High-affinity ligands of the colchicine domain in tubulin based on a structure-guided design

Microtubule-targeting agents that bind at the colchicine-site of tubulin are of particular interest in antitumoral therapy due to their dual mechanism of action as antimitotics and vascular disrupting agents. Cyclohexanediones derivatives have been described as a new family of colchicine-domain binders with an association constant to tubulin similar to that of colchicine. Here, the high-resolution structures of tubulin in complex with cyclohexanediones TUB015 and TUB075 were solved by X-ray crystallography. A detailed analysis of the tubulin-TUB075 interaction by means of computational affinity maps allowed the identification of two additional regions at the binding site that were addressed with the design and synthesis of a new series of cyclohexanediones with a distal 2-substituted benzofurane. These new compounds showed potent antiproliferative activity with IC50 values in the nM range, arrested cell cycle progression at the G2/M phase and induced apoptosis at sub μM concentrations. Moreover, they caused the destruction of a preformed vascular network in vitro and inhibited the migration of endothelial cells at non-toxic concentrations. Finally, these compounds displayed high affinity for tubulin as substantiated by a Kb value of 2.87 × 108 M−1 which, to the best of our knowledge, represents the highest binding constant measured to date for a colchicine-domain ligand.


Results and Discussion
Crystal structure of the tubulin-TUB015 and tubulin-TUB075 complexes. To obtain a detailed description of how TUB015 (7) 31 and TUB075 (8) 32 interact with tubulin and to establish a firm basis for the design of TUB-derivatives with improved affinities, we solved the structures of tubulin in complex with TUB015 and TUB075 ( Fig. 2A) by X-ray crystallography. To this end, we soaked the ligands into crystals formed by a protein complex composed of two bovine brain αβ-tubulin heterodimers, the rat stathmin-like protein RB3 and chicken tubulin tyrosine ligase (T 2 R-TTL) 24,25 . Using this approach, both the tubulin-TUB075 and tubulin-TUB015 complex structures were determined to 2.15 and 2.1 Å resolution, respectively (Figs S1 and 2B; Table S1). Comparison of the overall structure of tubulin in both the T 2 R-TTL-TUB075 and the T 2 R-TTL-TUB015 complexes with the one obtained in the absence of any ligand 25 revealed a good superimposition (rmsd T2R-TTL-TUB075 of 0.37 Å over 1925 C α -atoms; rmsd T2R-TTL-TUB015 of 0.34 Å over 1921 C α -atoms), suggesting that binding of both ligands does not affect the overall conformation of tubulin. Both ligands bind to the colchicine domain of tubulin, which is formed by residues of strands S8, S9 and S10, loop T7, and helices H7 and H8 of β-tubulin, and loop T5 of α-tubulin (Fig. 2C). The differences between TUBs and colchicine binding will be detailed in the next paragraphs. In both the tubulin-TUB075 and tubulin-TUB015 complexes, the 5-phenyl moiety of the cyclohexanedione ring (ring A in Fig. 2A) is buried in the pocket shaped by the side chains of βIle4, βTyr52, βGln136, βAsn167, βPhe169, βGlu200, βTyr202, βVal238, βThr239, βLeu242, and βLeu252 (Fig. 2C,D). Furthermore, the cyclohexanedione ring (ring B in Fig. 2A) is stacked between βTyr202 and βLeu255, and is further stabilized by two polar contacts, as follows. In both molecules one direct hydrogen bond is formed between the C3-carbonyl of the ligand and the side chain of βGlu200. A second hydrogen bond, water mediated, is formed by the C1-carbonyl to the main chain amide of βCys241. The water molecule is well defined in the TUB015 complex, and partially occupied in the tubulin-TUB075 complex. In agreement with the suggestion recently put forward by Guzmán-OCampo et al. 35 that, in the tubulin-nocodazole complex, the side-chain of βGlu200 is likely to exist as a carboxylic acid rather than as a carboxylate, we considered the C-3 carbonyl oxygen to be engaged in a hydrogen bonding interaction with this moiety, given the suitable distance and angle geometries. The 2-ethoxy/2-methoxy aniline moiety (ring D in Fig. 2A) of both TUB compounds is solely stabilized by hydrophobic contacts to βLeu248, βMet259, βAsn258, βAla317, βLys352, βala354, and to the T5 loop residue αThr179. This binding mode is maintained in both the occupied binding sites in our crystal system (TUB075: RMSD chain D onto chain B 0.37 Å over 420 C α -atoms; TUB015: RMSD chain D onto chain B 0.37 Å over 420 Cα-atoms).
Compared to the crystal structure of the tubulin-colchicine complex (PDB ID 4O2B; Fig. 2E) 26 , there are certain differences that need to be highlighted. Probably the most significant difference is the change in the αT5 loop, which is flipped out in the colchicine structure to avoid clashes with the acetamide moiety of the colchicine molecule while in the TUB complexes this loop is closing the binding site. Also the side chain of βLeu255 is blocking the access to the most internal subsite in the colchicine complex, while the binding of the TUB compounds causes a flip in this side chain to access this internal subpocket. Altogether, little overlap is observed among the tubulin-bound TUB and colchicine molecules (Fig. 2E).
According to the structural interaction fingerprints (SIFTs) reported by Massaroti et al. 36 , the colchicine domain can be divided into 3 zones: zone 1, located at the α-subunit interface; zone 2, that is the main zone located in the β-subunit which accommodates most of the structure of the ligands, and zone 3, an accessory pocket buried deeper in the β-subunit. For binding, TUB015 and TUB075 explore zones 2 and 3, as reported for other ligands, particularly TN-16, but the methoxy/ethoxy substituent points towards zone 1.
Notably, the binding of TUB015 and TUB075 is in full agreement with the structure-activity relationships (SAR) previously established for this family of compounds 31,32 which are graphically summarized in Supplementary Fig. S2.

Design of new ligands based on affinity maps.
To better analyze the binding of TUBs to tubulin, affinity maps were generated using our in-house computational tool cGRILL 33,34 , which relies on GRID functions as developed by Goodford et al. 37 . By applying different chemical probes, insight can be gained about both favorable binding interactions and forbidden areas. Using the lipophilic probe (CH 3 + no possibility of forming hydrogen bonds, akin to GRID's DRY probe) on the TUB075-tubulin complex, two new unexplored regions were identified in the proximity of the aniline moiety (ring D) as suitable for binding a short hydrocarbon chain (Fig. 3A). Region 1 covers the ethoxy substituent at position 2 and the adjacent position 3 of the aromatic ring. This zone could be occupied by a ring fused to positions 2 and 3 of the phenyl ring resulting in a 2-methylbenzofurane, as depicted in Fig. 3B. This substitution, besides keeping the existing favorable interactions of TUB075, should cover region 1 identified by cGRILL and fix the conformation of the ethoxy substituent of TUB075. The second region, with a tunnel shape, extends from the ethoxy substituent towards the αβ-tubulin interface, suggesting the 2-methyl substituent of the benzofurane could be enlarged to occupy this tunnel.
Synthesis of new derivatives. The synthesis of the 2-methylbenzofurane derivative was addressed as depicted in Fig. 4A. Intramolecular cyclization of the propyne derivative 10 38 at high temperature 39 afforded the nitro derivative 11 40 that was hydrogenated to provide the amino derivative 12 41 . Condensation of this amine with 2-acetyl-5-phenylcyclohexane-1,3-dione (13) 31 in refluxing toluene afforded the target compound 14 in 74% yield.
To include an additional functionalization at position 2 of the benzofurane that should allow elongation of the TUB structures towards the tunnel identified as region 2 in the cGRILL maps, new derivatives were envisioned where the methyl group of 14 was replaced by an ester, an amide or an alcohol. The 7-nitrobenzofuran-2-methyl ester (15) 42 (Fig. 4B) was synthesized following a described procedure 42 and further hydrogenated in the presence   43 Table 1 and expressed as IC 50 (50% inhibitory concentration) values, that is, the concentrations at which the compounds reduce cell proliferation by 50%. As reference compounds, colchicine (1) and CA-4P (3) are included. Our previous hit compound 8 has also been included for comparative purposes.
Compound 14 provided similar or lower IC 50 values than the reference compound 8 suggesting that the incorporation of the fused ring is beneficial for the antiproliferative activity. Incorporation of a methyl ester or methylamide functionality at position 2 of the benzofurane (compounds 20 and 21) had a negative impact on the cytostatic activity, while compound 22 with a hydroxymethyl group provided IC 50 values comparable to those of the methyl derivative 14. Interestingly further substitution of the hydroxymethyl group as in compounds 27 and 28 resulted in IC 50 values ranging from 0.009 to 0.051 μM. Thus, compounds 27 and 28 emerged as the most potent antiproliferative compounds among cyclohexanedione derivatives.
One of the most relevant causes of tumor resistance to chemotherapy is the overexpression of membrane pumps which remove the chemotherapeutic agents from the treated cells 45 . To evaluate the effect of the new chemotype on resistant cells, the most representative compounds were tested against A2780 and A2780AD (MDR overexpressing P-glycoprotein) human ovarian carcinomas and the results are collected in Table 2. All the compounds inhibit cell proliferation in the sub μM range against both non-resistant and resistant P-glycoprotein overexpressing, multidrug ovarian carcinoma cell lines, and no significant differences between both cell lines (R/S ratio close to 1) were found. Thus, P-glycoprotein overexpression does not seem to represent a problem for this chemotype.
Tubulin binding. Besides the antiproliferative activity, it was crucial to determine the capacity of the new compounds to bind tubulin. Binding affinities for the colchicine site in tubulin were determined by competition experiments with (R)-(+)-ethyl 5-amino 2-methyl-1,2-dihydro-3-phenylpyrido[3,4-b]pyrazin-7-yl carbamate (R-PT), as in previous examples 31,46 . However, under these conditions, compound 27 provided a strong association with tubulin that prevented the determination of the binding constant by displacement of R-PT 47 . Therefore a new reference ligand with a known higher binding affinity had to be used. Compound 22 with a Kb = 9.1 × 10 7 M −1 was selected as the reference compound for the competition assay with 27 instead of R-PT (data collected in Table 1).
Given the similar spectroscopic properties of 22 and 27, the competition experiment for determination of the binding constant of 27 was based on a centrifugation assay. Both compound 27 and the reference compound 22 Comp.  Table 1. Anti-proliferative activity of the benzofurane derivatives in endothelial and tumor cell lines and binding constants for αβ-tubulin. a IC 50 (50% inhibitory concentration) is given as the mean ± SD of three independent experiments. b Mean value of three experiments ± SD. c At 37 °C.
were incubated in molar excess with respect to tubulin to achieve a saturated equilibrium state. By centrifugation tubulin co-sediments with the bound ligands were discharged and the concentration of the free compounds in solution could be determined by reverse-phase HPLC. Using equations based on the law of mass action (see Methods), the binding affinity of compound 27 was calculated (Kb = 287 ± 106 × 10 6 M −1 ).

Molecular modelling studies. To gain insight into the binding determinants conferring on compound 27
high affinity for tubulin, molecular modelling studies were performed using the coordinates of an α 2 β 2 tubulin dimer extracted from the TUB075-(α 1 β 1 :α 2 β 2 ) tubulin complex solved by x-ray crystallography. The complex obtained with the best docking pose was then relaxed by means of a molecular dynamics (MD) simulation lasting 50 ns, during which a soft harmonic restraint (5.0 kcal·mol −1 Å −2 ) on the protein Cα atoms was used (Fig. 5A), followed by energy minimization. The resulting complex showed that the central core of 27 (regions A, B and C, Fig. 2A) shares the same pocket and orientation as TUB075, thus keeping hydrogen bonding interactions with Glu200 and Val238. Moreover, the 2-methylmethoxy group of 27, which lies at the interface between the α and β subunits, is found pointing towards the T5 loop of α-tubulin.
A detailed study of the contribution of individual protein residues to the free energy of binding along the MD trajectory (which was generated to improve the sampling of the ligand around the crystallographically determined positions of the protein) was carried out by means of program MM-ISMSA 48 . This approach takes into account not only van der Waals and electrostatic interactions but also ligand and receptor desolvation. As shown in Fig. 5B, the most important residue for the binding of 27 is Leu255, followed, in decreasing order of importance, by Leu248, Asn167 and Glu200.

Inhibition of cell cycle progression.
Tubulin-binding agents interfere with mitotic spindle formation during cell division. This results in the inhibition of cell proliferation and/or induction of apoptosis. Thus, we  Table 2. Anti-proliferative activity of the benzofurane derivatives in A2780 and A2780 AD cell lines. a IC 50 (50% inhibition of cell proliferation) values in ovarian carcinomas. IC 50 values are given as the mean ± standard error of three independent experiments. b Resistance index (the relative resistance of A2780AD cell line, obtained by dividing the IC 50 of the resistant cell line by that of the parental A2780 cell line).

Inhibition of endothelial cell migration.
Proper tubulin organization is not only required for cell division but also for cell motility and migration. Therefore, we evaluated the capacity of the compounds to inhibit the migration of HMEC-1 endothelial cells. To this end, we used the 96-well IncuCyte ® scratch wound assay.
After the creation of a cell-free zone (wound) in a confluent cell monolayer, compounds were added, and relative wound density was measured every minute and visualized in time-course plots. PI was added to all wells to visualize toxicity over time, i.e. only dead cells take up this cell-impermeable dye. As shown in Fig. 7, compounds 8, 27 and 28 inhibited wound closure at concentrations as low as 0.5, 0.06 and 0.06 µM, respectively. No increased uptake of PI was noted at these concentrations, indicating a specific, non-toxic effect.

Conclusions
We have solved the structures of tubulin in complex with two cyclohexanedione derivatives TUB015 and TUB075 by X-ray crystallography at 2.15 and 2.1 Å resolution. Most interactions between the ligand and the protein are hydrophobic in nature, as expected, although there are also some polar interactions, particularly between the C-3 carbonyl of the TUB ligands and the side chain of βGlu200 and a second water mediated hydrogen bond between the C1-carbonyl and the main chain amide of βCys241. Little overlap is observed among the tubulin-bound TUB and the colchicine molecules since TUB075 and TUB015 occupied zones 2 and 3 of the colchicine domain, as defined by Massarotti et al. 36 , while colchicine makes use of the most external zone, designated zone 1. Analysis of the binding site of TUB075 with the cGRILL program allowed the identification of two additional regions around fragment D of TUB075 that could be exploited to gain additional binding affinity. To occupy these two regions, a new series of cyclohexanediones with a 2-substituted benzofurane ring as fragment D were envisaged. Interestingly, from the new series, compound 27 with a 2-methoxymethyl group and compound 28 with a 2-(3-hydroxypropoxy)methyl substituent at the benzofurane, showed antiproliferative activity against tumor and endothelial cells in the nanomolar range (IC 50 = 8-31 nM), keeping full activity against P-glycoprotein overexpressing cells (A2780 AD). Additional studies performed to determine the mechanism of action of 27 and 28 revealed that they caused cell cycle arrest in G 2 /M at doses lower than those of the previous hit 8 (TUB075). In addition, they were able to induce apoptosis as shown in the caspase-3 assays. Compounds 27 and 28 further destroyed a preformed vascular network and inhibited the migration of endothelial cells at non-toxic concentrations. Very notably, compounds 27 and 28 demonstrated an exceptionally high affinity for tubulin in the binding assays, so that the K b of compound 27 (2.87 × 10 8 M −1 ), represents, to the best of our knowledge, the highest affinity constant described in the literature for a colchicine-site ligand.  13 C spectra were obtained using standard conditions. 2D inverse proton detected heteronuclear one-bond shift correlation spectra were obtained using the Pulsed Field Gradient HSQC pulse sequence. Data were collected in a 2048 × 512 matrix with a spectral width of 3460 Hz in the proton domain and 22500 Hz in the carbon domain, and processed in a 2048 × 1024 matrix. The experiment was optimized for one bond heteronuclear coupling constant of 150 Hz. 2D inverse proton detected heteronuclear long-range shift correlation spectra were obtained using the Pulsed Field Gradient HMBC pulse sequence. The HMBC experiment was acquired in the same conditions that HSQC experiment and optimized for long range coupling constants of 7 Hz. Separations on silica gel were performed by preparative centrifugal circular thin-layer chromatography (CCTLC) on a Chromatotron R (Kiesegel 60 PF 254 gipshaltig (Merck)), with layer thickness of 1 and 2 mm and flow rate of 4 or 8 mL/min, respectively. Flash column chromatography was performed in a Biotage Horizon instrument.

Methods
Microwave reactions were performed using the Biotage Initiator 2.0 single-mode cavity instrument from Biotage (Uppsala). Experiments were carried out in sealed microwave process vials utilizing the standard absorbance level (400 W maximum power). The temperature was measured with an IR sensor on the outside of the reaction vessel.
General procedure for the reaction of 2-acetylcyclohexane-1,3-dione with anilines (General procedure A). A solution of 2-acetylcyclohexane-1,3-dione (13) (1.0 mmol) and the appropriate aniline (1.5 mmol) in toluene (10 mL) was placed in an Ace pressure tube 31 . Then, 4 Å molecular sieves were added, the vessel was sealed and heated at 110 °C overnight. After cooling, the solvent was evaporated and the residue was purified by flash chromatography or by CCTLC in the Chromatotron.
General procedure for the synthesis of 7-aminobenzofuranes by reduction of 7-nitrobezofuranes anilines (General procedure B). The corresponding 7-nitrobenzofurane (1 eq) was dissolved in ethyl acetate (12 mL) in a pressure vessel and then 5% Pt/S (catalytic amount) was added. The mixture was hydrogenated at 30 psi for 1-8 h at room Figure 7. Inhibition of HMEC-1 migration. The 96-well IncuCyte ® scratch wound assay was used to create a cell-free zone (wound) in a confluent cell monolayer. Next, compounds were added and the relative wound density was measured every minute and visualized in time-course plots (left panels). PI is added to all wells to visualize toxicity over time, i.e. only dead cells take up this cell-impermeable dye (right panels). Average ± SEM (n = 3) are shown.

2-
Determination of binding constants. Proteins and ligands. Calf brain tubulin was purified as described 50 . (R)-(+)-ethyl 5-amino 2-methyl-1,2-dihydro-3-phenylpyrido[3,4-b]pyrazin-7-yl carbamate (R-PT) 51  Determination of binding constants. The binding constant of R-PT for dimeric tubulin was determined using the competition method in 10 mM sodium phosphate, 0.1 mM GTP pH 7.0 at 25 °C. To do so 0.2 µM of R-PT was incubated with increasing amounts of tubulin up to 10 µM and vice versa, 0.2 µM of tubulin was incubated with increasing amounts of R-PT up to 10 µM, the fluorescence emission spectra (excitation 374 nm) of the samples (5 nm excitation and emission slits) were determined using a Jobin-Ybon SPEX Fluoromax-2 (HORIBA, Ltd. Kyoto, Japan). Using these spectra it is possible to calculate the free and the bound R-PT concentration for each sample and thus to determine the binding constant of R-PT for tubulin.
Once K b of R-PT is determined (5.1 × 10 6 M −1 ) this compound could be used as a reference ligand as described in ref. 43  Samples were transferred to a fixed-angle TLA-100 rotor (Beckman-Coulter) and centrifuged for 2 h at 100.000 rpm at 25 °C. Samples were divided, splitting the upper 100 μL in which only free compound is found from the lower, in which soluble tubulin remains with bound compounds. Appropriate controls were included to confirm the solubility of the compounds under the assay conditions. Fractions were extracted with 3 rounds of 1:1 volume of dichloromethane, concentrated by evaporation in SpeedVac and resuspended in 100 μL of acetonitrile.
To determine the concentration of the compounds, reverse-phase chromatography was performed using a SUPELCOSIL ™ LC-18-DB HPLC Column coupled to an Agilent 1100 series HPLC system. Analytic separation of the compounds was performed in gradient, using as mobile phase CH 3  The concentration of the compounds in the upper part of the tube was considered the free concentration in the experiment (Cf), the difference in the concentration of the compounds in the lower part of the tube with these in the upper part of the tube was considered the concentration of compound co-sedimenting with tubulin and thus the bound concentration in the experiment (Cb). Assuming the colchicine binding site to have a 1:1 stoichiometry: where Tf is the concentration of free tubulin, Cf and Cb are molar concentrations for free and tubulin-bound compound. Using simultaneous mass-action equation, we obtain the following expression 47 : X-Ray Crystallography. Crystallization, Data Collection, and Structure Solution. Crystals of T 2 R-TTL were generated as described [24][25][26] . Suitable T 2 R-TTL crystals were exchanged into reservoir solutions containing 0.5 mM TUB-015 or TUB-075 and soaked overnight. Soaked crystals were flash cooled in liquid nitrogen following a brief transfer into cryo solution containing 20% glycerol. T 2 R-TTL-TUB015 and T 2 R-TTL-TUB075 data were collected at beamline X06SA at the Swiss Light Source (Paul Scherrer Institut, Villigen, Switzerland). Images were indexed and processed using XDS 52 . Structure solution using the difference Fourier method and refinement were performed using PHENIX 53 . Model building was carried out iteratively using the Coot software 54 . Data collection and refinement statistics for T 2 R-TTL-TUB015 and T 2 R-TTL-TUB075 are given in Supplemental  Table S1.
Computational Methods. Affinity maps calculation. Binding pocket analysis of the tubulin-TUB075 was carried out with cGRILL 33,34,55 , a computational tool formally equivalent to Goodford´s program GRID 37 . Hydrogen atoms, atom point charges and radii for all atoms in the complexes were calculated by submission to the H++ server (http://biophysics.cs.vt.edu) in order to obtain their pqr format files. Grid center was defined by selecting the corresponding ligand and a cubic box of 50 × 50 × 50 points and a grid spacing of 0.5 Å spacing was set for the calculations. cGRILL evaluates, at each grid point, the interaction energy between the whole receptor and five different probes combining van der Waals (Lennard-Jones potential), electrostatic (Coulombic), and geometry-based hydrogen bonding 56 terms. The five calculated affinity map, namely lipophilic (CH3), hydrogen bond acceptor (=O), hydrogen bond donor (NH4 + ), mixed hydrogen bond donor-aceptor (OH) and hydrophobic-like (hydrophobic), were visualized and analysed using the PyMOL plugin provided with the software.
Docking of compound 27. A three-dimensional cubic grid, consisting of 65 × 65 × 65 points with a spacing of 0.375 Å, was defined at the colchicine binding site in the α 2 β 2 tubulin dimer from the ligand-free T 2 R-TTL-TUB075 complex. In agreement with recent proposal 35 , the side chain carboxylic group of βGlu200 was considered to be protonated. Electrostatic, desolvation, and affinity maps for the atom types present in 27 were calculated using AutoGrid 4.2.6 and then the Lamarckian genetic algorithm implemented in the automated docking program AutoDock4 57 . Intra-and intermolecular energy evaluation of each configuration allowed the selection of the 10 best-scoring solutions, which were virtually superimposable. The resulting [(α 2 :GTP:Mg 2+ )-(β 2 :27)] complex was then immersed in an octahedral solvent box extending 12 Å away from any protein or ligand atom and neutralized by addition of 31 sodium ions. All hydrogens and water molecules were first reoriented in the electric field of the complex and then 27, all protein residues, water molecules and counterions were subjected to 2 000 steps of steepest descent followed by 50 000 steps of conjugate gradient energy minimization. To improve the sampling, the resulting geometry-optimized coordinate set was used as input for a C α -restrained (5.0 kcal·mol -1 Å -2 ) molecular dynamics (MD) simulation at 300 K and 1 atm using the GPU implementation of the pmemd_ cuda engine in AMBER 14 58 . The standard ff14SB force field parameter set 59 was used, which included updates for bioorganic phosphates and polyphosphates 60 , and the application of SHAKE to all bonds allowed an integration time step of 2 fs to be employed. The cutoff distance for the nonbonded interactions was fixed at 9 Å and the list of nonbonded pairs was updated every 25 steps. Periodic boundary conditions were applied and electrostatic interactions were represented using the smooth particle mesh Ewald method 61 with a grid spacing of 1 Å. The coupling constants for the temperature and pressure baths were 1.0 and 0.2 ps, respectively. 500 sets of coordinates from the resulting MD trajectory were analyzed by means of the cpptraj module in AMBER 62 and our in-house MM-ISMSA program 48 .

PDB ID Codes.
Coordinates have been deposited to the RCSB PDB (www.rcsb.org) under accession numbers 6FKJ (T2R-TTL-TUB075) and 6FKL (T2R-TTL-TUB015). Authors will release the atomic coordinates and experimental data upon article publication.