Real-Time Mechanics in Molecular Modeling

Mark Surles

Introduction

Recent increases in computer performance, visualization, and molecular mechanics algorithms now allow researchers to dock ligands or modify loops interactively while a computer continuously modifies the conformation to account for changes in energies. Interactive user manipulation of ligands and receptors, subject to real-time molecular mechanics, provides new insights into ligand-receptor interactions and increases productivity in model building and testing.

This close coupling between the user and molecular mechanics is a new paradigm in molecular modeling that is contained in the SCULPT modeling system. A goal with SCULPT is to bring the researcher closer to the modeling process. A researcher can directly modify and explore conformations by applying their knowledge of the chemistry to the problem, while the computer maintains geometric properties such as bond lengths and non-bonded interactions. This contrasts traditional molecular modeling development which has concentrated on improved algorithms; this had the unfortunate result of often imposing a barrier between the researcher and the modeling process.

User-directed manipulation with real-time mechanics lets a user spend more time researching proteins and less time with details of a computer program. We believe this approach helps researchers better understand protein structure and folding, ligands, and ligand-receptor interactions.

This paper begins with a comparison between interactive manipulation and tradition modeling systems. The application of this paradigm to the following areas of computer-aided drug design is then described:

• Ligand modeling including docking and conformational analysis;
• Model building using homology and site-directed mutagenesis.

Traditional Modeling Paradigm

Computer-aided molecular design is a promising field for building new therapeutic drugs based on the structure and function of molecules. Computer systems are used for database screening, graphical visualization, molecular mechanics, active-site modeling, and ligand design. Rational design of drugs based on known structure and pharmacophores has the potential of producing more effective lead compounds, shortening development time, and reducing costs. No commercially available drug has been developed solely from computer models. However, (Lam et al., 1994) and (DesJarlais et al., 1990) report the use of database screening, computer visualization, and molecular simulation to derive lead compounds in an HIV protease inhibitor. Similarly, many pharmaceutical and biotechnology companies have computational chemistry groups actively using molecular modeling to solve the structure of active sites and design novel ligands. The success of future efforts will in large part be predicated on more accurate and accessible software for computer-aided molecular design.

The following scenario outlines how existing computer-aided drug design software can be used in a project in which the structure of an active site is known. First, screen a massive database of known compounds for potential leads with a program such as DOCK (Kuntz et al., 1982) that ranks compounds based on steric and electrostatic fit in the receptor. Second, prune the list of leads further by visually inspecting their active-site fit. Third, using an interactive graphics system, modify or build new leads with better steric fit, more hydrogen bonds, and lower solvation energy. Throughout this process, run physical experiments to test hypotheses concerning receptor-ligand interactions.

A crucial missing component is that none of these steps model cooperative conformational changes caused by interactions between the ligand and receptor. Only the ligand, not the active site, can move in the database screening; neither can change conformations during the visual inspection or the interactive design. The reason is that traditional molecular modeling software separates the graphics and the simulation. The graphics are used for constructing molecules, editing topology, and visualizing properties, while the simulation carries out energy minimization or dynamics. A modeler uses the graphics package to add atoms or change the three-dimensional structure through bond rotations; little more than visual perception and chemical intuition guides such changes. Thus to modify a flexible compound a modeler must manually change torsion angles using the graphics package, and then submit the resulting structure to a mechanics package to relieve bad steric contacts that are invariably introduced. This approach is time consuming, breaks one's train of thought, and provides little intuition into the physical nature of the ligand-receptor interaction. This paradigm makes it difficult to discover how a change in a compound effects, for example, the tertiary structure of a receptor. Though we use the interaction of ligands with receptors as an example, many of the same problems are common in homology modeling, mutagenesis, and de novo design.

SCULPT Modeling Paradigm

The SCULPT paradigm, in contrast, tightly couples the simulation and the user interface. A user can guide the overall path of atoms while the computer maintains physically valid atom separations automatically. This lets one perform the following operations which are difficult in traditional modeling:

1. Move a functional group directly while maintaining favorable contacts without needing to change a complex set of dihedral angles.
2. Use van der Waals and electrostatic interactions to help guide or hold well-built components of a ligand.
3. Relax the receptor's conformation as the ligand moves.
4. Change backbone conformations directly without needing to break bonds, rotate (j,y) angles, and run a minimizer to fix the result.
5. Move secondary structure such as a helix or sheet to understand the protein mechanics.

One method for understanding steric hindrance and allowable degrees of freedom in a molecule is to build physical CPK and Kendrew models. However, CPK models have many drawbacks: they take a long time to build, fall apart easily, are difficult to hold for large structures, and do not account for non-bonded attractions and repulsions. Professors David and Jane Richardson of Duke University proposed the SCULPT paradigm to build a system that combines benefits of physical and computer models so that one can sculpt proteins into folded shapes by direct manipulation of models with realistic and intuitive behavior and feedback.

SCULPT constrains stiff properties (covalent bond lengths, angles, and planar peptides and rings) to an ideal value, and models weaker properties (multiple dihedral angles and van der Waals, hydrogen-bond, and electrostatic interactions) with potential energies (Surles et al., 1994). A user moves atoms by placing a spring between the cursor and atom, thus increasing the model's potential energy. SCULPT finds new atom positions that locally minimize the potential energy and satisfy the constraints, and then redisplays the screen.

Figure 1 shows a user tugging a Phenylalanine with the mouse, denoted by the spring between the atom and cursor. A previous spring inserted by the user pulls it towards the thumbtack; the lower-right thumbtack holds an atom in place. As the ring turns, it collides with the carbonyl carbon (wireframe shell) and has minor contact with the nearby leucine (dot surface), giving the user the impression of changing a real, physical model. This changes the backbone and sidechain conformations of the twenty-residue helix. The ring turns at approximately 11 updates per second on a Silicon Graphics Indigo2 (150-MHz R4400) or an Apple PowerMac 8100.

Figure 1

Benefits of Interactive Minimization

SCULPT provides new capabilities that complement, but do not replace traditional modeling systems. One can build models using sequence editors, small molecular editors, or homology modeling in other systems and then use SCULPT to explore and manipulate the structure. The results can then be passed back for molecular dynamics and further analysis. This section will survey some of the tasks facilitated by interactive minimization.

Ligand Modeling

Docking. With SCULPT one can interactively dock a ligand into a three-dimensional receptor. As this happens contacts are displayed to indicate favorable (blue) and bad (red) contacts (see Figure 2 for an example of Flavin in Flavodoxin). The user guides the overall path of the ligand, but changes in energies due to van der Waals and electrostatic interactions cause local changes to the ligand's conformation. This approach lets one quickly try various orientations of a ligand in a rigid receptor.

Figure 2

One can also let the receptor change conformation. One can either let the entire receptor move, or just isolate the changes to the sidechains, while keeping the backbone fixed. This lets one move a ligand within a more realistic, flexible receptor.

Conformational Analysis of Multiple Ligands. One can place multiple ligands into a receptor to gain an understanding of their relative interactions and conformational flexibility. A future version of SCULPT will let one superimpose multiple ligands and move the ensemble. The conformational analysis can also be run on individual ligands without a three-dimensional model of a receptor.

Receptor and Protein Modeling

SCULPT has been applied mainly to protein modeling because, in addition to tugging individual atoms, one can apply user tugs to groups of atoms such as helices and loops.

Membrane Proteins. Currently one can pick and drag individual atoms (i.e. apply a force to atoms) or apply a force to a group of atoms. This lets one, for example, move a helix while the attached loops at each end move to maintain a valid conformation. The helix moves as a rigid body while there are no opposing forces, but it deforms as the sidechains (or backbone) interact with the surrounding structure.

Homology and Site-Directed Mutagenesis. At the end of building a structure by homology modeling, one must repair splice points (between peptides), improve sidechain packing, and modify loop conformations. SCULPT can be applied to each of these steps to allow user guidance over the changes. One can isolate the changes made to a structure in SCULPT by freezing (holding in place) groups of atoms. In this case one could freeze the entire structure while only moving a loop, or one could freeze all of a structure while only manipulating internal sidechains. One can also "idealize" peptide bonds, so that one can fix splice points without causing numerous changes to the structure.

Mimic of HIV Protease Interface

The following example summarizes a modeling session by Professor Jane Richardson using SCULPT. SCULPT was used to design a thirty-residue peptide that mimics the interactions at the subunit interface of the dimer in HIV protease. Starting from Brookhaven Data Bank file 5HVP, we froze the backbone in the N- and C-terminal beta strands that form the intermingled beta sheet of the interface and a three-residue segment next to the active-site Asp; atoms in the other subunit were frozen but included in van der Waals interactions. We used SCULPT to find a suitable length, conformation, and sequence of polypeptide that could connect those segments, substituting for all contacts in the native dimer (except the "flap") of the newly modeled complex. The modeling progressed smoothly; in several cases the interactions forced a sequence change, and in one case a length change, from our initial guess. The figure shows the interface between the mainchain of the designed peptide and the trace of the HIV protease subunit.

Figure 3

System Information

More information related to SCULPT can be found on the World Wide Web using the URLhttp://www.intsim.com/~isigen. A trial version of the system is also available at that site. SCULPT currently runs on Silicon Graphics workstations running IRIX 5.2 or later and on Apple PowerMacs.

REFERENCES

DesJarlais, R., et. al., (1990). "Structure Based Design of Nonpeptide Inhibitors Specific for the Human Immunodeficiency Virus-1 Protease", PNAS, 87, 6644-6648.

Kuntz, I., et. al. (1982). "A Geometric Approach to Macromolecule-Ligand Interactions", J. Mol. Biol., 161, 269-288.

Lam, P., et. al. (1994). "Rational Design of Potent, Bioavailable, Nonpeptide Cyclic Ureas as HIV Protease Inhibitors." Science, 263, 380-84.

Surles, M., Richardson, J., Richardson, D. & Brooks, F. (1994). "Sculpting Proteins Interactively: Continual Energy Minimization Embedded in a Graphical Modeling System." Protein Science, 3, 198-210.