For our purposes, a pharmacophore is defined as the spatial arrangement of a minimal set of discriminating molecular features necessary to characterize the biological activity of a given system. Molecular fragments or macro classes such as hydrogen-bond acceptor/donor sites, ionized, or hydrophobic groups, can represent each of these features in three-dimensional space. A variety of techniques can be used to develop a pharmacophore model ranging from simple hypothesis generation via manual molecular overlays to completely automated procedures. References 25 provide good overviews of the strategies that will be discussed here.
Among the more popular commercially available automated procedures are programs such as DISCO (6), GASP (7) and Catalyst (8-9). Each of these programs attempts to determine the minimal pharmacophore given a small set of compounds, typically 5-10 diverse structures that span a range of biological activities. The underlying assumption in many of these methods is that most of the compounds in the training set will share a similar set of pharmacophore features (eg., hydrogen-bond acceptor or aromatic ring), though individual compounds need not necessarily contain all of the hypothetical pharmacophore elements. Once a pharmacophore is determined, it can be used to search compound databases, aid in virtual library design, or be used directly in lead optimization.
Recently, the performance of each of these pharmacophore generation programs was assessed based on their ability to reproduce target pharmacophores that were derived from various protein-ligand X-ray complexes (10). In that study, the authors considered five protein targets - Thrombin, CDK-2, DHFR, HIV-RT, and Thermolysin. The target pharmacophores were defined based on the observed protein-ligand interactions that were common to all molecules in each set. Generally, only low molecular weight non-peptidic compounds were considered, and were chosen to represent diverse structural classes; an average of 7 ligands were used for each protein target. Overall, the authors rank GASP = Catalyst > DISCO
ANNUAL REPORTS IN MEDICINAL CHEMISTRY-38 ISSN: 0065-7743
e 2003 Elsevier Inc All rights reserved.
based on their ability to reproduce the target pharmacophores. Though automated procedures for pharmacophore generation are very useful and efficient, they are by no means necessary to enable the searching of compound databases.
Structure-Based Pharmacophore Identification - Researchers at Aventis used a combination of peptide mutation data, NMR structural studies on the urotensin II (U-II) peptide, and a number of analogs to derive a three point pharmacophore model (11). The peptide mutation experiments identified Trp-7, Lys-8, and Tyr-9 in U-ll as important contributors to the recognition and activation activity of the peptide, while the NMR studies defined the overall three-dimensional arrangement of the peptide. By assuming the solution-based NMR structure resembles the bioactive conformation, a pharmacophore model consisting of two hydrophobic-aromatic groups and one positive ionizable group was constructed by centering the aromatic-hydrophobic groups on Trp-7 and Tyr-9 rings, and the positive ionizable feature centered on the Ne of Lys-8 (Catalyst definitions). The Catalyst query was then used to carry out a virtual screen of the Aventis compound collection. Biological testing of the 500 compounds selected resulted in identification of 10 highly active compounds with IC5o values between 400 nM (1) and 7 pM, and belonging to six different structural classes. This pharmacophore-based search resulted in a nearly 20x increase in hit rate / \ compared to conventional GPCR high-throughput screens as well as validation of the pharmacophore model (11).
A similar strategy was used to discover novel non-peptidic inhibitors of a4|31 (VLA-4) (12). The structure of a lead compound, 4-[N-(2-methylphenyl)-ureido]phenylacetyl-Leu-Asp-Val], was modeled based on the X-ray conformation of the Ile39-Asp40-Ser41 region of VCAM-1. A Catalyst query was generated based on the position of the carboxylate COO- in the Asp of the lead compound. A virtual library of possible replacements for Leu-Asp-Val was constructed from reagents in the ACD (13) and PAPU (4-[N-(2-methylphenyl)-ureido]phenylacetyl). The resulting collection of 8894 compounds was searched using the pharmacophore, and 12 compounds were selected based on fit, availability, and ease of synthesis. Of these compound 2 displayed inhibitory potency (1.3 nM) nearly equal to that of the original lead (0.6 nM).
Structure-Activity Based Pharmacophore Identification - Transporters represent an important class of targets for a variety of therapeutic indications such as depression, anxiety, Parkinson's disease, and substance abuse. Specifically, the dopamine transporter (DAT) has been implicated in cocaine addition, and the search for novel inhibitors has benefited from pharmacophore queries based on known antagonists (14-17). From extensive structure-activity relationship (SAR) studies on cocaine and related analogs, a pharmacophore was constructed, which consists of sp3 nitrogen required to be part of a ring, an aromatic ring, and a carbonyl group (14). A 3D search of the NCI database (18) of 206,876 compounds using the Chem-X program (19) found 4096 hits. Elimination of compounds with molecular weight > 1000, inappropriate ring N, and structural analogs resulted in the selection of 70 compounds for testing. Of these, 44 (63%) showed good activity in the primary binding assay. Further studies by these researchers (14-17) resulted in the development of a modified pharmacophore where the carbonyl group is replaced with an aromatic ring. Using a similar searching methodology, several novel classes of compounds based on 3,4-disubstituted pyrrolidines (15), substituted pyridines (16), and 2,3-disubstituted quinuclidines (17) have been reported for DAT inhibition.
GPCRs are an important class of targets for which no 3D structures currently exist. Novel leads are almost exclusively based on either modifications of natural ligands, or HTS of large databases. Based on a series of known muscarinic M3 antagonists, two pharmacophore models were constructed using DISCO, and consists of a tertiary nitrogen, a hydrogen-bond acceptor, and two hydrogen-bond donor sites (20). The pharmacophore models were used to query the Astra corporate database using UNITY (21). These queries produced 177 unique hits. Subsequent testing of 172 available compounds yielded the three most potent
Inhibitors of mesangial cell proliferation (MCP) are thought to be useful in the treatment of glomerular diseases such as diabetic nephropathy and lupus (22). Kurogi and co-workers (22) discovered a novel series of MCP inhibitors through construction of a pharmacophore from four benzylphosphonate compounds (represented by 6) with significant MCP inhibitory potency. The Catalyst pharmacophore consisted of two aromatic rings, two hydrophobic sites, and three hydrogen-bond acceptor sites. A search of the Maybridge database (23) of 47,045 compounds gave rise to 41 hits. Four of the best fitting compounds were tested for MCP inhibitory activity and demonstrated potency comparable to compounds in the original training set (exemplified by 7). Additionally, the new lead compounds were devoid of the cell toxicity present in the original series of compounds that were used to build the pharmacophore model.
Nearly all known chymase inhibitors suffer from poor stability in vivo, which will tend to lessen their potential usefulness as therapeutic agents (24). Efforts to improve the stability of a series of thiazolidinedione (8) and thiadiazole (9) chymase inhibitors proved difficult without sacrificing potency (24). Identification of a new class of inhibitors followed construction of a pharmacophore based on 26 compounds. The pharmacophore consisted of two hydrogen-bond acceptors flanked on each end by hydrophobic groups. A search of the ACD collection of compounds with Catalyst identified a number of hits, 45 of which were selected for testing. Three of the selected compounds showed potency >30% at 1 pM in the primary assay. Compound 10 showed 100% stability and an IC5o of 909 nM against chymase.
Was this article helpful?