhibitors except for the inhibitors targeting downstream signaling, such as MEK inhibitor trametinib (29–31), could inhibit the growth of G595R mutants (Fig. 3E). Consistent with the results of the focused drug screening, G667C mutants showed lower IC₅₀ to ponatinib, nintedanib, and foretinib than did the wild type (Fig. 3F–H).

We then examined the sensitivity of other identified resistant mutants to these inhibitors and obtained IC₅₀ values. The L564H mutant was highly sensitive to entrectinib but resistant to ponatinib, cabozantinib, and foretinib. The F646I mutant was sensitive to ponatinib but resistant to cabozantinib and foretinib. The D679G mutant was sensitive to entrectinib and LOXO-101 but resistant to ponatinib, cabozantinib, and foretinib. However, both G595R and G595L (solvent front) mutants were highly resistant to all the tested NTRK1 inhibitors (Fig. 4A, Supplementary Fig. S3A and S3B, and Supplementary Table S1). Consistent with the cell viability assay of Ba/F3 mutants, the inhibition of NTRK1 phosphorylation corresponded to the IC₅₀ of each mutant's growth to each NTRK1 inhibitor (Fig. 4B). When the cells were treated with higher concentrations of the focused inhibitor library, we found that the Hsp90 inhibitor 17-AAG caused marked growth inhibition in both wild-type and resistant mutants, and the growth inhibition by 17-AAG was also confirmed by treating KM12 parental cells with 17-AAG. When we examined the effect of 17-AAG by immunoblotting, we found that 17-AAG treatment not only decreased NTRK1 phosphorylation and phosphorylation of the downstream signaling molecules, but also drastically decreased total TPM3-NTRK1 protein, an effect different from the effects of other NTRK-TKIs, such as cabozantinib, crizotinib, and ponatinib (Supplementary Fig. S4A–SAC). To check whether ponatinib can inhibit NTRK1 and induce tumor growth inhibition, we performed in vivo experiments by administering ponatinib, cabozantinib, or entrectinib to nude mice into which KM12 cancer cells had been subcutaneously injected. As a result, the mice treated with entrectinib or cabozantinib showed significant inhibition of the growth of KM12 cells, and ponatinib also showed statistically significant suppression of tumor growth compared with vehicle controls without explicit toxicity (Supplementary Fig. S5A). Phosphorylation of NTRK1 and its downstream phosho-AKT were actually inhibited by entrectinib, cabozantinib, or ponatinib in tumor samples obtained from mice (Supplementary Fig. S5B). The observed effect and potential usefulness of the Hsp90 inhibitor, and the induction of the degradation of NTRK fusion proteins, could be similar phenomenon observed in ALK-rearranged lung cancer cells (20, 32–36).

To identify resistance mechanisms besides resistant mutations in NTRK1, such as bypass pathway activation, we exposed KM12 cells to 1 μmol/L of cabozantinib. A small number of KM12 cells remained after 1 week of cabozantinib treatment. Then the resistant cells that could survive in the presence of cabozantinib were selected and cloned in the presence of cabozantinib, and cabozantinib-resistant monoclonal cells were established. All the resultant-resistant clones had similar sensitivity to cabozantinib, with more than a 10-fold increase in IC₅₀ values (Supplementary Fig. S6A and Fig. 5A and B). None of the more than 20 isolated clones that we obtained had any mutations in the NTRK1 kinase domain. We mainly used cabozantinib-resistant clone 20 (CR20) for the following experiments. Unlike the mutants obtained from ENU mutagenesis screening, NTRK1 phosphorylation of CR20 cells was inhibited at the same concentration of cabozantinib, entrectinib, or LOXO-101 as was that of the parental KM12 cells (Figs. 1C and 5C). However, downstream AKT phosphorylation was maintained in CR20 and CR4 cells, even when they were treated with 1 μmol/L of cabozantinib (Fig. 5C and Supplementary Fig. S6B). CR20 and other CR clones also showed cross-resistance to the other NTRK1 inhibitors nintedanib, foretinib, entrectinib, and LOXO-101 (Fig. 5D and Supplementary Fig. S6C). We hypothesized that there was an alternative bypass pathway responsible for the resistance, and therefore we performed a phospho-RTK array to identify other active RTKs. We detected higher phosphorylation levels of IGF1R and ephrin receptor B3 (EphB3) in cabozantinib-treated CR20 cells (Fig. 5E). The same signals indicating phosphorylated RTKs were also detected in cabozantinib-treated CR4 cells (Supplementary Fig. S6D). To determine whether either of these two RTKs was inducing resistance, we treated the cells with either 1 μmol/L of dasatinib (a BCR-ABL inhibitor and also active against EphB3) or AEW541 (an IGF1R inhibitor) in combination with cabozantinib. We also performed focused drug screening in combination with cabozantinib, and IGF1R inhibitor AEW541, MEK inhibitor AZD6244, or ALK and IGF1R inhibitor ceritinib effectively resensitized the CR20 cells to cabozantinib (Supplementary Table S2). The combination of AEW541 and cabozantinib, nintedanib, foretinib, entrectinib, or LOXO-101 treatment was able to reverse the resistance in CR20 cells (Fig. 5F and Supplementary Fig. S7A–S7D). Similar results were observed when we treated CR4 cells with cabozantinib and AEW541 in combination (Supplementary Fig. S7E). We then confirmed this effect using OSI906, a distinct IGF1R-specific inhibitor, and compared the sensitivity of KM12 parental cells and CR20 cells to NTRK-TKIs and OSI906 cotreatment (Fig. 5G, Supplementary Fig. S8A–S8C, and Supplementary Table S2). With the combination of cabozantinib and 300 nmol/L OSI906, although the IC₅₀ of KM12 parental cells was slightly decreased from 25.7 to 12.2 nmol/L, CR20 cells improved IC₅₀ from 1,910 to 37.3 nmol/L, overcoming the resistance. Compared with KM12 parental cells, CR20 cells had higher phospho-IGF1R and total IGF1R levels, and although cabozantinib alone did not inhibit IGF1R or the downstream signaling pathways, the combination of cabozantinib with an IGF1R inhibitor inhibited IGF1R phosphorylation and downstream phosphorylation (Fig. 5H). Of note, AZD3463, reported as a dual inhibitor of ALK and IGF1R, also inhibited NTRK1. Indeed, AZD3463 monotherapy inhibited both NTRK1 and IGF1R phosphorylation and its downstream signaling (Fig. 5H), and AZD3463 showed similar IC₅₀ in KM12 parental cells and cabozantinib-resistant CR20 cells (Supplementary Fig. S8D).

Because the cabozantinib-resistant clones are thought to come from the surviving persister clones after 1 week of high-dose (1 μmol/L) cabozantinib treatment (Supplementary Fig. S5A), we examined whether the persister clones existed in KM12 parental cells and whether cotreatment with IGF1R inhibitor could eradicate these persister cells. When treated with OSI906 or vehicle (DMSO) only, parental KM12 cells formed many colonies after 7 days according to soft agar assay. Survival of the persistently resistant clones was observed when the cells were treated with cabozantinib, entrectinib, or LOXO-101 for 14 days, but the emergence of these cells was dramatically inhibited when they were treated with a combination of cabozantinib, entrectinib, or LOXO-101 with OSI906 (Fig. 5I and Supplementary Figure S9). To understand how the persistently resistant clones emerged, we subsequently performed exactly the same procedures using monoclonal KM12 cells, clone 5 (WT-5) obtained by limiting dilution (Supplementary Fig. S10A). The WT-5 clone had the same sensitivity as the parental cells and other sensitive clones (Supplementary Fig. S10B). Similar to the existence of parental KM12 cells, the existence of persister cells was confirmed in the colony formation assay by treating them with cabozantinib, and the persister cells were also eradicated by treating with cabozantinib and OSI-906 in combination (Fig. 5I).

To further examine the mechanisms of resistance to NTRK1 inhibitors, we chose a different sensitive clone, WT clone 6 (WT-6), which had the same sensitivity as the parental cells and other sensitive clones. WT-6 cells were also sensitive to cabozantinib and foretinib (Supplementary Fig. S10B and S10C). We then exposed the WT-6 cells to high concentrations of cabozantinib or entrectinib until the fully resistant cells proliferated. A small population of WT-6 cells survived after 2 months of exposure to 1 μmol/L of cabozantinib or 600 nmol/L of entrectinib, and the resultant 6-CR cells or 6-EntreR1 were able to grow normally in the presence of cabozantinib or entrectinib (Fig. 6A and Supplementary Fig. S11A). The cabozantinib-resistant 6-CR cells showed considerable resistance to all NTRK1 inhibitors, and the resistance could not be overcome by a combination of cabozantinib and OSI906 (Fig. 6B). Similarly, the entrectinib-resistant 6-EntreR1 cells were highly resistant to entrectinib (Supplementary Fig. S11B). To understand the mechanisms of resistance, cDNA was extracted from the 6-CR cells and 6-EntreR1 cells, and sequencing revealed that the both cells had a G595R (1783 G→A) mutation that did not exist initially in the sensitive monoclonal WT-6 cells (Fig. 6C and Supplementary Fig. S11C). This mutation was the same solvent front mutation that was detected in our ENU mutagenesis screening and reported from the entrectinib-resistant patient. Western blot analysis revealed the similar sensitivity to various inhibitors in the Ba/F3 cell line (Fig. 6D and Supplementary Fig. S11D). In this study, we have not identified a potential inhibitor candidate to overcome G595R resistance. Thus, further study is needed to overcome G595 solvent front mutation-mediated resistance.

The NTRK gene was first identified as a rearranged fusion oncogene, and various NTRK fusion oncogenes have been identified in many cancers, such as colorectal cancer, non–small cell lung cancer, thyroid cancer, and glioblastoma (5). The NTRK gene family comprises three genes: NTRK1, NTRK2, and NTRK3. All three genes have been observed in cancer and are thought to function as fusion oncogenes in similar fashion. To target these three fusion oncogenes, multiple NTRK inhibitors (cabozantinib, entrectinib, LOXO-101, DS-6051b, and others) have been evaluated in clinical trials. However, no NTRK inhibitor is approved for the treatment of NTRKs rearranged cancer patient at this moment. A few reports have found a dramatic effect in patients with cancer treated with NTRK-TKI (8, 14, 37). Currently, NTRK-rearranged cancer patients can be found by the NGS-based methods or the break apart FISH (37). The frequency and numbers of patients with NTRK-rearranged cancers are unclear. From the results of a next-generation sequencing (NGS)-based screening study, NTRK-rearranged cancer is predicted to exist in less than 1% of cases of lung and colorectal cancer. Thus, establishment of screening methods will be very important to deliver the potent NTRK inhibitors to the right patients. For patients with NTRK-rearranged cancers, even though marked tumor shrinkage can be achieved by treatment with NTRK-TKIs, the tumor will inevitably relapse due to acquired resistance. In one patient with entrectinib-resistant colorectal cancer, two acquired resistance mutations in the ntrk1 kinase domain, p.G595R and p.G667C, were found from analysis of the patient's ctDNA (14). This study suggested that tumor heterogeneity also exists and would be a critical issue for the development of effective therapy to overcome resistance.

In this study, we found multiple new resistance mutations to NTRK inhibitors by ENU mutagenesis screening and by the establishment of resistant cells from KM12, NTRK1-rearranged colorectal cancer cells. We also found an IGF1R activation-mediated resistance, which seemed to play an important role in innate cabozantinib-persistent clones of KM12 cells. By establishing monoclonal cabozantinib-sensitive KM12 clones by limiting dilution, we confirmed that IGF1R activation-mediated drug-tolerant clones were detectable even in the monoclone derived from KM12 cells, suggesting that IGF1R-mediated persistence emerges with a certain probability during expansion of the single-clone-derived cells. We also observed that cells harboring G595R-resistant mutations emerged from the single-clone-derived KM12 cells during 2 months of culture with high-dose cabozantinib treatment, which suggests that the bona fide resistant mutation could also be acquired. Hata and colleagues, using EGFR exon19 deletion positive lung cancer cells, recently showed that drug-resistant cancer cells are both preexisting and evolve from drug-tolerant cells (38). As we showed in this study, drug-tolerant cells existed (and emerged) at a low frequency by unknown mechanisms, such as altered epigenetic state and/or feedback reactivation of other RTKs such as IGF1R. Sharma and colleagues, using EGFR-mutated PC9 cells, showed that individual cancer cells transiently assume a reversible drug-tolerant state to protect the population from eradication by potentially lethal exposures. They also showed that IGF1R signaling is required for the emergence of reversibly drug-tolerant cells (39).

Among our identified or previously reported TKI-resistant mutations in NTRK1, the p.L564H mutation, located in the alpha-C helix within the N-lobe of the NTRK1 tyrosine kinase domain, would probably distort the alpha-C helix by the substitution of an amino acid from leucine to cationic histidine residue. This L564H showed higher resistance to cabozantinib and nintedanib but was sensitive to entrectinib. The F646 residue is equivalent to F1245 of ALK, which is required for sustaining the interaction between the alpha-C helix within the N-lobe and the C-lobe (π–π interaction between F1174 in the N-lobe of ALK and F1245 in the C-lobe, and hydrophobic interaction between L563, L567, and F646 in NTRK1), and the p.F646I mutation probably weakens the hydrophobic interaction between these residues. The F646I mutant was resistant to cabozantinib and nintedanib but sensitive to entrectinib and ponatinib. Interestingly, ponatinib showed higher sensitivity to F646I than wild-type NTRK1. D679G is the only resistant mutation that is located far from the ATP- and drug-binding pocket but close to the activation loop in the C-lobe of NTRK1. The p.D679G mutation is predicted to change the conformation of the ATP-binding pocket by affecting the structure of the activation loop. The D679G mutation conferred resistance to cabozantinib and foretinib, but not to entrectinib and LOXO-101, suggesting that the binding mode of inhibitors might differ between cabozantinib (or foretinib) and entrectinib (or LOXO-101). Foretinib and cabozantinib have similar structures, and it was reported that both are predicted to preferentially bind the inactive conformation of ROS1 with an apo-simulated docking pose different from that of other ROS1-TKIs, such as crizotinib (40). The G667 residue is the next of the conserved DFG motif, a core motif regulating kinase activity, and equivalent to G1269 in ALK and G2101 in ROS1. The p.G1269A mutation in ALK was found in a patient with crizotinib-resistant, ALK-rearranged lung cancer, and the p.G667C mutation was also found in the ctDNA of an entrectinib-resistant patient. Interestingly, however, we found that the G667C mutation confers higher sensitivity to cabozantinib and foretinib as well as to ponatinib and nintedanib; the sensitivity is even higher than that of wild-type NTRK1. The reported human plasma concentration of cabozantinib, ponatinib, and nintedanib are 3 to 5 μmol/L, 200, and 70 nmol/L, and the IC₅₀ of each drugs against G667C mutants were 12, 7, and 24 nmol/L, respectively. However, cabozantinib and ponatinib have extensive plasma protein binding (over 99% bound to plasma proteins) and the in vitro protein binding of nintedanib in human plasma is known to be more than 97%. In addition, when we consider the induction of tumor shrinkage, it is important to consider IC70s or IC90s rather than IC50s. From our in vivo experiment, cabozantinib and ponatinib were shown to be able to inhibit the growth of TPM3-NTRK (WT) expressing KM12 cells. However, it is important to carefully evaluate whether or not the exposures of these compounds in vivo experiments are achievable and tolerable dose in human. Therefore, the further studies are needed to investigate whether or not clinical dose of cabozantinib, ponatinib, or nintedanib could be able to potently inhibit NTRK1 especially G667C mutated NTRK1.

This glycine residue followed by DFG in NTRK1, ALK, or ROS1 is a cysteine in cKIT, PDGFRA, and FLT3; therefore, the change of glycine to cysteine might not impair the activity of NTRK1, but it might change the core conformation of the ATP-binding region of NTRK1 similar to that of the other related tyrosine kinases. Therefore, distinct TKIs for targeting other tyrosine kinases, such as ponatinib and nintedanib, might work better against resistant mutants than against the wild type. For example, ponatinib is known to potently inhibit ABL, KIT, FGFR, FLT3, and PDGFRa in vitro, and in our study, ponatinib had a 10-fold lower IC₅₀ to the G667C mutant than to wild-type NTRK1 (41). Of course, further studies will be needed to examine whether ponatinib, nintedanib, or cabozantinib can effectively inhibit G667C and/or wild-type NTRK1 at achievable concentrations in humans. These drugs will also have to be evaluated in vivo and, eventually, in clinical trials.

Compared with the above mutations, the mutations in the G595 residue, G595R and G595L, which are called solvent front mutations, showed significant resistance to all the tested NTRK inhibitors and TKIs in our focused screening, except for the inhibitors targeting downstream pathways, such as trametinib, an MEK inhibitor. The G595R mutation is equivalent to G1202R in ALK and G2032R in ROS1. The G1202R mutation of ALK has been shown to be relatively sensitive to lorlatinib (PF-06463922) (42), and the G2032R mutation is sensitive to cabozantinib and foretinib (43). However, NTRK1-G595R mutants were highly resistant to lorlatinib, cabozantinib, and foretinib. An Hsp90 inhibitor such as 17-AAG had similar potency to all the mutant and wild-type NTRK1 fusion proteins, which was accompanied by the induction of enhanced NTRK1 fusion protein degradation by the Hsp90 inhibitor, suggesting that the NTRK1 fusion protein is required for proper folding by Hsp90, which was also observed in ALK fusion. Therefore, an Hsp90 inhibitor might represent an alternative candidate to overcome secondary mutation-mediated NTRK-TKI resistance.

In this study, we identified five resistance mutations and an IGF1R-mediated resistance, and also provided potential therapeutic strategies to overcome the resistance of NTRK1-rearranged cancers. Considering that the resistance mutations identified in ALK and ROS1 using ENU mutagenesis screening have been observed in crizotinib-treated patients, there is a high possibility that the results in post-treated patients with NTRK1-rearranged cancers will agree with our results. However, the frequency of each resistance mechanism and the applicability of these findings to other NTRK2 fusion proteins has yet to be clarified. Further studies are needed to find additional resistance mechanisms, the clinical applicability of the findings, and potent inhibitors that can overcome the resistance mutations, especially G595R/L solvent front mutations.

No potential conflicts of interest were disclosed.

Conception and design: R. Katayama

Development of methodology: R. Katayama

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.J. Fuse, K. Okada, T. Oh-hara, H. Ogura, R. Katayama

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.J. Fuse, K. Okada, T. Oh-hara, R. Katayama

Writing, review, and/or revision of the manuscript: M.J. Fuse, R. Katayama

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.J. Fuse, K. Okada, N. Fujita

Study supervision: N. Fujita, R. Katayama

We thank Ms. Sumie Koike and Mr. Shigeo Sato at Japanese Foundation for Cancer Research (JFCR) for help with the in vitro and in vivo experiments.

This study was supported in part by MEXT/JSPS KAKENHI grant numbers 15H02368 and 17H06327 (to N. Fujita), 16H04715 (to R. Katayama), the grant from the MHLW/AMED grant numbers 17cm0106203h0002 and 17ck0106231h0002 (to R. Katayama), and the grant from the Vehicle Racing Commemorative Foundation (to R. Katayama).

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.