Articles pubs.acs.org/acschemicalbiology
Accelerated Discovery of Novel Benzodiazepine Ligands by Experiment-Guided Virtual Screening Simon J. Middendorp,† Roshan Puthenkalam,‡ Roland Baur,† Margot Ernst,‡,§ and Erwin Sigel*,†,§ †
Institute of Biochemistry and Molecular Medicine, University of Bern, CH-3012 Bern, Switzerland Department of Molecular Neurosciences, Center for Brain Research, Medical University of Vienna, A-1090 Vienna, Austria
‡
S Supporting Information *
ABSTRACT: High throughput discovery of ligand scaffolds for target proteins can accelerate development of leads and drug candidates enormously. Here we describe an innovative workflow for the discovery of high affinity ligands for the benzodiazepine-binding site on the so far not crystallized mammalian GABAA receptors. The procedure includes chemical biology techniques that may be generally applied to other proteins. Prerequisites are a ligand that can be chemically modified with cysteine-reactive groups, knowledge of amino acid residues contributing to the drug-binding pocket, and crystal structures either of proteins homologous to the target protein or, better, of the target itself. Part of the protocol is virtual screening that without additional rounds of optimization in many cases results only in low affinity ligands, even when a target protein has been crystallized. Here we show how the integration of functional data into structure-based screening dramatically improves the performance of the virtual screening. Thus, lead compounds with 14 different scaffolds were identified on the basis of an updated structural model of the diazepam-bound state of the GABAA receptor. Some of these compounds show considerable preference for the α3β2γ2 GABAA receptor subtype.
L
binding site for benzodiazepines have resulted in conflicting data (ref 13 and references therein).
igand-based and structure-based virtual screening both have been used to identify novel ligands either similar to known ligands or such as to fit a binding pocket in a crystallized target protein.1,2 Both types of screening suffer from the drawback that in many but not all cases they deliver only relatively low affinity hits, at least without optimization procedures or availability of a receptor structure with a cocrystallized ligand. This has been highlighted in two recent articles.1,2 We report here an attempt to structurally understand a drug-binding pocket in a non-crystallized protein and the use of an innovative procedure to predict high affinity ligands for this site. In essence it is a structure-based screening with a strong chemical biology input. It resulted in numerous high affinity lead compounds with as many as 14 different scaffolds. This experiment-guided virtual screening (EGVS) approach was developed in the process of investigating the positioning of ligands in the binding pocket for benzodiazepines in α1β2γ2 GABAA receptors. GABAA receptors are the major inhibitory receptors in the mammalian brain. Five homologous subunits surround a chloride-selective ion channel (for review see refs 3−6). Subunit composition and arrangement determine the properties of the receptors.7,8 The major adult isoform α1β2γ2 has the subunit arrangement γβαβα,9−11 and the subunit interface αγ harbors a high affinity binding site for benzodiazepines.12 These popular drugs have sedative, hypnotic, anxiolytic, anticonvulsive, and muscle relaxant properties. So far, attempts to elucidate the structure of the © 2014 American Chemical Society
■
RESULTS AND DISCUSSION Relative Positioning of Diazepam in the Benzodiazepine-Binding Pocket. We attempted to establish the relative positioning of benzodiazepines in their binding pocket on GABAA receptors. In the course of this work, we discovered that the combination of a biochemical approach with homology modeling performed remarkably well for the discovery of lead compounds. Figure 1 summarizes the procedure for the identification of novel high affinity ligands. As a first step, proximity accelerated chemical coupling reaction (PACCR) was used (reviewed by Foucaud et al.14). PACCR is ultimately derived from a method used for the investigation of water-accessibility of amino acid residues.15 Briefly, point mutations to cysteine were introduced in a total of 34 residues thought to reside in or near the binding pocket for benzodiazepines in α1β2γ2 GABAA receptors.12 Benzodiazepine ligands were modified with a cysteine-reactive -NCS group. Point-mutated receptors were combined with reactive ligands, and apposition of the cysteine with the -NCS group enables PACCR (Figure 2A). Thus, the observation of PACCR implies a short distance between the corresponding Received: March 10, 2014 Accepted: June 24, 2014 Published: June 24, 2014 1854
dx.doi.org/10.1021/cb5001873 | ACS Chem. Biol. 2014, 9, 1854−1859
ACS Chemical Biology
Articles
two different methods, radioligand binding assay (binding level) and two electrode voltage clamp experiments (functional level). At the level of radioactive ligand binding, covalent reaction leads to an occlusion of the binding site for reversible ligands and at the level of function, this leads to a permanent alteration of the latter. Figure 2B shows the structures of the used NCS derivatives and summarizes the eight covalent reactions identified so far. All point-mutated receptors leading to a covalent reaction retained high affinity to [3H]-Ro15-1788 and did not show a large change in the EC50 for GABA to open the channel. Therefore, it may be assumed that the protein structure is not strongly affected by the mutations. Compatibility with Poses of Diazepam Docked into Homology Models. In the next step we searched for structural models of the diazepam-bound pocket that are consistent with the covalent reactions we observed. One hundred diazepam docking poses were prepared for each of six homology models, based on AChBP16 and the nAChR.17 Each has about 19% sequence identity with the modeled subunits. Homology modeling and docking were performed as described earlier.13 Later we investigated also the recently published glutamate-gated Cl− channel.18 Results were consistently poor with the new template. While this protein has a high total sequence similarity to receptor subunits, the local similarity is obviously not convincing for the benzodiazepine-binding site. As indicated above, observation of PACCR restricts the distance between the mutated amino acid residue of the receptor and the reactive NCS group of the ligand. The eight distances predicted in each of the 600 poses were calculated with an svl script using MOE19 and scored for short distances as indicated in Supplementary Table S-1. The sum of the score was determined for each docking pose, and the 30 best rated poses were evaluated further. All of these poses based on eight covalent reactions shared a common gross binding mode and were quite similar to each other. Past use of three covalent reactions only (reactions 1, 3, and 4) resulted in a pool of poses representing two gross binding modes (called CBM I and II).13 These three reactions partly of neighboring residues only defined a line in space, leaving rotational freedom to the ligand position. A number of reactions defining the three-dimensional position of the ligand are thus mandatory for obtaining unique pose ranking. None of the 30 best rated poses based on the larger pool of reactions represented the binding mode CBMI proposed,13 indicating that the latter is not compatible with our observations (Supplementary Table S-2). Specifically, reactions 2, 5, 7, and 8 cannot be explained with the previously favored model. Three poses out of these 30 similar poses were selected by the following criteria and further evaluated: First, they were grouped into six structurally very closely related families of poses; then either one or no pose from each family was selected on the basis of good distance score, molecular interactions (hydrogen bridges, hydrophobic and aromatic interactions (arene−arene or H−arene)) between ligand, and receptor predicted by MOE19 and LigandScout;20 and last the performance of these poses was assessed in a validation screen of a substance library of 2400 compounds containing 20 positive allosteric modulators. One of these poses is illustrated in Figure 3. Figure 3A shows the front view of this diazepam docking pose. Figure 3B and C show the same pose turned by 90° to the left and down, respectively. Figure 3D shows the interactions between diazepam and receptor as detected by LigandScout in the same pose.
Figure 1. Scheme of the experiment-guided virtual screening (EGVS) leading to identification of high affinity ligands. We define lead compounds as substances of different scaffold with an affinity of
