Reversible binding of the anticancer drug KXO1 (tirbanibulin) to the colchicine-binding site of β-tubulin explains KXO1's low clinical toxicity

KXO1 (tirbanibulin or KX2–391) is as a non-ATP-competitive inhibitor of SRC proto-oncogene nonreceptor tyrosine kinase (SRC) and is being clinically investigated for the management of various cancers and actinic keratosis. Recently, KXO1 has also been shown to strongly inhibit tubulin. Interestingly, unlike conventional tubulin-targeting drugs, KXO1 has exhibited low toxicity in preclinical and clinical studies, but the reason for this remains elusive, as are the KXO1-binding site and other details of the interaction of KXO1 with tubulin. Here, cell-based experiments revealed that KXO1 induces tubulin depolymerization and G2/M phase cell cycle arrest at low nanomolar concentrations, similar to colchicine, used as a positive control. Results from biochemical experiments, including an N,N-ethylenebis(iodoacetamide) competition assay, disclosed that KXO1 binds to the colchicine-binding site on β-tubulin, further confirmed by the crystal structure of the tubulin–KXO1 complex at 2.5-Å resolution. A high-quality electron density map of the crystallographic data enabled us to unambiguously determine the position and orientation of KXO1 in the colchicine-binding site, revealing the detailed interactions between KXO1 and tubulin. We also found that KXO1 binds reversibly to purified tubulin, induces a totally reversible cellular effect (G2/M cell cycle arrest), and possesses no cellular toxicity 5 days after drug washout, explaining KXO1's low toxicity. In summary, we show that KXO1 binds to the colchicine-binding site of tubulin and resolved the crystal structure of the tubulin–KXO1 complex. Importantly, KXO1's reversible binding to tubulin explains its clinically low toxicity, an insight that could guide further clinical applications of KXO1.

KXO1 (tirbanibulin or KX2-391), an agent with good oral bioavailability, has been shown to inhibit primary tumor growth and metastasis in many preclinical animal models of cancer (1)(2)(3)(4)(5). Since 2007, studies have demonstrated the effectiveness of KXO1 in various solid and liquid tumor types, as well as its low toxicity (5)(6)(7)(8). In a phase I clinical trial in solid tumors, the maximum tolerance dose in patients with advanced cancer was determined to be 40 mg for a twice daily dose with a continuous dosing schedule; even the highest dose administered (80 mg/dose, twice daily) resulted in no evidence of drug accumulation after several days of administration (8). Another phase I study of KXO1 in elderly patients with advanced acute myeloid leukemia found a maximum tolerance dose of 120 mg/dose repeated once daily, and even a dose of 160 mg daily for 12 days caused no obvious side effects (7). Most recently, in two phase III pivotal efficacy studies of actinic keratosis (NCT03285490 and NCT03285477), KXO1 achieved its primary end point of 100% clearance of actinic keratosis lesions in the face or scalp areas within 2 months of treatment. Notably, KXO1 showed dramatically less toxicity compared with the existing standard therapy. Mechanistic studies initially identified KXO1 as a non-ATP-competitive SRC inhibitor (9). However, it also showed an effect on non-SRC-driving cells, suggesting that it may have other molecular targets (5). Photoaffinity labeling of KXO1 identified tubulin as another target of KXO1, and further study showed that KXO1 could inhibit tubulin polymerization in a low-nanomolar range, suggesting that it is a strong tubulin inhibitor (5).
Tubulin is an abundant protein in all cell types and is involved in various physiological activities, including cell migration, protein transport, and mitosis. It is the target of many chemotherapeutic drugs in clinical cancer treatment, including paclitaxel, vinblastine, and eribulin (10). There are six established binding sites on ␣␤-tubulin heterodimers, including the well-known paclitaxel (11), vinblastine (12), and colchicine (13) sites as well as the recently identified laulimalide (14), maytansine (15), and pironetin (16,17) sites. As tubulin inhibitors are nonselective cytotoxic compounds, the tubulinbinding agents in clinical use show good efficacy but exhibit nonnegligible side effects, including chemotherapy-induced peripheral neuropathy, stomatitis, bone marrow suppression, severe weakness, and neurotoxicity (18,19). This inevitable toxicity limits the clinical applications of tubulin inhibitors. KXO1 has exhibited a strong inhibitory activity toward tubulin together with extremely low toxicity in preclinical and clinical trials, in contrast to the characteristics of conventional toxic tubulin inhibitors. This deserves in-depth research, as insight into the underlying mechanism could be of great help in developing highly effective tubulin drugs with low toxicity. In addition, the binding site of KXO1 on tubulin and other details of the interaction are not well-understood, hindering further investigation.
In this study, through observing the changes in microtubule morphology induced by KXO1 and using an N,N-ethylenebis-(iodoacetamide) (EBI) 4 competitive binding assay, we found that KXO1 directly binds to the colchicine site of ␤-tubulin. X-ray data of the KXO1-tubulin complex confirmed the binding of KXO1 at the colchicine site and clearly showed the details of the interaction. Reversibility experiments demonstrated that KXO1 binds to tubulin in a totally reversible manner and induces reversible cellular effects, providing a good explanation for the low toxicity of KXO1. Overall, our results suggest that reversible tubulin inhibitors that bind to the colchicine site of ␤-tubulin could be developed as low-toxicity drugs.

KXO1 inhibits tubulin polymerization and induces G 2 /M cell cycle arrest in a low-nanomolar range
We first examined the effects of KXO1 on the morphology of microtubules in HeLa cells. A tubulin polymerization agent (paclitaxel) and tubulin depolymerization agents (colchicine and vinblastine) were used as positive controls. As shown in Fig.  1A, microtubules in nontreated HeLa cells had a normal network structure (a tubular structure formed by assembly of ␣␤-tubulin heterodimers). Colchicine and vinblastine inhibit polymerization of tubulin, and paclitaxel promotes its excessive polymerization into abnormal microtubules. KXO1 at 30 nM inhibited tubulin polymerization in a colchicine-or vinblastine-like manner, confirming that KXO1 is a strong tubulin depolymerization agent. All tubulin inhibitors inhibit mitosis and block the cell cycle in the G 2 /M phase (20). We found that KXO1 induced G 2 /M phase cell cycle arrest in a concentrationdependent manner. As shown in Fig. 1B, KXO1 caused G 2 /M phase cell cycle arrest with 50% of maximal effective concentration (EC 50 ) values of 53.2 Ϯ 3.73, 39.8 Ϯ 4.71, and 74.6 Ϯ 9.45 nM in HeLa, HepG2, and H460 cells, respectively, further confirming that KXO1 is a strong tubulin inhibitor.

KXO1 binds to the colchicine site of ␤-tubulin
A previous study using photoaffinity labeling of KXO1 showed that KXO1 directly bound to tubulin and inhibited tubulin polymerization in human peripheral blood mononuclear cells (5). However, the specific binding site and details of the interaction were not well-clarified. Tryptophan fluorescence decrease (TFD) is commonly used to detect drugs binding to tubulin (21,22). Using a TFD assay, we showed that KXO1 bound to purified tubulin with a dissociation constant (K d ) of 1.26 Ϯ 0.27 M, whereas the positive compound colchicine showed a K d value of 13.3 Ϯ 3.30 M ( Fig. 2A), suggesting that KXO1 probably binds more strongly to tubulin than does colchicine. In an in vitro tubulin polymerization assay, KXO1 inhibited tubulin polymerization in a concentration-dependent manner and performed a little better than colchicine (Fig. 2B). Next, we aimed to determine the binding site of KXO1 on tubulin. Tubulin polymerization can be inhibited by binding of inhibitors to either the colchicine or the vinblastine site; however, cells treated with high concentrations of vinblastine uniquely show formation of tubulin paracrystals (packing of unpolymerized tubulin in the cytoplasm) (12). High concentrations (2 M) of inhibitors were incubated with HeLa cells for 2 h, and tubulin was observed using immunofluorescence. As shown in Fig. 2C, vinblastine induced the formation of paracrystals in the cytoplasm, whereas colchicine and KXO1 did not, suggesting the possible binding of KXO1 to the colchicine site. For further confirmation, we carried out an EBI assay. One molecule of EBI bound covalently to cysteine 239 and cysteine 354 in the colchicine site of ␤-tubulin simultaneously to form an EBI-␤-tubulin complex, which migrated faster than ␤-tubulin in SDS-PAGE (23). As shown in Fig. 2D, colchicine and KXO1 inhibited the formation of the EBI-␤-tubulin complex, but vinblastine did not, thus confirming KXO1 binding to the colchicine site on tubulin.

X-ray crystal structure of the tubulin-KXO1 complex
Next, we determined the crystal structure of tubulin-bound KXO1 at 2.5-Å resolution by soaking KXO1 (Fig. 3A) in crystals of a protein complex composed of ␣␤-tubulin, stathmin-like protein RB3, and tubulin-tyrosine ligase (T2R-TTL) (24). Details of the data collection and refinement statistics are summarized in Table 1. The resulting high-quality electron density map enabled us to determine unambiguously the position and orientation of the inhibitor KXO1 (Fig. 3B), thus revealing in detail the interaction between KXO1 and tubulin ( Fig. 3, C and D). Consistent with the tubulin polymerization and EBI data ( Fig. 2, C and D), KXO1 bound to the colchicine site (Fig. 3, C and D). As could be seen in the crystal structure, upon binding to the colchicine site, KXO1 formed extensive hydrophobic interactions with various residues of ␤-tubulin, including ␤Ile-4, ␤Phe-169, ␤Leu-242, ␤Leu-248, ␤Leu-252, ␤Leu-255, ␤Met-259, ␤Ile-318, ␤Ile-376, and ␤Lys-352. The amide and pyridine moieties of KXO1 formed hydrogen bonds with the side chain of ␤Glu-200 via a water molecule (Fig. 3D).
KXO1 is much longer than colchicine. Superimposition of the tubulin-KXO1 structure onto that of tubulin-colchicine showed that KXO1 occupied a much broader position upon binding to the colchicine site than did colchicine (Fig. 4). We have proposed a structure-based colchicine-binding site inhibitor (CBSI) pharmacophore model consisting of three hydrophobic centers (I, II, and III) and two hydrogen bond centers (IV and V) (

KXO1 reversibly binds to colchicine site of ␤-tubulin
reported CBSI-tubulin structures, the morpholine group of KXO1 extended out of the colchicine pocket and approached the GTP molecule of ␣-tubulin.

KXO1 induced reversible G2/M cell cycle arrest and showed no toxicity to HeLa cells after washout
KXO1 has exhibited good pharmacological activity together with low toxicity in preclinical and clinical studies. The reversibility of its binding to tubulin is an effective predictor of the toxicity of a microtubule inhibitor (19,26). Generally, irreversible tubulin inhibitors accumulate in cells over a long period and affect the functions of normal cells, resulting in toxicity (26). We used a method developed by Towle et al. (27) to determine the reversibility of KXO1. As shown in Fig. 5A, HeLa cells were treated with a wide range of concentrations of KXO1 (1-10,000 nM) at Ϫ12 h, and compounds were totally washed out at 0 h. Cell cycling was detected at 0 and 10 h, and cell viability was detected at 5 days. Fig. 5B shows cell cycle profiles of KXO1 and positive compound colchicine (an irreversible tubulin inhibitor (27,28)). A proportion of cells in the G 2 /M phase greater than 50% was considered to indicate significant mitotic block (SMB). At 0 h, both KXO1 and colchicine presented SMB at the concentration of 30 nM. However, 10 h after washout, significant differences were observed regarding mitotic block; colchicine at 30 nM continued to induce SMB, whereas a very high concentration (10,000 nM) of KXO1 was required to sustain SMB at this time point. Detection of p-H3 (a marker of G 2 /M phase cell cycle arrest) by immunoblotting at 0 and 12 h gave similar results (Fig. S1). All these results indicate that colchicine is an irreversible tubulin inhibitor, but KXO1 is a totally reversible tubulin inhibitor. Fig. 5C shows 5-day cell viability data for colchicine and KXO1. After 5 days of culture, both colchicine and KXO1 reduced cell viability at the same concentrations that were capable of inducing SMB at 10 h; colchicine inhibited cell growth at concentrations higher than 30 nM, whereas KXO1 only decreased cell viability at 10,000 nM. Overall, KXO1 exhibited no toxicity to cells after washout at

KXO1 reversibly binds to colchicine site of ␤-tubulin
concentrations below 3000 nM, consistent with its observed low toxicity in clinical studies.

KXO1 binds reversibly to purified tubulin
We further investigated the mechanism underlying the difference in reversibility between colchicine and KXO1 inhibition by detecting the binding of compounds to purified tubulin following our previously published method (29). Purified tubulin (25 M) was incubated with 20 M colchicine or 20 M KXO1 for 1 h and then washed three times by ultrafiltration. Compounds in the filtrate and retentate were detected using high-performance LC (HPLC). As shown in Fig. 6A, 20 M colchicine and 20 M KXO1 were detectable by HPLC in controls. After ultrafiltration, we were able to detect almost the same amount of KXO1 in the filtrate as in the control group, whereas there was almost no detectable colchicine (Fig. 6B). By contrast, colchicine, but not KXO1, was detected in the retentate (Fig.  6C). This suggests that KXO1 can be easily washed from tubulin, as almost all KXO1 was detached from tubulin after washing three times, implying that KXO1 binds to purified tubulin in a totally reversible manner.

Discussion
Tubulin inhibitors are a major class of chemotherapy drugs for the treatment of cancer; examples include docetaxel, eribulin, and cabazitaxel (30). Microtubule drugs are cytotoxic and do not show selectivity between normal cells and tumor cells, leading to nonnegligible systemic side effects. Thus, there are severe restrictions on the use of tubulin inhibitors as chemotherapy drugs (20,31). KXO1 is a clinically effective and lowtoxicity microtubule inhibitor (5). It has strong tubulin-inhibitory activity and can significantly inhibit the polymerization of tubulin at low-nanomolar levels. In contrast to conventional

KXO1 reversibly binds to colchicine site of ␤-tubulin
microtubule inhibitors, however, preclinical and clinical trials have confirmed that the toxicity of KXO1 is very low (5). To determine the mechanism of KXO1's low toxicity, we systematically studied the interaction of KXO1 with tubulin. We first confirmed that KXO1 is indeed a highly effective tubulin inhibitor. KXO1 began to induce G 2 /M phase arrest at 10 nM and caused SMB and inhibition of tubulin polymerization at 30 nM in HeLa cells, demonstrating that KXO1 inhibited tubulin activity to a similar extent as the positive compound colchicine. TFD assays showed that KXO1 bound to tubulin with a stronger affinity (as indicated by a lower K d value) than colchicine, and in vitro tubulin polymerization experiments confirmed that KXO1 was slightly more potent than colchicine in inducing tubulin depolymerization. Of note, the TFD assay used to determine the K d values could only approximate the relative binding affinity of compounds to tubulin, and the results do not necessarily reflect the real binding affinity. We can only conclude that KXO1 showed greater affinity for tubulin than did colchicine. The increase of turbidity in in vitro tubulin polymerization requires high concentrations of tubulin to form high-polymerized microtubules. The concentration of tubulin (30 M) in the assay was high, but a lower concentration (5 M) of inhibitors could break down high-polymerized micro-tubules to a low-polymerized form, thus decreasing the turbidity. By monitoring changes in microtubule morphology and performing EBI competitive binding experiments, we determined that KXO1 binds to the colchicine site on tubulin; this was further confirmed by the X-ray structure of a tubulin-KXO1 complex. The X-ray data also allowed us to accurately determine specific details of the interaction between KXO1 and the tubulin colchicine site, providing guidance for the subsequent optimization of KXO1 design.
The reversibility of the binding of a tubulin inhibitor to tubulin can generally predict its toxicity. In general, irreversible compounds have higher toxicity, and reversible compounds are less toxic (26). Reversibility experiments showed that the irreversible inhibitor colchicine could maintain SMB at 30 nM after the drug had been washed out, whereas a much higher concentration of KXO1 (10,000 nM) was required to maintain SMB. That is, at concentrations below 10,000 nM, cell viability was not affected after drug washout in KXO1-treated cells, as confirmed by 5-day cell viability tests. We also found that colchicine bound irreversibly to purified tubulin, whereas KXO1 bound in a totally reversible manner (KXO1 could be easily eluted from purified tubulin). This reversible binding provides a plausible explanation for KXO1's low toxicity.

KXO1 reversibly binds to colchicine site of ␤-tubulin
Although we aimed to explain from a structural biology perspective why different compounds show different reversibility of binding to tubulin, we did not find significant differences when comparing the crystal structure of tubulin-KXO1 with that of tubulin-colchicine. It is worth noting that minor differences in the structures of compounds may lead to completely different reversibility of binding, as in the cases of colchicine (irreversible) and colcemid (reversible), eribulin (irreversible) and ER-076349 (partially reversible), and vinblastine (reversible) and vincristine (irreversible) (27,29). The differences in reversibility of these compounds are difficult to explain from a structural biology perspective. It is possible that the kinetics of binding to tubulin affect reversibility; however, this requires further validation.
In summary, in this study we showed that KXO1, a clinical phase III drug, binds to the colchicine site of tubulin and resolved the crystal structure of a tubulin-KXO1 complex. Importantly, we found that the binding of KXO1 to tubulin was fully reversible, which could explain the clinically low toxicity of KXO1. Our results have the potential to guide further clinical applications of KXO1.

Cell lines and cell culture
Human cervical adenocarcinoma cell line HeLa, human hepatoma cell line HepG2, and human large-cell lung cancer H460 were sourced from American Type Culture Collection. Cells were cultured with Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum; 100 units/ml penicillin and 100 g/ml streptomycin were also added to the medium to prevent bacterial contamination of cells. The culture environment was set to 5% CO 2 at 37°C. HeLa cells were authenticated by short tandem repeat testing by Cobioer Biosciences Co., Ltd.

Immunofluorescence
HeLa cells cultured in a 24-well plate were incubated with compounds for the appropriate times and then washed with phosphate-buffered saline (PBS) for 2 min. The cells were then incubated with 50% methanol, 50% acetone for 2 min for fixation and washed with PBS for another 2 min. Fixed cells were incubated with anti-␣-tubulin antibody (1:200) at 4°C for 12 h and then washed with 1ϫ PBST (PBS containing 0.1% Tween 20) for 20 min (5 min ϫ 4) before incubation with a second antibody for 45 min at room temperature. After washing for another 20 min (5 min ϫ 4), cells were stained with 4Ј,6-diamidino-2-phenylindole (DAPI) for 10 min before being washed and transferred to slides. Finally, cells were imaged using a fluorescence microscope (Olympus, Tokyo, Japan).

Cell cycle analysis
Cells cultured in 6-well plates were collected and washed with PBS before being fixed with 70% precooled ethanol overnight at 4°C. Cells were then washed three times with PBS again and stained with 50 g/ml PI for 20 min. Cell cycling in these samples was analyzed using a flow cytometer (BD FACSCalibur).

In vitro tubulin polymerization assay
Purified tubulin (30 M) in PEM buffer was incubated with different compounds at 4°C for 1 min before transfer to a prewarmed (37°C) 384-well plate for immediate analysis with a Biotech Gen5 spectrophotometer. Optical density values at 340 nm were recorded at 1-min intervals for 18 min. The polymerization curve was obtained from the absorption values at consecutive time points.

EBI competition assay
The assay was carried out following the method published by Fortin et al. (23). One molecule of EBI bound covalently to cysteine 239 and cysteine 354 in the colchicine site of ␤-tubulin simultaneously to form an EBI-␤-tubulin complex, which migrated faster than ␤-tubulin in SDS-PAGE. Tubulin inhibitors that could inhibit the formation of EBI-␤-tubulin complex were considered as CBSIs.
HeLa cells were cultured in 6-well plates and then treated with or without different concentrations of KXO1, vinblastine, or colchicine for 2 h before being treated with 100 M EBI for another 2 h. Total protein was extracted with radioimmunoprecipitation assay buffer and subjected to Western blot analysis of ␤-tubulin. GAPDH was used as a loading control. The Western blotting procedure was as described previously (29).

Structural biology
Protein expression and purification-The T2R-TTL complex, which includes two tubulins, one stathmin-like domain of RB3 (RB3-SLD), and one TTL, was prepared as described previously (25). Porcine brain tubulin was supplied at 10 mg/ml in PEM buffer and stored at 80°C until use. RB3-SLD was overexpressed in Escherichia coli strain BL21 (DE3) and purified sequentially by anion-exchange chromatography (Q Sepharose Fast Flow, GE Healthcare) and size-exclusion chromatography (Superdex 75, GE Healthcare). The purified RB3-SLD protein was concentrated to 10 mg/ml and stored at 80°C until use. TTL was overexpressed in E. coli strain BL21 (DE3) and purified by nickel-affinity chromatography followed by size-exclusion chromatography (Superdex 200, GE Healthcare). Purified TTL protein in Bistris propane (pH 6.5), 200 mM NaCl, 2.5 mM MgCl 2 , 5 mM ␤-mercaptoethanol, 1% glycerol was concentrated to 20 mg/ml and stored at 80°C until use. The T2R-TTL complex was formed by mixing tubulin, RB3-SLD, and TTL in a 2:1.3:1.2 molar ratio; 5 mM tyrosine, 10 mM DTT, and 1 mM ␤,␥-methyleneadenosine 5Ј-triphosphate disodium salt were then added; and the complex was concentrated to 15 mg/ml at 4°C.
Crystallization and crystal soaking-The T2R-TTL crystals were obtained by the sitting-drop vapor-diffusion method at 20°C in a buffer consisting of 6% PEG4000, 8% glycerol, 0.1 M MES (pH 6.7), 30 mM CaCl 2 , and 30 mM MgCl 2 . The seeding method was used to obtain single crystals. Rodlike crystals appeared within 2 days and grew to maximum dimensions within 1 week. For crystal soaking, 0.1 l of KXO1 (dissolved in DMSO at 10 mM concentration) was added to a 2-l crystalcontaining drop for 16 h at 20°C.
Data collection and structure determination-The crystals were transferred into cryoprotectant (reservoir solution supplemented with 20% (v/v) glycerol) for a few seconds, then mounted in nylon loops, and flash-cooled in liquid nitrogen. Diffraction data were collected on beamline BL19U1 at the Shanghai Synchrotron Radiation Facility (SSRF; Shanghai, China). Data were processed using HKL3000 (32). The structures were determined by the molecular replacement method using the tubulin-tivantinib structure (Protein Data Bank (PDB) code 5CB4) as a search model. The refinement was per-formed using Coot (33) and PHENIX (34). The model quality was evaluated with MolProbity (35). All figures were generated using PyMOL.

Mitotic block reversibility assay
The mitotic block reversibility assay was carried out following the protocol of Towle et al. (27). As shown in Fig. 5A, HeLa cells were treated with increasing concentrations (0, 10, 30, 100, 300, 1,000, 3,000, and 10,000 nM) of colchicine or KXO1 at the Ϫ12-h time point and cultured for 12 h. At 0 h, compounds were totally washed out, and cells were incubated with fresh medium and further cultured for 5 days. At 0 and 10 h, cell cycle analysis and p-H3 expression levels were detected by flow cytometry and Western blotting, respectively. At the 5-day time point, cell viability was examined with a trypan blue staining assay following a previously established method (36).

HPLC detection of compounds binding to tubulin
Purified tubulin (25 M) in PEM buffer (100 l) was incubated with 20 M colchicine or 20 M KXO1 for 60 min at room temperature, after which samples were transferred to ultrafiltration tubes (Millipore; 10 kDa) for centrifugation at 13,000 rpm for 15 min. Then 400 l of PEM buffer was added to the ultrafiltration tube to wash the tubulin with centrifugation at 13,000 rpm for 15 min again. This washing procedure was repeated three times. The filtrates were combined and dried under vacuum before being redissolved in 100 M methanol to obtain the filtrate fraction. The retentate (tubulin) was heated to 80°C to denature the tubulin protein and release the bound compounds. The denatured retentate was also dried under vacuum before being redissolved in 100 M methanol to yield the retentate fraction. The filtrate fraction and retentate fraction were subjected to HPLC for detection of compounds. HPLC was also performed on 20 M colchicine and 20 M KXO1 in 100 l of methanol as controls. HPLC was performed using a Waters Alliance 2695 separation module (Empower Software) with a Waters 2996 photodiode array detector.

Statistical analysis
Data are shown as mean Ϯ S.D. Statistical differences were determined using an unpaired, nonparametric Student's t test. p values are indicated in each graph or figure legend: *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001.