Predicting GPCR Promiscuity Using Binding Site Features Anat Levit, †,‡ Thijs Beuming, § Goran Krilov, § Woody Sherman, § and Masha Y. Niv* ,†,‡ † Institute of Biochemistry, Food Science and Nutrition, Robert H. Smith Faculty of Agriculture Food and Environment, The Hebrew University, Rehovot 76100, Israel ‡ Fritz Haber Center for Molecular Dynamics, The Hebrew University, Jerusalem 91904, Israel § Schrodinger Inc., 120 West Forty-Fifth Street, 17th Floor, New York, New York 10036, United States * S Supporting Information ABSTRACT: G protein-coupled receptors (GPCRs) repre- sent a large family of signaling proteins that includes many therapeutic targets. GPCR ligands include odorants, tastants, and neurotransmitters and vary in size and properties. Dramatic chemical diversity may occur even among ligands of the same receptor. Our goal is to unravel the structural and chemical features that determine GPCRs’ promiscuity toward their ligands. We perform statistical analysis using more than 30 descriptors related to the sequence, physicochemical, structural, and energetic properties of the GPCR binding siteswe find that the chemical variability of antagonists significantly correlates with the binding site hydrophobicity and anticorrelates with the number of hydrogen bond donors in the binding site. The number of disulfide bridges in the extracellular region of a receptor anticorrelates with the range of molecular weights of its antagonists, highlighting the role of the entrance pathway in determining the size selectivity for GPCR antagonists. The predictive capability of the model is successfully validated using a separate set of GPCRs, using either X-ray structures or homology models. ■ INTRODUCTION G protein-coupled receptors (GPCRs) form the largest family of cell surface receptors in the human genome. 1 They are key signaling molecules and are the targets of over 30% of currently approved and marketed drugs. 2 Endogenous agonists of GPCRs include bioamines, nucleotides, neurotransmitters, peptides, and many other chemical stimuli. Some GPCRs are narrowly tuned toward their agonists, such as the pheromone receptors 3 and subsets of the olfactory receptor subfamily, 4 while others have a broad receptive range. Striking examples of broadly tuned receptors include some olfactory receptors 5,6 and several bitter taste receptors, 7,8 in which a single receptor recognizes a broad range of ligands. A recent systematic analysis of ligand-contacting residues in the transmembrane (TM) ligand-binding pocket of GPCR X- ray structures has revealed that, except for the C-X-C chemokine type 4 receptor (CXCR4) and the neurotensin 1 receptor (NTSR1), all other X-ray structures of Family A GPCRs share similarity in ligand-contacting residues, with topologically equivalent positions in TM3, TM6, and TM7 typically contacting the ligand in nearly all receptors. 9 Variation in the amino acids occupying these positions accounts for ligand specificity in different receptors. Furthermore, individual GPCRs may accommodate diverse chemical matter by utilizing different subsets of binding site residues, 8,10 as well as different types of interactions (i.e., polar vs nonpolar). 8 These differences can be identified by crystallography 11,12 or via a combination of modeling and mutagenesis studies 8,10,13 on a case-by-case basis. Here, we attempt to address whether the receptive range of a receptor can be predicted based on the physicochemical properties of the binding site. Recently developed methods, such as ligand’s eye-view of protein similarity, 14,15 as well as chemoproteometric 16 or chemogenomic approaches, 17,18 analyze and predict the relationship between proteins and the ligands they bind based on either ligand similarities or both protein and ligand similarity information. For example, agonist-specific regions were shown to concentrate between TMs 2, 3, and the second extracellular loop (ECL2), while antagonist-specific regions are located at the top of TMs 5 and 6. 18 A ligand-based view of promiscuity can suggest ligand properties that determine its polypharmacol- ogy (the number of targets with which the ligand interacts). Studies have found that the most promiscuous drugs tend to be highly hydrophobic (clog P ≥ 3). The relation between ligand size (in terms of molecular weight (MW)) and its promiscuity has also been studied, but no consensus was reached. 19 In some cases, an inverse correlation between mean MW and ligand promiscuity toward targets was found, while another study showed that within a given clog P range, promiscuity decreases with increasing ligand size. A recent study 20 on a large set of over 40 000 molecules for which at least three measured affinities (pXC 50 ≥ 6) were available in ChEMBL did not find Received: September 25, 2013 Article pubs.acs.org/jcim © XXXX American Chemical Society A dx.doi.org/10.1021/ci400552z | J. Chem. Inf. Model. XXXX, XXX, XXX-XXX