Combining in silico and in vitro approaches to discover novel inhibitors of Acetylcholinesterase

Abstract

Alzheimer’s disease (AD) is the most common neurodegenerative disease which is a significant socio-economic problem. The number of patients affected by the disease is increasing at an alarming rate, largely due to expanding population sizes and longer life expectancy. While significant amounts of research into AD have been conducted the cause and pathogenesis of the disease are not well understood, with several hypotheses being noted in literature. To date, four drugs have been approved by the FDA, but these compounds only provide symptomatic relief. This study describes the uses of computer-aided drug discovery (CADD) techniques to identify novel inhibitors of Acetylcholinesterase (AChE), a target for AD. High throughput virtual screening (HTVS) was employed to predict potential inhibitors of AChE – an approach which, due to the associated difficulties of modelling the enzyme has to date not been reported to be successful in literature. Validation of enrichment was performed with the “Directory of Useful Decoys, enhanced” DUD-E dataset, showing that an ensemble of binding pocket conformations is critical when a diverse set of ligands are being screened. HTVS of a library of 20 000 compounds was performed. Cross-validation of the model was conducted by in vitro screening of 720 compounds, which led to 25 hits being identified with IC50 values of less than 50 μM. The majority of these hits belonged to two scaffolds: 1-ethyl-3-methoxy-3-methylpyrrolidine and 1H-pyrrolo[3,2-c]pyridin-6-amine, the latter being found through serendipity. Both scaffolds were noted to be promising compounds for further optimisation. Computational analysis of the active hits were performed to gain a deeper understanding of the binding pose to AChE. As various possible binding poses were suggested from molecular docking, molecular dynamic (MD) simulations were employed to validate the poses. In the case of the most active compounds identified, a critical, stable water bridge formed deep within the pocket. This, in part, explains the lack of activity for subsets of compounds that are not able to form this critical water bridge. The pKa analysis of AChE inhibitors showed a preference for pKa values higher than physiological pH leading to the ligands being cations and allowing the inhibitor to better mimic the substrate of AChE. Implications of using pKa as a guideline to improve potency and selectivity for AChE inhibitors are discussed. Further development of the docking protocol was performed with the use of a popular machine learning approach, Random Forest (RF). The approach is largely based on SIEVE-Score which takes the interaction energies between the ligand and residues in the pocket into account. By employing an ensemble of receptor conformations, significant enrichment over previous studies was obtained. Finally, additional secondary projects are reported which cover the computational analysis of compounds synthesised within the research group which inhibit either AChE or β-secretase (BACE1). Analysis of the interactions that inhibitors make with residues in the binding pocket allowed for an improved understanding of the system for future lead-optimisation studies.