Discovery of Ligand-Efficient Scaffolds for the Design of Novel Trichomonas vaginalis Uridine Nucleoside Ribohydrolase Inhibitors Using Fragment Screening.

Trichomoniasis is caused by the parasitic protozoan Trichomonas vaginalis. The increasing prevalence of strains resistant to the current 5-nitroimidazole treatments creates the need for novel therapies. T. vaginalis cannot synthesize purine and pyrimidine rings and requires salvage pathway enzymes to obtain them from host nucleosides. The uridine nucleoside ribohydrolase was screened using an 19F NMR-based activity assay against a 2000-compound fragment diversity library. Several series of inhibitors were identified including scaffolds based on acetamides, cyclic ureas or ureas, pyridines, and pyrrolidines. A number of potent singleton compounds were identified, as well. Eighteen compounds with IC50 values of 20 μM or lower were identified, including some with ligand efficiency values of 0.5 or greater. Detergent and jump-dilution counter screens validated all scaffold classes as target-specific, reversible inhibitors. Identified scaffolds differ substantially from 5-nitroimidazoles. Medicinal chemistry using the structure-activity relationship emerging from the fragment hits is being pursued to discover nanomolar inhibitors.

Introduction

Trichomoniasis is the most common sexually transmitted nonviral infection worldwide.[1] The World Health Organization estimated that there were 276.4 million cases in 2008 with 90% of these cases occurring in resource-limited areas.[1] In the United States, an estimated 3.7 million cases were reported by the Centers for Disease Control and Prevention.[2] Once infected, a person is more likely to become infected with chlamydia, gonorrhea, herpes simplex viruses type-1 and type-2, HIV, syphilis, and other sexually transmitted diseases.[2−4] Infections also increase the risk of developing bacterial vaginosis, candidiasis, pelvic inflammatory disease, and cervical and prostate cancer; pregnant women infected with trichomoniasis have an increased risk for low birth weight and preterm delivery.[2,3] Trichomonas is caused by the parasite Trichomonas vaginalis, a flagellated protozoan that is pyriform in shape. The majority of the time, T. vaginalis inhabits the squamous epithelium of the genital tract, performing fermentation using carbohydrates under both aerobic and anaerobic conditions.[3]

The current treatments for T. vaginalis infection are 5-nitroimidazoles.[5,6] This class of compounds, including metronidazole and tinidazole, is activated in the parasite’s hydrogenosomes; the nitro group is reduced by pyruvate–ferredoxin oxidoreductase creating toxic nitro radical anions, which damage thymine and adenine residues in the parasite’s DNA, causing the DNA to be cleaved and subsequent parasitic death.[5,6] Infections resulting from 5-nitroimidazole-resistant strains of T. vaginalis, however, are becoming more widespread, accounting for 5–17% of infections depending on the country.[7] New antitrichomonal agents with a mechanism of action distinct from existing drugs would provide a second line of therapy and would improve outcomes for the increasing number of patients with drug-resistant T. vaginalis infections.

Potential antitrichomonal drug targets include purine and pyrimidine salvage pathway enzymes. T. vaginalis lacks de novo biosynthetic pathways for purine and pyrimidine rings and relies on salvage pathway enzymes to metabolize nucleosides harvested from host cells.[8−10] The first step in these salvage pathways is the hydrolysis of nucleosides into their nucleobase and ribose components. The responsible enzymes belong to a superfamily of structurally related calcium-containing nucleoside ribohydrolases (NHs).[11] Previous studies have shown that all NHs have an active site highly specific to ribose, while specificity for the nucleobases is highly variable.[11] Despite the variability in substrate specificity of NHs, all enzymes of this class contain a Ca2+ cation deep within their active site coordinated by several conserved aspartic acid residues. The ribose portion of the nucleoside coordinates the Ca2+ cation via its 2′- and 3′-hydroxyl groups positioning the glycosidic bond for hydrolysis. A water molecule is also coordinated by the Ca2+ cation activating it for base-catalyzed hydrolysis of the N-glycosidic bond.[11]

While the ribose pockets of NH structures are highly conserved, the nucleobase pocket is more variable. The T. vaginalis genome[12] contains three confirmed NHs that we have cloned and characterized. Two are specific for purines, adenosine/guanosine nucleoside hydrolase (AGNH)[13] and guanosine/adenosine/cytidine nucleoside hydrolase (GACNH),[14] while the third is specific for pyrimidines, uridine nucleoside hydrolase (UNH).[15] Expressed sequence tags have been reported for all three T. vaginalis NHs.[16] The transcriptome of T. vaginalis under anaerobic conditions has been compared to that after exposure to oxygen and to vaginal epithelial cells[17] and has also been studied in response to glucose restriction.[18] Interestingly, transcripts for UNH were found to be up to 50-fold greater in number than those for either AGNH or GACNH depending on growth conditions. This might indicate the unique role played by this pyrimidine nucleoside ribohydrolase. Pyrimidine metabolism has been extensively studied in the related parasite Trypanosoma bruceibrucei, which, in contrast to T. vaginalis, is capable of both de novo biosynthesis and salvage of pyrimidines.[19] However, T. brucei brucei genetically modified to lack de novo pyrimidine biosynthesis capability was found to be completely dependent on salvage pathways, with the absence of pyrimidines in growth media rapidly lethal.[20] The addition of uracil returned growth rates to normal, while the addition of uridine only partially restored growth rates. This provides strong evidence that inhibiting the pyrimidine salvage pathway in T. vaginalis will be lethal to the parasite since this pathway represents its sole pyrimidine source.

We previously determined that UNH is a druggable target by developing an 19F NMR-based activity assay and then using it to screen the National Institutes of Health (NIH) Clinical Compound Collection for inhibitors.[15] Although the compounds in this collection have relatively large molar masses and lack chemical diversity, several benzimidazole-containing proton–pump inhibitors were identified as low micromolar inhibitors including omeprazole shown in Figure . Omeprazole has an IC50 value of 2.3 μM, but its relatively large molar mass of 345 g/mol combined with only modest ligand efficiency (LE)[21,22] of 0.36 (heavy atom count of 24) makes it a poor starting point for drug design.[23] A small hit deconstruction effort identified the fragment 2-methylthiobenzimidazole shown in Figure that has a molar mass of 164 g/mol and a much higher LE of 0.53 (heavy atom count of 11). These metrics suggest that 2-methylthiobenzimidazole could potentially be developed into a nanomolar inhibitor with a final molar mass less than 500 Da.[23,24] The relatively easy hit deconstruction of a modestly potent compound identified from a sampling of limited chemical diversity suggested that screening a large and diverse fragment library might identify multiple structure classes with better prospects for medicinal chemistry efforts.

Figure 1

Structures, potencies, and ligand efficiencies of omeprazole and its fragment 2-methylthiobenzimidazole.

Results and Discussion

The 19F NMR-based activity assay monitors the conversion of 5-fluorouridine to 5-fluorouracil and ribose.[15] While the reaction could also be monitored using 1H NMR of uridine/uracil, 19F NMR is advantageous because of the lack of possible overlaps with signals from the fragments themselves and its comparable sensitivity to 1H NMR.[25] Further, since the same two substrate and product 19F NMR signals are observed in every reaction, the effects of relaxation and chemical shift anisotropy that can complicate ligand-based 19F NMR screening methods are not a concern here.[26] The 50 μM concentration of 5-fluorouridine in the assay is three times its Km value of 15 μM creating assay conditions optimized for detecting inhibitors with competitive, noncompetitive, or uncompetitive mechanisms.[26] At the 333 μM fragment concentrations screened, a competitive fragment inhibitor would need to have a KI of only 77 μM to result in 50% inhibition.

Mixtures of six fragments were initially screened, with mixtures that exhibited at least 75% inhibition subsequently deconvoluted to determine the individual inhibitory fragments. Figure shows typical spectra for six mixtures along with the 0 and 30 min control spectra. Only the substrate signal at −165.8 ppm is observed in the 0 min control spectrum, while both the substrate signal and a new product signal at −169.2 ppm are observed in the 30 min control spectrum. The product peak is also present in all mixture spectra except that for mixture 5, suggesting that at least one compound in mixture 5 is an inhibitor of UNH. Figure shows the spectra for the deconvolution of mixture 5 using its six individual components. The product peak is present in all compounds tested with the exception of fragment G7. The absence of the product peak at −169.2 ppm indicates that fragment G7 fully inhibits UNH at 333 μM. The observation of residual substrate signals for fragments C8 and D8 indicates that these fragments are also inhibitors but much weaker. Fragments demonstrating 75% or greater inhibition in the deconvolution assays were assayed in five-point serial dilutions down to 1.3 μM. Fragment IC50 values or percent inhibition at 333 μM were then determined. A total of 33 fragments selected to represent the various chemical classes of inhibitors were obtained as solid samples to confirm activity. These compounds were dissolved in dimethyl sulfoxide (DMSO) and retested from 1.33 mM to 0.33 μM, as shown in Figure for fragment G7. The IC50 value for fragment G7 (subsequently referred to as fragment 7) calculated from this data is 45 μM.

Figure 2

19F NMR reaction spectra for the 0 and 30 min controls and six mixtures. The maximum intensities of the substrate (−165.8 ppm) and product (−169.2 ppm) signals are observed in the 0 and 30 min controls, respectively.

Figure 3

19F NMR reaction spectra for the 0 and 30 min controls and the individual components of mixture 5 (G7–D8).

Figure 4

19F NMR reaction spectra for 0 and 30 min controls and variable concentrations indicated of fragment 7.

A total of 97 fragments exhibited inhibition against UNH (4.9% hit rate). Several series of inhibitors with emerging structure–activity relationship (SAR) were identified including scaffolds based on acetamides, cyclic ureas or ureas, pyridines, and pyrrolidines. A number of potent singleton compounds were identified, as well. A singleton was defined as having no other closely related fragments in the screen, based on substructure searching of the core scaffold. Among the active fragments were 18 compounds with IC50 values of 20 μM or lower, including some with ligand efficiency values of 0.5 or greater. The structures, IC50 values, and LE values for nine fragments representative of the most common scaffolds are shown in Table . A total of 55 structural analogs of the most potent fragments were also obtained and tested. The emerging SAR from these fragments and the original screening hits are discussed below.

Table 1

Structures, IC50 Values, and LE Values of the Most Common Fragment Scaffoldsa

Values are the average of n = 2.

Interestingly, several of the fragment classes contain moieties that are components of active compounds identified in our previous screen of the NIH Clinical Compound Collection. For instance, the 2,3-substituted pyridine ring of fragments 4 and 5 is one component of the prazole class of compounds represented by omeprazole in Figure . The phenylpyridine component of fragment 7 is also very nearly the core 4-phenyl-1,4-dihydropyridine scaffold of the dipine compounds, such as nifedipine and nicardipine, which were the largest class of hits in the previous screen. Thus, the SAR from the previous NIH Clinical Compound Collection can be integrated in some circumstances in the context of the SAR from this emerging work to advance selected hit series.

All compound structural classes were validated as reversible, target-specific inhibitors based on four criteria.[27] First, the fragment library was designed to exclude PAINS chemotypes,[28] and this was verified for the fragments shown in Table using ZINC.[29] Further, the fragments shown in Table were analyzed by the program Badapple, an algorithm that detects likely patterns of promiscuity in molecular scaffolds.[30] High scores were indicated only for fragments containing biphenyl, phenylpyrrolidine, and pyridine chemotypes. Second, the lack of reporter enzymes in the NMR-based activity assays eliminates the possibility of false positives acting by this interference mechanism. Third, detergent counter screens were carried out to reduce the incidence of false positives arising from colloidal aggregates that can mimic inhibition by blocking the enzyme’s active site.[31] The Shoichet protocol was used to test for aggregation-based inhibition, where the nonionic detergent Triton X-100 will prevent aggregates from interacting with the enzyme nonspecifically.[31] If activity diminishes markedly with detergent, the compound is most likely an aggregator. Figure demonstrates the effect of detergent on the inhibition observed for fragment 7. Both control reactions have approximately 50% conversion of substrate to product, while reactions with 100 μM fragment 7 in the absence and presence of 0.01% Triton X-100 detergent show close to complete inhibition. Lack of significant change in potency with or without detergent indicates that the inhibition observed for fragment 7 is likely not aggregation-based. Similar results were obtained for all other fragments tested, indicating that all classes are target-specific inhibitors. In these experiments, 19F NMR is actually disadvantageous compared to 1H NMR. When using the latter, resonances from the fragments themselves are often simultaneously indicative of well-behaved, soluble compounds.[32,33] Fourth, jump-dilution assays were carried out to confirm that the fragment hits are noncovalent, reversible inhibitors.[34] Jump-dilution assays include a parallel incubation of the enzyme and compound at 200 μM before initiating normal reaction assays at 200 and 20 μM (10-fold dilution). Fragment 7 completely inhibited UNH at 200 μM, as shown in Figure . Full inhibition is expected since the fragment concentration is 4-fold higher than its IC50 value of 45 μM. However, upon rapid dilution to 20 μM before initiating the reaction, UNH inhibition dropped to 54%, as shown in Figure . Loss of activity indicates dissociation of the compound from the active site. Similar results were obtained for all other fragments tested, indicating that all classes are noncovalent, reversible inhibitors. Fragment SAR and dose–response curve shapes also suggest that the identified fragments are suitable for follow-up studies.

Figure 5

19F NMR reaction spectra for control (B, D) and 100 μM fragment 7 (A, C) in the absence or presence of detergent as indicated.

Figure 6

19F NMR reaction spectra for control (B, D) and fragment 7 (A, C) jump-dilution counter screens. Incubation and reaction concentrations are indicated.

The impetus to screen a fragment diversity library came from the observation that fragments with high LE values could be identified from larger-molecular-weight inhibitors, as shown in Figure . Screening a large set of diverse fragments might then lead to the identification of one or more scaffolds more optimal for medicinal chemistry elaboration. Identification of scaffold classes with more than five representatives indicates the success of the fragment approach and provides excellent starting points for ongoing work. Further, the scaffold classes identified are markedly different from those identified in our previous fragment screen of the same library against the purine-specific AGNH enzyme. Of the 60 fragments with IC50 values <100 μM for UNH, only nine also had IC50 values <100 μM for AGNH. This suggests that while the ribose binding pockets of the AGNH and UNH active sites are likely very similar, the nucleobase binding pockets possess markedly different molecular complementarities.

Several scaffold classes contain fragments with LE values greater than 0.5, indicating that the majority of the atoms make favorable interactions in the active site. It is important to start out with high LE values since during the optimization process, the efficiency will only remain the same or decrease as the size of the molecule is increased.[23] For instance, the molar masses and LE values for all of the fragments in Table with the exception of fragment 6 suggest that they can each be developed into nanomolar inhibitors with final molar masses less than 500 Da provided that LE remains constant as molar mass increases. Thus, the fragment screening output provides at least four chemical scaffolds that are attractive starting points for a chemical optimization program. Some fragments and fragment classes also appear to have overlapping structural features that may suggest fragment merging strategies, as well.

The 3-hydroxypyrrolidine fragment 9 is compelling for its combination of potency, LE, and emerging SAR. Figure compares the structure of fragment 9 with that of uridine. Interestingly, the inhibitor has almost one-third greater LE than the enzyme’s natural substrate. The lower LE for the substrate results from some of the binding energy being used to lower the activation energy for the reaction, thus reducing binding affinity. Inhibitors (nonsubstrates) do not have this limitation and thus can have higher LE values. There are, at present, no reported structures of pyrimidine-specific nucleoside hydrolases with a bound heterocyclic nucleobase in the active site. However, modeling studies on the pyrimidine-specific enzymes from Escherichia coli and Sulfolobus solfataricus indicate that both contain hydrophilic residues lining the active site that could potentially hydrogen-bond with the polar regions of fragment 9 and the substrate.[35,36] In addition, both fragment 9 and uridine have 6-membered aromatic/heteroaromatic rings that have a hydrophobic face that may make similar interactions in the active site. Both structures also contain 5-membered, saturated rings with attached hydroxyl groups. The pyrrolidine moiety of fragment 9 likely interacts with the ribose pocket of the active site. As previously discussed, nucleoside hydrolases have a highly conserved Ca2+ cation within their active sites.[11] Fragment 9 has a hydroxyl group in a similar position as uridine. It is highly probable that this hydroxyl group is interacting with the Ca2+ cation in the same manner as the 2′ hydroxyl group in uridine. Modeling and X-ray crystallography studies to validate these interactions and to guide inhibitor design are in progress.

Figure 7

Structures, IC50 or Km value, and LE values for fragment 9 and uridine.

A total of six 3-hydroxypyrrolidines were identified as inhibitors including fragments 8 and 9 from the original fragment library and fragments 10–13 shown in Table from the fragment hit structural analogs tested. Replacement of the methylamino group in fragment 9 with a nitrile group in fragment 10 improved potency, with fragment 10 having an IC50 value of 13 μM. By contrast, the addition of two meta methyl groups as in fragment 11, a para ethyl group as in fragment 8, and an ethoxy group as in fragment 12 resulted in a steady decrease in potency compared to that in fragment 9. This suggests either a steric hindrance or a limit to the hydrophobic character in this region of the active site and that the nonpolar edge of the uracil-like ring is a poor vector for picking up new interactions. Further, the pyridine ring fragment 13 had only barely detectable activity suggesting that the ring nitrogen is in the wrong position to pick up similar interactions in the active site that are responsible for substrate specificity. Adding hydrogen-bonding groups that mimic those of the uracil ring may improve activity.

Table 2

3-Hydroxypyrrolidine Structures and IC50 Valuesa

Values are the average of n = 2.

Conclusions

Fragment hits identified here provide ideal starting points to synthesize the tool compounds required to demonstrate that UNH inhibition is correlated with antitrichomonal activity. This will require improvements in UNH potency by several orders of magnitude, down into the 10 nM range. In addition to having small molecular weight and favorable aqueous solubility, the diverse compounds in the fragment library were selected for their potential to be elaborated on using medicinal chemistry protocols. Thus, the scaffolds identified with high ligand efficiencies can be chemically expanded using known synthetic organic chemistry approaches. This process will enable the development of larger compounds with improved UNH activity that meet the criteria for in vitro testing. Inhibitors active against T. vaginalis may also be broadly applicable to other neglected parasites that require nucleoside hydrolase enzymes for their survival such as Leishmania donovani.[37]

Experimental Section

NMR data sets were acquired on a Bruker AvanceIII 500 MHz spectrometer using a BBFO probe. 19F{1H} NMR spectra were acquired using inverse-gated decoupling with WALTZ-16.[38] Spectra were the average of 256 scans and included acquisition and relaxation delay times of 0.872 s and 4.0 s, respectively. 19F chemical shifts were referenced to external 50 μM trifluoroethanol at −76.7 ppm. The physical properties of the 1963 fragments screened, a diversity-based subset of the AstraZeneca fragment library, have been described previously.[33]

Sequentially, 3 μL of 10 mM 5-fluorouridine and 2 μL of each fragment to be tested were added to microfuge tubes. Reactions were then initiated using a stock solution consisting of 50 mM potassium phosphate and 0.3 M KCl at pH 6.5, 10% 2H2O, and 80 nM UNH to give a final volume of 600 μL. Reactions were quenched after 30 min with 10 μL of 1.5 M HCl. The highest DMSO concentrations used were 2%, which did not measurably affect enzyme activity. In all cases, control reactions were also run by quenching at 0 and 30 min in the presence of the same DMSO concentration but the absence of fragments. Serial dilution assays were carried out in duplicate and analyzed as previously described, maintaining a constant DMSO concentration for each dilution.[33]

Jump-dilution and detergent counter screens were carried out as previously described.[32,33] The IC50 values of the fragments used in these experiments were well suited to the 200 and 20 μM fragment concentrations used in the jump-dilution assays, as well as the 100 and 50 μM fragment concentrations used in the detergent assays.