Proc. Nati. Acad. Sci. USA Vol. 86, pp. 8165-8169, October 1989 Pharmacology Searching for pharmacophores in large coordinate data bases and its use in drug design (topographic/three-dimensional substructure/coNcoRD) ROBERT P. SHERIDAN, ANDREW RuSINKO III, RAMASWAMY NILAKANTAN, AND R. VENKATARAGHAVAN Medical Research Division, Lederle Laboratories, American Cyanamid, Pearl River, NY 10965 Communicated by F. W. McLafferty, May 25, 1989 (receivedfor review February 25, 1989) ABSTRACT Pharmacophores, three-dimensional ar- available (crystal coordinates or computer-generated confor- rangements ofchemical groups essential for biological activity, mations ofa few compounds) were -mall. Now, however, the are being proposed in increasing numbers. We have developed data base ofcrystal coordinates maintained by Cambridge (6) a system to search data bases of three-dimensional coordinates has >60,000 entries. Moreover, practical methods to trans- for compounds that contain a particular pharmacophore. The form large data bases ofconnection tables to coordinate form coordinates can be derived from experiment (e.g., Cambridge (7) have recently become available. Also, computer algo- Crystal Database) or be generated from data bases of connec- rithms for doing three-dimensional substructure searches tion tables (e.g., Cyanamid Laboratories proprietary com- (8-10) have improved. pounds) via the program CONCORD. We discuss the results of Searching for pharmacophores in large data bases has now searches for three sample pharmacophores. Two have been become a practical way to exploit geometric structure- proposed by others based on the conformational analysis of activity information. In this paper, we discuss the results of active compounds, and one is inferred from the crystal struc- searching for three sample pharmacophores. Our primary ture of a protein-igand complex. These examples show that goal is to demonstrate the type of insight that can come from such searches can identify classes of compounds that are pharmacophore searches. structurally different from the compounds from which the pharmacophore was derived but are known to have the ap- METHODS propriate biological activity. Occasionally, the searches find bond "frameworks" in which the important groups are rigidly We have been working with two large data bases of coordi- held in the proper geometry. These may suggest new structural nates: CCD, a subset of the Cambridge Crystal Database (6), classes for synthesis. and CLFIL, a three-dimensional version of the CL File, the set of proprietary structures from American Cyanamid. The 29,828 structures in CCD are those for which we could relate The function of rational drug design is to relate chemical the coordinate and connection table information provided by structure to a specific biological activity and then to use the the Cambridge Crystallographic Data Centre. The 223,988 relationship to select existing compounds for testing or to structures in CLFIL were generated from a MACCS data suggest new compounds for synthesis. Now that three- base by using CONCORD. [MACCS (Molecular ACCess Sys- dimensional molecular modeling is widely used, "pharma- tem) is the tradename for a chemical data base management cophore" models for a variety of biological activities are system supplied by Molecular Design Ltd., San Leandro, appearing in the literature in increasing numbers. The con- CA.] CONCORD is a rule-based system that generates a single cept of a pharmacophore, although based on many simplify- set of three-dimensional coordinates from a connection table ing assumptions, is a useful one: a particular spatial arrange- that contains only "organic" atoms. Structures in either data ment of chemical groups (usually atoms), common to all base are stored as a set of atom types and coordinates for active molecules, that is recognized by a single receptor. nonhydrogen atoms. The atom type consists of five fields: Pharmacophores can be inferred from the structure ofligand- element, number of neighbors (i.e., bonded atoms), the receptor complexes or, when the structure of the receptor is number of pi electrons, the number of attached hydrogens at not known, can be deduced by conformational analysis of physiological pH, and formal charge. Four types of dummy active compounds, as in the "active analog" approach (1-3). atoms were added: D5 and D6 (the centroids of planar 5- and A pharmacophore is usually expressed in three-dimensional 6-membered rings), DP (along the perpendiculars to these terms (distances, angles, volumes), and its atoms may be rings), and DL (attached to each heteroatom along the vector described by a physical property (cation, hydrogen bond sum ofthe bonds from its neighbors). These are equivalent to donor, etc.). Often, a pharmacophore will contain one or the types of dummy atoms most often used in describing more "dummy" atoms, which are used to define a geometric pharmacophore geometries. Details of how the data bases entity (centroid of a ring, a lone pair direction, etc.). were generated will be given elsewhere (7). More than 10 years ago, Gund (4, 5) pointed out how, once We have developed a system 3DSEARCH (8) that allows a a pharmacophore is defined for a particular biological activ- user to define a three-dimensional substructure "query," a ity, one could gain further insight by finding structures that set of atoms and a -description of the spatial relationship contain the pharmacophore but are structurally different between them, and to search for it in large data bases of from the compounds from which the pharmacophore was coordinates. The user can specify upper and lower bounds to derived. He described one of the first methods (MOLPAT) to selected distances, angles, and dihedral angles. Excluded search for pharmacophores in data bases of three- volumes (regions that are not allowed to be occupied) can dimensional coordinates. At that time, the usefulness of such also be taken into account. We can define several possible searches was limited since the three-dimensional data bases atom types for each query atom, making it possible to specify "generic" types (cations, hydrogen-bond donors, etc). The publication costs of this article were defrayed in part by page charge payment. This article must therefore be hereby marked "advertisement" Abbreviations: CNS, central nervous system; ACE, angiotensin- in accordance with 18 U.S.C. §1734 solely to indicate this fact. converting enzyme. 8165 Downloaded by guest on October 2, 2021 8166 Pharmacology: Sheridan et al. Proc. Natl. Acad. Sci. USA 86 (1989) 3DSEARCH uses a two-part search strategy: a rapid screen descriptors defined by pairs of atoms and the distances using a key-search algorithm and a slower atom-by-atom between them. We applied an "inverted" key-indexing geometric search on the structures that pass the screen scheme, in contrast to the "direct" scheme described by (typically a few percent of the original list). The keys are others (10) using similar keys. In an inverted key scheme, the a c e93 ~~~~~111 ej 4.9 to 5.9 1 70 to 1 10 90 to 130 -v 4 0.4 to0.6 \ ) /0.9 to 1.1 3 4.5 to 5.5 1.1 1--2--3--4 -30 to -90 Protein volume represented by 122 outriggers. Atom Type Element Neighbors Pi H's Chg Atom Type Element Neighbors Pi H's Chg 1 1 DP 0 0 0 0 1 1 N 1 0 3 1 2 1 D6 0 0 0 0 2 N 2 0 2 1 3 N 3 0 1 1 3 1 N 3 0 1 1 4 N 4 0 0 1 5 N 3 1 0 1 4 1 DL 0 0 0 0 6 P 4 0 0 1 7 S 3 0 0 1 2 1 C 3 1 0 0 b 2 N WI 1 0 0 3 S Wi WI0 0 4 P Wi WI0 0 3 1 D5 0 0 0 0 2 D6 0 0 0 0 4 1 0 1 0 0 -1 2 0 1 1 0 -1 FIG. 1. Pharmacophores and queries derived from them. In the queries, all distances are given in angstroms and all angle/dihedral 5 constraints are given in degrees. The distances to dummy atoms (D) are determined by the arbitrary distance we used to generate dummy .0to 1.7 6.6 to 8.3 atoms in the coordinate data bases (ref. 7). In each case the possible allowed atom types for each of the query atoms are shown. A "Wi" ("wild card") means any value is allowed. Pi, pi electrons; H's, 3.1to5.1 2.8to4.8 attached hydrogens at physiological pH; Chg, formal charge; Neigh- bors, bonded atoms. For explanations of D5, D6, DP, and DL, see 88 to 148 \0.9 to 1.0 Methods. (a) CNS pharmacophore as proposed by Lloyd and An- drews (12). Shown on the top is the relationship of a basic tertiary 1-2--3-4 -135to-1800r 135to 180 amine to the phenyl ring in their "common model." Below is the 1--2-3--5 -90 to 90 query derived from this relationship. Atom 2 is the centroid of a flat 6-membered ring, atom 3 is the basic amine, atom 1 indicates the perpendicular to the ring, and atom 4 indicates the direction of the Atom Type Element Neighbors Pi H's Chg long pair on the amine. (b) The angiotensin-converting enzyme inhibitor pharmacophore proposed by Mayer et al. (14). A structure 1 1 S 1 o Wi Wi ofa semirigid inhibitor (15), with distances indicating the relationship 2 0 1 0 0 -1 between essential atoms, including the enzyme-bound Zn, is shown 3 0 1 1 0 -1 at the top. Below this is shown the pharmacophore derived from 10 2 1 N 2 1 0 0 distance geometry solutions of the inhibitor given those distances. 2 N 1 2 0 0 Atom 1 is a potential Zn ligand (sulfhydryl or carboxylate oxygen), 3 0 2 0 0 0 atom 2 is a neutral H-bond acceptor, atom 3 is an anion (deprotonated 4 0 1 1 0 0 sulfur or charged oxygen), atom 4 indicates the direction of a 5 F 1 0 0 0 hydrogen bond to atom 2, and atom 5 is the central atom of a carboxylate, sulfate, or phosphate of which atom 3 is an oxygen or 3 1 S 1 0 0 -1 an unsaturated carbon when atom 3 is a deprotonated sulfur.