Effect of atomic charge, solvation, entropy, and ligand protonation state on MM‐PB(GB)SA binding energies of HIV protease

Abstract

The molecular mechanics-Poisson-Boltzmann surface area (MM-PBSA) and MM-generalized-Born surface area (MM-GBSA) approaches are commonly used in molecular modeling and drug design. Four critical aspects of these approaches have been investigated for their effect on calculated binding energies: (1) the atomic partial charge method used to parameterize the ligand force field, (2) the method used to calculate the solvation free energy, (3) inclusion of entropy estimates, and (4) the protonation state of the ligand. HIV protease has been used as a test case with six structurally different inhibitors covering a broad range of binding strength to assess the effect of these four parameters. Atomic charge methods are demonstrated to effect both the molecular dynamics (MD) simulation and MM-PB(GB)SA binding energy calculation, with a greater effect on the MD simulation. Coefficients of determination and Spearman rank coefficients were used to quantify the performance of the MM-PB(GB)SA methods relative to the experimental data. In general, better performance was achieved using (i) atomic charge models that produced smaller mean absolute atomic charges (Gasteiger, HF/STO-3G and B3LYP/cc-pVTZ), (ii) the MM-GBSA approach over MM-PBSA, while (iii) inclusion of entropy had a slightly positive effect on correlations with experiment. Accurate representation of the ligand protonation state was found to be important. It is demonstrated that these approaches can distinguish ligands according to binding strength, underlining the usefulness of these approaches in computer-aided drug design.

Computing Clinically Relevant Binding Free Energies of HIV-1 Protease Inhibitors

“…Gaussian 98 38 was used to perform geometric optimization of the inhibitor with 6-31G** basis functions, and the restrained electrostatic potential (RESP) procedure, also part of the AMBER package, was used to calculate the partial atomic charges. All ligands were found to be neutral except for Indinavir, which was protonated at pH 5, in agreement with the previous findings of Oehme et al 15 (structures for all nine inhibitors are shown in Figure 1). Before the production simulations reported here were run, all systems were minimized and equilibrated for 2 ns using the protocol defined by the BAC.…”

“…13−15 Here, we use ensembles of 50 simulations to determine the variability of the results produced by these analyses and how they depend upon the replica length and rate at which conformational samples are taken from the MD trajectories. This strategy is enabled by the increasing availability of supercomputing resources with many thousands of cores (sometimes referred to as petascale resources) which make the execution of many large simulations feasible.…”

“…The recent work of Oehme et al 15 examined the ability of MMPBSA and MMGBSA to rank a selection of HIV-1 protease inhibitors (three drugs are common to their work and the present study) using duplicate simulations with similar trajectory lengths to our individual replicas. The range of experimental Δ G values they attempted to evaluate was approximately double that of the difference between DRV and IDV in our chosen test set.…”

“…While the convergence and applicability of MMPBSA and MMGBSA have been widely studied 13,14 the computational cost of normal-mode analysis has meant that only recently has its performance begun to be similarly investigated. 15 Here, we present an extensive analysis of the convergence properties and ability to reproduce experimental values of the methods applied to HIV-1 protease binding to the nine FDA-approved inhibitors. These results build on our previous work that used these techniques to evaluate differences in binding strength of different protein sequences to individual drugs in both the HIV protease 16,17 and the anticancer therapy target epidermal growth factor receptor (EGFR).…”

Structural Study of a Flexible Active Site Loop in Human Indoleamine 2,3-Dioxygenase and Its Functional Implications

“…This kind of approach typically yields overestimated binding free energy values, but they are very useful for extracting comparative trends. 62 The results, presented in Table 1, show that the Trp binding energy is doubled in the closed loop conformation (most populated state in Cf1) compared to the open-extended loop conformation observed in Cf2. Interestingly, in the open-extended loop conformation, the interaction with Thr379 is weakend and the interaction with Arg231 is almost lost.…”