Cancer remains one of the most formidable challenges to global health, accounting for millions of deaths worldwide each year. While the disease manifests in various forms, the hallmark of cancer is uncontrolled cell growth and proliferation. Among the molecular players orchestrating these aberrant cellular behaviors, protein kinases emerge as key protagonists. These enzymes, which regulate critical signaling pathways, have become the focal point of modern cancer therapy due to their critical role in tumor initiation, progression, and metastasis [1] . In recent decades, the emergence of protein kinase inhibitors (PKIs) has revolutionized cancer treatment [2] - [4] . These targeted therapeutics act by inhibiting the activity of specific protein kinases, thus disrupting the signaling cascades that sustain malignant growth [5] . Imatinib, a breakthrough PKI developed to target the BCR-ABL fusion protein in chronic myeloid leukemia (CML), represents a paradigm-shifting success story in the field of oncology. Its introduction led to unprecedented responses in CML patients, effectively transforming a once-fatal disease into a manageable chronic condition [6] . The success of imatinib promoted intensive research into PKIs targeting various kinase-driven cancers as well as non-cancer conditions [7] . Constant identification of novel protein kinase targets and the revealing of target-related drug resistance mechanisms underscore the need for continuous innovation in PKI development. In recent decades, structure-based drug design (SBDD), accompanied by pharmacophore modeling and computation methods for binding affinity estimation, has emerged as an effective tool for the discovery and structure optimization of novel protein kinase inhibitors [8] . Crizotinib [9] , Osimertinib [10] , and Axitinib [11] , as well as many other FDA-approved PKIs, are valuable examples of SBDD applications. On the other hand, in silico methods offer cost-effective means for rapid lead identification. Novel inhibitors of ZAP70 [12] , CDK9 [13] , and ABL-T315I [14] are recent examples of successful use of computational methods for the development of novel protein kinase inhibitors with in vitro validation.

In this study, 25 new N-phenylbenzamide derivatives were specifically designed and evaluated as potential protein kinase inhibitors against 102 protein kinases using molecular docking and molecular dynamics methods.

Study Design. Chemical compounds reflecting several chemical classes were specially designed to resemble the pharmacophore properties of type-2 protein kinase inhibitors [15] . Three-dimensional structures of these compounds were generated. In the first step, binding affinities of these structures to various protein kinase receptors were studied using molecular docking. The obtained binding affinities were compared to those of known inhibitors, and the best structures were taken to the next step, which was to estimate the binding energy of selected structures using molecular dynamics simulation followed by MM-GBSA workflow. For each structure, six experiments of binding energy estimation were conducted: five short-term 2 ns simulations and one 20 ns simulation. Given that a substantial proportion of approved kinase inhibitors are, in fact, type-1 inhibitors, which encompass a diverse array of chemical scaffolds, our receptor set was not exclusively limited to type-2 receptors. Rather, it also included type-1 receptors to facilitate a more comprehensive investigation of the potential anti-kinase activity of the designed structures.

Docking. Structures of receptors were taken from the Protein Data Bank [16] . Three-dimensional structures of studied ligands were generated using the Cactus web server [17] . ADFR-suite was used for the preparation of the receptors and ligands for docking (prepare ligand, prepare receptor, and reduce programs) as well as for generating the coordinates and sizes of binding sites using coordinates of native ligands [18] . Qvina 2.1, with exhaustiveness set to 16, was used as docking software [19] . RMSDs for binding poses were calculated using RDKit (version 2020.09.1) Python package and custom scripting [20] .

Molecular Dynamics. GROMACS (version 2022.4) was used as a molecular dynamics package [21] . Ligands were converted from Qvina’s docking poses to sybyl mol2 format using OpenBabel [22] and protonated using Chimera [23] . Amber99sb-ildn force field was used for the simulations [24] . Ligands were parametrized to the Amber force field using the ACPYPE interface to Antechamber [25] . The protein-ligand system was prepared for the simulation using a dodecahedron box of 1 A, tip3p water model, and 0.15 mmol concentration of NaCl, followed by energy minimization and two equilibration steps (NPT and NVT) 200 ps each. The system state after the equilibration was used as a starting point for the separate production runs of different lengths. Parrinello-Rahman barostat and V-rescale thermostat were used for coupling. The Particle Mesh Ewald (PME) method was used for long-range electrostatic calculations. GMX-MMPBSA package was used to estimate binding energy [26] . The first 20 frames of the molecular dynamics trajectory of the protein-ligand complex were skipped. Interaction entropy was included in the binding energy calculations with the parameter “ie segment” being set to 25. The number of hydrogen bonds between receptor and ligand was estimated using the hbond module of the GROMACS package.

Design of Target Structures. Type-2 protein kinase inhibitors, such as imatinib, nilotinib, and sorafenib, bind to the inactive conformation of their target protein kinase. Such inactive conformations are characterized by the extension of the ATP-binding pocket from the active conformation to an adjacent, so-called ‘allosteric’ pocket. The examination of experimental structures of protein-ligand complexes reveals the most important pharmacophore properties of type-2 inhibitors. The structure of an inhibitor should contain a heterocyclic moiety that mimics the purine fragment, allowing it to bind in the ATP pocket of the inactive kinase. In the case of imatinib and nilotinib, 3-pyridyl-2-amino-pyrimidine is used to mimic the purine from the ATP molecule, while sorafenib uses pyridine and ponatinib uses imidazo-pyridazine. For the allosteric pocket, an amide group (in imatinib and nilotinib) or a carbamide group (in sorafenib) is necessary to form crucial hydrogen bonds, while a benzoic ring with different substituents provides van der Waals interactions. The linker, which is typically a benzoic ring, provides the necessary orientation for the ATP and allosteric fragments of an inhibitor. In this study, based on the abovementioned considerations, we propose novel chemical structures specifically designed as type-2 protein kinase inhibitors. The key idea behind our design approach is to retain the N-phenylbenzamide linker-allosteric framework of imatinib while replacing the pyrimidine with a benzimidazole or purine fragment, followed by the introduction of some substituents, primarily in the allosteric benzene ring ( Figure 1 , Table 1 ), The strategy of utilizing the N-phenylbenzamide linker previously led us to the discovery of several chemical scaffolds able to suppress the growth of tumor cell lines [27] [28] .

These structures were evaluated as protein kinase inhibitors using molecular docking and molecular dynamics.

Docking. The structures of 102 protein kinases from protein-ligand complexes of known inhibitors were used for docking. The structures were selected based on the approved protein kinase inhibitors list, followed by a manual examination of the PDB database. Selected complexes included both type-1 and type-2 inhibitors as native ligands.

First, docking of native ligands that were found in obtained PDB complexes was conducted to evaluate the performance of Qvina 2.1. The criteria for the docking pose of known ligands to be considered correct was the root mean squared deviation (RMSD) between atoms of the native ligand and the obtained docking pose. In our case, Qvina 2.1 was able to reproduce the binding pose of the original ligands according to the abovementioned criteria for 63 complexes out of 102 (62%, Figure 2 ).

Studied structures were docked into kinase receptors. Obtained docking poses that showed binding affinity higher than those for native ligands having proper RMSD were of particular interest. These “high-affinity” poses were found for 19 receptors, most of which are of type-1 or type-1/2 ( Figure 3 ).

Analysis of the structural features of the obtained compounds revealed that derivatives of N-methyl-1H-benzo[d]imidazole-6-carboxamide, series 2 (2a, 2b, 2d) showed the largest number of docking positions with high affinity in comparison with methyl esters (1a, 1b, 1f) or analogs containing purine bases 3a, 3b, 3c ( Figure 4 ). It is also interesting to note that the presence of the fluorine atom in the meta position of the benzene ring has increased the affinity with the active center of the protein. However, replacing the fluorine atom in the para position with a methoxy group led to a sharp drop in activity.

Example visualization of docking poses of 2a for type-1 (6bbu, Figure 5(А) ) and type-2 (6es0, Figure 5(B) ) receptors suggests that for this structure, type-2 binding mode is more favorable as there is better alignment with native ligands [29] [30] ( Figure 5 ). A notable difference exists in how the benzimidazole moiety of 2a is positioned relative to the adenine-binding region of ATP-pocket: it occupies this space only for type-2 receptor (6es0) while being pushed outside the location in the case of type-1 kinase conformation (6bbu). Also, a hydrogen bond was found for type 2 binding.

Studied structures from high-affinity docking poses and proceeded to the next step for the evaluation of binding energy using molecular dynamics workflow.

Molecular Dynamics. MM-PBSA binding energies against 12 kinase receptors were estimated for 71 studied structures and corresponding native ligands as a reference (a total of 83 protein-ligand complexes). The results are shown in Table 2 and visualized in Figure 6 . In general, studied structures showed poorer binding energy if compared to known reference ligands. In most cases, binding energies for type-2 receptors for both reference ligands and studied structures were better than those for type-1 receptors. For two structures, 4a for the 2f4j receptor and 3b for the 4i23 receptor, binding energies that were estimated from the 20 ns simulation exceeded the values for the reference ligands. It should be noted that during the calculation, GMX-MMPBSA was not able to establish obtained entropy values as reliable, so entropy was not part of the final binding energy.

Analysis of molecular dynamics trajectory also showed that an average number of hydrogen bonds between receptor and ligand was, in most cases, greater for known inhibitors ( Table 3 , Figure 7(A) ). Studied structures showed better average hydrogen bond numbers for receptors 2f4j (2a, 4а, 3b), 4i23 (1b, 2b, 3b) and 4oti (3c). All these receptors represent type-1 protein kinases. In terms of the maximum number of hydrogen bonds ( Figure 7(B) ), the results of studied structures and reference ligands could be, in general, considered as close. But in the case of type-2 receptors, known inhibitors showed a clear advantage, except for 7mfc, as structure 4a showed a maximum of 5 hydrogen bonds for this receptor, while native ligand vimseltinib showed 4 [31] . Nevertheless, vimseltinib maintained a higher average number of hydrogen bonds (2.58 versus 1.72 for compound 4a, Table 3 ).

Energy decomposition data is shown in Figure 8 .

Comparing the energy decomposition data in relation to the energy of hydrogen bonds with amino acid residues in the active center of the 7mfc protein of the native ligand (Vimseltinib) and compound 4a showed a decrease in the binding energy with Tyr-99, Val-81, Met-71, Gly-183, and Cys-100. In the case of 4a, new interactions were found for Lei-22, Val-30, and Ile-70 residues, and there was a notable elevation in binding energy associated with Glu-67 ( Figure 8 ). However, the appearance of new bonds apparently cannot compensate for the loss of binding with Leu-173, Ala-48, Lys-50, and Glu-98. As noted above, the average number of hydrogen bonds for Vimseltinib was 35% greater than for 4a ( Table 3 ).

For the 4i23 receptor, structures 2b and 3b, containing a fluorine atom in the meta position of the benzene ring, showed a maximum number of hydrogen bonds of four instead of two for the reference ligand dacomitinib [32] . Also, the average number of hydrogen bonds for structures 2b and 3b was 1.5 times higher compared to the native ligand ( Table 3 ). In general, the energy decomposition profile was similar for dacomitinib and structures 2b and 3b ( Figure 8 ).

Comparison of the energy decomposition of the binding energy by amino acid residues of oncoprotein 2f4j with structures 2a, 2b, 3b, and 4a ( Figure 8 ) showed their general similarity. However, the formation of strong additional interactions with Asn-322 is observed in the case of compound 2a, which probably affects the higher average number of hydrogen bonds compared to the reference ligand (1.58 versus 1.02) [33] .

As a result of this study, 25 novel hybrids of 4-methylbenzamide and benzimidazole or purine were designed as protein kinase inhibitors and evaluated using molecular docking and molecular dynamics methods to estimate their binding affinity to 102 protein kinase receptors. It was shown that, in most cases, Qvina 2.1. is capable of accurately reproducing binding modes of known protein kinase inhibitors that were used as a reference.

For studied structures, docking poses with a score better than those of reference ligands were found for 19 receptors that mostly represent type-1 or type-1/2 kinase conformations. N-methyl-1H-benzo[d]imidazole-6-carboxamide (series 2) was the most common scaffold in docking poses with the highest binding affinity.

At the same time, MM-GBSA binding energies obtained from a series of molecular dynamics simulations were, in most cases, better for reference ligands.

Analysis of protein-ligand complexes of studied structures that were constructed by virtual combinatorial chemistry methods indicates the presence of effective intermolecular interactions leading to the blockade of the catalytic site of kinase enzymes. Comparison of the profiles of binding energy decomposition by amino acid residues of the active center of oncoproteins with pdb ids 2f4j, 4i23, 7mfc shows their general similarity between reference ligands and structures 2a, 2b, 3b, and 4a. In some cases, for studied structures, the formation of additional bonds was observed, which probably may improve binding energy.

The data obtained show that the use of the 4-methylbenzamide linker makes it possible to create promising lead structures for the development of effective multikinase drugs for antitumor therapy.

The research was funded by the National Academy of Sciences of Belarus within the State Scientific Investigation Program “Chemical processes, reagents and technology, bioregulators and bioorganic chemistry, subprogram “Chemical basis of life processes”, project No. 2.3.2.1.

Elena Kalinichenko coordinated the project and was responsible for general supervision and the review of the manuscript; Aliaksandr Faryna performed experimental work, including target structure design and molecular modeling studies, wrote and edited the manuscript. All authors have read and agreed to the published version of the manuscript.