Three-Dimensional-QSAR and Relative Binding Affinity Estimation of Focal Adhesion Kinase Inhibitors

Abstract

Precise binding affinity predictions are essential for structure-based drug discovery (SBDD). Focal adhesion kinase (FAK) is a member of the tyrosine kinase protein family and is overexpressed in a variety of human malignancies. Inhibition of FAK using small molecules is a promising therapeutic option for several types of cancer. Here, we conducted computational modeling of FAK-targeting inhibitors using three-dimensional structure-activity relationship (3D-QSAR), molecular dynamics (MD), and hybrid topology-based free energy perturbation (FEP) methods. The structure-activity relationship (SAR) studies between the physicochemical descriptors and inhibitory activities of the chemical compounds were performed with reasonable statistical accuracy using CoMFA and CoMSIA. These are two well-known 3D-QSAR methods based on the principle of supervised machine learning (ML). Essential information regarding residue-specific binding interactions was determined using MD and MM-PB/GBSA methods. Finally, physics-based relative binding free energy (ΔΔGRBFEA→B) terms of analogous ligands were estimated using alchemical FEP simulation. An acceptable agreement was observed between the experimental and computed relative binding free energies. Overall, the results suggested that using ML and physics-based hybrid approaches could be useful in synergy for the rational optimization of accessible lead compounds with similar scaffolds targeting the FAK receptor.

Keywords: 3D-QSAR; MM-PB/GBSA; focal adhesion kinase; free energy perturbation; molecular dynamics.

Figure 1

MD simulation and MM-PB/GBSA binding energy calculation. (a) RMSD analysis of the protein backbone and TAE226 during 100 ns of MD simulation. The distances of the two intermolecular H-bonds (Hb-1 and Hb-2) with the carbonyl and amide groups of C502 are shown during the MD run. (b) Binding affinity calculation and residue-specific binding energy decomposition from the MM-PB/GBSA calculation. (c) Residues within 4 Å distance of the TAE226 atoms, that contribute critical binding energy to the ligand, are shown in the stick representation.

Figure 2

Molecular alignment of the dataset compounds, PLS plots, and applicability domain (AD) analysis. (a) Molecular alignment of the dataset compounds on the common chemical core by taking C36 inside the FAK binding cavity. (b) PLS correlation plots of the CoMFA (SET-D) study. (c) PLS correlation plots of the CoMSIA (SET-D) study. (d,e) Applicability domain analysis using the distance-based Williams plot using the data obtained from the CoMFA and CoMSIA models. The h* with dotted lines in red signifies the warning leverage values in both plots.

Figure 3

Contour map analysis and structure–activity relationship study from 3D-QSAR. (a) The steric contour map and (b) electrostatic contour map from CoMFA. (c–e) are steric, electrostatic, and H-bond (Hb) donor contour maps from CoMSIA. (f) Implementation of the SAR diagram from the CoMFA and CoMSIA analyses by taking TAE226 (C36) as a reference.

Figure 4

Overview of the FEP scheme and relative binding affinity estimation. (a) Thermodynamic pathway of ligand transformation from state-A to state-B in aqueous and complex form. The $\Delta \Delta {\mathrm{G}}_{\mathrm{RBFE}}^{\mathrm{A}\to \mathrm{B}}$ can be deduced from the free energy changes of both states in aqueous and complex systems. (b) Relative binding energy calculation of the ligands through alchemical transformation. The mismatched atoms between the ligand pairs, which need to be perturbed, are shown in red.

Figure 5

FEP energy convergence plots of (a) C36 → C28, (b) C36 → C38, (c) C36 → C64, (d) C36 → C73, (e) C36 → C76, (f) C36 → C80, (g) C36 → C83, (h) C70 → C45, (i) C36 → C89, and (j) C36 → C114 in isolated and complex form.