News and Publications

Molecular Graphics and Computation

A.J. Olson, B.S. Duncan, D.S. Goodsell, M. Pique, R. Huey, T.A. Larsen, T. Macke, G.M. Morris, M. Rao, R. Rosenstein, C.D. Rosin, M.F. Sanner, T.A. Soares, W.L. Walker*

* University of California, Los Angeles, CA

Molecular data coming from studies such as the Human Genome Project and from the biostructural research community continue to grow at exponential rates. With this growing information base comes the promise of a true structural and functional understanding of normal life processes and the nature of disease at the molecular and atomic level. However, the complexity of the systems involved, both from a structural and an informational point of view, cannot be underestimated. Computational technology coupled with biological understanding provides a pathway to realize the promise.

The Molecular Graphics Laboratory is addressing these challenges with advances in both technology and biomedical applications. We are shifting the technologic focus to a new, open paradigm that allows free use of many diverse methods within a consistent, interoperable environment. We have applied our computational and graphics technology to topics of biomedical interest, including the fight against HIV type 1 (HIV-1) drug resistance, the search for effective blood coagulation inhibitors, the characterization of protein-protein interfaces, and the design of optimal DNA-binding inhibitors.

OPEN COMPUTING ENVIRONMENTS

The Python environment described last year has held up to its promises. Python is a small, object-oriented scripting language that is simple to learn, yet powerful in its applications. Because it is dynamically interpreted and easily extendable, it is an excellent tool for prototyping applications. In the past year, we made most of our computational tools available within this programmable environment, substantially increasing their interoperability.

We have developed a number of new extensions to Python, including RAPID, a library to detect intersections between surface models; Molecule, a class to represent and manipulate molecular data; DejaVu, a 3-dimensional geometry viewer class; and Help, an on-line facility to aid in the use of all Python objects. We have also implemented a number of existing molecular tools within this environment. These include Harmony and SurfDock; Cosmo, a surface-based approach to electrostatics; and Babel, a general utility for translating molecular formats.

We added a visualization tool, DejaVu, to this collection, to allow display of data coming from any of the computational methods. The viewer uses standard OpenGL calls and is written in Python. DejaVu is tightly integrated into the environment, can be completely customized, and is reusable. In some sense, DejaVu is a new way to approach scientific visualization: the rendering capability is now simply another tool that can be slotted into our working environment.

COEVOLUTION ANALYSIS OF HIV-1 RESISTANCE

In collaboration with R.K. Belew, University of California, San Diego, we developed a coevolutionary method for designing compounds to inhibit an entire class of mutating targets. The goal is to design resistance-evading inhibitors of HIV-1 protease that are effective against both the wild-type and mutant enzymes. Coevolution is loosely based on coevolutionary "arms races" observed in Nature, such as the adaptations of herbivorous insects and the host plants of the insects. In our simulations of HIV-1 drug resistance, we allow a set of inhibitors and a set of mutant proteases to compete and ultimately find the best inhibitor to block the entire set of mutants and the best mutant in the face of all possible inhibitors.

Coevolution simulations have indicated that traditional drug-design techniques are often ineffective for targets that mutate. Although the goal of traditional methods is to fill the active site of the target enzyme, coevolution experiments suggest that inhibitors must be designed against the immutable properties of the active site, the properties that must be conserved in order to retain the ability to cleave the Gag and Pol processing sites. Coevolution simulations have also shown that powerful resistance-evading inhibitors can be designed by optimizing activity simultaneously against a large set of mutant enzymes.

MOLECULAR RECOGNITION IN BLOOD COAGULATION

In the past year, we analyzed recognition events in both the formation and the dissolution of blood clots. After successfully predicting the binary complex of tissue factor with factor VIIa, the first step in the cascade leading to formation of a blood clot, we looked to the mode of interaction of the binary complex with factor X, the next step in the cascade. A combination of homology modeling to generate a multidomain model of factor X, SurfDock analysis of protein-protein docking, and AutoDock analysis of peptide-protein docking has lead to a model of the ternary complex; this work was done in collaboration with W. Ruf and T. Edgington, Department of Immunology. This model is being tested experimentally by using site-directed mutagenesis of residues predicted to play key roles in the complex.

We also investigated the interaction of t-plasminogen activator, the enzyme that activates plasmin and leads to dissolution of fibrin clots, with its substrates and inhibitors. During the past few years, t-plasminogen activator has been widely used in the treatment of acute myocardial infarction. Using AutoDock, we explored the molecular basis of recognition of this important serine protease. As is the case with trypsin, our results underscore the importance of a "specificity pocket" in t-plasminogen activator for recognizing the amino acid adjacent to the scissile bond.

AUTODOCK: A GENERAL TOOL FOR PROTEIN-LIGAND DOCKING

Our AutoDock suite of programs, designed to dock flexible ligands to a protein target, is a mature computational tool and is in use in hundreds of laboratories. This past year, we added a Lamarckian genetic algorithm, which markedly improves the efficiency of AutoDock. AutoDock is now able to dock reliably ligands of the size and flexibility of tetrapeptides. The new method is in beta testing and will be released generally within the year.

One exciting advance has been the development of a Python-based version. This version allows the functionality of AutoDock to be used within the larger set of tools available in the laboratory. The Python version is more modular than the distributed version and allows rapid prototyping of new search methods and new energy evaluation methods.

In the past year, we used AutoDock in several applications. In collaboration with M. Fitzgerald, Department of Cell Biology, we analyzed the interaction between 2 enzymes involved in stereospecific ketonization and their substrates. Our purpose was to define the structural basis for the different substrate specificities. We also continued work on the atomic basis of substrate recognition in HIV-1 protease and the design of optimal inhibitors.

In collaboration with E. Madison, Department of Vascular Biology, we studied the subsite specificity of trypsin by docking the model substrates Phe-Gly-Arg and Phe-Phe-Arg. It has generally been accepted that the canonical conformation of the active center loop of standard inhibitors is a substratelike binding mode for the chymotrypsin enzyme family. On the basis of experimental results from the Madison laboratory and our AutoDock modeling results, we have proposed a noncanonical mode of binding of active substrates. In contrast to the canonical binding conformation, in our model, the P2 residue of the substrate extends toward the protease surface and contacts the S2 pocket, and the P3 residue extends into the solvent. Both the experimental and the computational results suggest that the widely accepted canonical conformation of the reactive-site loop of the standard inhibitors is a nonproductive binding mode for the chymotrypsin family of enzymes.

SURFDOCK: A GENERAL TOOL FOR PROTEIN-PROTEIN DOCKING

We have made considerable progress in the development of SurfDock, our program for analyzing and predicting protein-protein interactions. Although predicting complexes of proteins that have minor structural changes on binding is fairly straightforward, modeling proteins that have extensive rearrangements is difficult. As part of our research in component-based modeling techniques, our strategy to address this problem is to combine a wide variety of techniques within a uniform computational framework. Using the Python language framework, we extended SurfDock to include algorithms from computer vision, computational geometry, spherical harmonic analysis, and graph theory.

In collaboration with R. Beachy, Department of Cell Biology, we applied our modeling and visualization tools in the analysis of resistance to viral infection in transgenic plants that express a native or a mutant viral coat protein. We used our interactive symmetry tools to examine the interactions between molecules of native and mutant coat proteins of tobacco mosaic virus in an effort to understand the relationship between these interactions and the degree of resistance to viral infection.

Working with data provided by G. Feher and his coworkers at the University of California, San Diego, we used SurfDock to model the interaction of cytochrome c₂ with the photoreaction center. The structures of the individual components were known from x-ray crystallography, but a 4.5-Å map of the complex was ambiguous in terms of the interaction between the 2 molecules. Previous models had been based on the low-resolution diffraction data and suggested that cytochrome c₂ contacts mainly the M subunit of the photoreaction center. In our calculations, we weighted the electrostatic component as a dominant factor in evaluating the score of the complex. We submitted our top 15 docked complexes to Dr. Feher and his colleagues. Subsequently, in collaboration with D. Rees at Caltech, Dr. Feher determined the structure of the complex to 2.8-Å resolution. This complex differed from all previously published models. However, our computed model, the highest ranked complex, closely resembles the x-ray structure, showing contacts between both M and L subunits.

FROM ATOMS TO CELLS

Large, multisubunit assemblies mediate many of the important processes of life. We have developed computational and visualization tools to aid in the structure solution and analysis of multisubunit protein assemblies. We created a modular "symmetry server" that simplifies the creation and analysis of multisubunit models, and we added a "symmetry search" functionality that allows automated searching of conformations within a given custom symmetry space. These tools were used in collaboration with J. Johnson, Department of Molecular Biology, to explore pleomorphic virus structures and in collaboration with M. Yeager, Department of Cell Biology, to interpret low-resolution electron density of the cardiac gap junction.

To characterize the interaction important in the assembly of multimeric proteins, we did a visual survey of 136 homodimeric proteins, with images that highlight the features of the protein-protein interfaces. These images reveal the mechanisms by which individual proteins achieve stability in dimeric complexes. Nearly all the proteins have interfaces between 2 globular subunits; a minority, however, have extensive interdigitation of the 2 protein chains. Approximately one third of the proteins have a recognizable hydrophobic core, with a single large, contiguous hydrophobic patch surrounded by a ring of polar interactions. The remaining proteins have a varied mixture of small hydrophobic patches, polar interactions, and bridging water molecules, all scattered across the entire interface. The presence of a hydrophobic core does not appear to correlate with function, but the proteins with extensive interdigitation of chains all perform functions that require both stability and internal symmetry.

OPTIMAL DESIGN OF INHIBITORS

In collaboration with E. Landaw, University of California, Los Angeles, we analyzed the theoretical limits of DNA sequence recognition by linked polyamides. These polyamides are attractive candidates for design of chemotherapeutic and biosensing applications, because customized molecules can be synthesized to target any given sequence of DNA. In our analysis, we found, fortuitously, that the polyamide subunits currently in use--pyrrole, imidazole, and hydroxypyrrole--are among the theoretically best choices, and we have suggested how these rings might be used to their best advantage. Currently, we are weighing the advantages and disadvantages of different conformations of these molecules, including polyamides designed to bind in an overlapped mode vs those that bind in a staggered mode and homodimeric vs heterodimeric linked polyamides.

PUBLICATIONS

Goodsell, D.S. Molecules into cells: Depicting the cellular mesoscale. Leonardo, in press.

Larsen, T.A., Olson, A.J., Goodsell, D.S. Morphology of protein-protein interfaces. Structure 6:421, 1998.

Morris, G.M., Goodsell, D.S., Halliday, R.S., Huey, R., Hart, W.E., Belew, R.K., Olson, A.J. Automated docking using a Lamarckian genetic algorithm and an empirical free energy function. J. Comput. Chem., in press.

Olson, A.J., Goodsell, D.S. Automated docking and the search for HIV protease inhibitors. SAR QSAR Environ. Res. 8:273, 1998.

Reva, B.A., Finkelstein, A.V., Sanner, M.F., Olson, A.J. Residue-residue mean-force potentials for protein structure recognition. Protein Eng. 10:865, 1997.

Reva, B.A., Finkelstein, A.V., Sanner, M., Olson, A.J., Skolnick, J. Recognition of protein structure on coarse lattices with residue-residue energy functions. Protein Eng. 10:1123, 1997.

Rosin, C.D., Belew, R.K., Morris, G.M., Olson, A.J., Goodsell, D.S. Computational coevolution of antiviral drug resistance. In: Proceedings of the 6th International Conference on Artificial Life. Adami, C., et al. (Eds.). MIT Press, Boston, in press.

Walker, W.L., Kopka, M.L., Dickerson, R.E., Goodsell, D.S. Design of stapled DNA-minor-groove-binding molecules with a mutable atom simulated annealing method. J. Comp. Aided Mol. Design 11:539, 1997.

Walker, W.L., Kopka, M.L., Goodsell, D.S. Progress in the design of DNA-sequence-specific lexitropsins. Biopolymers Nucleic Acids, in press.

Walker, W.L., Landaw, E.M., Dickerson, R.E., Goodsell, D.S. The theoretical limits of DNA sequence discrimination by linked polyamides. Proc. Natl. Acad. Sci. U.S.A. 95:4315, 1998.