Discovery of Small Molecule Inhibitors that Interact with γ-Tubulin

Authors


Corresponding author: Jack A. Tuszynski, jackt@ualberta.ca

Abstract

Recent studies have shown an overexpression of γ-tubulin in human glioblastomas and glioblastoma cell lines. As the 2-year survival rate for glioblastoma is very poor, potential benefit exists for discovering novel chemotherapeutic agents that can inhibit γ-tubulin, which is known to form a ring complex that acts as a microtubule nucleation center. We present experimental evidence that colchicine and combretastatin A-4 bind to γ-tubulin, which are to our knowledge the first drug-like compounds known to interact with γ-tubulin. Molecular dynamics simulations and docking studies were used to analyze the hypothesized γ-tubulin binding domain of these compounds. The suitability of the potential binding modes was evaluated and suggests the subsequent rational design of novel targeted inhibitors of γ-tubulin.

Microtubules, actin filaments, and intermediate filaments form the basis of the eukaryotic cytoskeleton. Microtubules are the key to the correct functioning of the cell by providing structure, intracellular transport, and the creation of the mitotic spindle during cell division. They are macromolecules consisting mainly of αβ-tubulin heterodimers and γ-tubulin at the nucleation site of the microtubule. γ-Tubulin is most known for its role as a nucleating agent for the creation of microtubules. In healthy dividing cells, γ-tubulin is associated at microtubule organizing centers (1–3). Current models suggest that γ-tubulin associates into a large γ-tubulin ring complex (γTuRC), where αβ-tubulin dimers then proceed to polymerize into this template, creating a microtubule (4–7). While α- and β-tubulin are abundant proteins, found in approximately equal amounts and making up approximately 2.5% of the total protein in a cell, γ-tubulin is decidedly less prevalent and makes up <1% of the total tubulin content of the cell (3).

In humans, there are two main isoforms of γ-tubulin denoted: TUBG1 and TUBG2. The isoforms are very closely related, with over 97% amino acid sequence identity (8), and it has been suggested that any chemical agent targeting one isoform would probably target the other (8). The reason for two separate isoforms of γ-tubulin in humans has not yet been conclusively established (8,9), and both isoforms were expressed in all tissues studied by Wise et al. (8). In mice, it was found that TUBG1 was expressed ubiquitously, while TUBG2 was primarily found only in the brain (9). Additionally, in mouse knockout studies, it was found that TUBG1, as opposed to TUBG2, is critical to proper mitotic division, and thus, it was concluded that TUBG1 corresponds to conventional γ-tubulin, while TUBG2 has some still unknown function (9). Although isoforms of γ-tubulin have been shown to have different expression levels and probably distinct functions in various species (9,10), in mouse studies, it was found that mouse γ-tubulin genes are orthologs of human TUBG1 and TUBG2 (9). As both TUGB1 and TUBG2 are found to be overexpressed in human glioblastoma (11,12), and in mouse knockout studies TUBG1 as opposed to TUBG2 was critical to proper mitosis (9), the present study is entirely focused on TUBG1 γ-tubulin.

As microtubules play a pivotal role in mitosis by creating the mitotic spindle, tubulin has long been a natural target for chemotherapeutics (13,14). Chemotherapies targeting tubulin mainly involve the taxanes (e.g., paclitaxel and docetaxel) and the vinca alkaloids (e.g., vinblastine, vincristine). Paclitaxel is used as part of the treatment in breast, ovarian, prostate, non-small-cell lung cancer, and Kaposi’s sarcoma (13). Vinblastine is used to treat leukemia and lymphomas (13). Other drugs found to bind to tubulin include colchicine and, more recently, combretastatin, laulimalide, peloruside, and noscapine (15–18). However, all of these drugs are thought to primarily target β-tubulin. To our knowledge, there are currently no known inhibitors of γ-tubulin.

Potential for therapeutic importance exists for a sufficiently specific and selective inhibitor of γ-tubulin. It has been shown that γ-tubulin is absolutely required for the assembly of mitotic spindle microtubules and an absence of functional tubulin leads to inhibition of mitosis (9,19–21). TUBG1 γ-tubulin has been found to be overexpressed in glioblastoma multiforme (GBM) (11,22–24), the most prevalent and deadly form of brain cancer. While brain cancer is a relatively rare form of cancer (25), the prognosis for GBM is dismal, with a uniform survival of 3 months without treatment, and a median survival of only approximately 14.6 months with recent advances in treatment (26,27). γ-Tubulin is also altered in breast cancer cells (28) and overexpressed in preinvasive lesions and carcinomas of the breast (29,30). As γ-tubulin overexpression and altered cellular distribution are linked to tumor progression and anaplastic potential (11), a potent inhibitor of γ-tubulin would possibly halt mitosis in GBM and lead to increased understanding for the altered expression of γ-tubulin in multiple cancers.

Thus, in this study, we have targeted the discovery of a ligand interaction with γ-tubulin. This is currently possible in view of the recent crystallographic studies of γ-tubulin (31,32) that allow for a computational search of ligands with high binding affinities. Because γ-tubulin is structurally quite similar to β-tubulin (31), we concentrated our search on binding sites on γ-tubulin with homologous locations near binding sites of known β-tubulin inhibitors (see Figure 1). Our experimental results, using fluorescence spectroscopy and recombinant human γ-tubulin, indicate that colchicine and combretastatin A-4 bind to γ-tubulin (see Table 1 for compound structures). Computational studies identify both a possible binding site and a pose for these compounds bound to γ-tubulin. Experimental validation of the hypothesized binding site on γ-tubulin and subsequent virtual screening of derivatives of colchicine, combretastatin A-4, and various other drug compounds in this site are underway.

Figure 1.

 Homologous location of drugs known to target β-tubulin on γ-tubulin. This model was created using the crystal structure from PDB entry 1SA0, which contains two αβ-tubulin heterodimers bound together. The β-tubulin bound to the αβ-tubulin heterodimer was replaced with γ-tubulin in this model to illustrate hypothetical binding sites on γ-tubulin of ligands known to bind to β-tubulin. From this analysis, colchicine bound to γ-tubulin may bind in a region, which might interfere with the binding of γ-tubulin complex proteins (GCPs).

Table 1.   The structures of colchicine, colchicine derivatives, and combretastatin A-4 Thumbnail image of

Experimental Methods

Experimental apparatus and reagents

The human TUBG1 gene was inserted into a pET15b vector between the XhoI and HindIII restriction sites (see Figure S1). The identity of the mutation and the correctness of gene insertion (location and orientation) were verified by DNA sequencing. The recombinant protein was expressed in Eschericia coli BL21(DE3) host cells in LB medium supplemented with 100 μg/mL ampicillin. The cultures were grown at 37 °C until OD600 reached a value of 0.8 and then induced with 1.0 mm IPTG for 18 h at 25 °C. Subsequently, the cells were harvested by centrifugation (9060 g for 20 min in JS 7.5 rotor). The γ-tubulin protein was then isolated and purified from the inclusion bodies via refolding by dilution with immobilized metal ion affinity chromatography using a Ni-NTA column.

More specifically, the cell pellet from 1 L of culture with expressed γ-tubulin was resuspended in 25 mL of lysis buffer and lyzed by sonication (using Fisher Scientific Ultrasonic Dismembrator Model 500 with microtip probe for 4 × 15 seconds pulses at 45% power) on ice and centrifuged at 17 400 g for 20 min (at 4 °C) in a JA 25–50 rotor. The supernatant was removed, and inclusion bodies were cleaned by a series of washing procedures with buffer A (buffer A: 50 mm Tris, 50 mm MgSO4, 50 mm NaCl, pH 8.8) containing 0.1% Triton X-100, 25% glycerol, 500 mm NaCl an 2 m Urea as separate additives for every next washing step. Inclusion bodies were centrifuged at 17 400 g for 20 min (at 4 °C) in a JA 25–50 rotor after every wash, and the supernatant was removed. Clean protein pellets were solubilized in a buffer B (buffer B: 50 mm Tris, 50 mm NaCl, 1 mm CaCl2, 8 m Urea, 10 mm beta-mercaptoethanol, pH 8.8) and left for overnight incubation at room temperature with the next centrifugation at 48 400 g for 1 h (25 °C) in a JA 25–50 rotor. The γ-tubulin protein was refolded by fast dilution (1:10) into the buffer C (buffer C: 50 mm Tris, 50 mm NaCl, 10 mm MgSO4, 1 mm CaCl2, pH 8.8 and loaded onto a Ni-NTA column (12.5 mL bed volume) pre-equilibrated with buffer C. The loaded sample was incubated on a column for 1 h (at 4 °C) with rotation. The column was then washed with buffer D (buffer D: 50 mm Tris, 50 mm NaCl, 10 mm MgSO4, 1 mm CaCl2, 10 mm Imidazole, pH 7.2), and γ-tubulin was eluted with a linear gradient of 500 mm imidazole in buffer D. Fractions with protein were identified by SDS–PAGE mixed and dialyzed overnight (at 4 °C with two buffer changes) against 10 volumes of buffer E (buffer E: 25 mm Tris, 25 mm NaCl, 10 mm MgSO4, 1 mm MgCl2, 1 mm CaCl2, pH 7.3). The protein concentration was determined using an extinction coefficient at 280 nm of 47 705/m/cm. The extinction coefficient was calculated by Protparam program (33) based on recombinant γ-tubulin amino acid sequence. The TUBG1 protein was then concentrated using an Amicon Ultra-15 centrifugal device.

Fluorescence emission spectra were recorded on a PTI MODEL-MP1 spectrofluorometer using a 1-cm fluorescence cell in all measurements. The excitation wavelength of 295 nm was used, and the scan range was 310–450 nm. The gene for human γ-tubulin was purchased from DNA2.0 Inc. (Menlo Park, CA, USA). All reagents were purchased from Sigma-Aldrich Canada Ltd. (Oakville, ON, Canada) and Fisher Scientific Company (Ottawa, ON, Canada). NI-NTA resin was purchased from Qiagen Inc. (Toronto, ON, Canada). The purity of colchicine was >95%, and it was determined by liquid chromatography-Mass Spectrometry [Agilent Eclipse plus C18 column, 250 × 4.6 mm, 5 μm particle size; mobile phase, water/acetonitrile (0.1% HCOOH) 80:20–5:95 over 5 min and then held for 1.5 min; flow rate 0.5 mL/min]. The sample was dissolved in methanol.

Combretastatin A-4 was purchased from Sigma-Aldrich Canada Ltd. (Oakville, ON, Canada). The purity of combretastatin A-4 was ≥98%, determined by high-performance liquid chromatography.

Binding experiments and fluorescence assays

In a 96-well microplate placed in a shallow tray with ice, γ-tubulin was mixed with the assay buffer to reach a final protein concentration of 2 μm. Guanosine 5′-triphosphate (GTP) was added to the samples and control to keep a final concentration at 1 mm. The microplate was incubated on ice for 10 min, and then calculated amounts of stock solution of colchicine in DMSO were added to the samples to reach final ligand concentrations of 10, 40, and 80 μm, respectively. The control was ligand-free. The total volume of the sample was 100 μL. A glass bead was inserted in each well, and the microplate was covered with protective film, sealed with a lid, and incubated for 30 min at 25 °C. After that time, the microplate was transferred to a rotating platform, and it was vigorously rotated for 1 h at room temperature. From each well, 80 μL of samples and control were transferred to a 1-cm fluorescence cell. Fluorescence of the tubulin was monitored at 295 nm (excitation wavelength), and the scan range was 310–450 nm.

Estimation of the binding constant

Data from the fluorescence experiments were used to determine the apparent binding constant of γ-tubulin in the presence of the ligands according to the formula:

image(1)

where KS = KA, the formation constant of the donor-acceptor (quencher-fluorogen) complex. The concentration of the quencher [Q] after titration was taken to be its ratio to γ-tubulin concentration [Pt]: [Q]/[Pt] (34). From the slope of the linear plot of Fo/F versus [Q]/[Pt], the binding constant and dissociation constant (1/KA) were estimated. The results were expressed as mean values ±SD (n = 3). The inner filter effects were corrected empirically by measuring the change of fluorescence intensity of a tryptophan solution equivalent to the γ-tubulin concentration in the presence of colchicine, and the corrected fluorescence intensities were used for all calculations (35).

Amine cross-linking assay

In 1.5-mL Eppendorf tubes prechilled on ice, recombinant γ-tubulin (in 10 mm potassium phosphate, 1 mm MgSO4, 1 mm MgCl2, pH 7.6) was mixed with αβ-tubulin dimer (phosphocellulose purified from a bovine source, kindly provided by Dr Richard Ludueña, University of Texas Health Science Center, San Antonio, TX, USA) and 10 mm potassium phosphate buffer (pH 7.6) to reach the final protein concentration of 2 μm each. GTP was added at the same time to a final concentration of 1 mm. The protein mixture was incubated on ice for 5 min, and then calculated amounts of stock solution of disuccinimidyl glutarate (DSG) in DMSO were added to the samples to make cross-linking reagent/protein ratios of 1:1, 0.1:1, and 0.01:1, respectively. Two control samples were prepared for every DSG/tubulin combination. The first control sample included the cross-linking reagent and γ-tubulin only. The second control sample included the cross-linking reagent and the αβ-tubulin dimer only. The total volume of the sample was 100 μL. Samples were incubated for 30 min at room temperature. After that time, the cross-linking reaction was quenched by the addition of 1 m Tris, pH 8.0 (50 mm final concentration), and samples were incubated at room temperature for an additional 15 min; 12% SDS–PAGE protein gel was used to visualize the amine cross-linking assay results.

Comparison between β-tubulin and γ-tubulin structures

The first X-ray crystal structure of γ-tubulin (bound to GTPγS) was reported by Aldaz et al. in 2005 (31) at 2.7 Å resolution (PDB entries 1Z5V and 1Z5W). Rice et al. then reported a refined 2.3 Å crystal structure of γ-tubulin bound to GDP in 2008 (PDB entry 3CB2) (32). It was found that the structure of γ-tubulin bound to GDP was in a curved conformation identical to that of γ-tubulin bound to GTPγS (32). It was then concluded that γ-tubulin does not undergo a curved-to-straight domain change based on its nucleotide binding state (32). Thus, while γ- and β-tubulin bind to GTP with similar affinities, and have a similar preference to bind to GTP over GDP as found by Aldaz et al. (31), we used the latest crystal structure from PDB entry 3CB2 of γ-tubulin bound to GDP as the basis of our computational search for a γ-tubulin inhibitor because this file would give us the highest resolution of the γ-tubulin structure.

MD simulations

Molecular dynamics (MD) simulations followed the procedure outlined by Barakat et al. (36) using NAMD (37). PDB2PQR (38) and PTRAJ using the AMBER99SB force field (39) from Amber 10 (40) were used in setting up the system as described by Barakat et al. (36). γ-Tubulin (residues 2–446) and the associated GDP from chain A were taken from PDB entry 3CB2 (32). Swiss-PdbViewer 4.0 (41) was used to model the missing residues in the crystal structure (residues 1, 278–283, 311–312, 367–371, and 447–451). GDP was modeled using Amber 99 force field parameters developed by Meagher et al. (42). A TIP3P water cube buffer that ensured 15 Å of water surrounded the system added 26 499 water molecules. Counterions were added to reproduce physiological ionic concentrations. The system was equilibrated for 519 picoseconds (ps) by gradually releasing the constraints on the backbone atoms. The simulation, using periodic boundary conditions, then continued for 24 nanoseconds (ns) during which atomic co-ordinates were saved from the trajectory every 2 ps. To control the pressure at a standard atmospheric pressure value, a Langevin piston period of 300 femtoseconds was used.

Clustering

To generate a reduced representative sample of γ-tubulin conformations over the entire MD simulation trajectory, we followed the same method used by Barakat et al. (36,43). We examined the final 16 ns of our 24 ns MD simulation, when the RMSD of the backbone atoms of the structure had stabilized (see Figure S2). RMSD and B-factor calculations were performed using PTRAJ in Amber 11 (44) (see Figure S3).

To extract representative structures for the colchicine binding site, we performed RMSD clustering on the 20 residues (excluding their hydrogens) that line the hypothesized colchicine binding site in γ-tubulin, namely: 242, 243, 248, 250–256, 258, 259, 321, 323, 334, 337, 341, 357–359. Clustering was performed on 4001 structures every 4 ps in the final 16 ns of the MD simulation. As described below, using the Davies-Bouldin index (DBI) and sum of squares of the regression (SSR)/ total sum of squares (SST) metrics, an optimal number of 35 clusters were found, and the centroid of each cluster was chosen as the representative structure. These structures were later evaluated by our docking studies.

Docking

The colchicine ligand used for docking was the colchicine structure found from PDB entry 1SA0, chain B (45). This colchicine ligand in PDB entry 1SA0 was actually a colchicine derivative, with an added sulfur atom. The sulfur atom was removed, and the carbon previously bonded to the sulfur (C13) was modified to be located at 1.5046 Å from C12 to agree with the distance found in the colchicine structure from PubChem (CID_6167.sdf) and to generate a computational structure of the colchicine ligand.

All docking simulations were performed using AutoDock, version 4.0 (46). Partial atomic charges were assigned to both the protein and docked ligands using the Gasteiger-Marsili method (47). Atomic solvation parameters were assigned to the atoms of the protein using the AutoDock 4.0 utility ADDSOL. Docking grid maps with 54 × 60 × 126 points and grid point spacing of 0.31 Å were then centered on the hypothesized colchicine binding site within γ- and β-tubulin using the AutoGrid 4.0 program (46). Rotatable bonds of each ligand were then automatically assigned using the AUTOTORS utility of AutoDock 4.0. Docking was performed using the Lamarckian genetic algorithm (LGA) method with an initial population of 400 random individuals; a maximum number of 10 × 106 energy evaluations; 100 trials; 30 000 maximum generations; a mutation rate of 0.02; a crossover rate of 0.80 and the requirement that only one individual can survive into the next generation. Independent docking runs were performed for each protein structure and colchicine derivative with all residues of the receptors set being rigid during the simulations.

Colchicine derivatives

The structures of colchicine and colchicine derivatives used in this study are given in Table 1. Elsewhere, we have described in detail the generation and characterization of colchicine derivatives based on the differences observed among the most commonly expressed human tubulin isotypes: βI, βII, βIII, βIV, and βV (48). The colchicine derivatives were modified at either the C1 or C3 methoxy position of the A-ring (see Table 1). Acylation of the common intermediates 1- or 3-demethylcolchicine afforded ester derivatives, while alkylation gave ether derivatives. The general synthetic schemes for the ester and ether derivatives of colchicine were based on previously published schemes (49). The modification to the first analog was carried out by replacing the -OCH3 in the C13 position of the A-ring of colchicine by different -OX groups. In the second analog, the -OCH3 in the C11 position of the A-ring and the -OCH3 in the C-ring of colchicine were replaced by different -OY groups and -SCH3, respectively. All of these resulting derivative structures including colchicine were constructed using the molden program (50). In constructing the models, we used colchicine derived from PDB entry 1SA0 as the starting structure (see Docking section above) (45).

Equivalent binding site energies

Given the AutoDock predicted binding energies (ΔG) in kcal/mol, we used the formula: ΔG = RT ln (KD) to calculate the equivalent KD value.

Figure representations

All figures representing proteins were generated using visual molecular dynamics (VMD) (51) and were aligned using the RMSD Trajectory Tool plugin for VMD. Electrostatic calculations for figures were performed using APBS Plugin for VMD (52).

Results

Discovery of colchicine and combretastatin A-4 binding to γ-tubulin

Identification of a potent small molecule inhibitor of γ-tubulin first requires finding a molecule that binds to γ-tubulin. The crystal structures of both γ-tubulin (TUBG1) and β-tubulin (TUBB2B) are presently available (PDB entries 3CB2 (32) and 1SA0 (45), respectively), and comparison of the two yields a very similar tertiary structure, discussed by Aldaz et al., (31) with an RMSD of only 1.18 Å in the intermediate β-sheets (see Figure 2). There is a 33% sequence identity and a 75% structure similarity of the two structures (53). Given this similarity of these structures, we tested colchicine, combretastatin A-4, and paclitaxel, all known to bind to β-tubulin, using a fluorescence spectroscopy assay with recombinant human γ-tubulin, and discovered evidence that colchicine and combretastatin A-4, both inhibitors of β-tubulin, also bind to γ-tubulin. Paclitaxel showed no evidence of binding to γ-tubulin (data not shown). Colchicine and combretastatin A-4 are known to bind to the same binding site on β-tubulin.

Figure 2.

 β-tubulin from PDB entry 1SA0 (cyan) and γ-tubulin from PDB entry 3CB2 (purple) are aligned, showing tertiary structure similarity. Colchicine docked to β-tubulin is shown. Tryptophans from β-tubulin are shown in red, and tryptophans from γ-tubulin are shown in yellow.

The tubulin heterodimer, which contains four tryptophan residues in both the α- and β- monomer, exhibits intrinsic fluorescence, which can be used to detect ligand binding and to monitor conformational changes of the protein associated with binding (54–56). Fluorescence quenching as a result of colchicine binding to tubulin has been extensively observed and analyzed (55,56). Given that γ-tubulin contains four tryptophan residues in three of the same areas (see Figure 2) and has a high tertiary structure similarity to β-tubulin, fluorescence spectroscopy was selected as a reasonable method to detect a binding event of colchicine to γ-tubulin. The fluorescence spectrum of human γ-tubulin in the presence of colchicine is shown in Figure 3A. The final fluorescence value was corrected for any possible inner filter effect. A notable quenching of the tryptophan fluorescence of γ-tubulin was observed upon addition of increasing concentration of colchicine as a result of ligand binding. According to the graph shown in Figure 3B, the addition of 80 μm of the ligand quenched more than 70% of the fluorescence of the protein expressed as the percentage of fluorescence extinction of the tubulin tryptophans (denoted as F/Fo, %). Similar to the results found previously for other tubulin isoforms (57), colchicine induces structural changes in γ-tubulin upon binding because a blue shift of the maximum emission wavelength (Δλem = 11 nm) was observed (see Figure 3A), which suggests a change in the polarity of the environment surrounding one or more of the emitting tryptophan residues. The Stern-Volmer plot of the fluorescence data obtained from the colchicine binding is linear in the range of concentrations tested for this ligand. From this graph, the value of the apparent dissociation constant was estimated as, KD = 13.9 ± 0.4 μm (see Figure 3C), which is equivalent to a binding energy of −6.90 ± 0.03 kcal/mol at 310 K. This KD value is virtually identical to our experimental results of colchicine binding to αIβI-tubulin (V. Semenchenko, R. Perez-Pineiro, D.S. Wishart, Unpublished data, 2011), which suggests colchicine has an equal affinity to both γ-tubulin and β-tubulin. This direct comparison of fluorescence quenching data of colchicine interacting with both γ-tubulin and αIβI-tubulin is found in Figure S4. Published KD values of colchicine binding to bovine tubulin yield KD values of 1.1 μm (59,60), which is an order of magnitude less than our experimental results, but may be due to experimental conditions and differing methods of measurement.

Figure 3.

 Fluorescence emission spectra of γ-tubulin (2 μm) in the presence of colchicine: 0 μm (1), 10 μm (2), 40 μm (3), 80 μm (4). λexcit = 295 nm, slit width 5 nm (A). Tryptophan fluorescence quenching of γ-tubulin (2 μm) plotted as extinction of tubulin tryptophans (F/Fo, %) in the presence of increasing concentrations of colchicine (B). The Stern-Volmer plots of fluorescence quenching of γ-tubulin by colchicine (C). The value of the KD estimated from this plot by using the eqn 1 is 13.9 ± 0.4 μm.

The fluorescence spectrum of human γ-tubulin in the presence of combretastatin A-4 is shown in Figure 4. This shows combretastatin A-4 binding to γ-tubulin in a concentration-dependent fashion. However, the ligand has a strong fluorescence emission (450 nm) at the same excitation wavelength of tryptophan (295 nm) that interferes with the protein quenching. Thus, a quenching effect as shown for the interaction between colchicine and γ-tubulin has not been clearly demonstrated yet, and hence, KD values were not calculated. We plan to perform more careful measurements for combretastatin and its analogs in the near future. Published results show that combretastatin binds to tubulin with a KD of 0.40 ± 0.06 μm, making it a stronger binder than colchicine to the colchicine binding site on β-tubulin (60).

Figure 4.

 Fluorescence emission spectra of γ-tubulin (2 μm) in the presence of different concentrations of combretastatin A-4: 0 μm (1), 10 μm (2), 40 μm (3), 80 μm (4). λexcit = 295 nm, slit width 3 nm.

Validation of the quality of γ-tubulin created

As recombinant γ-tubulin studied in isolation is a relatively novel construct, we performed an αβ-tubulin dimer and γ-tubulin amine cross-linking assays to assess whether the γ-tubulin created by us was in a functionally correct conformational state. In theory, if we have an αβ-tubulin dimer and functional γ-tubulin, these proteins would combine to make a trimer. Each tubulin monomer’s molecular weight is ∼55 kDa. Our 12% SDS–PAGE gel showed evidence of the expected trimer of γ-tubulin bound to the αβ-tubulin dimer in the 130–170 kDa molecular weight range, in the assay involving the reactive mixture of αβ-tubulin, γ-tubulin, and DSG (see Figure S5). Controls of reactive mixtures with just γ-tubulin and just αβ-tubulin combined with DSG showed no bands in the 130–170 kDa molecular weight range as expected (see Figure S5). This suggests our prepared γ-tubulin binds as expected to the αβ-tubulin dimer, which is consistent with it being in a functional state. A more comprehensive analysis of γ-tubulin functionality was beyond the scope of this project, but we intend to perform appropriate assays in a future study. Multiple preparations of γ-tubulin showed the same fluorescence quenching in the presence of colchicine (see Figure S6).

Additionally, we have used the same methods for expressing recombinant tubulin in the case of β-tubulin single-residue mutants and tested their products by performing binding affinity assays with a host of ligands known to bind to αβ-tubulin dimers. The results obtained for binding colchicine to γ-tubulin closely reproduced the known values of the binding kinetics constants of colchicine to β-tubulin giving us a level of confidence that the method produces a protein that is properly folded.

Binding site prediction of colchicine to γ-tubulin

Given the experimental information that suggests colchicine and combretastatin A-4 bind to γ-tubulin, we used computational methods to predict the binding pose and location of these compounds bound to γ-tubulin. The crystal structure of γ-tubulin from PDB entry 3CB2 (TUBG1) was used for the computational docking process (32). This structure was missing 19 residues that were repaired using Swiss-PdbViewer 4.0 (see Experimental Methods) (41). An MD simulation of γ-tubulin bound to GDP in explicit water solvent was performed for 24 ns. The last 16 ns of this simulation were identified to have a stable RMSD from the initial reference structure for residues not in the N- and C-termini, that is, residues 2–438 (see Figure S2). Fluctuation analysis shows the movement of each residue in the simulation and clearly identifies the N- and C-termini (see Figure S3). The homologous region on γ-tubulin corresponding to the binding site of colchicine on β-tubulin was selected for investigation as to whether there was computational support for colchicine binding to this region. Combretastatin A-4 is known to bind to the same location as colchicine on β-tubulin so the same site on γ-tubulin was investigated for combretastatin A-4 binding as well.

Protein flexibility in docking

Proteins are structurally dynamic biomolecules in solution. This behavior is essential for recognition and binding to other molecules inside the cell. Although many attempts have been made to take into account the flexibility of the target with docking algorithms, there are many challenges to find a computationally efficient way of dealing with this flexibility. One solution is to use the relaxed complex scheme (62–65), which is an approach to accommodate receptor flexibility and allow for the use of accurate docking scoring techniques to implement a hybrid between static docking and MD simulations. We utilized this method by running a 24 ns MD simulation to explore the conformational space of the target, and then, we subsequently docked the ligand to our representative ensemble of 35 receptor conformations.

By extracted structures every 2 ps from our MD simulation, the final 16 ns of the trajectory that yielded a stable RMSD had 8000 resulting (snapshot) structures. To reduce this set to a manageable number of representative structures, we used average-linkage clustering, which has been documented to be effective in this task (66). This clustering method yields several metrics that help reveal the optimal number of clusters to be created and their population size. The metrics are the DBI (67) and the ratio of the SSR to the SST (66). A high-quality clustering is associated with a local minimum DBI value. As well, the percentage of variance explained by the data, shown by the SSR/SST curve, is expected to plateau for cluster counts exceeding the optimal number of clusters (66). Using these metrics, by varying the number of clusters, we looked for adequate clustering by selecting a local minimum for DBI and for a leveling off of the line for the percentage of variance explained by the data, shown by the SSR/SST curve, which is known as the ‘elbow criterion’ (66). Using this methodology, we reduced the 8000 MD trajectory structures into just 35 target structures (see Figure S7) to represent the conformational space of the binding site within γ-tubulin (see Figure S8).

Docking colchicine to γ-tubulin

Colchicine (denoted as colchicine derivative 1: see Table 1) was docked to the 35 representative structures of γ-tubulin created from our clustering method, which we refer to as targets 1 through 35. The targets were ordered based on the percentage of the trajectory that they represented, that is, target 1 represented the target representing the greatest proportion of the trajectory (at 33%). A binding mode resulting from the docking procedure was considered a hit if its presence in the results presented by AutoDock was more than or equal to 20% of the found poses. This indicates that it is a well-defined mode of binding and could be used for further analysis. Using a cutoff of docking cluster size of 20%, target 3 (which comprised 6.5% of the trajectory) gave the computational docking result with the greatest binding energy of −7.9 kcal/mol. However, colchicine docking to target 34 (comprising 0.1% of the trajectory) had the overall greatest binding energy of −8.7 kcal/mol but with a docking cluster size of only 6%. Both these targets had colchicine oriented deep within the binding pocket with the A-ring deepest in the pocket, similar to that of colchicine binding to β-tubulin (see Figure 5).

Figure 5.

 Colchicine bound to γ-tubulin in target 3 (white), and the relative position of colchicine bound to β-tubulin (purple). γ-Tubulin from target 3 is displayed (silver) and is aligned with β-tubulin to indicate the relative position of colchicine bound γ-tubulin versus β-tubulin. These computational findings suggest colchicine bound to γ-tubulin may be able to penetrate deeper than colchicine bound to β-tubulin. Both structures have the colchicine A-ring penetrating deepest into the protein.

Colchicine derivatives 2, 3a, 3b, and 3c (see Table 1) were also docked to the 35 representative structures to determine whether these derivatives had any potential to preferentially bind to γ-tubulin over β-tubulin. Results showed that all these derivatives had a best binding energy between −8.0 and −8.6 kcal/mol (although many of these cluster sizes were below 20%), which was comparable in binding energy to the results of colchicine binding to γ-tubulin. It is worth mentioning as a cautionary note that the standard error in binding energies given from our docking software, AutoDock, is empirically on the order of ±2.5 kcal/mol (46). The binding energies of colchicine and the corresponding derivatives that had the greatest binding energy are summarized in Table 2.

Table 2.   Best AutoDock binding energies compared for colchicine and colchicine derivatives docked to 35 representative structures of γ-tubulin. Best AutoDock binding energy of combretastatin A-4, which was docked to targets 3, 33, and 34, is also shown for comparison
Compoundγ-tubulin target structure (except first entry)Binding energy (kcal/mol)Equivalent KDm)Cluster size (%)
1β-Tubulin−6.81541
1Target 3−7.92.623
1Target 34−8.70.776
2Target 3−8.21.64
3aTarget 33−8.02.396
3aTarget 3−8.02.328
3aTarget 34−7.92.615
3bTarget 3−7.64.62
3bTarget 33−7.01116
3bTarget 34−8.31.48
3cTarget 34−8.60.833
4Target 33−5.96958

Docking colchicine to β-tubulin

As a positive control for our AutoDock docking studies, we computationally docked colchicine to β-tubulin and compared our computational results with the experimental result seen in PDB entry 1SA0 (45). Colchicine bound in the same orientation as in the crystal structure shown in Figure 6. Colchicine bound to β-tubulin with an AutoDock binding energy of −6.8 kcal/mol, which compares well with our experimental fluorescence quenching binding energy estimate of −6.90 ± 0.03 kcal/mol. This is also within the error of AutoDock’s energy calculating function and is comparable to other experimental binding energy estimates of −8.4 kcal/mol for colchicine bound to bovine tubulin (calculated from a KD of 1.1 μm) (59,60). This AutoDock binding energy result is summarized in Table 2.

Figure 6.

 Colchicine docked to β-tubulin (white) and colchicine from PDB entry 1SA0 (purple) in β-tubulin from PDB entry 1SA0. Distance between nitrogen in the two colchicines is 2.3 Å, indicating good agreement in the docked pose generated from AutoDock and the known pose given from crystal structure (PDB entry 1SA0).

Docking combretastatin A-4 to γ-tubulin

Combretastatin A-4 was docked to targets 3, 33, and 34, which were the targets found most amenable to colchicine and colchicine derivative binding. Computationally, combretastatin A-4 bound to γ-tubulin with a best AutoDock binding energy of −5.9 kcal/mol and a significant cluster size of 58% (see Table 2). Its computational binding pose is shown in Figure 7. This pose is deep inside the hypothesized colchicine binding pocket and shows promise for subsequent design of targeted derivatives that bind to this site.

Figure 7.

 Combretastatin A-4 bound to γ-tubulin in target 33. This pose is deep inside the γ-tubulin hypothesized colchicine binding site.

Computational results overview

Our AutoDock binding energies were comparable between colchicine binding to γ-tubulin and β-tubulin, with binding energies of −7.9 and −6.8 kcal/mol, respectively (see Table 2). It should be noted that the primary strength of AutoDock is in determining binding positions and not in calculating an accurate binding energy. The best docking trials of colchicine binding to γ-tubulin, in terms of binding energy, had colchicine consistently binding to γ-tubulin in a pose similar to that of colchicine binding to β-tubulin, with the A-rings of colchicine penetrating furthest into the protein (see Figure 5) (45). The electrostatic characteristics of these binding poses are shown in Figures 8 and 9. It is seen that colchicine binds to a predominantly electropositive region of β-tubulin, but our hypothesized binding location on γ-tubulin has both positive and negative electrostatic regions. This is not surprising, given β-tubulin and γ-tubulin have different electrostatic profiles, which may alter their preference to bind laterally or longitudinally (67). The different electrostatic profiles of the hypothesized γ-tubulin binding pocket suggest opportunities to design targeted colchicine and combretastatin derivatives to bind specifically to γ-tubulin, which is our intention for future work. The similarity of the binding poses and binding energies between our computational and experimental work for colchicine gives additional support to our computationally predicted binding poses. However, ultimately direct experimental validation of the binding poses of colchicine and combretastatin A-4 to γ-tubulin is required and is currently underway.

Figure 8.

 Colchicine binding pocket from β-tubulin (PDB entry 1SA0). Colors of protein convey electrostatics: positive (blue) and negative (red). This binding region is predominantly positive electrostatically.

Figure 9.

 Colchicine binding pocket hypothesized in γ-tubulin. Colors of protein convey electrostatics: positive (blue) and negative (red). This binding region contains both positive and negative regions.

Colchicine derivatives that were earlier investigated for their isoform-specific binding to β-tubulin (48) were analyzed here for their hitherto unknown potential to target γ-tubulin (see Table 2). The fact that the docking of colchicine derivatives yielded similar binding energies as colchicine itself neither indicates their potential for utilization as specific targets of γ-tubulin inhibition nor discounts their effectiveness in this regard. Further characterization of the binding site is needed for any definitive conclusions to be made as to whether an appropriate colchicine derivative can be designed with a sufficiently high affinity for the investigated γ-tubulin binding site.

Pharmacokinetic properties of ligands

The pharmacokinetic properties of colchicine, its derivatives, and combretastatin A-4 were analyzed to gauge their potential as chemotherapeutics for GBM (see Table 3). Colchicine and its derivatives had predicted partition coefficient (Log P) values in the range of 1.37–2.55, human jejunal permeability (Peff) in the range of 1.03–1.29 cm/second × 104, and solubility in intestinal fluid (FaSSIF) in the range of 0.071–0.559. Combretastatin A-4 is importantly smaller and has a decreased surface area (57.15 Å2) compared to that of colchicine and colchicine derivatives (73.86–95.98 Å2), making derivatives of combretastatin A-4 more amenable to passing the blood–brain barrier in a possible treatment for GBM.

Table 3.   Selected properties for compounds
Compound labelLog PPeff (cm/seconds × 104)FaSSIF (mg/mL)MW (g/mol)T_PSA (Å2)
  1. Log P, octanol-water partition coefficient; Peff, human jejunal effective permeability; FaSSIF, solubility in simulated fasted state intestinal fluid; MW, molecular weight; T_PSA, topological polar surface area.

  2. These results were generated by ADMET Predictor™ 5.5 software provided by Simulations Plus, Inc., Lancaster, CA, USA.

11.371.030.559399.4583.09
21.441.090.146476.5395.98
3a2.421.090.207401.4884.86
3b2.341.090.113429.5473.86
3c2.551.290.071443.5773.86
42.575.350.041316.3657.15

Discussion

As γ-tubulin is overexpressed in GBM and is required to nucleate microtubules that divide a cell in mitosis, it is a potential target for a novel chemotherapeutic agent. A potential inhibitor of γ-tubulin would ideally be highly specific for γ-tubulin and not have as great an affinity to the structurally similar α- and β-tubulin proteins. This is because γ-tubulin, being involved in nucleating the microtubule and not in the microtubule’s polymerization, is less abundant than the more prevalent α- and β-tubulins. Thus, any inhibitor that binds with similar affinity to both β- and γ-tubulin, for instance, might act with an effect similar to an inhibitor binding only to β-tubulin. Based on our fluorescence spectroscopy results that show colchicine having a similar affinity to both γ- and β-tubulin, this may indeed be the situation with colchicine. It is notable that colchicine did not pass clinical trials as a β-tubulin inhibitor because of toxicity. Therefore, a more targeted inhibitor of γ-tubulin might be able to have a therapeutic window at a lower dosage and thus provide a higher therapeutic index. It is also possible to speculate that inhibiting γ-tubulin results in extreme toxicity and that colchicine, in addition to inhibiting β-tubulin, may also be inhibiting γ-tubulin, leading to its extreme toxicity. Further studies are required to test the therapeutic value of a γ-tubulin inhibitor. However, even if inhibiting γ-tubulin results in a chemotherapeutic agent that is too toxic for a general therapy, it could possibly be used in the future as a chemotherapeutic agent in a targeted delivery system, for example, being conjugated with a protein or an antibody. In addition, a targeted inhibitor of γ-tubulin could provide an additional tool to discover the protein’s precise role in various cellular processes other than nucleation (11), such as affecting the dynamics of microtubules (68) or inactivation of the anaphase-promoting complex at the end of mitosis and G1 (21).

It is our hope that the experimental and computational results presented here showing that colchicine and combretastatin A-4 bind to γ-tubulin in vitro are a promising first step in the creation of an effective inhibitor of γ-tubulin that could be further developed as a chemotherapeutic agent for the treatment of GBM. Our computational work has confirmed that there is a potential binding pocket on γ-tubulin in the homologous location to the known binding site of colchicine on β-tubulin, and our analysis below suggests there is sufficient potential for rational drug design to develop an effective inhibitor of γ-tubulin at this in silico predicted binding location.

Binding energy calculations for an effective γ-tubulin inhibitor

It is estimated that γ-tubulin is 250–500 times less abundant than either α- or β-tubulin in cultured toad cells and frog eggs (3). Given this range, we calculate the binding affinity needed for an inhibitor to bind with equal probability to both γ-tubulin and β-tubulin. We use the Boltzmann distribution, given by

image

where ΔΔG is the change in binding energy of the inhibitor to γ-tubulin versus β-tubulin (i.e., ΔGγ − ΔGβ), Nβ is the number of β-tubulin proteins available to bind, and Nγ is the number of γ-tubulin proteins available to bind. Given γ-tubulin is 250–500 times less abundant than β-tubulin, ΔΔG = −3.4 to −3.8 kcal/mol. Thus, targeted inhibitors of γ-tubulin in order to be effective would ideally have increased binding affinity to γ-tubulin over β-tubulin by at least 3.8 kcal/mol. We have experience finding derivatives with comparable differences in the case of colchicine and noscapine binding to various isoforms of tubulin (70,71).

Colchicine binding site analysis

From the crystal structure found in PDB entry 1SA0, the colchicine binding site on β-tubulin is well characterized (45). It is surrounded by two β-sheets below-right (S8 and S9), a loop on the left (T7), alpha helix H7 below-left, and alpha helix H8 above-left (see Figure 10). Analysis of the residues immediately surrounding the site shows residues that are predominantly short and non-polar. Most notably, colchicine sits between T7 Leu 246 and H8 Leu 253.

Figure 10.

 Colchicine bound to β-tubulin from PDB entry 1SA0. Residues surrounding the binding site are shown.

The homologous γ-tubulin region contains a binding pocket also surrounded by predominantly short and non-polar residues and lies surrounded by β-sheets S8 and S9, alpha helix H8, and loop T7 (see Figure 11). The predominant exception is that Asn 251 and Tyr 248 are large residues, which in many of the conformations of γ-tubulin observed, blocked penetration of colchicine into the hypothesized binding pocket. The structure shown in Figure 11 has Asn 251 and Tyr 248 residues in the T7 loop oriented in a manner that allows an open pocket for colchicine to penetrate deep into the protein. Trp 22, Trp 446, and Trp 351 are located 1.9, 2.0, and 2.2 nm, respectively, from the approximate center of the hypothesized colchicine binding site in complex 3. Thus, while these tryptophans are not immediately adjacent to the hypothesized binding site, as shown in Figure 2, they are in similar locations to the tryptophans in β-tubulin and possibly similarly altered by colchicine binding, which would suggest we would obtain similar results of fluorescence quenching of colchicine interacting with β-tubulin and γ-tubulin, which is what was in fact observed (V. Semenchenko, R. Perez-Pineiro, D.S. Wishart, Unpublished data).

Figure 11.

In silico predicted colchicine binding site on γ-tubulin target 3. Asn 251 and Tyr 248 are large residues in T7, which in certain configurations of our MD simulation of γ-tubulin prevent a large enough space for colchicine to enter this site.

Recent work has characterized the crystal structure of the γ-tubulin complex protein 4 (GCP 4) and has found that it binds directly to γ-tubulin (72). The study suggests that the H7-H8 (or T7) loop, helix H8, and strand S9 interact with GCP2 and GCP3 (72,73), which are regions all associated with the hypothesized γ-tubulin binding pocket. Thus, the hypothesized colchicine binding pocket in γ-tubulin is in a position that might directly interfere with the binding of γ-tubulin to GCPs.

The recent work on combretastatin derivatives has yielded progress in finding water-soluble derivatives of combretastatin A-4 (61). As combretastatin derivatives have lower surface area than colchicine derivatives, they may also be promising targets to design an effective inhibitor of γ-tubulin that can penetrate the blood–brain barrier, and thus, investigation of this is underway. Our finding that combretastatin A-4 binds to γ-tubulin may also explain its antimitotic effects, although this as well must be further investigated.

Conclusion

Our combined experimental and computational results suggest that colchicine and combretastatin A-4 bind to γ-tubulin. These results provide the first evidence of a drug-like ligand interaction with γ-tubulin. Further work is still required to experimentally confirm the binding location of these compounds to γ-tubulin, so that focused virtual screening and derivatization can proceed to find a highly specific and selective inhibitor of γ-tubulin. Such an inhibitor may eventually yield therapeutic benefit to treat GBM. These studies might also explain an additional mode of action of these compounds, namely, their possible inhibition of γ-tubulin. A potential inhibitor could also be of benefit to study various biologic functions of γ-tubulin. Further study is underway to determine whether colchicine interferes with the creation of the γTuRC, which is employed to nucleate microtubules in metazoa (69).

Acknowledgments

All of the molecular dynamics simulations, docking, and computational work was done using SHARCNET, WESTGRID, and AICT (University of Alberta cluster) computational facilities. Funding for this work was obtained from CIHR, the Alberta Cancer Board, the National Institute for Nanotechnology (NRC-NINT), the Alberta Prion Research Institute (APRI) and the Allard Foundation. We thank Dr Chih-Yuan Tseng for his assistance in repairing missing residues in the crystal structures, Dr Travis Craddock for assistance with preparing Figure 1, and Mr Philip Winter for assistance in preparing Tables 1 and 3.

Conflict of interest

All authors confirm no competing financial or other interests.

[Correction added after online publication 27/03/2012: Reference numbers changed within the text]

Ancillary