Scientific Report 2008

Molecular Biology

Computation and Visualization in Structural Biology

A.J. Olson, D.S. Goodsell, M.F. Sanner, M. Chang, S. Cosconati,* S. Dallakyan, S. Forli, A. Gillet, R. Harris, R. Huey, J. Huntoon, G. Johnson, D. Keidel, W. Lindstrom,** G.M. Morris, A. Omelchenko, A. Perryman, M. Pique, R. Rosenstein, M. Utsintong,*** G. Vareille, Q. Zhang, Y. Zhao****

* University of Naples "Federico II," Naples, Italy
** Acelot, Inc., Santa Barbara, California
*** Mahidol University, Bangkok, Thailand
**** CambridgeSoft, Cambridge, Massachusetts

In the Molecular Graphics Laboratory, we develop novel computational methods to analyze, understand, and communicate the structure and interactions of complex biomolecular systems. Within our component-based simulation and visualization environment, we continue to develop 3-dimensional molecular models as a tangible human-computer interface in educational and research settings, methods for predicting biomolecular interactions and using these interactions in structure-based inhibitor design, analyzing biomolecular structure and function, and presenting the biomolecular world in education and outreach.

Component-Based Visualization Environments

To facilitate the integration and interoperation of computational models and techniques from a wide variety of scientific disciplines, we continue to expand our component-based software environment. The environment is centered on Python, a high-level, object-oriented, interpretive programming language. This approach allows the compartmentalization and reuse of software components. Python provides a powerful "glue” for assembling computational components and, at the same time, a flexible language for the interactive scripting of new applications.

In 2007, we released 4 versions of our software tools: 1.4.4, 1.4.5, 1.4.6, and 1.5.0. We are able to release new versions so quickly because of the state-of-the-art overnight software testing environment we have developed. Starting with version 1.4.6, we modified our installation mechanism to allow the simultaneous installation of multiple versions of the software. All software tools have been markedly improved. For example, we enhanced ADT in supporting the new AutoDock4 (described in more detail later); and in the Python Molecular Viewer, ADT modules no longer require a graphical user interface, rather they can now also function as stand-alone scripts or as nodes in Vision, our graphical network editor. Users can now load their own code as nodes in Vision. We have improved APBSCommands.py to use local installations of APBS (a software package for the numerical solution of the Poisson-Boltzmann equation) or remote APBS Web services through the Opal toolkit. We have added support for ambient occlusion for polygonal meshes in the visualization component, making it easier to visualize cavities on molecular surfaces (Fig. 1).

Fig. 1. A molecular surface for HIV protease, with traditional shading on the left and ambient occlusion on the right. Ambient occlusion simulates the reduced lighting in sheltered areas and gives useful cues to the shape of complex molecular surfaces. Image generated with the Python Molecular Viewer.

This software has been widely distributed. We had more than 132,000 downloads during the past 12 months, and 1761 unique users voluntarily registered in our user database (a sustained average of 5 users per day).

Tangible Interfaces In Structural Biology

We have continued to develop autofabricated physical models ("solid printing”) of biological molecules in the context of an augmented reality environment, with the goal of using the models in both research and education. In collaboration with E. Keinan, Department of Molecular Biology, we used tangible models to design self-assembling chemical structures that mimic the self-assembly of viral capsids. Our hypothesis was first developed and tested by using tangible models and then further characterized by molecular dynamics simulation. The models allowed direct physical testing of different possibilities for chemical complementarity of subunits, for instance, identifying 2 alternative binding modes for the self-assembled corannulene currently under development (Fig. 2).

Fig. 2. Physical models were used to study self-assembly of viruses and a designed corannulene complex. Left, Three frames from a movie showing self-assembly of a virus model. The subunits were fabricated on the basis of the molecular structure, and magnets were placed at the interfaces. When the models are shaken in a small bottle, the intact virus self-assembles. Right, Using these same structural model abstractions, a corannulene complex was designed. Two possible modes of assembly were discovered by model building and subsequent computational simulation.

We have also begun work on larger models that capture the dynamic characteristics of biomolecules. Most recently, we created an articulated model of MsbA, a bacterial ABC transporter that undergoes a large and complex conformational change in the course of its transport cycle.

Development of Inhibitors For HIV Protease and Integrase

We are using several approaches in our continuing work on the development of inhibitors of HIV protease. After analyzing the evolution of drug resistance against previously approved inhibitors, we developed a method to predict mutations associated with drug resistance that might be induced by new HIV protease inhibitors. With this technique, we were able to detect more than half of the major mutations in drug-resistant HIV proteases. Site-directed mutagenesis validated resistance of 3 predicted mutations (I47V, F53L, and I84V) for AB2, a novel inhibitor of HIV protease, and the clinically approved drug amprenavir.

In collaboration with ActiveSite, San Diego, California, and C.D. Stout, Department of Molecular Biology, we are developing methods to interpret electron density maps from crystallographic fragment screens. In our method, a combination of peak characterization and computational docking is used to deconvolute peaks in the crystallographic electron density maps that correspond to fragment binding in crystals soaked with a mixture of fragments. The method is very rapid, because docking is only performed in the immediate vicinity of the identified peaks, and so can be easily incorporated into the high-throughput work flow in use at ActiveSite.

To improve the efficiency of finding promising chemical lead compounds, the so-called hit rate in drug discovery, we developed a method that combines high-throughput screening and virtual screening. We performed a virtual screen of 1476 compounds obtained from a cell-based anti-HIV inhibitor screen. We used a novel selection procedure to choose compounds that bound preferentially either to the active site or to a putative allosteric exosite, but not to both. A total of 5 compounds showed micromolar binding constants; 3 were predicted to bind in the active site and 2 in the exosite. Kinetic analysis showed noncompetitive binding for the predicted exosite inhibitors, which are now being further characterized.

We are also collaborating with researchers at Pfizer, Inc., Sandwich, England, and with J.A. McCammon and his group at the University of California, San Diego, to explore the structure and function of HIV integrase (Fig. 3). We have modeled a form of integrase with a closed loop and 2 magnesium ions that is expected to be relevant to structure-based drug design. We are extending this research to model several drug-resistant mutants and explore how these mutations modify the dynamics of the enzyme. When combined with quantum mechanical calculations, these models will be useful for studying the mechanism of catalysis. Long-term goals are to characterize the interaction of the catalytic domain with DNA and ultimately use this information to characterize binding of DNA to the intact enzyme.

Fig. 3. Computational modeling and molecular dynamics were used to generate models of HIV integrase with a closed active-site loop and 2 hydrated magnesium ions (green). Two conformations of the enzyme are shown with ribbons, and water molecules and catalytic residues that coordinate the magnesium ions are shown with bonds. Image generated with the Python Molecular Viewer.

Protein-Ligand Docking with AutoDock

According to recent reports, the computer program AutoDock is the most widely cited docking method, and we have continued to make it a central tool for predicting biomolecular interactions. AutoDock4 was released in 2007 and is available through a convenient open-source license. AutoDock was the first docking code to be used in a public, Internet-based distributed computing project, and we have continued to use this massive computing resource to perform experiments that are beyond the reach of traditional computing. We are currently using the World Community Grid, a large Internet distributed computing project sponsored by IBM, to support FightAIDS@Home on more than 800,000 client computers. Personal computers are used by the program when the computers are not in use by their owners, providing an enormous, and largely untapped, computational resource.

To support our growing user community, we have hosted tutorials, workshops, and lectures presenting basic methods in AutoDock and specific application to virtual screening. We have also continued to expand the capabilities of AutoDock and the graphical user interface AutoDockTools, for instance, exploring the use of gradient information in the search, predicting covalent complexes between ligands and proteins, and streamlining all aspects of virtual screening. Most recently, we tested a method for using results from reiterated docking experiments to evaluate an empirical vibrational entropy of binding in ligand-protein complexes in collaboration with R.K. Bele, University of California, San Diego, and K.S. Carroll, University of Michigan, Ann Arbor. In addition, we have added new support for the relaxed complex method, in which snapshots are taken from a molecular dynamics simulation and used to sample the range of conformations available to a protein target.

We have continued development and application of AutoLigand, a program used to identify and quantify optimal binding sites in proteins. We showed that the method is effective for identifying binding sites in proteins and for optimizing the binding of drugs to protein targets. We have used the method to design exosite inhibitors of HIV protease; inhibitors for histidine deacetylase, in collaboration with J.M. Gottesfeld, Department of Molecular Biology; and exosite inhibitors for p38 MAP kinase, in collaboration with J.A. Tainer, Department of Molecular Biology.

Protein Flexibility and Protein-Protein Docking

In collaboration with C. Bajaj, University of Texas, Austin, we have continued the development of a new protein-protein docking technique that is now implemented in a software program called F2Dock (which stands for Fast Fourier-transform—based docking). This program is being tested on a set of protein-protein complexes to evaluate its performance and identify its shortcomings.

We have continued the development and testing of a new automated docking program called FLIPDock that uses a hierarchical and multiresolution representation of the flexibility of biological macromolecules to model protein motion and induced fit. We have further developed this software tool and applied it to a variety of docking problems in which receptor flexibility is known to cause the failure of rigid receptor-based docking simulations. We have optimized FLIPDock and fixed several problems in the software; we released version 0.1 beta to a selected set of users.

To test the hypothesis that low-resolution surfaces can be used to improve the success of protein-protein docking, we did a systematic study of the effects of multiresolution blurred surfaces on shape complementarity in a set of 66 protein-protein complexes for which complexed and unbound structures are available. We found that medium-resolution smoothing can reproduce about 88% of the shape complementarity of atomic resolution surfaces, and complexes formed from the free component structures show many overlaps and gaps with atomic resolution surfaces, which are improved by smoothing the surfaces to low resolution.

Computational Modeling of Extracellular Interactions of Tissue Factor

Signaling through protease-activated receptor 2 (PAR-2) by the complex composed of tissue factor and coagulation factor VIIa regulates gene transcription and protein translation, cell proliferation and survival, and cell motility and integrin activation. This signaling involves cleavage of the PAR-2 extracellular N-terminal tail between arginine at position 36 and serine at position 37 by the protease domain of factor VIIa. However, it is unclear how the protease domain of factor VIIa recognizes and binds the PAR-2 tail to facilitate the cleavage. We used molecular modeling and molecular dynamics simulations to derive the interactions between PAR-2 and factor VIIa. We found 3 types of key interactions at noncatalytic sites of the factor VIIa protease domain. Four of the key factor VIIa residues were then experimentally shown to be involved in PAR-2 activation.

Visual Methods from Atoms to Cells

Understanding structural molecular biology is essential to foster progress and critical decision making among students, policy makers, and the general public. In the past year, we continued our long-standing commitment to science education and outreach with a combination of presentations, popular and professional illustrations and animations, 3-dimensional tangible models, and a presence on the World Wide Web. In these projects, we use the diverse visualization tools developed in the Molecular Graphics Laboratory to disseminate results that range from atomic structure to cellular function.

In the past year, the "Molecule of the Month” at the Protein Data Bank entered its ninth year of providing an accessible introduction to the central database of biomolecular structure (Fig. 4). In collaboration with T. Herman, Milwaukee School of Engineering in Wisconsin, we continued work on "Protein Active Learning Modules” that provide educational materials for the high school and undergraduate level. We are also extending our modeling efforts from the realm of molecules and complexes into the realm of whole cells. Building on previous illustrative work, we are developing methods to model significant parts of living cells (Fig. 5). Currently, we are building these tools in the context of high-end computer animation software, to allow easy creation of educational materials.

Fig.4. The adrenergic receptor was presented as the 100th installment of the "Molecule of the Month” at the Protein Data Bank (http://www.pdb.org). Each month, a new molecule is presented with a description of its structure, function, and relevance to health and welfare. Visitors are then given suggestions on to how to begin their own exploration of the structures in the Protein Data Bank.

Fig. 5. A simulated 3-dimensional model of part of a cell, with a cell membrane at the top and generic cytoplasm at the bottom.

Publications

Amaro, R., Minh, D.L., Cheng, L., Lindstrom, W.M., Jr., Olson, A.J., Lin, J.-H., Li, W.W., McCammon, J.A. Remarkable loop flexibility in avian influenza N1 and its implications for antiviral drug design. J. Am. Chem. Soc. 129:7764, 2007.

Beuscher, A.E., Olson, A.J. Iterative docking strategies for virtual ligand screening. In: Computational and Structural Approaches to Drug Discovery: Ligand-Protein Interactions. Stroud, R.M., Finer-Moore, J. (Eds.). RSC Publishing, Cambridge, England, 2007, p. 242. A volume in the series RSC Biomolecular Sciences.

Bongini, L., Fanelli, D., Piazza, F., De Los Rios, P., Sanner, M., Skoglund, U. A dynamical study of antibody-antigen encounter reactions. Phys. Biol. 4:172, 2007.

Chang, M.W., Belew, R.K., Carroll, K.S., Olson, A.J., Goodsell, D.S. Empirical entropic contributions in computational docking: evaluation in APS reductase complexes. J. Comput. Chem. 29:1753, 2008.

Chang, M.W., Lindstrom, W., Olson, A.J., Belew, R.K. Analysis of HIV wild-type and mutant structures via in silico docking against diverse ligand libraries. J. Chem. Info. Model. 47:1258, 2007.

Evans, M.J., Morris, G.M., Wu, J., Olson, A.J., Sorensen, E.J., Cravatt, B.F. Mechanistic and structural requirements for active site labeling of phosphoglycerate mutase by spiroepoxides. Mol. Biosyst. 3:495, 2007.

Goodsell, D.S. Making the step from chemistry to biology and back. Nat. Chem. Biol. 3:681, 2007.

Goodsell, D.S., Johnson, G.T. Filling in the gaps: artistic license in education and outreach. PloS Biol. 5:e308, 2007.

Harris, R., Olson, A.J., Goodsell, D.S. Automated prediction of ligand-binding sites in proteins. Proteins 70:1506, 2008.

Illingworth, C.J.R., Morris, G.M., Parkes, K.E.B., Snell, C.R., Reynolds, C.A. Assessing the role of polarization in docking. J. Phys. Chem. A, in press.

Morris, G.M., Huey, R., Olson, A.J. Using AutoDock for ligand-receptor docking. Curr. Protoc. Bioinformatics, in press.

Olson, A.J., Hu, Y.H.U., Keinan, E. Chemical mimicry of viral capsid self-assembly. Proc. Natl. Acad. Sci. U. S. A. 104: 20731, 2007.

Zhao, Y., Sanner, M.F. FLIPDock: docking flexible ligands into flexible receptors. Proteins 68:726, 2007.

Zhao, Y., Sanner, M.F. Protein-ligand docking with multiple flexible side chains. J. Comput. Aided Mol. Des., in press.