In silico identification and assessment of novel allosteric protein binding sites to expand the “druggable” human proteome

Abstract

Throughout the last years there has been a considerable number of drugs that were discovered thanks to computer aided drug design (CADD) techniques. Using the 3D information, such as protein structures obtained by X-ray crystallography or nuclear magnetic resonance (NMR), it is possible to identify the binding sites and to design molecules that may specifically target these sites. This approach saves a lot of time and money, as the lead search is more accurate: less compounds need to be synthesised and tested. Although a great number of proteins have been successfully targeted with this structure-based approach, there are a lot of disease-linked proteins that have been considered “undruggable” by conventional structure-based techniques. This is mainly due to failure in detection of potential binding sites, which precludes the structure-guided design of suitable ligands. There is the presumption that the “druggable” human proteome may be larger than previously expected. Protein structures may present multiple binding sites (allosteric and/or cryptic) that cannot be targeted by the means of conventional CADD techniques. In the past years, several novel methods have been developed to identify and/or unveil these binding hotspots. Amongst them cosolvent Molecular Dynamics (MD) simulations are increasingly popular techniques developed for prediction and characterisation of allosteric and cryptic binding sites, which can be rendered “druggable” by small molecule ligands. Despite their conceptual simplicity and effectiveness, the analysis of cosolvent MD trajectories relies on pocket volume data, which requires a high level of manual investigation and may introduce a bias. The present study focused on the development of the novel cosolvent analysis toolkit (denoted as CAT), as an open-source, freely accessible analytical tool, suitable for automated analysis of cosolvent MD trajectories. CAT is compatible with popular molecular graphics software packages such as UCSF Chimera and VMD. Using a novel hybrid empirical force field scoring function, CAT accurately ranked the dynamic interactions between the macromolecular target and cosolvent molecular probes. Alongside the development of CAT, this work investigated the signal transducer activator of transcription 3 (STAT3) as the case study. STAT3 is among the most investigated oncogenic transcription factors, as it is highly associated with cancer initiation, progression, metastasis, chemoresistance, and immune evasion. Constitutive activation of STAT3 by mutations occurs frequently in tumour cells, and directly contributes to many malignant phenotypes. The evidence from both preclinical and clinical studies have demonstrated that STAT3 plays a critical role in several malignancies associated with poor prognosis such as glioblastoma and triple-negative breast cancer (TNBC), and STAT3 inhibitors have shown efficacy in inhibiting cancer growth and metastasis. Unfortunately, detailed structural biology studies on STAT3 as well as target-based drug discovery efforts have been hampered by difficulties in the expression and purification of the full length STAT3 and a lack of ligand-bound crystal structures. Considering these, computational methods offer an attractive strategy for the assessment of “druggability” of STAT3 dimers and allow investigations of reported activating and inhibiting STAT3 mutants at the atomistic level of detail. This work studied effects exerted by reported STAT3 mutations on the protein structure, dynamics, DNA binding and dimerisation, thus linking structure, dynamics, energetics, and the biological function. By employing a combination of equilibrium molecular dynamics (MD) and umbrella sampling (US) simulations to a series of human STAT3 dimers, which comprised wild-type protein and four mutations; the work presented herein explains the modulation of STAT3 activity by these mutations. The binding sites were mapped by the combination of MD simulations, molecular docking, and CAT analysis, and the binding mode of a clinical candidate napabucasin/BBI-608 at STAT3, which resembles the effect of D570K mutation, has been characterised. Collectively the results of this study demonstrate the robustness of the newly developed CAT methodology and its applicability in computational studies aiming at identification of protein “hotspots” in a wide range of protein targets, including the challenging ones. This work contributes to understanding the activation/inhibition mechanism of STAT3, and it explains the molecular mechanism of STAT3 inhibition by BBI-608. Alongside the characterisation of the BBI-608 binding mode, a novel binding site amenable to bind small molecule v ligands has been discovered in this work, which may pave the way to design novel STAT3 inhibitors and to suggest new strategies for pharmacological intervention to combat cancers associated with poor prognosis. It is expected that the results presented in this dissertation will contribute to an increase of the size of the potentially “druggable” human proteome.