Virtual screening of combinatorial libraries across a gene family: in search of inhibitors of Giardia lamblia guanine phosphoribosyltransferase.

Parasitic protozoa lack the ability to synthesize purine nucleotides de novo, relying instead on purine salvage enzymes for their survival. Guanine phosphoribosyltransferase (GPRT) from the protozoan parasite Giardia lamblia is a potential target for rational antiparasitic drug design, based on the experimental evidence, which indicates the lack of interconversion between adenine and guanine nucleotide pools. The present study is a continuation of our efforts to use three-dimensional structures of parasitic phosphoribosyltransferases (PRTs) to design novel antiparasitic agents. Two micromolar phthalimide-based GPRT inhibitors were identified by screening the in-house phthalimide library. A combination of structure-based scaffold selection using virtual library screening across the PRT gene family and solid phase library synthesis led to identification of smaller (molecular weight, <300) ligands with moderate to low specificity for GPRT; the best inhibitors, GP3 and GP5, had K(i) values in the 23 to 25 microM range. These results represent significant progress toward the goal of designing potent inhibitors of purine salvage in Giardia parasites. As a second step in this process, altering the phthalimide moiety to optimize interactions in the guanine-binding pocket of GPRT is expected to lead to compounds with promising activity against G. lamblia PRT.

DOCK 4.0: search strategies for automated molecular docking of flexible molecule databases.

In this paper we describe the search strategies developed for docking flexible molecules to macomolecular sites that are incorporated into the widely distributed DOCK software, version 4.0. The search strategies include incremental construction and random conformation search and utilize the existing Coulombic and Lennard-Jones grid-based scoring function. The incremental construction strategy is tested with a panel of 15 crystallographic testcases, created from 12 unique complexes whose ligands vary in size and flexibility. For all testcases, at least one docked position is generated within 2 A of the crystallographic position. For 7 of 15 testcases, the top scoring position is also within 2 A of the crystallographic position. The algorithm is fast enough to successfully dock a few testcases within seconds and most within 100 s. The incremental construction and the random search strategy are evaluated as database docking techniques with a database of 51 molecules docked to two of the crystallographic testcases. Incremental construction outperforms random search and is fast enough to reliably rank the database of compounds within 15 s per molecule on an SGI R10000 cpu.

Adenosine analogues as selective inhibitors of glyceraldehyde-3-phosphate dehydrogenase of Trypanosomatidae via structure-based drug design.

In our continuation of the structure-based design of anti-trypanosomatid drugs, parasite-selective adenosine analogues were identified as low micromolar inhibitors of glyceraldehyde-3-phosphate dehydrogenase (GAPDH). Crystal structures of Trypanosoma brucei, Trypanosoma cruzi, Leishmania mexicana, and human GAPDH's provided details of how the adenosyl moiety of NAD(+) interacts with the proteins, and this facilitated the understanding of the relative affinities of a series of adenosine analogues for the various GAPDH's. From exploration of modifications of the naphthalenemethyl and benzamide substituents of a lead compound, N(6)-(1-naphthalenemethyl)-2'-deoxy-2'-(3-methoxybenzamido)adenosine (6e), N(6)-(substituted-naphthalenemethyl)-2'-deoxy-2'-(substituted-benzamido)adenosine analogues were investigated. N(6)-(1-Naphthalenemethyl)-2'-deoxy-2'-(3,5-dimethoxybenzamido)adenosine (6m), N(6)-[1-(3-hydroxynaphthalene)methyl]-2'-deoxy-2'-(3,5-dimethoxybenzamido)adenosine (7m), N(6)-[1-(3-methoxynaphthalene)methyl]-2'-deoxy-2'-(3,5-dimethoxybenzamido)adenosine (9m), N(6)-(2-naphthalenemethyl)-2'-deoxy-2'-(3-methoxybenzamido)adenosine (11e), and N(6)-(2-naphthalenemethyl)-2'-deoxy-2'-(3,5-dimethoxybenzamido)adenosine (11m) demonstrated a 2- to 3-fold improvement over 6e and a 7100- to 25000-fold improvement over the adenosine template. IC(50)'s of these compounds were in the range 2-12 microM for T. brucei, T. cruzi, and L. mexicana GAPDH's, and these compounds did not inhibit mammalian GAPDH when tested at their solubility limit. To explore more thoroughly the structure-activity relationships of this class of compounds, a library of 240 N(6)-(substituted)-2'-deoxy-2'-(amido)adenosine analogues was generated using parallel solution-phase synthesis with N(6) and C2' substituents chosen on the basis of computational docking scores. This resulted in the identification of 40 additional compounds that inhibit parasite GAPDH's in the low micromolar range. We also explored adenosine analogues containing 5'-amido substituents and found that 2',5'-dideoxy-2'-(3,5-dimethoxybenzamido)-5'-(diphenylacetamido)adenosine (49) displays an IC(50) of 60-100 microM against the three parasite GAPDH's.

Conformation spaces of proteins.

We report a simple method for measuring the accessible conformational space explored by an ensemble of protein structures. The method is useful for diverse ensembles derived from molecular dynamics trajectories, molecular modeling, and molecular structure determinations. It can be used to examine a wide range of time scales. The central tactic we use, which has been previously employed, is to replace the true mechanical degrees of freedom of a molecular system with the conformationally effective degrees of freedom as measured by the root-mean squared cartesian distances among all pairs of conformations. Each protein conformation is treated as a point in a high dimensional euclidean space. In this article, we model this space in a novel way by representing it as an N-dimensional hypercube, describable with only two parameters: the number of dimensions and the edge length. To validate this approach, we provide a number of elementary test cases and then use the N-cube method for measuring the size and shape of conformational space covered by molecular dynamics trajectories spanning 10 orders of magnitude in time. These calculations were performed on a small protein, the villin headpiece subdomain, exploring both the native state and the misfolded/folding regime. Distinct features include single, vibrationally averaged, substate minima on the 0.1-1-ps time scale, thermally averaged conformational states that persist for 1-100 ps and transitions between these local minima on nanosecond time scales. Large-scale refolding modes appear to become uncorrelated on the microsecond time scale. Associated length scales for these events are 0.2 A for the vibrational minima; 0.5 A for the conformational minima; and 1-2 A for the nanosecond events. We find that the conformational space that is dynamically accessible during folding of villin has enough volume for approximately 10(9) minima of the variety that persist for picoseconds. Molecular dynamics trajectories of the native protein and experimentally derived solution ensembles suggest the native state to be composed of approximately 10(2) of these thermally accessible minima. Thus, based on random exploration of accessible folding space alone, protein folding for a small protein is predicted to be a milliseconds time scale event. This time can be compared with the experimental folding time for villin of 10-100 micros. One possible explanation for the 10-100-fold discrepancy is that the slope of the "folding funnel" increases the rate 1-2 orders of magnitude above random exploration of substates.

Design, docking, and evaluation of multiple libraries against multiple targets.

We present a general approach to the design, docking, and virtual screening of multiple combinatorial libraries against a family of proteins. The method consists of three main stages: docking the scaffold, selecting the best substituents at each site of diversity, and comparing the resultant molecules within and between the libraries. The core "divide-and-conquer" algorithm for side-chain selection, developed from an earlier version (Sun et al., J Comp Aided Mol Design 1998;12:597-604), provides a way to explore large lists of substituents with linear rather than combinatorial time dependence. We have applied our method to three combinatorial libraries and three serine proteases: trypsin, chymotrypsin, and elastase. We show that the scaffold docking procedure, in conjunction with a novel vector-based orientation filter, reproduces crystallographic binding modes. In addition, the free-energy-based scoring procedure (Zou et al., J Am Chem Soc 1999;121:8033-8043) is able to reproduce experimental binding data for P1 mutants of macromolecular protease inhibitors. Finally, we show that our method discriminates between a peptide library and virtual libraries built on benzodiazepine and tetrahydroisoquinolinone scaffolds. Implications of the docking results for library design are explored.

A genetic algorithm for structure-based de novo design.

Genetic algorithms have properties which make them attractive in de novo drug design. Like other de novo design programs, genetic algorithms require a method to reduce the enormous search space of possible compounds. Most often this is done using information from known ligands. We have developed the ADAPT program, a genetic algorithm which uses molecular interactions evaluated with docking calculations as a fitness function to reduce the search space. ADAPT does not require information about known ligands. The program takes an initial set of compounds and iteratively builds new compounds based on the fitness scores of the previous set of compounds. We describe the particulars of the ADAPT algorithm and its application to three well-studied target systems. We also show that the strategies of enhanced local sampling and re-introducing diversity to the compound population during the design cycle provide better results than conventional genetic algorithm protocols.

High throughput protein fold identification by using experimental constraints derived from intramolecular cross-links and mass spectrometry.

We have used intramolecular cross-linking, MS, and sequence threading to rapidly identify the fold of a model protein, bovine basic fibroblast growth factor (FGF)-2. Its tertiary structure was probed with a lysine-specific cross-linking agent, bis(sulfosuccinimidyl) suberate (BS(3)). Sites of cross-linking were determined by tryptic peptide mapping by using time-of-flight MS. Eighteen unique intramolecular lysine (Lys-Lys) cross-links were identified. The assignments for eight cross-linked peptides were confirmed by using post source decay MS. The interatomic distance constraints were all consistent with the tertiary structure of FGF-2. These relatively few constraints, in conjunction with threading, correctly identified FGF-2 as a member of the beta-trefoil fold family. To further demonstrate utility, we used the top-scoring homolog, IL-1beta, to build an FGF-2 homology model with a backbone error of 4.8 A (rms deviation). This method is fast, is general, uses small amounts of material, and is amenable to automation.

Rational design of selective submicromolar inhibitors of Tritrichomonas foetus hypoxanthine-guanine-xanthine phosphoribosyltransferase.

All parasitic protozoa lack the ability to synthesize purine nucleotides de novo, relying instead on purine salvage enzymes for their survival. Hypoxanthine-guanine-xanthine phosphoribosyltransferase (HGXPRT) from the protozoan parasite Tritrichomonas foetus is a rational target for antiparasitic drug design because it is the primary enzyme the parasite uses to salvage purine bases from the host. The study presented here is a continuation of our efforts to use the X-ray structure of the T. foetus HGXPRT-GMP complex to design compounds that bind tightly to the purine pocket of HGXPRT. The goal of the current project was to improve the affinity and selectivity of previously identified HGXPRT inhibitor TF1 [4-(3-nitroanilino)phthalic anhydride]. A virtual library of substituted 4-phthalimidocarboxanilides was constructed using methods of structure-based drug design, and was implemented synthetically on solid support. Compound 20 [(4'-phthalimido)carboxamido-3-benzyloxybenzene] was then used as a secondary lead for the second round of combinatorial chemistry, producing a number of low-micromolar inhibitors of HGXPRT. One of these compounds, TF2 [(4'-phthalimido)carboxamido-3-(4-bromobenzyloxy)benzene], was further characterized as a competitive inhibitor of T. foetus HGXPRT with respect to guanine with a K(I) of 0.49 microM and a 30-fold selectivity over the human HGPRT. TF2 inhibited the growth of cultured T. foetus cells in a concentration-dependent manner with an ED(50) of 2.8 microM, and this inhibitory effect could be reversed by addition of exogenous hypoxanthine. These studies underscore the efficiency of combining structure-based drug design with combinatorial chemistry to produce effective species-specific enzyme inhibitors of medicinal importance.

Novel cathepsin D inhibitors block the formation of hyperphosphorylated tau fragments in hippocampus.

Lysosomal disturbances may be a contributing factor to Alzheimer's disease. We used novel compounds to test if suppression of the lysosomal protease cathepsin D blocks production of known precursors to neurofibrillary tangles. Partial lysosomal dysfunction was induced in cultured hippocampal slices with a selective inhibitor of cathepsins B and L. This led within 48 h to hyperphosphorylated tau protein fragments recognized by antibodies against human tangles. Potent nonpeptidic cathepsin D inhibitors developed using combinatorial chemistry and structure-based design blocked production of the fragments in a dose-dependent fashion. Threshold was in the submicromolar range, with higher concentrations producing complete suppression. The effects were selective and not accompanied by pathophysiology. Comparable results were obtained with three structurally distinct inhibitors. These results support the hypothesis that cathepsin D links lysosomal dysfunction to the etiology of Alzheimer's disease and suggest a new approach to treating the disease.

Inhibitors of kinesin activity from structure-based computer screening.

Kinesin motor proteins use ATP hydrolysis for transport along microtubules in the cell. We sought to identify small organic ligands to inhibit kinesin's activity. Candidate molecules were identified by computational docking of commercially available compounds using the computer program DOCK. Compounds were docked at either of two sites, and a selection was tested for inhibition of microtubule-stimulated ATPase activity. Twenty-two submillimolar inhibitors were identified. Several inhibitors appeared to be competitive for microtubule binding and not for ATP binding, and three compounds showed 50% inhibition down to single-digit micromolar levels. Most inhibitors grouped into four distinct classes (fluoresceins, phenolphthaleins, anthraquinones, and naphthylene sulfonates). We measured the binding of one inhibitor, rose bengal lactone (RBL), to kinesin (dissociation constant 2.5 microM) by its increase in steady-state fluorescence anisotropy. The RBL binding site on kinesin was localized by fluorescent resonance energy transfer (FRET) using a donor fluorophore (coumarin) covalently attached at unique, surface-exposed cysteine residues engineered at positions 28, 149, 103, 220, or 330. RBL was found to bind in its original docked site: the pocket cradled by loop 8 and beta-strand 5 in kinesin's three-dimensional structure. These results confirm this region's role in microtubule binding and identify this pocket as a novel binding site for kinesin inhibition.

Development and screening of a polyketide virtual library for drug leads against a motilide pharmacophore.

A virtual library of macrocyclic polyketide molecules was generated and screened to identify novel, conformationally constrained potential motilin receptor agonists ("motilides"). A motilide pharmacophore model was generated from the potent 6,9-enol ether erythromycin and known derivatives from the literature. The pharmacophore for each molecular conformation was a point in a distance-volume space based on presentation of the putative binding moieties. Two methods, one fragment based method and the other reaction based, were explored for constructing the polyketide virtual library. First, a virtual library was assembled from monomeric fragments using the CHORTLES language. Second, the virtual library was assembled by the in silico application of all possible polyketide synthase enzyme reactions to generate the product library. Each library was converted to low-energy 3D conformations by distance geometry and standard minimization methods. The distance-volume metric was calculated for low-energy conformations of the members of the virtual polyketide library and screened against the enol ether pharmacophore. The goal was to identify novel macrocycles that satisfy the pharmacophore. We identified three conformationally constrained, novel polyketide series that have low-energy conformations satisfying the distance-volume constraints of the motilide pharmacophore.

DREAM++: flexible docking program for virtual combinatorial libraries.

We present a set of programs, DREAM+2 (Docking and Reaction programs using Efficient seArch Methods written in C++), for docking computationally generated ligands into macromolecular binding sites. DREAM++ is composed of three programs: ORIENT++, REACT++ and SEARCH++. The program ORIENT++ positions molecules in a binding site with the DOCK algorithm. Its output can be used as input to REACT++ and SEARCH+2. The program REACT++ performs user-specific chemical reactions on a docked molecule, so that reaction products can be evaluated for three dimensional complementarity with the macromolecular site. The program SEARCH++ performs an efficient conformation search on the reaction products using a hybrid backtrack and incremental construction algorithm. We have applied the programs to HIV protease-inhibitor complexes as test systems. We found that we can differentiate high-affinity ligands based on several measures: interaction energies, occupancy of protein subsites and the number of successfully docked conformations for each product. Encouraged by the results in the test case, we applied the programs to propose novel inhibitors of HIV protease. These inhibitors can be generated by organic reactions using commercially available reagents. They are alternatives to the inhibitors synthesized by Glaxo.

The maximal affinity of ligands.

We explore the question of what are the best ligands for macromolecular targets. A survey of experimental data on a large number of the strongest-binding ligands indicates that the free energy of binding increases with the number of nonhydrogen atoms with an initial slope of approximately -1.5 kcal/mol (1 cal = 4.18 J) per atom. For ligands that contain more than 15 nonhydrogen atoms, the free energy of binding increases very little with relative molecular mass. This nonlinearity is largely ascribed to nonthermodynamic factors. An analysis of the dominant interactions suggests that van der Waals interactions and hydrophobic effects provide a reasonable basis for understanding binding affinities across the entire set of ligands. Interesting outliers that bind unusually strongly on a per atom basis include metal ions, covalently attached ligands, and a few well known complexes such as biotin-avidin.

Flexible ligand docking: a multistep strategy approach.

A flexible ligand docking protocol based on a divide-and-conquer strategy is investigated. This approach first separates total search space into conformation and orientation space. It uses a grid-based method to sample the conformation of an unbound ligand and to select the low-energy conformers. Rigid docking is then carried out to locate the low-energy binding orientations for these conformers. These docking structures are subsequently subjected to structure refinement including molecular mechanics minimization, conformational scanning at the binding site and a short period of molecular dynamics-based simulated annealing. This approach has been applied to twelve ligand-protein complexes with three to sixteen rotatable bonds. The docked lowest-energy structures have root mean square deviations ranging from 0.64 A to 2.01 A with respect to the corresponding crystal structures. The effect of atomic charges and van der Waals parameters on the docking results, and the role of the dielectric constant in the conformation sampling are discussed in detail. A fragment-based docking approach that takes advantages of the divide-and-conquer strategy has also been explored and the results are compared with those produced by a whole molecule-based approach.

Macromolecular docking of a three-body system: the recognition of human growth hormone by its receptor.

Human growth hormone (hGH) binds to its receptor (hGHr) in a three-body interaction: one molecule of the hormone and two identical monomers of the receptor form a trimer. Curiously, the hormone-receptor interactions in the trimer are not equivalent and the formation of the complex occurs in a specific kinetic order (Cunningham BC, Ultsch M, De Vos AM, Mulkerrin MG, Clauser KR, Wells JA, 1991, Science 254:821-825). In this paper, we model the recognition of hGH to the hGHr using shape complementarity of the three-dimensional structures and macromolecular docking to explore possible binding modes between the receptor and hormone. The method, reported previously (Hendrix DK, Kuntz ID, 1998, Pacific symposium on biocomputing 1998, pp 1234-1244), is based upon matching complementary-shaped strategic sites on the molecular surface. We modify the procedure to examine three-body systems. We find that the order of binding seen experimentally is also essential to our model. We explore the use of mutational data available for hGH to guide our model. In addition to docking hGH to the hGHr, we further test our methodology by successfully reproducing 16 macromolecular complexes from X-ray crystal structures, including enzyme-inhibitor, antibody-antigen, protein dimer, and protein-DNA complexes.

Ligand solvation in molecular docking.

Solvation plays an important role in ligand-protein association and has a strong impact on comparisons of binding energies for dissimilar molecules. When databases of such molecules are screened for complementarity to receptors of known structure, as often occurs in structure-based inhibitor discovery, failure to consider ligand solvation often leads to putative ligands that are too highly charged or too large. To correct for the different charge states and sizes of the ligands, we calculated electrostatic and non-polar solvation free energies for molecules in a widely used molecular database, the Available Chemicals Directory (ACD). A modified Born equation treatment was used to calculate the electrostatic component of ligand solvation. The non-polar component of ligand solvation was calculated based on the surface area of the ligand and parameters derived from the hydration energies of apolar ligands. These solvation energies were subtracted from the ligand-receptor interaction energies. We tested the usefulness of these corrections by screening the ACD for molecules that complemented three proteins of known structure, using a molecular docking program. Correcting for ligand solvation improved the rankings of known ligands and discriminated against molecules with inappropriate charge states and sizes.

Potent, low-molecular-weight non-peptide inhibitors of malarial aspartyl protease plasmepsin II.

A number of single-digit nanomolar, low-molecular-weight plasmepsin II aspartyl protease inhibitors have been identified using combinatorial chemistry and structure-based design. By identifying multiple, small-molecule inhibitors using the parallel synthesis of several focused libraries, it was possible to select for compounds with desirable characteristics including enzyme specificity and minimal binding to serum proteins. The best inhibitors identified have Ki's of 2-10 nM, molecular weights between 594 and 650 Da, between 3- and 15-fold selectivity toward plasmepsin II over cathepsin D, the most closely related human protease, good calculated log P values (2.86-4.56), and no apparent binding to human serum albumin at 1 mg/mL in an in vitro assay. These compounds represent the most potent non-peptide plasmepsin II inhibitors reported to date.

A rapid method for exploring the protein structure universe.

We have developed an automatic protein fingerprinting method for the evaluation of protein structural similarities based on secondary structure element compositions, spatial arrangements, lengths, and topologies. This method can rapidly identify proteins sharing structural homologies as we demonstrate with five test cases: the globins, the mammalian trypsinlike serine proteases, the immunoglobulins, the cupredoxins, and the actinlike ATPase domain-containing proteins. Principal components analysis of the similarity distance matrix calculated from an all-by-all comparison of 1,031 unique chains in the Protein Data Bank has produced a distribution of structures within a high-dimensional structural space. Fifty percent of the variance observed for this distribution is bounded by six axes, two of which encode structural variability within two large families, the immunoglobulins and the trypsinlike serine proteases. Many aspects of the spatial distribution remain stable upon reduction of the database to 140 proteins with minimal family overlap. The axes correlated with specific structural families are no longer observed. A clear hierarchy of organization is seen in the arrangement of protein structures in the universe. At the highest level, protein structures populate regions corresponding to the all-alpha, all-beta, and alpha/beta superfamilies. Large protein families are arranged along family-specific axes, forming local densely populated regions within the space. The lowest level of organization is intrafamilial; homologous structures are ordered by variations in peripheral secondary structure elements or by conformational shifts in the tertiary structure.

Structure-based design of inhibitors specific for bacterial thymidylate synthase.

Thymidylate synthase is an attractive target for antiproliferative drug design because of its key role in the synthesis of DNA. As such, the enzyme has been widely targeted for anticancer applications. In principle, TS should also be a good target for drugs used to fight infectious disease. In practice, TS is highly conserved across species, and it has proven to be difficult to develop inhibitors that are selective for microbial TS enzymes over the human enzyme. Using the structure of TS from Lactobacillus casei in complex with the nonsubstrate analogue phenolphthalein, inhibitors were designed to take advantage of features of the bacterial enzyme that differ from those of the human enzyme. Upon synthesis and testing, these inhibitors were found to be up to 40-fold selective for the bacterial enzyme over the human enzyme. The crystal structures of two of these inhibitors in complex with TS suggested the design of further compounds. Subsequent synthesis and testing showed that these second-round compounds inhibit the bacterial enzyme at sub-micromolar concentrations, while the human enzyme was not inhibited at detectable levels (selectivities of 100-1000-fold or greater). Although these inhibitors share chemical similarities, X-ray crystal structures reveal that the analogues bind to the enzyme in substantially different orientations. Site-directed mutagenesis experiments suggest that the individual inhibitors may adopt multiple configurations in their complexes with TS.