Tivantinib (ARQ 197) Exhibits Antitumor Activity by Directly Interacting with Tubulin and Overcomes ABC Transporter–Mediated Drug Resistance Free

The receptor tyrosine kinase, MET proto-oncogene, receptor tyrosine kinase (c-MET) is a high-affinity receptor for hepatocyte growth factor (HGF), and its downstream v-akt murine thymoma viral oncogene homolog 1 (AKT) and mitogen-activated protein kinase 1 (ERK) pathways are regulated by HGF/c-MET. The HGF/c-MET axis is involved in cancer progression, metastasis, and acquired resistance. HGF/c-MET signaling is often highly activated in tumors because of various mechanisms (1). Because c-MET–addicted cancers have been shown to be highly sensitive to c-MET kinase inhibitors in vitro and in vivo, c-MET is recognized as a therapeutic target, and some c-MET inhibitors are currently being evaluated in clinical trials (2).

Tivantinib (ARQ197) was first reported to be a highly selective inhibitor of c-MET (3). Crystal structure analysis elucidated a unique mechanism in which tivantinib preferentially binds to the inactive form of c-MET. In addition, unlike other c-MET inhibitors, tivantinib inhibits c-MET through a non–ATP-competitive mechanism (4). From the results of a phase I clinical trial, tivantinib showed encouraging antitumor activity and tolerability (5). In early clinical trials, tivantinib increased overall survival (OS) and progression-free survival (PFS) in patients with hepatocellular cancer showing high c-MET expression. On the basis of these data, the phase III trial currently ongoing enrolls only MET-high patients (6). On the other hand, in a recent clinical trial of tivantinib combined with an epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor (TKI), there were no significant differences in PFS and OS between the study arm (tivantinib with EGFR-TKI) and control arm (EGFR-TKI only; ref. 7). Surprisingly, subgroup analysis showed that tivantinib with EGFR-TKI treatment significantly improved PFS among patients with non–small cell lung cancer (NSCLC) with Kirsten rat sarcoma viral oncogene homolog (KRAS) mutation, but the latest phase III data presented at European Cancer Organisation-European Society for Medical Oncology (ECCO-ESMO) 2013, no benefit was observed in KRAS-mutant patients (7, 8). In addition, an original study by Munshi and colleagues (3) showed that tivantinib inhibited the cell growth of KRAS-mutated A549 cells that had been reported to be insensitive to another c-MET inhibitor, PHA-665752 (9, 10). These results suggest that tivantinib shows cytotoxic activity in addition to c-MET inhibition. Furthermore, we and Basilico and colleagues independently reported that tivantinib showed cytotoxic activity in various cancer cells by disrupting microtubule function independent of cellular c-MET status (11–13). Related to the report (13), there is published letter explaining that the activity of tivantinib in clinical trials is related to the MET inhibition; however, there is the counterargument (14, 15). In addition, recently Remsing Rix and colleagues (16) reported that tivantinib bound and inhibited glycogen synthase kinase 3 (GSK3) alpha and beta to a greater degree than did c-MET. Molecular target of tivantinib is still uncertain.

Because microtubules have a crucial role in cancer cell division and motility, microtubules have been recognized as highly attractive targets for cancer chemotherapy. Tubulin binding agents, such as vincristine and paclitaxel, have been commonly used as chemotherapeutic drugs in various cancers. However, as with all chemotherapeutic agents, the appearance of drug-resistant cells is a major obstacle. It has been reported that microtubule inhibitor–treated cells acquired resistance through various mechanisms, including overexpression of βIII-tubulin, alteration of the β-tubulin gene, and overexpression of ABC transporters, such as P-glycoprotein (17–20). Thus, to overcome chemoresistance to tubulin binding agents, a new class of microtubule-targeting agents is currently being evaluated in clinical trials, and several drugs have recently been approved and used clinically. For example, microtubule-stabilizing agent ixabepilone, which is an analogue of epothilone B, was approved by the FDA in 2007 for treatment of patients with metastatic or advanced breast cancer whose tumors are resistant or refractory to anthracyclines, taxanes, and capecitabine. Ixabepilone showed low susceptibility against taxane-resistant cells caused by overexpression of P-glycoprotein in vitro and in vivo (21). Eribulin, a synthetic analogue of halichondrin B, has been approved by the FDA for treatment of patients with metastatic breast cancer who have previously received at least two chemotherapeutic regimens to treat metastatic disease. It has been reported that the underlying mechanism of eribulin activity is unique and that the drug suppressed the growth rates of microtubules, but unlike vinblastine, eribulin did not affect shortening rates of microtubules (22).

In the present study, we aimed to elucidate the mechanism underlying the cytotoxic activity of tivantinib in which cellular microtubules are disrupted. To determine if tivantinib shows cytotoxic activity by inhibition of tubulin polymerization, we quantified assembled microtubules in cells and a xenograft model. To determine if tivantinib directly binds to tubulin, we performed a tubulin binding competition assay using tivantinib and other ³H-labeled microtubule inhibitors. We found that tivantinib directly bound to tubulin and competitively inhibited binding of colchicine to tubulin. We also used computational modeling to predict a structural model of the tubulin–tivantinib complex in the colchicine binding pocket of α- and β-tubulin complexes. On the other hand, tivantinib showed cytotoxic activity against ATP-binding cassette sub-family B member 1 (MDR1)-overexpressing cells that acquired resistance to vincristine, vinblastine, and colchicine. Tivantinib equally inhibited growth and induced apoptosis in ATP-binding cassette sub-family C member 1 (MRP1)- or ATP-binding cassette sub-family G member 2 (BCRP)-overexpressed cells. These results suggest that tivantinib, differently from other tubulin inhibitors, might be effective against multidrug-resistant cells that overexpress ABC transporters.

EBC-1, SK-MEL-28, DLD-1, K562, K562/VCR, CEM, and CEM/VBL were cultured in RPMI-1640 medium with 10% FBS. 293T, KB3-1, KB3-1 MDR1, KB3-1 MRP1, and KB3-1 BCRP were cultured in RPMI medium with 10% FBS. All cells were maintained at 37°C in a humidified atmosphere at 5% CO₂. EBC-1 and CEM cells were obtained from the Japanese Cancer Research Resources Bank, SK-MEL-28 was obtained from the ATCC, and 293FT was purchased from Invitrogen. K562/VCR and CEM/VBL were established in our institute (23, 24). KB3-1 cells were provided by Dr. I. Pastan (National Cancer Institute), and MDR1-, MRP1-, or BCRP-expressing subclones were provided by Dr. Y. Sugimoto (Keio University, Japan). K562 and DLD-1 cells were provided by Drs. E. Ezaki and T. Tsuruo in our institute, respectively. SK-MEL-28, DLD-1, K562, K562/VCR, 293T, KB3-1, KB3-1 MDR1, KB3-1 MRP1, and KB3-1 BCRP cells were authenticated by short tandem repeat analysis. CEM and CEM/VBL cells have not been tested by the authors. Tivantinib (ARQ 197) and crizotinib were purchased from ChemieTek. Vincristine, vinblastine, colchicine, and paclitaxel were purchased from Sigma. Streptavidin-coated yttrium scintillation proximity assay (SPA) beads, [³H]colchicine, and [³H]vincristine were purchased from PerkinElmer. [³H]vinblastine was purchased from American Radiolabeled Chemicals. The CSII-EF-MCS vectors encoding mCherry-hCdt1 and mVenus-hGem were kindly provided by Dr. Atsushi Miyawaki (RIKEN).

Lentiviral plasmid for fluorescent ubiquitination-based cell cycle indicator (Fucci) expression [mCherry-hCdt1 (30/120) or mVenus-hGem (1/110)] was transfected by using a packaging plasmid into 293FT cells (25). Both viral solutions were prepared and infected into EBC-1 or SK-MEL 28 cells. Fucci-expressed clones were obtained by limiting dilution. For time-lapse fluorescence microscopy, Fucci-expressing cells were plated onto a 35-mm glass-bottom dish. After 24 hours, the medium was replaced with drug-containing medium, and dishes were placed in 37°C humidified chamber of an FV10i confocal laser microscope (Olympus).

Cell lysates were prepared as described previously (11, 26). Equal amounts of lysates were electrophoresed and immunoblotted with the antibodies against phospho-c-MET (Tyr1234/1235) (3D7), c-MET (25H2), phospho-p42/44 ERK/MAPK (Thr202/Tyr204), p42/44 ERK/MAPK, phospho-AKT (Ser473) (D9E), panAKT (C67E7), β-actin (13E5), cleaved PARP (Asp214; Cell Signaling Technology), MDR1 (C219; MILLIPORE), MRP1 (MONOSAN), BCRP (C219; Santa Cruz Biotechnology), and α-tubulin (DM1A; Sigma). An ECL Prime Western Blotting Detection Reagent (GE Healthcare) or SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Scientific) was used to detect immunoreactive signal.

Separation of polymerized tubulin from tubulin dimmers and analysis of the effect of tivantinib on tubulin polymerization in cells were performed as described previously with some modification (27). In brief, EBC-1 or DLD-1 cells were treated with the indicated concentrations of agents. After treatment, the cells were washed with PBS, and subsequently lysis buffer containing 20 mmol/L Tris-HCl, pH 8.6, 1 mmol/L MgCl₂, 2 mmol/L EGTA, 1 mmol/L PMSF, 20 μg/mL aprotinin, and 0.5% Nonidet P-40 (tubulin detection buffer) was added to the treated agents. Supernatants were collected after centrifugation at 15,000 rpm for 15 minutes at 4°C. The pellets were dissolved in a Laemmli Sample Buffer. Equal amounts of supernatants and dissolved pellets were electrophoresed and immunoblotted with the indicated antibodies.

Six-week-old female athymic nude mice (Charles River Laboratory) were inoculated s.c. at 1 × 10⁶ EBC cells per mouse. Five days after inoculation, tivantinib was administered at 200 mg/kg twice per day for 5 consecutive days, followed by a 2-day dosing holiday. Mouse body weight and tumor volume (0.5 × length × width²) were measured twice per week. Tivantinib was formulated in polyethylene glycol 400/20% Vitamin E tocopheryl polyethylene glycol succinate (60:40) at 30 mg/mL as previously described (3). The nonparametric Mann–Whitney U test was used to perform statistical analysis. Nineteen days after drug treatment, the tumors were resected from the nude mice and homogenized with tubulin detection buffer (above). Cell lysates were filtered through a 0.80-μm filter, centrifuged, and subsequently immunoblotted according to the same protocol described above. All animal procedures were performed according to protocols approved by the Japanese Foundation for Cancer Research Animal Care and Use Committee.

The tubulin binding assay was performed as reported previously with slight modification (28). In brief, [³H]colchicine was diluted with ethanol and concentrated by evaporation on a SpeedVac centrifuge for 40 minutes. After centrifugation, [³H]colchicine with or without unlabeled compounds was diluted in binding buffer [1 mmol/L GTP, 1 mmol/L EGTA, 1 mmol/L MgCl₂, and 80 mmol/L PIPES (pH 6.8)], and the buffer was subsequently transferred into a 96-well plate (90 μL per well). After transfer, 0.5 μg of biotin-labeled tubulin prepared in 10 μL of binding buffer was added. The total reaction buffer volume was 100 μL/well, and the final concentration of [³H]colchicine was indicated in figure legends. The plates were incubated for 2 hours at 37°C with gentle shaking. After incubation, 0.08 mg of SPA beads in 20 μL of binding buffer was added. After further incubation for 10 minutes under agitation at 37°C, the SPA beads were allowed to settle for 1 hour. A TopCount microplate scintillation counter (PACKARD) was used to perform scintillation counting.

We predicted the binding poses of tivantinib for the colchicine binding site of tubulin using docking software GOLD version 5.1. The docking score was calculated by GOLD version 5.1. Higher genetic optimization for ligand docking (GOLD) scores correspond to better pose ranks. The X-ray crystal structure of tubulin in a complex with colchicine (PDB code: 1SA0) was used as the initial structure. The statistics program package R was used to perform clustering of 50 tivantinib structures.

Cell survival assay was performed as described previously (11). In brief, 3,000 cells were seeded into 96-well plates in triplet or sextuplet. On the following day, the cells were incubated with the indicated drugs for 72 hours. Cell viability was determined by adding the CellTiter-Glo assay reagent (Promega) for 10 minutes, and using a TriStar LB 941 Multimode Microplate Reader (Berthold Technologies). GraphPad Prism version 5.0 (GraphPad Software) was used to graphically display the data and to determine IC₅₀ values.

A total of 1 × 10⁵ cells were seeded into 6-well plates. On the following day, the cells were treated with the indicated concentrations of drugs and incubated for another 72 hours. After incubation, the cells were collected and stained with FITC-labeled Annexin V and 1 μg/mL propidium iodide (PI) for 10 minutes. A Cytomics 500 flow cytometer (Beckman Coulter) was used to assay the cells, and FlowJo software (Tree Star) was used to analyze the data.

Previously, we reported that tivantinib disrupts microtubule assembly in vitro and in vivo (11). However, detailed mechanism underlying inhibition of tubulin polymerization by tivantinib and whether tivantinib shows antitumor activity by microtubule depolymerization were not known.

Because drugs that disrupt microtubules induce aberrant mitotic phenotypes, it is expected that the influence of tivantinib on the cell cycle would be different from other c-MET inhibitors. To monitor the cell cycle in real time, we established Fucci-expressing EBC-1 or SK-MEL-28 cells and observed the cell-cycle progression after treatment with tivantinib or crizotinib. In the Fucci system, human Cdt1 deletion mutants were fused to red fluorescent protein mKO2 (monomeric Kusabira Orange 2), and the N-terminus of human geminin was fused to green fluorescent protein mAG1 (monomeric Azami Green 1). Because Cdt1, a substrate of SCF^Skp2 complex, expression is highest during G₁ and geminin, a substrate of APC^cdh1 complex, expression is highest during S–G₂–M phases, Fucci-expressing cells emit red light when the cells are in G₁ phase and emit green fluorescence when the cells are in S–G₂–M phase (25). EBC-1 cells are reported as c-MET–addicted cells because of c-MET amplification, and SK-MEL-28 cells are known to have BRAF mutation and barely express c-MET (Supplementary Fig. S1; refs. 11, 29). After exposure to tivantinib, EBC-1 cells induced G₂–M cell-cycle arrest, whereas crizotinib induced G₀–G₁ arrest (Fig. 1A; Supplementary Fig. S2A; Supplementary Movies 1–3). Consistent with the report that microtubule inhibitors induce metaphase arrest (30), tivantinib treatment reduced cyclin A (which is required for S phase entry) and markedly upregulated cyclin B (which functions in late M phase). Furthermore, crizotinib treatment decreased both cyclin A and B, suggesting the accumulation of G₀–G₁-phase cells (Fig. 1B). In addition, Fucci-expressing SK-MEL-28 cells also showed G₂–M cell-cycle arrest after 3 μmol/L tivantinib treatment, similar to the behavior of the cells treated with 30 nmol/L of vincristine (Supplementary Fig. S2B and S2C; Supplementary Movies 4–6). To quantify the G₀–G₁-phase cells and G₂–M-phase cells, we counted the ratio of the cells emit red or green fluorescence. Consistent with the previous result, tivantinib treatment increased G₂–M-phase cells.

To further clarify the effect of tivantinib on microtubule dynamics, we examined the microtubule assembly state in the cells treated with tivantinib or other drugs. First, c-MET–amplified lung cancer EBC-1 cells were treated with 1 μmol/L tivantinib or 100 nmol/L of crizotinib, anaplastic lymphoma receptor tyrosine kinase (ALK) and c-MET inhibitor, for 16 hours and subsequently immunostained with anti–α-tubulin antibody. The tivantinib-treated cells showed cellular microtubule depolymerization with short microtubules in the cytoplasm, whereas the microtubules in the crizotinib-treated cells were not affected (Supplementary Fig. S3). Second, GFP α-tubulin–expressing KRAS-mutant lung cancer A549 cells or melanoma SK-MEL-28 cells were treated with tivantinib and observed by time-lapse confocal microscopy. Tivantinib treatment caused disruption of microtubules within 30 minutes (Supplementary Fig. S4).

These results suggest that tivantinib inhibits tubulin polymerization and induces G₂–M cell-cycle arrest independent of c-MET inhibition.

In our previous study, we performed an in vitro tubulin polymerization assay and observed that tivantinib inhibited tubulin polymerization in a dose-dependent manner similar to that by vincristine (11). To determine if tivantinib inhibited tubulin polymerization in cells, we quantified assembled microtubules in cells after tivantinib, paclitaxel, vincristine, or crizotinib treatment. As previously reported, 6-hour paclitaxel treatment significantly increased polymerized tubulin in EBC-1 cells. In contrast, vincristine treatment decreased assembled microtubules (31). As expected, tivantinib treatment inhibited tubulin polymerization in a concentration-dependent manner similar to that by vincristine treatment but surprisingly did not inhibit c-MET and its downstream AKT and ERK signaling pathways. On the other hand, crizotinib treatment did not affect tubulin polymerization but significantly suppressed phosphorylation of c-MET, AKT, and ERK (Fig. 1C). After 24-hour treatment, tivantinib still inhibited tubulin polymerization (Supplementary Fig. S5A). To confirm whether the reduction of polymerized tubulin was independent of c-MET activity, we used c-MET non-addicted H460 cells which expressed c-MET and had weak activity. In previous report, we showed H460 cells were unaffected by knockdown of c-MET (11). Equally to EBC-1 cells, tivantinib treatment inhibited tubulin polymerization dose-dependently in H460 cells (Supplementary Fig. S5B). It was recently reported that tivantinib induced microtubule stabilization in colon cancer DLD-1 cells when the cells were treated with a lower concentration of tivantinib (13). To see if tivantinib would similarly affect DLD-1 cells, we treated DLD-1 cells and quantified the amount of assembled microtubules. Similar to EBC-1 cells, the DLD-1 cells treated with tivantinib also showed a decrease in assembled microtubules in a dose-dependent manner (Supplementary Fig. S5C). These results suggest that tivantinib disrupts microtubule assembly and inhibits tubulin polymerization in cells without regard to c-MET activity.

From the results of phase I clinical trials, the mean C_max in plasma from patients treated with tivantinib was approximately 5 μmol/L (5, 32–34). After confirming that <5 μmol/L tivantinib induced reduction of polymerized tubulin in cells, we assessed the effect of tivantinib on xenografted human tumors in mice to see if tivantinib inhibited tubulin polymerization in vivo. Tivantinib showed antitumor effects at doses of 200 mg/kg twice per day for 5 consecutive days, followed by a 2-day dosing holiday (Fig. 2A). Following this dosing regimen, tivantinib did not show serious side effects, such as body weight loss (Supplementary Fig. S6A). After 19 days of drug treatment, the tumor xenografts were resected from nude mice. The resected tumors were minced, homogenized, and subsequently lysed with tubulin detection buffer according to the same method as used in the experiment with cultured cells. The tivantinib-treated tumors showed significant decreases in tubulin polymerization (insoluble α-tubulin) compared with the levels in the control tumors (Fig. 2B). Furthermore, tivantinib-treated tumor xenografts of the H460 cells also showed antitumor activity and reduction of polymerized tubulin (Fig. 2C and D and Supplementary Fig. S6B). In addition, tivantinib did not induce obvious change on c-MET and GSK-3 status (Supplementary Fig. S6C and S6D). These results suggest that tivantinib exhibits the antitumor effect and disrupting microtubule assembly in vivo as well as in vitro independent of c-MET status.

It is well known that microtubule-destabilizing agents, such as vincristine and colchicine, directly bind to tubulins at several different binding sites (35, 36). To determine if tivantinib directly binds to tubulin in a manner similar to that by other microtubule inhibitors, we estimated the effect of tivantinib or other small-molecule inhibitors on [³H]colchicine binding to tubulin using a competition binding SPA. Interestingly, tivantinib inhibited [³H]colchicine binding to tubulin in a dose-dependent manner similar to that by unlabeled colchicine (Fig. 3A). As expected, vincristine, vinblastine, and crizotinib did not affect [³H]colchicine binding to tubulin. To determine if tivantinib interferes with the tubulin binding of vincristine or vinblastine, which have been shown to bind to the vinca domain of tubulin, DEAE-cellulose filters were used to test the effect of tivantinib on [³H]vincristine or [³H]vinblastine binding to tubulin. As expected, unlabeled vincristine and vinblastine, but not tivantinib, colchicine, or crizotinib, inhibited both [³H]vincristine and [³H]vinblastine binding to tubulin (Supplementary Fig. S7A and S7B). These results suggest that tivantinib affects colchicine–tubulin binding. To determine if tivantinib competitively or noncompetitively inhibits colchicine binding to tubulin, we performed a binding assay with increasing concentrations of [³H]colchicine in the presence of tivantinib and calculated the B_max and Kd values for colchicine binding to tubulin using nonlinear regression analysis. Consistent with the increase in tivantinib concentration, the Kd values for colchicine but not for B_max binding to tubulin increased (Fig. 3B). These results suggest that tivantinib directly interacts with tubulin via the colchicine binding site, and competitively inhibits colchicine binding to tubulin.

On the basis of these in vitro results, we perform a computational simulation of the binding model in which tivantinib binds to the colchicine binding site of tubulin using GOLD docking program. In this study, we used the well-known tubulin–colchicine structure reported by Ravelli and colleagues (35). To validate the GOLD simulation findings, we pulled colchicine out of the binding site and consequently redocked colchicine into the structure. The top-ranked tubulin–colchicine structure simulated by the GOLD program showed a conformation quite similar to that of the original crystal structure. Next, we extracted colchicine from the reported structure and docked tivantinib (Fig. 3C). In this docking study, the top 50 structures calculated by the GOLD program were clustered into 10 groups (Fig. 3D and Supplementary Fig. S8). As shown in Fig. 3D, the top-scored tivantinib molecule was highly overlapped with the colchicine molecule in the binding pocket of tubulin. These results predict that tivantinib directly binds to the colchicine site of tubulin without steric hindrance.

The microtubule inhibitors, paclitaxel, vincristine, and vinblastine, are commonly used in the treatment of various cancers, but development of acquired resistance can prevent success. For example, overexpression of the drug efflux pump is one major mechanism of drug resistance. To determine if tivantinib could overcome drug resistance, we tested the cytotoxic activity of tivantinib against tubulin binding agent-resistant cells. In this assay, we used clones of K562 and CEM cells that showed acquired resistance to vinca alkaloid. These resistant clones of K562/VCR and CEM/VBL were obtained by growing the cells in the presence of sublethal concentrations of drug, and the cells were shown to acquire resistance by overexpressing MDR1 (23, 24; Fig. 4A). As expected, K562/VCR or CEM/VBL cells showed cross-resistance to vinblastine, vincristine, or colchicine, but surprisingly, these microtubule inhibitor–resistant cells were still sensitive to tivantinib (Fig. 4B).

Next, we examined the sensitivity of tivantinib in KB3-1 cells that overexpressed various ABC transporters (Fig. 5A; ref. 37). Ectopic overexpression of MDR1 resulted in acquired resistance to vincristine and slight resistance to adriamycin in KB3-1 cells. In addition, BCRP-overexpressed KB3-1 cells exhibited resistance to SN-38, and MRP1 overexpression resulted in resistance to adriamycin and slightly to vincristine (38). To our surprise, tivantinib showed the same cytotoxic activity in these ABC transporter–overexpressing cells (Fig. 5B). Next, we used PI and FITC-labeled Annexin V staining to measure induction of apoptosis. Consistent with the results from a cell viability assay, MDR1-overexpressing KB3-1 cells acquired resistance to vincristine, but all of the cells similarly induced apoptosis after tivantinib treatment (Fig. 6A and Supplementary Fig. S9A). Because of the fluorescence of SN-38 or adriamycin, we could not assess the apoptosis by FITC-Annexin V/PI staining. Therefore, we measured apoptotic cells by sub-G₁ analysis. As expected, we observed that KB3-1 cells that overexpressed BCRP did not undergo apoptosis by SN-38 treatment and KB3-1 cells that overexpressed MRP1 did not increase apoptosis by adriamycin (Supplementary Fig. S9B). After 24-hour treatment with tivantinib, cleaved PARP expression was equally induced in all of the cells, whereas vincristine did not induce PARP cleavage in the MDR1-overexpressed cells, SN-38 did not induce PARP cleavage in BCRP-overexpressed cells, and adriamycin did not induce PARP cleavage in MRP1-overexpressed cells (Fig. 6B).

Tivantinib was first reported as a highly selective inhibitor of c-MET and is currently being investigated in a phase III clinical trial based on the result of early clinical trials that increased OS and PFS in MET-high population (3, 6, 8). Because c-MET activation by c-MET gene amplification or HGF upregulation is related to acquired resistance to EGFR-TKI, a phase II study of EGFR-TKI erlotinib with or without tivantinib treatment for patients with advanced NSCLC was initiated. Although that study did not meet the primary endpoint, a small cohort of patients with KRAS mutations significantly improved PFS in the tivantinib combined with erlotinib-treated patients (7). This result is surprising because KRAS is downstream of c-MET, and KRAS-mutated cells have been reported to be insensitive to other c-MET inhibitors, such as PHA-665752 and crizotinib (9, 13). To explain these inconsistencies, we and Basilico and colleagues recently showed that tivantinib inhibited cell proliferation independent of addiction to c-MET and disrupted microtubule dynamics (11, 13, 14). Considering that tivantinib is currently in a phase III clinical trial and has shown encouraging antitumor activity, it is very important to clarify the true target of tivantinib. To investigate the mechanism underlying how tivantinib affects cell growth, we first used Fucci-expressing cells to compare the effect of tivantinib on cell cycling. The tivantinib-treated cells primarily emitted green fluorescence, which showed that tivantinib treatment induced G₂–M arrest and that the treatment with the potent c-MET/ALK inhibitor, crizotinib, induced G₁ arrest (Fig. 1A and B and Supplementary Fig. S2A–S2C). It has been reported that treatment with microtubule inhibitors, such as vincristine or paclitaxel, induced G₂–M arrest in cancer cells, and cells treated with tyrosine kinase inhibitors, such as erlotinib or gefitinib, often showed G₁ arrest in various EGFR-mutated NSCLC cell lines (39, 40). Consistent with our previous report, these results indicate that the mode of action of tivantinib is different from that of other tyrosine kinase inhibitors (11).

To clarify the effect of tivantinib on microtubule status in cells and in vivo, we quantified polymerized tubulin after tivantinib treatment. In EBC-1 cells, tivantinib decreased polymerized tubulin in a dose-dependent manner (Fig. 1C). Moreover, at this concentration of tivantinib, c-MET and downstream AKT/ERK were not inhibited. From the phase I dose-escalation study, the reported C_max of tivantinib was approximately 5 μmol/L (5, 32–34). Because we confirmed that <5 μmol/L tivantinib could inhibit tubulin polymerization in EBC-1 cells, we determined if tivantinib also could inhibit tubulin polymerization in vivo. Similar to the in vitro study results, tumor xenografts of tivantinib-treated mice showed reduction of polymerized tubulin compared with the level in the controls (Fig. 2B and D). This result suggests that the antitumor effect of tivantinib is associated with tubulin polymerization inhibition. Previous confocal analysis suggested that treatment with several hundred nanomolar concentrations of tivantinib induced microtubule stabilization in DLD-1 cell (13). To determine if tivantinib inhibited tubulin polymerization in DLD-1 cells, we measured cellular polymerized tubulin and confirmed that tivantinib reduced tubulin polymerization at an approximate 3 μmol/L concentration, which is lower than the C_max of tivantinib (Supplementary Fig. S5C). These results again suggested that tivantinib impairs microtubules by inhibiting tubulin polymerization.

Most tubulin-targeting drugs are known to directly bind to β-tubulin at three different sites, such as the vinca binding site, taxane binding site, or colchicine binding site (36). In this study, we showed that tivantinib dose-dependently inhibited [³H]colchicine binding to tubulin and did not affect [³H]vincristine or [³H]vinblastine binding to tubulin (Fig. 3A; Supplementary Fig. S7A and S7B). Moreover, the B_max for colchicine binding to tubulin was not changed, and the Kd value for colchicine binding to tubulin was increased in the presence of tivantinib (Fig. 3B). These results imply that tivantinib directly binds to tubulin at the colchicine binding site. It is known that classical tubulin inhibitors, such as paclitaxel and vincristine, produce a painful peripheral neuropathy; however, previous clinical trials have shown that there is no neurotoxicity at all in patients treated with tivantinib. This finding of minimal neuropathy is similar to that for recently approved tubulin-targeted inhibitors, such as eribulin and the semi-synthetic vinca alkaloid, vinflunine, for which the incidence of peripheral neuropathy was lower than those for paclitaxel or vinblastine. In addition, the Ki value of tivantinib was calculated to be approximately 9 μmol/L by using the Cheng–Prusoff equation, and the Ki value was lower than those of other colchicine-site binding agents previously reported (28), which suggests that tivantinib might have a slightly weak effect on tubulin polymerization. However, vinflunine is known to have less affinity for the vinca binding site of tubulin than does vincristine in mouse models and to show a potent antitumor effect with lower neurotoxicity in clinical trials (41, 42). The mechanism of microtubule inhibition by vinflunine has been reported to involve increase of microtubule growth duration and suppression of the treadmilling rate, but the effects were weaker than vinblastine. In addition, vinflunine has been reported to induce reduction of the microtubule growth rate (43). On the other hand, the molecular mechanism underlying how tivantinib inhibits microtubules remains unknown. Moreover, Remsing Rix and colleagues (16) recently reported that GSK3α and GSK3β are new target kinases of tivantinib. Further study is needed to elucidate the mechanism in detail.

Consistent with the results of the tubulin binding competition assay, computational prediction of tubulin–tivantinib binding also suggested that tivantinib directly binds to the colchicine site of tubulin without steric hindrance (Fig. 3C and D). However, because the resolution of the reported tubulin–DAMA colchicine structure was low, the predicted model of tivantinib–tubulin has a higher degree of uncertainty. Further experiments would be required to elucidate the binding mode of tivantinib.

Microtubule-targeting agents have been widely used in cancer chemotherapy; however, the emergence of resistant tumors is a major obstacle. As one of the resistant mechanisms to microtubule-targeting drugs, MDR1 overexpression was identified in various cancer cells (18). MDR1 is one of the ABC transporter proteins that confers resistance by pumping out various microtubule inhibitors, such as colchicine, taxol, and vincristine. In our previous report, the result of a COMPARE analysis involving in silico screening of a database of drug sensitivities across 39 cancer cell lines showed that tivantinib exhibited similar IC₅₀ values across 39 cancer cells including HCT-15 cells known to naturally overexpress MDR1 (11, 44). Consistent with this previous result, tivantinib showed cytotoxic activity equally against vinca alkaloid–resistant cells or MDR1-overexpressed cells compared with the effect on parental cells (Figs. 4B and 5B). In addition, it is widely known that other ABC transporter BCRP confers resistance to topotecan and that MRP1 confers resistance to adriamycin. Tivantinib equally inhibited growth of cells overexpressing BCRP or MRP1 (Figs. 5B and 6A and B).

In clinical trials, tivantinib showed encouraging antitumor activity and was well tolerated. However, the additional possible targets of tivantinib were unknown. In this study, we showed that tivantinib inhibited tubulin polymerization in cells and tumors. Moreover, we suggest that tivantinib directly interacts with tubulin at the colchicine binding site and overcomes the effects of drug resistance–mediating ABC transporters. However, the detailed mechanism underlying inhibition of tubulin polymerization by tivantinib remains unknown. Moreover, it is expected that cells may acquire resistance to tivantinib in a manner similar to that for other drugs. However, the mechanism of acquired resistance to tivantinib remains unknown. Currently, tivantinib is under clinical evaluation as a selective c-MET inhibitor and showed lower toxicity. Our finding that tivantinib shows cytotoxic activity by disrupting tubulin polymerization, directly interacts with tubulin, and overcomes ABC transporter–mediated drug resistance suggests that tivantinib may be useful for treatment of a variety of other cancers.

No potential conflicts of interest were disclosed.

Conception and design: A. Aoyama, R. Katayama, N. Fujita

Development of methodology: A. Aoyama, R. Katayama, N. Fujita

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Aoyama, R. Katayama, T. Oh-hara, S. Sato

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Aoyama, R. Katayama, T. Oh-hara, S. Sato, Y. Okuno, N. Fujita

Writing, review, and/or revision of the manuscript: A. Aoyama, R. Katayama, Y. Okuno, N. Fujita

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Aoyama, T. Oh-hara, S. Sato, N. Fujita

Study supervision: R. Katayama, N. Fujita

The authors thank Ms. Sumie Koike for helping with data analysis and Drs. Atsushi Miyawaki and Hiroyuki Miyoshi, RIKEN, for providing the CSII-EF-MCS vectors encoding mCherry-hCdt1 and mVenus-hGem.

The study was supported in part by Japan Society for the Promotion of Science KAKENHI grant numbers 24300344 and 22112008 (to N. Fujita) and 25710015 (to R. Katayama) and a research grant of the Princess Takamatsu Cancer Research Fund (to N. Fujita).