Two orders of magnitude more protein sequences can be modeled by comparative modeling than have been determined by X-ray crystallography and NMR spectroscopy. Investigators have nevertheless been cautious about using comparative models for ligand discovery because of concerns about model errors. We suggest how to exploit comparative models for molecular screens, based on docking against a wide range of crystallographic structures and comparative models with known ligands. To account for the variation in the ligand-binding pocket as it binds different ligands, we calculate “consensus” enrichment by ranking each library compound by its best docking score against all available comparative models and/or modeling templates. For the majority of the targets, the consensus enrichment for multiple models was better than or comparable to that of the holo and apo X-ray structures. Even for single models, the models are significantly more enriching than the template structure if the template is paralogous and shares more than 25% sequence identity with the target.

Applications in structural biology and medicinal chemistry require protein–ligand scoring functions for two distinct tasks: (i) ranking different poses of a small molecule in a protein binding site and (ii) ranking different small molecules by their complementarity to a protein site. Using probability theory, we developed two atomic distance-dependent statistical scoring functions: PoseScore was optimized for recognizing native binding geometries of ligands from other poses and RankScore was optimized for distinguishing ligands from nonbinding molecules. Both scores are based on a set of 8,885 crystallographic structures of protein–ligand complexes but differ in the values of three key parameters. Factors influencing the accuracy of scoring were investigated, including the maximal atomic distance and non-native ligand geometries used for scoring, as well as the use of protein models instead of crystallographic structures for training and testing the scoring function. For the test set of 19 targets, RankScore improved the ligand enrichment (logAUC) and early enrichment (EF1) scores computed by DOCK 3.6 for 13 and 14 targets, respectively. In addition, RankScore performed better at rescoring than each of seven other scoring functions tested. Accepting both the crystal structure and decoy geometries with all-atom root-mean-square errors of up to 2 Å from the crystal structure as correct binding poses, PoseScore gave the best score to a correct binding pose among 100 decoys for 88% of all cases in a benchmark set containing 100 protein–ligand complexes. PoseScore accuracy is comparable to that of DrugScoreCSD and ITScore/SE and superior to 12 other tested scoring functions. Therefore, RankScore can facilitate ligand discovery, by ranking complexes of the target with different small molecules; PoseScore can be used for protein–ligand complex structure prediction, by ranking different conformations of a given protein–ligand pair. The statistical potentials are available through the Integrative Modeling Platform (IMP) software package (http://salilab.org/imp/) and the LigScore Web server (http://salilab.org/ligscore/).

The binding of drugs and reagents to off-targets is well-known. Whereas many off-targets are related to the primary target by sequence and fold, many ligands bind to unrelated pairs of proteins, and these are harder to anticipate. If the binding site in the off-target can be related to that of the primary target, this challenge resolves into aligning the two pockets. However, other cases are possible: the ligand might interact with entirely different residues and environments in the off-target, or wholly different ligand atoms may be implicated in the two complexes. To investigate these scenarios at atomic resolution, the structures of 59 ligands in 116 complexes (62 pairs in total), where the protein pairs were unrelated by fold but bound an identical ligand, were examined. In almost half of the pairs, the ligand interacted with unrelated residues in the two proteins (29 pairs), and in 14 of the pairs wholly different ligand moieties were implicated in each complex. Even in those 19 pairs of complexes that presented similar environments to the ligand, ligand superposition rarely resulted in the overlap of related residues. There appears to be no single pattern-matching “code” for identifying binding sites in unrelated proteins that bind identical ligands, though modeling suggests that there might be a limited number of different patterns that suffice to recognize different ligand functional groups.

A prospective, large library virtual screen against an activated β2-adrenergic receptor (β2AR) structure returned potent agonists to the exclusion of inverse-agonists, providing the first complement to the previous virtual screening campaigns against inverse-agonist-bound G protein coupled receptor (GPCR) structures, which predicted only inverse-agonists. In addition, two hits recapitulated the signaling profile of the co-crystal ligand with respect to the G protein and arrestin mediated signaling. This functional fidelity has important implications in drug design, as the ability to predict ligands with predefined signaling properties is highly desirable. However, the agonist-bound state provides an uncertain template for modeling the activated conformation of other GPCRs, as a dopamine D2 receptor (DRD2) activated model templated on the activated β2AR structure returned few hits of only marginal potency.

The substrate specificities of two incorrectly annotated enzymes belonging to cog3964 from the amidohydrolase superfamily were determined. This group of enzymes are currently misannotated as either dihydroorotases or adenine deaminases. Atu3266 from Agrobacterium tumefaciens C58 and Oant2987 from Ochrobactrum anthropi ATCC 49188 were found to catalyze the hydrolysis of acetyl-(R)-mandelate and similar esters with values of kcat/Km that exceed 105 M–1 s–1. These enzymes do not catalyze the deamination of adenine or the hydrolysis of dihydroorotate. Atu3266 was crystallized and the structure determined to a resolution of 2.62 Å. The protein folds as a distorted (β/α)8 barrel and binds two zincs in the active site. The substrate profile was determined via a combination of computational docking to the three-dimensional structure of Atu3266 and screening of a highly focused library of potential substrates. The initial weak hit was the hydrolysis of N-acetyl-d-serine (kcat/Km = 4 M–1 s–1). This was followed by the progressive identification of acetyl-(R)-glycerate (kcat/Km = 4 × 102 M–1 s–1), acetyl glycolate (kcat/Km = 1.3 × 104 M–1 s–1), and ultimately acetyl-(R)-mandelate (kcat/Km = 2.8 × 105 M–1 s–1).

Of the over 22 million protein sequences in the nonredundant TrEMBL database, fewer than 1% have experimentally confirmed functions. Structure-based methods have been used to predict enzyme activities from experimentally determined structures; however, for the vast majority of proteins, no such structures are available. Here, homology models of a functionally uncharacterized amidohydrolase from Agrobacterium radiobacter K84 (Arad3529) were computed on the basis of a remote template structure. The protein backbone of two loops near the active site was remodeled, resulting in four distinct active site conformations. Substrates of Arad3529 were predicted by docking of 57 672 high-energy intermediate (HEI) forms of 6440 metabolites against these four homology models. On the basis of docking ranks and geometries, a set of modified pterins were suggested as candidate substrates for Arad3529. The predictions were tested by enzymology experiments, and Arad3529 deaminated many pterin metabolites (substrate, kcat/Km [M–1 s–1]): formylpterin, 5.2 × 106; pterin-6-carboxylate, 4.0 × 106; pterin-7-carboxylate, 3.7 × 106; pterin, 3.3 × 106; hydroxymethylpterin, 1.2 × 106; biopterin, 1.0 × 106; d-(+)-neopterin, 3.1 × 105; isoxanthopterin, 2.8 × 105; sepiapterin, 1.3 × 105; folate, 1.3 × 105, xanthopterin, 1.17 × 105; and 7,8-dihydrohydroxymethylpterin, 3.3 × 104. While pterin is a ubiquitous oxidative product of folate degradation, genomic analysis suggests that the first step of an undescribed pterin degradation pathway is catalyzed by Arad3529. Homology model-based virtual screening, especially with modeling of protein backbone flexibility, may be broadly useful for enzyme function annotation and discovering new pathways and drug targets.

A key metric to assess molecular docking remains ligand enrichment against challenging decoys. Whereas the directory of useful decoys (DUD) has been widely used, clear areas for optimization have emerged. Here we describe an improved benchmarking set that includes more diverse targets such as GPCRs and ion channels, totaling 102 proteins with 22886 clustered ligands drawn from ChEMBL, each with 50 property-matched decoys drawn from ZINC. To ensure chemotype diversity, we cluster each target’s ligands by their Bemis–Murcko atomic frameworks. We add net charge to the matched physicochemical properties and include only the most dissimilar decoys, by topology, from the ligands. An online automated tool (http://decoys.docking.org) generates these improved matched decoys for user-supplied ligands. We test this data set by docking all 102 targets, using the results to improve the balance between ligand desolvation and electrostatics in DOCK 3.6. The complete DUD-E benchmarking set is freely available at http://dude.docking.org.

Many small molecules, including bioactive molecules and approved drugs, spontaneously form colloidal aggregates in aqueous solution at micromolar concentrations. Though it is widely accepted that aggregation leads to artifacts in screens for ligands of soluble proteins, the effects of colloid formation in cell-based assays have not been studied. Here, seven anticancer drugs and one diagnostic reagent were found to form colloids in both biochemical buffer and in cell culture media. In cell-based assays, the antiproliferative activities of three of the drugs were substantially reduced when in colloidal form as compared to monomeric form; a new formulation method ensured the presence of drug colloids versus drug monomers in solution. We also found that Evans Blue, a dye classically used to measure vascular permeability and to demonstrate the “enhanced permeability and retention (EPR) effect” in solid tumors, forms colloids that adsorb albumin, as opposed to older literature that suggested the reverse.

Two enzymes of unknown function from the cog1735 subset of the amidohydrolase superfamily (AHS), LMOf2365_2620 (Lmo2620) from Listeria monocytogenes str. 4b F2365 and Bh0225 from Bacillus halodurans C-125, were cloned, expressed, and purified to homogeneity. The catalytic functions of these two enzymes were interrogated by an integrated strategy encompassing bioinformatics, computational docking to three-dimensional crystal structures, and library screening. The three-dimensional structure of Lmo2620 was determined at a resolution of 1.6 Å with two phosphates and a binuclear zinc center in the active site. The proximal phosphate bridges the binuclear metal center and is 7.1 Å from the distal phosphate. The distal phosphate hydrogen bonds with Lys-242, Lys-244, Arg-275, and Tyr-278. Enzymes within cog1735 of the AHS have previously been shown to catalyze the hydrolysis of substituted lactones. Computational docking of the high-energy intermediate form of the KEGG database to the three-dimensional structure of Lmo2620 highly enriched anionic lactones versus other candidate substrates. The active site structure and the computational docking results suggested that probable substrates would likely include phosphorylated sugar lactones. A small library of diacid sugar lactones and phosphorylated sugar lactones was synthesized and tested for substrate activity with Lmo2620 and Bh0225. Two substrates were identified for these enzymes, d-lyxono-1,4-lactone-5-phosphate and l-ribono-1,4-lactone-5-phosphate. The kcat/Km values for the cobalt-substituted enzymes with these substrates are ∼105 M–1 s–1.

G-protein–coupled receptors (GPCRs) are key signaling molecules and are intensely studied. Whereas GPCRs recognizing small-molecules have been successfully targeted for drug discovery, protein-recognizing GPCRs, such as the chemokine receptors, claim few drugs or even useful small molecule reagents. This reflects both the difficulties that attend protein–protein interface inhibitor discovery, and the lack of structures for these targets. Imminent structure determination of chemokine receptor CXCR4 motivated docking screens for new ligands against a homology model and subsequently the crystal structure. More than 3 million molecules were docked against the model and then against the crystal structure; 24 and 23 high-scoring compounds from the respective screens were tested experimentally. Docking against the model yielded only one antagonist, which resembled known ligands and lacked specificity, whereas the crystal structure docking yielded four that were dissimilar to previously known scaffolds and apparently specific. Intriguingly, several were potent and relatively small, with IC50 values as low as 306 nM, ligand efficiencies as high as 0.36, and with efficacy in cellular chemotaxis. The potency and efficiency of these molecules has few precedents among protein–protein interface inhibitors, and supports structure-based efforts to discover leads for chemokine GPCRs.

Synthetic cavitands and protein cavities have been widely studied as models for ligand recognition. Here we investigate the Met102 → His substitution in the artificial L99A cavity in T4 lysozyme as a Kemp eliminase. The resulting enzyme had kcat/KM = 0.43 M-1 s-1 and a (kcat/KM)/kuncat = 107 at pH 5.0. The crystal structure of this enzyme was determined at 1.30 Å, as were the structures of four complexes of substrate and product analogs. The absence of ordered waters or hydrogen bonding interactions, and the presence of a common catalytic base (His102) in an otherwise hydrophobic, buried cavity, facilitated detailed analysis of the reaction mechanism and its optimization. Subsequent substitutions increased eliminase activity by an additional four-fold. As activity-enhancing substitutions were engineered into the cavity, protein stability decreased, consistent with the stability-function trade-off hypothesis. This and related model cavities may provide templates for studying protein design principles in radically simplified environments.

Notwithstanding their key roles in therapy and as biological probes, 7% of approved drugs are purported to have no known primary target, and up to 18% lack a well-defined mechanism of action. Using a chemoinformatics approach, we sought to “de-orphanize” drugs that lack primary targets. Surprisingly, targets could be easily predicted for many: Whereas these targets were not known to us nor to the common databases, most could be confirmed by literature search, leaving only 13 Food and Drug Administration—approved drugs with unknown targets; the number of drugs without molecular targets likely is far fewer than reported. The number of worldwide drugs without reasonable molecular targets similarly dropped, from 352 (25%) to 44 (4%). Nevertheless, there remained at least seven drugs for which reasonable mechanism-of-action targets were unknown but could be predicted, including the antitussives clemastine, cloperastine, and nepinalone; the antiemetic benzquinamide; the muscle relaxant cyclobenzaprine; the analgesic nefopam; and the immunomodulator lobenzarit. For each, predicted targets were confirmed experimentally, with affinities within their physiological concentration ranges. Turning this question on its head, we next asked which drugs were specific enough to act as chemical probes. Over 100 drugs met the standard criteria for probes, and 40 did so by more stringent criteria. A chemical information approach to drug-target association can guide therapeutic development and reveal applications to probe biology, a focus of much current interest.

Fragment-based design was used to guide derivatization of a lead series of β-lactamase inhibitors that had heretofore resisted optimization for in vivo activity. X-ray structures of fragments overlaid with the lead suggested new, unanticipated functionality and points of attachment. Synthesis of three derivatives improved affinity over 20-fold and improved efficacy in cell culture. Crystal structures were consistent with the fragment-based design, enabling further optimization to a Ki of 50 pM, a 500-fold improvement that required the synthesis of only six derivatives. One of these, compound 5, was tested in mice. Whereas cefotaxime alone failed to cure mice infected with β-lactamase-expressing Escherichia coli, 65% were cleared of infection when treated with a cefotaxime:5 combination. Fragment complexes offer a path around design hurdles, even for advanced molecules; the series described here may provide leads to overcome β-lactamase-based resistance, a key clinical challenge.

The similarity ensemble approach (SEA) relates proteins based on the set-wise chemical similarity among their ligands. It can be used to rapidly search large compound databases and to build cross-target similarity maps. The emerging maps relate targets in ways that reveal relationships one might not recognize based on sequence or structural similarities alone. SEA has previously revealed cross talk between drugs acting primarily on G-protein coupled receptors (GPCRs). Here we used SEA to look for potential off-target inhibition of the enzyme protein farnesyltransferase (PFTase) by commercially available drugs. The inhibition of PFTase has profound consequences for oncogenesis, as well as a number of other diseases. In the present study, two commercial drugs, Loratadine and Miconazole, were identified as potential ligands for PFTase and subsequently confirmed as such experimentally. These results point toward the applicability of SEA for the prediction of not only GPCR−GPCR drug cross talk but also GPCR−enzyme and enzyme−enzyme drug cross talk.

We investigated a series of sulfonamide boronic acids that resulted from the merging of two unrelated AmpC β-lactamase inhibitor series. The new boronic acids differed in the replacement of the canonical carboxamide, found in all penicillin and cephalosporin antibiotics, with a sulfonamide. Surprisingly, these sulfonamides had a highly distinct structure−activity relationship from the previously explored carboxamides, high ligand efficiencies (up to 0.91), and Ki values down to 25 nM and up to 23 times better for smaller analogues. Conversely, Ki values were 10−20 times worse for larger molecules than in the carboxamide congener series. X-ray crystal structures (1.6−1.8 Å) of AmpC with three of the new sulfonamides suggest that this altered structure−activity relationship results from the different geometry and polarity of the sulfonamide versus the carboxamide. The most potent inhibitor reversed β-lactamase-mediated resistance to third generation cephalosporins, lowering their minimum inhibitory concentrations up to 32-fold in cell culture.

Virtual and high-throughput screens (HTS) should have complementary strengths and weaknesses, but studies that prospectively and comprehensively compare them are rare. We undertook a parallel docking and HTS screen of 197861 compounds against cruzain, a thiol protease target for Chagas disease, looking for reversible, competitive inhibitors. On workup, 99% of the hits were eliminated as false positives, yielding 146 well-behaved, competitive ligands. These fell into five chemotypes: two were prioritized by scoring among the top 0.1% of the docking-ranked library, two were prioritized by behavior in the HTS and by clustering, and one chemotype was prioritized by both approaches. Determination of an inhibitor/cruzain crystal structure and comparison of the high-scoring docking hits to experiment illuminated the origins of docking false-negatives and false-positives. Prioritizing molecules that are both predicted by docking and are HTS-active yields well-behaved molecules, relatively unobscured by the false-positives to which both techniques are individually prone.

Many organic molecules form colloidal aggregates in aqueous solution at micromolar concentrations. These aggregates promiscuously inhibit soluble proteins and are a major source of false positives in high-throughput screening. Several drugs also form colloidal aggregates, and there has been speculation that this may affect the absorption and distribution of at least one drug in vivo. Here we investigate the ability of drugs to form aggregates in simulated intestinal fluid. Thirty-three Biopharmaceutics Classification System (BCS) class II and class IV drugs, spanning multiple pharmacological activities, were tested for promiscuous aggregation in biochemical buffers. The 22 that behaved as aggregators were then tested for colloid formation in simulated intestinal fluid, a buffer mimicking conditions in the small intestine. Six formed colloids at concentrations equal to or lower than the concentrations reached in the gut, suggesting that aggregation may have an effect on the absorption and distribution of these drugs, and potentially others, in vivo.

The recent determination of X-ray structures of pharmacologically relevant GPCRs has made these targets accessible to structure-based ligand discovery. Here we explore whether novel chemotypes may be discovered for the A2A adenosine receptor, based on complementarity to its recently determined structure. The A2A adenosine receptor signals in the periphery and the CNS, with agonists explored as anti-inflammatory drugs and antagonists explored for neurodegenerative diseases. We used molecular docking to screen a 1.4 million compound database against the X-ray structure computationally and tested 20 high-ranking, previously unknown molecules experimentally. Of these 35% showed substantial activity with affinities between 200 nM and 9 μM. For the most potent of these new inhibitors, over 50-fold specificity was observed for the A2A versus the related A1 and A3 subtypes. These high hit rates and affinities at least partly reflect the bias of commercial libraries toward GPCR-like chemotypes, an issue that we attempt to investigate quantitatively. Despite this bias, many of the most potent new ligands were novel, dissimilar from known ligands, providing new lead structures for modulation of this medically important target.

The perceived and actual burden of false positives in high-throughput screening has received considerable attention; however, few studies exist on the contributions of distinct mechanisms of nonspecific effects like chemical reactivity, assay signal interference, and colloidal aggregation. Here, we analyze the outcome of a screen of 197861 diverse compounds in a concentration−response format against the cysteine protease cruzain, a target expected to be particularly sensitive to reactive compounds, and using an assay format with light detection in the short-wavelength region where significant compound autofluorescence is typically encountered. Approximately 1.9% of all compounds screened were detergent-sensitive inhibitors. The contribution from autofluorescence and compounds bearing reactive functionalities was dramatically lower: of all hits, only 1.8% were autofluorescent and 1.5% contained reactive or undesired functional groups. The distribution of false positives was relatively constant across library sources. The simple step of including detergent in the assay buffer suppressed the nonspecific effect of approximately 93% of the original hits.

In structure-based screens for new ligands, a molecular docking algorithm must rapidly score many molecules in multiple configurations, accounting for both the ligand’s interactions with receptor and its competing interactions with solvent. Here we explore a context-dependent ligand desolvation scoring term for molecular docking. We relate the Generalized-Born effective Born radii for every ligand atom to a fractional desolvation and then use this fraction to scale an atom-by-atom decomposition of the full transfer free energy. The fractional desolvation is precomputed on a scoring grid by numerically integrating over the volume of receptor proximal to a ligand atom, weighted by distance. To test this method’s performance, we dock ligands versus property-matched decoys over 40 DUD targets. Context-dependent desolvation better enriches ligands compared to both the raw full transfer free energy penalty and compared to ignoring desolvation altogether, though the improvement is modest. More compellingly, the new method improves docking performance across receptor types. Thus, whereas entirely ignoring desolvation works best for charged sites and overpenalizing with full desolvation works well for neutral sites, the physically more correct context-dependent ligand desolvation is competitive across both types of targets. The method also reliably discriminates ligands from highly charged molecules, where ignoring desolvation performs poorly. Since this context-dependent ligand desolvation may be precalculated, it improves docking reliability with minimal cost to calculation time and may be readily incorporated into any physics-based docking program.

Molecular biology now dominates pharmacology so thoroughly that it is difficult to recall that only a generation ago the field was very different. To understand drug action today, we characterize the targets through which they act and new drug leads are discovered on the basis of target structure and function. Until the mid-1980s the information often flowed in reverse: investigators began with organic molecules and sought targets, relating receptors not by sequence or structure but by their ligands. Recently, investigators have returned to this chemical view of biology, bringing to it systematic and quantitative methods of relating targets by their ligands. This has allowed the discovery of new targets for established drugs, suggested the bases for their side effects, and predicted the molecular targets underlying phenotypic screens. The bases for these new methods, some of their successes and liabilities, and new opportunities for their use are described.

Penicillin-binding protein 6 (PBP6) is one of the two main dd-carboxypeptidases in Escherichia coli, which are implicated in maturation of bacterial cell wall and formation of cell shape. Here, we report the first X-ray crystal structures of PBP6, capturing its apo state (2.1 Å), an acyl-enzyme intermediate with the antibiotic ampicillin (1.8 Å), and for the first time for a PBP, a preacylation complex (a “Michaelis complex”, determined at 1.8 Å) with a peptidoglycan substrate fragment containing the full pentapeptide, NAM-(l-Ala-d-isoGlu-l-Lys-d-Ala-d-Ala). These structures illuminate the molecular interactions essential for ligand recognition and catalysis by dd-carboxypeptidases, and suggest a coupling of conformational flexibility of active site loops to the reaction coordinate. The substrate fragment complex structure, in particular, provides templates for models of cell wall recognition by PBPs, as well as substantiating evidence for the molecular mimicry by β-lactam antibiotics of the peptidoglycan acyl-d-Ala-d-Ala moiety.

A docking screen identified reversible, noncovalent inhibitors (e.g., 1) of the parasite cysteine protease cruzain. Chemical optimization of 1 led to a series of oxadiazoles possessing interpretable SAR and potencies as much as 500-fold greater than 1. Detailed investigation of the SAR series subsequently revealed that many members of the oxadiazole class (and surprisingly also 1) act via divergent modes of inhibition (competitive or via colloidal aggregation) depending on the assay conditions employed.

One of the leading sources of false positives in early drug discovery is the formation of organic small molecule aggregates, which inhibit enzymes nonspecifically at micromolar concentrations in aqueous solution. The molecular basis for this widespread problem remains hazy. To investigate the mechanism of inhibition at a molecular level, we determined changes in solvent accessibility that occur when an enzyme binds to an aggregate using hydrogen−deuterium exchange mass spectrometry. For AmpC β-lactamase, binding to aggregates of the small molecule rottlerin increased the deuterium exchange of all 10 reproducibly detectable peptides, which covered 41% of the sequence of β-lactamase. This suggested a global increase in proton accessibility upon aggregate binding, consistent with denaturation. We then investigated whether enzyme−aggregate complexes were more susceptible to proteolysis than uninhibited enzyme. For five aggregators, trypsin degradation of β-lactamase increased substantially when β-lactamase was inhibited by aggregates, whereas uninhibited enzyme was generally stable to digestion. Combined, these results suggest that the mechanism of action of aggregate-based inhibitors proceeds via partial protein unfolding when bound to an aggregate particle.

Dr0930, a member of the amidohydrolase superfamily in Deinococcus radiodurans, was cloned, expressed, and purified to homogeneity. The enzyme crystallized in the space group P3121, and the structure was determined to a resolution of 2.1 Å. The protein folds as a (β/α)7β-barrel, and a binuclear metal center is found at the C-terminal end of the β-barrel. The purified protein contains a mixture of zinc and iron and is intensely purple at high concentrations. The purple color was determined to be due to a charge transfer complex between iron in the β-metal position and Tyr-97. Mutation of Tyr-97 to phenylalanine or complexation of the metal center with manganese abolished the absorbance in the visible region of the spectrum. Computational docking was used to predict potential substrates for this previously unannotated protein. The enzyme was found to catalyze the hydrolysis of δ- and γ-lactones with an alkyl substitution at the carbon adjacent to the ring oxygen. The best substrate was δ-nonanoic lactone with a kcat/Km of 1.6 × 106 M−1 s−1. Dr0930 was also found to catalyze the very slow hydrolysis of paraoxon with values of kcat and kcat/Km of 0.07 min−1 and 0.8 M−1 s−1, respectively. The amino acid sequence identity to the phosphotriesterase (PTE) from Pseudomonas diminuta is ∼30%. The eight substrate specificity loops were transplanted from PTE to Dr0930, but no phosphotriesterase activity could be detected in the chimeric PTE-Dr0930 hybrid. Mutation of Phe-26 and Cys-72 in Dr0930 to residues found in the active site of PTE enhanced the kinetic constants for the hydrolysis of paraoxon. The F26G/C72I mutant catalyzed the hydrolysis of paraoxon with a kcat of 1.14 min−1, an increase of 16-fold over the wild-type enzyme. These results support previous proposals that phosphotriesterase activity evolved from an ancestral parent enzyme possessing lactonase activity.

Aminergic G protein-coupled receptors (GPCRs) have been a major focus of pharmaceutical research for many years. Due partly to the lack of reliable receptor structures, drug discovery efforts have been largely ligand-based. The recently determined X-ray structure of the β2-adrenergic receptor offers an opportunity to investigate the advantages and limitations inherent in a structure-based approach to ligand discovery against this and related GPCR targets. Approximately 1 million commercially available, “lead-like” molecules were docked against the β2-adrenergic receptor structure. On testing of 25 high-ranking molecules, 6 were active with binding affinities <4 μM, with the best molecule binding with a Ki of 9 nM (95% confidence interval 7–10 nM). Five of these molecules were inverse agonists. The high hit rate, the high affinity of the most potent molecule, the discovery of unprecedented chemotypes among the new inhibitors, and the apparent bias toward inverse agonists among the docking hits, have implications for structure-based approaches against GPCRs that recognize small organic molecules.

Drugs that are chemically quite similar often bind to biologically diverse protein targets, and it is unclear how selective many of these compounds are. Because many drug–target combinations exist, it would be useful to explore possible interactions computationally. Here, 3,665 drugs are tested against hundreds of targets; chemical similarities between drugs and ligand sets are found to predict thousands of unanticipated associations.

Fragment screens for new ligands have had wide success, notwithstanding their constraint to libraries of 1,000–10,000 molecules. Larger libraries would be addressable were molecular docking reliable for fragment screens, but this has not been widely accepted. To investigate docking's ability to prioritize fragments, a library of >137,000 such molecules were docked against the structure of β-lactamase. Forty-eight fragments highly ranked by docking were acquired and tested; 23 had Ki values ranging from 0.7 to 9.2 mM. X-ray crystal structures of the enzyme-bound complexes were determined for 8 of the fragments. For 4, the correspondence between the predicted and experimental structures was high (RMSD between 1.2 and 1.4 Å), whereas for another 2, the fidelity was lower but retained most key interactions (RMSD 2.4–2.6 Å). Two of the 8 fragments adopted very different poses in the active site owing to enzyme conformational changes. The 48% hit rate of the fragment docking compares very favorably with “lead-like” docking and high-throughput screening against the same enzyme. To understand this, we investigated the occurrence of the fragment scaffolds among larger, lead-like molecules. Approximately 1% of commercially available fragments contain these inhibitors whereas only 10−7% of lead-like molecules do. This suggests that many more chemotypes and combinations of chemotypes are present among fragments than are available among lead-like molecules, contributing to the higher hit rates. The ability of docking to prioritize these fragments suggests that the technique can be used to exploit the better chemotype coverage that exists at the fragment level.

A current weakness in docking is the treatment of water-mediated protein−ligand interactions. We explore switching ordered water molecules “on” and “off” during docking screens of a large library. The method assumes additivity and scales linearly with the number of waters sampled despite the exponential growth in configurations. It is tested for ligand enrichment against 24 targets, exploring up to 256 water configurations. Water inclusion increased enrichment substantially for 12 targets, while most others were largely unaffected.

High-throughput screening (HTS) is widely used in drug discovery. Especially for screens of unbiased libraries, false positives can dominate “hit lists”; their origins are much debated. Here we determine the mechanism of every active hit from a screen of 70,563 unbiased molecules against β-lactamase using quantitative HTS (qHTS). Of the 1274 initial inhibitors, 95% were detergent-sensitive and were classified as aggregators. Among the 70 remaining were 25 potent, covalent-acting β-lactams. Mass spectra, counter-screens, and crystallography identified 12 as promiscuous covalent inhibitors. The remaining 33 were either aggregators or irreproducible. No specific reversible inhibitors were found. We turned to molecular docking to prioritize molecules from the same library for testing at higher concentrations. Of 16 tested, 2 were modest inhibitors. Subsequent X-ray structures corresponded to the docking prediction. Analog synthesis improved affinity to 8 µM. These results suggest that it may be the physical behavior of organic molecules, not their reactivity, that accounts for most screening artifacts. Structure-based methods may prioritize weak-but-novel chemotypes in unbiased library screens.

The similarity of drug targets is typically measured using sequence or structural information. Here, we consider chemo-centric approaches that measure target similarity on the basis of their ligands, asking how chemoinformatics similarities differ from those derived bioinformatically, how stable the ligand networks are to changes in chemoinformatics metrics, and which network is the most reliable for prediction of pharmacology. We calculated the similarities between hundreds of drug targets and their ligands and mapped the relationship between them in a formal network. Bioinformatics networks were based on the BLAST similarity between sequences, while chemoinformatics networks were based on the ligand-set similarities calculated with either the Similarity Ensemble Approach (SEA) or a method derived from Bayesian statistics. By multiple criteria, bioinformatics and chemoinformatics networks differed substantially, and only occasionally did a high sequence similarity correspond to a high ligand-set similarity. In contrast, the chemoinformatics networks were stable to the method used to calculate the ligand-set similarities and to the chemical representation of the ligands. Also, the chemoinformatics networks were more natural and more organized, by network theory, than their bioinformatics counterparts: ligand-based networks were found to be small-world and broad-scale.

At micromolar concentrations, many molecules form aggregates in aqueous solution. In this form, they inhibit enzymes non-specifically leading to false positive “hits” in enzyme assays, especially when screened in high-throughput. This inhibition can be attenuated by bovine serum albumin (BSA); the mechanism of this effect is not understood. Here we present evidence that BSA, lysozyme, and trypsin prevent inhibition when incubated at milligram per millilitre concentrations with aggregates prior to the addition of the monitored enzyme. These solutions still contained aggregates by dynamic light scattering (DLS), suggesting that inhibition is prevented by saturating the aggregate, rather than disrupting it. For most combinations of aggregate and protein, inhibition was not reversed if the competing protein was added after the incubation of aggregates with the monitored enzyme. In the one exception where modest reversal was observed, DLS and flow cytometry indicated that the effect was due to the disruption of aggregates. These results suggest that aggregate-bound enzyme is not in dynamic equilibrium with free enzyme and that bound enzyme cannot be displaced by a competing protein. To further test this hypothesis, we incubated aggregate-bound enzyme with a specific, irreversible inhibitor and then disrupted the aggregates with detergent. Most enzyme activity was restored on aggregate disruption, indicating no modification by the irreversible inhibitor. These results suggest that enzyme is bound to aggregate so tightly as to prevent any noticeable dissociation and that furthermore, aggregates are stable at physiologically relevant concentrations of protein.

With many genomes sequenced, a pressing challenge in biology is predicting the function of the proteins that the genes encode. When proteins are unrelated to others of known activity, bioinformatics inference for function becomes problematic. It would thus be useful to interrogate protein structures for function directly. Here, we predict the function of an enzyme of unknown activity, Tm0936 from Thermotoga maritima, by docking high-energy intermediate forms of thousands of candidate metabolites. The docking hit list was dominated by adenine analogues, which appeared to undergo C6-deamination. Four of these, including 5-methylthioadenosine and S-adenosylhomocysteine (SAH), were tested as substrates, and three had substantial catalytic rate constants (105 M-1 s-1). The X-ray crystal structure of the complex between Tm0936 and the product resulting from the deamination of SAH, S-inosylhomocysteine, was determined, and it corresponded closely to the predicted structure. The deaminated products can be further metabolized by T. maritima in a previously uncharacterized SAH degradation pathway. Structure-based docking with high-energy forms of potential substrates may be a useful tool to annotate enzymes for function.

Computer-based docking screens are now widely used to discover new ligands for targets of known structure; in the last two years alone, the discovery of ligands for more than 20 proteins has been reported. Recently, investigators have also turned to predicting new substrates for enzymes of unknown function, taking docking in a wholly new direction. Increasingly, the hit rates, the true-positives, and the false-positives from the docking screens are being compared to those from empirical, high-throughput screens, revealing the strengths, weaknesses, and complementarities of both techniques. The recent efflorescence of GPCR structures has made these quintessential drug targets available to structure-based approaches. Consistent with their ‘druggability’, the docking screens have returned high hit rates and potent molecules. Finally, in the last several years, an approach almost exactly opposite to docking has also appeared; this pharmacological network approach begins not with the structure of the target but rather those of drug molecules and asks, given a pattern of chemistry in the ligands, what targets may a particular drug bind to? This method, which returns to an older, pharmacology logic, has been surprisingly successful in predicting new ‘off-targets’ for established drugs.

Molecular docking computationally screens thousands to millions of organic molecules against protein structures, looking for those with complementary fits. Many approximations are made, often resulting in low “hit rates.” A strategy to overcome these approximations is to rescore top-ranked docked molecules using a better but slower method. One such is afforded by molecular mechanics–generalized Born surface area (MM–GBSA) techniques. These more physically realistic methods have improved models for solvation and electrostatic interactions and conformational change compared to most docking programs. To investigate MM–GBSA rescoring, we re-ranked docking hit lists in three small buried sites: a hydrophobic cavity that binds apolar ligands, a slightly polar cavity that binds aryl and hydrogen-bonding ligands, and an anionic cavity that binds cationic ligands. These sites are simple; consequently, incorrect predictions can be attributed to particular errors in the method, and many likely ligands may actually be tested. In retrospective calculations, MM–GBSA techniques with binding-site minimization better distinguished the known ligands for each cavity from the known decoys compared to the docking calculation alone. This encouraged us to test rescoring prospectively on molecules that ranked poorly by docking but that ranked well when rescored by MM–GBSA. A total of 33 molecules highly ranked by MM–GBSA for the three cavities were tested experimentally. Of these, 23 were observed to bind—these are docking false negatives rescued by rescoring. The 10 remaining molecules are true negatives by docking and false positives by MM–GBSA. X-ray crystal structures were determined for 21 of these 23 molecules. In many cases, the geometry prediction by MM–GBSA improved the initial docking pose and more closely resembled the crystallographic result; yet in several cases, the rescored geometry failed to capture large conformational changes in the protein. Intriguingly, rescoring not only rescued docking false positives, but also introduced several new false positives into the top-ranking molecules. We consider the origins of the successes and failures in MM–GBSA rescoring in these model cavity sites and the prospects for rescoring in biologically relevant targets.

A central challenge in structure-based ligand design is the accurate prediction of binding free energies. Here we apply alchemical free energy calculations in explicit solvent to predict ligand binding in a model cavity in T4 lysozyme. Even in this simple site, there are challenges. We made systematic improvements, beginning with single poses from docking, then including multiple poses, additional protein conformational changes, and using an improved charge model. Computed absolute binding free energies had an RMS error of 1.9 kcal/mol relative to previously determined experimental values. In blind prospective tests, the methods correctly discriminated between several true ligands and decoys in a set of putative binders identified by docking. In these prospective tests, the RMS error in predicted binding free energies relative to those subsequently determined experimentally was only 0.6 kcal/mol. X-ray crystal structures of the new ligands bound in the cavity corresponded closely to predictions from the free energy calculations, but sometimes differed from those predicted by docking. Finally, we examined the impact of holding the protein rigid, as in docking, with a view to learning how approximations made in docking affect accuracy and how they may be improved.

•Free-energy calculations are the state of the art for binding affinity prediction.•We blindly predicted affinities for charged compounds to a charged protein site.•New experiments measured 19 affinities and defined 13 ligand binding poses.•All predicted affinities were systematically too strong, based on ligand polarity.•Results show directions for improvement in modeling charges and affinity prediction.Predicting absolute protein–ligand binding affinities remains a frontier challenge in ligand discovery and design. This becomes more difficult when ionic interactions are involved because of the large opposing solvation and electrostatic attraction energies. In a blind test, we examined whether alchemical free-energy calculations could predict binding affinities of 14 charged and 5 neutral compounds previously untested as ligands for a cavity binding site in cytochrome c peroxidase. In this simplified site, polar and cationic ligands compete with solvent to interact with a buried aspartate. Predictions were tested by calorimetry, spectroscopy, and crystallography. Of the 15 compounds predicted to bind, 13 were experimentally confirmed, while 4 compounds were false negative predictions. Predictions had a root-mean-square error of 1.95 kcal/mol to the experimental affinities, and predicted poses had an average RMSD of 1.7 Å to the crystallographic poses. This test serves as a benchmark for these thermodynamically rigorous calculations at predicting binding affinities for charged compounds and gives insights into the existing sources of error, which are primarily electrostatic interactions inside proteins. Our experiments also provide a useful set of ionic binding affinities in a simplified system for testing new affinity prediction methods.Download high-res image (282KB)Download full-size image

Preorganization of enzyme active sites for substrate recognition typically comes at a cost to the stability of the folded form of the protein; consequently, enzymes can be dramatically stabilized by substitutions that attenuate the size and preorganization “strain” of the active site. How this stability–activity tradeoff constrains enzyme evolution has remained less certain, and it is unclear whether one should expect major stability insults as enzymes mutate towards new activities or how these new activities manifest structurally. These questions are both germane and easy to study in β-lactamases, which are evolving on the timescale of years to confer resistance to an ever-broader spectrum of β-lactam antibiotics. To explore whether stability is a substantial constraint on this antibiotic resistance evolution, we investigated extended-spectrum mutants of class C β-lactamases, which had evolved new activity versus third-generation cephalosporins. Five mutant enzymes had between 100-fold and 200-fold increased activity against the antibiotic cefotaxime in enzyme assays, and the mutant enzymes all lost thermodynamic stability (from 1.7 kcal mol− 1 to 4.1 kcal mol− 1), consistent with the stability–function hypothesis. Intriguingly, several of the substitutions were 10–20 Å from the catalytic serine; the question of how they conferred extended-spectrum activity arose. Eight structures, including complexes with inhibitors and extended-spectrum antibiotics, were determined by X-ray crystallography. Distinct mechanisms of action, including changes in the flexibility and ground-state structures of the enzyme, are revealed for each mutant. These results explain the structural bases for the antibiotic resistance conferred by these substitutions and their corresponding decrease in protein stability, which will constrain the evolution of new antibiotic resistance.

We present a combined experimental and modeling study of organic ligand molecules binding to a slightly polar engineered cavity site in T4 lysozyme (L99A/M102Q). For modeling, we computed alchemical absolute binding free energies. These were blind tests performed prospectively on 13 diverse, previously untested candidate ligand molecules. We predicted that eight compounds would bind to the cavity and five would not; 11 of 13 predictions were correct at this level. The RMS error to the measurable absolute binding energies was 1.8 kcal/mol. In addition, we computed “relative” binding free energies for six phenol derivatives starting from two known ligands: phenol and catechol. The average RMS error in the relative free energy prediction was 2.5 kcal/mol (phenol) and 1.1 kcal/mol (catechol). To understand these results at atomic resolution, we obtained x-ray co-complex structures for nine of the diverse ligands and for all six phenol analogs. The average RMSD of the predicted pose to the experiment was 2.0 Å (diverse set), 1.8 Å (phenol-derived predictions), and 1.2 Å (catechol-derived predictions). We found that predicting accurate affinities and rank-orderings required near-native starting orientations of the ligand in the binding site. Unanticipated binding modes, multiple ligand binding, and protein conformational change all proved challenging for the free energy methods. We believe that these results can help guide future improvements in physics-based absolute binding free energy methods.