Computational Tools for Allosteric Drug Discovery: Site Identification and Focus Library Design

Abstract

Allostery is an intrinsic phenomenon of biological macromolecules involving regulation and/or signal transduction induced by a ligand binding to an allosteric site distinct from a molecule’s active site. Allosteric drugs are currently receiving increased attention in drug discovery because drugs that target allosteric sites can provide important advantages over the corresponding orthosteric drugs including specific subtype selectivity within receptor families. Consequently, targeting allosteric sites, instead of orthosteric sites, can reduce drug-related side effects and toxicity. On the down side, allosteric drug discovery can be more challenging than traditional orthosteric drug discovery due to difficulties associated with determining the locations of allosteric sites and designing drugs based on these sites and the need for the allosteric effects to propagate through the structure, reach the ligand binding site and elicit a conformational change. In this study, we present computational tools ranging from the identification of potential allosteric sites to the design of “allosteric-like” modulator libraries. These tools may be particularly useful for allosteric drug discovery.

1. Introduction

Allostery, which is also known as allosteric regulation, is an essential biological phenomenon that plays significant roles in signal transduction pathways, metabolic processes, and genomic transcription [1, 2]. Perturbation at an allosteric site can rapidly shift the equilibrium of a protein conformational ensemble towards another state, thereby inducing local conformation change at an active site [3–5]. Potential perturbations include the binding of small molecules/ions and local chemical modifications [6–8]. Thus, allostery is the most direct mechanism for regulating the function of biological macromolecules. Insight into allostery can lead to new ideas for method development in allosteric drug discovery [9, 10].

Unlike orthosteric drugs, which compete with the substrates of target proteins at the active sites, allosteric drugs bind at a location other than an active site and influence the affinity or catalytic efficiency of biological macromolecules through the propagation of a perturbation signal [11–14]. Allosteric drugs have several advantages relative to orthosteric drugs. First, according to sequence conservation analyses [15, 16], allosteric sites are significantly less conserved than orthosteric sites; this phenomenon allows allosteric modulators to selectively target specific subtypes within receptor families [17, 18], resulting in higher selectivity and fewer side effects than orthosteric drugs. Second, allosteric drugs do not block substrate-protein interactions, and there is an upper bound to allosteric regulation. In addition, allosteric modulators can enhance the efficiency of orthosteric drugs [19]. For instance, the allosteric modulator GNF-2 binds to the myristate-binding site of T315I human Bcr-Abl. GNF-2 and the substrate-competitive inhibitor imatinib exhibit additive inhibitory activity against this mutated Bcr-Abl; as a result, a combination of these two drugs can be used to overcome drug resistance in cases of chronic myelogenous leukemia (CML) [20]. Thus, the identification of modulators targeting allosteric sites receives increasing attention in the field of drug discovery, and several allosteric drugs have been approved by the US FDA [21, 22]. For example, Genzyme’s plerixafor is an allosteric antagonist of the C-X-C chemokine receptor type 4 (CXCR4) that enhances the mobilization of hematopoietic stem cells (HSCs).

Allosteric drug discovery also presents new challenges relative to traditional drug discovery approaches. The identification and characterization of drug binding sites is the first step of structure-based drug discovery. However, the locations of allosteric sites remain unclear for most drug targets [23]. Moreover, the discovery of allosteric modulators is hampered by several obstacles, such as the low affinities and unknown structural features of potential small allosteric molecules. In our prior work, we summarized the properties of allosteric sites [24] and allosteric modulators [25] and developed an allosteric site identification method named Allosite [26]. In addition, a preliminary filter for allosteric modulator discovery was also established. In this protocol, we introduce practical guidelines describing how to obtain predictions for allosteric sites and build a focused library of “allosteric-like” molecules.

1.1. Theory

The fundamental strategy for Allosite is to use the topology and physiochemical properties of protein pockets to build a classification model relating allosteric sites to other sites. We have extracted 90 nonredundant allosteric protein–allosteric modulator co-crystals from the AlloSteric Database [27, 28]. After feature selection, 21 pocket descriptors were characterized for each pocket identified by FPocket [29]. The classification model for allosteric site identification was then trained and tested using a support vector machine [30]. In a cross-validation test, the success metrics for the Support Vector Machine (SVM) model were a sensitivity of ~83 % and a specificity of ~96 %. We have made the final model available on the Allosite Web server.

To reveal the structural specificity of allosteric modulators, 3916 known structurally diverse allosteric modulators in the Allosteric Database were compared with compounds from other databases (the Accelrys Available Chemicals Directory, the Accelrys Comprehensive Medicinal Chemistry database, the Chinese Natural Product Database, DrugBank, the MDDR database, and the NCI Open Database). Interestingly, relative to other modulators, allosteric modulators exhibit higher structure rigidity, with less rotatable bonds and more rings from ring systems. In addition, higher hydrophobicity is also observed for allosteric modulators; this finding is consistent with the hydrophobic characteristics of allosteric sites [25]. In summary, we established the following rule for differentiating allosteric modulators from other modulators: (1) molecular weight (MW) ≤ 600; (2) number of rotatable bonds (nRB) ≤ 6; (3) 2 ≤ number of rings (nR) ≤ 5; (4) number of rings in the largest ring Systems (nRIS) = 1 or 2; and (5) 3 ≤ SlogP ≤ 7.

2. Materials

2.1. Software for Visualizing Protein Structures

The PyMOL molecular graphics system is required for visualizing PDB files and allosteric sites. This system, which is a open-source software, is available at http://www.pymol.org.

2.2. Browser

The Allosite server requires a Web browser with JavaScript and cookies enabled. A recommendation to ensure that protein structures can be visualized correctly is to use the latest version of Firefox or Chrome to access Allosite.

3. Methods

In the following subsections, we first describe the individual steps that the Allosite server uses to identify allosteric sites and then describe how to construct focused libraries for the screening of allosteric modulators with a preliminary filter.

3.1. Input File Preparation

Allosite utilizes a method based on the proteins’ three-dimensional structures, which can be obtained from the Protein Data Bank database (see Note 1). If there is no crystal structure for the query protein, homology modeling methods will be helpful for building the protein’s 3D structure. The following considerations should be taken into account to ensure the quality of the prediction. (1) We recommend using an X-ray structure with a resolution <2.5 Å. (2) There should be no missing loops in the main chain of the protein. (3) Small molecules, ions and solvents within the PDB structure will automatically be removed.

3.2. Job Submission

The Allosite Web server is freely available for use at http://mdl.shsmu.edu.cn/AST. Jobs can be submitted either by “PDB ID” or by “PDB File” (see Fig. 1). In “PDB ID” mode, users can specify their input by simply entering the 4-character PDB ID of their query protein. Users with their own experimental/model-based structures can choose the “PDB File” mode to browse their local hard drives and provide a protein structure file. A submitted query protein structure should be in standard PDB format. Another parameter, “Job Name”, must be set before running the job; this parameter can then be used to check the status of the job and retrieve calculated results for the job at any time. After “Job Name” has been specified, users can click “Run” and select PDB chain(s) to submit the job (see Note 2).

Fig. 1.

The Web interface and workflow of Allosite

3.3. Retrieving the Results

Once the job has been submitted, detailed job information, including a unique Job ID, will appear on the “Select PDB chains” page. Users can also track the progress of a job or access the results page from the “Job Queue” page by searching for their Job ID. The status of a job is refreshed in the “Job List” every 10 s until the “Finished” button appears. The “Finished” button indicates that the job has finished, and results can be viewed by clicking this button. The Allosite approach features rapid calculation times that depend on the size of the query protein. A typical Allosite job for a 400-residue protein will require ~15 s.

3.4. Analyzing the Results

The job will redirect to the calculation result page after the “Finished” button has been clicked. The GLmol applet will load automatically and provide a default color-coded representation of the query protein. The predicted allosteric site can be viewed in the GLmol applet by clicking the “Show Pocket” button. The predicted allosteric site is displayed as white spheres, and allosteric site residues are represented using a stick model. The result page also contains the following pocket properties for the predicted allosteric site: “Pocket Volume”, “Pocket Total SASA”, “Pocket Polar SASA”, and “Pocket Druggability Score”. A representative run of an Allosite job provides 0–4 potential allosteric sites.

3.5. Analyzing the Results Using PyMOL

Result files can be downloaded for offline analysis by clicking the “Download Report” link. After tar archives have been extracted, three files are obtained: a structure file for the query protein, site information for the predicted results, and a .pml PyMOL script for visualization. Users can then analyze the predicted allosteric site in PyMOL (see Fig. 2).

Fig. 2.

An analysis of Allosite results using PyMOL. The allosteric pocket is represented by white points. Red lines are used to highlight residues in this site

3.6. Designing Focus Libraries of Allosteric Modulators

Based on our prior work, molecules that satisfy the following criteria are more likely to be allosteric modulators: (1) MW ≤ 600; (2) RBN ≤ 6; (3) 2 ≤ nR ≤ 5; (4) nRIS = 1 or 2; and (5) 3 ≤ SlogP ≤ 7. To fetch potential allosteric modulators from a database of chemical molecules, we developed a Web server that can be accessed at http://mdl.shsmu.edu.cn/ASD/. For each job, users can upload their molecular database of interest with either 2D or 3D structures. Three file types are acceptable for uploading: MOL, SDF, and SMILES. When our Web server completes a job, users are redirected to a results page where they can click “Download” to download molecules that have passed the “allosteric-like” filter. Histograms indicate the distribution of five calculated molecular properties. The filter’s local script runs quickly and is recommended for users who intend to filter large molecular databases.