Impact of the Halogen PB Radii in the Estimation of Protein-Ligand Binding Energies Using MM-PBSA Calculations

Abstract

Halogenation is a widely used strategy in drug design, not only to improve ADME properties but also because halogens can engage in halogen bonds (XBs) with biological targets. To predict protein--ligand binding free energies (ΔG_bind), paramount in drug discovery, MM-PBSA remains a popular intermediate approach between less accurate docking and more rigorous free-energy methods. However, the use of extra-points of charge (EPs) to describe halogen anisotropy and enable XB sampling in MM-PBSA calculations is still uncommon. Optimized halogen radii (r_opt) for PBSA calculations, which are compatible with EPs and reproduce hydration free energies with good accuracy, were recently developed. Yet, their impact on protein--ligand binding free energy calculations has not been systematically assessed. Here, we evaluate the performance of r_opt in estimating ΔG_bind values for three sets of halogenated inhibitors of casein kinase-2 with available experimental binding data. We compared three EP models and standard RESP charges (no EPs), while also analyzing the effect of the internal dielectric constant ( ε_in) and sampling time. Our results show that direct use of the X-ray structures generally leads to poor correlations, whereas relaxation through MM minimization, particularly with ε_in, yields substantially improved agreement with experiment. Incorporating configurational sampling via MD further enhances the correlations, though Pearson coefficients varied notably with both sampling length and PBSA setup. Interestingly, longer MD sampling did not consistently improve correlations, highlighting the sensitivity of MM-PBSA to the simulation conditions. Nonetheless, optimized radii (r_opt) provided slight but systematic improvements, while inclusion of halogen anisotropy through EPs considerably increased correlations in two of the three ligand sets studied. The choice of ε_in strongly influenced the quality of predictions: the use of EPs was sufficient to mimic polarizability in systems with multiple halogens, but for ligands with fewer halogens, a higher ε_in proved beneficial. To our knowledge, this is the first comprehensive benchmark of MM-PBSA calculations using EP-based simulations, and our findings provide practical guidelines on how to best combine EPs, optimized radii, dielectric constants, and sampling strategies for an improved description of halogen bonding in protein--ligand complexes.