Molecular dynamics in mixed solvents as a tool for protein-ligand binding studies (aceptado para ser presentado)

MODENUTTI, CARLOS P.; BARRIL, XAVIER; ARCON, JUAN PABLO; LOPEZ, ELIAS D.; DEFELIPE, LUCAS A.; TURJANSKI, ADRIÁN G.; ALVAREZ-GARCIA, DANIEL; MARTÍ, MARCELO A.

Lausana

Congreso; SBDD2017: Computational Advances in Drug Discovery; 2017

École Polytechnique Fédérale de Lausanne - Instituto Italiano di Tecnología - Swiss Institute of Bioinformatics

Small water miscible molecules (or  cosolvents), that mimic specific types of ligand molecular interactions, have been proven useful to reveal protein−ligand interaction hot spots and thus are a relevant source of information for drug development projects. In the present work we performed molecular dynamics simulations in mixed solvents for 18 different proteins to determine solvent sites, i.e. space regions adjacent to the protein surface where the probability of finding a probe solvent atom is higher than that in the bulk solvent. The solvents used were water mixtures of ethanol, acetamide, acetonitrile and methylammonium acetate, as well as pure water. When analyzing the probe capacity to reveal known protein-ligand interactions, results showed remarkable sensitivity for water (0.64) and specificity for ethanol-OH (0.56). These values were further improved when we only considered the ligand-based pharmacophoric points (0.73 and 0.61, respectively). Hydrophobic probes, such as methyl ends of ethanol and acetonitrile, also displayed great performance. Therefore, ethanol specificity makes it a great probe to use as a pharmacophoric bias in docking experiments, with ethanol−OH sites guiding the position of ligand groups with hydrogen bonding capabilities and ethanol-CH3 sites guiding the location of ligand aromatic hydrophobic rings. Biased redocking and cross docking experiments showed 70% vs. 20% success when compared with the conventional docking method. Furthermore, for targets extracted from the DUD-E data set, ligand enrichment factors were also increased (e.g. EF1% from 2.0 to 12.2 for AmpC β-lactamase). Finally, accurate predictions of binding free energies were obtained by adding the ∆G contribution of solvent sites replaced by crystallized ligand groups capable of establishing the same type of interactions.