VEGFR-2 protein preparations

The human VEGFR2 kinase domain crystal structure comprises 359 amino acids and a novel pyrrolopyrimidine inhibitor (1-2-fluoro-4-[(5-methyl-5H-pyrrolo[3,2-d]pyrimidin-4-yl)oxy]phenyl-3-[3-(trifluoromethyl)phenyl]urea) 42Q (PDB ID: 3VHE) [18]. It was obtained from the "Protein Data Bank" (https://www.rcsb.org/) and cleaned with the "Biovia Discovery Studio Visualizer" (https://disCover.3ds.com/discovery-studio-visualizer-download). Swiss-pdb viewer 4.1.0 and UCSF Chimera v1.15 were used to optimizing the retrieved protein. Intrinsic water molecules were first removed from the protein’s pocket area and the co-crystallized ligand. Polar hydrogens were inserted, which was proceeded by energy reduction utilizing conjugate gradient and the steepest method. Gasteiger charge potentials were introduced after minimization. All of the protein’s heteroatoms are also eliminated, which may interfere with the receptor’s binding site and cause problems with protein-ligand interaction. The macromolecule’s output structure is then stored in pdbqt format.

Ligand preparation

The structure of ABZ and reference standard 42Q (1-{2-fluoro-4-[(5-methyl-5H-pyrrolo[3,2-d]pyrimidin-4-yl)oxy]phenyl}-3-[3-(trifluoromethyl)phenyl]urea) was downloaded from the public open-access database “PubChem” (https://pubchem.ncbi.nlm.nih.gov/) in 3D-SDF format. The structures of molecules were then translated into PDB format using the "Open Babel GUI" graphic user interface (http://openbabel.org/wiki/MainPage) [19]. AM1 semi-empirical force fields were used to optimize ABZ and 42Q, stored in mol2 file format utilizing the ChemMOP module of the ChemDes webserver (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Molecular structures of Albendazole and standard ligand after geometry optimization.

(a) Albendazole (ABZ)- (methyl N-(6-propylsulfanyl-1H-benzimidazol-2-yl)carbamate). (b) 42Q-1-{2-Fluoro-4-[(5-Methyl-5h-Pyrrolo[3,2-D]pyrimidin-4-Yl)oxy]phenyl}-3-[3-(Trifluoromethyl)phenyl]urea.

https://doi.org/10.1371/journal.pone.0287198.g001

Predictions of molecular targets

By entering smiles of desired drugs, the SwissTargetPrediction service was used to investigate specific targets gives an idea about their physiological side effects and cross reactivity for anticancer efficacy. A three—dimensional & two-dimensional similarity index with known ligands is used to determine the probable targets of the provided chemical [20].

Prediction of biological activity spectra for substances

Prediction of Activity Spectra for Substances (PASS) is a free web service that predicts the biological activities of drug-like chemicals (http://www.way2drug.com/passonline). Based on multilevel neighbourhoods of atom (MNA) descriptors and a modified Bayesian algorithm containing more than 300,000 organic compounds with more than 6,825 distinct biological activities, it also predicts a structure-activity relationship (SAR). Thus, PASS may be applied to improve and well-target chemical production and biological testing. The anticancer activity of the ABZ and standard drug 42Q was anticipated, and the findings are shown in Table 1. A substance’s estimated activity is forecasted as probable activity (Pa) and probable inactivity (Pi). Only compounds with a Pa greater than Pi are considered suitable for a certain therapeutic activity [21].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. PASS prediction coefficients of ABZ and 42Q against tumor cell lines and other protein receptors related to cancer progression.

https://doi.org/10.1371/journal.pone.0287198.t001

Molecular docking

This research aims to shed insight on the binding properties of 3VHE protein using geometrically optimized chemicals like ABZ and standard 42Q. Before being employed in molecular interaction studies, proteins were checked for missing side-chain residues using the open Molecular Mechanics (MM) simulation tool (https://openmm.org/). Autodock v 4.2.6 was used for molecular docking investigations. The binding cavity was established using the co-crystallized X-ray structure of VEGFR-2 protein from the RCSB PDB. Within three spaces of the co-crystallized ligand, the residue locations were computed. The energy was minimized using the steepest descent and conjugate gradient techniques after cavity selection. Finally, the receptor and target molecules were stored in pdbqt format after combining the nonpolar hydrogens. Grid boxes with a spacing of 0.3 were made. To get the minimum free energy of binding (G), docking studies of the protein-ligand complex were performed utilizing the Lamarckian Genetic Algorithm (LGA). Three replicates of molecular modelling experiments were carried out, with default parameters like number of solutions (50 in each), number of evaluations (2500000), maximum number of generations (2700) etc. After docking, the Root Mean Square Deviation clustering maps were created by re-clustering with clustering tolerances of 0.25, 0.5, and 1 in order to obtain the best cluster with the lowest energy score and a high number of populations, respectively.

Molecular dynamics simulation

The Desmond software Vs 2020.1 from Schrödinger was used to do MD simulations on docked complexes for ABZ and 42Q with protein (PDB ID: 3VHE). Triplicate sampling was performed with the same condition for each MD run to achieve better results. This system employed the OPLS-2005 force field [22–24] and an explicit solvent model with simple point charger (SPC) water molecules [25]. To neutralise the charge, Na+ ions were added. NaCl solution (0.15M) was used to simulate the physiological environment. To retrain over the protein-ligand complex, the system was first equilibrated using NVT ensemble for 200 ps. This system was then exposed to a 12 ps run for equilibration and minimization using the NPT ensemble. The NPT ensemble was built using the Nose-Hoover chain coupling method [26]. All simulations were performed at a temperature of 27°C, with a relaxation time of 1.0 ps and a pressure of 1 bar. A time step of 2 fs was chosen. For pressure regulation, the Martyna-Tuckerman–Klein chain coupling scheme [27] barostat technique with a relaxation duration of 2 ps was utilised. The particle mesh Ewald method [28] was used to determine long-range electrostatic interactions, with the Coulomb interaction radius set at 9 cm. For each trajectory, the bonded forces were estimated using the RESPA integrator with a time step of 2 fs. To measure the stability of MD simulations, the root mean square deviation (RMSD), radius of gyration (Rg), root mean square fluctuation (RMSF), solvent accessible surface area (SASA), and a number of hydrogen bonds were computed. RMSD clustering of individual trajectories for 200 ns of 3VHE+apo, 3VHE+ABZ and 3VHE+42Q were carried out using Desmond trajectory clustering module. The RMSD matrix was built by considering frequency of 20 and possible clusters 10. Initially the highest populated cluster from all the three apo. ABZ and 42Q bound proteins (3VHE) and compared them by superimposing to investigate the conformational changes. Then top two highest and second highest populated clusters are compared for conformation variation at binding cavities of the ligand.

MMGBSA calculations

Molecular Mechanics Generalized Born surface area (MM-GBSA) was used to calculate the binding free energy (ΔG_bind) of docked complexes during MD simulations of 3VHE complexed with ABZ and 42Q. (Schrodinger suite, LLC, New York, NY, 2017–4). The binding free energy was estimated using the OPLS 2005 force field, the VSGB solvent model, and rotamer search techniques. The MD trajectories frames were chosen at 10 ns intervals after the MD run. Eq 1 was used to compute the total free energy binding: (1)

Where,

ΔG_bind = binding free energy,

G_complex = free energy of the complex,

G_protein = free energy of the target protein, and

G_ligand = free energy of the ligand.

The MMGBSA outcome trajectories were analyzed further for post dynamics structure modifications.

Principal component analysis (PCA)

During a 200 ns simulation of the human VEGFR2 kinase domain (PDB ID: 3VHE) complexed with ABZ, PCA analysis was used to recover the global movements of the trajectories. A covariance matrix was generated as described to calculate the PCA. Conformational analysis of Albendazole in-bound complex was done by using 10 appropriate modes of conformation. The main component was calculated as trajectories, and a comparison of the first highest mode (PC1) versus the second highest mode (PC2), followed by PC2 and PC3, and finally, PC9 and PC10 modes were investigated. Geo measures v0.8 were used to calculate the principal components based on eigenvectors from the correlation matrix on an ABZ-bound complex. The MD trajectory versus principal components of trajectory motion was recorded in a 2D plot using the Matplotlib python package using Geo measures, including a comprehensive library of g_sham [29].

Molecular target predictions

Molecular target studies were projected based on their similarities to known compounds to estimate their targets to conduct the molecular mechanisms behind a certain phenotypic or bioactivity and rationalize any side effects. The top 50 results of the closely related receptors in 2D/3D were shown as a pie chart based on Target, Popular Name, Uniprot ID, ChEMBL-ID, Target Class, Probability, and Reported actives (Fig 2). As can be seen, ABZ predicts 26% Kinase receptor, 18% family α G protein-coupled receptor, and 18% Enzymes while 42Q has 34% Kinases and 14% protease, and 6% Enzymes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Pie-chart of top-50 of target predicted for selected molecules (A: ABZ, B: 42Q).

https://doi.org/10.1371/journal.pone.0287198.g002

Overall, the findings demonstrated the efficacy of ABZ as an anticancer treatment that targets the MAP kinase p38 alpha, protein-tyrosine kinase erbB-2, nuclear factor-kappa kinase beta subunit, Serine/threonine-protein kinase, Tyrosine-protein kinase JAK3, Tyrosine-protein kinase JAK2, VEGFR-1.

Prediction of activity spectra for substances (PASS) prediction

PASS employed both ABZ and 42Q to forecast the biological spectrum. Out of a maximum likelihood score of 1, it calculates the probability of activity and inactivity against cancer and non-cancerous cells. With active coefficients of 0.348, 0.320, 0.248, and 0.212, ABZ showed considerable anticarcinogenic action against melanoma, solid tumors, lymphocytic leukemia, and endocrine cancer in the PASS filter (Table 1). 42Q was shown to be effective against melanoma (0.302) and solid tumors (0.348). Nonetheless, both ABZ and 42Q displayed significant activity on endocrine cancer; Platelet-derived growth factor receptor kinase, MAP3K5, Transcription factor STAT, Angiogenesis, etc. These findings point to a high likelihood of anticarcinogenic action. This bioactivity investigation confirms albendazole’s importance in protecting human health against tumor development and inflammation.

Molecular docking

All the binding energy scores are calculated from the best cluster (95%) that falls within the lowest RMSD 0.25 Å. With the lowest binding energy, (G– 7.12 kcal/mol) and inhibitory concentration, Ki (6.04 M), ABZ showed a considerable binding affinity for the human VEGFR2 kinase domain (PDB ID: 3VHE). During the interaction of the ligand ABZ, Asp1046 residues at the binding cavity of the protein were involved in conventional hydrogen bonding. Apart from hydrogen bonding, van der Waal’s interactions by amino acid residues were also involved in non-bonded interaction with the ligand (Fig 3A and 3B left panel). While the standard ligand 42Q displayed a high binding affinity for 3VHE protein where free energy binds were recorded as -12.35 kcal/mol and inhibitory concentration (Ki) 881 pM, respectively. Glu885, Cys919, Asp1044 residues in the binding cavity of3VHE formed conventional hydrogen bonds (Fig 3B, right panel and 3C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Binding pose of molecular docking.

(A) 3VHE+ABZ. A left panel displays the ribbon structure of 3VHE associated with ABZ in the stick model. A middle panel displays the 2D interaction of ABZ (disc-shaped) surrounding the binding cavity residues in 3VHE. A right panel displays the surface view of the 3VHE displaying binding cavity and accommodation of ABZ. (B) A left panel displays ABZ 2D interaction at the binding cavity in the wireframe. A right panel displays 42Q 2D interaction at the binding cavity in the wireframe. (C) 3VHE+42Q. A left panel displays the ribbon structure of 3VHE associated with 42Q in the stick model. A middle panel displays the 2D interaction of 42Q (disc-shaped) surrounding the binding cavity residues in 3VHE. A right panel displays the surface view of the 3VHE displaying binding cavity and accommodation of 42Q.

https://doi.org/10.1371/journal.pone.0287198.g003

Molecular dynamics and simulation

The stability and convergence of the 3VHE+ABZ and 3VHE+42Q complexes were investigated using molecular dynamics and simulation (MD). When comparing the root mean square deviation (RMSD) data, a simulation of 200 ns revealed stable conformation. The RMSD of Cα-backbone of 3VHE bound to ABZ exhibited a deviation of 0.2 Å (Fig 4A (i)), while 42Q displayed a deviation of 0.64 Å. However, 3VHE protein without ligand (3VHE-Apo) exhibited a deviation from the beginning to the end of simulation 0.63 Å. Similar RMSD of ligand bound protein and apo-protein suggested least perturbations and stable conformations. Therefore, it can be suggested that ABZ and 42Q bound 3VHE is quite stable in complex due to the higher affinity of the ligand. Comparison of RMSD clustering having highest population of Apo protein and ABZ bound to 3VHE proteins displayed little conformational changes (Fig 4A (ii)). At the binding cavity of ligand ABZ, couple of residue forming conventional hydrogen bonds viz. Glu917 and Cys919 where we have observed Glu917 pushed away (distance 2.1 Å) from the original position in ligand bound state while comparing with the position in apo protein where the orientation of the residue is much straighter (Fig 4A (ii), left panel, up). RMSD of the deviated residue of ligand bound state and apo state was measured to be RMSD 0.23 Å. The other residue Cys919 is pulled toward ligand ABZ(distance 1.1 Å) while in the apo protein Cys919 seemed to be straighter and unaltered (RMSD 0.21 Å) (Fig 4A (ii), left panel, up). On the other hand, at the C-terminal end residues Asp994-Lys997 of the apo protein conformed randomly into a loop structure whereas, in ABZ bound 3VHE protein a partial helix formed at that position (RMSD 1.5Å) (Fig 4A (ii), left panel, down). Formation of helix structure made the C-terminal of 3VHE more stable than the loop orientation. The top three clusters having highest populations are also compared to see the positional variances of the ligand. Comparison of conformation of top two cluster 1 and cluster 2 displayed aligned ligands having small positional deviation (RMSD 0.25Å) (Fig 4A (ii), right panel, green-cyan, up). Whereas, comparison of cluster 1 and cluster 3 showed significant deviation of ligand twisted around the axis (Fig 4A (ii), right panel, green-magenta, down). Therefore, conformational stability is achieved in the untwisted form of the ligand at the binding cavity than the twisted conformation. Apo and 42Q bound top most clusters having highest population showed significant movements of residues to achieve high stability. In 42Q bound 3VHE (Fig 4A (ii), cyan) two key residues Cys919 and Asp1046 moved apart to form spatial arrangement with the ligand for hydrogen bond formation. While Glu885 pulled toward the ligand to attain a stable hydrogen bond formation where in apo protein the residue lie distantly apart (Fig 4A (ii)). Comparative analysis of top two cluster 1 and cluster 2 having highest population of 42Q bound 3VHE trajectories displayed least ligand mobility from its axis exhibiting very stable complexes (Fig 4A(ii)). The plots for root mean square fluctuations (RMSF) displayed small spikes of fluctuation except at amino acid residue 60 (4.5 Å) and 130–140 (8 Å) in 3VHE-ABZ complex, while the rest of the residues less fluctuated during the entire 200 ns simulation (Fig 4B, dark pink) indicating the stable amino acid conformations during the simulation time. The high fluctuations in ABZ bound protein residues due to loop conformation. However, in 42Q bound protein showed significant fluctuation at 130–140 residues (4.3 Å) (Fig 1B, dark green). In contrast, RMSF values of the 3VHE-apo displayed significant fluctuations, and at 130–140 residues, the fluctuation was found to be 9 Å, signifying very flexible unstable residues ordered into a loop conformation (Fig 4B, orange). Therefore, from RMSF plots, it can be suggested that the structures of 3VHE were stable during simulation in ABZ and 42Q bound conformations. The number of hydrogen bonds between protein and ligand suggests the significant interaction and stability of the complex. A significant number of hydrogen bonds showed between 3VHE & ABZ and 42Q throughout the simulation time 200 ns (Fig 4C). Consistent numbers of hydrogen bonds are observed between 3VHE and ABZ (Average 2 numbers) and 42Q (Average 4 numbers) that might facilitate to conform into stable complex (Fig 4C). The gyration radius is a measurement of the protein’s compactness. Here in this study, 3VHE Cα-backbone bound to ABZ displayed lowering of radius of gyration (Rg) from 20.3 Å to 20.0 Å (Fig 4D, dark pink), and 3VHE Cα-backbone bound to 4Q2 displayed stable as well as lowering of peaks from 20.2 to 20.0 Å (Fig 4D). Similarly, 42Q bound protein displayed the lowering of Rg from 20.3 to 20.2 Å (Fig 4D). Lowering of RMSD indicates the compactness of protein ligand complex. On the other hand, apoprotein (3VHE-apo) displayed significant uneven fluctuations and was devoid of stability, indicating the less compact orientation of the protein (Fig 4D, orange). After Rg analysis, similar patterns were identified in solvent accessible surface area (SASA) in both ligand-bound and unbound states. The unbound state of ligand 3VHE revealed a large surface area accessible to solvent (Fig 4E and 4F, black), but binding with 42Q and ABZ resulted in a lower SASA value than the unbound state (Fig 4E and 4F, dark green and dark pink, respectively). The ligand ABZ and 42Q binding causes the proteins to become more compact and less flexible.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Analysis of non parametric clustering of RMSD for ABZ and 42Q and MD simulation trajectories of 200 ns time scale.

[A(i)] RMSD plot displaying the molecular vibration of Cα backbone of 3VHE-apo (orange), 3VHE+ABZ (dark pink), and 3VHE+42Q (dark green). A(ii) a. Comparison of conformations Apo and ABZ bound top RMSD cluster. A (ii) b. Comparison of conformations ABZ bound top thee RMSD clusters. A (ii) d. Comparison of conformations 42Q bound top three RMSD clusters. A (ii) c. Comparison of conformations Apo and 42Q bound top RMSD cluster. [A (ii)] Non parametric clustering of RMSD for ABZ and 42Q and conformational comparison with apo-protein highest populated cluster as well as intra top three highest populated clusters. B: RMSF plots show the fluctuations of respective amino acids throughout the simulation time 200 ns for -apo (orange), 3VHE+ABZ (dark pink), and 3VHE+42Q (dark green). (C) Number of hydrogen bonds formed between -apo (orange), 3VHE+ABZ (dark pink), and 3VHE+42Q (dark green) during 200 ns simulation time scale. (D) The radius of gyration plots for the deduction of compactness of protein -apo (orange), 3VHE+ABZ (dark pink), and 3VHE+42Q (dark green). Solvent accessible surface area (SAS Area) displaying the ligand-bound and unbound area (Black) at the binding pocket. (E) 3VHE+ABZ (dark green). (F) 3VHE+42Q (dark green).

https://doi.org/10.1371/journal.pone.0287198.g004

Molecular mechanics generalized born and surface area (MMGBSA) calculations

The MMGBSA technique is widely used to determine the binding energy of ligands to protein molecules. The binding free energy of each of the 3VHE+ABZ and 3VHE+42Q complexes was calculated and the involvement of additional non-bonded interaction energies. The average binding energy of the ABZ bound to 3VHE was found to be -42.07 ± 2.4 kcal/mol, whereas the average binding energy of 42Q was determined to be -80.94 ± 2.25 kcal/mol (Table 2). G_bindCoulomb, G_bindCovalent, G_bindHbond, G_bindLipo, G_{bindSolvation}, and G_bindvdW interactions are all examples of non-bonded interactions that affect G_bind. G_bindvdW, G_bindLipo, G_bindHbond, and G_bindCoulomb energies contributed the most to obtain the average binding energy across all types of interactions. The G_{bindSolvation} and G_bindcovalent energies, on the other hand, contributed the least to the final average binding energies.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Non-bonded interactions between ligand molecules and the target protein contribute binding energy to MMGBSA.

https://doi.org/10.1371/journal.pone.0287198.t002

MMGBSA trajectories of 3VHE+ABZ and 3VHE+42Q complexes are further analyzed for positional variances of the respective ligand at the binding pocket. Complex 3VHE with ABZ displayed change in orientation in the ligand position from 0 to 200 ns (Fig 5A). While least arrangement is observed in the 42Q ligand (Fig 5B). MMGBSA trajectory of ABZ and 42Q with 3VHE displayed a very stable configuration throughout the simulation. Moreover, high binding energies were observed due to stable conformation with the protein-ligand complex. Therefore, for the entire simulation studies and MMGBSA analysis, it can be suggested that both ABZ and 42Q have significant affinity for the target 3VHE and can be potential inhibitors of drug molecules.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Positional changes comparison of first (0 ns) and last frame (200 ns) of ligand (A) ABZ and (B) 42Q bound to 3VHE from MMGBSA trajectories.

https://doi.org/10.1371/journal.pone.0287198.g005

Analysis of principal component (PCA) with free energy surface/landscape (FEL)

Principal component analysis (PCA) of the MD simulation trajectories for ABZ bound to the VEGFR2 kinase domain was analyzed to interpret the randomized, global motion of the atoms of amino acid residues along with free energy. This analysis interprets the more flexible scattered trajectories due to the protein structure’s randomness due to non-correlated global motion. The internal coordinate’s mobility into three-dimensional space in the spatial time of 200 ns was recorded in a covariance matrix. The rational motion of each trajectory is interpreted in the form of orthogonal sets or Eigenvectors. MD simulation trajectory of Cα atoms of ABZ bound to VEGFR2 kinase domain displayed more unordered orientation in PC1 and PC2 modes where, the free energies are not converged localized (Fig 6A, blue, as per scale). Following this, PC2 and PC3 mode displayed better order Eigenvalues for the trajectories with partially ordered free energy surface (Fig 6B). The Eigenvectors displayed a less scattered, more correlated and partially converged free energy surface of the last 50 trajectories (Fig 6B). Most of the trajectories finally settled in PC9 and PC10, where the global motion centred toward the origin of the plot (Fig 6C). Eigenvectors are observed. The global movement of MD trajectories toward a positive correlation Eigen vector indicates that the VEGFR2 kinase domain complex is compact and converged. The high periodic global motion was observed along the PC9 and PC10 planes due to the grouping of trajectories in a single free energy surface at the centre of the PCA plot toward origin. Cantering of the trajectories in a single cluster indicates the periodic motion of MD trajectories due to stable conformational global motion (Fig 6C, blue, as per scale).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6.

Comparative free energy surface/landscape (FEL) of human VEGFR2 kinase bound with ABZ displaying (a) PC1 and PC2, (b) PC2 and PC3, and (e) PC9 and PC10, for 200 ns simulation trajectories. 42Q with human VEGFR2 kinase domain displaying (d) PC1 and PC2, (e) PC2 and PC3 and (f) PC9 and PC10, for 200 ns simulation trajectories.

https://doi.org/10.1371/journal.pone.0287198.g006

On the other hand, the 42Q bound VEGFR2 kinase domain displayed a similar observation for PCA as found in ABZ bound state. The MD trajectories in PC1 and PC2 modes displayed more disoriented distribution and less free energy converged zones (Fig 6D, blue), while in PC2 and PC3 modes, trajectories started organizing more into partially definite free energies (Fig 6E, blue). Interestingly increment of PC modes up to PC9 and PC10 displayed more ordered trajectories. The last 50 trajectories (yellow dots) in PC9 and PC10 showed ordered orientation toward the origin of the Eigenvalues (Fig 6F, blue). Ordered orientation signifies the ordered global motion of the trajectories that arose due to a stable converged structure and high free energies toward zero. Therefore, at this stage global minimization achieved.