Virtual screening for protein function annotation and drug discovery

Abstract

Molecular docking is a computational procedure that attempts to efficiently predict non-covalent binding of macromolecules or, more frequently, of a macromolecule (receptor) and a small molecule (ligand), starting with their unbound structures. This approach is widely used in virtual screening, an in silico procedure that has become increasingly popular in the pharmaceutical research for lead identification. The basic goal of the virtual screening is the reduction of the massive virtual chemical space of small organic molecules, to be screened against a specific target protein, to a manageable number of compounds that have the highest chance to lead to a drug candidate. Virtual screening applied to the discovery of new drugs involves docking, computational fitting of structures of compounds to the active site of a protein and scoring and ranking of each compound. In this context, databases of chemical and drug-like compounds are very helpful because, in most cases, contain many compounds in ready-to-dock formats (three-dimensional structures). The present work was aimed at studying molecular docking protocols, and specifically protein-ligand docking, and its application to biological systems of biomedical relevance, such as human serum albumin (HSA), to provide complementary and supporting evidence to experimental studies. For this purpose was used the program AutoDock Vina, an open-source program for drug discovery, molecular docking and virtual screening, offering multi-core capability (or, in other words, the possibility to use at the same time all the processors of the computer used), high performance, enhanced accuracy, and ease of use. Docking simulations on HSA were carried out to investigate the interaction of this protein with the drug imatinib and with retinoids. Results obtained in these simulations are in agreement with available experimental data and allowed both to identify the preferential binding sites of the ligand/drug, and to investigate the interaction at atomic level. In addition, a user-friendly procedure was developed, which allows to perform automatic molecular docking simulations and virtual screening analyses. This method was tested on experimental structures of potential Pseudomonas aeruginosa protein targets for the identification of novel lead compounds with potential antibacterial activity. In fact, bacterial resistance is a growing threat and, at present, very few new antibiotics active against multi-resistant bacteria are available. A combination of falling profits, regulatory mechanisms and irrational and injudicious use of antibiotics has led to an alarming situation where some infections have no cure. Finding novel targets, possibly with no cross-resistance, and/or identification of novel compounds that could reduce or inhibit bacterial virulence, has elements of uncertainty but successful outcomes may translate into significant results. Given the huge number of compounds to be screened (1466 FDA-approved compounds from DrugBank), an automated procedure for docking studies was developed, using the Python programming language, which allows to calculate the input parameters to perform the docking phase, to rank the compounds according to their predicted binding energy and to visualize the predicted drug binding site on the protein of interest. This procedure has been incorporated in an application named “DockingApp”. DockingApp allows to perform this kind of analyses also to users without any expertise in bioinformatics. In fact, a Java graphical user interface (GUI) was developed to guide the user through all the steps required to perform the docking simulations and analyze the results. In particular, the input of the virtual screening/docking application is a protein with still unknown function or a target of interest from a pharmaceutical viewpoint. The user needs only the three-dimensional structure of the target (i.e. the PDB file); the input is then processed to obtain a structure with all the information needed by the docking algorithm (hydrogen atoms, atomic charges). The core of the algorithm is an automatic protein-ligand docking procedure or a virtual screening analysis against a library of compounds, in particular the subset of FDA-approved compounds retrieved from DrugBank, provided in the download folder. In addition, the user can customize the library, or use his own, simply indicating the relative folder in the GUI. The output is a list (stored in a text file) containing the best hits resulting from the screening, based on the binding affinity calculated. This file also contains some details, such as residues in contact with the ligand at different distance thresholds and details about the compounds. Obviously, in the output are stored also the complexes between the input protein and the best hit compounds with their respective atomic coordinates. To run the application the user must install MGLtools (ADT) and download Autodock Vina. In particular, MGLtools is quite useful in receptor and ligand preparation. In fact, PDB files of the receptor and the ligand must be converted in PDBQT file format, required by Vina to run docking simulations. Through the ADT graphical user interface, it is possible to obtain this conversion as well as to set the desired search space; however, for a user without any (or low) expertise in bioinformatics these steps could be difficult. To overcome this limitation, some Python scripts provided by Vina developers were used in this work, allowing the automatic preparation of PDBQT files. Regarding the search space (docking grid), that is the portion of the receptor that Vina will explore to place the ligand, a specific function was developed to allow to automatically set the search space, either local, relative to a limited number of residues, or extended to cover the entire structure of the receptor. Although tests are still in progress, the virtual screening version of the application was already tested on five protein targets from Pseudomonas aeruginosa that are involved in bacterial virulence or are essential for bacterial growth (PqsE, PqsR, LasR, WspR, MurG), with the aim to identify FDA-approved compounds with potential antibacterial activity. The results obtained are still under investigation but they are promising because in most cases the best poses of each compound are placed in the target's binding site with a better affinity if compared with the co-crystallized ligand. Summarizing, with this application it should be possible to achieve two objectives. First, starting from the information derived from docking results it may be possible to predict protein function. For enzymes of unknown function, substrate prediction based on structural complementarity becomes attractive when the target enzyme has little relationship to orthologues of known activity, making inference unreliable. From this viewpoint, the identification of few compounds that bind the target with the highest probability, that is the identification of potential substrate/inhibitor candidates, could lead to protein function annotation. Then, drug discovery and/or repositioning, that is the main goal of the present project, could be achieved. In fact, the automated virtual screening against a library of FDA-approved compounds reduces the cost of the initial experimental screening and accelerates lead compounds discovery. Moreover, this analysis can lead to drug repositioning, which aims to identify new therapeutic indications for existing drugs.