Computational Approaches for the identification of potential inhibitors of the AcrAB-TolC Multidrug Efflux Pump

Antibiotic resistance is a major burden constituting a global public health hazard; it occurs when bacteria no longer respond to antibacterial medicines. It poses a substantial strain on both the healthcare system and the economy worldwide; researchers have identified its driving factor to be the widespread phenomenon of antibiotic overuse. Development of new antibacterial agents is therefore needed to overcome the spread of bacterial strains resistant to currently available antibiotics. Mechanisms of antibiotic resistance include both enzymatic and non-enzymatic paths. This thesis project focuses on the modulation of a non-enzymatic mechanism, specifically on the inhibition of a resistance-nodulation-division (RND) family efflux pump through a Computer-Aided Drug Discovery (CADD) approach. In particular, the target of this study is a tripartite protein complex named AcrAB-TolC and present in Gram-negative Enterobacteriaceae such as Escherichia Coli. Overexpression of RND efflux pumps is a key factor in mediating antibiotic resistance due to their peculiar substrate polyspecificity; thus inhibiting efflux pumps in a fight against antibiotic resistance has been a crucial research topic ever since they were discovered in the 1990s. Drugs acting as inhibitors of this protein superfamily reduce the extrusion of antibiotics, giving them the chance to exert their pharmacological action in situ. Several efflux pump inhibitors (EPIs) have been developed; however, none have reached the clinic stage yet. This thesis aims to identify potential AcrAB-TolC inhibitors through molecular modelling techniques such as molecular docking-based virtual screening, performed selectively on the inner membrane protein of the complex, AcrB, both in its transmembrane and periplasmic regions. Experimentally resolved AcrB structures were retrieved from the Protein Data Bank and categorized based on their co-crystallized ligand, focusing on high-resolution structures presenting existing EPIs as ligands. Selected entries were employed in self-docking and cross-docking benchmarks to evaluate the ability of both protein structures and molecular docking protocols (e.g.: PLANTS, GOLD) and their scoring functions to reproduce binding poses of crystallographic inhibitors. The best performing PDB structures – with lowest average root-mean-square-deviation (RMSD) values of predicted vs. experimental ligand poses – were selected. The combination of best performing structures and docking protocols were then used in the subsequent virtual screening. The initial screening protocol focused on a library of FDA-approved molecules pursuing a “drug repurposing” strategy, that is the employment of existing medications beyond typical clinical indications. The experimental structures of AcrB-ligand complexes were used to build pharmacophore- and interaction-based filters for the virtual screening output. Hit candidates were selected based on patterns of interaction with the protein through computed van der Waals, electrostatic, hydrophobic ligand-target interactions, pharmacophore filters and protein-ligand interaction fingerprints built on X-ray inhibitors. Lastly, stability of protein-ligand complexes was assessed for few selected compounds through molecular dynamics simulations leading to the selection of a subset of compounds for future experimental studies.