In Silico Exploration of Novel Tubulin Inhibitors: A Combination of Docking and Molecular Dynamics Simulations, Pharmacophore Modeling, and Virtual Screening

Microtubules play a critical role in mitosis and cell division and are regarded as an excellent target for anticancer therapy. Although microtubule-targeting agents have been widely used in the clinical treatment of different human cancers, their clinical application in cancer therapy is limited by both intrinsic and acquired drug resistance and adverse toxicities. In a previous work, we synthesized compound 9IV-c, ((E)-2-(3,4-dimethoxystyryl)-6,7,8-trimethoxy-N-(3,4,5-trimethoxyphenyl)quinoline-4-amine) that showed potent activity against multiple human tumor cell lines, by targeting spindle formation and/or the microtubule network. Accordingly, in this study, to identify potent tubulin inhibitors, at first, molecular docking and molecular dynamics studies of compound 9IV-c were performed into the colchicine binding site of tubulin; then, a pharmacophore model of the 9IV-c-tubulin complex was generated. The pharmacophore model was then validated by Güner–Henry (GH) scoring methods and receiver operating characteristic (ROC) analysis. The IBScreen database was searched by using this pharmacophore model as a screening query. Finally, five retrieved compounds were selected for molecular docking studies. These efforts identified two compounds (b and c) as potent tubulin inhibitors. Investigation of pharmacokinetic properties of these compounds (b and c) and compound 9IV-c displayed that ligand b has better drug characteristics compared to the other two ligands.


Introduction
Microtubules made of αand β-tubulin heterodimers in eukaryotic cells and are vital components of the cytoskeleton which are involved in numerous cellular processes such as cell signaling, cell motility, and intracellular vesicle transport [1][2][3].
Microtubules form highly dynamics mitotic spindles, which are vital for the suitable orientation and segregation of chromosomes; disruption of this equilibrium will lead to cell cycle arrest or cell apoptosis [4][5][6].
Microtubule-targeting agents (MTAs) are divided into two classes including microtubule stabilizers such as taxanes which bind to the paclitaxel binding site that stabilize the tubulin polymer form and microtubule-destabilizing agents like vinca alkaloids and colchicine which bind to the colchicine and vinblastine binding site which inhibit tubulin polymerization into microtubules [6][7][8]. Despite the progress in the administration of microtubule targeting agents for the treatment of patients with cancer, currently, there are no FDA-approved tubulin inhibitors targeting the colchicine binding site [9]. This has encouraged medicinal chemists to design and discover the novel antimitotic agents that bind to the colchicine binding site for cancer therapy [10][11][12][13][14][15][16].
Pharmacophore is defined as an ensemble of steric and electronic features that is necessary to reach the optimal interactions of a ligand with the catalytic site of a protein and very well accepted in the medicinal chemistry laboratory [17][18][19].
In a previous work, we synthesized a new series of styrylquinolines that among them, compound 9IV-c, ((E)-2-(3,4-dimethoxystyryl)-6,7,8-trimethoxy-N-(3,4,5-trimethoxyphenyl)quinoline-4-amine) ( Figure 1) exhibited a potent activity against multiple human tumor cell lines, by targeting spindle formation and/or the microtubule network [4]. In this work, to identify potent tubulin inhibitors, at first, to explain the possible binding mode of compound 9IV-c and to evaluate the validity of docking results, molecular docking and molecular dynamics studies of this compound, respectively, were performed into the colchicine binding site of tubulin; then, a pharmacophore model of 9IV-c-tubulin complex was generated. It was then used as a screening query to obtain potent small molecules from an IBScreen database. Five hits were obtained in this step for a match to the pharmacophore model which was subjected to molecular docking studies. Finally, two compounds b and c with the lowest binding-free energies were selected as potent tubulin inhibitors. The pharmacokinetic properties of these compounds were studied using Swis-sADME software.

Molecular Modeling.
To study the manner of interactions between compound 9ІV-c and tubulin, this compound was docked into the colchicine binding site of tubulin (PDB ID: 4O2B) via MOE 2015.10 [4]. The studies exhibited that the compound 9ІV-c docked well in the colchicine-binding site with a binding free energy of -9.759 kcal/mol (Figures 2(a) and 2(b)). The hydrogen-bonding interactions (purple dashed line in Figure 2(a)) were seen between the methoxy group of A-ring of compound 9ІV-c with residue Asn 258β and also nitrogen atom of quinoline ring with Ala 180α and the methoxy group of D-ring with Asn 249β. Further, the hydrophobic interactions between quinolines with different amino acids such as Gln11α, Asn258β, Thr352β, Lys254β, Ser178α, Ala250β, Asn350β, Val351β, Lys352β, and Leu255β have been observed (Figure 2(b)). Figure 2(c) shows the binding mode of compound 9ІV-c with colchicine, the standard tubulin inhibitor, at the active site of colchicine in tubulin. The binding mode observed for 9ІV-c was very similar to that for colchicine in the cocrystallized tubulin structure. The trimethoxyphenyl group of 9ІV-c, which is an essential pharmacophore for tubulin inhibitors, was placed in the same position as the corresponding moiety of colchicine.

Molecular Dynamics
Simulation. Molecular dynamics (MD) simulation studies were performed to obtain a more exact ligand-receptor model in a state near to the natural situations [7]. For this goal, the docked structure of compound 9ІV-c was used as the initial structure for 100 ns MD simulations. After accomplishment of the simulation process, the complex was examined for root mean square deviation (RMSD), radius of gyration (Rg), and hydrogen bonds [20].
RMSD determines the stability of the constructions. The RMSD of backbone (Cα, C, and N) of apo-form and ligand-bond protein was shown in Figure 3. The average RMSD value of ligand-bond protein was 1:536 ± 0:212 Å, whereas the apo-form was with an average RMSD value of 1:776 ± 0:292 Å. In RMSD of apo-form, the change was larger than ligand-bond protein. It seemed that binding of the ligand with protein decreased the conformation flexibility of protein and the complex reached stability during the simulation.
Rg (readius of gyration) is an indicator of the protein structure density and measures the distance of region's parts from its center of gravity. As shown in Figure 4. Rg values of the apo-form and ligand-bond protein were 30:24 ± 0:125 and 30:089 ± 0:0:091 Å (mean ± SD), respectively. This confirms the stabilization and nonsignificant conformational changes in the structure of the ligand-bound protein [20].
Hydrogen bonds have a key role in the stability of the ligand-protein complex. This stability of the hydrogenbond network was because of the presence of the ligand in the protein binding site. The findings revealed that ligand made one to two hydrogen bonds with protein ( Figure 5). These hydrogen bonds increased binding affinity of the ligand with protein. The mean of intramolecular hydrogen bonds of the protein was 196:96 ± 11:96 (Figure 6.). These hydrogen bonds stabilized the secondary structure of protein.
The 2D representation of the interaction between compound 9ІV-c with the colchicine binding site after 100 ns simulation by LigPlot has been depicted in Figure 7.

Validation of the Pharmacophore Model.
To validate the reliability of the pharmacophore model constructed in this study, Güner-Henry (GH) scoring methods and receiver operating characteristic (ROC) analysis were used.
2.4.1. GH Scoring Method. The GH score method was used to evaluate the discriminative ability of the pharmacophore model in distinguishing active compounds from the inactive compounds from a testing set database including 20 known tubulin inhibitors and 600 inactive molecules [3,12,[22][23][24][25]. The GH analysis by computing parameters such as total hits (Ht), active hits (Ha), enrichment factor (E), and goodness    (Table 1). GH score was observed to be 0.69 for the model, representing that the pharmacophore model has a good ability to discriminate active compounds from inactive ones.

ROC
Studies. The pharmacophore model was used to receiver operating curve (ROC) analysis to assess its ability to suitably classify a list of compounds as active or inactive [26]. The ROC curve obtained for the validation showed an AUC value of 0.84 (Figure 9), indicating that the model differentiated the active compounds from the inactive ones efficiently (p < 0:001). The sensitivity and specificity of the model were 80% and 98.66%, respectively.

Virtual Screening and Identifying Potential Inhibitors.
The IBScreen database (containing 158210 compounds) was searched by using the 2D chemical structure of compound 9ІV-c (Figure 1.) as a query to find compounds with similar structures to this compound [27]. At this step, a total of 1257 compounds were obtained based on the most similarity to compound 9ІV-c; this set was then screened based on the created pharmacophore model. As a result, five ligands (a-e) were confirmed to fit with the pharmacophore model ( Table 2). These compounds were then docked into the colchicine binding site, and two ligands b and c were selected with the lowest binding-free energies and the best modes of interactions with the colchicine binding site.
Interactions of compounds b and c with tubulin protein are shown in Figures 10 and 11. It can be observed that compounds b and c have better free binding energies in comparison with compound 9ІV-c (-10.498 and -10.083 kcal/mol for b and c, respectively, and -9.759 for 9ІV-c). This subject suggests that these compounds (b and c) may interact with higher potencies than compound 9ІV-c.
As seen in Figure 11(a) and 11(b), molecular docking analysis of compound b displayed two hydrogen bond interactions between acidic substituents with residues of Lys 352β and the nitrogen atom of quinoline ring with residues of Thr 179α and a cation-π interaction between residue Leu 248β and one of the phenyl rings (Figure 11(a)). The sulfonami group of compound c formed two hydrogen bond interactions with residues Lys254β and Asn249β (Figure 11(b)). These compounds formed hydrophobic interactions with residues Leu255β, Lys248β, Asn258β, Ser178α and Asn101α, and Gln11α.
2.6. Pharmacokinetic Property Prediction. In this study, by using the SwissADME software, drug development involves the process of adsorption, distribution, metabolism, and excretion (ADME), for three ligands b, c, and 9ІV-c was performed [28]. The test results can be seen in Table 3.
The results of the tests in the gastrointestinal absorption parameter showed that ligand b (with the high result) are suitable for oral use. It was also found that ligands c and 9ІV-c with low results are not well used orally, but may be used as an injection.
The results of yes, in the cytochrome inhibitor parameter, will indicate that ligand can act as an inhibitor in the process of cytochrome metabolism, while no results will show that the ligand cannot act as an inhibitor in the process of cytochrome metabolism [29]. The test results show that three ligands were cytochrome inhibitors.
The higher bioavailability of the compound indicates that it can be better used orally [29]. Based on the results obtained from this parameter (Table 3), all three ligands have good bioavailability values, and among these, ligand b, which exhibited better results in docking studies, has better bioavailability.

Conclusion
In this research, molecular docking and molecular dynamics studies of compound 9IV-c (a potent inhibitor of tubulin polymerization found in previous work) were performed into the colchicine binding site of tubulin. The studies exhibited modes of interaction of compound 9ІV-c with tubulin including the formation of hydrogen bonds and hydrophobic interactions and also characterized the stability of this compound during a 100 ns simulation run. Then, a pharmacophore model of the 9IV-c-tubulin complex was generated. The created model included two hydrogen bond donor features (F1 & F2: Don), one aromatic feature (F3: Aro), and one hydrophobic feature (F4: Hyd). The pharmacophore model was then validated by Güner-Henry (GH) scoring methods and receiver operating characteristic (ROC) analysis. GH score was observed to be 0.69 for the model, representing that the pharmacophore model has a good ability to discriminate active compounds from inactive ones.
The IBScreen database (containing 158210 compounds) was searched by using the 2D chemical structure of compound 9ІV-c as a query to find compounds with similar structures to this compound and was then screened based on the generated pharmacophore model, and five ligands (a-e) verified to fit with the pharmacophore model. These compounds were then docked into the colchicine binding site, and two ligands b and c were selected with the lowest binding free energies and the best modes of interactions with the colchicine binding site. These ligands (ligands b and c)   Computational and Mathematical Methods in Medicine

Molecular
Modeling. The 2D structure of the compound was organized in ChemDraw Ultra 8.0 software, and the 3D structure was obtained using Hyperchem 7 software through molecular mechanic force filed preoptimization and then by AM1 semiempirical calculation. The X-ray crystal structure of tubulin (PDB ID: 4O2B) was downloaded from the PDB (Protein Data Bank). Further changes (addition of polar hydrogen or deletion of water molecules) were done by MOE software. Compounds were docked into the binding site of tubulin using MOE software. The top-score docking poses were selected for final ligand-target interaction analysis using the LigX module in MOE Software [30].

Molecular Dynamics
Simulation. Molecular dynamics (MD) simulation studies were performed to find the interaction between protein and ligand in atomic details in the real natural situation (aqueous solution at T = 37°C, P = 1 atm). Calculations were performed using NAMD 2.12 program (http://www.ks.uiuc.edu/Research/namd) with the CHARMM27 force field. To analyze results Visual Molecular Dynamics (VMD) (http://www.ks.uiuc.edu/Research/ VMD) was used. The force field parameters of ligand 9ІV-c were provided by SwissParam (http://swissparam.chr). The whole systems were immersed in the center of a TIP3 water box with dimensions 97:117 Å × 113:789 Å × 97:287 Å, using the VMD program. The systems were neutralized by adding sodium and chloride ions. The Particle-Mesh Ewald (PME) algorithm with a grid spacing of 1 Å and periodic boundary conditions was applied. A cut-off of 15 (Å) was used for the short-range Lennard-Jones interactions. Finally, MD simulations were performed with a time step of 2 fs for 100 ns. The trajectory of the system was stored at every 1 ps and analyzed by VMD analyzer tools. The system was complex and required significant computational. A system with a processor: AMD Ryzen Threadripper 1950X 16-Core, 3.40 GHz and installed memory (RAM) 32.0 GB, was used in this study [31].

Pharmacophore Model Generation.
The docked structure of compound 9IV-c in the colchicine binding site of tubulin was used for pharmacophore model generation. To create the most descriptive features of the colchicine binding site of tubulin, the Pharmacophore Query Editor of MOE was employed. These features are indicated as spheres that characterize the important interaction points of the ligand with the active site of tubulin [18,29]. The Güner-Henry (GH) scoring method quantifies the importance of the generated model by retrieving the active compounds from a database containing known active and inactive molecules [23,32]. To evaluate the discriminating ability of the generated pharmacophore model a testing set database including 20 known tubulin inhibitors with experimental activity and 600 inactive molecules was searched by the pharmacophore model as a 3D structural search query. The retrieved compounds are selected based on according to their RMSD value. Lower RMSD values show better quality of the matching of the molecule to the pharmacophore model. The GH score was calculated using the following formulae: [24].
Here, D is the number of compounds in the database, A is the number of active compounds, H t is the number of hits retrieved, H a is the number of actives in hit list, EF is the enrichment factor, and GH is the Güner-Henry score. The GH score ranging from 0.6 to 1 would indicate an optimal pharmacophore model.

Analysis Using ROC.
To evaluate the pharmacophore model, receiver operating characteristic (ROC) curve analysis was also performed using MedCalc statistical software (http://www.medcalc.org). In ROC analysis, the ability of the obtained pharmacophore model was indicated with the area under the curve (AUC) to distinguish a list of compounds as active or inactive compounds in terms of two parameters, sensitivity and specificity.

4.5.
Screening of the Database Based on Pharmacophore Models. The IBScreen database containing 158210 compounds was used. At first, using ChemFinder software the 2D chemical structure (SMILES file format) of compound 9ІV-c (Figure 1.) was used as a query in the IBScreen database to find compounds with similar structures to this compound. The applied filter gave a total of

Data Availability
Data are available upon a reasonable request from the corresponding author.

Conflicts of Interest
The authors declare that they have no conflicts of interest.