Background

Chronic myeloid leukemia (CML) is a myeloproliferative disease characterized by the translocation of a section of human chromosome 9 which contains ABL kinase domain with specific breakpoint cluster region (BCR) on chromosome 22 (Faderl et al. 1999). The ABL1 gene on chromosome 9 and BCR gene on chromosome 22 together involved in the formation of BCR–ABL which is basically the active oncogene tyrosine kinase (TK) (Kang et al. 2016). BCR–ABL tyrosine kinase plays a crucial role in the development of CML (Jabbour and Kantarjian 2014). Thus, the inhibition of BCR–ABL is the best possible option for the management of CML.

Imatinib, a tyrosine kinase inhibitor, inhibits ATP binding site of BCR–ABL. It prevents tyrosine phosphorylation and downstream signaling (Shah et al. 2002). Additionally, the conformation of tyrosine kinase changes over time due to mutation and made completely inaccessible to imatinib (Mauro 2006; Shah et al. 2013; Le Coutre et al. 2008; Steinberg 2007). BCR–ABL escalation at the genomic level has been associated with imatinib and other kinase inhibitors’ failure (Eadie et al. 2018). This failure may overcome by targeting specific protein instead of BCR–ABL (Druker et al. 2001). The effects of imatinib on a number of tyrosine kinases depicted in Fig. 1 (Szklarczyk et al. 2015).

Fig. 1
figure 1

Numerous targeted proteins for imatinib which were generated by STITCH server

Full size image

Proto-oncogene tyrosine-protein kinase (KIT), Abelson (ABL2), Colony-stimulating factor 1 receptor (CSF1R), Lymphocyte-specific protein tyrosine kinase (LCK), Platelet-derived growth factor receptors A (PDGFRA), Platelet-derived growth factor receptors B (PDGFRB) are the notable target of interests which are further divided into two categories. Receptor protein kinase including BCR–ABL, ABL2, KIT, a CSF1R kinase domain and Non-Receptor protein kinase including PDGFRA, PDGFRB, LCK kinase domain (Rask-Andersen et al. 2014). Imatinib exhibited increased activity toward ABL. However, activity against PDGFR, KIT, and LCK never performed before. Thus, screening new inhibitors by targeting all these proteins may reveal a new, wider possibility for the management of CML.

Biologically active molecules were screened out from kinase database and further subjected to molecular docking, ADMET profiling, and molecular dynamics simulation in search of new inhibitor that could provide better affinity than imatinib (Quintás-Cardama et al. 2007; Goldman and Melo 2003; Krieger and Vriend 2014).

Materials and method

Selection and preparation of proteins

Targeted proteins had carefully chosen from the stitch database and SwissTargetPrediction web server. Crystal structure of proteins were retrieved from PDB is ABL1, ABL2, CSF1R, KIT, LCK, PDGFRA, PDGFRB, and their corresponding PDB IDs are 4WA9, 3GVU, 2I1 M, 4U0I, 2PL0, 5K5X and 3MJG. Ligand and water molecules removed using pymol. Missing residues were corrected using MOE (Molecular Operating Environment) protein preparation wizard. The bound Zn2+ and Cl− ions removed and bond orders were determined. Proteins preparation performed by adding polar hydrogens and Gasteiger charge using Auto Dock Tool 1.5.6.

Binding site prediction

MetaPocket 2.0 has been used to predict the top three binding cavities of ABL2, CSF1R, KIT, LCK, PDGFRA, and PDGFRB proteins (Huang 2009). The accuracy of predicting binding pockets enhanced by means of the LIGSITE. Z-score evaluated to select the top binding cavities. Desired bioactive molecules had docked with the first pocket of the protein. In order to validate the pockets of the protein obtained from MetaPocket, data compared with the CASTp server (Dundas et al. 2006).

Virtual screening and preparation of ligand molecules

Ligand-based virtual screening was carried out using a kinase database comprising 177,000 molecules (biological activity < 10 µM) and kinase inhibitor (ChEMBL) database. About 135 compounds were screened out which were closely similar to imatinib based on the cut-off value of above 80%. Selected structures downloaded from the PubChem database and prepared for docking studies. Polar hydrogen and Gasteiger charge added using Auto Dock Tool 1.5.6., and the energy of candidates minimized for docking purposes.

Molecular docking

135 bioactive molecules subjected for consensus docking studies using the AutoDock, Achilles, Mcule server (Morris et al. 1996; Nagasundaram et al. 2015; Pettersen et al. 2004; Hassan et al. 2017). The docking program calculated energies to obtain the best binding mode. Based on the binding energies better than Imatinib, 22 molecules had sorted out for further ADME/T profiling.

ADMET prediction

SwissADME, AdmetSAR, and DruLiTo tool were used to predict the topological polar surface area (TPSA), rotatable bonds, H-bond acceptors, H-bond donors, fraction Csp3 from physicochemical properties, iLOGP, WLOGP for lipophilicity, ESOL Class for solubility, Pharmacokinetics profile, Lipinski rule of 5, druglikeliness, AMES toxicity, carcinogenicity and acute oral toxicity. Finally, after ADME/T screening, four molecules passed all the criteria and selected as the top hits for molecular simulation study and multi-target analysis with ABL2, CSF1R, KIT, LCK, PDGFRA, and PDGFRB.

Molecular dynamics (MD) simulation of the complexes

ABL1 complexed with the top hits were subjected to MD simulation using YASARA in Windows 64-bit operating system (Krieger and Vriend 2015; Mitra and Dash 2018). Each complex cleaned and hydrogen bond network had optimized. Refinement of these complexes performed with subsequent relaxation by steepest descent minimization and subjected for full potential energy minimization for 5000 cycles until convergence reached. After that complexes were set for simulation annealing minimization to adapt with deleted water (Isa et al. 2018; Mandlik and Singh 2016; Jakalian et al. 2002; Pascoini et al. 2018). At constant pressure, MD simulation of these complexes had carried out for a period of 50 ns. Force field parameters for the complexes obtained by using the AMBER force field (Krieger et al. 2002, 2006; Dickson et al. 2014). The complexes had placed in a cubic box and filled with transferable intermolecular potential 3P (TIP3P) water molecules (Skelton et al. 2011). All simulations had performed at 298 K under certain periodic boundary conditions. The pH set to 7.4 and 0.9% NaCl had maintained. Time step had used with force cut-off 8 Å. Solvent density assigned for 0.997 g/ml. Each simulation system consisted of 29,400 ± 140 atoms. Using the YASARA structure built-in macros the resulting trajectories subjected to analyze the stability. To understand the relative stability of the ligand inside its binding pocket, hydrogen bonds between the solute and solvent analyzed. For further experimental validation, the radius of gyration (Rg) of the solute, RMSD and RMSF of the ligand bound protein had generated (Krieger and Vriend 2015; Krieger et al. 2002; Maier et al. 2015; Stewart 1990; Wang et al. 2004; Meng et al. 2011; Sánchez-Linares et al. 2012).

Result and discussion

Consensus docking of bioactive ligands

In this study, consensus docking had performed to find out the best scores possible from 135 bioactive molecules obtain from the PubChem database. Consensus binding energies of 22 molecules showed in Table 1 and Fig. 2.

Table 1 Consensus binding affinities of 22 bioactive molecules with ABL1
Full size table
Fig. 2
figure 2

The chart showing consensus (Autodock, Chimera, Achilles, Mcule) docking score of 22 molecules

Full size image

Results of Lipinski’s rule, and Veber filter demonstrated in Table 2A&B. These consensus-binding scores had compared with imatinib and presented in Fig. 3. 22 bioactive molecule and ABL1 protein residual interactions showed in Table 3.

Table 2 molecules under Lipinski’s rule (A) and molecules under Veber filter (B)
Full size table
Fig. 3
figure 3

Consensus (Autodock, Chimera, Achilles, Mcule) docking score of Imatinib

Full size image
Table 3 Residual interaction results of ABL1 with 22 bioactive molecules
Full size table

ADMET analysis

The 22 compounds selected for further analysis to predict the ADME/T parameter, such as, predict physicochemical descriptors, drug likeliness, and toxicity. Two servers had used to calculate the important information required to validate the selected bioactive ligands. The results achieved from the two servers listed in Table 4.

Table 4 ADMET profile of 22 bioactive molecules
Full size table

According to the results, GI absorption of all molecules was high. TPSA is an important interpreter for BBB penetration. The standard acceptable range of TPSA is from 90 to 140. If the TPSA value is less than 90 then it has the possibility to cross BBB. Boiled egg from SwissADME for all the molecules had analyzed. The boiled egg rendered according to WLOGP vs TPSA. This graphical representation illustrated that molecules, which presented in the yellow portion, literally can penetrate the blood–brain barrier (BBB). Red and blue colored dots (molecules) indicated PGP positive or PGP negative. Blue colored molecules have the possibility to efflux by PGP positive and red-colored molecules cannot efflux. According to TPSA score from Boiled egg analysis, PCID 9822042, PCID 10110378, PCID 10181075, PCID 24970384, PCID 25235817, PCID 44273619, PCID 2931814 and PCID 46291494 molecules were BBB permeate positive and dropped out. Based on ESOL LogS scale all molecules were within − 4.2 to − 5.97, which exhibited normal water solubility. All molecules were within range (from − 0.4 to 5.6) of MLogP except PCID 24848331, which showed − 1.53. Log Kp (the skin permeability coefficient) of all molecules were ranging from − 5.3 cm/s to − 7.02 cm/s. The more negative the log kp, the less skin permeate was the molecule. Based on Log Kp, PCID 24971209 dropped out. According to a carcinogen and AMES toxicity profile, PCID 25128185, PCID 10111108, PCID 25125572 and PCID 25125902 molecules showed positive toxicity signs and dropped out.

According to acute oral toxicity (Level I Fatal if swallowed, Level II Fatal if swallowed, Level III Toxic if swallowed, Level IV Harmful if swallowed, Level V may be Harmful if swallowed), PCID 44340691, PCID 51351016, PCID 51351018 and PCID 73353596 were dropped out as they had shown Level II toxicity.

According to the synthetic accessibility profile (SA score 1 means very easy to synthesize and 10 means very difficult to synthesize), PGP Substrate profile, and TPSA value from boiled egg analysis and considering other parameters, PCID 4369496, PCID 10181160, PCID 58172700 and PCID 72724706 had fulfilled all of the requirements of ADMET. The ADMET data (Boiled egg analysis from SwissADME) of 22 molecules presented in Fig. 4 and the structure of the selected four molecules showed in Fig. 5.

Fig. 4
figure 4

Boiled egg analysis from SwissADME for 22 bioactive molecules

Full size image
Fig. 5
figure 5

Structures of desired four molecules (a) PCID 4369496, b 10181160, c PCID 58172700 and d PCID 72724706

Full size image

Multi-target analysis

Though ABL1 is the key target for CML, there are other proteins CSF1R, KIT, LCK, ABL2, PDGFRA, and PDGFRB, which could be also important target to combat CML. After ADMET analysis, PCID 72724706, PCID 10181160, PCID 4369496 and PCID 58172700 bioactive molecules found as potential hit molecules. Consensus docking scores of imatinib and hits with the CSF1R, KIT, LCK, ABL2, PDGFRA, and PDGFRB illustrated in Figs. 6 and 3D & 2D interaction of selected four molecules with ABL1 presented in Fig. 7.

Fig. 6
figure 6

Consensus docking scores of top 4 bioactive molecules and imatinib with ABL2 (a), CSF1R (b), KIT (c), LCK (d), PDGFRA (e) and PDGFRB (f)

Full size image
Fig. 7
figure 7

3D and 2D interaction diagram of PCID 4369496 (a, b), PCID 10181160 (c, d), PCID 58172700 (e, f), PCID 72724706 (g, h) with ABL1

Full size image

Molecular dynamics simulation study

The protein ABL1 is an important drug target in the treatment of CML. To account for the flexibility of the protein and selected four ligands, a 50 ns molecular dynamics simulation of the docked complexes carried out. Four main parameters analyzed throughout the simulations, which were Root Mean Square Deviation (RMSD) and Root Mean Square Fluctuations (RMSF), Radius of gyration of the complexes.

Root mean square deviation (RMSD)

The stability of the complexes found validated by calculating the root mean square deviation. The RMSD value showed in Fig. 8 and in which observed range up to 6.5 Å showing that the systems are well converged. —PCID 4369496 was stable up to 34 ns then deviated. PCID 10181160 became stable after 23 ns, PCID 58172700 remained stable from 8 ns to 43 ns then deviation occurred. PCID 72724706 stabilized after 33 ns. PCID 10181160 and PCID 72724706 showed greater stability than the rest two.

Fig. 8
figure 8

RMSD of the ABL1 with ligand PCID 4369496 (a), PCID 10181160 (b), PCID 58172700 (c) and PCID 72724706 (d)

Full size image

Root mean square fluctuation (RMSF)

The root mean square fluctuation evident in Fig. 9. On this plot, peaks demonstrate the areas of the protein that fluctuated most in the entire simulation period. The overall fluctuations of the RMSF of the ligands found from a range of 1–14 Å throughout the simulation. The average RMSF values for the generated molecules achieved 13.07 Å (PCID 4369496), 11.65 Å (PCID 10181160), 6.08 Å (PCID 58172700) and 5.78 Å (PCID 72724706) in which the protein flexibility had conferred.

Fig. 9
figure 9

RMSF data for each complex

Full size image

The radius of gyration (Rg)

The radius of gyration shows the compactness of protein and protein–ligand complex i.e., how much protein is folded or unfolded with or without ligand. From Fig. 10, it had found that the radius of gyration of the molecules is constant over time and maintained the same mean value of 21.0 ± 0.5 Å. However, the oscillations were greater for PCID 4369496. Rg may suggest that this molecule was less stable than the rest. From the overall molecular dynamic study, these two compounds PCID 10181160 and PCID 72724706 formed stable complexes with ABL1 during 50-ns MD simulations.

Fig. 10
figure 10

Radius of gyration of ABL1 with four selected molecules

Full size image

Conclusion

The present, in silico studies, provides insight into the inhibition of ABL1, ABL2, CSF1R, KIT, LCK, PDGFRA, and PDGFRB by imatinib and its bioactive molecules. Virtual screening, Consensus docking studies, Multi-target analysis, and ADMET profiling suggest that PCID 4369496, PCID 10181160, PCID 58172700, and PCID 72724706, among 22 bioactive molecules from 135 bioactive molecules with the best docking scores than imatinib from a database of 1,77,000 molecules, are the most active inhibitors. These four molecules further selected for molecular dynamics (MD) simulation analysis, which revealed that the two compounds PCID 10181160 and PCID 72724706 formed stable complexes with ABL1 during 50-ns MD simulations. These lead molecules are the potential CML inhibitors that could resolve the resistance pattern. Further chemical synthesis, wet lab analysis, and experimental validation deserve the utmost attention.