When There is No Receptor Crystal Structure:
Building Useful Models of Receptor Sites

D. Eric Walters

Department of Biological Chemistry
Finch University of Health Sciences
The Chicago Medical School
3333 Green Bay Road
North Chicago, IL 60064 USA
tel. + 1 708-578-3000 (ext. 498)
ax + 1 708-578-3240
E-mail: walterse@mis.finchcms.edu



http://www.netsci.org/Science/Compchem/feature03.html

ABSTRACT

When the drug design target is a receptor for which there is no experimentally determined three-dimensional structure, it is often useful to construct models of the receptor site. Here, I review a number of methods which have been used successfully in constructing practical receptor models solely on the basis of structure-activity results for a series of ligands. These models may take the form of graphical surfaces, grid points surrounding the ligands, or model atoms and functional groups.


KEY WORDS: receptor, receptor model, drug design, quantitative structure-activity relationships (QSAR), Comparative Molecular Field Analysis (CoMFA), Generating Optimal Linear PLS Estimates (GOLPE), Yak, Genetically Evolved Receptor Models (GERM).


INTRODUCTION

Those of us who work on problems for which there is no X-ray crystallographic structure of a receptor may sometimes feel that we are at a disadvantage. Receptor structures can be used to understand structure-activity relationships, to make predictions about the likely activities of analogs before their synthesis, and even to design completely novel ligands, for instance using de novo design software. It is possible, however, to use structure-activity relationships to construct models of receptor sites, and to use these models in exactly the same way. Usually the first attempt at making such a model is only partially successful, but the errors point to ways in which the model can be refined. After a few cycles of modeling-testing-refining, it is usually possible to obtain models which are quite good. In this paper I survey some of the different approaches to receptor model construction. This review is selective rather than comprehensive, and is intended to illustrate what is possible using receptor models. I have divided the subject of receptor model building into three categories:

  1. Models based on a surface which is constructed over one or more active compounds.
  2. Models based on a grid of points surrounding one or more active compounds.
  3. Models based on a set of atoms, molecular fragments, or amino acid side chains surrounding one or more active compounds. This category could be broadly interpreted to include homology modeling from crystal structures, but I have chosen in this paper to discuss models based only on ligand structures.

An important issue in constructing receptor models is whether to use a few very active structures as templates, hoping to capture the most important structural requirements for bioactivity, or whether to use ligands with a range of bioactivities, hoping to identify features which discriminate between active and inactive compounds. The latter may be preferable, but there are difficulties inherent in using inactive structures. There are reasonable arguments for either approach; one simply has to be aware which approach is being taken so as not to mis-interpret the results.


SURFACE MODELS

One of the simplest approaches to receptor model construction is to superimpose a series of active compounds and then to construct a van der Waals surface over the set. Such a surface can visually convey the steric requirements of the receptor binding site (to the extent that the structure- activity series has probed those requirements). With computer graphics, it is possible to map properties such as electrostatic potential onto this surface, to provide some information about the electronic properties of active analogs. This approach was used in building a receptor model for high potency sweeteners by Culberson and Walters [1]. The starting point was a set of five structurally diverse sweet-tasting compounds which are believed to act at a common receptor site because they share a number of common structural features. Low-energy conformers of these compounds were superimposed in such a way as to get optimal overlap of the carboxylate group, one or two polar N-H groups, and a hydrophobic group. A van der Waals surface was constructed over the superimposed structures, and a composite electrostatic potential from the five compounds was mapped onto the surface (Figure 1).

[Fig 1]

Figure 1



This model was constructed on the basis of compounds with a fairly broad range of potencies. Qualitatively, it was found that compounds which place the carboxylate and N-H groups in the correct orientation are highly likely to taste sweet, and that the more of the hydrophobic pocket which was filled (without exceeding the boundaries), the higher the potency went. This model was successfully used in the pre-synthesis evaluation of analogs and in the design of novel high potency sweeteners [2].

Recently Hahn has reported a program which interactively constructs receptor surface models [3]. Both closed surfaces and partially open surfaces can be generated. Properties such as electrostatic potential, hydrogen bonding potential, and hydrophobicity are readily mapped onto the surface models using color coding. This method starts with one or a few highly active compounds. Hahn's surfaces are implemented in such a way that structures can be energy-minimized with respect to the surface. This feature should be most useful in designing and evaluating new analogs. Hahn and Rogers employed such receptor surface models quantitatively in QSAR studies of corticosteroid binding and dopamine beta-hydroxylase inhibition [4]. After construction of the model, each analog in the series was docked into the receptor model and binding energies were calculated for use as QSAR descriptors.

A low-tech aside: It is also possible to make physical receptor models without the benefit of any computation at all. In 1986 we described a procedure [5] for making receptor models using CPK models of ligands as the templates, and forming a receptor model surface around the ligands using a thermoplastic which is rigid at room temperature but can be cut and shaped when dipped in hot water (Figure 2). The conformations of the CPK models may be determined computationally, experimentally, or using "chemical good sense." There is something inherently satisfying about holding a real, three-dimensional receptor site model, particularly for those who have difficulty seeing stereo pictures!

[Fig 2]

Figure 2



GRID-BASED MODELS

Another way to construct a model around a series of superimposed analogs is to place the structures in a three-dimensional grid and to look at those grid points near the surface of the ligand set. This is the starting point for the Comparative Molecular Field Analysis (CoMFA) approach developed by Cramer [6, 7]. Field properties (e.g., steric, electrostatic) are calculated with respect to the ligands at each of the grid points. Quantitative structure-activity relationship (QSAR) methods such as principal component analysis and partial least squares are used to carry out a statistical analysis of the relationship between field interaction energies and bioactivity.

Allen et al. [8] described the use of CoMFA in constructing a predictive model for beta-carboline agonists and antagonists acting at the benzodiazepine receptor. One of the problems inherent in using CoMFA models is that the number of variables (field descriptors at all of the grid points) is invariably much larger than the number of compounds in the training set. These authors employed a procedure called GOLPE (Generating Optimal Linear PLS Estimates) [9] to eliminate irrelevant descriptors and improve the predictive quality of the CoMFA models. For a set of 37 ligands, these authors reported an estimated standard error in leave-one-out cross-validation of 0.6 log IC50 units.


ATOM-BASED AND AMINO ACID-BASED MODELS

In addition to surfaces and grids, receptor models can be made by placing atoms or groups of atoms (such as amino acid side chains) around a set of active ligands. Holtje and Anzali constructed a tripeptide model of the receptor for cardiac glycosides such as digitalis [10] in the following way. First, they identified eight binding regions around their superimposed ligands, using Goodford's GRID program [11]. Then they docked into these regions amino acid side chains, based on mutagenesis studies which had identified specific amino acids of the Na+/K+ ATPase with which these compounds interact. Finally, three of these amino acids were sufficiently close to each other to be formed into a tripeptide, and possible tripeptide conformations were identified in the Brookhaven Protein Data Bank. The final tripeptide receptor model exhibited a very high correlation between the calculated binding energies (ligand + tripeptide) and the experimentally measured binding affinities (ligand + ATPase).

In a study which carefully examines the potential pitfalls as well as the potential value of quantitative modeling, pseudoreceptor model for the NMDA receptor was built by Snyder and coworkers [12, 13]. Phenol, lysine, an aza- crown ether, and a cyclic diamine were linked with methylene groups to form a binding pocket which is complementary to the agonists and antagonists under consideration. This assembly was surrounded with a sphere of water molecules, and free energy perturbation calculations were carried out to determine the free energy of binding of the ligands. They compared their calculated free energies of binding with experimentally measured Ki values for a series of high-affinity ligands.

An interactive program called Yak has been developed by Vedani and coworkers for the construction of receptor models from amino acid side chains [14]. Starting from the structures of four sulfonamide inhibitors, Yak produced a model of the carbonic anhydrase I binding site which was remarkably similar in some respects to the X-ray crystallographic structure of human carbonic anhydrase I. Yak was also used by Snyder et al. [15] to construct an amino acid-based model of the NMDA receptor.

Whenever one sets out to construct a receptor model using atoms or groups of atoms, one is immediately struck by the arbitrary nature of the choices which must be made. If one considers making a model using a shell of, for example, 50 atoms, and each of those atoms can have one of perhaps a dozen typical protein atom types, the number of possible models which could be constructed is beyond comprehension. This is a highly combinatorial problem, for which systematic solution is not possible. We have recently approached this problem using a genetic algorithm, in a program called GERM (Genetically Evolved Receptor Models) [16, 17]. The genetic algorithm method was developed by Holland [18] to deal specifically with complex, highly combinatorial problems for which systematic solution is impractical. A genetic algorithm does not promise ever to find the global "best" solution, but it often finds many "very good" solutions very quickly. We used the GERM method to produce predictive models for a high-potency sweetener receptor. These models work for a very diverse set of structures (peptides, ureas, guanidines, amino acids). In addition, since genetic algorithms deal with sets of hundreds of models, it is possible to compare many different models and to identify the receptor model features which are most important in discriminating between more active and less active compounds. Figure 3 shows calculated versus experimentally determined log potency results for a set of high-potency sweeteners. The filled blue data points represent 11 compounds which were used as templates for receptor model generation; the open red data points are 11 other compounds which were not included in model generation but whose potencies were predicted from the evolved models.

[Fig 3]

Figure 3



ACKNOWLEDGMENTS

I would like to thank Molecular Simulations, Inc., for providing Quanta/CHARMm and Cerius2 software used in my research; The NutraSweet Company and G.D. Searle Co. for gifts of hardware; Procter & Gamble for financial support.


LITERATURE CITED

  1. Culberson, J.C.; Walters, D.E. In Sweeteners: Discovery, Molecular Design, and Chemoreception; Walters, D.E.; Orthoefer, F.T.; DuBois, G.E., Eds.; American Chemical Society: Washington, DC, 1991; pp 214-223.

  2. Muller, G.W.; Madigan, D.L.; Culberson, J.C.; Walters, D.E.; Carter, J.S.; Klade, C.A.; DuBois, G.E.; Kellogg, M.S. In Sweeteners: Discovery, Molecular Design, and Chemoreception; Walters, D.E.; Orthoefer, F.T.; DuBois, G.E., Eds.; American Chemical Society: Washington, DC, 1991; pp 113-125.

  3. Hahn, M. Receptor Surface Models. 1. Definition and Construction. J. Med. Chem., 1995, 38, 2080-2090.

  4. Hahn, M.; Rogers, D. Receptor Surface Models. 2. Application to Quantitative Structure-Activity Relationships. J. Med. Chem. 1995, 38, 2091-2102.

  5. Walters, D.E.; Pearlstein, R.A.; Krimmel, C.P. A Procedure for Preparing Models of Receptor Sites. J. Chem. Ed., 1986, 63, 869-872

  6. Cramer III, R.D.; Patterson, D.E.; Bunce, J.D. Comparative Molecular Field Analysis (CoMFA). 1. Effect of Shape on Binding of Steroids to Carrier Proteins. J. Amer. Chem. Soc., 1988, 110, 5959-5967.

  7. Cramer III, R.D.; DePriest, S.A.; Patterson, D.E.; Hecht, P. The Developing Practice of Comparative Molecular Field Analysis. In 3D QSAR in Drug Design, Kubinyi, H., Ed.; Escom: Leiden, 1993, pp. 443-485.

  8. Allen, M.S.; LaLoggia, A.J.; Dorn, L.J.; Martin, M.J.; Costantino, G.; Hagen, T.J.; Koehler, K.F.; Skolnick, P.; Cook, J.M. Predictive Binding of beta- Carboline Inverse Agonists and Antagonists via the CoMFA/GOLPE Approach. J. Med. Chem., 1992, 35, 4001-4010.

  9. Baroni, M.; Costantino, G.; Cruciani, G.; Riganelli, D.; Valigi, R.; Clementi, S.; Generating Optimal Linear PLS Estimations (GOLPE): An Advanced Chemometric Tool for Handling 3D-QSAR Problems. Quant. Struct.- Act. Relat. 1993, 12, 9-20.

  10. Holtje, H.-D.; Anzali, S. Molecular Modelling Studies on the Digitalis Binding Site of the Na+/K+ ATPase. In Trends in QSAR and Molecular Modelling 92. Proceedings of the 9th European Symposium on Structure-Activity Relationships: QSAR and Molecular Modelling, Wermuth, C.G., Ed.; Escom: Leiden, 1993, pp. 180-185.

  11. Goodford, P.J. A Computational Procedure for Determining Energetically Favorable Binding Sites on Biologically Important Macromolecules. J. Med. Chem., 1985, 28, 849-857.

  12. Snyder, J.P.; Rao, S.N.; Pellicciari, R.; Monahan, J.B. Receptor Modeling of Highly Charged Excitatory Amino Acids using the Molecular Dynamics/Free Energy Perturbation Approach. In Frontiers in Drug Research, Alfred Benzon Symposium 28, Jensen, B.; Joorgensen, F.S.; Kofod, H., Eds.; Munksgaard: Copenhagen, 1990, pp 109-123.

  13. Snyder, J.P.; Rao, S.N.; Koehler, K.F.; Pellicciari, R. Drug Modeling at Cell Membrane Receptors: The Concept of Pseudoreceptors. In Trends in Receptor Research, Angeli, P.; Gulini, U.; Quaglia, W., Eds.; Elsevier: Amsterdam, 1992, pp. 367-403.

  14. Vedani, A.; Zbinden, P.; Snyder, J.P. Pseudoreceptor modeling: A New Concept for the Three-dimensional Construction of Receptor Binding Sites. J. Receptor Res., 1993, 13, 163-177.

  15. Snyder, J.P.; Rao, S.N.; Koehler, K.F.; Vedani, A. Minireceptors and Pseudoreceptors. In 3D QSAR in Drug Design. Theory, Methods and Applications, Kubinyi, H., Ed.; Escom: Leiden, 1993, pp. 336-354.

  16. Walters, D.E.; Hinds, R.M. Genetically Evolved Receptor Models: A Computational Approach to Construction of Receptor Models. J. Med. Chem., 1994, 37, 2527-2536.

  17. Walters, D.E.; Muhammad, T.D. Genetically Evolved Receptor Models (GERM): A Procedure for Construction of Atomic-level Receptor Site Models in the Absence of a Receptor Crystal Structure. In Genetic Algorithms in Molecular Modeling, Devillers, J., Ed.; Academic Press (in press).

  18. Holland, J.H. Genetic Algorithms. Scientific American, 1992, 267 (7), 66-72.


NetSci, ISSN 1092-7360, is published by Network Science Corporation. Except where expressly stated, content at this site is copyright (© 1995 - 2010) by Network Science Corporation and is for your personal use only. No redistribution is allowed without written permission from Network Science Corporation. This web site is managed by:

Network Science Corporation
4411 Connecticut Avenue NW, STE 514
Washington, DC 20008
Tel: (828) 817-9811
E-mail: TheEditors@netsci.org
Website Hosted by Total Choice