Enhancing binding affinity predictions through efficient sampling with a re-engineered BAR method: a test on GPCR targets

Abstract

Computational approaches for predicting the binding affinity of ligand-receptor complex structures often fail to validate experimental results satisfactorily due to insufficient sampling. To address these challenges, recent emphasis has been placed on the re-sampling of new trajectories. In this study, we propose a simulation protocol that achieves efficient sampling by re-engineering the widely used Bennett acceptance ratio (BAR) method as a representative approach. We tested its efficacy across various membrane protein targets, including G-protein coupled receptors (GPCRs) with diverse structural landscapes and experimentally validated binding affinities, to verify its efficient applicability. Subsequently, using BAR-based binding free energy calculations, we confirmed correlations with experimental data, demonstrating the validity and performance of this computational approach.

Fig. 1. Structural representation of the agonist-bound β₁AR complex. (A) The superimposition of the active state (gray color) of the β₁AR-nanobody complex with four agonists bound and corresponding inactive state (green color), as depicted in panel B. The structure of the nanobody in the active state is highlighted in black. (B) The structures of four ligands co-crystallized with eight β₁ARs. (C) Linear correlation between the computational and experimental results of the binding free energies for four agonist-bound inactive (L) and active (H) states of β₁AR. See ESI Table S5 for numerical data on BAR binding free energies.

Fig. 2. Polar and nonpolar interactions in agonists bound to each β₁AR in the inactive and active states, respectively. For the interaction analysis, the residues within the distance of 3.9 Å from the ligand and over the 40% of an interaction population in the entire MD trajectory were considered. Van der Waals contacts are depicted with light blue rays and hydrogen bonds were highlighted as red dotted lines with favorable geometry.

Fig. 3. Comparison of binding modes of PSB36 bound to A₁AR and A_2AAR. (A) Superposition of PSB36 bound A₁AR (blue ribbon) and A_2AAR (pink ribbon). The PSB36 ligand in A₁AR is represented with cyan carbon atoms, while the same ligand in A_2AAR is shown with orange carbon atoms. (B) 2D structure of the PSB36 used in this study. (C) Competition binding experiments for wild-type and mutated A2A and A1 receptor bound PSB36. (D)–(E) Comparison of competition binding experiments (pK_i) and binding free energies (ΔG) of the wild type and T270M (A₁AR)/M270T (A_2AAR) mutant for A₁AR and A_2AAR. In particular, a detailed comparison of experimental and computational binding affinities is presented in (E), with experimental and computational pK_i predictions highlighted in blue and yellow colors. See ESI Table S6 for numerical data on BAR binding free energies.

Fig. 4. Comparison of ligand-binding sites in the A₁AR crystal structure and A₃AR obtained by Alphafold2 prediction. (A) Overview of the A₁AR in complex with DU172 crystal structure. (B) and (C) Superimposition of backbone ribbons of the A₁AR structure in complex with PSB36 with A₃AR structure ((B) side view, (C) extracellular view).

Fig. 5. Molecular basis of adenosine receptors and binding free energy calculations in A₁AR and A₃AR. (A) and (B) Correlation between calculated binding free energies of A₁AR (A) and A₃AR ligands (B) and experimental K_i values. The coefficient of determination (R²) is shown in the figures as red. (C) and (D) Representative binding mode structures of the adenosine receptors bound to #18 in A₁AR (C) and A₃AR (D). See Table 1 for numerical data on BAR binding free energies.

Fig. 6. Evaluation of the H264–E169 salt bridge using molecular dynamics followed by binding free energy calculation. (A) Difference in the orthosteric site accessibility between the A_2AAR crystal structure bound to ZMA and a SAR structure (25d) used in this study for BAR free energy calculation. The SAR structure (25d) with the salt bridge formed and a 3PWH crystal structure without a salt bridge were colored in cyan and magenta, respectively. (B) Measurement of the minimum distance between the nitrogen atoms in the sidechain of H264 and the oxygen atoms in the sidechain of E169 for all SAR data. (C) Comparison of experimental binding affinity (K_i) with calculated BAR binding free energy results from the SAR data used, and relative energy stabilities (ΔΔG) from the 25d compound. (D) Linear correlation between experimental pK_i and calculated binding free energy ΔG for a total of 9 SAR data.