DOCK 6: Impact of new features and current docking performance.

This manuscript presents the latest algorithmic and methodological developments to the structure-based design program DOCK 6.7 focused on an updated internal energy function, new anchor selection control, enhanced minimization options, a footprint similarity scoring function, a symmetry-corrected root-mean-square deviation algorithm, a database filter, and docking forensic tools. An important strategy during development involved use of three orthogonal metrics for assessment and validation: pose reproduction over a large database of 1043 protein-ligand complexes (SB2012 test set), cross-docking to 24 drug-target protein families, and database enrichment using large active and decoy datasets (Directory of Useful Decoys [DUD]-E test set) for five important proteins including HIV protease and IGF-1R. Relative to earlier versions, a key outcome of the work is a significant increase in pose reproduction success in going from DOCK 4.0.2 (51.4%) → 5.4 (65.2%) → 6.7 (73.3%) as a result of significant decreases in failure arising from both sampling 24.1% → 13.6% → 9.1% and scoring 24.4% → 21.1% → 17.5%. Companion cross-docking and enrichment studies with the new version highlight other strengths and remaining areas for improvement, especially for systems containing metal ions. The source code for DOCK 6.7 is available for download and free for academic users at http://dock.compbio.ucsf.edu/.

Darwinian docking.

The Darwinian model of evolution is an optimization strategy that can be adapted to docking. It differs from the common use of genetic algorithms, primarily in its acceptance of diverse solutions over finding "global" optima. A related problem is selecting compounds using multiple criteria. I discuss these ideas and present the outlines of a protocol for selecting "hits" and "leads" in drug discovery.

DOCK 6: combining techniques to model RNA-small molecule complexes.

With an increasing interest in RNA therapeutics and for targeting RNA to treat disease, there is a need for the tools used in protein-based drug design, particularly DOCKing algorithms, to be extended or adapted for nucleic acids. Here, we have compiled a test set of RNA-ligand complexes to validate the ability of the DOCK suite of programs to successfully recreate experimentally determined binding poses. With the optimized parameters and a minimal scoring function, 70% of the test set with less than seven rotatable ligand bonds and 26% of the test set with less than 13 rotatable bonds can be successfully recreated within 2 A heavy-atom RMSD. When DOCKed conformations are rescored with the implicit solvent models AMBER generalized Born with solvent-accessible surface area (GB/SA) and Poisson-Boltzmann with solvent-accessible surface area (PB/SA) in combination with explicit water molecules and sodium counterions, the success rate increases to 80% with PB/SA for less than seven rotatable bonds and 58% with AMBER GB/SA and 47% with PB/SA for less than 13 rotatable bonds. These results indicate that DOCK can indeed be useful for structure-based drug design aimed at RNA. Our studies also suggest that RNA-directed ligands often differ from typical protein-ligand complexes in their electrostatic properties, but these differences can be accommodated through the choice of potential function. In addition, in the course of the study, we explore a variety of newly added DOCK functions, demonstrating the ease with which new functions can be added to address new scientific questions.

Assessment of the role of computations in structural biology.

Computations are now an integrated part of structural biology and are used in data gathering, data processing, and data storage as well as in a full spectrum of theoretical pursuits. In this review, we focus on areas of great promise and call attention to important issues of internal consistency and error analysis.

Hierarchical database screenings for HIV-1 reverse transcriptase using a pharmacophore model, rigid docking, solvation docking, and MM-PB/SA.

In this work, an efficient strategy was presented to search drug leads for human immunodeficiency virus type 1 reverse transcriptase (HIV-1 RT) using hierarchical database screenings, which included a pharmacophore model, multiple-conformation rigid docking, solvation docking, and molecular mechanics-Poisson-Boltzmann/surface area (MM-PB/SA) sequentially. Encouraging results were achieved in searching a refined available chemical directory (ACD) database: the enrichment factor after the first three filters was estimated to be 25-fold; the hit rate for all the four filters was predicted to be 41% in a control test using 37 known HIV-1 non-nucleoside reverse transcriptase inhibitors; 10 out of 30 promising solvation-docking hits had MM-PB/SA binding free energies better than -6.8 kcal/mol and the best one, HIT15, had -17.0 kcal/mol. In conclusion, the hierarchical multiple-filter database searching strategy is an attractive strategy in drug lead exploration.

Evaluating the high-throughput screening computations.

The judges evaluated the submissions for the McMaster University High-Throughput Data-Mining and Docking Competition based on 3 criteria: identification of active compounds, percent enrichment, and overview of the competition. Using these metrics, 4 of the participating groups found meaningful enrichment, and 3 groups made perceptive comments about the general nature of the competition.

Small molecule affinity fingerprinting. A tool for enzyme family subclassification, target identification, and inhibitor design.

Classifying proteins into functionally distinct families based only on primary sequence information remains a difficult task. We describe here a method to generate a large data set of small molecule affinity fingerprints for a group of closely related enzymes, the papain family of cysteine proteases. Binding data was generated for a library of inhibitors based on the ability of each compound to block active-site labeling of the target proteases by a covalent activity based probe (ABP). Clustering algorithms were used to automatically classify a reference group of proteases into subfamilies based on their small molecule affinity fingerprints. This approach was also used to identify cysteine protease targets modified by the ABP in complex proteomes by direct comparison of target affinity fingerprints with those of the reference library of proteases. Finally, experimental data were used to guide the development of a computational method that predicts small molecule inhibitors based on reported crystal structures. This method could ultimately be used with large enzyme families to aid in the design of selective inhibitors of targets based on limited structural/function information.

Stability of macromolecular complexes.

Macromolecular interactions are crucial in numerous biologic processes, yet few general principles are available that establish firm expectations for the strength of these interactions or the expected contribution of specific forces. The simplest principle would be a monotonic increase in interactions as the size of the interface grows. The exact relationship might be linear or nonlinear depending on the nature of the forces involved. Simple "linear-free energy" relationships based on atomic properties have been well documented, for example, additivity for the interaction of small molecules with solvent, and, recently, have been explored for ligand-receptor interactions. Horton and Lewis propose such additivity based on buried surface area for protein-protein complexes. We investigated macromolecular interactions and found that the highest-affinity complexes do not fulfill this simple expectation. Instead, binding free energies of the tightest macromolecular complexes are roughly constant, independent of interface size, with the notable exception of DNA duplexes. By comparing these results to an earlier study of protein-ligand interactions we find that: (1) The maximum affinity is approximately 1.5 kcal/mol per nonhydrogen atom or 120 cal/mol A(2) of buried surface area, comparable to results of our earlier work; (2) the lack of an increase in affinity with interface size is likely due to nonthermodynamic factors, such as functional and evolutionary constraints rather than some fundamental physical limitation. The implication of these results have some importance for molecular design because they suggest that: (1) The stability of any given complex can be increased significantly if desired; (2) small molecule inhibitors of macromolecular interactions are feasible; and (3) different functional classes of protein-protein complexes exhibit differences in maximal stability, perhaps in response to differing evolutionary pressures. These results are consistent with the widespread observation that proteins have not evolved to maximize thermodynamic stability, but are only marginally stable.

A molecular basis for the selectivity of thiadiazole urea inhibitors with stromelysin-1 and gelatinase-A from generalized born molecular dynamics simulations.

Matrix metalloproteinases (MMPs) represent a potentially important class of therapeutic targets for the treatment of diseases such as cancer. Selective inhibition of MMPs will be required given the high sequence identity across the family and the discovery that individual MMPs also regulate the natural angiogenesis inhibitor angiostatin. In this study, we have used computational methods to model the selectivity for six thiadiazole urea inhibitors with stromelysin-1 and gelatinase-A, two homologous MMPs that have been implicated in breast cancer. From continuum Generalized Born molecular dynamics (GB-MD) and MM-GBSA analysis, we estimated ligand free energies of binding using 200 snapshots obtained from a short 40 ps simulation of the relevant protein-ligand complex. The MM-GBSA free energies, computed from the continuum GB-MD trajectories, show strong correlation with the experimental affinities (r(2) = 0.74); prior studies have employed explicit water MD simulations. Including estimates for changes in solute entropy in the binding calculations slightly diminishes the overall correlation with experiment (r2 = 0.71). Notably, in every case, the simulation results correctly predict that a given ligand will bind selectively to stromelysin-1 over gelatinase-A which is gratifying given the high degree of structural homology between the two proteins. The increased selectivity for stromelysin-1 appears to be driven by (1) increased favorable van der Waals interactions, (2) increased favorable Coulombic interactions, and (3) decreased unfavorable total electrostatic energies (Coulombic plus desolvation).

Investigation of neuraminidase-substrate recognition using molecular dynamics and free energy calculations.

Development of the new generation of therapeutics against the influenza viral coat protein neuraminidase is a response to the continuing threat of influenza epidemics. A variety of structurally similar compounds have been reported that vary greatly in their ability to inhibit neuraminidase, a critical enzyme that cleaves sialic acid and promotes virion release. To determine how neuraminidase exhibits this wide range of affinities with structurally similar compounds, molecular dynamic simulations, coupled with free energy calculations, were used to determine the binding components of a series of neuraminidase inhibitors. Using four cocrystal structures of neuraminidase-inhibitor complexes, we examined the structural and energetic components of ligand potency and selectivity. An in-depth energetic analysis, including internal energy, entropy, and nonbonded interactions, reveals that potency of ligand binding is governed by nonpolar contacts. Electrostatic components generally oppose binding, although two of the best inhibitors use electrostatic interactions to orient the ligand. This investigation suggests that the enhanced selectivity and potency of the better ligands may arise from an improved positioning of their ligand atoms in the active site due to polar and hydrophobic functionalities. Simulations that included crystal water molecules in the active site indicate that the more potent ligands make less use of water-mediated interactions.

Computer-assisted design of selective imidazole inhibitors for cytochrome p450 enzymes.

A modified version of the DOCK program has been used to predict inhibitors for cytochrome P450cam and its L244A mutant. A library of azole compounds was designed in silico and screened for binding to wild-type P450cam. Lead compounds were synthesized and found to inhibit wild-type P450cam. To test our approach to designing ligands that discriminate between closely related sites, the azole library was DOCKed into both the active sites of wild-type P450cam and its L244A mutant. The L244A active site is predicted to be slightly larger than that of wild-type P450cam. Ligands predicted to have a high affinity for the mutant alone were synthesized and assayed with the recombinant enzymes. All of the compounds showed inhibition of the L244A enzyme (IC(50) = 6-40 microM), and the compounds that were predicted to be too large to bind to the wild-type showed poor inhibition (IC(50) > or = 1 mM). The binding mode was shown to be similar to that predicted by our modified version of DOCK by spectroscopic analysis. A discrepancy between the IC(50) values and spectroscopic K(s) values indicates that the spectroscopic binding constants do not accurately estimate inhibitory activity. This study, the first report of computer-assisted ligand (drug) design for P450 enzymes in which the coordination bond between imidazole and the heme is explicitly considered in structural modeling, opens a promising design avenue because azole compounds are widely used as P450 enzyme inhibitors and drugs.

Molecular dynamics and free energy analyses of cathepsin D-inhibitor interactions: insight into structure-based ligand design.

In this study, we compare the calculated and experimental binding free energies for a combinatorial library of inhibitors of cathepsin D (CatD), an aspartyl protease. Using a molecular dynamics (MD)-based, continuum solvent method (MM-PBSA), we are able to reproduce the experimental binding affinity for a set of seven inhibitors with an average error of ca. 1 kcal/mol and a correlation coefficient of 0.98. By comparing the dynamical conformations of the inhibitors complexed with CatD with the initial conformations generated by CombiBuild (University of California, San Francisco, CA, 1995), we have found that the docking conformation observed in an X-ray structure of one peptide inhibitor bound to CatD (Proc. Natl. Acad. Sci. U.S.A. 1993, 90, 6796-6800) is in good agreement with our MD simulation. However, the DOCK scoring function, based on intermolecular van der Waals and electrostatics, using a distance-dependent dielectric constant (J. Comput. Chem. 1992, 13, 505-524), poorly reproduces the trend of experimental binding affinity for these inhibitors. Finally, the use of PROFEC (J. Comput.-Aided Mol. Des. 1998, 12, 215-227) analysis allowed us to identify two possible substitutions to improve the binding of one of the better inhibitors to CatD. This study offers hope that current methods of estimating the free energy of binding are accurate enough to be used in a multistep virtual screening protocol.

Computational investigation of kinetics of cross-linking reactions in proteins: importance in structure prediction.

The determination of protein structure using distance constraints is a new and promising field of study. One implementation involves attaching residues of a protein using a cross-linking agent, followed by protease digestion, analysis of the resulting peptides by mass spectroscopy, and finally sequence threading to detect the protein folds. In the present work, we carry out computational modeling of the kinetics of cross-linking reactions in proteins using the master equation approach. The rate constants of the cross-linking reactions are estimated using the pKas and the solvent-accessible surface areas of the residues involved. This model is tested with fibroblast growth factor (FGF) and cytochrome C. It is consistent with the initial experimental rate data for individual lysine residues for cytochrome C. Our model captures all observed cross-links for FGF and almost 90% of the observed cross-links for cytochrome C, although it also predicts cross-links that were not observed experimentally (false positives). However, the analysis of the false positive results is complicated by the fact that experimental detection of cross-links can be difficult and may depend on specific experimental conditions such as pH, ionic strength. Receiver operator characteristic plots showed that our model does a good job in predicting the observed cross-links. Molecular dynamics simulations showed that for cytochrome C, in general, the two lysines come closer for the observed cross-links as compared to the false positive ones. For FGF, no such clear pattern exists. The kinetic model and MD simulation can be used to study proposed cross-linking protocols.

Partial acetylation of lysine residues improves intraprotein cross-linking.

Intramolecular cross-linking coupled with mass spectrometric identification of cross-linked amino acids is a rapid method for elucidating low-resolution protein tertiary structures or fold families. However, previous cross-linking studies on model proteins, such as cytochrome c and ribonuclease A, identified a limited number of peptide cross-links that are biased toward only a few of the potentially reactive lysine residues. Here, we report an approach to improve the diversity of intramolecular protein cross-linking starting with a systematic quantitation of the reactivity of lysine residues of a model protein, bovine cytochrome c. Relative lysine reactivities among the 18 lysine residues of cytochrome c were determined by the ratio of d0 and acetyl-d3 groups at each lysine after partial acetylation with sulfosuccinimidyl acetate followed by denaturation and quantitative acetylation of remaining unmodified lysines with acetic-d6 anhydride. These lysine reactivities were then compared with theoretically derived pKa and relative solvent accessibility surface values. To ascertain if partial N-acetylation of the most reactive lysine residues prior to cross-linking can redirect and increase the observable Lys-Lys cross-links, partially acetylated bovine cytochrome c was cross-linked with the amine-specific, bis-functional reagent, bis(sulfosuccinimidyl)suberate. After proteolysis and mass spectrometry analysis, partial acetylation was shown to significantly increase the number of observable peptides containing Lys-Lys cross-links, shifting the pattern from the most reactive lysine residues to less reactive ones. More importantly, these additional cross-linked peptides contained novel Lys-Lys cross-link information not seen in the non-acetylated protein and provided additional distance constraints that were consistent with the crystal structure and facilitated the identification of the proper protein fold.

The collaboratory for MS3D: a new cyberinfrastructure for the structural elucidation of biological macromolecules and their assemblies using mass spectrometry-based approaches.

Modern biomedical research is evolving with the rapid growth of diverse data types, biophysical characterization methods, computational tools and extensive collaboration among researchers spanning various communities and having complementary backgrounds and expertise. Collaborating researchers are increasingly dependent on shared data and tools made available by other investigators with common interests, thus forming communities that transcend the traditional boundaries of the single research laboratory or institution. Barriers, however, remain to the formation of these virtual communities, usually due to the steep learning curve associated with becoming familiar with new tools, or with the difficulties associated with transferring data between tools. Recognizing the need for shared reference data and analysis tools, we are developing an integrated knowledge environment that supports productive interactions among researchers. Here we report on our current collaborative environment, which focuses on bringing together structural biologists working in the area of mass spectrometric based methods for the analysis of tertiary and quaternary macromolecular structures (MS3D) called the Collaboratory for MS3D (C-MS3D). C-MS3D is a Web-portal designed to provide collaborators with a shared work environment that integrates data storage and management with data analysis tools. Files are stored and archived along with pertinent meta data in such a way as to allow file handling to be tracked (data provenance) and data files to be searched using keywords and modification dates. While at this time the portal is designed around a specific application, the shared work environment is a general approach to building collaborative work groups. The goal of this is to not only provide a common data sharing and archiving system, but also to assist in the building of new collaborations and to spur the development of new tools and technologies.

Rational design of a nonpeptide general chemical scaffold for reversible inhibition of PDZ domain interactions.

Novel small molecules were designed to specifically target the ligand-binding pocket of a PDZ domain. Iterative molecular docking and modeling allowed the design of an indole scaffold 10a as a reversible inhibitor of ligand binding. The 10a scaffold inhibited the interaction between MAGI-3 and PTEN and showed cellular activities that are consistent with the inhibition of NHERF-1 function.

Estimation of Absolute Free Energies of Hydration Using Continuum Methods: Accuracy of Partial Charge Models and Optimization of Nonpolar Contributions.

Absolute free energies of hydration (ΔGhyd) for more than 500 neutral and charged compounds have been computed, using Poisson-Boltzmann (PB) and Generalized Born (GB) continuum methods plus a solvent-accessible surface area (SA) term, to evaluate the accuracy of eight simple point-charge models used in molecular modeling. The goal is to develop improved procedures and protocols for protein-ligand binding calculations and virtual screening (docking). The best overall PBSA and GBSA results, in comparison with experimental ΔGhyd values for small molecules, were obtained using MSK, RESP, or ChelpG charges obtained from ab initio calculations using 6-31G* wave functions. Correlations using semiempirical (AM1BCC, AM1CM2, and PM3CM2) or empirical (Gasteiger-Marsili and MMFF94) methods yielded mixed results, particularly for charged compounds. For neutral compounds, the AM1BCC method yielded the best agreement with experimental results. In all cases, the PBSA and GBSA results are highly correlated (overall r(2) = 0.94), which highlights the fact that various partial charge models influence the final results much more than which continuum method is used to compute hydration free energies. Overall improved agreement with experimental results was demonstrated using atom-based constants in place of a single surface area term. Sets of optimized SA constants, suitable for use with a given charge model, were derived by fitting to the difference in experimental free energies and polar continuum results. The use of optimized atom-based SA constants for the computation of ΔGhyd can fine-tune already reasonable agreement with experimental results, ameliorate gross deficiencies in any particular charge model, account for nonoptimal radii, or correct for systematic errors.

Development and validation of a modular, extensible docking program: DOCK 5.

We report on the development and validation of a new version of DOCK. The algorithm has been rewritten in a modular format, which allows for easy implementation of new scoring functions, sampling methods and analysis tools. We validated the sampling algorithm with a test set of 114 protein-ligand complexes. Using an optimized parameter set, we are able to reproduce the crystal ligand pose to within 2 A of the crystal structure for 79% of the test cases using our rigid ligand docking algorithm with an average run time of 1 min per complex and for 72% of the test cases using our flexible ligand docking algorithm with an average run time of 5 min per complex. Finally, we perform an analysis of the docking failures in the test set and determine that the sampling algorithm is generally sufficient for the binding pose prediction problem for up to 7 rotatable bonds; i.e. 99% of the rigid ligand docking cases and 95% of the flexible ligand docking cases are sampled successfully. We point out that success rates could be improved through more advanced modeling of the receptor prior to docking and through improvement of the force field parameters, particularly for structures containing metal-based cofactors.

Proteomimetic libraries: design, synthesis, and evaluation of p53-MDM2 interaction inhibitors.

The p53-MDM2 interaction regulates p53-mediated cellular responses to DNA damage, and MDM2 is overexpressed in 7% of all cancers. Structure-based computational design was applied to this system to design libraries centered on a scaffold that projects side chain functionalities with distance and angular relationships equivalent to those seen in the MDM2 interacting motif of p53. A library of 173 such compounds was synthesized using solution phase parallel chemistry. The in vitro competitive ability of the compounds to block p53 peptide binding to MDM2 was determined using a fluorescence polarization competition assay. The most active compound bound with K(d) = 12 microM, and its binding was characterized by (15)N-(1)H HSQC NMR.

An information theoretic approach to macromolecular modeling: I. Sequence alignments.

We are interested in applying the principles of information theory to structural biology calculations. In this article, we explore the information content of an important computational procedure: sequence alignment. Using a reference state developed from exhaustive sequences, we measure alignment statistics and evaluate gap penalties based on first-principle considerations and gap distributions. We show that there are different gap penalties for different alphabet sizes and that the gap penalties can depend on the length of the sequences being aligned. In a companion article, we examine the information content of molecular force fields.