Barry L . Stoddard and Daniel

Protein docking protocols are used for the prediction of both small molecule binding to DNA and protein macromolecules and of complexes between macromolecules. These protocols are becoming increasingly automated and powerful tools for computer-aided drug design. We revew the basic methodologies and strategies used for analyzing molecular recognition by computer docking algorithms and discuss recent experiments in which (i) substrate and substrate analogues are docked to the active site of isocitrate dehydrogenase and (ii) maltose binding protein is docked to the extraceflular domain of the receptor, which signals maltose chemotaxis. Why Protein Docking? The terms "rational drug design" and "computer-aided drug design" refer in their most specific sense to the systematic exploration of the three-dimensional structure of a macromolecule of pharmacological importance, in order to design potential ligands that might bind to the target with high affinity and specificity. This goal is largely carried out through docking protocols, which quantitate the affinity between the macromolecule and a ligand bound in specific locations and conformations. This discipline is currently used to examine molecular recognition with some success for the following purposes: (i) The screening of a large number of small molecule species for binding activity against a single target molecule (1-4). A number of data bases of small molecule structures currently exists for such docking searches, including the Cambridge Structural Database (5), the Fine Chemicals Directory by Molecular Design Limited (4), and the Chemical Abstracts registry. (ii) Detailed statistical and energetic analyses of an individual small molecule and its binding interactions to a specific macromolecular site (6-8). Such an analysis often begins where the initial computational screening ofmany candidates ends, allowing us to quantitate the binding of individual compounds and to design and test closely related moleculess registry. (ii) Detailed statistical and energetic analyses of an individual small molecule and its binding interactions to a specific macromolecular site (6-8). Such an analysis often begins where the initial computational screening ofmany candidates ends, allowing us to quantitate the binding of individual compounds and to design and test closely related molecules that exploit the architecture and specificity of the protein binding site. Computational paradigms are usually needed, which use more robust conformational searches and energy calculations than those used for rapid screening of large data bases. For this paper we discuss substrate analogue binding studies pursued through a combination of computational techniques and actual enzymatic analysis, using isocitrate dehydrogenase (IDH) as a model system. (iii) The determination of the structure of protein-protein complexes (9-12). This is one ofthe most important emerging problems in structural biochemistry, due to the rapidly increasing number of structures being solved and the even more rapidly increasing number of gene products being identified, characterized, and sequenced that recognize and associate with one another. It is also one of the most difficult problems, due to the challenge ofperforming actual structural analyses of large multiprotein complexes and of computationally modeling structures with such a large degree of topographical and thermodynamic complexity. We discuss in this paper recent results from computationally predicting the protein-protein interactions between the tar protein, which has been shown to be a membrane-bound receptor that mediates both aspartate and maltose chemotaxis, and the maltose binding protein (MBP; which binds to the receptor). Modeling and Simplifying the System To determine and characterize molecular recognition and binding by a large macromolecule, simplified computational strategies currently must be followed in order to keep the calculations within reason. The simplifications that are used are severalfold. (i) Rigid body docking searches. The number of possible conformational isomers of a macromolecule of even limited size is so large that the target molecule is usually treated initially as a collection ofunmoving atoms (7, 10). In addition, for most data base-screening algorithms, the small molecule probe structures are also held rigid (a compromise that reduces the success rate of identifying potential drugs by an unknown amount). During the refinement of the low-energy docking solutions, some macromolecular dynamic motion is sometimes allowed in the region of the docked complex. The methods that use rigid atoms reduce the computational demand of docking searches but can be misleading since virtually all substrates and other ligands to macromolecular surfaces induce conformational changes upon binding. Agard and coworkers (13, 14) have shown that modeling algorithms that make use of multiple side-chain rotamers provide an energy calculation that is powerful in predicting conformational variation in the active site. Such calculations should allow the design and manipulation of engineered enzymeAbbreviations: IDH, isocitrate dehydrogenase; MBP, maltose bind-